Findings of the Association for Computational Linguistics: ACL 2023

Anna Rogers, Jordan Boyd-Graber, Naoaki Okazaki (Editors)

Anthology ID:: 2023.findings-acl
Month:: July
Year:: 2023
Address:: Toronto, Canada
Venue:: Findings
SIG:
Publisher:: Association for Computational Linguistics
URL:: https://aclanthology.org/2023.findings-acl
DOI:
Bib Export formats:: BibTeX

pdf bib
Findings of the Association for Computational Linguistics: ACL 2023
Anna Rogers | Jordan Boyd-Graber | Naoaki Okazaki

pdf bib abs
Investigating Glyph-Phonetic Information for Chinese Spell Checking: What Works and What’s Next?
Xiaotian Zhang | Yanjun Zheng | Hang Yan | Xipeng Qiu

While pre-trained Chinese language models have demonstrated impressive performance on a wide range of NLP tasks, the Chinese Spell Checking (CSC) task remains a challenge. Previous research has explored using information such as glyphs and phonetics to improve the ability of CSC models to distinguish misspelled characters, with good results at the accuracy level on public datasets. However, the generalization ability of these CSC models has not been well understood: it is unclear whether they incorporate glyph-phonetic information and, if so, whether this information is fully utilized. In this paper, we aim to better understand the role of glyph-phonetic information in the CSC task and suggest directions for improvement. Additionally, we propose a new, more challenging, and practical setting for testing the generalizability of CSC models. All code is made publicly available.

pdf bib abs
A Self-Supervised Integration Method of Pretrained Language Models and Word Definitions
Hwiyeol Jo

We investigate the representation of pretrained language models and humans, using the idea of word definition modeling–how well a word is represented by its definition, and vice versa. Our analysis shows that a word representation in pretrained language models does not successfully map its human-written definition and its usage in example sentences. We then present a simple method DefBERT that integrates pretrained models with word semantics in dictionaries. We show its benefits on newly-proposed tasks of definition ranking and definition sense disambiguation. Furthermore, we present the results on standard word similarity tasks and short text classification tasks where models are required to encode semantics with only a few words. The results demonstrate the effectiveness of integrating word definitions and pretrained language models.

pdf abs
Conformal Nucleus Sampling
Shauli Ravfogel | Yoav Goldberg | Jacob Goldberger

Language models generate text based on successively sampling the next word. A decoding procedure based on nucleus (top-p) sampling chooses from the smallest possible set of words whose cumulative probability exceeds the probability p. In this work, we assess whether a top-p set is indeed aligned with its probabilistic meaning in various linguistic contexts.We employ conformal prediction, a calibration procedure that focuses on the construction of minimal prediction sets according to a desired confidence level, to calibrate the parameter p as a function of the entropy of the next word distribution. We find that OPT models are overconfident, and that calibration shows a moderate inverse scaling with model size.

Implicit Discourse Relation Recognition (IDRR) is a sophisticated and challenging task to recognize the discourse relations between the arguments with the absence of discourse connectives. The sense labels for each discourse relation follow a hierarchical classification scheme in the annotation process (Prasad et al., 2008), forming a hierarchy structure. Most existing works do not well incorporate the hierarchy structure but focus on the syntax features and the prior knowledge of connectives in the manner of pure text classification. We argue that it is more effective to predict the paths inside the hierarchical tree (e.g., “Comparison -> Contrast -> however”) rather than flat labels (e.g., Contrast) or connectives (e.g., however). We propose a prompt-based path prediction method to utilize the interactive information and intrinsic senses among the hierarchy in IDRR. This is the first work that injects such structure information into pre-trained language models via prompt tuning, and the performance of our solution shows significant and consistent improvement against competitive baselines.

pdf abs
Modularized Zero-shot VQA with Pre-trained Models
Rui Cao | Jing Jiang

Large-scale pre-trained models (PTMs) show great zero-shot capabilities. In this paper, we study how to leverage them for zero-shot visual question answering (VQA).Our approach is motivated by a few observations. First, VQA questions often require multiple steps of reasoning, which is still a capability that most PTMs lack. Second, different steps in VQA reasoning chains require different skills such as object detection and relational reasoning, but a single PTM may not possess all these skills. Third, recent work on zero-shot VQA does not explicitly consider multi-step reasoning chains, which makes them less interpretable compared with a decomposition-based approach. We propose a modularized zero-shot network that explicitly decomposes questions into sub reasoning steps and is highly interpretable. We convert sub reasoning tasks to acceptable objectives of PTMs and assign tasks to proper PTMs without any adaptation. Our experiments on two VQA benchmarks under the zero-shot setting demonstrate the effectiveness of our method and better interpretability compared with several baselines.

Lifelogs are descriptions of experiences that a person had during their life. Lifelogs are created by fusing data from the multitude of digital services, such as online photos, maps, shopping and content streaming services. Question answering over lifelogs can offer personal assistants a critical resource when they try to provide advice in context. However, obtaining answers to questions over lifelogs is beyond the current state of the art of question answering techniques for a variety of reasons, the most pronounced of which is that lifelogs combine free text with some degree of structure such as temporal and geographical information. We create and publicly release TimelineQA, a benchmark for accelerating progress on querying lifelogs. TimelineQA generates lifelogs of imaginary people. The episodes in the lifelog range from major life episodes such as high school graduation to those that occur on a daily basis such as going for a run. We describe a set of experiments on TimelineQA with several state-of-the-art QA models. Our experiments reveal that for atomic queries, an extractive QA system significantly out-performs a state-of-the-art retrieval-augmented QA system. For multi-hop queries involving aggregates, we show that the best result is obtained with a state-of-the-art table QA technique, assuming the ground truth set of episodes for deriving the answer is available.

pdf abs
Abstractive Text Summarization Using the BRIO Training Paradigm
Khang Lam | Thieu Doan | Khang Pham | Jugal Kalita

Summary sentences produced by abstractive summarization models may be coherent and comprehensive, but they lack control and rely heavily on reference summaries. The BRIO training paradigm assumes a non-deterministic distribution to reduce the model’s dependence on reference summaries, and improve model performance during inference. This paper presents a straightforward but effective technique to improve abstractive summaries by fine-tuning pre-trained language models, and training them with the BRIO paradigm. We build a text summarization dataset for Vietnamese, called VieSum. We perform experiments with abstractive summarization models trained with the BRIO paradigm on the CNNDM and the VieSum datasets. The results show that the models, trained on basic hardware, outperform all existing abstractive summarization models, especially for Vietnamese.

pdf abs
Modeling the Q-Diversity in a Min-max Play Game for Robust Optimization
Ting Wu | Rui Zheng | Tao Gui | Qi Zhang | Xuanjing Huang

Models trained with empirical risk minimization (ERM) are revealed to easily rely on spurious correlations, resulting in poor generalization. Group distributionally robust optimization (group DRO) can alleviate this problem by minimizing the worst-case loss over pre-defined groups. While promising, in practice factors like expensive annotations and privacy preclude the availability of group labels. More crucially, when taking a closer look at the failure modes of out-of-distribution generalization, the typical procedure of reweighting in group DRO loses efficiency. Hinged on the limitations, in this work, we reformulate the group DRO framework by proposing Q-Diversity. Characterized by an interactive training mode, Q-Diversity relaxes the group identification from annotation into direct parameterization. Furthermore, a novel mixing strategy across groups is presented to diversify the under-represented groups. In a series of experiments on both synthetic and real-world text classification tasks, results demonstrate that Q-Diversity can consistently improve worst-case accuracy under different distributional shifts, outperforming state-of-the-art alternatives.

ELECTRA, the generator-discriminator pre-training framework, has achieved impressive semantic construction capability among various downstream tasks. Despite the convincing performance, ELECTRA still faces the challenges of monotonous training and deficient interaction. Generator with only masked language modeling (MLM) leads to biased learning and label imbalance for discriminator, decreasing learning efficiency; no explicit feedback loop from discriminator to generator results in the chasm between these two components, underutilizing the course learning. In this study, a multi-perspective course learning (MCL) method is proposed to fetch a many degrees and visual angles for sample-efficient pre-training, and to fully leverage the relationship between generator and discriminator. Concretely, three self-supervision courses are designed to alleviate inherent flaws of MLM and balance the label in a multi-perspective way. Besides, two self-correction courses are proposed to bridge the chasm between the two encoders by creating a “correction notebook” for secondary-supervision. Moreover, a course soups trial is conducted to solve the “tug-of-war” dynamics problem of MCL, evolving a stronger pre-trained model. Experimental results show that our method significantly improves ELECTRA’s average performance by 2.8% and 3.2% absolute points respectively on GLUE and SQuAD 2.0 benchmarks, and overshadows recent advanced ELECTRA-style models under the same settings. The pre-trained MCL model is available at https://huggingface.co/McmanusChen/MCL-base.

pdf abs
Layerwise universal adversarial attack on NLP models
Olga Tsymboi | Danil Malaev | Andrei Petrovskii | Ivan Oseledets

In this work, we examine the vulnerability of language models to universal adversarial triggers (UATs). We propose a new white-box approach to the construction of layerwise UATs (LUATs), which searches the triggers by perturbing hidden layers of a network. On the example of three transformer models and three datasets from the GLUE benchmark, we demonstrate that our method provides better transferability in a model-to-model setting with an average gain of 9.3% in the fooling rate over the baseline. Moreover, we investigate triggers transferability in the task-to-task setting. Using small subsets from the datasets similar to the target tasks for choosing a perturbed layer, we show that LUATs are more efficient than vanilla UATs by 7.1% in the fooling rate.

pdf abs
Scene-robust Natural Language Video Localization via Learning Domain-invariant Representations
Zehan Wang | Yang Zhao | Haifeng Huang | Yan Xia | Zhou Zhao

Natural language video localization(NLVL) task involves the semantic matching of a text query with a moment from an untrimmed video. Previous methods primarily focus on improving performance with the assumption of independently identical data distribution while ignoring the out-of-distribution data. Therefore, these approaches often fail when handling the videos and queries in novel scenes, which is inevitable in real-world scenarios. In this paper, we, for the first time, formulate the scene-robust NLVL problem and propose a novel generalizable NLVL framework utilizing data in multiple available scenes to learn a robust model. Specifically, our model learns a group of generalizable domain-invariant representations by alignment and decomposition. First, we propose a comprehensive intra- and inter-sample distance metric for complex multi-modal feature space, and an asymmetric multi-modal alignment loss for different information densities of text and vision. Further, to alleviate the conflict between domain-invariant features for generalization and domain-specific information for reasoning, we introduce domain-specific and domain-agnostic predictors to decompose and refine the learned features by dynamically adjusting the weights of samples. Based on the original video tags, we conduct extensive experiments on three NLVL datasets with different-grained scene shifts to show the effectiveness of our proposed methods.

pdf abs
Exploiting Pseudo Image Captions for Multimodal Summarization
Chaoya Jiang | Rui Xie | Wei Ye | Jinan Sun | Shikun Zhang

Multimodal summarization with multimodal output (MSMO) faces a challenging semantic gap between visual and textual modalities due to the lack of reference images for training. Our pilot investigation indicates that image captions, which naturally connect texts and images, can significantly benefit MSMO. However, exposure of image captions during training is inconsistent with MSMO’s task settings, where prior cross-modal alignment information is excluded to guarantee the generalization of cross-modal semantic modeling. To this end, we propose a novel coarse-to-fine image-text alignment mechanism to identify the most relevant sentence of each image in a document, resembling the role of image captions in capturing visual knowledge and bridging the cross-modal semantic gap. Equipped with this alignment mechanism, our method easily yet impressively sets up state-of-the-art performances on all intermodality and intramodality metrics (e.g., more than 10% relative improvement on image recommendation precision). Further experiments reveal the correlation between image captions and text summaries, and prove that the pseudo image captions we generated are even better than the original ones in terms of promoting multimodal summarization.

pdf abs
Cross-Lingual Transfer with Target Language-Ready Task Adapters
Marinela Parovic | Alan Ansell | Ivan Vulić | Anna Korhonen

Adapters have emerged as a modular and parameter-efficient approach to (zero-shot) cross-lingual transfer. The established MAD-X framework employs separate language and task adapters which can be arbitrarily combined to perform the transfer of any task to any target language. Subsequently, BAD-X, an extension of the MAD-X framework, achieves improved transfer at the cost of MAD-X’s modularity by creating ‘bilingual’ adapters specific to the source-target language pair. In this work, we aim to take the best of both worlds by (i) fine-tuning *task* adapters adapted to the target language(s) (so-called *‘target language-ready’ (TLR)* adapters) to maintain high transfer performance, but (ii) without sacrificing the highly modular design of MAD-X. The main idea of ‘target language-ready’ adapters is to resolve the training-vs-inference discrepancy of MAD-X: the task adapter ‘sees’ the target language adapter for the very first time during inference, and thus might not be fully compatible with it. We address this mismatch by exposing the task adapter to the target language adapter during training, and empirically validate several variants of the idea: in the simplest form, we alternate between using the source and target language adapters during task adapter training, which can be generalized to cycling over any set of language adapters. We evaluate different TLR-based transfer configurations with varying degrees of generality across a suite of standard cross-lingual benchmarks, and find that the most general (and thus most modular) configuration consistently outperforms MAD-X and BAD-X on most tasks and languages.

Dynamic topic models (DTMs) analyze text streams to capture the evolution of topics. Despite their popularity, existing DTMs are either fully supervised, requiring expensive human annotations, or fully unsupervised, producing topic evolutions that often do not cater to a user’s needs. Further, the topic evolutions produced by DTMs tend to contain generic terms that are not indicative of their designated time steps. To address these issues, we propose the task of discriminative dynamic topic discovery. This task aims to discover topic evolutions from temporal corpora that distinctly align with a set of user-provided category names and uniquely capture topics at each time step. We solve this task by developing DynaMiTE, a framework that ensembles semantic similarity, category indicative, and time indicative scores to produce informative topic evolutions. Through experiments on three diverse datasets, including the use of a newly-designed human evaluation experiment, we demonstrate that DynaMiTE is a practical and efficient framework for helping users discover high-quality topic evolutions suited to their interests.

pdf abs
Boost Transformer-based Language Models with GPU-Friendly Sparsity and Quantization
Chong Yu | Tao Chen | Zhongxue Gan

Along with the performance improvement in NLP domain, the sizes of transformer-based language models (TLM) are also dramatically increased. Some prior works intend to compress TLM models into more compact forms, but do not fully consider the hardware characters may not support the efficient execution for these forms, leading to the deployment of TLM on hardware with noticeable acceleration is still challenging. This paper thoroughly designs a compression scheme named GPUSQ-TLM to maximally utilize the GPU-friendly 2:4 fine-grained structured sparsity and quantization characters. Especially, a dense TLM model is first pruned to meet the GPU’s acceleration constraint of sparse patterns with FP16 type, then it is further quantized into a fixed-point one by quantization-aware training, to provide an extra speedup for integer tensors on GPU. A mixed-strategy knowledge distillation of labels, logits and feature maps is used for best accuracy compensation during pruning and quantization process. Experiment results show GPUSQ-TLM scheme achieves state-of-the-art compression on TLM model of various encoder and decoder blocks with negligible accuracy degradation on SQuAD, GLUE, CNN-DM & XSum and WikiText benchmarking tasks. Moreover, GPUSQ-TLM can boost actual deployment performance by up to 4.08-4.25x latency and 6.18-6.79x throughput on A100 GPU.

We are interested in a challenging task, Realistic-Music-Score based Singing Voice Synthesis (RMS-SVS). RMS-SVS aims to generate high-quality singing voices given realistic music scores with different note types (grace, slur, rest, etc.). Though significant progress has been achieved, recent singing voice synthesis (SVS) methods are limited to fine-grained music scores, which require a complicated data collection pipeline with time-consuming manual annotation to align music notes with phonemes. % Furthermore, existing approaches cannot synthesize rhythmic singing voices given realistic music scores due to the domain gap between fine-grained music scores and realistic music scores. Furthermore, these manual annotation destroys the regularity of note durations in music scores, making fine-grained music scores inconvenient for composing. To tackle these challenges, we propose RMSSinger, the first RMS-SVS method, which takes realistic music scores as input, eliminating most of the tedious manual annotation and avoiding the aforementioned inconvenience. Note that music scores are based on words rather than phonemes, in RMSSinger, we introduce word-level modeling to avoid the time-consuming phoneme duration annotation and the complicated phoneme-level mel-note alignment. Furthermore, we propose the first diffusion-based pitch modeling method, which ameliorates the naturalness of existing pitch-modeling methods. To achieve these, we collect a new dataset containing realistic music scores and singing voices according to these realistic music scores from professional singers. Extensive experiments on the dataset demonstrate the effectiveness of our methods. Audio samples are available at https://rmssinger.github.io/.

pdf abs
Zero-Shot Prompting for Implicit Intent Prediction and Recommendation with Commonsense Reasoning
Hui-Chi Kuo | Yun-Nung Chen

The current generation of intelligent assistants require explicit user requests to perform tasks or services, often leading to lengthy and complex conversations. In contrast, human assistants can infer multiple implicit intents from utterances via their commonsense knowledge, thereby simplifying interactions. To bridge this gap, this paper proposes a framework for multi-domain dialogue systems. This framework automatically infers implicit intents from user utterances, and prompts a large pre-trained language model to suggest suitable task-oriented bots. By leveraging commonsense knowledge, our framework recommends associated bots in a zero-shot manner, enhancing interaction efficiency and effectiveness. This approach substantially reduces interaction complexity, seamlessly integrates various domains and tasks, and represents a significant step towards creating more human-like intelligent assistants that can reason about implicit intents, offering a superior user experience.

Target-oriented dialogue guides the dialogue to a target quickly and smoothly. The latest approaches focus on global planning, which plans toward the target before the conversation instead of adopting a greedy strategy during the conversation. However, the global plan in existing works is fixed to certain turns by generating paths with certain nodes, which limits the optimization of turns and coherence of the target-oriented process. Toward flexible global planning, we propose to generate a global path as a natural language sentence instead of a sequence of nodes. With this path, the dialog is guided to the target with flexible turns of dialog. For model training, we also extract targetoriented dialogues from the chit-chat corpus with a knowledge graph. We conduct experiments on three datasets and simulate scenarios with and without user participation. The results show that our method has fewer turns, more coherent semantics, and a higher success rate in reaching the target than baselines.

pdf abs
The Larger they are, the Harder they Fail: Language Models do not Recognize Identifier Swaps in Python
Antonio Valerio Miceli Barone | Fazl Barez | Shay B. Cohen | Ioannis Konstas

Large Language Models (LLMs) have successfully been applied to code generation tasks, raising the question of how well these models understand programming. Typical programming languages have invariances and equivariances in their semantics that human programmers intuitively understand and exploit, such as the (near) invariance to the renaming of identifiers. We show that LLMs not only fail to properly generate correct Python code when default function names are swapped, but some of them even become more confident in their incorrect predictions as the model size increases, an instance of the recently discovered phenomenon of Inverse Scaling, which runs contrary to the commonly observed trend of increasing prediction quality with increasing model size. Our findings indicate that, despite their astonishing typical-case performance, LLMs still lack a deep, abstract understanding of the content they manipulate, making them unsuitable for tasks that statistically deviate from their training data, and that mere scaling is not enough to achieve such capability.

Intent detection, which estimates diverse intents behind user utterances, is an essential component of task-oriented dialogue systems. Previous intent detection models are usually trained offline, which can only handle predefined intent classes. In the real world, new intents may keep challenging deployed models. For example, with the prevalence of the COVID-19 pandemic, users may pose various issues related to the pandemic to conversational systems, which brings many new intents. A general intent detection model should be intelligent enough to continually learn new data and recognize new arriving intent classes. Therefore, this work explores Class Lifelong Learning for Intent Detection (CLL-ID), where the model continually learns new intent classes from new data while avoiding catastrophic performance degradation on old data. To this end, we propose a novel lifelong learning method, called Structure Consolidation Networks (SCN), which consists of structure-based retrospection and contrastive knowledge distillation to handle the problems of expression diversity and class imbalance in the CLL-ID task. In addition to formulating the new task, we construct 3 benchmarks based on 8 intent detection datasets. Experimental results demonstrate the effectiveness of SCN, which significantly outperforms previous lifelong learning methods on the three benchmarks.

pdf abs
On Evaluating and Mitigating Gender Biases in Multilingual Settings
Aniket Vashishtha | Kabir Ahuja | Sunayana Sitaram

While understanding and removing gender biases in language models has been a long-standing problem in Natural Language Processing, prior research work has primarily been limited to English. In this work, we investigate some of the challenges with evaluating and mitigating biases in multilingual settings which stem from a lack of existing benchmarks and resources for bias evaluation beyond English especially for non-western context. In this paper, we first create a benchmark for evaluating gender biases in pre-trained masked language models by extending DisCo to different Indian languages using human annotations. We extend various debiasing methods to work beyond English and evaluate their effectiveness for SOTA massively multilingual models on our proposed metric. Overall, our work highlights the challenges that arise while studying social biases in multilingual settings and provides resources as well as mitigation techniques to take a step toward scaling to more languages.

pdf abs
Rethinking Round-Trip Translation for Machine Translation Evaluation
Terry Yue Zhuo | Qiongkai Xu | Xuanli He | Trevor Cohn

Automatic evaluation methods for translation often require model training, and thus the availability of parallel corpora limits their applicability to low-resource settings. Round-trip translation is a potential workaround, which can reframe bilingual evaluation into a much simpler monolingual task. Early results from the era of statistical machine translation (SMT) raised fundamental concerns about the utility of this approach, based on poor correlation with human translation quality judgments. In this paper, we revisit this technique with modern neural translation (NMT) and show that round-trip translation does allow for accurate automatic evaluation without the need for reference translations. These opposite findings can be explained through the copy mechanism in SMT that is absent in NMT. We demonstrate that round-trip translation benefits multiple machine translation evaluation tasks: i) predicting forward translation scores; ii) improving the performance of a quality estimation model; and iii) identifying adversarial competitors in shared tasks via cross-system verification.

We present a framework called G3R for complex and cross-domain Text-to-SQL generation. G3R aims to address two limitations of current approaches: (1) The structure of the abstract syntax tree (AST) is not fully explored during the decoding process which is crucial for complex SQL generation; (2) Domain knowledge is not incorporated to enhance their ability to generalise to unseen domains. G3R consists of a graph-guided SQL generator and a knowledge-enhanced re-ranking mechanism. Firstly, during the decoding process, An AST-Grammar bipartite graph is constructed for both the AST and corresponding grammar rules of the generated partial SQL query. The graph-guided SQL generator captures its structural information and fuses heterogeneous information to predict the action sequence which can construct the AST for the corresponding SQL query uniquely. Then, in the inference stage, a knowledge-enhanced re-ranking mechanism is proposed to introduce domain knowledge to re-rank candidate SQL queries from the beam output and choose the final answer. The SQL ranker is based on pre-trained language models (PLM) and contrastive learning with hybrid prompt tuning is incorporated to stimulate the knowledge of PLMs and make it more discriminative. The proposed approach achieves state-of-the-art results on the Spider and Spider-DK benchmarks, which are challenging complex and cross-domain benchmarks for Text-to-SQL semantic analysis.

pdf abs
A Unified Knowledge Graph Augmentation Service for Boosting Domain-specific NLP Tasks
Ruiqing Ding | Xiao Han | Leye Wang

By focusing the pre-training process on domain-specific corpora, some domain-specific pre-trained language models (PLMs) have achieved state-of-the-art results. However, it is under-investigated to design a unified paradigm to inject domain knowledge in the PLM fine-tuning stage. We propose KnowledgeDA, a unified domain language model development service to enhance the task-specific training procedure with domain knowledge graphs. Given domain-specific task texts input, KnowledgeDA can automatically generate a domain-specific language model following three steps: (i) localize domain knowledge entities in texts via an embedding-similarity approach; (ii) generate augmented samples by retrieving replaceable domain entity pairs from two views of both knowledge graph and training data; (iii) select high-quality augmented samples for fine-tuning via confidence-based assessment. We implement a prototype of KnowledgeDA to learn language models for two domains, healthcare and software development. Experiments on domain-specific text classification and QA tasks verify the effectiveness and generalizability of KnowledgeDA.

pdf abs
Dialogue Planning via Brownian Bridge Stochastic Process for Goal-directed Proactive Dialogue
Jian Wang | Dongding Lin | Wenjie Li

Goal-directed dialogue systems aim to proactively reach a pre-determined target through multi-turn conversations. The key to achieving this task lies in planning dialogue paths that smoothly and coherently direct conversations towards the target. However, this is a challenging and under-explored task. In this work, we propose a coherent dialogue planning approach that uses a stochastic process to model the temporal dynamics of dialogue paths. We define a latent space that captures the coherence of goal-directed behavior using a Brownian bridge process, which allows us to incorporate user feedback flexibly in dialogue planning. Based on the derived latent trajectories, we generate dialogue paths explicitly using pre-trained language models. We finally employ these paths as natural language prompts to guide dialogue generation. Our experiments show that our approach generates more coherent utterances and achieves the goal with a higher success rate.

pdf abs
A Match Made in Heaven: A Multi-task Framework for Hyperbole and Metaphor Detection
Naveen Badathala | Abisek Rajakumar Kalarani | Tejpalsingh Siledar | Pushpak Bhattacharyya

Hyperbole and metaphor are common in day-to-day communication (e.g., “I am in deep trouble”: how does trouble have depth?), which makes their detection important, especially in a conversational AI setting. Existing approaches to automatically detect metaphor and hyperbole have studied these language phenomena independently, but their relationship has hardly, if ever, been explored computationally. In this paper, we propose a multi-task deep learning framework to detect hyperbole and metaphor simultaneously. We hypothesize that metaphors help in hyperbole detection, and vice-versa. To test this hypothesis, we annotate two hyperbole datasets- HYPO and HYPO-L- with metaphor labels. Simultaneously, we annotate two metaphor datasets- TroFi and LCC- with hyperbole labels. Experiments using these datasets give an improvement of the state of the art of hyperbole detection by 12%. Additionally, our multi-task learning (MTL) approach shows an improvement of up to 17% over single-task learning (STL) for both hyperbole and metaphor detection, supporting our hypothesis. To the best of our knowledge, ours is the first demonstration of computational leveraging of linguistic intimacy between metaphor and hyperbole, leading to showing the superiority of MTL over STL for hyperbole and metaphor detection.

pdf abs
Prompt Tuning for Unified Multimodal Pretrained Models
Hao Yang | Junyang Lin | An Yang | Peng Wang | Chang Zhou

Prompt tuning has become a new paradigm for model tuning and it has demonstrated success in natural language pretraining and even vision pretraining. The parameter-efficient prompt tuning methods that optimize soft embeddings while keeping the pretrained model frozen demonstrate advantages in low computation costs and almost lossless performance. In this work, we explore the transfer of prompt tuning to multimodal pretrained models. Specifically, we implement prompt tuning to a unified sequence-to-sequence pretrained model by adding a sequence of learnable embeddings to each layer and finetuning the pretrained model on downstream task with only the learnable embeddings being optimized. Experimental results on a series of multimodal understanding and generation tasks demonstrate that our method OFA-PT can achieve comparable performance with finetuning across a series of multimodal generation and understanding tasks. Additionally, it significantly outperforms the unified multimodal pretrained model with other parameter-efficient tuning methods, e.g., Adapter, BitFit. etc. Besides, in comparison with finetuned models, the prompt-tuned models demonstrate improved robustness against adversarial attacks. We further figure out that experimental factors, including prompt length, prompt depth, and reparameteratization, have great impacts on the model performance, and thus we empirically provide a recommendation for the setups of prompt tuning.

Temporal knowledge graph completion that predicts missing links for incomplete temporal knowledge graphs (TKG) is gaining increasing attention. Most existing works have achieved good results by incorporating time information into static knowledge graph embedding methods. However, they ignore the contextual nature of the TKG structure, i.e., query-specific subgraph contains both structural and temporal neighboring facts. This paper presents the SToKE, a novel method that employs the pre-trained language model (PLM) to learn joint Structural and Temporal Contextualized Knowledge Embeddings.Specifically, we first construct an event evolution tree (EET) for each query to enable PLMs to handle the TKG, which can be seen as a structured event sequence recording query-relevant structural and temporal contexts. We then propose a novel temporal embedding and structural matrix to learn the time information and structural dependencies of facts in EET.Finally, we formulate TKG completion as a mask prediction problem by masking the missing entity of the query to fine-tune pre-trained language models. Experimental results on three widely used datasets show the superiority of our model.

The development of large language models (LLMs) such as ChatGPT has brought a lot of attention recently. However, their evaluation in the benchmark academic datasets remains under-explored due to the difficulty of evaluating the generative outputs produced by this model against the ground truth. In this paper, we aim to present a thorough evaluation of ChatGPT’s performance on diverse academic datasets, covering tasks like question-answering, text summarization, code generation, commonsense reasoning, mathematical problem-solving, machine translation, bias detection, and ethical considerations. Specifically, we evaluate ChatGPT across 140 tasks and analyze 255K responses it generates in these datasets. This makes our work the largest evaluation of ChatGPT in NLP benchmarks. In short, our study aims to validate the strengths and weaknesses of ChatGPT in various tasks and provide insights for future research using LLMs. We also report a new emergent ability to follow multi-query instructions that we mostly found in ChatGPT and other instruction-tuned models. Our extensive evaluation shows that even though ChatGPT is capable of performing a wide variety of tasks, and may obtain impressive performance in several benchmark datasets, it is still far from achieving the ability to reliably solve many challenging tasks. By providing a thorough assessment of ChatGPT’s performance across diverse NLP tasks, this paper sets the stage for a targeted deployment of ChatGPT-like LLMs in real-world applications.

This paper proposes a new task of commonsense question generation, which aims to yield deep-level and to-the-point questions from the text. Their answers need to reason over disjoint relevant contexts and external commonsense knowledge, such as encyclopedic facts and causality. The knowledge may not be explicitly mentioned in the text but is used by most humans for problem-shooting. Such complex reasoning with hidden contexts involves deep semantic understanding. Thus, this task has great application value, such as making high-quality quizzes in advanced exams. Due to the lack of modeling complexity, existing methods may produce shallow questions that can be answered by simple word matching. To address these challenges, we propose a new QG model by simultaneously considering asking contents, expressive ways, and answering complexity. We first retrieve text-related commonsense context. Then we disentangle the key factors that control questions in terms of reasoning content and verbalized way. Independence priors and constraints are imposed to facilitate disentanglement. We further develop a discriminator to promote the deep results by considering their answering complexity. Through adversarial inference, we learn the latent factors from data. By sampling the expressive factor from the data distributions, diverse questions can be yielded. Evaluations of two typical data sets show the effectiveness of our approach.

pdf abs
TADA: Efficient Task-Agnostic Domain Adaptation for Transformers
Chia-Chien Hung | Lukas Lange | Jannik Strötgen

Intermediate training of pre-trained transformer-based language models on domain-specific data leads to substantial gains for downstream tasks. To increase efficiency and prevent catastrophic forgetting alleviated from full domain-adaptive pre-training, approaches such as adapters have been developed. However, these require additional parameters for each layer, and are criticized for their limited expressiveness. In this work, we introduce TADA, a novel task-agnostic domain adaptation method which is modular, parameter-efficient, and thus, data-efficient. Within TADA, we retrain the embeddings to learn domain-aware input representations and tokenizers for the transformer encoder, while freezing all other parameters of the model. Then, task-specific fine-tuning is performed. We further conduct experiments with meta-embeddings and newly introduced meta-tokenizers, resulting in one model per task in multi-domain use cases. Our broad evaluation in 4 downstream tasks for 14 domains across single- and multi-domain setups and high- and low-resource scenarios reveals that TADA is an effective and efficient alternative to full domain-adaptive pre-training and adapters for domain adaptation, while not introducing additional parameters or complex training steps.

pdf abs
Robust Natural Language Understanding with Residual Attention Debiasing
Fei Wang | James Y. Huang | Tianyi Yan | Wenxuan Zhou | Muhao Chen

Natural language understanding (NLU) models often suffer from unintended dataset biases. Among bias mitigation methods, ensemble-based debiasing methods, especially product-of-experts (PoE), have stood out for their impressive empirical success. However, previous ensemble-based debiasing methods typically apply debiasing on top-level logits without directly addressing biased attention patterns. Attention serves as the main media of feature interaction and aggregation in PLMs and plays a crucial role in providing robust prediction. In this paper, we propose REsidual Attention Debiasing (READ), an end-to-end debiasing method that mitigates unintended biases from attention. Experiments on three NLU benchmarks show that READ significantly improves the OOD performance of BERT-based models, including +12.9% accuracy on HANS, +11.0% accuracy on FEVER-Symmetric, and +2.7% F1 on PAWS. Detailed analyses demonstrate the crucial role of unbiased attention in robust NLU models and that READ effectively mitigates biases in attention.

Dialogue state tracking (DST) aims to convert the dialogue history into dialogue states which consist of slot-value pairs. As condensed structural information memorizes all history information, the dialogue state in the previous turn is typically adopted as the input for predicting the current state by DST models. However, these models tend to keep the predicted slot values unchanged, which is defined as state momentum in this paper. Specifically, the models struggle to update slot values that need to be changed and correct wrongly predicted slot values in the previous turn. To this end, we propose MoNET to tackle state momentum via noise-enhanced training. First, the previous state of each turn in the training data is noised via replacing some of its slot values. Then, the noised previous state is used as the input to learn to predict the current state, improving the model’s ability to update and correct slot values. Furthermore, a contrastive contextmatching framework is designed to narrow the representation distance between a state and itscorresponding noised variant, which reduces the impact of noised state and makes the model better understand the dialogue history. Experimental results on MultiWOZ datasets show that MoNET outperforms previous DST methods. Ablations and analysis verify the effectiveness of MoNET in alleviating state momentum issues and improving the anti-noise ability.

pdf abs
PAL: Persona-Augmented Emotional Support Conversation Generation
Jiale Cheng | Sahand Sabour | Hao Sun | Zhuang Chen | Minlie Huang

Due to the lack of human resources for mental health support, there is an increasing demand for employing conversational agents for support. Recent work has demonstrated the effectiveness of dialogue models in providing emotional support. As previous studies have demonstrated that seekers’ persona is an important factor for effective support, we investigate whether there are benefits to modeling such information in dialogue models for support. In this paper, our empirical analysis verifies that persona has an important impact on emotional support. Therefore, we propose a framework for dynamically inferring and modeling seekers’ persona. We first train a model for inferring the seeker’s persona from the conversation history. Accordingly, we propose PAL, a model that leverages persona information and, in conjunction with our strategy-based controllable generation method, provides personalized emotional support. Automatic and manual evaluations demonstrate that PAL achieves state-of-the-art results, outperforming the baselines on the studied benchmark. Our code and data are publicly available at https://github.com/chengjl19/PAL.

Pretrained language models have achieved remarkable success in various natural language processing tasks. However, pretraining has recently shifted toward larger models and larger data, which has resulted in significant computational and energy costs. In this paper, we propose Influence Subset Selection (ISS) for language model, which explicitly utilizes end-task knowledge to select a tiny subset of the pretraining corpus. Specifically, the ISS selects the samples that will provide the most positive influence on the performance of the end task. Furthermore, we design a gradient matching-based influence estimation method, which can drastically reduce the computation time of influence. With only 0.45% of the data and a three-orders-of-magnitude lower computational cost, ISS outperformed pretrained models (e.g., RoBERTa) on eight datasets covering four domains.

pdf abs
Exclusive Supermask Subnetwork Training for Continual Learning
Prateek Yadav | Mohit Bansal

Continual Learning (CL) methods focus on accumulating knowledge over time while avoiding catastrophic forgetting. Recently, Wortsman et al. (2020) proposed a CL method, SupSup, which uses a randomly initialized, fixed base network (model) and finds a supermask for each new task that selectively keeps or removes each weight to produce a subnetwork. They prevent forgetting as the network weights are not being updated. Although there is no forgetting, the performance of SupSup is sub-optimal because fixed weights restrict its representational power. Furthermore, there is no accumulation or transfer of knowledge inside the model when new tasks are learned. Hence, we propose ExSSNeT (Exclusive Supermask SubNetwork Training), that performs exclusive and non-overlapping subnetwork weight training. This avoids conflicting updates to the shared weights by subsequent tasks to improve performance while still preventing forgetting. Furthermore, we propose a novel KNN-based Knowledge Transfer (KKT) module that utilizes previously acquired knowledge to learn new tasks better and faster. We demonstrate that ExSSNeT outperforms strong previous methods on both NLP and Vision domains while preventing forgetting. Moreover, ExSSNeT is particularly advantageous for sparse masks that activate 2-10% of the model parameters, resulting in an average improvement of 8.3% over SupSup. Furthermore, ExSSNeT scales to a large number of tasks (100).

This paper proposes a new method, OFA-OCR, to transfer multimodal pretrained models to text recognition. Specifically, we recast text recognition as image captioning and directly transfer a unified vision-language pretrained model to the end task. Without pretraining on large-scale annotated or synthetic text recognition data, OFA-OCR outperforms the baselines and achieves state-of-the-art performance in the Chinese text recognition benchmark. Additionally, we construct an OCR pipeline with OFA-OCR, and we demonstrate that it can achieve competitive performance with the product-level API.

Byte-Pair Encoding (BPE) is a popular algorithm used for tokenizing data in NLP, despite being devised initially as a compression method.BPE appears to be a greedy algorithm at face value, but the underlying optimization problem that BPE seeks to solve has not yet been laid down. We formalize BPE as a combinatorial optimization problem. Via submodular functions, we prove that the iterative greedy version is a 1/sigma*(1-e(-sigma))-approximation of an optimal merge sequence, where sigma is the total backward curvature with respect to the optimal merge sequence. Empirically the lower bound of the approximation is approx0.37.We provide a faster implementation of BPE which improves the runtime complexity from O(NM) to O(N log M), where N is the sequence length and M is the merge count. Finally, we optimize the brute-force algorithm for optimal BPE using memoization.

pdf abs
Automatic Named Entity Obfuscation in Speech
Judita Preiss

Sharing data containing personal information often requires its anonymization, even when consent for sharing was obtained from the data originator. While approaches exist for automated anonymization of text, the area is not as thoroughly explored in speech. This work focuses on identifying, replacing and inserting replacement named entities synthesized using voice cloning into original audio thereby retaining prosodic information while reducing the likelihood of deanonymization. The approach employs a novel named entity recognition (NER) system built directly on speech by training HuBERT (Hsu et al, 2021) using the English speech NER dataset (Yadav et al, 2020). Name substitutes are found using a masked language model and are synthesized using text to speech voice cloning (Eren and team, 2021), upon which the substitute named entities are re-inserted into the original text. The approach is prototyped on a sample of the LibriSpeech corpus (Panyatov et al, 2015) with each step evaluated individually.

pdf abs
Recursion of Thought: A Divide-and-Conquer Approach to Multi-Context Reasoning with Language Models
Soochan Lee | Gunhee Kim

Generating intermediate steps, or Chain of Thought (CoT), is an effective way to significantly improve language models’ (LM) multi-step reasoning capability. However, the CoT lengths can grow rapidly with the problem complexity, easily exceeding the maximum context size. Instead of increasing the context limit, which has already been heavily investigated, we explore an orthogonal direction: making LMs divide a problem into multiple contexts. We propose a new inference framework, called Recursion of Thought (RoT), which introduces several special tokens that the models can output to trigger context-related operations. Extensive experiments with multiple architectures including GPT-3 show that RoT dramatically improves LMs’ inference capability to solve problems, whose solution consists of hundreds of thousands of tokens.

Multimodal learning aims to imitate human beings to acquire complementary information from multiple modalities for various downstream tasks. However, traditional aggregation-based multimodal fusion methods ignore the inter-modality relationship, treat each modality equally, suffer sensor noise, and thus reduce multimodal learning performance. In this work, we propose a novel multimodal contrastive method to explore more reliable multimodal representations under the weak supervision of unimodal predicting. Specifically, we first capture task-related unimodal representations and the unimodal predictions from the introduced unimodal predicting task. Then the unimodal representations are aligned with the more effective one by the designed multimodal contrastive method under the supervision of the unimodal predictions. Experimental results with fused features on two image-text classification benchmarks UPMC-Food-101 and N24News show that our proposed Unimodality-Supervised MultiModal Contrastive UniS-MMC learning method outperforms current state-of-the-art multimodal methods. The detailed ablation study and analysis further demonstrate the advantage of our proposed method.

pdf abs
Robustness-Aware Word Embedding Improves Certified Robustness to Adversarial Word Substitutions
Yibin Wang | Yichen Yang | Di He | Kun He

Natural Language Processing (NLP) models have gained great success on clean texts, but they are known to be vulnerable to adversarial examples typically crafted by synonym substitutions. In this paper, we target to solve this problem and find that word embedding is important to the certified robustness of NLP models. Given the findings, we propose the Embedding Interval Bound Constraint (EIBC) triplet loss to train robustness-aware word embeddings for better certified robustness. We optimize the EIBC triplet loss to reduce distances between synonyms in the embedding space, which is theoretically proven to make the verification boundary tighter. Meanwhile, we enlarge distances among non-synonyms, maintaining the semantic representation of word embeddings. Our method is conceptually simple and componentized. It can be easily combined with IBP training and improves the certified robust accuracy from 76.73% to 84.78% on the IMDB dataset. Experiments demonstrate that our method outperforms various state-of-the-art certified defense baselines and generalizes well to unseen substitutions. The code is available at https://github.com/JHL-HUST/EIBC-IBP/.

In the context-dependent Text-to-SQL task, the generated SQL statements are refined iteratively based on the user input utterance from each interaction. The input text from each interaction can be viewed as component modifications to the previous SQL statements, which could be further extracted as the modification patterns. Since these modification patterns could also be combined with other SQL statements, the models are supposed to have the compositional generalization to these novel combinations. This work is the first exploration of compositional generalization in context-dependent Text-to-SQL scenarios. To facilitate related studies, we constructed two challenging benchmarks named CoSQL-CG and SParC-CG by recombining the modification patterns and existing SQL statements. The following experiments show that almost all current models struggle on our proposed benchmarks. Furthermore, we found that better aligning the previous SQL statements with the input utterance could give models better combinatorial generalization ability. Based on these observations, we propose a method name p-align to improve the combinatorial generalization of Text-to-SQL models. Further experiments validate the effectiveness of our model.

pdf abs
Towards Generative Event Factuality Prediction
John Murzaku | Tyler Osborne | Amittai Aviram | Owen Rambow

We present a novel end-to-end generative task and system for predicting event factuality holders, targets, and their associated factuality values. We perform the first experiments using all sources and targets of factuality statements from the FactBank corpus. We perform multi-task learning with other tasks and event-factuality corpora to improve on the FactBank source and target task. We argue that careful domain specific target text output format in generative systems is important and verify this with multiple experiments on target text output structure. We redo previous state-of-the-art author-only event factuality experiments and also offer insights towards a generative paradigm for the author-only event factuality prediction task.

pdf abs
Can Language Models Be Specific? How?
Jie Huang | Kevin Chen-Chuan Chang | Jinjun Xiong | Wen-mei Hwu

“He is a person”, “Paris is located on the earth”. Both statements are correct but meaningless - due to lack of specificity. In this paper, we propose to measure how specific the language of pre-trained language models (PLMs) is. To achieve this, we introduce a novel approach to build a benchmark for specificity testing by forming masked token prediction tasks with prompts. For instance, given “Toronto is located in [MASK].”, we want to test whether a more specific answer will be better filled in by PLMs, e.g., Ontario instead of Canada. From our evaluations, we show that existing PLMs have only a slight preference for more specific answers. We identify underlying factors affecting the specificity and design two prompt-based methods to improve the specificity. Results show that the specificity of the models can be improved by the proposed methods without additional training. We hope this work can bring to awareness the notion of specificity of language models and encourage the research community to further explore this important but understudied problem.

Pretrained language models (PLMs) encode a large amount of world knowledge. However, as such knowledge is frozen at the time of model training, the models become static and limited by the training data at that time. In order to further improve the capacity of PLMs for knowledge-intensive tasks, we consider augmenting PLMs with the large-scale web using search engine. Unlike previous augmentation sources (e.g., Wikipedia data dump), the web provides broader, more comprehensive and constantly updated information. In this paper, we present a web-augmented PLM – UniWeb, which is trained over 16 knowledge-intensive tasks in a unified text-to-text format. Instead of simply using the retrieved contents from web, our approach has made two major improvements. Firstly, we propose an adaptive search engine assisted learning method that can self-evaluate the confidence level of PLM’s predictions, and adaptively determine when to refer to the web for more data, which can avoid useless or noisy augmentation from web. Secondly, we design a pretraining task, i.e., continual knowledge learning, based on salient spans prediction, to reduce the discrepancy between the encoded and retrieved knowledge. Experiments on a wide range of knowledge-intensive tasks show that our model significantly outperforms previous retrieval-augmented methods.

pdf abs
Enhancing Few-shot Cross-lingual Transfer with Target Language Peculiar Examples
Hwichan Kim | Mamoru Komachi

Few-shot cross-lingual transfer, fine-tuning Multilingual Masked Language Model (MMLM) with source language labeled data and a small amount of target language labeled data, provides excellent performance in the target language. However, if no labeled data in the target language are available, they need to be created through human annotations. In this study, we devise a metric to select annotation candidates from an unlabeled data pool that efficiently enhance accuracy for few-shot cross-lingual transfer. It is known that training a model with hard examples is important to improve the model’s performance. Therefore, we first identify examples that MMLM cannot solve in a zero-shot cross-lingual transfer setting and demonstrate that it is hard to predict peculiar examples in the target language, i.e., the examples distant from the source language examples in cross-lingual semantic space of the MMLM.We then choose high peculiarity examples as annotation candidates and perform few-shot cross-lingual transfer. In comprehensive experiments with 20 languages and 6 tasks, we demonstrate that the high peculiarity examples improve the target language accuracy compared to other candidate selection methods proposed in previous studies.

Real-life multilingual systems should be able to efficiently incorporate new languages as data distributions fed to the system evolve and shift over time. To do this, systems need to handle the issue of catastrophic forgetting, where the model performance drops for languages or tasks seen further in its past. In this paper, we study catastrophic forgetting, as well as methods to minimize this, in a massively multilingual continual learning framework involving up to 51 languages and covering both classification and sequence labeling tasks. We present LR ADJUST, a learning rate scheduling method that is simple, yet effective in preserving new information without strongly overwriting past knowledge. Furthermore, we show that this method is effective across multiple continual learning approaches. Finally, we provide further insights into the dynamics of catastrophic forgetting in this massively multilingual setup.

Vision-language tasks, such as VQA, SNLI-VE, and VCR are challenging because they require the model’s reasoning ability to understand the semantics of the visual world and natural language. Supervised methods working for vision-language tasks have been well-studied. However, solving these tasks in a zero-shot setting is less explored. Since Contrastive Language-Image Pre-training (CLIP) has shown remarkable zero-shot performance on image-text matching, previous works utilized its strong zero-shot ability by converting vision-language tasks into an image-text matching problem, and they mainly consider global-level matching (e.g., the whole image or sentence). However, we find visual and textual fine-grained information, e.g., keywords in the sentence and objects in the image, can be fairly informative for semantics understanding. Inspired by this, we propose a unified framework to take advantage of the fine-grained information for zero-shot vision-language learning, covering multiple tasks such as VQA, SNLI-VE, and VCR. Our experiments show that our framework outperforms former zero-shot methods on VQA and achieves substantial improvement on SNLI-VE and VCR. Furthermore, our ablation studies confirm the effectiveness and generalizability of our proposed method.

pdf abs
Aligning Instruction Tasks Unlocks Large Language Models as Zero-Shot Relation Extractors
Kai Zhang | Bernal Jimenez Gutierrez | Yu Su

Recent work has shown that fine-tuning large language models (LLMs) on large-scale instruction-following datasets substantially improves their performance on a wide range of NLP tasks, especially in the zero-shot setting. However, even advanced instruction-tuned LLMs still fail to outperform small LMs on relation extraction (RE), a fundamental information extraction task. We hypothesize that instruction-tuning has been unable to elicit strong RE capabilities in LLMs due to RE’s low incidence in instruction-tuning datasets, making up less than 1% of all tasks (Wang et al. 2022). To address this limitation, we propose QA4RE, a framework that aligns RE with question answering (QA), a predominant task in instruction-tuning datasets. Comprehensive zero-shot RE experiments over four datasets with two series of instruction-tuned LLMs (six LLMs in total) demonstrate that our QA4RE framework consistently improves LLM performance, strongly verifying our hypothesis and enabling LLMs to outperform strong zero-shot baselines by a large margin. Additionally, we provide thorough experiments and discussions to show the robustness, few-shot effectiveness, and strong transferability of our QA4RE framework. This work illustrates a promising way of adapting LLMs to challenging and underrepresented tasks by aligning these tasks with more common instruction-tuning tasks like QA.

pdf abs
TADA : Task Agnostic Dialect Adapters for English
William Held | Caleb Ziems | Diyi Yang

Large Language Models, the dominant starting point for Natural Language Processing (NLP) applications, fail at a higher rate for speakers of English dialects other than Standard American English (SAE). Prior work addresses this using task specific data or synthetic data augmentation, both of which require intervention for each dialect and task pair. This poses a scalability issue that prevents the broad adoption of robust dialectal English NLP. We introduce a simple yet effective method for task-agnostic dialect adaptation by aligning non-SAE dialects using adapters and composing them with task-specific adapters from SAE. Task-Agnostic Dialect Adapters (TADA) improve dialectal robustness on 4 dialectal variants of the GLUE benchmark without task-specific supervision.

Zero-shot cross-domain slot filling aims to transfer knowledge from the labeled source domain to the unlabeled target domain. Existing models either encode slot descriptions and examples or design handcrafted question templates using heuristic rules, suffering from poor generalization capability or robustness. In this paper, we propose a generative zero-shot prompt learning framework for cross-domain slot filling, both improving generalization and robustness than previous work. Besides, we introduce a novel inverse prompting strategy to distinguish different slot types to avoid the multiple prediction problem, and an efficient prompt tuning strategy to boost higher performance only training fewer prompt parameters. Experiments and analysis demonstrate the effectiveness of our proposed framework, especially huge improvements (+13.44% F1) on the unseen slots.

pdf abs
Re-appraising the Schema Linking for Text-to-SQL
Yujian Gan | Xinyun Chen | Matthew Purver

Most text-to-SQL models, even though based on the same grammar decoder, generate the SQL structure first and then fill in the SQL slots with the correct schema items. This second step depends on schema linking: aligning the entity references in the question with the schema columns or tables. This is generally approached via Exact Match based Schema Linking (EMSL) within a neural network-based schema linking module. EMSL has become standard in text-to-SQL: many state-of-the-art models employ EMSL, with performance dropping significantly when the EMSL component is removed. In this work, however, we show that EMSL reduces robustness, rendering models vulnerable to synonym substitution and typos. Instead of relying on EMSL to make up for deficiencies in question-schema encoding, we show that using a pre-trained language model as an encoder can improve performance without using EMSL, giving a more robust model. We also study the design choice of the schema linking module, finding that a suitable design benefits performance and interoperability. Finally, based on the above study of schema linking, we introduce the grammar linking to help model align grammar references in the question with the SQL keywords.

pdf abs
Echoes from Alexandria: A Large Resource for Multilingual Book Summarization
Alessandro Scirè | Simone Conia | Simone Ciciliano | Roberto Navigli

In recent years, research in text summarization has mainly focused on the news domain, where texts are typically short and have strong layout features. The task of full-book summarization presents additional challenges which are hard to tackle with current resources, due to their limited size and availability in English only. To overcome these limitations, we present “Echoes from Alexandria”, or in shortened form, “Echoes”, a large resource for multilingual book summarization. Echoes featuresthree novel datasets: i) Echo-Wiki, for multilingual book summarization, ii) Echo-XSum, for extremely-compressive multilingual book summarization, and iii) Echo-FairySum, for extractive book summarization. To the best of our knowledge, Echoes – with its thousands of books and summaries – is the largest resource, and the first to be multilingual, featuring 5 languages and 25 language pairs. In addition to Echoes, we also introduce a new extractive-then-abstractive baseline, and, supported by our experimental results and manual analysis of the summaries generated, we argue that this baseline is more suitable for book summarization than purely-abstractive approaches. We release our resource and software at https://github.com/Babelscape/echoes-from-alexandria in the hope of fostering innovative research in multilingual booksummarization.

Large pre-trained language models (PLMs) have garnered significant attention for their versatility and potential for solving a wide spectrum of natural language processing (NLP) tasks. However, the cost of running these PLMs may be prohibitive. Furthermore, PLMs may not be open-sourced due to commercial considerations and potential risks of misuse, such as GPT-3. The parameters and gradients of PLMs are unavailable in this scenario. To solve the issue, black-box tuning has been proposed, which utilizes derivative-free optimization (DFO), instead of gradient descent, for training task-specific continuous prompts. However, these gradient-free methods still exhibit a significant gap compared to gradient-based methods. In this paper, we introduce gradient descent into black-box tuning scenario through knowledge distillation. Furthermore, we propose a novel method GDFO, which integrates gradient descent and derivative-free optimization to optimize task-specific continuous prompts in a harmonized manner. Experimental results show that GDFO can achieve significant performance gains over previous state-of-the-art methods.

Entailment graphs (EGs) with predicates as nodes and entailment relations as edges are typically incomplete, while EGs in different languages are often complementary to each other. In this paper, we propose a new task, multilingual entailment graph enhancement, which aims to utilize the entailment information from one EG to enhance another EG in a different language. The ultimate goal is to obtain an enhanced EG containing richer and more accurate entailment information. We present an align-then-enhance framework (ATE) to achieve accurate multilingual entailment graph enhancement, which first exploits a cross-graph guided interaction mechanism to automatically discover potential equivalent predicates between different EGs and then constructs more accurate enhanced entailment graphs based on soft predicate alignments. Extensive experiments show that ATE achieves better and more robust predicate alignment results between different EGs, and the enhanced entailment graphs generated by ATE outperform the original graphs for entailment detection.

Metric-based meta-learning is one of the de facto standards in few-shot learning. It composes of representation learning and metrics calculation designs. Previous works construct class representations in different ways, varying from mean output embedding to covariance and distributions. However, using embeddings in space lacks expressivity and cannot capture class information robustly, while statistical complex modeling poses difficulty to metric designs. In this work, we use tensor fields (“areas”) to model classes from the geometrical perspective for few-shot learning. We present a simple and effective method, dubbed as hypersphere prototypes (HyperProto), where class information is represented by hyperspheres with dynamic sizes with two sets of learnable parameters: the hypersphere’s center and the radius. Extending from points to areas, hyperspheres are much more expressive than embeddings. Moreover, it is more convenient to perform metric-based classification with hypersphere prototypes than statistical modeling, as we only need to calculate the distance from a data point to the surface of the hypersphere. Following this idea, we also develop two variants of prototypes under other measurements. Extensive experiments and analysis on few-shot NLP tasks and comparison with 20+ competitive baselines demonstrate the effectiveness of our approach.

pdf abs
Structured Mean-Field Variational Inference for Higher-Order Span-Based Semantic Role Labeling
Wei Liu | Songlin Yang | Kewei Tu

In this work, we enhance higher-order graph-based approaches for span-based semantic role labeling (SRL) by means of structured modeling. To decrease the complexity of higher-order modeling, we decompose the edge from predicate word to argument span into three different edges, predicate-to-head (P2H), predicate-to-tail (P2T), and head-to-tail (H2T), where head/tail means the first/last word of the semantic argument span. As such, we use a CRF-based higher-order dependency parser and leverage Mean-Field Variational Inference (MFVI) for higher-order inference. Moreover, since semantic arguments of predicates are often constituents within a constituency parse tree, we can leverage such nice structural property by defining a TreeCRF distribution over all H2T edges, using the idea of partial marginalization to define structural training loss. We further leverage structured MFVI to enhance inference. We experiment on span-based SRL benchmarks, showing the effectiveness of both higher-order and structured modeling and the combination thereof. In addition, we show superior performance of structured MFVI against vanilla MFVI.

Argument mining involves multiple sub-tasks that automatically identify argumentative elements, such as claim detection, evidence extraction, stance classification, etc. However, each subtask alone is insufficient for a thorough understanding of the argumentative structure and reasoning process. To learn a complete view of an argument essay and capture the interdependence among argumentative components, we need to know what opinions people hold (i.e., claims), why those opinions are valid (i.e., supporting evidence), which source the evidence comes from (i.e., evidence type), and how those claims react to the debating topic (i.e., stance). In this work, we for the first time propose a challenging argument quadruplet extraction task (AQE), which can provide an all-in-one extraction of four argumentative components, i.e., claims, evidence, evidence types, and stances. To support this task, we construct a large-scale and challenging dataset. However, there is no existing method that can solve the argument quadruplet extraction. To fill this gap, we propose a novel quad-tagging augmented generative approach, which leverages a quadruplet tagging module to augment the training of the generative framework. The experimental results on our dataset demonstrate the empirical superiority of our proposed approach over several strong baselines.

Large language models are known to produce output which sounds fluent and convincing, but is also often wrong, e.g. “unfaithful” with respect to a rationale as retrieved from a knowledge base. In this paper, we show that task-based systems which exhibit certain advanced linguistic dialog behaviors, such as lexical alignment (repeating what the user said), are in fact preferred and trusted more, whereas other phenomena, such as pronouns and ellipsis are dis-preferred. We use open-domain question answering systems as our test-bed for task based dialog generation and compare several open- and closed-book models. Our results highlight the danger of systems that appear to be trustworthy by parroting user input while providing an unfaithful response.

pdf abs
Discrete Prompt Optimization via Constrained Generation for Zero-shot Re-ranker
Sukmin Cho | Soyeong Jeong | Jeong yeon Seo | Jong Park

Re-rankers, which order retrieved documents with respect to the relevance score on the given query, have gained attention for the information retrieval (IR) task. Rather than fine-tuning the pre-trained language model (PLM), the large-scale language model (LLM) is utilized as a zero-shot re-ranker with excellent results. While LLM is highly dependent on the prompts, the impact and the optimization of the prompts for the zero-shot re-ranker are not explored yet. Along with highlighting the impact of optimization on the zero-shot re-ranker, we propose a novel discrete prompt optimization method, Constrained Prompt generation (Co-Prompt), with the metric estimating the optimum for re-ranking. Co-Prompt guides the generated texts from PLM toward optimal prompts based on the metric without parameter update. The experimental results demonstrate that Co-Prompt leads to outstanding re-ranking performance against the baselines. Also, Co-Prompt generates more interpretable prompts for humans against other prompt optimization methods.

pdf abs
Triggering Multi-Hop Reasoning for Question Answering in Language Models using Soft Prompts and Random Walks
Kanishka Misra | Cicero Nogueira dos Santos | Siamak Shakeri

Despite readily memorizing world knowledge about entities, pre-trained language models (LMs) struggle to compose together two or more facts to perform multi-hop reasoning in question-answering tasks. In this work, we propose techniques that improve upon this limitation by relying on random-walks over structured knowledge graphs. Specifically, we use soft-prompts to guide LMs to chain together their encoded knowledge by learning to map multi-hop questions to random-walk paths that lead to the answer. Applying our methods on two T5 LMs shows substantial improvements over standard tuning approaches in answering questions that require multi-hop reasoning.

Goal-oriented generative script learning aims to generate subsequent steps to reach a particular goal, which is an essential task to assist robots or humans in performing stereotypical activities. An important aspect of this process is the ability to capture historical states visually, which provides detailed information that is not covered by text and will guide subsequent steps. Therefore, we propose a new task, Multimedia Generative Script Learning, to generate subsequent steps by tracking historical states in both text and vision modalities, as well as presenting the first benchmark containing 5,652 tasks and 79,089 multimedia steps. This task is challenging in three aspects: the multimedia challenge of capturing the visual states in images, the induction challenge of performing unseen tasks, and the diversity challenge of covering different information in individual steps. We propose to encode visual state changes through a selective multimedia encoder to address the multimedia challenge, transfer knowledge from previously observed tasks using a retrieval-augmented decoder to overcome the induction challenge, and further present distinct information at each step by optimizing a diversity-oriented contrastive learning objective. We define metrics to evaluate both generation and inductive quality. Experiment results demonstrate that our approach significantly outperforms strong baselines.

Conventional approaches to text classification typically assume the existence of a fixed set of predefined labels to which a given text can be classified. However, in real-world applications, there exists an infinite label space for describing a given text. In addition, depending on the aspect (sentiment, topic, etc.) and domain of the text (finance, legal, etc.), the interpretation of the label can vary greatly. This makes the task of text classification, particularly in the zero-shot scenario, extremely challenging. In this paper, we investigate the task of zero-shot text classification with the aim of improving the ability of pre-trained language models (PLMs) to generalize to both seen and unseen data across varying aspects and domains. To solve this we introduce two new simple yet effective pre-training strategies, Implicit and Explicit pre-training. These methods inject aspect-level understanding into the model at train time with the goal of conditioning the model to build task-level understanding. To evaluate this, we construct and release UTCD, a new benchmark dataset for evaluating text classification in zero-shot settings. Experimental results on UTCD show that our approach achieves improved zero-shot generalization on a suite of challenging datasets across an array of zero-shot formalizations.

pdf abs
Click: Controllable Text Generation with Sequence Likelihood Contrastive Learning
Chujie Zheng | Pei Ke | Zheng Zhang | Minlie Huang

It has always been an important yet challenging problem to control language models to avoid generating texts with undesirable attributes, such as toxic language and unnatural repetition. We introduce Leo for controllable text generation, which needs no modification to the model architecture and facilitates out-of-the-box use of trained models. It employs a contrastive loss on sequence likelihood, which fundamentally decreases the generation probability of negative samples (i.e., generations with undesirable attributes). It also adopts a novel likelihood ranking-based strategy to construct contrastive samples from model generations. On the tasks of language detoxification, sentiment steering, and repetition reduction, we show that Leo outperforms strong baselines of controllable text generation and demonstrate the superiority of Leo’s sample construction strategy.

pdf abs
Improving Embedding-based Unsupervised Keyphrase Extraction by Incorporating Structural Information
Mingyang Song | Huafeng Liu | Yi Feng | Liping Jing

Keyphrase extraction aims to extract a set of phrases with the central idea of the source document. In a structured document, there are certain locations (e.g., the title or the first sentence) where a keyphrase is most likely to appear. However, when extracting keyphrases from the document, most existing embedding-based unsupervised keyphrase extraction models ignore the indicative role of the highlights in certain locations, leading to wrong keyphrases extraction. In this paper, we propose a new Highlight-Guided Unsupervised Keyphrase Extraction model (HGUKE) to address the above issue. Specifically, HGUKE first models the phrase-document relevance via the highlights of the documents. Next, HGUKE calculates the cross-phrase relevance between all candidate phrases. Finally, HGUKE aggregates the above two relevance as the importance score of each candidate phrase to rank and extract keyphrases. The experimental results on three benchmarks demonstrate that HGUKE outperforms the state-of-the-art unsupervised keyphrase extraction baselines.

pdf abs
Towards Reasoning in Large Language Models: A Survey
Jie Huang | Kevin Chen-Chuan Chang

Reasoning is a fundamental aspect of human intelligence that plays a crucial role in activities such as problem solving, decision making, and critical thinking. In recent years, large language models (LLMs) have made significant progress in natural language processing, and there is observation that these models may exhibit reasoning abilities when they are sufficiently large. However, it is not yet clear to what extent LLMs are capable of reasoning. This paper provides a comprehensive overview of the current state of knowledge on reasoning in LLMs, including techniques for improving and eliciting reasoning in these models, methods and benchmarks for evaluating reasoning abilities, findings and implications of previous research in this field, and suggestions on future directions. Our aim is to provide a detailed and up-to-date review of this topic and stimulate meaningful discussion and future work.

pdf abs
Transitioning from benchmarks to a real-world case of information-seeking in Scientific Publications
Chyrine Tahri | Aurore Bochnakian | Patrick Haouat | Xavier Tannier

Although recent years have been marked by incredible advances in the whole development process of NLP systems, there are still blind spots in characterizing what is still hampering real-world adoption of models in knowledge-intensive settings. In this paper, we illustrate through a real-world zero-shot text search case for information seeking in scientific papers, the masked phenomena that the current process of measuring performance might not reflect, even when benchmarks are, in appearance, faithfully representative of the task at hand. In addition to experimenting with TREC-COVID and NFCorpus, we provide an industrial, expert-carried/annotated, case of studying vitamin B’s impact on health. We thus discuss the misalignment between solely focusing on single-metric performance as a criterion for model choice and relevancy as a subjective measure for meeting a user’s need.

pdf abs
CLIPText: A New Paradigm for Zero-shot Text Classification
Libo Qin | Weiyun Wang | Qiguang Chen | Wanxiang Che

While CLIP models are useful for zero-shot vision-and-language (VL) tasks or computer vision tasks, little attention has been paid to the application of CLIP for language tasks. Intuitively, CLIP model have a rich representation pre-trained with natural language supervision, in which we argue that it is useful for language tasks. Hence, this work bridge this gap by investigating a CLIP model for zero-shot text classification. Specifically, we introduce CLIPText, a novel paradigm for zero-shot text classification, which reformulates zero-shot text classification into a text-image matching problem that CLIP can be applied to. In addition, we further incorporate prompt into CLIPText (Prompt-CLIPText) to better derive knowledge from CLIP. Experimental results on seven publicly available zero-shot text classification datasets show that both CLIPText and Prompt-CLIPText attain promising performance. Besides, extensive analysis further verifies that knowledge from CLIP can benefit zero-shot text classification task. We hope this work can attract more breakthroughs on applying VL pre-trained models for language tasks.

pdf abs
Rethinking Dictionaries and Glyphs for Chinese Language Pre-training
Yuxuan Wang | Jack Wang | Dongyan Zhao | Zilong Zheng

We introduce CDBert, a new learning paradigm that enhances the semantics understanding ability of the Chinese PLMs with dictionary knowledge and structure of Chinese characters. We name the two core modules of CDBert as Shuowen and Jiezi, where Shuowen refers to the process of retrieving the most appropriate meaning from Chinese dictionaries and Jiezi refers to the process of enhancing characters’ glyph representations with structure understanding. To facilitate dictionary understanding, we propose three pre-training tasks, i.e.„ Masked Entry Modeling, Contrastive Learning for Synonym and Antonym, and Example Learning. We evaluate our method on both modern Chinese understanding benchmark CLUE and ancient Chinese benchmark CCLUE. Moreover, we propose a new polysemy discrimination task PolyMRC based on the collected dictionary of ancient Chinese. Our paradigm demonstrates consistent improvements on previous Chinese PLMs across all tasks. Moreover, our approach yields significant boosting on few-shot setting of ancient Chinese understanding.

We introduce INSTRUCTOR, a new method for computing text embeddings given task instructions: every text input is embedded together with instructions explaining the use case (e.g., task and domain descriptions). Unlike encoders from prior work that are more specialized, INSTRUCTOR is a single embedder that can generate text embeddings tailored to different downstream tasks and domains, without any further training. We first annotate instructions for 330 diverse tasks and train INSTRUCTOR on this multitask mixture with a contrastive loss. We evaluate INSTRUCTOR on 70 embedding evaluation tasks (66 of which are unseen during training), ranging from classification and information retrieval to semantic textual similarity and text generation evaluation. INSTRUCTOR, while having an order of magnitude fewer parameters than the previous best model, achieves state-of-the-art performance, with an average improvement of 3.4% compared to the previous best results on the 70 diverse datasets. Our analysis suggests that INSTRUCTOR is robust to changes in instructions, and that instruction finetuning mitigates the challenge of training a single model on diverse datasets. Our model, code, and data are available at https://instructor-embedding.github.io.

pdf abs
Towards Speech Dialogue Translation Mediating Speakers of Different Languages
Shuichiro Shimizu | Chenhui Chu | Sheng Li | Sadao Kurohashi

We present a new task, speech dialogue translation mediating speakers of different languages. We construct the SpeechBSD dataset for the task and conduct baseline experiments. Furthermore, we consider context to be an important aspect that needs to be addressed in this task and propose two ways of utilizing context, namely monolingual context and bilingual context. We conduct cascaded speech translation experiments using Whisper and mBART, and show that bilingual context performs better in our settings.

pdf abs
Adaptation Approaches for Nearest Neighbor Language Models
Rishabh Bhardwaj | George Polovets | Monica Sunkara

Semi-parametric Nearest Neighbor Language Models (kNN-LMs) have produced impressive gains over purely parametric LMs, by leveraging large-scale neighborhood retrieval over external memory datastores. However, there has been little investigation into adapting such models for new domains. This work attempts to fill that gap and suggests the following approaches for adapting kNN-LMs — 1) adapting the underlying LM (using Adapters), 2) expanding neighborhood retrieval over an additional adaptation datastore, and 3) adapting the weights (scores) of retrieved neighbors using a learned Rescorer module. We study each adaptation strategy separately, as well as the combined performance improvement through ablation experiments and an extensive set of evaluations run over seven adaptation domains. Our combined adaptation approach consistently outperforms purely parametric adaptation and zero-shot (kNN-LM) baselines that construct datastores from the adaptation data. On average, we see perplexity improvements of 17.1% and 16% for these respective baselines, across domains.

pdf abs
Language Models for German Text Simplification: Overcoming Parallel Data Scarcity through Style-specific Pre-training
Miriam Anschütz | Joshua Oehms | Thomas Wimmer | Bartłomiej Jezierski | Georg Groh

Automatic text simplification systems help to reduce textual information barriers on the internet. However, for languages other than English, only few parallel data to train these systems exists. We propose a two-step approach to overcome this data scarcity issue. First, we fine-tuned language models on a corpus of German Easy Language, a specific style of German. Then, we used these models as decoders in a sequence-to-sequence simplification task. We show that the language models adapt to the style characteristics of Easy Language and output more accessible texts. Moreover, with the style-specific pre-training, we reduced the number of trainable parameters in text simplification models. Hence, less parallel data is sufficient for training. Our results indicate that pre-training on unaligned data can reduce the required parallel data while improving the performance on downstream tasks.

pdf abs
Client-Customized Adaptation for Parameter-Efficient Federated Learning
Yeachan Kim | Junho Kim | Wing-Lam Mok | Jun-Hyung Park | SangKeun Lee

Despite the versatility of pre-trained language models (PLMs) across domains, their large memory footprints pose significant challenges in federated learning (FL), where the training model has to be distributed between a server and clients. One potential solution to bypass such constraints might be the use of parameter-efficient fine-tuning (PEFT) in the context of FL. However, we have observed that typical PEFT tends to severely suffer from heterogeneity among clients in FL scenarios, resulting in unstable and slow convergence. In this paper, we propose Client-Customized Adaptation (C2A), a novel hypernetwork-based FL framework that generates client-specific adapters by conditioning the client information. With the effectiveness of the hypernetworks in generating customized weights through learning to adopt the different characteristics of inputs, C2A can maximize the utility of shared model parameters while minimizing the divergence caused by client heterogeneity. To verify the efficacy of C2A, we perform extensive evaluations on FL scenarios involving heterogeneity in label and language distributions. Comprehensive evaluation results clearly support the superiority of C2A in terms of both efficiency and effectiveness in FL scenarios.

Understanding users’ intentions in e-commerce platforms requires commonsense knowledge. In this paper, we present FolkScope, an intention knowledge graph construction framework, to reveal the structure of humans’ minds about purchasing items. As commonsense knowledge is usually ineffable and not expressed explicitly, it is challenging to perform information extraction. Thus, we propose a new approach that leverages the generation power of large language models (LLMs) and human-in-the-loop annotation to semi-automatically construct the knowledge graph. LLMs first generate intention assertions via e-commerce specific prompts to explain shopping behaviors, where the intention can be an open reason or a predicate falling into one of 18 categories aligning with ConceptNet, e.g., IsA, MadeOf, UsedFor, etc. Then we annotate plausibility and typicality labels of sampled intentions as training data in order to populate human judgments to all automatic generations. Last, to structurize the assertions, we propose pattern mining and conceptualization to form more condensed and abstract knowledge. Extensive evaluations and study demonstrate that our constructed knowledge graph can well model e-commerce knowledge and have many potential applications.

pdf abs
I am PsyAM: Modeling Happiness with Cognitive Appraisal Dimensions
Xuan Liu | Kokil Jaidka

This paper proposes and evaluates PsyAM (https://anonymous.4open.science/r/BERT-PsyAM-10B9), a framework that incorporates adaptor modules in a sequential multi-task learning setup to generate high-dimensional feature representations of hedonic well-being (momentary happiness) in terms of its psychological underpinnings. PsyAM models emotion in text through its cognitive antecedents through auxiliary models that achieve multi-task learning through novel feature fusion methods. We show that BERT-PsyAM has cross-task validity and cross-domain generalizability through experiments with emotion-related tasks – on new emotion tasks and new datasets, as well as against traditional methods and BERT baselines. We further probe the robustness of BERT-PsyAM through feature ablation studies, as well as discuss the qualitative inferences we can draw regarding the effectiveness of the framework for representing emotional states. We close with a discussion of a future agenda of psychology-inspired neural network architectures.

pdf abs
Value type: the bridge to a better DST model
Gao Qixiang | Mingyang Sun | Yutao Mou | Chen Zeng | Weiran Xu

Value type of the slots can provide lots of useful information for DST tasks. However, it has been ignored in most previous works. In this paper, we propose a new framework for DST task based on these value types. Firstly, we extract the type of token from each turn. Specifically, we divide the slots in the dataset into 9 categories according to the type of slot value, and then train a Ner model to extract the corresponding type-entity from each turn of conversation according to the token. Secondly, we improve the attention mode which is integrated into value type information between the slot and the conversation history to help each slot pay more attention to the turns that contain the same value type. Meanwhile, we introduce a sampling strategy to integrate these types into the attention formula, which decrease the error of Ner model. Finally, we conduct a comprehensive experiment on two multi-domain task-oriented conversation datasets, MultiWOZ 2.1 and MultiWOZ 2.4. The ablation experimental results show that our method is effective on both datasets, which verify the necessity of considering the type of slot value.

Machine Reading Comprehension (MRC) models easily learn spurious correlations from complex contexts such as tabular data. Counterfactual training—using the factual and counterfactual data by augmentation—has become a promising solution. However, it is costly to construct faithful counterfactual examples because it is tricky to maintain the consistency and dependency of the tabular data. In this paper, we take a more efficient fashion to ask hypothetical questions like “in which year would the net profit be larger if the revenue in 2019 were $38,298?”, whose effects on the answers are equivalent to those expensive counterfactual tables. We propose a hypothetical training framework that uses paired examples with different hypothetical questions to supervise the direction of model gradient towards the counterfactual answer change. The superior generalization results on tabular MRC datasets, including a newly constructed stress test and MultiHiertt, validate our effectiveness.

pdf abs
BanglaBook: A Large-scale Bangla Dataset for Sentiment Analysis from Book Reviews
Mohsinul Kabir | Obayed Bin Mahfuz | Syed Rifat Raiyan | Hasan Mahmud | Md Kamrul Hasan

The analysis of consumer sentiment, as expressed through reviews, can provide a wealth of insight regarding the quality of a product. While the study of sentiment analysis has been widely explored in many popular languages, relatively less attention has been given to the Bangla language, mostly due to a lack of relevant data and cross-domain adaptability. To address this limitation, we present BanglaBook, a large-scale dataset of Bangla book reviews consisting of 158,065 samples classified into three broad categories: positive, negative, and neutral. We provide a detailed statistical analysis of the dataset and employ a range of machine learning models to establish baselines including SVM, LSTM, and Bangla-BERT. Our findings demonstrate a substantial performance advantage of pre-trained models over models that rely on manually crafted features, emphasizing the necessity for additional training resources in this domain. Additionally, we conduct an in-depth error analysis by examining sentiment unigrams, which may provide insight into common classification errors in under-resourced languages like Bangla. Our codes and data are publicly available at https://github.com/mohsinulkabir14/BanglaBook.

pdf abs
Risks and NLP Design: A Case Study on Procedural Document QA
Nikita Haduong | Alice Gao | Noah A. Smith

As NLP systems are increasingly deployed at scale, concerns about their potential negative impacts have attracted the attention of the research community, yet discussions of risk have mostly been at an abstract level and focused on generic AI or NLP applications. We argue that clearer assessments of risks and harms to users—and concrete strategies to mitigate them—will be possible when we specialize the analysis to more concrete applications and their plausible users. As an illustration, this paper is grounded in cooking recipe procedural document question answering (ProcDocQA), where there are well-defined risks to users such as injuries or allergic reactions. Our case study shows that an existing language model, applied in “zero-shot” mode, quantitatively answers real-world questions about recipes as well or better than the humans who have answered the questions on the web. Using a novel questionnaire informed by theoretical work on AI risk, we conduct a risk-oriented error analysis that could then inform the design of a future system to be deployed with lower risk of harm and better performance.

Transformer-based masked language models such as BERT, trained on general corpora, have shown impressive performance on downstream tasks. It has also been demonstrated that the downstream task performance of such models can be improved by pretraining larger models for longer on more data. In this work, we empirically evaluate the extent to which these results extend to tasks in science. We use 14 domain-specific transformer-based models (including ScholarBERT, a new 770Mparameter science-focused masked language model pretrained on up to 225B tokens) to evaluate the impact of training data, model size, pretraining and finetuning time on 12 downstream scientific tasks. Interestingly, we find that increasing model size, training data, or compute time does not always lead to significant improvements (i.e., >1% F1), if any, in scientific information extraction tasks. We offer possible explanations for this surprising result.

A promising approach to estimate the causal effects of peer review policies is to analyze data from publication venues that shift policies from single-blind to double-blind from one year to the next. However, in these settings the content of the manuscript is a confounding variable—each year has a different distribution of scientific content which may naturally affect the distribution of reviewer scores. To address this textual confounding, we extend variable ratio nearest neighbor matching to incorporate text embeddings. We compare this matching method to a widely-used causal method of stratified propensity score matching and a baseline of randomly selected matches. For our case study of the ICLR conference shifting from single- to double-blind review from 2017 to 2018, we find human judges prefer manuscript matches from our method in 70% of cases. While the unadjusted estimate of the average causal effect of reviewers’ scores is -0.25, our method shifts the estimate to -0.17, a slightly smaller difference between the outcomes of single- and double-blind policies. We hope this case study enables exploration of additional text-based causal estimation methods and domains in the future.

pdf abs
Learning to Generalize for Cross-domain QA
Yingjie Niu | Linyi Yang | Ruihai Dong | Yue Zhang

There have been growing concerns regarding the out-of-domain generalization ability of natural language processing (NLP) models, particularly in question-answering (QA) tasks. Current synthesized data augmentation methods for QA are hampered by increased training costs. To address this issue, we propose a novel approach that combines prompting methods and linear probing with fine-tuning strategy, which does not entail additional cost. Our method has been theoretically and empirically shown to be effective in enhancing the generalization ability of both generative and discriminative models. Our approach outperforms state-of-the-art baselines, with an average increase in F1 score of 4.5%-7.9%. Furthermore, our method can be easily integrated into any pre-trained models and offers a promising solution to the under-explored cross-domain QA task.

Building cross-model intelligence that can understand charts and communicate the salient information hidden behind them is an appealing challenge in the vision and language (V+L) community. The capability to uncover the underlined table data of chart figures is a critical key to automatic chart understanding. We introduce ChartT5, a V+L model that learns how to interpret table information from chart images via cross-modal pre-training on plot table pairs. Specifically, we propose two novel pre-training objectives: Masked Header Prediction (MHP) and Masked Value Prediction (MVP) to facilitate the model with different skills to interpret the table information. We have conducted extensive experiments on chart question answering and chart summarization to verify the effectiveness of the proposed pre-training strategies. In particular, on the ChartQA benchmark, our ChartT5 outperforms the state-of-the-art non-pretraining methods by over 8% performance gains.

There has been increasing interest in synthesizing data to improve downstream text-to-SQL tasks. In this paper, we examined the existing synthesized datasets and discovered that state-of-the-art text-to-SQL algorithms did not further improve on popular benchmarks when trained with augmented synthetic data. We observed three shortcomings: illogical synthetic SQL queries from independent column sampling, arbitrary table joins, and language gaps between the synthesized SQL and natural language question (NLQ) pair. To address these issues, we propose a novel synthesis framework that imposes strong typing constraints, incorporates key relationships from schema, and conducts schema-distance-weighted column sampling. We also adopt an intermediate representation (IR) for the SQL-to-text task to further improve the quality of the generated NLQ. When existing powerful text-to-SQL parsers are pretrained on our high-quality synthesized data, these models have significant accuracy boosts and achieve new state-of-the-art performance on Spider. We also demonstrate the effectiveness of our techniques with ablation studies

Exploring the generalizability of a text-to-SQL parser is essential for a system to automatically adapt the real-world databases. Previous investigation works mostly focus on lexical diversity, including the influence of the synonym and perturbations in both natural language questions and databases. However, the structural variability of database schema (DS), as a widely seen real-world scenario, is yet underexplored. Specifically, confronted with the same input question, the target SQL may be represented in different ways when the DS comes to a different structure. In this work, we provide in-depth discussions about the schema generalizability challenge of text-to-SQL tasks. We observe that current datasets are too templated to study schema generalization. To collect suitable test data, we propose a framework to generate novel text-to-SQL data via automatic and synchronous (DS, SQL) pair altering. When evaluating state-of-the-art text-to-SQL models on the synthetic samples, performance is significantly degraded, which demonstrates the limitation of current research regarding schema generalization.

Cross-lingual natural language inference is a fundamental problem in cross-lingual language understanding. Many recent works have used prompt learning to address the lack of annotated parallel corpora in XNLI.However, these methods adopt discrete prompting by simply translating the templates to the target language and need external expert knowledge to design the templates. Besides, discrete prompts of human-designed template words are not trainable vectors and can not be migrated to target languages in the inference stage flexibly. In this paper, we propose a novel Soft prompt learning framework with the Multilingual Verbalizer (SoftMV) for XNLI. SoftMV first constructs cloze-style question with soft prompts for the input sample. Then we leverage bilingual dictionaries to generate an augmented multilingual question for the original question. SoftMV adopts a multilingual verbalizer to align the representations of original and augmented multilingual questions into a unified semantic space with consistency regularization. Experimental results on XNLI demonstrate that SoftMV can achieve state-of-the-art performance and significantly outperform the previous methods under the few-shot and full-shot cross-lingual transfer settings.

Existing models for named entity recognition (NER) are mainly based on large-scale labeled datasets, which always obtain using crowdsourcing. However, it is hard to obtain a unified and correct label via majority voting from multiple annotators for NER due to the large labeling space and complexity of this task. To address this problem, we aim to utilize the original multi-annotator labels directly. Particularly, we propose a CONfidence-based partial Label Learning (CONLL) method to integrate the prior confidence (given by annotators) and posterior confidences (learned by models) for crowd-annotated NER. This model learns a token- and content-dependent confidence via an Expectation–Maximization (EM) algorithm by minimizing empirical risk. The true posterior estimator and confidence estimator perform iteratively to update the true posterior and confidence respectively. We conduct extensive experimental results on both real-world and synthetic datasets, which show that our model can improve performance effectively compared with strong baselines.

Much work has been done to improve persona consistency by finetuning a pretrained dialogue model on high-quality human-annoated persona datasets. However, these methods still face the challenges of high cost and poor scalability. To this end, we propose a simple-yet-effective approach to significantly improve zero-shot persona consistency via in-context learning. Specifically, we first pre-train a persona-augmented dialogue generation model and then utilize in-context prompting mechanism to realize zero-shot persona customization. Experimental results demonstrate that our method can dramatically improve persona consistency without compromising coherence and informativeness in zero-shot settings.

pdf abs
Grammar-based Decoding for Improved Compositional Generalization in Semantic Parsing
Jing Zheng | Jyh-Herng Chow | Zhongnan Shen | Peng Xu

Sequence-to-sequence (seq2seq) models have achieved great success in semantic parsing tasks, but they tend to struggle on out-of-distribution (OOD) data. Despite recent progress, robust semantic parsing on large-scale tasks with combined challenges from both compositional generalization and natural language variations remains an unsolved problem. To promote research in this area, this work presents CUDON, a large-scale dialogue dataset in Chinese language, particularly designed for evaluating compositional generalization of semantic parsing. The dataset contains about ten thousand multi-turn complex queries, and provides multiple splits with different degrees of train-test distribution divergence. We have investigated improving compositional generalization with grammar-based decodering on this dataset. With specially designed grammars leveraging program schema, we are able to substantially improve accuracy of seq2seq semantic parsers on OOD splits: A LSTM-based parser using a Context-free Grammar (CFG) achieves over 25% higher accuracy than a standard seq2seq baseline; a parser using Tree-Substitution Grammar (TSG) improves parsing speed five to seven times over the CFG parser with only a small accuracy loss. The grammar-based LSTM parsers also outperforms BART- and T5-based seq2seq parsers on the OOD splits, despite having less than one tenth of parameters and no pretraining. We also verified our approach on the SMCalflow-CS dataset, particularly, on the zero-shot learning task.

pdf abs
Exploiting Rich Textual User-Product Context for Improving Personalized Sentiment Analysis
Chenyang Lyu | Linyi Yang | Yue Zhang | Yvette Graham | Jennifer Foster

User and product information associated with a review is useful for sentiment polarity prediction. Typical approaches incorporating such information focus on modeling users and products as implicitly learned representation vectors. Most do not exploit the potential of historical reviews, or those that currently do require unnecessary modifications to model architectureor do not make full use of user/product associations. The contribution of this work is twofold: i) a method to explicitly employ historical reviews belonging to the same user/product in initializing representations, and ii) efficient incorporation of textual associations between users and products via a user-product cross-context module. Experiments on the IMDb, Yelp-2013 and Yelp-2014 English benchmarks with BERT, SpanBERT and Longformer pretrained language models show that our approach substantially outperforms previous state-of-the-art.

Sequence-to-sequence (seq2seq) models based on the Transformer architecture have become a ubiquitous tool applicable not only to classical text generation tasks such as machine translation and summarization but also to any other task where an answer can be represented in a form of a finite text fragment (e.g., question answering). However, when deploying a model in practice, we need not only high performance but also an ability to determine cases where the model is not applicable. Uncertainty estimation (UE) techniques provide a tool for identifying out-of-domain (OOD) input where the model is susceptible to errors. State-of-the-art UE methods for seq2seq models rely on computationally heavyweight and impractical deep ensembles. In this work, we perform an empirical investigation of various novel UE methods for large pre-trained seq2seq models T5 and BART on three tasks: machine translation, text summarization, and question answering. We apply computationally lightweight density-based UE methods to seq2seq models and show that they often outperform heavyweight deep ensembles on the task of OOD detection.

pdf abs
Emotion Cause Extraction on Social Media without Human Annotation
Debin Xiao | Rui Xia | Jianfei Yu

In social media, there is a vast amount of information pertaining to people’s emotions and the corresponding causes. The emotion cause extraction (ECE) from social media data is an important research area that has not been thoroughly explored due to the lack of fine-grained annotations. Early studies referred to either unsupervised rule-based methods or supervised machine learning methods using a number of manually annotated data in specific domains. However, the former suffers from limitations in extraction performance, while the latter is constrained by the availability of fine-grained annotations and struggles to generalize to diverse domains. To address these issues, this paper proposes a new ECE framework on Chinese social media that achieves high extraction performance and generalizability without relying on human annotation. Specifically, we design a more dedicated rule-based system based on constituency parsing tree to discover causal patterns in social media. This system enables us to acquire large amounts of fine-grained annotated data. Next, we train a neural model on the rule-annotated dataset with a specific training strategy to further improve the model’s generalizability. Extensive experiments demonstrate the superiority of our approach over other methods in unsupervised and weakly-supervised settings.

For real-world language applications, detecting an out-of-distribution (OOD) sample is helpful to alert users or reject such unreliable samples. However, modern over-parameterized language models often produce overconfident predictions for both in-distribution (ID) and OOD samples. In particular, language models suffer from OOD samples with a similar semantic representation to ID samples since these OOD samples lie near the ID manifold.A rejection network can be trained with ID and diverse outlier samples to detect test OOD samples, but explicitly collecting auxiliary OOD datasets brings an additional burden for data collection. In this paper, we propose a simple but effective method called Pseudo Outlier Exposure (POE) that constructs a surrogate OOD dataset by sequentially masking tokens related to ID classes. The surrogate OOD sample introduced by POE shows a similar representation to ID data, which is most effective in training a rejection network. Our method does not require any external OOD data and can be easily implemented within off-the-shelf Transformers.A comprehensive comparison with state-of-the-art algorithms demonstrates POE’s competitiveness on several text classification benchmarks.

pdf abs
Adversarial Multi-task Learning for End-to-end Metaphor Detection
Shenglong Zhang | Ying Liu

Metaphor detection (MD) suffers from limited training data. In this paper, we started with a linguistic rule called Metaphor Identification Procedure and then proposed a novel multi-task learning framework to transfer knowledge in basic sense discrimination (BSD) to MD. BSD is constructed from word sense disambiguation (WSD), which has copious amounts of data. We leverage adversarial training to align the data distributions of MD and BSD in the same feature space, so task-invariant representations can be learned. To capture fine-grained alignment patterns, we utilize the multi-mode structures of MD and BSD. Our method is totally end-to-end and can mitigate the data scarcity problem in MD. Competitive results are reported on four public datasets. Our code and datasets are available.

pdf abs
SERENGETI: Massively Multilingual Language Models for Africa
Ife Adebara | AbdelRahim Elmadany | Muhammad Abdul-Mageed | Alcides Alcoba Inciarte

Multilingual pretrained language models (mPLMs) acquire valuable, generalizable linguistic information during pretraining and have advanced the state of the art on task-specific finetuning. To date, only ~31 out of ~2,000 African languages are covered in existing language models. We ameliorate this limitation by developing SERENGETI, a set of massively multilingual language model that covers 517 African languages and language varieties. We evaluate our novel models on eight natural language understanding tasks across 20 datasets, comparing to 4 mPLMs that cover 4-23 African languages. SERENGETI outperforms other models on 11 datasets across the eights tasks, achieving 82.27 average F_1. We also perform analyses of errors from our models, which allows us to investigate the influence of language genealogy and linguistic similarity when the models are applied under zero-shot settings. We will publicly release our models for research. Anonymous link

pdf abs
Prompt- and Trait Relation-aware Cross-prompt Essay Trait Scoring
Heejin Do | Yunsu Kim | Gary Geunbae Lee

Automated essay scoring (AES) aims to score essays written for a given prompt, which defines the writing topic. Most existing AES systems assume to grade essays of the same prompt as used in training and assign only a holistic score. However, such settings conflict with real-education situations; pre-graded essays for a particular prompt are lacking, and detailed trait scores of sub-rubrics are required. Thus, predicting various trait scores of unseen-prompt essays (called cross-prompt essay trait scoring) is a remaining challenge of AES. In this paper, we propose a robust model: prompt- and trait relation-aware cross-prompt essay trait scorer. We encode prompt-aware essay representation by essay-prompt attention and utilizing the topic-coherence feature extracted by the topic-modeling mechanism without access to labeled data; therefore, our model considers the prompt adherence of an essay, even in a cross-prompt setting. To facilitate multi-trait scoring, we design trait-similarity loss that encapsulates the correlations of traits. Experiments prove the efficacy of our model, showing state-of-the-art results for all prompts and traits. Significant improvements in low-resource-prompt and inferior traits further indicate our model’s strength.

pdf abs
AugESC: Dialogue Augmentation with Large Language Models for Emotional Support Conversation
Chujie Zheng | Sahand Sabour | Jiaxin Wen | Zheng Zhang | Minlie Huang

Crowdsourced dialogue corpora are usually limited in scale and topic coverage due to the expensive cost of data curation. This would hinder the generalization of downstream dialogue models to open-domain topics. In this work, we leverage large language models for dialogue augmentation in the task of emotional support conversation (ESC). By treating dialogue augmentation as a dialogue completion task, we prompt a fine-tuned language model to complete full dialogues from available dialogue posts of various topics, which are then postprocessed based on heuristics. Applying this approach, we construct AugESC, an augmented dataset for the ESC task, which largely extends the scale and topic coverage of the crowdsourced ESConv corpus. Through comprehensive human evaluation, we demonstrate that our approach is superior to strong baselines of dialogue augmentation and that AugESC has comparable dialogue quality to the crowdsourced corpus. We also conduct human interactive evaluation and prove that post-training on AugESC improves downstream dialogue models’ generalization ability to open-domain topics. These results suggest the utility of AugESC and highlight the potential of large language models in improving data-scarce dialogue generation tasks.

pdf abs
2*n is better than n²: Decomposing Event Coreference Resolution into Two Tractable Problems
Shafiuddin Rehan Ahmed | Abhijnan Nath | James H. Martin | Nikhil Krishnaswamy

Event Coreference Resolution (ECR) is the task of linking mentions of the same event either within or across documents. Most mention pairs are not coreferent, yet many that are coreferent can be identified through simple techniques such as lemma matching of the event triggers or the sentences in which they appear. Existing methods for training coreference systems sample from a largely skewed distribution, making it difficult for the algorithm to learn coreference beyond surface matching. Additionally, these methods are intractable because of the quadratic operations needed. To address these challenges, we break the problem of ECR into two parts: a) a heuristic to efficiently filter out a large number of non-coreferent pairs, and b) a training approach on a balanced set of coreferent and non-coreferent mention pairs. By following this approach, we show that we get comparable results to the state of the art on two popular ECR datasets while significantly reducing compute requirements. We also analyze the mention pairs that are “hard” to accurately classify as coreferent or non-coreferentcode repo: \mathtt{github.com/ahmeshaf/lemma\_ce\_coref}.

Multimedia summarization with multimodal output (MSMO) is a recently explored application in language grounding. It plays an essential role in real-world applications, i.e., automatically generating cover images and titles for news articles or providing introductions to online videos. However, existing methods extract features from the whole video and article and use fusion methods to select the representative one, thus usually ignoring the critical structure and varying semantics with video/document. In this work, we propose a Semantics-Consistent Cross-domain Summarization (SCCS) model based on optimal transport alignment with visual and textual segmentation. Our method first decomposes both videos and articles into segments in order to capture the structural semantics, and then follows a cross-domain alignment objective with optimal transport distance, which leverages multimodal interaction to match and select the visual and textual summary. We evaluated our method on three MSMO datasets, and achieved performance improvement by 8% & 6% of textual and 6.6% &5.7% of video summarization, respectively, which demonstrated the effectiveness of our method in producing high-quality multimodal summaries.

pdf abs
General-to-Specific Transfer Labeling for Domain Adaptable Keyphrase Generation
Rui Meng | Tong Wang | Xingdi Yuan | Yingbo Zhou | Daqing He

Training keyphrase generation (KPG) models require a large amount of annotated data, which can be prohibitively expensive and often limited to specific domains. In this study, we first demonstrate that large distribution shifts among different domains severely hinder the transferability of KPG models. We then propose a three-stage pipeline, which gradually guides KPG models’ learning focus from general syntactical features to domain-related semantics, in a data-efficient manner. With domain-general phrase pre-training, we pre-train Sequence-to-Sequence models with generic phrase annotations that are widely available on the web, which enables the models to generate phrases in a wide range of domains. The resulting model is then applied in the Transfer Labeling stage to produce domain-specific pseudo keyphrases, which help adapt models to a new domain. Finally, we fine-tune the model with limited data with true labels to fully adapt it to the target domain. Our experiment results show that the proposed process can produce good quality keyphrases in new domains and achieve consistent improvements after adaptation with limited in-domain annotated data. All code and datasets are available at https://github.com/memray/OpenNMT-kpg-release.

Most named entity recognition (NER) systems focus on improving model performance, ignoring the need to quantify model uncertainty, which is critical to the reliability of NER systems in open environments. Evidential deep learning (EDL) has recently been proposed as a promising solution to explicitly model predictive uncertainty for classification tasks. However, directly applying EDL to NER applications faces two challenges, i.e., the problems of sparse entities and OOV/OOD entities in NER tasks. To address these challenges, we propose a trustworthy NER framework named E-NER by introducing two uncertainty-guided loss terms to the conventional EDL, along with a series of uncertainty-guided training strategies. Experiments show that E-NER can be applied to multiple NER paradigms to obtain accurate uncertainty estimation. Furthermore, compared to state-of-the-art baselines, the proposed method achieves a better OOV/OOD detection performance and better generalization ability on OOV entities.

pdf abs
LMCap: Few-shot Multilingual Image Captioning by Retrieval Augmented Language Model Prompting
Rita Ramos | Bruno Martins | Desmond Elliott

Multilingual image captioning has recently been tackled by training with large-scale machine translated data, which is an expensive, noisy, and time-consuming process. Without requiring any multilingual caption data, we propose LMCap, an image-blind few-shot multilingual captioning model that works by prompting a language model with retrieved captions. Specifically, instead of following the standard encoder-decoder paradigm, given an image, LMCap first retrieves the captions of similar images using a multilingual CLIP encoder. These captions are then combined into a prompt for an XGLM decoder, in order to generate captions in the desired language. In other words, the generation model does not directly process the image, instead it processes retrieved captions. Experiments on the XM3600 dataset of geographically diverse images show that our model is competitive with fully-supervised multilingual captioning models, without requiring any supervised training on any captioning data.

pdf abs
Boosting Text Augmentation via Hybrid Instance Filtering Framework
Heng Yang | Ke Li

Text augmentation is an effective technique for addressing the problem of insufficient data in natural language processing. However, existing text augmentation methods tend to focus on few-shot scenarios and usually perform poorly on large public datasets. Our research indicates that existing augmentation methods often generate instances with shifted feature spaces, which leads to a drop in performance on the augmented data (for example, EDA generally loses approximately 2% in aspect-based sentiment classification). To address this problem, we propose a hybrid instance-filtering framework (BoostAug) based on pre-trained language models that can maintain a similar feature space with natural datasets. BoostAug is transferable to existing text augmentation methods (such as synonym substitution and back translation) and significantly improves the augmentation performance by 2-3% in classification accuracy. Our experimental results on three classification tasks and nine public datasets show that BoostAug addresses the performance drop problem and outperforms state-of-the-art text augmentation methods. Additionally, we release the code to help improve existing augmentation methods on large datasets.

Multimodal Review Helpfulness Prediction (MRHP) aims to rank product reviews based on predicted helpfulness scores and has been widely applied in e-commerce via presenting customers with useful reviews. Previous studies commonly employ fully-connected neural networks (FCNNs) as the final score predictor and pairwise loss as the training objective. However, FCNNs have been shown to perform inefficient splitting for review features, making the model difficult to clearly differentiate helpful from unhelpful reviews. Furthermore, pairwise objective, which works on review pairs, may not completely capture the MRHP goal to produce the ranking for the entire review list, and possibly induces low generalization during testing. To address these issues, we propose a listwise attention network that clearly captures the MRHP ranking context and a listwise optimization objective that enhances model generalization. We further propose gradient-boosted decision tree as the score predictor to efficaciously partition product reviews’ representations. Extensive experiments demonstrate that our method achieves state-of-the-art results and polished generalization performance on two large-scale MRHP benchmark datasets.

While Neural Machine Translation (NMT) has achieved great progress in recent years, it still suffers from inaccurate translation of entities (e.g., person/organization name, location), due to the lack of entity training instances. When we humans encounter an unknown entity during translation, we usually first look up in a dictionary and then organize the entity translation together with the translations of other parts to form a smooth target sentence. Inspired by this translation process, we propose an Extract-and-Attend approach to enhance entity translation in NMT, where the translation candidates of source entities are first extracted from a dictionary and then attended to by the NMT model to generate the target sentence. Specifically, the translation candidates are extracted by first detecting the entities in a source sentence and then translating the entities through looking up in a dictionary. Then, the extracted candidates are added as a prefix of the decoder input to be attended to by the decoder when generating the target sentence through self-attention. Experiments conducted on En-Zh and En-Ru demonstrate that the proposed method is effective on improving both the translation accuracy of entities and the overall translation quality, with up to 35% reduction on entity error rate and 0.85 gain on BLEU and 13.8 gain on COMET.

pdf abs
Real-World Compositional Generalization with Disentangled Sequence-to-Sequence Learning
Hao Zheng | Mirella Lapata

Compositional generalization is a basic mechanism in human language learning, which current neural networks struggle with. A recently proposed Disentangled sequence-to-sequence model (Dangle) shows promising generalization capability by learning specialized encodings for each decoding step. We introduce two key modifications to this model which encourage more disentangled representations and improve its compute and memory efficiency, allowing us to tackle compositional generalization in a more realistic setting. Specifically, instead of adaptively re-encoding source keys and values at each time step, we disentangle their representations and only re-encode keys periodically, at some interval. Our new architecture leads to better generalization performance across existing tasks and datasets, and a new machine translation benchmark which we create by detecting naturally occurring compositional patterns in relation to a training set. We show this methodology better emulates real-world requirements than artificial challenges.

pdf abs
Cross-lingual AMR Aligner: Paying Attention to Cross-Attention
Abelardo Carlos Martínez Lorenzo | Pere Lluís Huguet Cabot | Roberto Navigli

This paper introduces a novel aligner for Abstract Meaning Representation (AMR) graphs that can scale cross-lingually, and is thus capable of aligning units and spans in sentences of different languages. Our approach leverages modern Transformer-based parsers, which inherently encode alignment information in their cross-attention weights, allowing us to extract this information during parsing. This eliminates the need for English-specific rules or the Expectation Maximization (EM) algorithm that have been used in previous approaches. In addition, we propose a guided supervised method using alignment to further enhance the performance of our aligner. We achieve state-of-the-art results in the benchmarks for AMR alignment and demonstrate our aligner’s ability to obtain them across multiple languages. Our code will be available at [https://www.github.com/babelscape/AMR-alignment](https://www.github.com/babelscape/AMR-alignment).

Existing solutions to zero-shot text classification either conduct prompting with pre-trained language models, which is sensitive to the choices of templates, or rely on large-scale annotated data of relevant tasks for meta-tuning. In this work, we propose a new paradigm based on self-supervised learning to solve zero-shot text classification tasks by tuning the language models with unlabeled data, called self-supervised tuning. By exploring the inherent structure of free texts, we propose a new learning objective called first sentence prediction to bridge the gap between unlabeled data and text classification tasks. After tuning the model to learn to predict the first sentence in a paragraph based on the rest, the model is able to conduct zero-shot inference on unseen tasks such as topic classification and sentiment analysis. Experimental results show that our model outperforms the state-of-the-art baselines on 7 out of 10 tasks. Moreover, the analysis reveals that our model is less sensitive to the prompt design. Our code and pre-trained models are publicly available at https://github.com/DAMO-NLP-SG/SSTuning.

Natural language contains rich logical structures and logical information, and correctly detecting and accurately understanding these logical structures and information underlying natural language texts is very crucial for NLP models’ performance on many important NLU and NLG tasks. Existing pre-trained language models based on the transformer architecture mostly adopt a classical design for constructing their input embeddings that ignores the logical structures underlying natural language texts, thus limiting their ability to better capture and encode key logical information in the input sequences. To overcome such limitations, in this paper we first propose a novel approach to construct logic-aware input embeddings for transformer language models through a combination of logic detection, logic mapping and hierarchical logical projections, and then develop a corresponding new modeling paradigm that can upgrade existing transformer language models into logical transformers to boost their performance on different NLU and NLG tasks. Our empirical experiments on four important and challenging NLU and NLG tasks demonstrate that our proposed logical transformer language models can achieve superior performance over their baseline transformer models through a deeper understanding of the logical structures of texts.

Large language models (LLMs) have led to a series of breakthroughs in natural language processing (NLP), partly owing to the massive amounts of world knowledge they memorize during pretraining. While many downstream applications provide the model with an informational context to aid its underlying task, how the model’s world knowledge interacts with the factual information presented in the context remains under explored. As a desirable behavior, an LLM should give precedence to the context whenever it contains task-relevant information that conflicts with the model’s memorized knowledge. This enables model predictions to be grounded in the context, which then facilitates updating specific model predictions without frequently retraining the model. By contrast, when the context is irrelevant to the task, the model should ignore it and fall back on its internal knowledge. In this paper, we undertake a first joint study of the aforementioned two properties, namely controllability and robustness, in the context of LLMs. We demonstrate that state-of-the-art T5 and PaLM models (both pretrained and finetuned) could exhibit low controllability and robustness that does not improve with increasing the model size. As a solution, we propose a simple yet effective method – knowledge aware finetuning (KAFT) – to strengthen both controllability and robustness by injecting counterfactual and irrelevant contexts to standard supervised datasets. Our comprehensive evaluation showcases the utility of KAFT across model architectures and sizes.

pdf abs
A Unified Evaluation Framework for Novelty Detection and Accommodation in NLP with an Instantiation in Authorship Attribution
Neeraj Varshney | Himanshu Gupta | Eric Robertson | Bing Liu | Chitta Baral

State-of-the-art natural language processing models have been shown to achieve remarkable performance in ‘closed-world’ settings where all the labels in the evaluation set are known at training time. However, in real-world settings, ‘novel’ instances that do not belong to any known class are often observed. This renders the ability to deal with novelties crucial. To initiate a systematic research in this important area of ‘dealing with novelties’, we introduce NoveltyTask, a multi-stage task to evaluate a system’s performance on pipelined novelty ‘detection’ and ‘accommodation’ tasks. We provide mathematical formulation of NoveltyTask and instantiate it with the authorship attribution task that pertains to identifying the correct author of a given text. We use amazon reviews corpus and compile a large dataset (consisting of 250k instances across 200 authors/labels) for NoveltyTask. We conduct comprehensive experiments and explore several baseline methods for the task. Our results show that the methods achieve considerably low performance making the task challenging and leaving sufficient room for improvement. Finally, we believe our work will encourage research in this underexplored area of dealing with novelties, an important step en route to developing robust systems.

Alzheimer’s Disease (AD) is a neurodegenerative disorder that significantly impacts a patient’s ability to communicate and organize language. Traditional methods for detecting AD, such as physical screening or neurological testing, can be challenging and time-consuming. Recent research has explored the use of deep learning techniques to distinguish AD patients from non-AD patients by analysing the spontaneous speech. These models, however, are limited by the availability of data. To address this, we propose a novel contrastive data augmentation method, which simulates the cognitive impairment of a patient by randomly deleting a proportion of text from the transcript to create negative samples. The corrupted samples are expected to be in worse conditions than the original by a margin. Experimental results on the benchmark ADReSS Challenge dataset demonstrate that our model achieves the best performance among language-based models.

pdf abs
Disentangling Aspect and Stance via a Siamese Autoencoder for Aspect Clustering of Vaccination Opinions
Lixing Zhu | Runcong Zhao | Gabriele Pergola | Yulan He

Mining public opinions about vaccines from social media has been increasingly relevant to analyse trends in public debates and to provide quick insights to policy-makers. However, the application of existing models has been hindered by the wide variety of users’ attitudes and the new aspects continuously arising in the public debate. Existing approaches, frequently framed via well-known tasks, such as aspect classification or text span detection, make direct usage of the supervision information constraining the models to predefined aspect classes, while still not distinguishing those aspects from users’ stances. As a result, this has significantly hindered the dynamic integration of new aspects. We thus propose a model, namely Disentangled Opinion Clustering (DOC), for vaccination opinion mining from social media. DOC is able to disentangle users’ stances from opinions via a disentangling attention mechanism and a Swapping-Autoencoder, and is designed to process unseen aspect categories via a clustering approach, leveraging clustering-friendly representations induced by out-of-the-box Sentence-BERT encodings and disentangling mechanisms. We conduct a thorough experimental assessment demonstrating the benefit of the disentangling mechanisms and cluster-based approach on both the quality of aspect clusters and the generalization across new aspect categories, outperforming existing methodologies on aspect-based opinion mining.

pdf abs
Temporal Relation Classification using Boolean Question Answering
Omer Cohen | Kfir Bar

Classifying temporal relations between a pair of events is crucial to natural language understanding and a well-known natural language processing task. Given a document and two event mentions, the task is aimed at finding which one started first. We propose an efficient approach for temporal relation classification (TRC) using a boolean question answering (QA) model which we fine-tune on questions that we carefully design based on the TRC annotation guidelines, thereby mimicking the way human annotators approach the task. Our new QA-based TRC model outperforms previous state-of-the-art results by 2.4%.

pdf abs
Are Synonym Substitution Attacks Really Synonym Substitution Attacks?
Cheng-Han Chiang | Hung-yi Lee

In this paper, we explore the following question: Are synonym substitution attacks really synonym substitution attacks (SSAs)?We approach this question by examining how SSAs replace words in the original sentence and show that there are still unresolved obstacles that make current SSAs generate invalid adversarial samples. We reveal that four widely used word substitution methods generate a large fraction of invalid substitution words that are ungrammatical or do not preserve the original sentence’s semantics. Next, we show that the semantic and grammatical constraints used in SSAs for detecting invalid word replacements are highly insufficient in detecting invalid adversarial samples.

pdf abs
DivHSK: Diverse Headline Generation using Self-Attention based Keyword Selection
Venkatesh E | Kaushal Maurya | Deepak Kumar | Maunendra Sankar Desarkar

Diverse headline generation is an NLP task where given a news article, the goal is to generate multiple headlines that are true to the content of the article but are different among themselves. This task aims to exhibit and exploit semantically similar one-to-many relationships between a source news article and multiple target headlines. Toward this, we propose a novel model called DIVHSK. It has two components:KEYSELECT for selecting the important keywords, and SEQGEN, for finally generating the multiple diverse headlines. In KEYSELECT, we cluster the self-attention heads of the last layer of the pre-trained encoder and select the most-attentive theme and general keywords from the source article. Then, cluster-specific keyword sets guide the SEQGEN, a pre-trained encoder-decoder model, to generate diverse yet semantically similar headlines. The proposed model consistently outperformed existing literature and our strong baselines and emerged as a state-of-the-art model. We have also created a high-quality multi-reference headline dataset from news articles.

Automatically scoring student answers is an important task that is usually solved using instance-based supervised learning. Recently, similarity-based scoring has been proposed as an alternative approach yielding similar perfor- mance. It has hypothetical advantages such as a lower need for annotated training data and better zero-shot performance, both of which are properties that would be highly beneficial when applying content scoring in a realistic classroom setting. In this paper we take a closer look at these alleged advantages by comparing different instance-based and similarity-based methods on multiple data sets in a number of learning curve experiments. We find that both the demand on data and cross-prompt performance is similar, thus not confirming the former two suggested advantages. The by default more straightforward possibility to give feedback based on a similarity-based approach may thus tip the scales in favor of it, although future work is needed to explore this advantage in practice.

pdf abs
Pragmatic Inference with a CLIP Listener for Contrastive Captioning
Jiefu Ou | Benno Krojer | Daniel Fried

We propose a simple yet effective and robust method for contrastive captioning: generating discriminative captions that distinguish target images from very similar alternative distractor images. Our approach is built on a pragmatic inference procedure that formulates captioning as a reference game between a speaker, which produces possible captions describing the target, and a listener, which selects the target given the caption. Unlike previous methods that derive both speaker and listener distributions from a single captioning model, we leverage an off-the-shelf CLIP model to parameterize the listener. Compared with captioner-only pragmatic models, our method benefits from rich vision-language alignment representations from CLIP when reasoning over distractors. Like previous methods for discriminative captioning, our method uses a hyperparameter to control the tradeoff between the informativity (how likely captions are to allow a human listener to discriminate the target image) and the fluency of the captions. However, we find that our method is substantially more robust to the value of this hyperparameter than past methods, which allows us to automatically optimize the captions for informativity — outperforming past methods for discriminative captioning by 11% to 15% accuracy in human evaluations.

We present a pipeline for a statistical stylometric exploration of a hypothesized partition of a text. Given a parameterization of the text, our pipeline: (1) detects literary features yielding the optimal overlap between the hypothesized and unsupervised partitions, (2) performs a hypothesis-testing analysis to quantify the statistical significance of the optimal overlap, while conserving implicit correlations between units of text that are more likely to be grouped, and (3) extracts and quantifies the importance of features most responsible for the classification, estimates their statistical stability and cluster-wise abundance. We apply our pipeline to the first two books in the Bible, where one stylistic component stands out in the eyes of biblical scholars, namely, the Priestly component. We identify and explore statistically significant stylistic differences between the Priestly and non-Priestly components.

pdf abs
A Language-First Approach for Procedure Planning
Jiateng Liu | Sha Li | Zhenhailong Wang | Manling Li | Heng Ji

Procedure planning, or the ability to predict a series of steps that can achieve a given goal conditioned on the current observation, is critical for building intelligent embodied agents that can assist users in everyday tasks. Encouraged by the recent success of language models (LMs) for zero-shot and few-shot planning, we hypothesize that LMs may be equipped with stronger priors for planning compared to their visual counterparts. To this end, we propose a language-first procedure planning framework with a modularized design: we first align the current and goal observations with corresponding steps and then use a pre-trained LM to predict the intermediate steps. Under this framework, we find that using an image captioning model for alignment can already match state-of-the-art performance and by designing a double retrieval model conditioned over current and goal observations jointly, we can achieve large improvements (19.2%-98.9% relatively higher success rate than state-of-the-art) on both COIN and CrossTask benchmarks. Our work verifies the planning ability of LMs and demonstrates how LMs can serve as a powerful “reasoning engine” even when the input is provided in another modality.

Task-oriented semantic parsing has drawn a lot of interest from the NLP community, and especially the voice assistant industry as it enables representing the meaning of user requests with arbitrarily nested semantics, including multiple intents and compound entities. SOTA models are large seq2seq transformers and require hundreds of thousands of annotated examples to be trained. However annotating such data to bootstrap new domains or languages is expensive and error-prone, especially for requests made of nested semantics. In addition large models easily break the tight latency constraints imposed in a user-facing production environment. As part of this work we explore leveraging external knowledge to improve model accuracy in low-resource and low-compute settings. We demonstrate that using knowledge-enhanced encoders inside seq2seq models does not result in performance gains by itself, but jointly learning to uncover entities in addition to the parse generation is a simple yet effective way of improving performance across the board. We show this is especially true in the low-compute scarce-data setting and for entity-rich domains, with relative gains up to 74.48% on the TOPv2 dataset.

pdf abs
TempLM: Distilling Language Models into Template-Based Generators
Tianyi Zhang | Mina Lee | Xiang Lisa Li | Ende Shen | Tatsunori Hashimoto

While pretrained language models (PLMs) have greatly improved text generation, they have also been known to produce unfaithful or inappropriate content. In contrast, classic template-based systems provide strong guarantees of faithfulness at the cost of fluency. We propose TempLM, which achieves the best of both worlds by distilling a PLM into a template-based generator. On the E2E and SynthBio data-to-text datasets, we show that TempLM is more faithful than the original PLM and is more fluent than prior template systems. Notably, on an out-of-domain evaluation, TempLM reduces a finetuned BART model’s unfaithfulness rate from 83% to 0%. In a human study, we find that TempLM’s templates substantially improve upon human-written ones in BERTScore.

pdf abs
Incorporating Graph Information in Transformer-based AMR Parsing
Pavlo Vasylenko | Pere Lluís Huguet Cabot | Abelardo Carlos Martínez Lorenzo | Roberto Navigli

Abstract Meaning Representation (AMR) is a Semantic Parsing formalism that aims at providing a semantic graph abstraction representing a given text. Current approaches are based on autoregressive language models such as BART or T5, fine-tuned through Teacher Forcing to obtain a linearized version of the AMR graph from a sentence. In this paper, we present LeakDistill, a model and method that explores a modification to the Transformer architecture, using structural adapters to explicitly incorporate graph information into the learned representations and improve AMR parsing performance. Our experiments show how, by employing word-to-node alignment to embed graph structural information into the encoder at training time, we can obtain state-of-the-art AMR parsing through self-knowledge distillation, even without the use of additional data. We release the code at [http://www.github.com/sapienzanlp/LeakDistill](http://www.github.com/sapienzanlp/LeakDistill).

pdf abs
Rethinking the Word-level Quality Estimation for Machine Translation from Human Judgement
Zhen Yang | Fandong Meng | Yuanmeng Yan | Jie Zhou

Word-level Quality Estimation (QE) of Machine Translation (MT) aims to detect potential translation errors in the translated sentence without reference. Typically, conventional works on word-level QE are usually designed to predict the quality of translated words in terms of the post-editing effort, where the word labels in the dataset, i.e., OK or BAD, are automatically generated by comparing words between MT sentences and the post-edited sentences through a Translation Error Rate (TER) toolkit. While the post-editing effort can be used to measure the translation quality to some extent, we find it usually conflicts with human judgment on whether the word is well or poorly translated. To investigate this conflict, we first create a golden benchmark dataset, namely HJQE (Human Judgement on Quality Estimation), where the source and MT sentences are identical to the original TER-based dataset and the expert translators directly annotate the poorly translated words on their judgments. Based on our analysis, we further propose two tag-correcting strategies which can make the TER-based artificial QE corpus closer to HJQE. We conduct substantial experiments based on the publicly available WMT En-De and En-Zh corpora. The results not only show our proposed dataset is more consistent with human judgment but also confirm the effectiveness of the proposed tag-correcting strategies.For reviewers, the corpora and codes can be found in the attached files.

Information extraction, e.g., attribute value extraction, has been extensively studied and formulated based only on text. However, many attributes can benefit from image-based extraction, like color, shape, pattern, among others. The visual modality has long been underutilized, mainly due to multimodal annotation difficulty. In this paper, we aim to patch the visual modality to the textual-established attribute in- formation extractor. The cross-modality integration faces several unique challenges: (C1) images and textual descriptions are loosely paired intra-sample and inter-samples; (C2) images usually contain rich backgrounds that can mislead the prediction; (C3) weakly supervised labels from textual-established ex- tractors are biased for multimodal training. We present PV2TEA, an encoder-decoder architecture equipped with three bias reduction schemes: (S1) Augmented label-smoothed contrast to improve the cross-modality alignment for loosely-paired image and text; (S2) Attention-pruning that adaptively distinguishes the visual foreground; (S3) Two-level neighborhood regularization that mitigates the label textual bias via reliability estimation. Empirical results on real-world e-Commerce datasets1 demonstrate up to 11.74% absolute (20.97% relatively) F1 increase over unimodal baselines.

To present, multilingual language models trained using various pre-training tasks like mask language modeling (MLM) have yielded encouraging results on a wide range of downstream tasks. Despite the promising performances, structural knowledge in cross-lingual corpus is less explored in current works, leading to the semantic misalignment. In this paper, we propose a new pre-training task named Structural Contrast Pretraining (SCP) to align the structural words in a parallel sentence, enhancing the models’ ability to comprehend cross-lingual representations. Concretely, each structural word in source and target languages is regarded as a positive pair in SCP. Since contrastive learning compares positive and negative pairs, an increase in the frequency of negative pairings could enhance the performance of the resulting model. Therefore, we further propose Cross-lingual Momentum Contrast (CL-MoCo) to increase the number of negative pairs by maintaining a large size of the queue. CL-MoCo extends the original Moco approach into cross-lingual training and jointly optimizes the source-to-target language and target-to-source language representations, resulting in a more suitable encoder for cross-lingual transfer. We conduct extensive experiments to validate the proposed approach on three cross-lingual tasks across five datasets such as MLQA, WikiAnn, etc, and results prove the effectiveness of our method.

pdf abs
Reducing Sensitivity on Speaker Names for Text Generation from Dialogues
Qi Jia | Haifeng Tang | Kenny Zhu

Changing speaker names consistently throughout a dialogue should not affect its meaning and corresponding outputs for text generation from dialogues. However, pre-trained language models, serving as the backbone for dialogue-processing tasks, have shown to be sensitive to nuances. This may result in unfairness in real-world applications. No comprehensive analysis of this problem has been done in the past. In this work, we propose to quantitatively measure a model’s sensitivity on speaker names, and comprehensively evaluate a number of known methods for reducing speaker name sensitivity, including a novel approach of our own. Extensive experiments on multiple datasets provide a benchmark for this problem and show the favorable performance of our approach in sensitivity reduction and quality of generation.

pdf abs
Topic and Style-aware Transformer for Multimodal Emotion Recognition
Shuwen Qiu | Nitesh Sekhar | Prateek Singhal

Understanding emotion expressions in multimodal signals is key for machines to have a better understanding of human communication. While language, visual and acoustic modalities can provide clues from different perspectives, the visual modality is shown to make minimal contribution to the performance in the emotion recognition field due to its high dimensionality. Therefore, we first leverage the strong multimodality backbone VATT to project the visual signal to the common space with language and acoustic signals. Also, we propose content-oriented features Topic and Speaking style on top of it to approach the subjectivity issues. Experiments conducted on the benchmark dataset MOSEI show our model can outperform SOTA results and effectively incorporate visual signals and handle subjectivity issues by serving as content “normalization”.

The Open-Domain Question Answering (ODQA) task involves retrieving and subsequently generating answers from fine-grained relevant passages within a database. Current systems leverage Pretrained Language Models (PLMs) to model the relationship between questions and passages. However, the diversity in surface form expressions can hinder the model’s ability to capture accurate correlations, especially within complex contexts. Therefore, we utilize Abstract Meaning Representation (AMR) graphs to assist the model in understanding complex semantic information. We introduce a method known as Graph-as-Token (GST) to incorporate AMRs into PLMs. Results from Natural Questions (NQ) and TriviaQA (TQ) demonstrate that our GST method can significantly improve performance, resulting in up to 2.44/3.17 Exact Match score improvements on NQ/TQ respectively. Furthermore, our method enhances robustness and outperforms alternative Graph Neural Network (GNN) methods for integrating AMRs. To the best of our knowledge, we are the first to employ semantic graphs in ODQA.

Existing language models (LMs) predict tokens with a softmax over a finite vocabulary, which can make it difficult to predict rare tokens or phrases. We introduce NPM, the first nonparametric masked language model that replaces this softmax with a nonparametric distribution over every phrase in a reference corpus. NPM fills in the [MASK] solely from retrieving a token from a text corpus. We show that NPM can be efficiently trained with a contrastive objective and an in-batch approximation to full corpus retrieval. Zero-shot evaluation on 16 tasks including classification, fact probing and question answering demonstrates that NPM outperforms significantly larger parametric models, either with or without a retrieve-and-generate approach. It is particularly better at dealing with rare patterns (word senses or facts) and predicting rare or nearly unseen words (e.g., non-Latin script). We release the model and code at github.com/facebookresearch/NPM.

Knowledge base question answering (KBQA) is a challenging task that aims to retrieve correct answers from large-scale knowledge bases. Existing attempts primarily focus on entity representation and final answer reasoning, which results in limited supervision for this task. Moreover, the relations, which empirically determine the reasoning path selection, are not fully considered in recent advancements. In this study, we propose a novel framework, RE-KBQA, that utilizes relations in the knowledge base to enhance entity representation and introduce additional supervision. We explore guidance from relations in three aspects, including (1) distinguishing similar entities by employing a variational graph auto-encoder to learn relation importance; (2) exploring extra supervision by predicting relation distributions as soft labels with a multi-task scheme; (3) designing a relation-guided re-ranking algorithm for post-processing. Experimental results on two benchmark datasets demonstrate the effectiveness and superiority of our framework, improving the F1 score by 5.8% from 40.5 to 46.3 on CWQ and 5.7% from 62.8 to 68.5 on WebQSP, better or on par with state-of-the-art methods.

Pre-trained multilingual language models (PMLMs) are commonly used when dealing with data from multiple languages and cross-lingual transfer. However, PMLMs are trained on varying amounts of data for each language. In practice this means their performance is often much better on English than many other languages. We explore to what extent this also applies to moral norms. Do the models capture moral norms from English and impose them on other languages? Do the models exhibit random and thus potentially harmful beliefs in certain languages? Both these issues could negatively impact cross-lingual transfer and potentially lead to harmful outcomes. In this paper, we (1) apply the MORALDIRECTION framework to multilingual models, comparing results in German, Czech, Arabic, Chinese, and English, (2) analyse model behaviour on filtered parallel subtitles corpora, and (3) apply the models to a Moral Foundations Questionnaire, comparing with human responses from different countries. Our experiments demonstrate that, indeed, PMLMs encode differing moral biases, but these do not necessarily correspond to cultural differences or commonalities in human opinions. We release our code and models.

pdf abs
Retrieving Relevant Context to Align Representations for Cross-lingual Event Detection
Chien Nguyen | Linh Ngo | Thien Nguyen

We study the problem of cross-lingual transfer learning for event detection (ED) where models trained on a source language are expected to perform well on data for a new target language. Among a few recent works for this problem, the main approaches involve representation matching (e.g., adversarial training) that aims to eliminate language-specific features from the representations to achieve the language-invariant representations. However, due to the mix of language-specific features with event-discriminative context, representation matching methods might also remove important features for event prediction, thus hindering the performance for ED. To address this issue, we introduce a novel approach for cross-lingual ED where representations are augmented with additional context (i.e., not eliminating) to bridge the gap between languages while enriching the contextual information to facilitate ED. At the core of our method involves a retrieval model that retrieves relevant sentences in the target language for an input sentence to compute augmentation representations. Experiments on three languages demonstrate the state-of-the-art performance of our model for cross-lingual ED.

pdf abs
NormNet: Normalize Noun Phrases for More Robust NLP
Minlong Peng | Mingming Sun

A critical limitation of deep NLP models is their over-fitting over spurious features. Previous work has proposed several approaches to debunk such features and reduce their impact on the learned models. In this work, a normalization strategy is proposed to eliminate the false features caused by the textual surfaces of noun phrases. The motivation for this strategy is that noun phrases often play the role of slots in textual expressions and their exact forms are often not that important for performing the final task. As an intuitive example, consider the expression ”x like eating y". There are a huge number of suitable instantiations for x and y in the locale. However, humans can already infer the sentiment polarity of x toward y without knowing their exact forms.Based on this intuition, we introduce NormNet, a pretrained language model based network, to implement the normalization strategy. NormNet learns to replace as many noun phrases in the input sentence as possible with pre-defined base forms. The output of NormNet is then fed as input to a prompt-based learning model to perform label prediction. To evaluate the effectiveness of our strategy, we conducted experimental studies on several tasks, including aspect sentiment classification (ASC), semantic text similarity (STS), and natural language inference (NLI). The experimental results confirm the effectiveness of our strategy.

Text classification in education, usually called auto-tagging, is the automated process of assigning relevant tags to educational content, such as questions and textbooks. However, auto-tagging suffers from a data scarcity problem, which stems from two major challenges: 1) it possesses a large tag space and 2) it is multi-label. Though a retrieval approach is reportedly good at low-resource scenarios, there have been fewer efforts to directly address the data scarcity problem. To mitigate these issues, here we propose a novel retrieval approach CEAA that provides effective learning in educational text classification. Our main contributions are as follows: 1) we leverage transfer learning from question-answering datasets, and 2) we propose a simple but effective data augmentation method introducing cross-encoder style texts to a bi-encoder architecture for more efficient inference. An extensive set of experiments shows that our proposed method is effective in multi-label scenarios and low-resource tags compared to state-of-the-art models.

pdf abs
Adversarial Robustness of Prompt-based Few-Shot Learning for Natural Language Understanding
Venkata Prabhakara Sarath Nookala | Gaurav Verma | Subhabrata Mukherjee | Srijan Kumar

State-of-the-art few-shot learning (FSL) methods leverage prompt-based fine-tuning to obtain remarkable results for natural language understanding (NLU) tasks. While much of the prior FSL methods focus on improving downstream task performance, there is a limited understanding of the adversarial robustness of such methods. In this work, we conduct an extensive study of several state-of-the-art FSL methods to assess their robustness to adversarial perturbations. To better understand the impact of various factors towards robustness (or the lack of it), we evaluate prompt-based FSL methods against fully fine-tuned models for aspects such as the use of unlabeled data, multiple prompts, number of few-shot examples, model size and type. Our results on six GLUE tasks indicate that compared to fully fine-tuned models, vanilla FSL methods lead to a notable relative drop in task performance (i.e., are less robust) in the face of adversarial perturbations. However, using (i) unlabeled data for prompt-based FSL and (ii) multiple prompts flip the trend – the few-shot learning approaches demonstrate a lesser drop in task performance than fully fine-tuned models. We further demonstrate that increasing the number of few-shot examples and model size lead to increased adversarial robustness of vanilla FSL methods. Broadly, our work sheds light on the adversarial robustness evaluation of prompt-based FSL methods for NLU tasks.

pdf abs
This prompt is measuring <mask>: evaluating bias evaluation in language models
Seraphina Goldfarb-Tarrant | Eddie Ungless | Esma Balkir | Su Lin Blodgett

Bias research in NLP seeks to analyse models for social biases, thus helping NLP practitioners uncover, measure, and mitigate social harms. We analyse the body of work that uses prompts and templates to assess bias in language models. We draw on a measurement modelling framework to create a taxonomy of attributes that capture what a bias test aims to measure and how that measurement is carried out. By applying this taxonomy to 90 bias tests, we illustrate qualitatively and quantitatively that core aspects of bias test conceptualisations and operationalisations are frequently unstated or ambiguous, carry implicit assumptions, or be mismatched. Our analysis illuminates the scope of possible bias types the field is able to measure, and reveals types that are as yet under-researched. We offer guidance to enable the community to explore a wider section of the possible bias space, and to better close the gap between desired outcomes and experimental design, both for bias and for evaluating language models more broadly.

pdf abs
Towards Open Environment Intent Prediction
Yunhua Zhou | Jiawei Hong | Xipeng Qiu

Out-of-Domain (OOD) Intent Classification and New Intent Discovering are two basic and critical tasks in the Task-Oriented Dialogue System, which are typically treated two independent tasks. Classification focuses on identifying intents beyond the predefined set of the dialog system, but it will not further differentiate detected OOD intents in fine granularity. Discovering focuses on how to cluster unlabeled samples according to their semantic representation, which relies heavily on prior knowledge and can not provide label information for the formed clusters. To be closer to the real user-facing scenarios, we introduce a task paradigm to extend Classification with Discovering referred as Open Environment Intent Prediction, which is to make a further fine-grained discovery of OOD based on OOD Intent Classification. Using various widely-used generative models as an archetype, we propose a general scheme for Open Environment Intent Prediction. In a nutshell, we first perform intent detection to identify the In-domain (IND) samples and then generate labels for those identified as OOD. With these generated labels, we can discover new general intents and provide label information for them. We develop a suite of benchmarks on the existing intent datasets and present a simple yet effective implementation. Extensive experiments demonstrate that our method establishes substantial improvement compared to the baselines.

pdf abs
Teamwork Is Not Always Good: An Empirical Study of Classifier Drift in Class-incremental Information Extraction
Minqian Liu | Lifu Huang

Class-incremental learning (CIL) aims to develop a learning system that can continually learn new classes from a data stream without forgetting previously learned classes. When learning classes incrementally, the classifier must be constantly updated to incorporate new classes, and the drift in decision boundary may lead to severe forgetting. This fundamental challenge, however, has not yet been studied extensively, especially in the setting where no samples from old classes are stored for rehearsal. In this paper, we take a closer look at how the drift in the classifier leads to forgetting, and accordingly, design four simple yet (super-) effective solutions to alleviate the classifier drift: an Individual Classifiers with Frozen Feature Extractor (ICE) framework where we individually train a classifier for each learning session, and its three variants ICE-PL, ICE-O, and ICE-PL&O which further take the logits of previously learned classes from old sessions or a constant logit of an Other class as constraint to the learning of new classifiers. Extensive experiments and analysis on 6 class-incremental information extraction tasks demonstrate that our solutions, especially ICE-O, consistently show significant improvement over the previous state-of-the-art approaches with up to 44.7% absolute F-score gain, providing a strong baseline and insights for future research on class-incremental learning.

pdf abs
C-XNLI: Croatian Extension of XNLI Dataset
Leo Obadić | Andrej Jertec | Marko Rajnović | Branimir Dropuljić

Comprehensive multilingual evaluations have been encouraged by emerging cross-lingual benchmarks and constrained by existing parallel datasets. To partially mitigate this limitation, we extended the Cross-lingual Natural Language Inference (XNLI) corpus with Croatian. The development and test sets were translated by a professional translator, and we show that Croatian is consistent with other XNLI dubs. The train set is translated using Facebook’s 1.2B parameter m2m_100 model. We thoroughly analyze the Croatian train set and compare its quality with the existing machine-translated German set. The comparison is based on 2000 manually scored sentences per language using a variant of the Direct Assessment (DA) score commonly used at the Conference on Machine Translation (WMT). Our findings reveal that a less-resourced language like Croatian is still lacking in translation quality of longer sentences compared to German. However, both sets have a substantial amount of poor quality translations, which should be considered in translation-based training or evaluation setups.

pdf abs
AVATAR: A Parallel Corpus for Java-Python Program Translation
Wasi Uddin Ahmad | Md Golam Rahman Tushar | Saikat Chakraborty | Kai-Wei Chang

Program translation refers to migrating source code from one programming language to another. It has tremendous practical value in software development, as porting software across languages is time-consuming and costly. Automating program translation is of paramount importance in software migration, and recently researchers explored unsupervised approaches due to the unavailability of parallel corpora. However, the availability of pre-trained language models for programming languages enables supervised fine-tuning with a small number of labeled examples. Therefore, we present AVATAR, a collection of 9,515 programming problems and their solutions written in two popular languages, Java and Python. AVATAR is collected from competitive programming sites, online platforms, and open-source repositories. Furthermore, AVATAR includes unit tests for 250 examples to facilitate functional correctness evaluation. We benchmark several pre-trained language models fine-tuned on AVATAR. Experiment results show that the models lack in generating functionally accurate code.

pdf abs
On Dataset Transferability in Active Learning for Transformers
Fran Jelenić | Josip Jukić | Nina Drobac | Jan Snajder

Active learning (AL) aims to reduce labeling costs by querying the examples most beneficial for model learning. While the effectiveness of AL for fine-tuning transformer-based pre-trained language models (PLMs) has been demonstrated, it is less clear to what extent the AL gains obtained with one model transfer to others. We consider the problem of transferability of actively acquired datasets in text classification and investigate whether AL gains persist when a dataset built using AL coupled with a specific PLM is used to train a different PLM. We link the AL dataset transferability to the similarity of instances queried by the different PLMs and show that AL methods with similar acquisition sequences produce highly transferable datasets regardless of the models used. Additionally, we show that the similarity of acquisition sequences is influenced more by the choice of the AL method than the choice of the model.

pdf abs
Structured Persuasive Writing Support in Legal Education: A Model and Tool for German Legal Case Solutions
Florian Weber | Thiemo Wambsganss | Seyed Parsa Neshaei | Matthias Soellner

We present an annotation approach for capturing structured components and arguments inlegal case solutions of German students. Based on the appraisal style, which dictates the structured way of persuasive writing in German law, we propose an annotation scheme with annotation guidelines that identify structured writing in legal case solutions. We conducted an annotation study with two annotators and annotated legal case solutions to capture the structures of a persuasive legal text. Based on our dataset, we trained three transformer-based models to show that the annotated components can be successfully predicted, e.g. to provide users with writing assistance for legal texts. We evaluated a writing support system in which our models were integrated in an online experiment with law students and found positive learning success and users’ perceptions. Finally, we present our freely available corpus of 413 law student case studies to support the development of intelligent writing support systems.

Building robust deep neural networks (DNNs) against adversarial attacks is an important but challenging task. Previous defense approaches mainly focus on developing new model structures or training algorithms, but they do little to tap the potential of training instances, especially instances with robust patterns carring innate robustness. In this paper, we show that robust and non-robust instances in the training dataset, though are both important for test performance, have contrary impacts on robustness, which makes it possible to build a highly robust model by leveraging the training dataset in a more effective way. We propose a new method that can distinguish between robust instances from non-robust ones according to the model’s sensitivity to perturbations on individual instances during training. Surprisingly, we find that the model under standard training easily overfits the robust instances by relying on their simple patterns before the model completely learns their robust features. Finally, we propose a new mitigation algorithm to further release the potential of robust instances. Experimental results show that proper use of robust instances in the original dataset is a new line to achieve highly robust models.

The open-ended Visual Question Answering (VQA) task requires AI models to jointly reason over visual and natural language inputs using world knowledge. Recently, pre-trained Language Models (PLM) such as GPT-3 have been applied to the task and shown to be powerful world knowledge sources. However, these methods suffer from low knowledge coverage caused by PLM bias – the tendency to generate certain tokens over other tokens regardless of prompt changes, and high dependency on the PLM quality – only models using GPT-3 can achieve the best result. To address the aforementioned challenges, we propose RASO: a new VQA pipeline that deploys a generate-then-select strategy guided by world knowledge for the first time. Rather than following the de facto standard to train a multi-modal model that directly generates the VQA answer, {pasted macro ‘MODEL’}name first adopts PLM to generate all the possible answers, and then trains a lightweight answer selection model for the correct answer. As proved in our analysis, RASO expands the knowledge coverage from in-domain training data by a large margin. We provide extensive experimentation and show the effectiveness of our pipeline by advancing the state-of-the-art by 4.1% on OK-VQA, without additional computation cost.

Syllogistic reasoning, a typical form of deductive reasoning, is a critical capability widely required in natural language understanding tasks, such as text entailment and question answering. To better facilitate research on syllogistic reasoning, we develop a benchmark called SylloBase that differs from existing syllogistic datasets in three aspects: (1) Covering a complete taxonomy of syllogism reasoning patterns; (2) Containing both automatically and manually constructed samples; and (3) Involving both the generation and understanding tasks. We automatically construct 50k template-based syllogism samples by mining syllogism patterns from Wikidata and ConceptNet. To improve our dataset’s naturalness and challenge, we apply GPT-3 to paraphrase the template-based data and further manually rewrite 1,000 samples as the test set. State-of-the-art pre-trained language models can achieve the best generation ROUGE-L of 38.72 by T5 and the best multi-choice accuracy of 72.77% by RoBERTa on SylloBase, which indicates the great challenge of learning diverse syllogistic reasoning types on SylloBase. Our datasets are released at https://github.com/casually-PYlearner/SYLLOBASE.

pdf abs
Categorial grammar induction from raw data
Christian Clark | William Schuler

Grammar induction, the task of learning a set of grammatical rules from raw or minimally labeled text data, can provide clues about what kinds of syntactic structures are learnable without prior knowledge. Recent work (e.g., Kim et al., 2019; Zhu et al., 2020; Jin et al., 2021a) has achieved advances in unsupervised induction of probabilistic context-free grammars (PCFGs). However, categorial grammar induction has received less recent attention, despite allowing inducers to support a larger set of syntactic categories—due to restrictions on how categories can combine—and providing a transparent interface with compositional semantics, opening up possibilities for models that jointly learn form and meaning. Motivated by this, we propose a new model for inducing a basic (Ajdukiewicz, 1935; Bar-Hillel, 1953) categorial grammar. In contrast to earlier categorial grammar induction systems (e.g., Bisk and Hockenmaier, 2012), our model learns from raw data without any part-of-speech information. Experiments on child-directed speech show that our model attains a recall-homogeneity of 0.33 on average, which dramatically increases to 0.59 when a bias toward forward function application is added to the model.

pdf abs
Attribute Controlled Dialogue Prompting
Runcheng Liu | Ahmad Rashid | Ivan Kobyzev | Mehdi Rezagholizadeh | Pascal Poupart

Prompt-tuning has become an increasingly popular parameter-efficient method for adapting large pretrained language models to downstream tasks. However, both discrete prompting and continuous prompting assume fixed prompts for all data samples within a task, neglecting the fact that inputs vary greatly in some tasks such as open-domain dialogue generation. In this paper, we present a novel, instance-specific prompt-tuning algorithm for dialogue generation. Specifically, we generate prompts based on instance-level control code, rather than the conversation history, to explore their impact on controlled dialogue generation. Experiments on popular open-domain dialogue datasets, evaluated on both automated metrics and human evaluation, demonstrate that our method is superior to prompting baselines and comparable to fine-tuning with only 5%-6% of total parameters.

pdf abs
Open-World Factually Consistent Question Generation
Himanshu Maheshwari | Sumit Shekhar | Apoorv Saxena | Niyati Chhaya

Question generation methods based on pre-trained language models often suffer from factual inconsistencies and incorrect entities and are not answerable from the input paragraph. Domain shift – where the test data is from a different domain than the training data - further exacerbates the problem of hallucination. This is a critical issue for any natural language application doing question generation. In this work, we propose an effective data processing technique based on de-lexicalization for consistent question generation across domains. Unlike existing approaches for remedying hallucination, the proposed approach does not filter training data and is generic across question-generation models. Experimental results across six benchmark datasets show that our model is robust to domain shift and produces entity-level factually consistent questions without significant impact on traditional metrics.

pdf abs
Contrastive Learning of Sociopragmatic Meaning in Social Media
Chiyu Zhang | Muhammad Abdul-Mageed | Ganesh Jawahar

Recent progress in representation and contrastive learning in NLP has not widely considered the class of sociopragmatic meaning (i.e., meaning in interaction within different language communities). To bridge this gap, we propose a novel framework for learning task-agnostic representations transferable to a wide range of sociopragmatic tasks (e.g., emotion, hate speech, humor, sarcasm). Our framework outperforms other contrastive learning frameworks for both in-domain and out-of-domain data, across both the general and few-shot settings. For example, compared to two popular pre-trained language models, our model obtains an improvement of 11.66 average F1 on 16 datasets when fine-tuned on only 20 training samples per dataset. We also show that our framework improves uniformity and preserves the semantic structure of representations. Our code is available at: https://github.com/UBC-NLP/infodcl

This paper studies speculative reasoning task on real-world knowledge graphs (KG) that contain both false negative issue (i.e., potential true facts being excluded) and false positive issue (i.e., unreliable or outdated facts being included). State-of-the-art methods fall short in the speculative reasoning ability, as they assume the correctness of a fact is solely determined by its presence in KG, making them vulnerable to false negative/positive issues. The new reasoning task is formulated as a noisy Positive-Unlabeled learning problem. We propose a variational framework, namely nPUGraph, that jointly estimates the correctness of both collected and uncollected facts (which we call label posterior) and updates model parameters during training. The label posterior estimation facilitates speculative reasoning from two perspectives. First, it improves the robustness of a label posterior-aware graph encoder against false positive links. Second, it identifies missing facts to provide high-quality grounds of reasoning. They are unified in a simple yet effective self-training procedure. Empirically, extensive experiments on three benchmark KG and one Twitter dataset with various degrees of false negative/positive cases demonstrate the effectiveness of nPUGraph.

This research addresses the challenges of Cross-Lingual Summarization (CLS) in low-resource scenarios and over imbalanced multilingual data. Existing CLS studies mostly resort to pipeline frameworks or multi-task methods in bilingual settings. However, they ignore the data imbalance in multilingual scenarios and do not utilize the high-resource monolingual summarization data. In this paper, we propose the Aligned CROSs-lingual Summarization (ACROSS) model to tackle these issues. Our framework aligns low-resource cross-lingual data with high-resource monolingual data via contrastive and consistency loss, which help enrich low-resource information for high-quality summaries. In addition, we introduce a data augmentation method that can select informative monolingual sentences, which facilitates a deep exploration of high-resource information and introduce new information for low-resource languages. Experiments on the CrossSum dataset show that ACROSS outperforms baseline models and obtains consistently dominant performance on 45 language pairs.

pdf abs
RFiD: Towards Rational Fusion-in-Decoder for Open-Domain Question Answering
Cunxiang Wang | Haofei Yu | Yue Zhang

Open-Domain Question Answering (ODQA) systems necessitate a reader model capable of generating answers by simultaneously referring to multiple passages. Although representative models like Fusion-in-Decoder (FiD) have been proposed to address this challenge, these systems can inadvertently rely on spurious features instead of genuine causal relationships between the question and the passages to generate answers. To counter this problem, we introduce the Rational Fusion-in-Decoder (RFiD) model. Our model leverages the encoders of FiD to differentiate between causal relationships and spurious features, subsequently guiding the decoder to generate answers informed by this discernment. Experimental results on two ODQA datasets, Natural Questions (NQ) and TriviaQA (TQ), demonstrate that our model surpasses previous methods, achieving improvements of up to 1.5 and 0.7 in Exact Match scores on NQ, and exhibits an enhanced ability to identify causal relationships.

pdf abs
Unsupervised Keyphrase Extraction by Learning Neural Keyphrase Set Function
Mingyang Song | Haiyun Jiang | Lemao Liu | Shuming Shi | Liping Jing

We create a paradigm shift concerning building unsupervised keyphrase extraction systems in this paper. Instead of modeling the relevance between an individual candidate phrase and the document as in the commonly used framework, we formulate the unsupervised keyphrase extraction task as a document-set matching problem from a set-wise perspective, in which the document and the candidate set are globally matched in the semantic space to particularly take into account the interactions among all candidate phrases. Since it is intractable to exactly extract the keyphrase set by the matching function during the inference, we propose an approximate approach, which obtains the candidate subsets via a set extractor agent learned by reinforcement learning. Exhaustive experimental results demonstrate the effectiveness of our model, which outperforms the recent state-of-the-art unsupervised keyphrase extraction baselines by a large margin.

Pre-trained Language Models (PLMs) may be poisonous with backdoors or bias injected by the suspicious attacker during the fine-tuning process. A core challenge of purifying potentially poisonous PLMs is precisely finding poisonous dimensions. To settle this issue, we propose the Fine-purifying approach, which utilizes the diffusion theory to study the dynamic process of fine-tuning for finding potentially poisonous dimensions. According to the relationship between parameter drifts and Hessians of different dimensions, we can detect poisonous dimensions with abnormal dynamics, purify them by resetting them to clean pre-trained weights, and then fine-tune the purified weights on a small clean dataset. To the best of our knowledge, we are the first to study the dynamics guided by the diffusion theory for safety or defense purposes. Experimental results validate the effectiveness of Fine-purifying even with a small clean dataset.

pdf abs
Retrieving Multimodal Prompts for Generative Visual Question Answering
Timothy Ossowski | Junjie Hu

Recent years have witnessed impressive results of pre-trained vision-language models on knowledge-intensive tasks such as visual question answering (VQA). Despite the recent advances in VQA, existing methods mainly adopt a discriminative formulation that predicts answers within a pre-defined label set, leading to easy overfitting on low-resource domains (e.g., medicine) and poor generalization under domain shift to another dataset. To tackle this limitation, we propose a novel generative model enhanced by multimodal prompt retrieval (MPR) that integrates retrieved prompts and multimodal features to generate answers in free text. Our generative model enables rapid zero-shot dataset adaptation to unseen data distributions and open-set answer labels across datasets. Our experiments on medical VQA tasks show that MPR outperforms its non-retrieval counterpart by up to 30% accuracy points in a few-shot domain adaptation setting.

pdf abs
InfoSync: Information Synchronization across Multilingual Semi-structured Tables
Siddharth Khincha | Chelsi Jain | Vivek Gupta | Tushar Kataria | Shuo Zhang

Information Synchronization of semi-structured data across languages is challenging. For example, Wikipedia tables in one language need to be synchronized with others. To address this problem, we introduce a new dataset InfoSync and a two-step method for tabular synchronization. InfoSync contains 100K entity-centric tables (Wikipedia Infoboxes) across 14 languages, of which a subset (~3.5K pairs) are manually annotated. The proposed method includes 1) Information Alignment to map rows and 2) Information Update for updating missing/outdated information for aligned tables across multilingual tables. When evaluated on InfoSync, information alignment achieves an F1 score of 87.91 (en <-> non-en). To evaluate information updation, we perform human-assisted Wikipedia edits on Infoboxes for 532 table pairs. Our approach obtains an acceptance rate of 77.28% on Wikipedia, showing the effectiveness of the proposed method.

pdf abs
T2IAT: Measuring Valence and Stereotypical Biases in Text-to-Image Generation
Jialu Wang | Xinyue Liu | Zonglin Di | Yang Liu | Xin Wang

*Warning: This paper contains several contents that may be toxic, harmful, or offensive.*In the last few years, text-to-image generative models have gained remarkable success in generating images with unprecedented quality accompanied by a breakthrough of inference speed. Despite their rapid progress, human biases that manifest in the training examples, particularly with regard to common stereotypical biases, like gender and skin tone, still have been found in these generative models. In this work, we seek to measure more complex human biases exist in the task of text-to-image generations. Inspired by the well-known Implicit Association Test (IAT) from social psychology, we propose a novel Text-to-Image Association Test (T2IAT) framework that quantifies the implicit stereotypes between concepts and valence, and those in the images. We replicate the previously documented bias tests on generative models, including morally neutral tests on flowers and insects as well as demographic stereotypical tests on diverse social attributes. The results of these experiments demonstrate the presence of complex stereotypical behaviors in image generations.

pdf abs
An Investigation of Evaluation Methods in Automatic Medical Note Generation
Asma Ben Abacha | Wen-wai Yim | George Michalopoulos | Thomas Lin

Recent studies on automatic note generation have shown that doctors can save significant amounts of time when using automatic clinical note generation (Knoll et al., 2022). Summarization models have been used for this task to generate clinical notes as summaries of doctor-patient conversations (Krishna et al., 2021; Cai et al., 2022). However, assessing which model would best serve clinicians in their daily practice is still a challenging task due to the large set of possible correct summaries, and the potential limitations of automatic evaluation metrics. In this paper we study evaluation methods and metrics for the automatic generation of clinical notes from medical conversation. In particular, we propose new task-specific metrics and we compare them to SOTA evaluation metrics in text summarization and generation, including: (i) knowledge-graph embedding-based metrics, (ii) customized model-based metrics with domain-specific weights, (iii) domain-adapted/fine-tuned metrics, and (iv) ensemble metrics. To study the correlation between the automatic metrics and manual judgments, we evaluate automatic notes/summaries by comparing the system and reference facts and computing the factual correctness, and the hallucination and omission rates for critical medical facts. This study relied on seven datasets manually annotated by domain experts. Our experiments show that automatic evaluation metrics can have substantially different behaviors on different types of clinical notes datasets. However, the results highlight one stable subset of metrics as the most correlated with human judgments with a relevant aggregation of different evaluation criteria.

This paper rethinks translation memory augmented neural machine translation (TM-augmented NMT) from two perspectives, i.e., a probabilistic view of retrieval and the variance-bias decomposition principle. The finding demonstrates that TM-augmented NMT is good at the ability of fitting data (i.e., lower bias) but is more sensitive to the fluctuations in the training data (i.e., higher variance), which provides an explanation to a recently reported contradictory phenomenon on the same translation task: TM-augmented NMT substantially advances NMT without TM under the high resource scenario whereas it fails under the low resource scenario. Then this paper proposes a simple yet effective TM-augmented NMT model to promote the variance and address the contradictory phenomenon. Extensive experiments show that the proposed TM-augmented NMT achieves consistent gains over both conventional NMT and existing TM-augmented NMT under two variance-preferable (low resource and plug-and-play) scenarios as well as the high resource scenario.

pdf abs
Controlling Styles in Neural Machine Translation with Activation Prompt
Yifan Wang | Zewei Sun | Shanbo Cheng | Weiguo Zheng | Mingxuan Wang

Controlling styles in neural machine translation (NMT) has attracted wide attention, as it is crucial for enhancing user experience. Earlier studies on this topic typically concentrate on regulating the level of formality and achieve some progress in this area. However, they still encounter two major challenges. The first is the difficulty in style evaluation. The style comprises various aspects such as lexis, syntax, and others that provide abundant information. Nevertheless, only formality has been thoroughly investigated. The second challenge involves excessive dependence on incremental adjustments, particularly when new styles are necessary. To address both challenges, this paper presents a new benchmark and approach. A multiway stylized machine translation (MSMT) benchmark is introduced, incorporating diverse categories of styles across four linguistic domains. Then, we propose a method named style activation prompt (StyleAP) by retrieving prompts from stylized monolingual corpus, which does not require extra fine-tuning. Experiments show that StyleAP could effectively control the style of translation and achieve remarkable performance.

Few-shot named entity recognition (NER), identifying named entities with a small number of labeled data, has attracted much attention. Frequently, entities are nested within each other. However, most of the existing work on few-shot NER addresses flat entities instead of nested entities. To tackle nested NER in a few-shot setting, it is crucial to utilize the limited labeled data to mine unique features of nested entities, such as the relationship between inner and outer entities and contextual position information. Therefore, in this work, we propose a novel method based on focusing, bridging and prompting for few-shot nested NER without using source domain data. Both focusing and bridging components provide accurate candidate spans for the prompting component. The prompting component leverages the unique features of nested entities to classify spans based on soft prompts and contrastive learning. Experimental results show that the proposed approach achieves state-of-the-art performance consistently on the four benchmark datasets (ACE2004, ACE2005, GENIA and KBP2017) and outperforms several competing baseline models on F1-score by 9.33% on ACE2004, 6.17% on ACE2005, 9.40% on GENIA and 5.12% on KBP2017 on the 5-shot setting.

pdf abs
Together We Make Sense–Learning Meta-Sense Embeddings
Haochen Luo | Yi Zhou | Danushka Bollegala

Sense embedding learning methods learn multiple vectors for a given ambiguous word, corresponding to its different word senses. For this purpose, different methods have been proposed in prior work on sense embedding learning that use different sense inventories, sense-tagged corpora and learning methods. However, not all existing sense embeddings cover all senses of ambiguous words equally well due to the discrepancies in their training resources. To address this problem, we propose the first-ever meta-sense embedding method – Neighbour Preserving Meta-Sense Embeddings, which learns meta-sense embeddings by combining multiple independently trained source sense embeddings such that the sense neighbourhoods computed from the source embeddings are preserved in the meta-embedding space. Our proposed method can combine source sense embeddings that cover different sets of word senses. Experimental results on Word Sense Disambiguation (WSD) and Word-in-Context (WiC) tasks show that the proposed meta-sense embedding method consistently outperforms several competitive baselines. An anonymised version of the source code implementation for our proposed method is submitted to reviewing system. Both source code and the learnt meta-sense embeddings will be publicly released upon paper acceptance.

Generating an informative and attractive title for the product is a crucial task for e-commerce. Most existing works follow the standard multimodal natural language generation approaches, e.g., image captioning, and employ the large scale of human-labelled datasets to train desirable models. However, for novel products, especially in a different domain, there are few existing labelled data. In this paper, we propose a prompt-based approach, i.e., the Multimodal Prompt Learning framework, to accurately and efficiently generate titles for novel products with limited labels. We observe that the core challenges of novel product title generation are the understanding of novel product characteristics and the generation of titles in a novel writing style. To this end, we build a set of multimodal prompts from different modalities to preserve the corresponding characteristics and writing styles of novel products. As a result, with extremely limited labels for training, the proposed method can retrieve the multimodal prompts to generate desirable titles for novel products. The experiments and analyses are conducted on five novel product categories under both the in-domain and out-of-domain experimental settings. The results show that, with only 1% of downstream labelled data for training, our proposed approach achieves the best few-shot results and even achieves competitive results with fully-supervised methods trained on 100% of training data; With the full labelled data for training, our method achieves state-of-the-art results.

pdf abs
Large Language Models are Built-in Autoregressive Search Engines
Noah Ziems | Wenhao Yu | Zhihan Zhang | Meng Jiang

Document retrieval is a key stage of standard Web search engines. Existing dual-encoder dense retrievers obtain representations for questions and documents independently, allowing for only shallow interactions between them. To overcome this limitation, recent autoregressive search engines replace the dual-encoder architecture by directly generating identifiers for relevant documents in the candidate pool. However, the training cost of such autoregressive search engines rises sharply as the number of candidate documents increases. In this paper, we find that large language models (LLMs) can follow human instructions to directly generate URLs for document retrieval. Surprisingly, when providing a few Query-URL pairs as in-context demonstrations, LLMs can generate Web URLs where nearly 90% of the corresponding documents contain correct answers to open-domain questions. In this way, LLMs can be thought of as built-in search engines, since they have not been explicitly trained to map questions to document identifiers. Experiments demonstrate that our method can consistently achieve better retrieval performance than existing retrieval approaches by a significant margin on three open-domain question answering benchmarks, under both zero and few-shot settings. The code for this work can be found at https://github.com/Ziems/llm-url.

Multimodal machine translation (MMT) aims to improve translation quality by incorporating information from other modalities, such as vision. Previous MMT systems focus on better access and use of visual information and tend to validate their methods on image-related datasets. However, these studies face two challenges. First, they can only utilize a limited amount of data that is composed of bilingual texts and images (referred to as “triple data”), which is scarce. Second, current benchmarks for MMT are restricted and do not correspond to realistic scenarios. Therefore, this paper correspondingly establishes new methods and a new dataset for MMT. We propose a novel framework for MMT that addresses these challenges by utilizing large-scale non-triple data, such as monolingual image-text and parallel text-only data. Additionally, we construct a new e-commercial multimodal translation dataset, named EMMT, of which the test set is specifically designed to include ambiguous words that require visual context for accurate translation. Experiments show that our method is well-suited for real-world scenarios and can significantly improve translation performance with more non-triple data. In addition, our model also rivals or surpasses various SOTA models in conventional multimodal translation benchmarks.

pdf abs
From chocolate bunny to chocolate crocodile: Do Language Models Understand Noun Compounds?
Albert Coil | Vered Shwartz

Noun compound interpretation is the task of expressing a noun compound (e.g. chocolate bunny) in a free-text paraphrase that makes the relationship between the constituent nouns explicit (e.g. bunny-shaped chocolate). We propose modifications to the data and evaluation setup of the standard task (Hendrickx et al., 2013), and show that GPT-3 solves it almost perfectly. We then investigate the task of noun compound conceptualization, i.e. paraphrasing a novel or rare noun compound. E.g., chocolate crocodile is a crocodile-shaped chocolate. This task requires creativity, commonsense, and the ability to generalize knowledge about similar concepts. While GPT-3’s performance is not perfect, it is better than that of humans—likely thanks to its access to vast amounts of knowledge, and because conceptual processing is effortful for people (Connell and Lynott, 2012). Finally, we estimate the extent to which GPT-3 is reasoning about the world vs. parroting its training data. We find that the outputs from GPT-3 often have significant overlap with a large web corpus, but that the parroting strategy is less beneficial for novel noun compounds.

Data-driven analyses of biases in historical texts can help illuminate the origin and development of biases prevailing in modern society. However, digitised historical documents pose a challenge for NLP practitioners as these corpora suffer from errors introduced by optical character recognition (OCR) and are written in an archaic language. In this paper, we investigate the continuities and transformations of bias in historical newspapers published in the Caribbean during the colonial era (18th to 19th centuries). Our analyses are performed along the axes of gender, race, and their intersection. We examine these biases by conducting a temporal study in which we measure the development of lexical associations using distributional semantics models and word embeddings. Further, we evaluate the effectiveness of techniques designed to process OCR-generated data and assess their stability when trained on and applied to the noisy historical newspapers. We find that there is a trade-off between the stability of the word embeddings and their compatibility with the historical dataset. We provide evidence that gender and racial biases are interdependent, and their intersection triggers distinct effects. These findings align with the theory of intersectionality, which stresses that biases affecting people with multiple marginalised identities compound to more than the sum of their constituents.

pdf abs
Incomplete Utterance Rewriting by A Two-Phase Locate-and-Fill Regime
Zitong Li | Jiawei Li | Haifeng Tang | Kenny Zhu | Ruolan Yang

Rewriting incomplete and ambiguous utterances can improve dialogue models’ understanding of the context and help them generate better results. However, the existing end-to-end models will have the problem of too large search space, resulting in poor quality of rewriting results. We propose a 2-phase rewriting framework which first predicts the empty slots in the utterance that need to be completed, and then generate the part to be filled into each positions. Our framework is simple to implement, fast to run, and achieves the state-of-the-art results on several public rewriting datasets.

pdf abs
Exploring Variation of Results from Different Experimental Conditions
Maja Popović | Mohammad Arvan | Natalie Parde | Anya Belz

It might reasonably be expected that running multiple experiments for the same task using the same data and model would yield very similar results. Recent research has, however, shown this not to be the case for many NLP experiments. In this paper, we report extensive coordinated work by two NLP groups to run the training and testing pipeline for three neural text simplification models under varying experimental conditions, including different random seeds, run-time environments, and dependency versions, yielding a large number of results for each of the three models using the same data and train/dev/test set splits. From one perspective, these results can be interpreted as shedding light on the reproducibility of evaluation results for the three NTS models, and we present an in-depth analysis of the variation observed for different combinations of experimental conditions. From another perspective, the results raise the question of whether the averaged score should be considered the ‘true’ result for each model.

pdf abs
Playing the Part of the Sharp Bully: Generating Adversarial Examples for Implicit Hate Speech Detection
Nicolás Benjamín Ocampo | Elena Cabrio | Serena Villata

Research on abusive content detection on social media has primarily focused on explicit forms of hate speech (HS), that are often identifiable by recognizing hateful words and expressions. Messages containing linguistically subtle and implicit forms of hate speech still constitute an open challenge for automatic hate speech detection. In this paper, we propose a new framework for generating adversarial implicit HS short-text messages using Auto-regressive Language Models. Moreover, we propose a strategy to group the generated implicit messages in complexity levels (EASY, MEDIUM, and HARD categories) characterizing how challenging these messages are for supervised classifiers. Finally, relying on (Dinan et al., 2019; Vidgen et al., 2021), we propose a “build it, break it, fix it”, training scheme using HARD messages showing how iteratively retraining on HARD messages substantially leverages SOTA models’ performances on implicit HS benchmarks.

Task-oriented dialogue research has mainly focused on a few popular languages like English and Chinese, due to the high dataset creation cost for a new language. To reduce the cost, we apply manual editing to automatically translated data. We create a new multilingual benchmark, X-RiSAWOZ, by translating the Chinese RiSAWOZ to 4 languages: English, French, Hindi, Korean; and a code-mixed English-Hindi language.X-RiSAWOZ has more than 18,000 human-verified dialogue utterances for each language, and unlike most multilingual prior work, is an end-to-end dataset for building fully-functioning agents. The many difficulties we encountered in creating X-RiSAWOZ led us to develop a toolset to accelerate the post-editing of a new language dataset after translation. This toolset improves machine translation with a hybrid entity alignment technique that combines neural with dictionary-based methods, along with many automated and semi-automated validation checks. We establish strong baselines for X-RiSAWOZ by training dialogue agents in the zero- and few-shot settings where limited gold data is available in the target language. Our results suggest that our translation and post-editing methodology and toolset can be used to create new high-quality multilingual dialogue agents cost-effectively. Our dataset, code, and toolkit are released open-source.

pdf abs
Subword Segmental Machine Translation: Unifying Segmentation and Target Sentence Generation
Francois Meyer | Jan Buys

Subword segmenters like BPE operate as a preprocessing step in neural machine translation and other (conditional) language models. They are applied to datasets before training, so translation or text generation quality relies on the quality of segmentations. We propose a departure from this paradigm, called subword segmental machine translation (SSMT). SSMT unifies subword segmentation and MT in a single trainable model. It learns to segment target sentence words while jointly learning to generate target sentences. To use SSMT during inference we propose dynamic decoding, a text generation algorithm that adapts segmentations as it generates translations. Experiments across 6 translation directions show that SSMT improves chrF scores for morphologically rich agglutinative languages. Gains are strongest in the very low-resource scenario. SSMT also learns subwords that are closer to morphemes compared to baselines and proves more robust on a test set constructed for evaluating morphological compositional generalisation.

Deep Neural Networks (DNNs) are becoming integral components of real world services relied upon by millions of users. Unfortunately, architects of these systems can find it difficult to ensure reliable performance as irrelevant details like random initialization can unexpectedly change the outputs of a trained system with potentially disastrous consequences. We formulate the model stability problem by studying how the predictions of a model change, even when it is retrained on the same data, as a consequence of stochasticity in the training process. For Natural Language Understanding (NLU) tasks, we find instability in predictions for a significant fraction of queries. We formulate principled metrics, like per-sample “label entropy” across training runs or within a single training run, to quantify this phenomenon. Intriguingly, we find that unstable predictions do not appear at random, but rather appear to be clustered in data-specific ways. We study data-agnostic regularization methods to improve stability and propose new data-centric methods that exploit our local stability estimates. We find that our localized data-specific mitigation strategy dramatically outperforms data-agnostic methods, and comes within 90% of the gold standard, achieved by ensembling, at a fraction of the computational cost.

pdf abs
What Knowledge Is Needed? Towards Explainable Memory for kNN-MT Domain Adaptation
Wenhao Zhu | Shujian Huang | Yunzhe Lv | Xin Zheng | Jiajun Chen

kNN-MT presents a new paradigm for domain adaptation by building an external datastore, which usually saves all target language token occurrences in the parallel corpus. As a result, the constructed datastore is usually large and possibly redundant. In this paper, we investigate the interpretability issue of this approach: what knowledge does the NMT model need? We propose the notion of local correctness (LAC) as a new angle, which describes the potential translation correctness for a single entry and for a given neighborhood. Empirical study shows that our investigation successfully finds the conditions where the NMT model could easily fail and need related knowledge. Experiments on six diverse target domains and two language-pairs show that pruning according to local correctness brings a light and more explainable memory for kNN-MT domain adaptation.

Aspect Sentiment Triplet Extraction (ASTE) is widely used in various applications. However, existing ASTE datasets are limited in their ability to represent real-world scenarios, hindering the advancement of research in this area. In this paper, we introduce a new dataset, named DMASTE, which is manually annotated to better fit real-world scenarios by providing more diverse and realistic reviews for the task. The dataset includes various lengths, diverse expressions, more aspect types, and more domains than existing datasets. We conduct extensive experiments on DMASTE in multiple settings to evaluate previous ASTE approaches. Empirical results demonstrate that DMASTE is a more challenging ASTE dataset. Further analyses of in-domain and cross-domain settings provide some promising directions for future research.

pdf abs
Grounding the Lexical Substitution Task in Entailment
Talgat Omarov | Grzegorz Kondrak

Existing definitions of lexical substitutes are often vague or inconsistent with the gold annotations. We propose a new definition which is grounded in the relation of entailment; namely, that the sentence that results from the substitution should be in the relation of mutual entailment with the original sentence. We argue that the new definition is well-founded and supported by previous work on lexical entailment. We empirically validate our definition by verifying that it covers the majority of gold substitutes in existing datasets. Based on this definition, we create a new dataset from existing semantic resources. Finally, we propose a novel context augmentation method motivated by the definition, which relates the substitutes to the sense of the target word by incorporating glosses and synonyms directly into the context. Experimental results demonstrate that our augmentation approach improves the performance of lexical substitution systems on the existing benchmarks.

pdf abs
Operator Selection and Ordering in a Pipeline Approach to Efficiency Optimizations for Transformers
Ji Xin | Raphael Tang | Zhiying Jiang | Yaoliang Yu | Jimmy Lin

There exists a wide variety of efficiency methods for natural language processing (NLP) tasks, such as pruning, distillation, dynamic inference, quantization, etc. From a different perspective, we can consider an efficiency method as an operator applied on a model. Naturally, we may construct a pipeline of operators, i.e., to apply multiple efficiency methods on the model sequentially. In this paper, we study the plausibility of this idea, and more importantly, the commutativity and cumulativeness of efficiency operators. We make two interesting observations from our experiments: (1) The operators are commutative—the order of efficiency methods within the pipeline has little impact on the final results; (2) The operators are also cumulative—the final results of combining several efficiency methods can be estimated by combining the results of individual methods. These observations deepen our understanding of efficiency operators and provide useful guidelines for building them in real-world applications.

Developing monolingual large Pre-trained Language Models (PLMs) is shown to be very successful in handling different tasks in Natural Language Processing (NLP). In this work, we present AraMUS, the largest Arabic PLM with 11B parameters trained on 529GB of high-quality Arabic textual data. AraMUS achieves state-of-the-art performances on a diverse set of Arabic classification and generative tasks. Moreover, AraMUS shows impressive few-shot learning abilities compared with the best existing Arabic PLMs.

pdf abs
Leveraging Explicit Procedural Instructions for Data-Efficient Action Prediction
Julia White | Arushi Raghuvanshi | Yada Pruksachatkun

Task-oriented dialogues often require agents to enact complex, multi-step procedures in order to meet user requests. While large language models have found success automating these dialogues in constrained environments, their widespread deployment is limited by the substantial quantities of task-specific data required for training. The following paper presents a data-efficient solution to constructing dialogue systems, leveraging explicit instructions derived from agent guidelines, such as company policies or customer service manuals. Our proposed Knowledge-Augmented Dialogue System (KADS) combines a large language model with a knowledge retrieval module that pulls documents outlining relevant procedures from a predefined set of policies, given a user-agent interaction. To train this system, we introduce a semi-supervised pre-training scheme that employs dialogue-document matching and action-oriented masked language modeling with partial parameter freezing. We evaluate the effectiveness of our approach on prominent task-oriented dialogue datasets, Action-Based Conversations Dataset and Schema-Guided Dialogue, for two dialogue tasks: action state tracking and workflow discovery. Our results demonstrate that procedural knowledge augmentation improves accuracy predicting in- and out-of-distribution actions while preserving high performance in settings with low or sparse data.

pdf abs
Quantifying Train-Evaluation Overlap with Nearest Neighbors
Gauri Kambhatla | Thuy Nguyen | Eunsol Choi

Characterizing benchmark datasets is crucial to interpreting model performance. In this work, we study train-evaluation overlap as a measure of an individual dataset’s adequacy to evaluate model generalization over a wide range of datasets. We quantify the overlap with a simple novel metric based on a nearest neighbors approach between the training and evaluation sets. We identify nearest training examples for each evaluation example by mapping instances with generic and task-specific embedding methods. Our study on eleven classification and extractive QA tasks reveals a wide range of train-evaluation overlap, and we show that the data collection method of the dataset and the difficulty of the task may play a role in the amount of overlap. Lastly, we use our nearest neighbor analysis to identify challenging or potentially mislabeled examples. Our analysis quantifies train-evaluation overlap, providing insights for constructing datasets to study generalization.

pdf abs
Unsupervised Mapping of Arguments of Deverbal Nouns to Their Corresponding Verbal Labels
Aviv Weinstein | Yoav Goldberg

Deverbal nouns are nominal forms of verbs commonly used in written English texts to describe events or actions, as well as their arguments. However, many NLP systems, and in particular pattern-based ones, neglect to handle such nominalized constructions. The solutions that do exist for handling arguments of nominalized constructions are based on semantic annotation and require semantic ontologies, making their applications restricted to a small set of nouns. We propose to adopt instead a more syntactic approach, which maps the arguments of deverbal nouns to the universal-dependency relations of the corresponding verbal construction. We present an unsupervised mechanism—based on contextualized word representations—which allows to enrich universal-dependency trees with dependency arcs denoting arguments of deverbal nouns, using the same labels as the corresponding verbal cases. By sharing the same label set as in the verbal case, patterns that were developed for verbs can be applied without modification but with high accuracy also to the nominal constructions.

pdf abs
The Decades Progress on Code-Switching Research in NLP: A Systematic Survey on Trends and Challenges
Genta Winata | Alham Fikri Aji | Zheng Xin Yong | Thamar Solorio

Code-Switching, a common phenomenon in written text and conversation, has been studied over decades by the natural language processing (NLP) research community. Initially, code-switching is intensively explored by leveraging linguistic theories and, currently, more machine-learning oriented approaches to develop models. We introduce a comprehensive systematic survey on code-switching research in natural language processing to understand the progress of the past decades and conceptualize the challenges and tasks on the code-switching topic. Finally, we summarize the trends and findings and conclude with a discussion for future direction and open questions for further investigation.

pdf abs
Learning to Predict Persona Information for Dialogue Personalization without Explicit Persona Description
Wangchunshu Zhou | Qifei Li | Chenle Li

Personalizing dialogue agents is important for dialogue systems to generate more specific,consistent, and engaging responses. However, most current dialogue personalization approaches rely on explicit persona descriptions during inference, which severely restricts its application. In this paper, we propose a novel approach that learns to predict persona information based on the dialogue history to personalize the dialogue agent without relying on any explicit persona descriptions during inference. Experimental results on the PersonaChat dataset show that the proposed method can improve the consistency of generated responses when conditioning on the predicted profile of the dialogue agent (i.e. “self persona”), and improve the engagingness of the generated responses when conditioning on the predicted persona of the dialogue partner (i.e. “their persona”). We also find that a trained persona prediction model can be successfully transferred to other datasets and help generate more relevant responses.

pdf abs
Automated Refugee Case Analysis: A NLP Pipeline for Supporting Legal Practitioners
Claire Barale | Michael Rovatsos | Nehal Bhuta

In this paper, we introduce an end-to-end pipeline for retrieving, processing, and extracting targeted information from legal cases. We investigate an under-studied legal domain with a case study on refugee law Canada. Searching case law for past similar cases is a key part of legal work for both lawyers and judges, the potential end-users of our prototype. While traditional named-entity recognition labels such as dates are meaningful information in law, we propose to extend existing models and retrieve a total of 19 categories of items from refugee cases. After creating a novel data set of cases, we perform information extraction based on state-of-the-art neural named-entity recognition (NER). We test different architectures including two transformer models, using contextual and non-contextual embeddings, and compare general purpose versus domain-specific pre-training. The results demonstrate that models pre-trained on legal data perform best despite their smaller size, suggesting that domain-matching had a larger effect than network architecture. We achieve a F1- score superior to 90% on five of the targeted categories and superior to 80% on an additional 4 categories.

Self-attention-based models have achieved remarkable progress in short-text mining. However, the quadratic computational complexities restrict their application in long text processing. Prior works have adopted the chunking strategy to divide long documents into chunks and stack a self-attention backbone with the recurrent structure to extract semantic representation. Such an approach disables parallelization of the attention mechanism, significantly increasing the training cost and raising hardware requirements. Revisiting the self-attention mechanism and the recurrent structure, this paper proposes a novel long-document encoding model, Recurrent Attention Network (RAN), to enable the recurrent operation of self-attention. Combining the advantages from both sides, the well-designed RAN is capable of extracting global semantics in both token-level and document-level representations, making it inherently compatible with both sequential and classification tasks, respectively. Furthermore, RAN is computationally scalable as it supports parallelization on long document processing. Extensive experiments demonstrate the long-text encoding ability of the proposed RAN model on both classification and sequential tasks, showing its potential for a wide range of applications.

pdf abs
Exploring the Relationship between Alignment and Cross-lingual Transfer in Multilingual Transformers
Felix Gaschi | Patricio Cerda | Parisa Rastin | Yannick Toussaint

Without any explicit cross-lingual training data, multilingual language models can achieve cross-lingual transfer. One common way to improve this transfer is to perform realignment steps before fine-tuning, i.e., to train the model to build similar representations for pairs of words from translated sentences. But such realignment methods were found to not always improve results across languages and tasks, which raises the question of whether aligned representations are truly beneficial for cross-lingual transfer. We provide evidence that alignment is actually significantly correlated with cross-lingual transfer across languages, models and random seeds. We show that fine-tuning can have a significant impact on alignment, depending mainly on the downstream task and the model. Finally, we show that realignment can, in some instances, improve cross-lingual transfer, and we identify conditions in which realignment methods provide significant improvements. Namely, we find that realignment works better on tasks for which alignment is correlated with cross-lingual transfer when generalizing to a distant language and with smaller models, as well as when using a bilingual dictionary rather than FastAlign to extract realignment pairs. For example, for POS-tagging, between English and Arabic, realignment can bring a +15.8 accuracy improvement on distilmBERT, even outperforming XLM-R Large by 1.7. We thus advocate for further research on realignment methods for smaller multilingual models as an alternative to scaling.

The ability to converse with humans and follow natural language commands is crucial for intelligent unmanned aerial vehicles (a.k.a. drones). It can relieve people’s burden of holding a controller all the time, allow multitasking, and make drone control more accessible for people with disabilities or with their hands occupied. To this end, we introduce Aerial Vision-and-Dialog Navigation (AVDN), to navigate a drone via natural language conversation. We build a drone simulator with a continuous photorealistic environment and collect a new AVDN dataset of over 3k recorded navigation trajectories with asynchronous human-human dialogs between commanders and followers. The commander provides initial navigation instruction and further guidance by request, while the follower navigates the drone in the simulator and asks questions when needed. During data collection, followers’ attention on the drone’s visual observation is also recorded. Based on the AVDN dataset, we study the tasks of aerial navigation from (full) dialog history and propose an effective Human Attention Aided Transformer model (HAA-Transformer), which learns to predict both navigation waypoints and human attention.

pdf abs
Improved Logical Reasoning of Language Models via Differentiable Symbolic Programming
Hanlin Zhang | Jiani Huang | Ziyang Li | Mayur Naik | Eric Xing

Pre-trained large language models (LMs) struggle to perform logical reasoning reliably despite advances in scale and compositionality. In this work, we tackle this challenge through the lens of symbolic programming. We propose DSR-LM, a Differentiable Symbolic Reasoning framework where pre-trained LMs govern the perception of factual knowledge, and a symbolic module performs deductive reasoning. In contrast to works that rely on hand-crafted logic rules, our differentiable symbolic reasoning framework efficiently learns weighted rules and applies semantic loss to further improve LMs. DSR-LM is scalable, interpretable, and allows easy integration of prior knowledge, thereby supporting extensive symbolic programming to robustly derive a logical conclusion. The results of our experiments suggest that DSR-LM improves the logical reasoning abilities of pre-trained language models, resulting in a significant increase in accuracy of over 20% on deductive reasoning benchmarks. Furthermore, DSR-LM outperforms a variety of competitive baselines when faced with systematic changes in sequence length.

pdf abs
B2T Connection: Serving Stability and Performance in Deep Transformers
Sho Takase | Shun Kiyono | Sosuke Kobayashi | Jun Suzuki

In the perspective of a layer normalization (LN) position, the architecture of Transformers can be categorized into two types: Post-LN and Pre-LN.Recent Transformers prefer to select Pre-LN because the training in Post-LN with deep Transformers, e.g., ten or more layers, often becomes unstable, resulting in useless models. However, in contrast, Post-LN has also consistently achieved better performance than Pre-LN in relatively shallow Transformers, e.g., six or fewer layers. This study first investigates the reason for these discrepant observations empirically and theoretically and discovers 1, the LN in Post-LN is the source of the vanishing gradient problem that mainly leads the unstable training whereas Pre-LN prevents it, and 2, Post-LN tends to preserve larger gradient norms in higher layers during the back-propagation that may lead an effective training. Exploiting the new findings, we propose a method that can equip both higher stability and effective training by a simple modification from Post-LN.We conduct experiments on a wide range of text generation tasks and demonstrate that our method outperforms Pre-LN, and stable training regardless of the shallow or deep layer settings.

pdf abs
Boosting Zero-shot Cross-lingual Retrieval by Training on Artificially Code-Switched Data
Robert Litschko | Ekaterina Artemova | Barbara Plank

Transferring information retrieval (IR) models from a high-resource language (typically English) to other languages in a zero-shot fashion has become a widely adopted approach. In this work, we show that the effectiveness of zero-shot rankers diminishes when queries and documents are present in different languages. Motivated by this, we propose to train ranking models on artificially code-switched data instead, which we generate by utilizing bilingual lexicons. To this end, we experiment with lexicons induced from (1) cross-lingual word embeddings and (2) parallel Wikipedia page titles. We use the mMARCO dataset to extensively evaluate reranking models on 36 language pairs spanning Monolingual IR (MoIR), Cross-lingual IR (CLIR), and Multilingual IR (MLIR). Our results show that code-switching can yield consistent and substantial gains of 5.1 MRR@10 in CLIR and 3.9 MRR@10 in MLIR, while maintaining stable performance in MoIR. Encouragingly, the gains are especially pronounced for distant languages (up to 2x absolute gain). We further show that our approach is robust towards the ratio of code-switched tokens and also extends to unseen languages. Our results demonstrate that training on code-switched data is a cheap and effective way of generalizing zero-shot rankers for cross-lingual and multilingual retrieval.

pdf abs
Domain-specific Attention with Distributional Signatures for Multi-Domain End-to-end Task-Oriented Dialogue
Xing Ma | Peng Zhang | Feifei Zhao

The end-to-end task-oriented dialogue system has achieved great success in recent years. Most of these dialogue systems need to accommodate multi-domain dialogue in real-world scenarios. However, due to the high cost of dialogue data annotation and the scarcity of labeled dialogue data, existing methods are difficult to extend to new domains. Therefore, it is important to use limited data to construct multi-domain dialogue systems. To solve this problem, we propose a novel domain attention module. It use the distributional signatures to construct a multi-domain dialogue system effectively with limited data, which has strong extensibility. We also define a adjacent n-gram pattern to explore potential patterns for dialogue entities. Experimental results show that our approach outperforms the baseline models on most metrics. In the few-shot scenario, we show our method get a great improvement compared with previous methods while keeping smaller model scale.

Distilling knowledge from a high-resource task, e.g., machine translation, is an effective way to alleviate the data scarcity problem of end-to-end speech translation. However, previous works simply use the classical knowledge distillation that does not allow for adequate transfer of knowledge from machine translation. In this paper, we propose a comprehensive knowledge distillation framework for speech translation, CKDST, which is capable of comprehensively and effectively distilling knowledge from machine translation to speech translation from two perspectives: cross-modal contrastive representation distillation and simultaneous decoupled knowledge distillation. In the former, we leverage a contrastive learning objective to optmize the mutual information between speech and text representations for representation distillation in the encoder. In the later, we decouple the non-target class knowledge from target class knowledge for logits distillation in the decoder. Experiments on the MuST-C benchmark dataset demonstrate that our CKDST substantially improves the baseline by 1.2 BLEU on average in all translation directions, and outperforms previous state-of-the-art end-to-end and cascaded speech translation models.

pdf abs
Follow the leader(board) with confidence: Estimating p-values from a single test set with item and response variance
Shira Wein | Christopher Homan | Lora Aroyo | Chris Welty

Among the problems with leaderboard culture in NLP has been the widespread lack of confidence estimation in reported results. In this work, we present a framework and simulator for estimating p-values for comparisons between the results of two systems, in order to understand the confidence that one is actually better (i.e. ranked higher) than the other. What has made this difficult in the past is that each system must itself be evaluated by comparison to a gold standard. We define a null hypothesis that each system’s metric scores are drawn from the same distribution, using variance found naturally (though rarely reported) in test set items and individual labels on an item (responses) to produce the metric distributions. We create a test set that evenly mixes the responses of the two systems under the assumption the null hypothesis is true. Exploring how to best estimate the true p-value from a single test set under different metrics, tests, and sampling methods, we find that the presence of response variance (from multiple raters or multiple model versions) has a profound impact on p-value estimates for model comparison, and that choice of metric and sampling method is critical to providing statistical guarantees on model comparisons.

pdf abs
Parallel Data Helps Neural Entity Coreference Resolution
Gongbo Tang | Christian Hardmeier

Coreference resolution is the task of finding expressions that refer to the same entity in a text. Coreference models are generally trained on monolingual annotated data but annotating coreference is expensive and challenging. Hardmeier et al. (2013) have shown that parallel data contains latent anaphoric knowledge, but it has not been explored in end-to-end neural models yet. In this paper, we propose a simple yet effective model to exploit coreference knowledge from parallel data. In addition to the conventional modules learning coreference from annotations, we introduce an unsupervised module to capture cross-lingual coreference knowledge. Our proposed cross-lingual model achieves consistent improvements, up to 1.74 percentage points, on the OntoNotes 5.0 English dataset using 9 different synthetic parallel datasets. These experimental results confirm that parallel data can provide additional coreference knowledge which is beneficial to coreference resolution tasks.

pdf abs
Towards Open-Domain Twitter User Profile Inference
Haoyang Wen | Zhenxin Xiao | Eduard Hovy | Alexander Hauptmann

Twitter user profile inference utilizes information from Twitter to predict user attributes (e.g., occupation, location), which is controversial because of its usefulness for downstream applications and its potential to reveal users’ privacy. Therefore, it is important for researchers to determine the extent of profiling in a safe environment to facilitate proper use and make the public aware of the potential risks. Contrary to existing approaches on limited attributes, we explore open-domain Twitter user profile inference. We conduct a case study where we collect publicly available WikiData public figure profiles and use diverse WikiData predicates for profile inference. After removing sensitive attributes, our data contains over 150K public figure profiles from WikiData, over 50 different attribute predicates, and over 700K attribute values. We further propose a prompt-based generation method, which can infer values that are implicitly mentioned in the Twitter information. Experimental results show that the generation-based approach can infer more comprehensive user profiles than baseline extraction-based methods, but limitations still remain to be applied for real-world use. We also enclose a detailed ethical statement for our data, potential benefits and risks from this work, and our efforts to mitigate the risks.

pdf abs
Eliciting Affective Events from Language Models by Multiple View Co-prompting
Yuan Zhuang | Ellen Riloff

Prior research on affective event classification showed that exploiting weakly labeled data for training can improve model performance. In this work, we propose a simpler and more effective approach for generating training data by automatically acquiring and labeling affective events with Multiple View Co-prompting, which leverages two language model prompts that provide independent views of an event. The approach starts with a modest amount of gold data and prompts pre-trained language models to generate new events. Next, information about the probable affective polarity of each event is collected from two complementary language model prompts and jointly used to assign polarity labels. Experimental results on two datasets show that the newly acquired events improve a state-of-the-art affective event classifier. We also present analyses which show that using multiple views produces polarity labels of higher quality than either view on its own.

pdf abs
ZeroAE: Pre-trained Language Model based Autoencoder for Transductive Zero-shot Text Classification
Kaihao Guo | Hang Yu | Cong Liao | Jianguo Li | Haipeng Zhang

Many text classification tasks require handling unseen domains with plenty of unlabeled data, thus giving rise to the self-adaption or the so-called transductive zero-shot learning (TZSL) problem. However, current methods based solely on encoders or decoders overlook the possibility that these two modules may promote each other. As a first effort to bridge this gap, we propose an autoencoder named ZeroAE. Specifically, the text is encoded with two separate BERT-based encoders into two disentangled spaces, i.e., label-relevant (for classification) and label-irrelevant respectively. The two latent spaces are then decoded by prompting GPT-2 to recover the text as well as to further generate text with labels in the unseen domains to train the encoder in turn. To better exploit the unlabeled data, a novel indirect uncertainty-aware sampling (IUAS) approach is proposed to train ZeroAE. Extensive experiments show that ZeroAE largely surpasses the SOTA methods by 15.93% and 8.70% on average respectively in the label-partially-unseen and label-fully-unseen scenario. Notably, the label-fully-unseen ZeroAE even possesses superior performance to the label-partially-unseen SOTA methods.

Zero-resource cross-lingual named entity recognition (ZRCL-NER) aims to leverage rich labeled source language data to address the NER problem in the zero-resource target language. Existing methods are built either based on data transfer or representation transfer. However, the former usually leads to additional computation costs, and the latter lacks explicit optimization specific to the NER task. To overcome the above limitations, we propose a novel prototype-based representation alignment model (PRAM) for the challenging ZRCL-NER task. PRAM models the cross-lingual (CL) NER task and transfers knowledge from source languages to target languages in a unified neural network, and performs end-to-end training, avoiding additional computation costs. Moreover, PRAM borrows the CL inference ability of multilingual language models and enhances it with a novel training objective—attribution-prediction consistency (APC)—for explicitly enforcing the entity-level alignment between entity representations and predictions, as well as that across languages using prototypes as bridges. The experimental results show that PRAM significantly outperforms existing state-of-the-art methods, especially in some challenging scenarios.

pdf abs
It Takes Two to Tango: Navigating Conceptualizations of NLP Tasks and Measurements of Performance
Arjun Subramonian | Xingdi Yuan | Hal Daumé III | Su Lin Blodgett

Progress in NLP is increasingly measured through benchmarks; hence, contextualizing progress requires understanding when and why practitioners may disagree about the validity of benchmarks. We develop a taxonomy of disagreement, drawing on tools from measurement modeling, and distinguish between two types of disagreement: 1) how tasks are conceptualized and 2) how measurements of model performance are operationalized. To provide evidence for our taxonomy, we conduct a meta-analysis of relevant literature to understand how NLP tasks are conceptualized, as well as a survey of practitioners about their impressions of different factors that affect benchmark validity. Our meta-analysis and survey across eight tasks, ranging from coreference resolution to question answering, uncover that tasks are generally not clearly and consistently conceptualized and benchmarks suffer from operationalization disagreements. These findings support our proposed taxonomy of disagreement. Finally, based on our taxonomy, we present a framework for constructing benchmarks and documenting their limitations.

pdf abs
Task-adaptive Label Dependency Transfer for Few-shot Named Entity Recognition
Shan Zhang | Bin Cao | Tianming Zhang | Yuqi Liu | Jing Fan

Named Entity Recognition (NER), as a crucial subtask in natural language processing (NLP), suffers from limited labeled samples (a.k.a. few-shot). Meta-learning methods are widely used for few-shot NER, but these existing methods overlook the importance of label dependency for NER, resulting in suboptimal performance. However, applying meta-learning methods to label dependency learning faces a special challenge, that is, due to the discrepancy of label sets in different domains, the label dependencies can not be transferred across domains. In this paper, we propose the Task-adaptive Label Dependency Transfer (TLDT) method to make label dependency transferable and effectively adapt to new tasks by a few samples. TLDT improves the existing optimization-based meta-learning methods by learning general initialization and individual parameter update rule for label dependency. Extensive experiments show that TLDT achieves significant improvement over the state-of-the-art methods.

pdf abs
WYWEB: A NLP Evaluation Benchmark For Classical Chinese
Bo Zhou | Qianglong Chen | Tianyu Wang | Xiaomi Zhong | Yin Zhang

To fully evaluate the overall performance of different NLP models in a given domain, many evaluation benchmarks are proposed, such as GLUE, SuperGLUE and CLUE. The field of natural language understanding has traditionally focused on benchmarks for various tasks in languages such as Chinese, English, and multilingual, however, there has been a lack of attention given to the area of classical Chinese, also known as "wen yan wen (文言文)", which has a rich history spanning thousands of years and holds significant cultural and academic value. For the prosperity of the NLP community, in this paper, we introduce the WYWEB evaluation benchmark, which consists of nine NLP tasks in classical Chinese, implementing sentence classification, sequence labeling, reading comprehension, and machine translation. We evaluate the existing pre-trained language models, which are all struggling with this benchmark. We also introduce a number of supplementary datasets and additional tools to help facilitate further progress on classical Chinese NLU. The github repository is https://github.com/baudzhou/WYWEB.

pdf abs
A Fused Gromov-Wasserstein Framework for Unsupervised Knowledge Graph Entity Alignment
Jianheng Tang | Kangfei Zhao | Jia Li

Entity alignment is the task of identifying corresponding entities across different knowledge graphs (KGs). Although recent embedding-based entity alignment methods have shown significant advancements, they still struggle to fully utilize KG structural information. In this paper, we introduce FGWEA, an unsupervised entity alignment framework that leverages the Fused Gromov-Wasserstein (FGW) distance, allowing for a comprehensive comparison of entity semantics and KG structures within a joint optimization framework. To address the computational challenges associated with optimizing FGW, we devise a three-stage progressive optimization algorithm. It starts with a basic semantic embedding matching, proceeds to approximate cross-KG structural and relational similarity matching based on iterative updates of high-confidence entity links, and ultimately culminates in a global structural comparison between KGs. We perform extensive experiments on four entity alignment datasets covering 14 distinct KGs across five languages. Without any supervision or hyper-parameter tuning, FGWEA surpasses 21 competitive baselines, including cutting-edge supervised entity alignment methods. Our code is available at https://github.com/squareRoot3/FusedGW-Entity-Alignment.

Prompting has gained tremendous attention as an efficient method for the adaptation of large-scale language models. However, prompts often act against human intuition and report unstable performances, which has motivated methods that automatically find effective prompts. One popular approach is gradient-based search, which iteratively updates a (randomly) initialized prompt towards the optimal one with the guide of gradients. We propose a novel regularization method, CoRe, for gradient-based prompt tuning techniques, which guides a prompt to produce a task context properly. CoRe realizes two regularization effects — context attuning and context filtering — that improve prediction performance in a zero-shot in-context learning setting where a model makes inferences only with the prompt tuned by CoRe, without any demonstration examples for in-context learning. Context attuning guides the context generated by the input and the tuned prompt toward embedding the appropriate context for the task. In our theoretical analysis, regularizing the context extends to improving zero-shot in-context learning performance. Context filtering steers the prompt to select only the task-related context so that context attuning solely focuses on creating and sending the right task context. We evaluate CoRe on natural language understanding datasets and two large language models, GPT2-XL and GPT-J.Our training scheme shows performance improvements up to 11.9% on GPT2-XL, and up to 6.3% on GPT-J in zero-shot settings.

pdf abs
An Investigation of Noise in Morphological Inflection
Adam Wiemerslage | Changbing Yang | Garrett Nicolai | Miikka Silfverberg | Katharina Kann

With a growing focus on morphological inflection systems for languages where high-quality data is scarce, training data noise is a serious but so far largely ignored concern. We aim at closing this gap by investigating the types of noise encountered within a pipeline for truly unsupervised morphological paradigm completion and its impact on morphological inflection systems: First, we propose an error taxonomy and annotation pipeline for inflection training data. Then, we compare the effect of different types of noise on multiple state-of-the- art inflection models. Finally, we propose a novel character-level masked language modeling (CMLM) pretraining objective and explore its impact on the models’ resistance to noise. Our experiments show that various architectures are impacted differently by separate types of noise, but encoder-decoders tend to be more robust to noise than models trained with a copy bias. CMLM pretraining helps transformers, but has lower impact on LSTMs.

Answering complex questions often requires reasoning over knowledge graphs (KGs). State-of-the-art methods often utilize entities in questions to retrieve local subgraphs, which are then fed into KG encoder, e.g. graph neural networks (GNNs), to model their local structures and integrated into language models for question answering. However, this paradigm constrains retrieved knowledge in local subgraphs and discards more diverse triplets buried in KGs that are disconnected but useful for question answering. In this paper, we propose a simple yet effective method to first retrieve the most relevant triplets from KGs and then rerank them, which are then concatenated with questions to be fed into language models. Extensive results on both CommonsenseQA and OpenbookQA datasets show that our method can outperform state-of-the-art up to 4.6% absolute accuracy.

pdf abs
End-to-End Argument Mining over Varying Rhetorical Structures
Elena Chistova

Rhetorical Structure Theory implies no single discourse interpretation of a text, and the limitations of RST parsers further exacerbate inconsistent parsing of similar structures. Therefore, it is important to take into account that the same argumentative structure can be found in semantically similar texts with varying rhetorical structures. In this work, the differences between paraphrases within the same argument scheme are evaluated from a rhetorical perspective. The study proposes a deep dependency parsing model to assess the connection between rhetorical and argument structures. The model utilizes rhetorical relations; RST structures of paraphrases serve as training data augmentations. The method allows for end-to-end argumentation analysis using a rhetorical tree instead of a word sequence. It is evaluated on the bilingual Microtexts corpus, and the first results on fully-fledged argument parsing for the Russian version of the corpus are reported. The results suggest that argument mining can benefit from multiple variants of discourse structure.

pdf abs
Unsupervised Task Graph Generation from Instructional Video Transcripts
Lajanugen Logeswaran | Sungryull Sohn | Yunseok Jang | Moontae Lee | Honglak Lee

This work explores the problem of generating task graphs of real-world activities. Different from prior formulations, we consider a setting where text transcripts of instructional videos performing a real-world activity (e.g., making coffee) are provided and the goal is to identify the key steps relevant to the task as well as the dependency relationship between these key steps. We propose a novel task graph generation approach that combines the reasoning capabilities of instruction-tuned language models along with clustering and ranking components to generate accurate task graphs in a completely unsupervised manner. We show that the proposed approach generates more accurate task graphs compared to a supervised learning approach on tasks from the ProceL and CrossTask datasets.

pdf abs
Exploiting Hierarchically Structured Categories in Fine-grained Chinese Named Entity Recognition
Jiuding Yang | Jinwen Luo | Weidong Guo | Di Niu | Yu Xu

Chinese Named Entity Recognition (CNER) is a widely used technology in various applications. While recent studies have focused on utilizing additional information of the Chinese language and characters to enhance CNER performance, this paper focuses on a specific aspect of CNER known as fine-grained CNER (FG-CNER). FG-CNER involves the use of hierarchical, fine-grained categories (e.g. Person-MovieStar) to label named entities. To promote research in this area, we introduce the FiNE dataset, a dataset for FG-CNER consisting of 30,000 sentences from various domains and containing 67,651 entities in 54 fine-grained flattened hierarchical categories. Additionally, we propose SoftFiNE, a novel approach for FG-CNER that utilizes a custom-designed relevance scoring function based on label structures to learn the potential relevance between different flattened hierarchical labels. Our experimental results demonstrate that the proposed SoftFiNE method outperforms the state-of-the-art baselines on the FiNE dataset. Furthermore, we conduct extensive experiments on three other datasets, including OntoNotes 4.0, Weibo, and Resume, where SoftFiNE achieved state-of-the-art performance on all three datasets.

pdf abs
Adversarial Textual Robustness on Visual Dialog
Lu Yu | Verena Rieser

Adversarial robustness evaluates the worst-case performance scenario of a machine learning model to ensure its safety and reliability. For example, cases where the user input contains a minimal change, e.g. a synonym, which causes the previously correct model to return a wrong answer. Using this scenario, this study is the first to investigate the robustness of visually grounded dialog models towards textual attacks. We first aim to understand how multimodal input components contribute to model robustness. Our results show that models which encode dialog history are more robust by providing redundant information. This is in contrast to prior work which finds that dialog history is negligible for model performance on this task. We also evaluate how to generate adversarial test examples which successfully fool the model but remain undetected by the user/software designer. Our analysis shows that the textual, as well as the visual context are important to generate plausible attacks.

pdf abs
Language Model Analysis for Ontology Subsumption Inference
Yuan He | Jiaoyan Chen | Ernesto Jimenez-Ruiz | Hang Dong | Ian Horrocks

Investigating whether pre-trained language models (LMs) can function as knowledge bases (KBs) has raised wide research interests recently. However, existing works focus on simple, triple-based, relational KBs, but omit more sophisticated, logic-based, conceptualised KBs such as OWL ontologies. To investigate an LM’s knowledge of ontologies, we propose OntoLAMA, a set of inference-based probing tasks and datasets from ontology subsumption axioms involving both atomic and complex concepts. We conduct extensive experiments on ontologies of different domains and scales, and our results demonstrate that LMs encode relatively less background knowledge of Subsumption Inference (SI) than traditional Natural Language Inference (NLI) but can improve on SI significantly when a small number of samples are given. We will open-source our code and datasets.

pdf abs
Exploring Automatically Perturbed Natural Language Explanations in Relation Extraction
Wanyun Cui | Xingran Chen

Previous research has demonstrated that natural language explanations provide valuable inductive biases that guide models, thereby improving the generalization ability and data efficiency. In this paper, we undertake a systematic examination of the effectiveness of these explanations. Remarkably, we find that corrupted explanations with diminished inductive biases can achieve competitive or superior performance compared to the original explanations. Our findings furnish novel insights into the characteristics of natural language explanations in the following ways: (1) the impact of explanations varies across different training styles and datasets, with previously believed improvements primarily observed in frozen language models. (2) While previous research has attributed the effect of explanations solely to their inductive biases, our study shows that the effect persists even when the explanations are completely corrupted. We propose that the main effect is due to the provision of additional context space. (3) Utilizing the proposed automatic perturbed context, we were able to attain comparable results to annotated explanations, but with a significant increase in computational efficiency, 20-30 times faster.

pdf abs
Varta: A Large-Scale Headline-Generation Dataset for Indic Languages
Rahul Aralikatte | Ziling Cheng | Sumanth Doddapaneni | Jackie Chi Kit Cheung

We present Varta, a large-scale multilingual dataset for headline generation in Indic languages. This dataset includes more than 41 million pairs of headlines and articles in 14 different Indic languages (and English), which come from a variety of high-quality news sources. To the best of our knowledge, this is the largest collection of curated news articles for Indic languages currently available. We use the collected data in a series of experiments to answer important questions related to Indic NLP and multilinguality research in general. We show that the dataset is challenging even for state-of-the-art abstractive models and that they perform only slightly better than extractive baselines. Owing to its size, we also show that the dataset can be used to pre-train strong language models that outperform competitive baselines in both NLU and NLG benchmarks.

pdf abs
Better Zero-Shot Reasoning with Self-Adaptive Prompting
Xingchen Wan | Ruoxi Sun | Hanjun Dai | Sercan Arik | Tomas Pfister

Modern large language models (LLMs) have demonstrated impressive capabilities at sophisticated tasks, often through step-by-step reasoning similar to humans. This is made possible by their strong few- and zero-shot abilities – they can effectively learn from a handful of handcrafted, completed responses (“in-context examples”), or are prompted to reason spontaneously through specially designed triggers. Nonetheless, some limitations have been observed. First, performance in the few-shot setting is sensitive to the choice of the examples, whose design requires significant human effort. Moreover, given the diverse downstream tasks of LLMs, it may be difficult or laborious to handcraft per-task labels. Second, while the zero-shot setting does not require handcrafting, its performance is limited due to the lack of guidance to the LLMs. To address these limitations, we propose Consistency-based Self-adaptive Prompting (COSP), a novel prompt design method for LLMs. Requiring neither handcrafted responses nor ground-truth labels, COSP selects and builds the set of examples from the LLM zero-shot outputs via carefully designed criteria combining consistency, diversity and repetition. In the zero-shot setting for three different LLMs, we show that using only LLM predictions, COSP significantly improves performance up to 15% compared to zero-shot baselines and matches or exceeds few-shot baselines at a range of reasoning tasks.

Existing multimodal task-oriented dialog data fails to demonstrate the diverse expressions of user subjective preferences and recommendation acts in the real-life shopping scenario. This paper introduces a new dataset SURE (Multimodal Recommendation Dialog with Subjective Preference), which contains 12K shopping dialogs in complex store scenes. The data is built in two phases with human annotations to ensure quality and diversity. SURE is well-annotated with subjective preferences and recommendation acts proposed by sales experts. A comprehensive analysis is given to reveal the distinguishing features of SURE. Three benchmark tasks are then proposed on the data to evaluate the capability of multimodal recommendation agents. Basing on the SURE, we propose a baseline model, powered by a state-of-the-art multimodal model, for these tasks.

Over the past decade, analogies, in the form of word-level analogies, have played a significant role as an intrinsic measure of evaluating the quality of word embedding methods such as word2vec. Modern large language models (LLMs), however, are primarily evaluated on extrinsic measures based on benchmarks such as GLUE and SuperGLUE, and there are only a few investigations on whether LLMs can draw analogies between long texts. In this paper, we present ANALOGICAL, a new benchmark to intrinsically evaluate LLMs across a taxonomy of analogies of long text with six levels of complexity – (i) word, (ii) word vs. sentence, (iii) syntactic, (iv) negation, (v) entailment, and (vi) metaphor. Using thirteen datasets and three different distance measures, we evaluate the abilities of eight LLMs in identifying analogical pairs in the semantic vector space. Our evaluation finds that it is increasingly challenging for LLMs to identify analogies when going up the analogy taxonomy.

The U.S. Securities and Exchange Commission (SEC) mandates all public companies to file periodic financial statements that should contain numerals annotated with a particular label from a taxonomy. In this paper, we formulate the task of automating the assignment of a label to a particular numeral span in a sentence from an extremely large label set. Towards this task, we release a dataset, Financial Numeric Extreme Labelling (FNXL), annotated with 2,794 labels. We benchmark the performance of the FNXL dataset by formulating the task as (a) a sequence labelling problem and (b) a pipeline with span extraction followed by Extreme Classification. Although the two approaches perform comparably, the pipeline solution provides a slight edge for the least frequent labels.

Abstractive summarization has enjoyed renewed interest in recent years, thanks to pre-trained language models and the availability of large-scale datasets. Despite promising results, current models still suffer from generating factually inconsistent summaries, reducing their utility for real-world application. Several recent efforts attempt to address this by devising models that automatically detect factual inconsistencies in machine generated summaries. However, they focus exclusively on English, a language with abundant resources. In this work, we leverage factual consistency evaluation models to improve multilingual summarization. We explore two intuitive approaches to mitigate hallucinations based on the signal provided by a multilingual NLI model, namely data filtering and controlled generation. Experimental results in the 45 languages from the XLSum dataset show gains over strong baselines in both automatic and human evaluation. We release models and human judgements of summaries to foster progress towards more factually consistent multilingual summarization.

pdf abs
Enhancing Out-of-Vocabulary Estimation with Subword Attention
Raj Patel | Carlotta Domeniconi

Word embedding methods like word2vec and GloVe have been shown to learn strong representations of words. However, these methods only learn representations for words in the training corpus and therefore struggle to handle unknown and new words, known as out-of-vocabulary (OOV) words. As a result, there have been multiple attempts to learn OOV word representations in a similar fashion to how humans learn new words, using word roots/subwords and/or surrounding words. However, while most of these approaches use advanced architectures like attention on the context of the OOV word, they tend to use simple structures like ngram addition or character based convolutional neural networks (CNN) to handle processing subword information. In response to this, we propose SubAtt, a transformer based OOV estimation model that uses attention mechanisms on both the context and the subwords. In addition to attention, we also show that pretraining subword representations also leads to improvement in OOV estimation. We show SubAtt outperforms current state-of-the-art OOV estimation models.

pdf abs
Encoder and Decoder, Not One Less for Pre-trained Language Model Sponsored NMT
Sufeng Duan | Hai Zhao

Well pre-trained contextualized representations from pre-trained language model (PLM) have been shown helpful for enhancing various natural language processing tasks, surely including neural machine translation (NMT). However, existing methods either consider encoder-only enhancement or rely on specific multilingual PLMs, which leads to a much larger model or give up potentially helpful knowledge from target PLMs. In this paper, we propose a new monolingual PLM-sponsored NMT model to let both encoder and decoder enjoy PLM enhancement to alleviate such obvious inconvenience. Especially, incorporating a newly proposed frequency-weighted embedding transformation algorithm, PLM embeddings can be effectively exploited in terms of the representations of the NMT decoder. We evaluate our model on IWSLT14 En-De, De-En, WMT14 En-De, and En-Fr tasks, and the results show that our proposed PLM enhancement gives significant improvement and even helps achieve new state-of-the-art.

Data augmentation is an effective way to improve model performance of grammatical error correction (GEC). This paper identifies a critical side-effect of GEC data augmentation, which is due to the style discrepancy between the data used in GEC tasks (i.e., texts produced by non-native speakers) and data augmentation (i.e., native texts). To alleviate this issue, we propose to use an alternative data source, translationese (i.e., human-translated texts), as input for GEC data augmentation, which 1) is easier to obtain and usually has better quality than non-native texts, and 2) has a more similar style to non-native texts. Experimental results on the CoNLL14 and BEA19 English, NLPCC18 Chinese, Falko-MERLIN German, and RULEC-GEC Russian GEC benchmarks show that our approach consistently improves correction accuracy over strong baselines. Further analyses reveal that our approach is helpful for overcoming mainstream correction difficulties such as the corrections of frequent words, missing words, and substitution errors. Data, code, models and scripts are freely available at https://github.com/NLP2CT/TransGEC.

Hot news is one of the most popular topics in daily conversations. However, news grounded conversation has long been stymied by the lack of well-designed task definition and scarce data. In this paper, we propose a novel task, Proactive News Grounded Conversation, in which a dialogue system can proactively lead the conversation based on some key topics of the news. In addition, both information-seeking and chit-chat scenarios are included realistically, where the user may ask a series of questions about the news details or express their opinions and be eager to chat. To further develop this novel task, we collect a human-to-human Chinese dialogue dataset NewsDialogues, which includes 1K conversations with a total of 14.6K utterances and detailed annotations for target topics and knowledge spans. Furthermore, we propose a method named Predict-Generate-Rank, consisting of a generator for grounded knowledge prediction and response generation, and a ranker for the ranking of multiple responses to alleviate the exposure bias. We conduct comprehensive experiments to demonstrate the effectiveness of the proposed method and further present several key findings and challenges to prompt future research.

We study the problem of retrieval with instructions, where users provide explicit descriptions of their intent along with their queries to guide a retrieval system. Our solution is a general-purpose task-aware retrieval system, trained using multi-task instruction tuning and can follow human-written instructions to find relevant documents to a given query. We introduce the first large-scale collection of 37 retrieval datasets with instructions, BERRI, and present TART, a single multi-task retrieval system trained on BERRI with instructions that can adapt to a new task without any parameter updates. TART advances the state of the art on two zero-shot retrieval benchmarks, BEIR and LOTTE, outperforming models up to three times larger. We further introduce a new evaluation setup, X2-Retrieval, to better reflect real-world scenarios in which diverse domains and tasks are pooled. TART significantly outperforms competitive baselines in this setup, further highlighting the effectiveness of guiding retrieval with instructions.

pdf abs
Non-Repeatable Experiments and Non-Reproducible Results: The Reproducibility Crisis in Human Evaluation in NLP
Anya Belz | Craig Thomson | Ehud Reiter | Simon Mille

Human evaluation is widely regarded as the litmus test of quality in NLP. A basic requirementof all evaluations, but in particular where they are used for meta-evaluation, is that they should support the same conclusions if repeated. However, the reproducibility of human evaluations is virtually never queried, let alone formally tested, in NLP which means that their repeatability and the reproducibility of their results is currently an open question. This focused contribution reports our review of human evaluation experiments reported in NLP papers over the past five years which we assessed in terms oftheir ability to be rerun. Overall, we estimatethat just 5% of human evaluations are repeatable in the sense that (i) there are no prohibitivebarriers to repetition, and (ii) sufficient information about experimental design is publicly available for rerunning them. Our estimate goesup to about 20% when author help is sought. We complement this investigation with a survey of results concerning the reproducibilityof human evaluations where those are repeatable in the first place. Here we find worryinglylow degrees of reproducibility, both in terms ofsimilarity of scores and of findings supportedby them. We summarise what insights can begleaned so far regarding how to make humanevaluations in NLP more repeatable and morereproducible.

pdf abs
Define, Evaluate, and Improve Task-Oriented Cognitive Capabilities for Instruction Generation Models
Lingjun Zhao | Khanh Nguyen | Hal Daumé III

Recent work studies the cognitive capabilities of language models through psychological tests designed for humans. While these studies are helpful for understanding the general capabilities of these models, there is no guarantee that a model possessing sufficient capabilities to pass those tests would actually use those capabilities in performing real-life tasks. In this work, we formulate task-oriented cognitive capabilities, which are human-like cognitive capabilities that language models leverage to perform tasks. These capabilities are (i) the ability to quickly generate good candidate utterances (the search capability) (ii) the ability to predict how a listener interprets those utterances and choose the most appropriate one (the pragmatic capability). We design an evaluation scheme for comparing these capabilities of a language model with those of a human. Applying this scheme to examine various models in a navigation instruction generation problem, we find that their pragmatic capability is severely lacking. This insight leads us to augment them with better models of the listener and obtain a significant boost of 11% in success rate in guiding real humans. Our work advocates for having a principled procedure for aligning language models with humans that involves (i) formulating task-oriented capabilities, (ii) devising a method to quantify their deficiency, and (iii) iteratively improving them.

pdf abs
Robustness of Multi-Source MT to Transcription Errors
Dominik Macháček | Peter Polák | Ondřej Bojar | Raj Dabre

Automatic speech translation is sensitive to speech recognition errors, but in a multilingual scenario, the same content may be available in various languages via simultaneous interpreting, dubbing or subtitling. In this paper, we hypothesize that leveraging multiple sources will improve translation quality if the sources complement one another in terms of correct information they contain. To this end, we first show that on a 10-hour ESIC corpus, the ASR errors in the original English speech and its simultaneous interpreting into German and Czech are mutually independent. We then use two sources, English and German, in a multi-source setting for translation into Czech to establish its robustness to ASR errors. Furthermore, we observe this robustness when translating both noisy sources together in a simultaneous translation setting. Our results show that multi-source neural machine translation has the potential to be useful in a real-time simultaneous translation setting, thereby motivating further investigation in this area.

pdf abs
Not The End of Story: An Evaluation of ChatGPT-Driven Vulnerability Description Mappings
Xin Liu | Yuan Tan | Zhenghang Xiao | Jianwei Zhuge | Rui Zhou

As the number of vulnerabilities increases day by day, security management requires more and more structured data. In addition to textual descriptions of vulnerabilities, security engineers must classify and assess vulnerabilities and clarify their associated techniques. Vulnerability Description Mapping (VDM) refers to mapping vulnerabilities to Common Weakness Enumeration (CWE), Common Attack Pattern Enumeration and Classification, ATT&CK Techniques, and other classifications. Accurate VDM is necessary to reduce the pressure of security management and improve the speed of security emergency response. ChatGPT is the latest state-of-the-art closed-source conversational large language model (LLM), which performs excellently on many tasks. This paper explores the application of closed-source LLMs to real-world security management scenarios by evaluating ChatGPT’s performance on VDM tasks. The results show that although ChatGPT may be close to the level of human experts on some tasks, it still cannot replace the critical role of professional security engineers in vulnerability analysis. In a word, closed-source LLM is not the end of story.

Task-oriented dialogue (ToD) systems have been widely deployed in many industries as they deliver more efficient customer support. These systems are typically constructed for a single domain or language and do not generalise well beyond this. To support work on Natural Language Understanding (NLU) in ToD across multiple languages and domains simultaneously, we constructed Multi3NLU++, a multilingual, multi-intent, multi-domain dataset. Multi3NLU++ extends the English-only NLU++ dataset to include manual translations into a range of high, medium, and low resource languages (Spanish, Marathi, Turkish and Amharic), in two domains (banking and hotels). Because of its multi-intent property, Multi3NLU++ represents complex and natural user goals, and therefore allows us to measure the realistic performance of ToD systems in a varied set of the world’s languages. We use Multi3NLU++ to benchmark state-of-the-art multilingual models for the NLU tasks of intent detection and slot labeling for ToD systems in the multilingual setting. The results demonstrate the challenging nature of the dataset, particularly in the low-resource language setting, offering ample room for future experimentation in multi-domain multilingual ToD setups.

pdf abs
A Robust Information-Masking Approach for Domain Counterfactual Generation
Pengfei Hong | Rishabh Bhardwaj | Navonil Majumder | Somak Aditya | Soujanya Poria

Domain shift is a big challenge in NLP. Many approaches, thus, resort to learning domain-invariant features to mitigate the hurdles of domain shift during inference. Such methods, however, inexorably fail to leverage the domain-specific nuances relevant to the task at hand. To avoid such drawbacks, domain counterfactual generation has recently been proposed that aims to transform a text from the source domain to a given target domain. To achieve this, the existing method uses a frequency-based approach to identify and mask the source-domain-specific tokens in a text. A pretrained LM is then prompted to fill the masks with target-domain-specific tokens. We, however, have observed that, due to limitations of the available data, such a frequency-based method may either miss some domain-token associations or lead to some spurious domain-token associations. To this end, we additionally employ attention norm-based scores to identify additional token-domain associations from a domain classifier. To minimize spurious associations, we also devise an iterative unmasking heuristic that unmasks the masked tokens to minimize the confidence of a domain classifier in the source domain. Our experiments empirically show that the counterfactual samples sourced from our masked text lead to improved domain transfer across various classification tasks. The proposed approach outperforms the baselines on 10 out of 12 domain-counterfactual classification settings with an average of 1.7% improvement in accuracy metric.

pdf abs
Misleading Relation Classifiers by Substituting Words in Texts
Tian Jiang | Yunqi Liu | Yan Feng | Yuqing Li | Xiaohui Cui

Relation classification is to determine the semantic relationship between two entities in a given sentence. However, many relation classifiers are vulnerable to adversarial attacks, which is using adversarial examples to lead victim models to output wrong results. In this paper, we propose a simple but effective method for misleading relation classifiers. We first analyze the most important parts of speech (POSs) from the syntax and morphology perspectives, then we substitute words labeled with these POS tags in original samples with synonyms or hyponyms. Experimental results show that our method can generate adversarial texts of high quality, and most of the relationships between entities can be correctly identified in the process of human evaluation. Furthermore, the adversarial examples generated by our method possess promising transferability, and they are also helpful for improving the robustness of victim models.

Given a data lake of tabular data as well as a query table, how can we retrieve all the tables in the data lake that can be unioned with the query table? Table union search constitutes an essential task in data discovery and preparation as it enables data scientists to navigate massive open data repositories. Existing methods identify uniability based on column representations (word surface forms or token embeddings) and column relation represented by column representation similarity. However, the semantic similarity obtained between column representations is often insufficient to reveal latent relational features to describe the column relation between pair of columns and not robust to the table noise. To address these issues, in this paper, we propose a multi-stage self-supervised table union search framework called AutoTUS, which represents column relation as a vector– column relational representation and learn column relational representation in a multi-stage manner that can better describe column relation for unionability prediction. In particular, the large language model powered contextualized column relation encoder is updated by adaptive clustering and pseudo label classification iteratively so that the better column relational representation can be learned. Moreover, to improve the robustness of the model against table noises, we propose table noise generator to add table noise to the training table data. Experiments on real-world datasets as well as synthetic test set augmented with table noise show that AutoTUS achieves 5.2% performance gain over the SOTA baseline.

pdf abs
Bidirectional Transformer Reranker for Grammatical Error Correction
Ying Zhang | Hidetaka Kamigaito | Manabu Okumura

Pre-trained seq2seq models have achieved state-of-the-art results in the grammatical error correction task. However, these models still suffer from a prediction bias due to their unidirectional decoding. Thus, we propose a bidirectional Transformer reranker (BTR), that re-estimates the probability of each candidate sentence generated by the pre-trained seq2seq model. The BTR preserves the seq2seq-style Transformer architecture but utilizes a BERT-style self-attention mechanism in the decoder to compute the probability of each target token by using masked language modeling to capture bidirectional representations from the target context. For guiding the reranking, the BTR adopts negative sampling in the objective function to minimize the unlikelihood. During inference, the BTR gives final results after comparing the reranked top-1 results with the original ones by an acceptance threshold. Experimental results show that, in reranking candidates from a pre-trained seq2seq model, T5-base, the BTR on top of T5-base could yield 65.47 and 71.27 F0.5 scores on the CoNLL-14 and BEA test sets, respectively, and yield 59.52 GLEU score on the JFLEG corpus, with improvements of 0.36, 0.76 and 0.48 points compared with the original T5-base. Furthermore, when reranking candidates from T5-large, the BTR on top of T5-base improved the original T5-large by 0.26 points on the BEA test set.

pdf abs
Not Enough Data to Pre-train Your Language Model? MT to the Rescue!
Gorka Urbizu | Iñaki San Vicente | Xabier Saralegi | Ander Corral

In recent years, pre-trained transformer-based language models (LM) have become a key resource for implementing most NLP tasks. However, pre-training such models demands large text collections not available in most languages. In this paper, we study the use of machine-translated corpora for pre-training LMs. We answer the following research questions: RQ1: Is MT-based data an alternative to real data for learning a LM?; RQ2: Can real data be complemented with translated data and improve the resulting LM? In order to validate these two questions, several BERT models for Basque have been trained, combining real data and synthetic data translated from Spanish.The evaluation carried out on 9 NLU tasks indicates that models trained exclusively on translated data offer competitive results. Furthermore, models trained with real data can be improved with synthetic data, although further research is needed on the matter.

Summarization quality evaluation is a non-trivial task in text summarization. Contemporary methods can be mainly categorized into two scenarios: (1) reference-based: evaluating with human-labeled reference summary; (2) reference-free: evaluating the summary consistency of the document. Recent studies mainly focus on one of these scenarios and explore training neural models built on PLMs to align with human criteria. However, the models from different scenarios are optimized individually, which may result in sub-optimal performance since they neglect the shared knowledge across different scenarios. Besides, designing individual models for each scenario caused inconvenience to the user. Inspired by this, we propose Unified Multi-scenario Summarization Evaluation Model (UMSE). More specifically, we propose a perturbed prefix tuning method to share cross-scenario knowledge between scenarios and use a self-supervised training paradigm to optimize the model without extra human labeling. Our UMSE is the first unified summarization evaluation framework engaged with the ability to be used in three evaluation scenarios. Experimental results across three typical scenarios on the benchmark dataset SummEval indicate that our UMSE can achieve comparable performance with several existing strong methods which are specifically designed for each scenario.

pdf abs
Maximum Entropy Loss, the Silver Bullet Targeting Backdoor Attacks in Pre-trained Language Models
Zhengxiao Liu | Bowen Shen | Zheng Lin | Fali Wang | Weiping Wang

Pre-trained language model (PLM) can be stealthily misled to target outputs by backdoor attacks when encountering poisoned samples, without performance degradation on clean samples. The stealthiness of backdoor attacks is commonly attained through minimal cross-entropy loss fine-tuning on a union of poisoned and clean samples. Existing defense paradigms provide a workaround by detecting and removing poisoned samples at pre-training or inference time. On the contrary, we provide a new perspective where the backdoor attack is directly reversed. Specifically, maximum entropy loss is incorporated in training to neutralize the minimal cross-entropy loss fine-tuning on poisoned data. We defend against a range of backdoor attacks on classification tasks and significantly lower the attack success rate. In extension, we explore the relationship between intended backdoor attacks and unintended dataset bias, and demonstrate the feasibility of the maximum entropy principle in de-biasing.

pdf abs
Improving Named Entity Recognition via Bridge-based Domain Adaptation
Jingyun Xu | Changmeng Zheng | Yi Cai | Tat-Seng Chua

Recent studies have shown remarkable success in cross-domain named entity recognition (cross-domain NER). Despite the promising results, existing methods mainly utilize pre-training language models like BERT to represent words. As such, the original chaotic representations may challenge them to distinguish entity types of entities, leading to entity type misclassification. To this end, we attempt to utilize contrastive learning to refine the original representations and propose a model-agnostic framework named MoCL for cross-domain NER. Additionally, we respectively combine MoCL with two distinctive cross-domain NER methods and two pre-training language models to explore its generalization ability. Empirical results on seven domains show the effectiveness and good generalization ability of MoCL.

Distantly-Supervised Named Entity Recognition effectively alleviates the burden of time-consuming and expensive annotation in the supervised setting. But the context-free matching process and the limited coverage of knowledge bases introduce inaccurate and incomplete annotation noise respectively. Previous studies either considered only incomplete one or indiscriminately handle two types of noise with the same strategy. In this paper, we argue that the different causes of two types of noise bring up the requirement of different strategies in model architecture. Therefore, we propose the SANTA to handle these two types of noise separately with (1) Memory-smoothed Focal Loss and Entity-aware KNN to relieve the entity ambiguity problem caused by inaccurate annotation, and (2) Boundary Mixup to alleviate decision boundary shifting problem caused by incomplete annotation and a noise-tolerant loss to improve the model’s robustness. Benefiting from our separate tailored strategies, we confirm in the experiment that the two types of noise are well mitigated.SANTA also achieves a new state-of-the-art on five public datasets.

pdf abs
The State of Profanity Obfuscation in Natural Language Processing Scientific Publications
Debora Nozza | Dirk Hovy

Work on hate speech has made considering rude and harmful examples in scientific publications inevitable. This situation raises various problems, such as whether or not to obscure profanities. While science must accurately disclose what it does, the unwarranted spread of hate speech can harm readers and increases its internet frequency. While maintaining publications’ professional appearance, obfuscating profanities makes it challenging to evaluate the content, especially for non-native speakers. Surveying 150 ACL papers, we discovered that obfuscation is usually used for English but not other languages, and even then, quite unevenly. We discuss the problems with obfuscation and suggest a multilingual community resource called PrOf with a Python module to standardize profanity obfuscation processes. We believe PrOf can help scientific publication policies to make hate speech work accessible and comparable, irrespective of language.

pdf abs
Teacher and Student Models of Offensive Language in Social Media
Tharindu Ranasinghe | Marcos Zampieri

State-of-the-art approaches to identifying offensive language online make use of large pre-trained transformer models. However, the inference time, disk, and memory requirements of these transformer models present challenges for their wide usage in the real world. Even the distilled transformer models remain prohibitively large for many usage scenarios. To cope with these challenges, in this paper, we propose transferring knowledge from transformer models to much smaller neural models to make predictions at the token- and at the post-level. We show that this approach leads to lightweight offensive language identification models that perform on par with large transformers but with 100 times fewer parameters and much less memory usage

pdf abs
A Simple Yet Strong Domain-Agnostic De-bias Method for Zero-Shot Sentiment Classification
Yang Zhao | Tetsuya Nasukawa | Masayasu Muraoka | Bishwaranjan Bhattacharjee

Zero-shot prompt-based learning has made much progress in sentiment analysis, and considerable effort has been dedicated to designing high-performing prompt templates. However, two problems exist; First, large language models are often biased to their pre-training data, leading to poor performance in prompt templates that models have rarely seen. Second, in order to adapt to different domains, re-designing prompt templates is usually required, which is time-consuming and inefficient. To remedy both shortcomings, we propose a simple yet strong data construction method to de-bias a given prompt template, yielding a large performance improvement in sentiment analysis tasks across different domains, pre-trained language models, and prompt templates. Also, we demonstrate the advantage of using domain-agnostic generic responses over the in-domain ground-truth data.

pdf abs
Balancing the Effect of Training Dataset Distribution of Multiple Styles for Multi-Style Text Transfer
Debarati Das | David Ma | Dongyeop Kang

Text style transfer is an exciting task within the field of natural language generation that is often plagued by the need for high-quality paired datasets. Furthermore, training a model for multi-attribute text style transfer requires datasets with sufficient support across all combinations of the considered stylistic attributes, adding to the challenges of training a style transfer model. This paper explores the impact of training data input diversity on the quality of the generated text from the multi-style transfer model. We construct a pseudo-parallel dataset by devising heuristics to adjust the style distribution in the training samples. We balance our training dataset using marginal and joint distributions to train our style transfer models. We observe that a balanced dataset produces more effective control effects over multiple styles than an imbalanced or skewed one. Through quantitative analysis, we explore the impact of multiple style distributions in training data on style-transferred output. These findings will better inform the design of style-transfer datasets.

pdf abs
A Benchmark on Extremely Weakly Supervised Text Classification: Reconcile Seed Matching and Prompting Approaches
Zihan Wang | Tianle Wang | Dheeraj Mekala | Jingbo Shang

Extremely Weakly Supervised Text Classification (XWS-TC) refers to text classification based on minimal high-level human guidance, such as a few label-indicative seed words or classification instructions. There are two mainstream approaches for XWS-TC, however, never being rigorously compared: (1) training classifiers based on pseudo-labels generated by (softly) matching seed words (Seed) and (2) prompting (and calibrating) language models using classification instruction (and raw texts) to decode label words (Prompt). This paper presents the first XWS-TC benchmark to compare the two approaches on fair grounds, where the datasets, supervisions, and hyperparameter choices are standardized across methods. Our benchmarking results suggest that (1) Both Seed and Prompt approaches are competitive and there is no clear winner; (2) Seed is empirically more tolerant than Prompt to human guidance (e.g., seed words, classification instructions, and label words) changes; (3) Seed is empirically more selective than Prompt to the pre-trained language models; (4) Recent Seed and Prompt methods have close connections and a clustering post-processing step based on raw in-domain texts is a strong performance booster to both. We hope this benchmark serves as a guideline in selecting XWS-TC methods in different scenarios and stimulate interest in developing guidance- and model-robust XWS-TC methods.

pdf abs
Ambiguity Meets Uncertainty: Investigating Uncertainty Estimation for Word Sense Disambiguation
Zhu Liu | Ying Liu

Word sense disambiguation (WSD), which aims to determine an appropriate sense for a target word given its context, is crucial for natural language understanding. Existing supervised methods treat WSD as a classification task and have achieved remarkable performance. However, they ignore uncertainty estimation (UE) in the real-world setting, where the data is always noisy and out of distribution. This paper extensively studies UE on the benchmark designed for WSD. Specifically, we first compare four uncertainty scores for a state-of-the-art WSD model and verify that the conventional predictive probabilities obtained at the end of the model are inadequate to quantify uncertainty. Then, we examine the capability of capturing data and model uncertainties by the model with the selected UE score on well-designed test scenarios and discover that the model reflects data uncertainty satisfactorily but underestimates model uncertainty. Furthermore, we explore numerous lexical properties that intrinsically affect data uncertainty and provide a detailed analysis of four critical aspects: the syntactic category, morphology, sense granularity, and semantic relations.

Although large language models have exhibited impressive zero-shot ability, the huge model size generally incurs high cost. Recently, semi-parametric language models, which augment a smaller language model with retrieved related background knowledge, alleviate the need for storing everything into the model parameters. Although existing semi-parametric language models have demonstrated promising language modeling capabilities, it remains unclear whether they can exhibit competitive zero-shot abilities as their fully-parametric counterparts. In this work, we introduce Zemi, a semi-parametric language model for zero-shot task generalization. To our best knowledge, this is the first semi-parametric language model that can demonstrate strong zero-shot performance on a wide range of held-out unseen tasks. We train Zemi with semi-parametric multitask training, which shows significant improvement compared with the parametric multitask training as proposed by T0. Specifically, during both training and inference, Zemi is equipped with a retrieval system based on the unlabeled pretraining corpus of our backbone model. To address the unique challenges from large-scale retrieval, we further propose a novel retrieval-augmentation fusion module that can effectively incorporate noisy retrieved documents. Finally, we show detailed analysis and ablation studies on the key ingredients towards building effective zero-shot semi-parametric language models. Notably, our proposed Zemi_Large model outperforms T0-3B by 16% across seven diverse evaluation tasks while being 3.8x smaller in scale.

Large pretrained language models have shown surprising in-context learning (ICL) ability. With a few demonstration input-label pairs, they can predict the label for an unseen input without parameter updates. Despite the great success in performance, its working mechanism still remains an open question. In this paper, we explain language models as meta-optimizers and understand in-context learning as implicit finetuning. Theoretically, we figure out that Transformer attention has a dual form of gradient descent. On top of it, we understand ICL as follows: GPT first produces meta-gradients according to the demonstration examples, and then these meta-gradients are applied to the original GPT to build an ICL model. We comprehensively compare the behaviors of in-context learning and explicit finetuning on real tasks to provide empirical evidence that supports our understanding. Experimental results show that in-context learning behaves similarly to explicit finetuning from multiple perspectives. Inspired by the dual form between Transformer attention and gradient descent, we design a momentum-based attention by analogy with gradient descent with momentum. The improved performance over vanilla attention further supports our understanding from another perspective, and more importantly, shows the potential to utilize our understanding for future model design. The code is available at https://aka.ms/icl.

pdf abs
Dramatic Conversation Disentanglement
Kent Chang | Danica Chen | David Bamman

We present a new dataset for studying conversation disentanglement in movies and TV series. While previous work has focused on conversation disentanglement in IRC chatroom dialogues, movies and TV shows provide a space for studying complex pragmatic patterns of floor and topic change in face-to-face multi-party interactions. In this work, we draw on theoretical research in sociolinguistics, sociology, and film studies to operationalize a conversational thread (including the notion of a floor change) in dramatic texts, and use that definition to annotate a dataset of 10,033 dialogue turns (comprising 2,209 threads) from 831 movies. We compare the performance of several disentanglement models on this dramatic dataset, and apply the best-performing model to disentangle 808 movies. We see that, contrary to expectation, average thread lengths do not decrease significantly over the past 40 years, and characters portrayed by actors who are women, while underrepresented, initiate more new conversational threads relative to their speaking time.

pdf abs
Injecting Comparison Skills in Task-Oriented Dialogue Systems for Database Search Results Disambiguation
Yongil Kim | Yerin Hwang | Joongbo Shin | Hyunkyung Bae | Kyomin Jung

In task-oriented dialogue (TOD) systems designed to aid users accomplish specific goals in one or more domains, the agent retrieves entities that satisfy user constraints from the database. However, when multiple database search results exist, an ambiguity occurs regarding which results to select and present to the user. Existing TOD systems handle this ambiguity by randomly selecting one or few results and presenting their names to the user. However, in a real scenario, users do not always accept a randomly recommended entity, and users should have access to more comprehensive information about the search results. To address this limitation, we propose a novel task called Comparison-Based database search Ambiguity handling (CBA), which handles ambiguity in database search results by comparing the properties of multiple entities to enable users to choose according to their preferences. Accordingly, we introduce a new framework for automatically collecting high-quality dialogue data along with the Disambiguating Schema-guided Dialogue (DSD) dataset, an augmented version of the SGD dataset. Experimental studies on the DSD dataset demonstrate that training baseline models with the dataset effectively address the CBA task. Our dataset and code will be publicized.

This work examines the presence of modularity in pre-trained Transformers, a feature commonly found in human brains and thought to be vital for general intelligence. In analogy to human brains, we consider two main characteristics of modularity: (1) functional specialization of neurons: we evaluate whether each neuron is mainly specialized in a certain function, and find that the answer is yes. (2) function-based neuron grouping: we explore to find a structure that groups neurons into modules by function, and each module works for its corresponding function. Given the enormous amount of possible structures, we focus on Mixture-of-Experts as a promising candidate, which partitions neurons into experts and usually activates different experts for different inputs. Experimental results show that there are functional experts, where clustered are the neurons specialized in a certain function. Moreover, perturbing the activations of functional experts significantly affects the corresponding function. Finally, we study how modularity emerges during pre-training, and find that the modular structure is stabilized at the early stage, which is faster than neuron stabilization. It suggests that Transformer first constructs the modular structure and then learns fine-grained neuron functions. Our code and data are available at https://github.com/THUNLP/modularity-analysis.

Universal Information Extraction (Universal IE) aims to solve different extraction tasks in a uniform text-to-structure generation manner. Such a generation procedure tends to struggle when there exist complex information structures to be extracted. Retrieving knowledge from external knowledge bases may help models to overcome this problem but it is impossible to construct a knowledge base suitable for various IE tasks. Inspired by the fact that large amount of knowledge are stored in the pretrained language models (PLM) and can be retrieved explicitly, in this paper, we propose MetaRetriever to retrieve task-specific knowledge from PLMs to enhance universal IE. As different IE tasks need different knowledge, we further propose a Meta-Pretraining Algorithm which allows MetaRetriever to quicktly achieve maximum task-specific retrieval performance when fine-tuning on downstream IE tasks. Experimental results show that MetaRetriever achieves the new state-of-the-art on 4 IE tasks, 12 datasets under fully-supervised, low-resource and few-shot scenarios.

pdf abs
SETI: Systematicity Evaluation of Textual Inference
Xiyan Fu | Anette Frank

We propose SETI (Systematicity Evaluation of Textual Inference), a novel and comprehensive benchmark designed for evaluating pre-trained language models (PLMs) for their systematicity capabilities in the domain of textual inference. Specifically, SETI offers three different NLI tasks and corresponding datasets to evaluate various types of systematicity in reasoning processes. In order to solve these tasks, models are required to perform compositional inference based on known primitive constituents. We conduct experiments of SETI on six widely used PLMs. Results show that various PLMs are able to solve unseen compositional inferences when having encountered the knowledge of how to combine primitives, with good performance. However, they are considerably limited when this knowledge is unknown to the model (40-100 % points decrease). Furthermore, we find that PLMs are able to improve dramatically once exposed to crucial compositional knowledge in minimalistic shots. These findings position SETI as the first benchmark for measuring the future progress of PLMs in achieving systematicity generalization in the textual inference.

Recently, Few-shot Named Entity Recognition has received wide attention with the growing need for NER models to learn new classes with minimized annotation costs. However, one common yet understudied situation is to transfer a model trained with coarse-grained classes to recognize fine-grained classes, such as separating a product category into sub-classes. We find that existing few-shot NER solutions are not suitable for such a situation since they do not consider the sub-class discrimination during coarse training and various granularity of new classes during few-shot learning. In this work, we introduce the Coarse-to-fine Few-shot NER (C2FNER) task and propose an effective solution. Specifically, during coarse training, we propose a cluster-based prototype margin loss to learn group-wise discriminative representations, so as to benefit fine-grained learning. Targeting various granularity of new classes, we separate the coarse classes into extra-fine clusters and propose a novel prototype retrieval and bootstrapping algorithm to retrieve representative clusters for each fine class. We then adopt a mixture prototype loss to efficiently learn the representations of fine classes. We conduct experiments on both in-domain and cross-domain C2FNER settings with various target granularity, and the proposed method shows superior performance over the baseline methods.

pdf abs
Self-Evolution Learning for Discriminative Language Model Pretraining
Qihuang Zhong | Liang Ding | Juhua Liu | Bo Du | Dacheng Tao

Masked language modeling, widely used in discriminative language model (e.g., BERT) pretraining, commonly adopts a random masking strategy. However, random masking does not consider the importance of the different words in the sentence meaning, where some of them are more worthy to be predicted. Therefore, various masking strategies (e.g., entity-level masking) are proposed, but most of them require expensive prior knowledge and generally train from scratch without reusing existing model weights. In this paper, we present Self-Evolution learning (SE), a simple and effective token masking and learning method to fully and wisely exploit the knowledge from data. SE focuses on learning the informative yet under-explored tokens and adaptively regularizes the training by introducing a novel Token-specific Label Smoothing approach. Experiments on 10 tasks show that our SE brings consistent and significant improvements (+1.43 2.12 average scores) upon different PLMs. In-depth analyses demonstrate that SE improves linguistic knowledge learning and generalization.

Zero-shot transfer learning for document understanding is a crucial yet under-investigated scenario to help reduce the high cost involved in annotating document entities. We present a novel query-based framework, QueryForm, that extracts entity values from form-like documents in a zero-shot fashion. QueryForm contains a dual prompting mechanism that composes both the document schema and a specific entity type into a query, which is used to prompt a Transformer model to perform a single entity extraction task. Furthermore, we propose to leverage large-scale query-entity pairs generated from form-like webpages with weak HTML annotations to pre-train QueryForm. By unifying pre-training and fine-tuning into the same query-based framework, QueryForm enables models to learn from structured documents containing various entities and layouts, leading to better generalization to target document types without the need for target-specific training data. QueryForm sets new state-of-the-art average F1 score on both the XFUND (+4.6% 10.1%) and the Payment (+3.2% 9.5%) zero-shot benchmark, with a smaller model size and no additional image input.

Conversational query rewriting (CQR) realizes conversational search by reformulating the search dialogue into a standalone rewrite. However, existing CQR models either are not learned toward improving the downstream search performance or inefficiently generate the rewrite token-by-token from scratch while neglecting the fact that the search dialogue often has a large overlap with the rewrite. In this paper, we propose EdiRCS, a new text editing-based CQR model tailored for conversational search. In EdiRCS, most of the rewrite tokens are selected from the dialogue in a non-autoregressive fashion and only a few new tokens are generated to supplement the final rewrite, which makes EdiRCS highly efficient. In particular, the learning of EdiRCS is augmented with two search-oriented objectives, including contrastive ranking augmentation and contextualization knowledge transfer, which effectively improve it to select and generate more useful tokens from the view of retrieval. We show that EdiRCS outperforms state-of-the-art CQR models on three conversational search benchmarks while having low rewriting latency, and is robust to out-of-domain search dialogues and long dialogue contexts.

pdf abs
TAPIR: Learning Adaptive Revision for Incremental Natural Language Understanding with a Two-Pass Model
Patrick Kahardipraja | Brielen Madureira | David Schlangen

Language is by its very nature incremental in how it is produced and processed. This property can be exploited by NLP systems to produce fast responses, which has been shown to be beneficial for real-time interactive applications. Recent neural network-based approaches for incremental processing mainly use RNNs or Transformers. RNNs are fast but monotonic (cannot correct earlier output, which can be necessary in incremental processing). Transformers, on the other hand, consume whole sequences, and hence are by nature non-incremental. A restart-incremental interface that repeatedly passes longer input prefixes can be used to obtain partial outputs, while providing the ability to revise. However, this method becomes costly as the sentence grows longer. In this work, we propose the Two-pass model for AdaPtIve Revision (TAPIR) and introduce a method to obtain an incremental supervision signal for learning an adaptive revision policy. Experimental results on sequence labelling show that our model has better incremental performance and faster inference speed compared to restart-incremental Transformers, while showing little degradation on full sequences.

pdf abs
Speaking the Language of Your Listener: Audience-Aware Adaptation via Plug-and-Play Theory of Mind
Ece Takmaz | Nicolo’ Brandizzi | Mario Giulianelli | Sandro Pezzelle | Raquel Fernandez

Dialogue participants may have varying levels of knowledge about the topic under discussion. In such cases, it is essential for speakers to adapt their utterances by taking their audience into account. Yet, it is an open question how such adaptation can be modelled in computational agents. In this paper, we model a visually grounded referential game between a knowledgeable speaker and a listener with more limited visual and linguistic experience. Inspired by psycholinguistic theories, we endow our speaker with the ability to adapt its referring expressions via a simulation module that monitors the effectiveness of planned utterances from the listener’s perspective. We propose an adaptation mechanism building on plug-and-play approaches to controlled language generation, where utterance generation is steered on the fly by the simulator without finetuning the speaker’s underlying language model. Our results and analyses show that our approach is effective: the speaker’s utterances become closer to the listener’s domain of expertise, which leads to higher communicative success.

pdf abs
A Semi-Autoregressive Graph Generative Model for Dependency Graph Parsing
Ye Ma | Mingming Sun | Ping Li

Recent years have witnessed the impressive progress in Neural Dependency Parsing. According to the different factorization approaches to the graph joint probabilities, existing parsers can be roughly divided into autoregressive and non-autoregressive patterns. The former means that the graph should be factorized into multiple sequentially dependent components, then it can be built up component by component. And the latter assumes these components to be independent so that they can be outputted in a one-shot manner. However, when treating the directed edge as an explicit dependency relationship, we discover that there is a mixture of independent and interdependent components in the dependency graph, signifying that both aforementioned models fail to precisely capture the explicit dependencies among nodes and edges. Based on this property, we design a Semi-Autoregressive Dependency Parser to generate dependency graphs via adding node groups and edge groups autoregressively while pouring out all group elements in parallel. The model gains a trade-off between non-autoregression and autoregression, which respectively suffer from the lack of target inter-dependencies and the uncertainty of graph generation orders. The experiments show the proposed parser outperforms strong baselines on Enhanced Universal Dependencies of multiple languages, especially achieving 4% average promotion at graph-level accuracy. Also, the performances of model variations show the importance of specific parts.

Abstract Meaning Representation (AMR) is a semantic representation that can enhance natural language generation (NLG) by providing a logical semantic input. In this paper, we propose the AMR-TST, an AMR-based text style transfer (TST) technique. The AMR-TST converts the source text to an AMR graph and generates the transferred text based on the AMR graph modified by a TST policy named style rewriting. Our method combines both the explainability and diversity of explicit and implicit TST methods. The experiments show that the proposed method achieves state-of-the-art results compared with other baseline models in automatic and human evaluations. The generated transferred text in qualitative evaluation proves the AMR-TST have significant advantages in keeping semantic features and reducing hallucinations. To the best of our knowledge, this work is the first to apply the AMR method focusing on node-level features to the TST task.

pdf abs
Understanding the Cooking Process with English Recipe Text
Yi Fan | Anthony Hunter

Translating procedural text, like recipes, into a graphical representation can be important for visualizing the text, and can offer a machine-readable formalism for use in software. There are proposals for translating recipes into a flow graph representation, where each node represents an ingredient, action, location, or equipment, and each arc between the nodes denotes the steps of the recipe. However, these proposals have had performance problems with both named entity recognition and relationship extraction. To address these problems, we propose a novel framework comprising two modules to construct a flow graph from the input recipe. The first module identifies the named entities in the input recipe text using BERT, Bi-LSTM and CRF, and the second module uses BERT to predict the relationships between the entities. We evaluate our framework on the English recipe flow graph corpus. Our framework can predict the edge label and achieve the overall F1 score of 92.2, while the baseline F1 score is 43.3 without the edge label predicted.

pdf abs
Follow the Wisdom of the Crowd: Effective Text Generation via Minimum Bayes Risk Decoding
Mirac Suzgun | Luke Melas-Kyriazi | Dan Jurafsky

In open-ended natural-language generation, existing text decoding methods typically struggle to produce text which is both diverse and high-quality. Greedy and beam search are known to suffer from text degeneration and linguistic diversity issues, while temperature, top-k, and nucleus sampling yield diverse but often lower-quality outputs. In this work, we build upon Minimum Bayes Risk Decoding (MBRD), a family of decoding methods based on Bayesian risk minimization, to address this diversity-quality trade-off. Inspired by the principle of the wisdom of the crowd, MBRD seeks to select a candidate from a pool of candidates that has the least expected risk under a generative model according to a given utility function. The crowd of candidates serves as an approximation for the distribution over human-generated references. We show that MBRD generalizes numerous decoding methods, including majority voting, and can be used as a drop-in replacement for existing sampling methods. Across a wide range of tasks—such as summarization, data-to-text, translation, and textual style transfer—MBRD yields 3-7 ROUGE and BLEU point improvements, including state-of-the-art results on WebNLG and WMT’16.

Open-domain question answering (ODQA) is a crucial task in natural language processing. A typical ODQA system relies on a retriever module to select relevant contexts from a large corpus for a downstream reading comprehension model. Existing ODQA datasets consist mainly of Wikipedia corpus, and are insufficient to study models’ generalizability across diverse domains as models are trained and evaluated on the same genre of data. We propose **RobustQA**, a novel benchmark consisting of datasets from 8 different domains, which facilitates the evaluation of ODQA’s domain robustness. To build **RobustQA**, we annotate QA pairs in retrieval datasets with rigorous quality control. We further examine improving QA performances by incorporating unsupervised learning methods with target-domain corpus and adopting large generative language models. These methods can effectively improve model performances on **RobustQA**. However, experimental results demonstrate a significant gap from in-domain training, suggesting that **RobustQA** is a challenging benchmark to evaluate ODQA domain robustness.

pdf abs
SenteCon: Leveraging Lexicons to Learn Human-Interpretable Language Representations
Victoria Lin | Louis-Philippe Morency

Although deep language representations have become the dominant form of language featurization in recent years, in many settings it is important to understand a model’s decision-making process. This necessitates not only an interpretable model but also interpretable features. In particular, language must be featurized in a way that is interpretable while still characterizing the original text well. We present SenteCon, a method for introducing human interpretability in deep language representations. Given a passage of text, SenteCon encodes the text as a layer of interpretable categories in which each dimension corresponds to the relevance of a specific category. Our empirical evaluations indicate that encoding language with SenteCon provides high-level interpretability at little to no cost to predictive performance on downstream tasks. Moreover, we find that SenteCon outperforms existing interpretable language representations with respect to both its downstream performance and its agreement with human characterizations of the text.

pdf abs
Reinforcement Learning for Topic Models
Jeremy Costello | Marek Reformat

We apply reinforcement learning techniques to topic modeling by replacing the variational autoencoder in ProdLDA with a continuous action space reinforcement learning policy. We train the system with a policy gradient algorithm REINFORCE. Additionally, we introduced several modifications: modernize the neural network architecture, weight the ELBO loss, use contextual embeddings, and monitor the learning process via computing topic diversity and coherence for each training step. Experiments areperformed on 11 data sets. Our unsupervised model outperforms all other unsupervised models and performs on par with or better than most models using supervised labeling. Our model is outperformed on certain data sets by a model using supervised labeling and contrastive learning. We have also conducted an ablation study to provide empirical evidence of performance improvements from changes we made to ProdLDA and found that the reinforcement learning formulation boosts performance. We open-source our code implementation.

pdf abs
Contextualized Soft Prompts for Extraction of Event Arguments
Chien Nguyen | Hieu Man | Thien Nguyen

Event argument extraction (EAE) is a sub-task of event extraction where the goal is to identify roles of entity mentions for events in text. The current state-of-the-art approaches for this problem explore prompt-based methods to prompt pre-trained language models for arguments over input context. However, existing prompt-based methods mainly rely on discrete and manually-designed prompts that cannot exploit specific context for each example to improve customization for optimal performance. In addition, the discrete nature of current prompts prevents the incorporation of relevant context from multiple external documents to enrich prompts for EAE. To this end, we propose a novel prompt-based method for EAE that introduces soft prompts to facilitate the encoding of individual example context and multiple relevant documents to boost EAE. We extensively evaluate the proposed method on benchmark datasets for EAE to demonstrate its benefits with state-of-the-art performance.

pdf abs
TextVerifier: Robustness Verification for Textual Classifiers with Certifiable Guarantees
Siqi Sun | Wenjie Ruan

When textual classifiers are deployed in safety-critical workflows, they must withstand the onslaught of AI-enabled model confusion caused by adversarial examples with minor alterations. In this paper, the main objective is to provide a formal verification framework, called TextVerifier, with certifiable guarantees on deep neural networks in natural language processing against word-level alteration attacks. We aim to provide an approximation of the maximal safe radius by deriving provable bounds both mathematically and automatically, where a minimum word-level L_0 distance is quantified as a guarantee for the classification invariance of victim models. Here, we illustrate three strengths of our strategy: i) certifiable guarantee: effective verification with convergence to ensure approximation of maximal safe radius with tight bounds ultimately; ii) high-efficiency: it yields an efficient speed edge by a novel parallelization strategy that can process a set of candidate texts simultaneously on GPUs; and iii) reliable anytime estimation: the verification can return intermediate bounds, and robustness estimates that are gradually, but strictly, improved as the computation proceeds. Furthermore, experiments are conducted on text classification on four datasets over three victim models to demonstrate the validity of tightening bounds. Our tool TextVerifier is available at https://github.com/TrustAI/TextVerifer.

Aspect or query-based summarization has recently caught more attention, as it can generate differentiated summaries based on users’ interests. However, the current dataset for aspect or query-based summarization either focuses on specific domains, on a relatively small scale, or contains only a few aspect types. Such limitations hinder further explorations in this direction. In this work, we take advantage of crowd-sourcing knowledge on Wikipedia and automatically create a high-quality, large-scale open-domain aspect-based summarization dataset named OASum, which contains more than 3.7 million instances with around 1 million different aspects on 2 million Wikipedia pages. We provide benchmark results on OASum and demonstrate its ability for diverse aspect-based summarization generation. To overcome the data scarcity problem on specific domains, we also perform zero-shot, few-shot, and fine-tuning on seven downstream datasets. Specifically, zero/few-shot and fine-tuning results show that the model pre-trained on our corpus demonstrates a strong aspect or query-focused generation ability compared with the backbone model. Our dataset and pre-trained checkpoints are publicly available.

pdf abs
On the Limitations of Simulating Active Learning
Katerina Margatina | Nikolaos Aletras

Active learning (AL) is a human-and-model-in-the-loop paradigm that iteratively selects informative unlabeled data for human annotation, aiming to improve data efficiency over random sampling. However, performing AL experiments with human annotations on-the-fly is a laborious and expensive process, thus unrealistic for academic research. An easy fix to this impediment is to simulate AL, by treating an already labeled and publicly available dataset as the pool of unlabeled data. In this position paper, we first survey recent literature and highlight the challenges across all different steps within the AL loop. We further unveil neglected caveats in the experimental setup that can significantly affect the quality of AL research. We continue with an exploration of how the simulation setting can govern empirical findings, arguing that it might be one of the answers behind the ever posed question “Why do Active Learning algorithms sometimes fail to outperform random sampling?”. We argue that evaluating AL algorithms on available labeled datasets might provide a lower bound as to their effectiveness in real data. We believe it is essential to collectively shape the best practices for AL research, especially now that the stellar engineering advances (e.g. ChatGPT) shift the research focus to data-driven approaches. To this end, we present guidelines for future work, hoping that by bringing these limitations to the community’s attention, we can explore ways to address them.

pdf abs
Towards Alleviating the Object Bias in Prompt Tuning-based Factual Knowledge Extraction
Yuhang Wang | Dongyuan Lu | Chao Kong | Jitao Sang

Many works employed prompt tuning methods to automatically optimize prompt queries and extract the factual knowledge stored in Pre-trained Language Models. In this paper, we observe that the optimized prompts, including discrete prompts and continuous prompts, exhibit undesirable object bias. To handle this problem, we propose a novel prompt tuning method called MeCoD consisting of three modules: Prompt Encoder, Object Equalization and Biased Object Obstruction. Experimental results show that MeCoD can significantly reduce the object bias and at the same time improve accuracy of factual knowledge extraction.

pdf abs
vONTSS: vMF based semi-supervised neural topic modeling with optimal transport
Weijie Xu | Xiaoyu Jiang | Srinivasan Sengamedu Hanumantha Rao | Francis Iannacci | Jinjin Zhao

Recently, Neural Topic Models (NTM), inspired by variational autoencoders, have attracted a lot of research interest; however, these methods have limited applications in the real world due to the challenge of incorporating human knowledge. This work presents a semi-supervised neural topic modeling method, vONTSS, which uses von Mises-Fisher (vMF) based variational autoencoders and optimal transport. When a few keywords per topic are provided, vONTSS in the semi-supervised setting generates potential topics and optimizes topic-keyword quality and topic classification. Experiments show that vONTSS outperforms existing semi-supervised topic modeling methods in classification accuracy and diversity. vONTSS also supports unsupervised topic modeling. Quantitative and qualitative experiments show that vONTSS in the unsupervised setting outperforms recent NTMs on multiple aspects: vONTSS discovers highly clustered and coherent topics on benchmark datasets. It is also much faster than the state-of-the-art weakly supervised text classification method while achieving similar classification performance. We further prove the equivalence of optimal transport loss and cross-entropy loss at the global minimum.

pdf abs
Bias Beyond English: Counterfactual Tests for Bias in Sentiment Analysis in Four Languages
Seraphina Goldfarb-Tarrant | Adam Lopez | Roi Blanco | Diego Marcheggiani

Sentiment analysis (SA) systems are used in many products and hundreds of languages. Gender and racial biases are well-studied in English SA systems, but understudied in other languages, with few resources for such studies. To remedy this, we build a counterfactual evaluation corpus for gender and racial/migrant bias in four languages. We demonstrate its usefulness by answering a simple but important question that an engineer might need to answer when deploying a system: What biases do systems import from pre-trained models when compared to a baseline with no pre-training? Our evaluation corpus, by virtue of being counterfactual, not only reveals which models have less bias, but also pinpoints changes in model bias behaviour, which enables more targeted mitigation strategies. We release our code and evaluation corpora to facilitate future research.

Large language models (LLMs) have exhibited remarkable capabilities in learning from expla- nations in prompts, but there has been limited understanding of exactly how these explana- tions function or why they are effective. This work aims to better understand the mechanisms by which explanations are used for in-context learning. We first study the impact of two dif- ferent factors on the performance of prompts with explanations: the computation trace (the way the solution is decomposed) and the natural language used to express the prompt. By per- turbing explanations on three controlled tasks, we show that both factors contribute to the ef- fectiveness of explanations. We further study how to form maximally effective sets of expla- nations for solving a given test query. We find that LLMs can benefit from the complemen- tarity of the explanation set: diverse reasoning skills shown by different exemplars can lead to better performance. Therefore, we propose a maximal marginal relevance-based exemplar selection approach for constructing exemplar sets that are both relevant as well as comple- mentary, which successfully improves the in- context learning performance across three real- world tasks on multiple LLMs.

With the growing interest in large language models, the need for evaluating the quality of machine text compared to reference (typically human-generated) text has become focal attention. Most recent works focus either on task-specific evaluation metrics or study the properties of machine-generated text captured by the existing metrics. In this work, we propose a new evaluation scheme to model human judgments in 7 NLP tasks, based on the fine-grained mismatches between a pair of texts. Inspired by the recent efforts in several NLP tasks for fine-grained evaluation, we introduce a set of 13 mismatch error types such as spatial/geographic errors, entity errors, etc, to guide the model for better prediction of human judgments. We propose a neural framework for evaluating machine texts that uses these mismatch error types as auxiliary tasks and re-purposes the existing single-number evaluation metrics as additional scalar features, in addition to textual features extracted from the machine and reference texts. Our experiments reveal key insights about the existing metrics via the mismatch errors. We show that the mismatch errors between the sentence pairs on the held-out datasets from 7 NLP tasks align well with the human evaluation.

Dialogue systems can leverage large pre-trained language models and knowledge to generate fluent and informative responses. However, these models are still prone to produce hallucinated responses not supported by the input source, which greatly hinders their application. The heterogeneity between external knowledge and dialogue context challenges representation learning and source integration, which further contributes to unfaithfulness. To handle this challenge and generate more faithful responses, this paper presents RHO (ρ) utilizing the representations of linked entities and relation predicates from a knowledge graph (KG). We propose (1) local knowledge grounding to combine textual embeddings with the corresponding KG embeddings; and (2) global knowledge grounding to equip RHO with multi-hop reasoning abilities via the attention mechanism. In addition, we devise a response re-ranking technique based on walks over KG sub-graphs for better conversational reasoning. Experimental results on OpenDialKG (Moon et al., 2019) show that our approach significantly outperforms state-of-the-art methods on both automatic and human evaluation by a large margin, especially in hallucination reduction (17.54% in FeQA (Durmus et al., 2020)).

pdf abs
Transformer Language Models Handle Word Frequency in Prediction Head
Goro Kobayashi | Tatsuki Kuribayashi | Sho Yokoi | Kentaro Inui

Prediction head is a crucial component of Transformer language models. Despite its direct impact on prediction, this component has often been overlooked in analyzing Transformers.In this study, we investigate the inner workings of the prediction head, specifically focusing on bias parameters. Our experiments with BERT and GPT-2 models reveal that the biases in their word prediction heads play a significant role in the models’ ability to reflect word frequency in a corpus, aligning with the logit adjustment method commonly used in long-tailed learning. We also quantify the effect of controlling the biases in practical auto-regressive text generation scenarios;under a particular setting, more diverse text can be generated without compromising text quality.

pdf abs
Prompted LLMs as Chatbot Modules for Long Open-domain Conversation
Gibbeum Lee | Volker Hartmann | Jongho Park | Dimitris Papailiopoulos | Kangwook Lee

In this paper, we propose MPC (Modular Prompted Chatbot), a new approach for creating high-quality conversational agents without the need for fine-tuning. Our method utilizes pre-trained large language models (LLMs) as individual modules for long-term consistency and flexibility, by using techniques such as few-shot prompting, chain-of-thought (CoT), and external memory. Our human evaluation results show that MPC is on par with fine-tuned chatbot models in open-domain conversations, making it an effective solution for creating consistent and engaging chatbots.

pdf abs
Prompt to be Consistent is Better than Self-Consistent? Few-Shot and Zero-Shot Fact Verification with Pre-trained Language Models
Fengzhu Zeng | Wei Gao

Few-shot or zero-shot fact verification only relies on a few or no labeled training examples. In this paper, we propose a novel method called ProToCo, to Prompt pre-trained language models (PLMs) To be Consistent, for improving the factuality assessment capability of PLMs in the few-shot and zero-shot settings. Given a claim-evidence pair, ProToCo generates multiple variants of the claim with different relations and frames a simple consistency mechanism as constraints for making compatible predictions across these variants. We update PLMs by using parameter-efficient fine-tuning (PEFT), leading to more accurate predictions in few-shot and zero-shot fact verification tasks. Our experiments on three public verification datasets show that ProToCo significantly outperforms state-of-the-art few-shot fact verification baselines. With a small number of unlabeled instances, ProToCo also outperforms the strong zero-shot learner T0 on zero-shot verification. Compared to large PLMs using in-context learning (ICL) method, ProToCo outperforms OPT-30B and the Self-Consistency-enabled OPT-6.7B model in both few- and zero-shot settings.

pdf abs
Model Analysis & Evaluation for Ambiguous Question Answering
Konstantinos Papakostas | Irene Papadopoulou

Ambiguous questions are a challenge for Question Answering models, as they require answers that cover multiple interpretations of the original query. To this end, these models are required to generate long-form answers that often combine conflicting pieces of information. Although recent advances in the field have shown strong capabilities in generating fluent responses, certain research questions remain unanswered. Does model/data scaling improve the answers’ quality? Do automated metrics align with human judgment? To what extent do these models ground their answers in evidence? In this study, we aim to thoroughly investigate these aspects, and provide valuable insights into the limitations of the current approaches. To aid in reproducibility and further extension of our work, we open-source our code.

pdf abs
Debiasing should be Good and Bad: Measuring the Consistency of Debiasing Techniques in Language Models
Robert Morabito | Jad Kabbara | Ali Emami

Debiasing methods that seek to mitigate the tendency of Language Models (LMs) to occasionally output toxic or inappropriate text have recently gained traction. In this paper, we propose a standardized protocol which distinguishes methods that yield not only desirable results, but are also consistent with their mechanisms and specifications. For example, we ask, given a debiasing method that is developed to reduce toxicity in LMs, if the definition of toxicity used by the debiasing method is reversed, would the debiasing results also be reversed? We used such considerations to devise three criteria for our new protocol: Specification Polarity, Specification Importance, and Domain Transferability. As a case study, we apply our protocol to a popular debiasing method, Self-Debiasing, and compare it to one we propose, called Instructive Debiasing, and demonstrate that consistency is as important an aspect to debiasing viability as is simply a desirable result. We show that our protocol provides essential insights into the generalizability and interpretability of debiasing methods that may otherwise go overlooked.

Steering language generation towards objectives or away from undesired content has been a long-standing goal in utilizing language models (LM). Recent work has demonstrated reinforcement learning and weighted decoding as effective approaches to achieve a higher level of language control and quality with pros and cons. In this work, we propose a novel critic decoding method for controlled language generation (CriticControl) that combines the strengths of reinforcement learning and weighted decoding. Specifically, we adopt the actor-critic framework and train an LM-steering critic from reward models. Similar to weighted decoding, our method freezes the language model and manipulates the output token distribution using a critic to improve training efficiency and stability. Evaluation of our method on three controlled generation tasks, topic control, sentiment control, and detoxification, shows that our approach generates more coherent and well-controlled texts than previous methods. In addition, CriticControl demonstrates superior generalization ability in zero-shot settings. Human evaluation studies also corroborate our findings.

pdf abs
MedNgage: A Dataset for Understanding Engagement in Patient-Nurse Conversations
Yan Wang | Heidi Donovan | Sabit Hassan | Malihe Alikhani

Patients who effectively manage their symptoms often demonstrate higher levels of engagement in conversations and interventions with healthcare practitioners. This engagement is multifaceted, encompassing cognitive and social dimensions. Consequently, it is crucial for AI systems to understand the engagement in natural conversations between patients and practitioners to better contribute toward patient care. In this paper, we present a novel dataset (MedNgage), which consists of patient-nurse conversations about cancer symptom management. We manually annotate the dataset with a novel framework of categories of patient engagement from two different angles, namely: i) socio-affective engagement (3.1K spans), and ii) cognitive engagement (1.8K spans). Through statistical analysis of the data that is annotated using our framework, we show a positive correlation between patient symptom management outcomes and their engagement in conversations. Additionally, we demonstrate that pre-trained transformer models fine-tuned on our dataset can reliably predict engagement categories in patient-nurse conversations. Lastly, we use LIME (Ribeiro et al., 2016) to analyze the underlying challenges of the tasks that state-of-the-art transformer models encounter. The de-identified data is available for research purposes upon request.

Extracting event causality underlies a broad spectrum of natural language processing applications. Cutting-edge methods break this task into Event Detection and Event Causality Identification. Although the pipelined solutions succeed in achieving acceptable results, the inherent nature of separating the task incurs limitations. On the one hand, it suffers from the lack of cross-task dependencies and may cause error propagation. On the other hand, it predicts events and relations separately, undermining the integrity of the event causality graph (ECG). To address such issues, in this paper, we propose an approach for Structure-Aware Event Causality Generation (SEAG). With a graph linearization module, we generate the ECG structure in a way of text2text generation based on a pre-trained language model. To foster the structural representation of the ECG, we introduce the novel Causality Structural Discrimination training paradigm in which we perform structural discriminative training alongside auto-regressive generation enabling the model to distinguish from constructed incorrect ECGs. We conduct experiments on three datasets. The experimental results demonstrate the effectiveness of structural event causality generation and the causality structural discrimination training.

pdf abs
Large Language Models Can be Lazy Learners: Analyze Shortcuts in In-Context Learning
Ruixiang Tang | Dehan Kong | Longtao Huang | Hui Xue

Large language models (LLMs) have recently shown great potential for in-context learning, where LLMs learn a new task simply by conditioning on a few input-label pairs (prompts). Despite their potential, our understanding of the factors influencing end-task performance and the robustness of in-context learning remains limited. This paper aims to bridge this knowledge gap by investigating the reliance of LLMs on shortcuts or spurious correlations within prompts. Through comprehensive experiments on classification and extraction tasks, we reveal that LLMs are “lazy learners” that tend to exploit such shortcuts. Additionally, we uncover a surprising finding that larger models are more likely to utilize shortcuts in prompts during inference. Our findings provide a new perspective on evaluating robustness in in-context learning and pose new challenges for detecting and mitigating the use of shortcuts in prompts.

pdf abs
A Two-Stage Decoder for Efficient ICD Coding
Thanh-Tung Nguyen | Viktor Schlegel | Abhinav Ramesh Kashyap | Stefan Winkler

Clinical notes in healthcare facilities are tagged with the International Classification of Diseases (ICD) code; a list of classification codes for medical diagnoses and procedures. ICD coding is a challenging multilabel text classification problem due to noisy clinical document inputs and long-tailed label distribution. Recent automated ICD coding efforts improve performance by encoding medical notes and codes with additional data and knowledge bases. However, most of them do not reflect how human coders generate the code: first, the coders select general code categories and then look for specific subcategories that are relevant to a patient’s condition. Inspired by this, we propose a two-stage decoding mechanism to predict ICD codes. Our model uses the hierarchical properties of the codes to split the prediction into two steps: At first, we predict the parent code and then predict the child code based on the previous prediction. Experiments on the public MIMIC-III data set have shown that our model performs well in single-model settings without external data or knowledge.

pdf abs
Asymmetric feature interaction for interpreting model predictions
Xiaolei Lu | Jianghong Ma | Haode Zhang

In natural language processing (NLP), deep neural networks (DNNs) could model complex interactions between context and have achieved impressive results on a range of NLP tasks. Prior works on feature interaction attribution mainly focus on studying symmetric interaction that only explains the additional influence of a set of words in combination, which fails to capture asymmetric influence that contributes to model prediction. In this work, we propose an asymmetric feature interaction attribution explanation model that aims to explore asymmetric higher-order feature interactions in the inference of deep neural NLP models. By representing our explanation with an directed interaction graph, we experimentally demonstrate interpretability of the graph to discover asymmetric feature interactions. Experimental results on two sentiment classification datasets show the superiority of our model against the state-of-the-art feature interaction attribution methods in identifying influential features for model predictions.

pdf abs
Disagreement Matters: Preserving Label Diversity by Jointly Modeling Item and Annotator Label Distributions with DisCo
Tharindu Cyril Weerasooriya | Alexander Ororbia | Raj Bhensadadia | Ashiqur KhudaBukhsh | Christopher Homan

Annotator disagreement is common whenever human judgment is needed for supervised learning. It is conventional to assume that one label per item represents ground truth. However, this obscures minority opinions, if present. We regard “ground truth” as the distribution of all labels that a population of annotators could produce, if asked (and of which we only have a small sample). We next introduce DisCo (Distribution from Context), a simple neural model that learns to predict this distribution. The model takes annotator-item pairs, rather than items alone, as input, and performs inference by aggregating over all annotators. Despite its simplicity, our experiments show that, on six benchmark datasets, our model is competitive with, and frequently outperforms, other, more complex models that either do not model specific annotators or were not designed for label distribution learning.

pdf abs
Domain Aligned Prefix Averaging for Domain Generalization in Abstractive Summarization
Pranav Nair | Sukomal Pal | Pradeepika Verma

Domain generalization is hitherto an underexplored area applied in abstractive summarization. Moreover, most existing works on domain generalization have sophisticated training algorithms. In this paper, we propose a lightweight, weight averaging based, Domain Aligned Prefix Averaging approach to domain generalization for abstractive summarization. Given a number of source domains, our method first trains a prefix for each one of them. These source prefixes generate summaries for a small number of target domain documents. The similarity of the generated summaries to their corresponding source documents is used for calculating weights required to average source prefixes. In DAPA, prefix tuning allows for lightweight finetuning, and weight averaging allows for the computationally efficient addition of new source domains. When evaluated on four diverse summarization domains, DAPA shows comparable or better performance against the baselines demonstrating the effectiveness of its prefix averaging scheme.

With the growing importance of detecting misinformation, many studies have focused on verifying factual claims by retrieving evidence. However, canonical fact verification tasks do not apply to catching subtle differences in factually consistent claims, which might still bias the readers, especially on contentious political or economic issues. Our underlying assumption is that among the trusted sources, one’s argument is not necessarily more true than the other, requiring comparison rather than verification. In this study, we propose ClaimDIff, a novel dataset that primarily focuses on comparing the nuance between claim pairs. In ClaimDiff, we provide human-labeled 2,941 claim pairs from 268 news articles. We observe that while humans are capable of detecting the nuances between claims, strong baselines struggle to detect them, showing over a 19% absolute gap with the humans. We hope this initial study could help readers to gain an unbiased grasp of contentious issues through machine-aided comparison.

pdf abs
Unsupervised Paraphrasing of Multiword Expressions
Takashi Wada | Yuji Matsumoto | Timothy Baldwin | Jey Han Lau

We propose an unsupervised approach to paraphrasing multiword expressions (MWEs) in context. Our model employs only monolingual corpus data and pre-trained language models (without fine-tuning), and does not make use of any external resources such as dictionaries. We evaluate our method on the SemEval 2022 idiomatic semantic text similarity task, and show that it outperforms all unsupervised systems and rivals supervised systems.

pdf abs
G-Tuning: Improving Generalization of Pre-trained Language Models with Generative Adversarial Network
Rongxiang Weng | Wen Sen Cheng | Min Zhang

The generalization ability of pre-trained language models (Plms) in downstream tasks is heavily influenced by fine-tuning. The objective of fine-tuning is to transform the latent representation of Plms from a universal space to a target space, allowing the model to be applied to downstream tasks with the capability of generalizing to unseen samples. However, the effect of Plms will be diminished when the training data coverage is insufficient, in which fine-tuning is inadequate to learn the complete mapping. In this study, we propose a new fine-tuning framework, referred to as G-Tuning, that aims to preserve the generalization ability of Plms in downstream tasks. Specifically, we integrate a generative adversarial network into the fine-tuning process to aid in the transformation of the latent representation in the entire space. Empirical evaluations on the GLUE benchmark, as well as two additional demanding scenarios involving domain and language generalization, demonstrate that G-Tuning can accurately map the universal representation to the target space, thus effectively enhancing the generalization performance of Plms across various downstream tasks.

pdf abs
Unified Language Representation for Question Answering over Text, Tables, and Images
Bowen Yu | Cheng Fu | Haiyang Yu | Fei Huang | Yongbin Li

When trying to answer complex questions, people often rely on multiple sources of information, such as visual, textual, and tabular data. Previous approaches to this problem have focused on designing input features or model structure in the multi-modal space, which is inflexible for cross-modal reasoning or data-efficient training. In this paper, we call for an alternative paradigm, which transforms the images and tables into unified language representations, so that we can simplify the task into a simpler textual QA problem that can be solved using three steps: retrieval, ranking, and generation, all within a language space. This idea takes advantage of the power of pre-trained language models and is implemented in a framework called Solar. Our experimental results show that Solar outperforms all existing methods by 10.6-32.3 pts on two datasets, MultimodalQA and MMCoQA, across ten different metrics. Additionally, Solar achieves the best performance on the WebQA leaderboard.

pdf abs
A Set Prediction Network For Extractive Summarization
Xiaoxia Cheng | Yongliang Shen | Weiming Lu

Extractive summarization focuses on extracting salient sentences from the source document and incorporating them in the summary without changing their wording or structure. The naive approach for extractive summarization is sentence classification, which makes independent binary decisions for each sentence, resulting in the model cannot detect the dependencies between sentences in the summary. Recent approaches introduce an autoregressive decoder to detect redundancy relationship between sentences by step-by-step sentence selection, but bring train-inference gap. To address these issues, we formulate extractive summarization as a salient sentence set recognition task. To solve the sentence set recognition task, we propose a set prediction network (SetSum), which sets up a fixed set of learnable queries to extract the entire sentence set of the summary, while capturing the dependencies between them.Different from previous methods with an auto-regressive decoder, we employ a non-autoregressive decoder to predict the sentences within the summary in parallel during both the training and inference process, which eliminates the train-inference gap. Experimental results on both single-document and multi-document extracted summary datasets show that our approach outperforms previous state-of-the-art models.

pdf abs
Geo-Seq2seq: Twitter User Geolocation on Noisy Data through Sequence to Sequence Learning
Jingyu Zhang | Alexandra DeLucia | Chenyu Zhang | Mark Dredze

Location information can support social media analyses by providing geographic context. Some of the most accurate and popular Twitter geolocation systems rely on rule-based methods that examine the user-provided profile location, which fail to handle informal or noisy location names. We propose Geo-Seq2seq, a sequence-to-sequence (seq2seq) model for Twitter user geolocation that rewrites noisy, multilingual user-provided location strings into structured English location names. We train our system on tens of millions of multilingual location string and geotagged-tweet pairs. Compared to leading methods, our model vastly increases coverage (i.e., the number of users we can geolocate) while achieving comparable or superior accuracy. Our error analysis reveals that constrained decoding helps the model produce valid locations according to a location database. Finally, we measure biases across language, country of origin, and time to evaluate fairness, and find that while our model can generalize well to unseen temporal data, performance does vary by language and country.

pdf abs
Predicting Numerals in Text Using Nearest Neighbor Language Models
Taku Sakamoto | Akiko Aizawa

Commonsense about quantitative properties is essential for a deep understanding of texts containing numerals. However, naive language models (LMs) treat numerals as string tokens; therefore, they lack an understanding of the magnitudes of numerals, resulting in a difficulty in acquiring the commonsense. In this study, we apply the k-nearest neighbor LM (kNN-LM) to the masked numeral prediction (MNP) task, which measures the quantitative commonsense of LMs.kNN-LM extends pre-trained neural LMs with the k-nearest neighbor (kNN) search.Since it can utilize patterns that appear in the datastore for prediction, we expect an improvement in numeral prediction accuracy, which is associated with a high rate of occurrence of out-of-vocabulary (OOV) words.Through experiments, we verified that the retrieval-based method is effective for fine-grained predictions of numerals from context, especially for the OOV numerals.We also compared two different context spans for context representations to improve the accuracy of kNN search by using only the words that are closely related to the masked numeral: the mask and its surrounding words, and the mask and its subsequent words.Our results reveal that using only the embeddings of mask tokens for numerals in kNN search is the most effective approach for realizing MNP tasks.

pdf abs
HonestBait: Forward References for Attractive but Faithful Headline Generation
Chih Yao Chen | Dennis Wu | Lun-Wei Ku

Current methods for generating attractive headlines often learn directly from data, which bases attractiveness on the number of user clicks and views. Although clicks or views do reflect user interest, they can fail to reveal how much interest is raised by the writing style and how much is due to the event or topic itself. Also, such approaches can lead to harmful inventions by over-exaggerating the content, aggravating the spread of false information. In this work, we propose HonestBait, a novel framework for solving these issues from another aspect: generating headlines using forward references (FRs), a writing technique often used for clickbait. A self-verification process is included during training to avoid spurious inventions. We begin with a preliminary user study to understand how FRs affect user interest, after which we present PANCO, an innovative dataset containing pairs of fake news with verified news for attractive but faithful news headline generation. Auto matic metrics and human evaluations show that our framework yields more attractive results (+11.25% compared to human-written verified news headlines) while maintaining high veracity, which helps promote real information to fight against fake news.

pdf abs
Few Shot Rationale Generation using Self-Training with Dual Teachers
Aditya Srikanth Veerubhotla | Lahari Poddar | Jun Yin | György Szarvas | Sharanya Eswaran

Self-rationalizing models that also generate a free-text explanation for their predicted labels are an important tool to build trustworthy AI applications. Since generating explanations for annotated labels is a laborious and costly process, recent models rely on large pretrained language models (PLMs) as their backbone and few-shot learning. In this work we explore a self-training approach leveraging both labeled and unlabeled data to further improve few-shot models, under the assumption that neither human written rationales nor annotated task labels are available at scale. We introduce a novel dual-teacher learning framework, which learns two specialized teacher models for task prediction and rationalization using self-training and distills their knowledge into a multi-tasking student model that can jointly generate the task label and rationale. Furthermore, we formulate a new loss function, Masked Label Regularization(MLR) which promotes explanations to be strongly conditioned on predicted labels. Evaluation on three public datasets demonstrate that the proposed methods are effective in modeling task labels and generating faithful rationales.

Lexically-constrained NMT (LNMT) aims to incorporate user-provided terminology into translations. Despite its practical advantages, existing work has not evaluated LNMT models under challenging real-world conditions. In this paper, we focus on two important but understudied issues that lie in the current evaluation process of LNMT studies. The model needs to cope with challenging lexical constraints that are “homographs” or “unseen” during training. To this end, we first design a homograph disambiguation module to differentiate the meanings of homographs. Moreover, we propose PLUMCOT which integrates contextually rich information about unseen lexical constraints from pre-trained language models and strengthens a copy mechanism of the pointer network via direct supervision of a copying score. We also release HOLLY, an evaluation benchmark for assessing the ability of model to cope with “homographic” and “unseen” lexical constraints. Experiments on HOLLY and the previous test setup show the effectiveness of our method. The effects of PLUMCOT are shown to be remarkable in “unseen” constraints. Our dataset is available at https://github.com/papago-lab/HOLLY-benchmark.

Large-scale datasets in the real world inevitably involve label noise. Deep models can gradually overfit noisy labels and thus degrade model generalization. To mitigate the effects of label noise, learning with noisy labels (LNL) methods are designed to achieve better generalization performance. Due to the lack of suitable datasets, previous studies have frequently employed synthetic label noise to mimic real-world label noise. However, synthetic noise is not instance-dependent, making this approximation not always effective in practice. Recent research has proposed benchmarks for learning with real-world noisy labels. However, the noise sources within may be single or fuzzy, making benchmarks different from data with heterogeneous label noises in the real world. To tackle these issues, we contribute NoisywikiHow, the largest NLP benchmark built with minimal supervision. Specifically, inspired by human cognition, we explicitly construct multiple sources of label noise to imitate human errors throughout the annotation, replicating real-world noise, whose corruption is affected by both ground-truth labels and instances. Moreover, we provide a variety of noise levels to support controlled experiments on noisy data, enabling us to evaluate LNL methods systematically and comprehensively. After that, we conduct extensive multi-dimensional experiments on a broad range of LNL methods, obtaining new and intriguing findings.

pdf abs
Sampling Better Negatives for Distantly Supervised Named Entity Recognition
Lu Xu | Lidong Bing | Wei Lu

Distantly supervised named entity recognition (DS-NER) has been proposed to exploit the automatically labeled training data instead of human annotations. The distantly annotated datasets are often noisy and contain a considerable number of false negatives. The recent approach uses a weighted sampling approach to select a subset of negative samples for training. However, it requires a good classifier to assign weights to the negative samples. In this paper, we propose a simple and straightforward approach for selecting the top negative samples that have high similarities with all the positive samples for training. Our method achieves consistent performance improvements on four distantly supervised NER datasets. Our analysis also shows that it is critical to differentiate the true negatives from the false negatives.

pdf abs
Prototype-Based Interpretability for Legal Citation Prediction
Chu Fei Luo | Rohan Bhambhoria | Samuel Dahan | Xiaodan Zhu

Deep learning has made significant progress in the past decade, and demonstrates potential to solve problems with extensive social impact. In high-stakes decision making areas such as law, experts often require interpretability for automatic systems to be utilized in practical settings. In this work, we attempt to address these requirements applied to the important problem of legal citation prediction (LCP). We design the task with parallels to the thought-process of lawyers, i.e., with reference to both precedents and legislative provisions. After initial experimental results, we refine the target citation predictions with the feedback of legal experts. Additionally, we introduce a prototype architecture to add interpretability, achieving strong performance while adhering to decision parameters used by lawyers. Our study builds on and leverages the state-of-the-art language processing models for law, while addressing vital considerations for high-stakes tasks with practical societal impact.

pdf abs
LMs stand their Ground: Investigating the Effect of Embodiment in Figurative Language Interpretation by Language Models
Philipp Wicke

Figurative language is a challenge for language models since its interpretation is based on the use of words in a way that deviates from their conventional order and meaning. Yet, humans can easily understand and interpret metaphors, similes or idioms as they can be derived from embodied metaphors. Language is a proxy for embodiment and if a metaphor is conventional and lexicalised, it becomes easier for a system without a body to make sense of embodied concepts. Yet, the intricate relation between embodiment and features such as concreteness or age of acquisition has not been studied in the context of figurative language interpretation concerning language models. Hence, the presented study shows how larger language models perform better at interpreting metaphoric sentences when the action of the metaphorical sentence is more embodied. The analysis rules out multicollinearity with other features (e.g. word length or concreteness) and provides initial evidence that larger language models conceptualise embodied concepts to a degree that facilitates figurative language understanding.

pdf abs
Making Better Use of Training Corpus: Retrieval-based Aspect Sentiment Triplet Extraction via Label Interpolation
Guoxin Yu | Lemao Liu | Haiyun Jiang | Shuming Shi | Xiang Ao

In this paper, we aim to adapt the idea of retrieval-based neural approaches to the Aspect Sentiment Triplet Extraction (ASTE) task. Different from previous studies retrieving semantic similar neighbors, the ASTE task has its specialized challenges when adapting, i.e., the purpose includes predicting the sentiment polarity and it is usually aspect-dependent. Semantic similar neighbors with different polarities will be infeasible even counterproductive. To tackle this issue, we propose a retrieval-based neural ASTE approach, named RLI (Retrieval-based Aspect Sentiment Triplet Extraction via Label Interpolation), which exploits the label information of neighbors. Given an aspect-opinion term pair, we retrieve semantic similar triplets from the training corpus and interpolate their label information into the augmented representation of the target pair. The retriever is jointly trained with the whole ASTE framework, and neighbors with both similar semantics and sentiments can be recalled with the aid of this distant supervision. In addition, we design a simple yet effective pre-train method for the retriever that implicitly encodes the label similarities. Extensive experiments and analysis on two widely-used benchmarks show that the proposed model establishes a new state-of-the-art on ASTE.

pdf abs
Multi-Domain Dialogue State Tracking with Disentangled Domain-Slot Attention
Longfei Yang | Jiyi Li | Sheng Li | Takahiro Shinozaki

As the core of task-oriented dialogue systems, dialogue state tracking (DST) is designed to track the dialogue state through the conversation between users and systems. Multi-domain DST has been an important challenge in which the dialogue states across multiple domains need to consider. In recent mainstream approaches, each domain and slot are aggregated and regarded as a single query feeding into attention with the dialogue history to obtain domain-slot specific representations. In this work, we propose disentangled domain-slot attention for multi-domain dialogue state tracking. The proposed approach disentangles the domain-slot specific information extraction in a flexible and context-dependent manner by separating the query about domains and slots in the attention component. Through a series of experiments on MultiWOZ 2.0 and MultiWOZ 2.4 datasets, we demonstrate that our proposed approach outperforms the standard multi-head attention with aggregated domain-slot query.

Diffusion models developed on top of powerful text-to-image generation models like Stable Diffusion achieve remarkable success in visual story generation. However, the best-performing approach considers historically generated results as flattened memory cells, ignoring the fact that not all preceding images contribute equally to the generation of the characters and scenes at the current stage. To address this, we present a simple method that improves the leading system with adaptive context modeling, which is not only incorporated in the encoder but also adopted as additional guidance in the sampling stage to boost the global consistency of the generated story. We evaluate our model on PororoSV and FlintstonesSV datasets and show that our approach achieves state-of-the-art FID scores on both story visualization and continuation scenarios. We conduct detailed model analysis and show that our model excels at generating semantically consistent images for stories.

We focus on dialogue reading comprehension (DRC) that extracts answers from dialogues. Compared to standard RC tasks, DRC has raised challenges because of the complex speaker information and noisy dialogue context. Essentially, the challenges come from the speaker-centric nature of dialogue utterances — an utterance is usually insufficient in its surface form, but requires to incorporate the role of its speaker and the dialogue context to fill the latent pragmatic and intention information. We propose to deal with these problems in two folds. First, we propose a new key-utterances-extracting method, which can realize more answer-contained utterances. Second, based on the extracted utterances, we then propose a Question-Interlocutor Scope Realized Graph (QuISG). QuISG involves the question and question-mentioning speaker as nodes. To realize interlocutor scopes, utterances are connected with corresponding speakers in the dialogue. Experiments on the benchmarks show that our method achieves state-of-the-art performance against previous works.

We study speech-to-speech translation (S2ST) that translates speech from one language into another language and focuses on building systems to support languages without standard text writing systems. We use English-Taiwanese Hokkien as a case study, and present an end-to-end solution from training data collection, modeling choices to benchmark dataset release. First, we present efforts on creating human annotated data, automatically mining data from large unlabeled speech datasets, and adopting pseudo-labeling to produce weakly supervised data. On the modeling, we take advantage of recent advances in applying self-supervised discrete representations as target for prediction in S2ST and show the effectiveness of leveraging additional text supervision from Mandarin, a language similar to Hokkien, in model training. Finally, we release an S2ST benchmark set to facilitate future research in this field.

Code execution is a fundamental aspect of programming language semantics that reflects the exact behavior of the code. However, most pre-trained models for code intelligence ignore the execution trace and only rely on source code and syntactic structures. In this paper, we investigate how well pre-trained models can understand and perform code execution. We develop a mutation-based data augmentation technique to create a large-scale and realistic Python dataset and task for code execution, which challenges existing models such as Codex. We then present CodeExecutor, a Transformer model that leverages code execution pre-training and curriculum learning to enhance its semantic comprehension. We evaluate CodeExecutor on code execution and show its promising performance and limitations. We also demonstrate its potential benefits for code intelligence tasks such as zero-shot code-to-code search and text-to-code generation. Our analysis provides insights into the learning and generalization abilities of pre-trained models for code execution.

It is crucial to automatically construct knowledge graphs (KGs) of diverse new relations to support knowledge discovery and broad applications. Previous KG construction methods, based on either crowdsourcing or text mining, are often limited to a small predefined set of relations due to manual cost or restrictions in text corpus. Recent research proposed to use pretrained language models (LMs) as implicit knowledge bases that accept knowledge queries with prompts. Yet, the implicit knowledge lacks many desirable properties of a full-scale symbolic KG, such as easy access, navigation, editing, and quality assurance. In this paper, we propose a new approach of harvesting massive KGs of arbitrary relations from pretrained LMs. With minimal input of a relation definition (a prompt and a few shot of example entity pairs), the approach efficiently searches in the vast entity pair space to extract diverse accurate knowledge of the desired relation. We develop an effective search-and-rescore mechanism for improved efficiency and accuracy. We deploy the approach to harvest KGs of over 400 new relations, from LMs of varying capacities such as RoBERTaNet. Extensive human and automatic evaluations show our approach manages to extract diverse accurate knowledge, including tuples of complex relations (e.g., “A is capable of but not good at B”). The resulting KGs as a symbolic interpretation of the source LMs also reveal new insights into the LMs’ knowledge capacities.

pdf abs
Sequential Path Signature Networks for Personalised Longitudinal Language Modeling
Talia Tseriotou | Adam Tsakalidis | Peter Foster | Terence Lyons | Maria Liakata

Longitudinal user modeling can provide a strong signal for various downstream tasks. Despite the rapid progress in representation learning, dynamic aspects of modelling individuals’ language have only been sparsely addressed. We present a novel extension of neural sequential models using the notion of path signatures from rough path theory, which constitute graduated summaries of continuous paths and have the ability to capture non-linearities in trajectories. By combining path signatures of users’ history with contextual neural representations and recursive neural networks we can produce compact time-sensitive user representations. Given the magnitude of mental health conditions with symptoms manifesting in language, we show the applicability of our approach on the task of identifying changes in individuals’ mood by analysing their online textual content. By directly integrating signature transforms of users’ history in the model architecture we jointly address the two most important aspects of the task, namely sequentiality and temporality. Our approach achieves state-of-the-art performance on macro-average F1 score on the two available datasets for the task, outperforming or performing on-par with state-of-the-art models utilising only historical posts and even outperforming prior models which also have access to future posts of users.

Recent studies have pointed out that many well-developed Visual Question Answering (VQA) systems suffer from bias problem. Despite the remarkable performance gained on In-Distribution (ID) datasets, the VQA model might merely capture the superficial correlation from question to answer rather than showing real reasoning abilities. Therefore, when switching to Out-of-Distribution (OOD) dataset, whose test distribution is unknown or even reversed with the training set, significant drops might be demonstrated. Although efforts have been devoted to easing the negative bias effect brought by language prior and analysing its inherent cause, they are still limited by the following two aspects. First, most current debiasing methods achieve promising OOD generalization ability with a major sacrifice of the ID performance. Second, existing researches are restricted by exploiting comprehensive biases, since weakening the language bias is mainly focused, while only a few works consider vision bias. In this paper, we investigate a straightforward way to mitigate bias problem for VQA task. Specifically, we reduce bias effect by subtracting bias score from standard VQA base score. Based on such a direct strategy, we design two bias learning branches to detect more bias information, which are combined with a dynamical constraint loss to alleviate the problem of over-correction and insufficient debiasing effect. We evaluate our method on the challenging VQA v2.0 and VQA-CP V2,0 datasets and the proposed method achievessignificant improvement.

Event Detection (ED) is a critical task that aims to identify events of certain types in plain text. Neural models have achieved great success on ED, thus coming with a desire for higher interpretability. Existing works mainly exploit words or phrases of the input text to explain models’ inner mechanisms. However, for ED, the event structure, comprising of an event trigger and a set of arguments, are more enlightening clues to explain model behaviors. To this end, we propose a Trigger-Argument based Explanation method (TAE), which can utilize event structure knowledge to uncover a faithful interpretation for the existing ED models at neuron level. Specifically, we design group, sparsity, support mechanisms to construct the event structure from structuralization, compactness, and faithfulness perspectives. We evaluate our model on the large-scale MAVEN and the widely-used ACE 2005 datasets, and observe that TAE is able to reveal the process by which the model predicts. Experimental results also demonstrate that TAE can not only improve the interpretability on standard evaluation metrics, but also effectively facilitate the human understanding.

pdf abs
Interactive Concept Learning for Uncovering Latent Themes in Large Text Collections
Maria Leonor Pacheco | Tunazzina Islam | Lyle Ungar | Ming Yin | Dan Goldwasser

Experts across diverse disciplines are often interested in making sense of large text collections. Traditionally, this challenge is approached either by noisy unsupervised techniques such as topic models, or by following a manual theme discovery process. In this paper, we expand the definition of a theme to account for more than just a word distribution, and include generalized concepts deemed relevant by domain experts. Then, we propose an interactive framework that receives and encodes expert feedback at different levels of abstraction. Our framework strikes a balance between automation and manual coding, allowing experts to maintain control of their study while reducing the manual effort required.

pdf abs
NormMark: A Weakly Supervised Markov Model for Socio-cultural Norm Discovery
Farhad Moghimifar | Shilin Qu | Tongtong Wu | Yuan-Fang Li | Gholamreza Haffari

Norms, which are culturally accepted guidelines for behaviours, can be integrated into conversational models to generate utterances that are appropriate for the socio-cultural context. Existing methods for norm recognition tend to focus only on surface-level features of dialogues and do not take into account the interactions within a conversation. To address this issue, we propose NormMark, a probabilistic generative Markov model to carry the latent features throughout a dialogue. These features are captured by discrete and continuous latent variables conditioned on the conversation history, and improve the model’s ability in norm recognition. The model is trainable on weakly annotated data using the variational technique. On a dataset with limited norm annotations, we show that our approach achieves higher F1 score, outperforming current state-of-the-art methods, including GPT3.

pdf abs
VoteTRANS: Detecting Adversarial Text without Training by Voting on Hard Labels of Transformations
Hoang-Quoc Nguyen-Son | Seira Hidano | Kazuhide Fukushima | Shinsaku Kiyomoto | Isao Echizen

Adversarial attacks reveal serious flaws in deep learning models. More dangerously, these attacks preserve the original meaning and escape human recognition. Existing methods for detecting these attacks need to be trained using original/adversarial data. In this paper, we propose detection without training by voting on hard labels from predictions of transformations, namely, VoteTRANS. Specifically, VoteTRANS detects adversarial text by comparing the hard labels of input text and its transformation. The evaluation demonstrates that VoteTRANS effectively detects adversarial text across various state-of-the-art attacks, models, and datasets.

Existing approaches in the vision-and-language pre-training (VLP) paradigm mainly deploy either fusion-based encoders or dual-encoders, failing to achieve both effectiveness and efficiency in downstream multimodal tasks. In this paper, we build a flexible VLP model by incorporating cross-modal fusions into a dual-encoder architecture, where the introduced fusion modules can be easily decoupled from the dual encoder so as to switch the model to a fusion-free one. To better absorb cross-modal features from the fusion modules, we design a cross-modal knowledge transfer strategy along with other comprehensive pre-training tasks to guide the training process, which can further strengthen both the fusion-based and fusion-free representation learning. Extensive experiments conducted on various downstream vision-language tasks show that our proposed model is well-equipped with effectiveness as well as efficiency, demonstrating a superior performance compared with other strong VLP models.

Transformer architectures are complex and their use in NLP, while it has engendered many successes, makes their interpretability or explainability challenging. Recent debates have shown that attention maps and attribution methods are unreliable (Pruthi et al., 2019; Brunner et al., 2019). In this paper, we present some of their limitations and introduce COCKATIEL, which successfully addresses some of them. COCKATIEL is a novel, post-hoc, concept-based, model-agnostic XAI technique that generates meaningful explanations from the last layer of a neural net model trained on an NLP classification task by using Non-Negative Matrix Factorization (NMF) to discover the concepts the model leverages to make predictions and by exploiting a Sensitivity Analysis to estimate accurately the importance of each of these concepts for the model. It does so without compromising the accuracy of the underlying model or requiring a new one to be trained. We conduct experiments in single and multi-aspect sentiment analysis tasks and we show COCKATIEL’s superior ability to discover concepts that align with humans’ on Transformer models without any supervision, we objectively verify the faithfulness of its explanations through fidelity metrics, and we showcase its ability to provide meaningful explanations in two different datasets. Our code is freely available: https://github.com/fanny-jourdan/cockatiel

pdf abs
Code-Switched Text Synthesis in Unseen Language Pairs
I-Hung Hsu | Avik Ray | Shubham Garg | Nanyun Peng | Jing Huang

Existing efforts on text synthesis for code-switching mostly require training on code-switched texts in the target language pairs, limiting the deployment of the models to cases lacking code-switched data. In this work, we study the problem of synthesizing code-switched texts for language pairs absent from the training data. We introduce GLOSS, a model built on top of a pre-trained multilingual machine translation model (PMMTM) with an additional code-switching module. This module, either an adapter or extra prefixes, learns code-switching patterns from code-switched data during training, while the primary component of GLOSS, i.e., the PMMTM, is frozen. The design of only adjusting the code-switching module prevents our model from overfitting to the constrained training data for code-switching. Hence, GLOSS exhibits the ability to generalize and synthesize code-switched texts across a broader spectrum of language pairs. Additionally, we develop a self-training algorithm on target language pairs further to enhance the reliability of GLOSS. Automatic evaluations on four language pairs show that GLOSS achieves at least 55% relative BLEU and METEOR scores improvements compared to strong baselines. Human evaluations on two language pairs further validate the success of GLOSS.

pdf abs
Imagination is All You Need! Curved Contrastive Learning for Abstract Sequence Modeling Utilized on Long Short-Term Dialogue Planning
Justus-Jonas Erker | Stefan Schaffer | Gerasimos Spanakis

Inspired by the curvature of space-time, we introduce Curved Contrastive Learning (CCL), a novel representation learning technique for learning the relative turn distance between utterance pairs in multi-turn dialogues. The resulting bi-encoder models can guide transformers as a response ranking model towards a goal in a zero-shot fashion by projecting the goal utterance and the corresponding reply candidates into a latent space. Here the cosine similarity indicates the distance/reachability of a candidate utterance toward the corresponding goal. Furthermore, we explore how these forward-entailing language representations can be utilized for assessing the likelihood of sequences by the entailment strength i.e. through the cosine similarity of its individual members (encoded separately) as an emergent property in the curved space. These non-local properties allow us to imagine the likelihood of future patterns in dialogues, specifically by ordering/identifying future goal utterances that are multiple turns away, given a dialogue context. As part of our analysis, we investigate characteristics that make conversations (un)plannable and find strong evidence of planning capability over multiple turns (in 61.56% over 3 turns) in conversations from the DailyDialog dataset. Finally, we show how we achieve higher efficiency in sequence modeling tasks compared to previous work thanks to our relativistic approach, where only the last utterance needs to be encoded and computed during inference.

pdf abs
Data-Efficient French Language Modeling with CamemBERTa
Wissam Antoun | Benoît Sagot | Djamé Seddah

Recent advances in NLP have significantly improved the performance of language models on a variety of tasks. While these advances are largely driven by the availability of large amounts of data and computational power, they also benefit from the development of better training methods and architectures. In this paper, we introduce CamemBERTa, a French DeBERTa model that builds upon the DeBERTaV3 architecture and training objective. We evaluate our model’s performance on a variety of French downstream tasks and datasets, including question answering, part-of-speech tagging, dependency parsing, named entity recognition, and the FLUE benchmark, and compare against CamemBERT, the state-of-the-art monolingual model for French. Our results show that, given the same amount of training tokens, our model outperforms BERT-based models trained with MLM on most tasks. Furthermore, our new model reaches similar or superior performance on downstream tasks compared to CamemBERT, despite being trained on only 30% of its total number of input tokens. In addition to our experimental results, we also publicly release the weights and code implementation of CamemBERTa, making it the first publicly available DeBERTaV3 model outside of the original paper and the first openly available implementation of a DeBERTaV3 training objective.

pdf abs
Coupling Large Language Models with Logic Programming for Robust and General Reasoning from Text
Zhun Yang | Adam Ishay | Joohyung Lee

While large language models (LLMs), such as GPT-3, appear to be robust and general, their reasoning ability is not at a level to compete with the best models trained for specific natural language reasoning problems. In this study, we observe that a large language model can serve as a highly effective few-shot semantic parser. It can convert natural language sentences into a logical form that serves as input for answer set programs, a logic-based declarative knowledge representation formalism. The combination results in a robust and general system that can handle multiple question-answering tasks without requiring retraining for each new task. It only needs a few examples to guide the LLM’s adaptation to a specific task, along with reusable ASP knowledge modules that can be applied to multiple tasks. We demonstrate that this method achieves state-of-the-art performance on several NLP benchmarks, including bAbI, StepGame, CLUTRR, and gSCAN. Additionally, it successfully tackles robot planning tasks that an LLM alone fails to solve.

While large language models (LLMs) have proven to be effective on a large variety of tasks, they are also known to hallucinate information. To measure whether an LLM prefers factually consistent continuations of its input, we propose a new benchmark called FIB (Factual Inconsistency Benchmark) that focuses on the task of summarization. Specifically, our benchmark involves comparing the scores an LLM assigns to a factually consistent versus a factually inconsistent summary for an input news article. For factually consistent summaries, we use human-written reference summaries that we manually verify as factually consistent. To generate summaries that are factually inconsistent, we generate summaries from a suite of summarization models that we have manually annotated as factually inconsistent. A model’s factual consistency is then measured according to its accuracy, i.e. the proportion of documents where it assigns a higher score to the factually consistent summary. To validate the usefulness of {pasted macro ‘BENCHMARK’}, we evaluate 23 large language models ranging from 1B to 176B parameters from six different model families including BLOOM and OPT. We find that existing LLMs generally assign a higher score to factually consistent summaries than to factually inconsistent summaries. However, if the factually inconsistent summaries occur verbatim in the document, then LLMs assign a higher score to these factually inconsistent summaries than factually consistent summaries. We validate design choices in our benchmark including the scoring method and source of distractor summaries.

pdf abs
Text Generation Model Enhanced with Semantic Information in Aspect Category Sentiment Analysis
Tu Tran | Kiyoaki Shirai | Natthawut Kertkeidkachorn

Aspect Category Sentiment Analysis (ACSA) is one of the main subtasks of sentiment analysis, which aims at predicting polarity over a given aspect category. Recently, generative methods emerge as an efficient way to utilize a pre-trained language model for solving ACSA. However, those methods fail to model relations of target words and opinion words in a sentence including multiple aspects. To tackle this problem, this paper proposes a method to incorporate Abstract Meaning Representation (AMR), which describes semantic representation of a sentence as a directed graph, into a text generation model. Furthermore, two regularizers are designed to guide cross attention weights allocation over AMR graphs. One is the identical regularizer that constrains attention weights of aligned nodes, the other is the entropy regularizer that helps the decoder generate tokens by heavily considering only a few related nodes in the AMR graph. Experimental results on three datasets show that the proposed method outperforms state-of-the-art methods, proving the effectiveness of our model.

pdf abs
Mind the Biases: Quantifying Cognitive Biases in Language Model Prompting
Ruixi Lin | Hwee Tou Ng

We advocate the importance of exposing uncertainty on results of language model prompting which display bias modes resembling cognitive biases, and propose to help users grasp the level of uncertainty via simple quantifying metrics. Cognitive biases in the human decision making process can lead to flawed responses when we are under uncertainty. Not surprisingly, we have seen biases in language models resembling cognitive biases as a result of training on biased textual data, raising dangers in downstream tasks that are centered around people’s lives if users trust their results too much. In this work, we reveal two bias modes leveraging cognitive biases when we prompt BERT, accompanied by two bias metrics. On a drug-drug interaction extraction task, our bias measurements reveal an error pattern similar to the availability bias when the labels for training prompts are imbalanced, and show that a toning-down transformation of the drug-drug description in a prompt can elicit a bias similar to the framing effect, warning users to distrust when prompting language models for answers.

pdf abs
CodePrompt: Task-Agnostic Prefix Tuning for Program and Language Generation
YunSeok Choi | Jee-Hyong Lee

In order to solve the inefficient parameter update and storage issues of fine-tuning in Natural Language Generation (NLG) tasks, prompt-tuning methods have emerged as lightweight alternatives. Furthermore, efforts to reduce the gap between pre-training and fine-tuning have shown successful results in low-resource settings. As large Pre-trained Language Models (PLMs) for Program and Language Generation (PLG) tasks are constantly being developed, prompt tuning methods are necessary for the tasks. However, due to the gap between pre-training and fine-tuning different from PLMs for natural language, a prompt tuning method that reflects the traits of PLM for program language is needed. In this paper, we propose a Task-Agnostic prompt tuning method for the PLG tasks, CodePrompt, that combines Input-Dependent Prompt Template (to bridge the gap between pre-training and fine-tuning of PLMs for program and language) and Corpus-Specific Prefix Tuning (to update the parameters of PLMs for program and language efficiently).Also, we propose a method to provide richer prefix word information for limited prefix lengths. We prove that our method is effective in three PLG tasks, not only in the full-data setting but also in the low-resource setting and cross-domain setting.

pdf abs
Honey, I Shrunk the Language: Language Model Behavior at Reduced Scale.
Vijeta Deshpande | Dan Pechi | Shree Thatte | Vladislav Lialin | Anna Rumshisky

In recent years, language models have drastically grown in size, and the abilities of these models have been shown to improve with scale. The majority of recent scaling laws studies focused on high-compute high-parameter count settings, leaving the question of when these abilities begin to emerge largely unanswered. In this paper, we investigate whether the effects of pre-training can be observed when the problem size is reduced, modeling a smaller, reduced-vocabulary language. We show the benefits of pre-training with masked language modeling (MLM) objective in models as small as 1.25M parameters, and establish a strong correlation between pre-training perplexity and downstream performance (GLUE benchmark). We examine downscaling effects, extending scaling laws to models as small as ~1M parameters. At this scale, we observe a break of the power law for compute-optimal models and show that the MLM loss does not scale smoothly with compute-cost (FLOPs) below 2.2 × 10¹⁵ FLOPs. We also find that adding layers does not always benefit downstream performance.Our filtered pre-training data, reduced English vocabulary, and code are available at https://github.com/text-machine-lab/mini_bertgithub.com/text-machine-lab/mini_bert

Federated Multilingual Neural Machine Translation (Fed-MNMT) has emerged as a promising paradigm for institutions with limited language resources. This approach allows multiple institutions to act as clients and train a unified model through model synchronization, rather than collecting sensitive data for centralized training. This significantly reduces the cost of corpus collection and preserves data privacy. However, as pre-trained language models (PLMs) continue to increase in size, the communication cost for transmitting parameters during synchronization has become a training speed bottleneck. In this paper, we propose a communication-efficient Fed-MNMT framework that addresses this issue by keeping PLMs frozen and only transferring lightweight adapter modules between clients. Since different language pairs exhibit substantial discrepancies in data distributions, adapter parameters of clients may conflict with each other. To tackle this, we explore various clustering strategies to group parameters for integration and mitigate the negative effects of conflicting parameters. Experimental results demonstrate that our framework reduces communication cost by over 98% while achieving similar or even better performance compared to competitive baselines. Further analysis reveals that clustering strategies effectively solve the problem of linguistic discrepancy and pruning adapter modules further improves communication efficiency.

pdf abs
Cross-task Knowledge Transfer for Extremely Weakly Supervised Text Classification
Seongmin Park | Kyungho Kim | Jihwa Lee

Text classification with extremely weak supervision (EWS) imposes stricter supervision constraints compared to regular weakly supervise classification. Absolutely no labeled training samples or hand-crafted rules specific to the evaluation data are allowed. Such restrictions limit state-of-the-art EWS classification methods to indirect weak labeling techniques that assign unnatural label uncertainty estimates. We present PLAT, a framework that creates weak labels by leveraging recent developments in zero-shot text classification. PLAT employs models trained for sub-tasks other than classification to label documents. Most importantly, PLAT refrains from assigning overly confident weak labels and improves soft-label training performance for downstream classifiers. Classifiers trained with PLAT significantly outperform those trained on weak labels generated by the previous state-of-the-art in extremely weakly supervised text classification.

pdf abs
GVdoc - Graph-based Visual DOcument Classification
Fnu Mohbat | Mohammed J Zaki | Catherine Finegan-Dollak | Ashish Verma

The robustness of a model for real-world deployment is decided by how well it performs on unseen data and distinguishes between in-domain and out-of-domain samples. Visual document classifiers have shown impressive performance on in-distribution test sets. However, they tend to have a hard time correctly classifying and differentiating out-of-distribution examples. Image-based classifiers lack the text component, whereas multi-modality transformer-based models face the token serialization problem in visual documents due to their diverse layouts. They also require a lot of computing power during inference, making them impractical for many real-world applications. We propose, GVdoc, a graph-based document classification model that addresses both of these challenges. Our approach generates a document graph based on its layout, and then trains a graph neural network to learn node and graph embeddings. Through experiments, we show that our model, even with fewer parameters, outperforms state-of-the-art models on out-of-distribution data while retaining comparable performance on the in-distribution test set.

Recently, the text-to-table generation task has attracted increasing attention due to its wide applications. In this aspect, the dominant model formalizes this task as a sequence-to-sequence generation task and serializes each table into a token sequence during training by concatenating all rows in a top-down order. However, it suffers from two serious defects: 1) the predefined order introduces a wrong bias during training, which highly penalizes shifts in the order between rows; 2) the error propagation problem becomes serious when the model outputs a long token sequence. In this paper, we first conduct a preliminary study to demonstrate the generation of most rows is order-insensitive. Furthermore, we propose a novel sequence-to-sequence&set text-to-table generation model. Specifically, in addition to a text encoder encoding the input text, our model is equipped with a table header generator to first output a table header, i.e., the first row of the table, in the manner of sequence generation. Then we use a table body generator with learnable row embeddings and column embeddings to generate a set of table body rows in parallel. Particularly, to deal with the issue that there is no correspondence between each generated table body row and target during training, we propose a target assignment strategy based on the bipartite matching between the first cells of generated table body rows and targets. Experiment results show that our model significantly surpasses the baselines, achieving state-of-the-art performance on commonly-used datasets.

pdf abs
Automatic Readability Assessment for Closely Related Languages
Joseph Marvin Imperial | Ekaterina Kochmar

In recent years, the main focus of research on automatic readability assessment (ARA) has shifted towards using expensive deep learning-based methods with the primary goal of increasing models’ accuracy. This, however, is rarely applicable for low-resource languages where traditional handcrafted features are still widely used due to the lack of existing NLP tools to extract deeper linguistic representations. In this work, we take a step back from the technical component and focus on how linguistic aspects such as mutual intelligibility or degree of language relatedness can improve ARA in a low-resource setting. We collect short stories written in three languages in the Philippines—Tagalog, Bikol, and Cebuano—to train readability assessment models and explore the interaction of data and features in various cross-lingual setups. Our results show that the inclusion of CrossNGO, a novel specialized feature exploiting n-gram overlap applied to languages with high mutual intelligibility, significantly improves the performance of ARA models compared to the use of off-the-shelf large multilingual language models alone. Consequently, when both linguistic representations are combined, we achieve state-of-the-art results for Tagalog and Cebuano, and baseline scores for ARA in Bikol.

A neural ranker plays an indispensable role in the de facto ‘retrieval & rerank’ pipeline, but its training still lags behind due to the weak negative mining during contrastive learning. Compared to retrievers boosted by self-adversarial (i.e., in-distribution) negative mining, the ranker’s heavy structure suffers from query-document combinatorial explosions, so it can only resort to the negative sampled by the fast yet out-of-distribution retriever. Thereby, the moderate negatives compose ineffective contrastive learning samples, becoming the main barrier to learning a robust ranker. To alleviate this, we propose a multi-adversarial training strategy that leverages multiple retrievers as generators to challenge a ranker, where i) diverse hard negatives from a joint distribution are prone to fool the ranker for more effective adversarial learning and ii) involving extensive out-of-distribution label noises renders the ranker against each noise distribution, leading to more challenging and robust contrastive learning. To evaluate our robust ranker (dubbed R2anker), we conduct experiments in various settings on the passage retrieval benchmarks, including BM25-reranking, full-ranking, retriever distillation, etc. The empirical results verify the new state-of-the-art effectiveness of our model.

pdf abs
Semi-Supervised Domain Adaptation for Emotion-Related Tasks
Mahshid Hosseini | Cornelia Caragea

Semi-supervised domain adaptation (SSDA) adopts a model trained from a label-rich source domain to a new but related domain with a few labels of target data. It is shown that, in an SSDA setting, a simple combination of domain adaptation (DA) with semi-supervised learning (SSL) techniques often fails to effectively utilize the target supervision and cannot address distribution shifts across different domains due to the training data bias toward the source-labeled samples. In this paper, inspired by the co-learning of multiple classifiers for the computer vision tasks, we propose to decompose the SSDA framework for emotion-related tasks into two subcomponents of unsupervised domain adaptation (UDA) from the source to the target domain and semi-supervised learning (SSL) in the target domain where the two models iteratively teach each other by interchanging their high confident predictions. We further propose a novel data cartography-based regularization technique for pseudo-label denoising that employs training dynamics to further hone our models’ performance. We publicly release our code.

pdf abs
Boosting Distress Support Dialogue Responses with Motivational Interviewing Strategy
Anuradha Welivita | Pearl Pu

AI-driven chatbots have become an emerging solution to address psychological distress. Due to the lack of psychotherapeutic data, researchers use dialogues scraped from online peer support forums to train them. But since the responses in such platforms are not given by professionals, they contain both conforming and non-conforming responses. In this work, we attempt to recognize these conforming and non-conforming response types present in online distress-support dialogues using labels adapted from a well-established behavioral coding scheme named Motivational Interviewing Treatment Integrity (MITI) code and show how some response types could be rephrased into a more MI adherent form that can, in turn, enable chatbot responses to be more compliant with the MI strategy. As a proof of concept, we build several rephrasers by fine-tuning Blender and GPT3 to rephrase MI non-adherent Advise without permission responses into Advise with permission. We show how this can be achieved with the construction of pseudo-parallel corpora avoiding costs for human labor. Through automatic and human evaluation we show that in the presence of less training data, techniques such as prompting and data augmentation can be used to produce substantially good rephrasings that reflect the intended style and preserve the content of the original text.

Since conventional knowledge embedding models cannot take full advantage of the abundant textual information, there have been extensive research efforts in enhancing knowledge embedding using texts. However, existing enhancement approaches cannot apply to temporal knowledge graphs (tKGs), which contain time-dependent event knowledge with complex temporal dynamics. Specifically, existing enhancement approaches often assume knowledge embedding is time-independent. In contrast, the entity embedding in tKG models usually evolves, which poses the challenge of aligning temporally relevant texts with entities. To this end, we propose to study enhancing temporal knowledge embedding with textual data in this paper. As an approach to this task, we propose Enhanced Temporal Knowledge Embeddings with Contextualized Language Representations (ECOLA), which takes the temporal aspect into account and injects textual information into temporal knowledge embedding. To evaluate ECOLA, we introduce three new datasets for training and evaluating ECOLA. Extensive experiments show that ECOLA significantly enhances temporal KG embedding models with up to 287% relative improvements regarding Hits@1 on the link prediction task. The code and models are publicly available on https://github.com/mayhugotong/ECOLA.

pdf abs
Gender-tuning: Empowering Fine-tuning for Debiasing Pre-trained Language Models
Somayeh Ghanbarzadeh | Yan Huang | Hamid Palangi | Radames Cruz Moreno | Hamed Khanpour

Recent studies have revealed that the widely-used Pre-trained Language Models (PLMs) propagate societal biases from the large unmoderated pre-training corpora. Existing solutions require debiasing training processes and datasets for debiasing, which are resource-intensive and costly. Furthermore, these methods hurt the PLMs’ performance on downstream tasks. In this study, we propose Gender-tuning, which debiases the PLMs through fine-tuning on downstream tasks’ datasets. For this aim, Gender-tuning integrates Masked Language Modeling (MLM) training objectives into fine-tuning’s training process. Comprehensive experiments show that Gender-tuning outperforms the state-of-the-art baselines in terms of average gender bias scores in PLMs while improving PLMs’ performance on downstream tasks solely using the downstream tasks’ dataset. Also, Gender-tuning is a deployable debiasing tool for any PLM that works with original fine-tuning.

In real-world applications, pre-trained language models are typically deployed on the cloud, allowing clients to upload data and perform compute-intensive inference remotely. To avoid sharing sensitive data directly with service providers, clients can upload numerical representations rather than plain text to the cloud. However, recent text reconstruction techniques have demonstrated that it is possible to transform representations into original words, suggesting that privacy risk remains. In this paper, we propose TextObfuscator, a novel framework for protecting inference privacy by applying random perturbations to clustered representations. The random perturbations make the representations indistinguishable from surrounding clustered representations, thus obscuring word information while retaining the original word functionality. To achieve this, we utilize prototypes to learn clustered representation, where tokens of similar functionality are encouraged to be closer to the same prototype during training. Additionally, we design different methods to find prototypes for token-level and sentence-level tasks, which can improve performance by incorporating semantic and task information. Experimental results on token and sentence classification tasks show that TextObfuscator achieves improvement over compared methods without increasing inference cost.

pdf abs
Mini-Model Adaptation: Efficiently Extending Pretrained Models to New Languages via Aligned Shallow Training
Kelly Marchisio | Patrick Lewis | Yihong Chen | Mikel Artetxe

Prior work shows that it is possible to expand pretrained Masked Language Models (MLMs) to new languages by learning a new set of embeddings, while keeping the transformer body frozen. Despite learning a small subset of parameters, this approach is not compute-efficient, as training the new embeddings requires a full forward and backward pass over the entire model. We propose mini-model adaptation, a compute-efficient alternative that builds a shallow mini-model from a fraction of a large model’s parameters. New language-specific embeddings can then be efficiently trained over the mini-model and plugged into the aligned large model for rapid cross-lingual transfer. We explore two approaches to learn mini-models: MINIJOINT, which jointly pretrains the primary model and the mini-model using a single transformer with a secondary MLM head at a middle layer; and MINIPOST, where we start from a regular pretrained model, build a mini-model by extracting and freezing a few layers, and learn a small number of parameters on top. Experiments on XNLI, MLQA and PAWS-X show that mini-model adaptation matches the performance of the standard approach using up to 2.3x less compute on average.

Prompt-based methods have shown their efficacy in transferring general knowledge within pre-trained language models (PLMs) for low-resource scenarios. Typically, prompt-based methods convert downstream tasks to cloze-style problems and map all labels to verbalizers.However, when applied to zero-shot entity and relation extraction, vanilla prompt-based methods may struggle with the limited coverage of verbalizers to labels and the slow inference speed. In this work, we propose a novel Discriminate Soft Prompts (DSP) approach to take advantage of the prompt-based methods to strengthen the transmission of general knowledge. Specifically, we develop a discriminative prompt method, which reformulates zero-shot tasks into token discrimination tasks without having to construct verbalizers.Furthermore, to improve the inference speed of the prompt-based methods, we design a soft prompt co-reference strategy, which leverages soft prompts to approximately refer to the vector representation of text tokens. The experimental results show that, our model outperforms baselines on two zero-shot entity recognition datasets with higher inference speed, and obtains a 7.5% average relation F1-score improvement over previous state-of-the-art models on Wiki-ZSL and FewRel.

pdf abs
Exploring Robust Overfitting for Pre-trained Language Models
Bin Zhu | Yanghui Rao

We identify the robust overfitting issue for pre-trained language models by showing that the robust test loss increases as the epoch grows. Through comprehensive exploration of the robust loss on the training set, we attribute robust overfitting to the model’s memorization of the adversarial training data. We attempt to mitigate robust overfitting by combining regularization methods with adversarial training. Following the philosophy that prevents the model from memorizing the adversarial data, we find that flooding, a regularization method with loss scaling, can mitigate robust overfitting for pre-trained language models. Eventually, we investigate the effect of flooding levels and evaluate the models’ adversarial robustness under textual attacks. Extensive experiments demonstrate that our methods can mitigate robust overfitting upon three top adversarial training methods and further promote adversarial robustness.

There has been great progress in unifying various table-to-text tasks using a single encoder-decoder model trained via multi-task learning (Xie et al., 2022).However, existing methods typically encode task information with a simple dataset name as a prefix to the encoder. This not only limits the effectiveness of multi-task learning, but also hinders the model’s ability to generalize to new domains or tasks that were not seen during training, which is crucial for real-world applications. In this paper, we propose compositional task configurations, a set of prompts prepended to the encoder to improve cross-task generalization of unified models. We design the task configurations to explicitly specify the task type, as well as its input and output types. We show that this not only allows the model to better learn shared knowledge across different tasks at training, but also allows us to control the model by composing new configurations that apply novel input-output combinations in a zero-shot manner. We demonstrate via experiments over ten table-to-text tasks that our method outperforms the UnifiedSKG baseline by noticeable margins in both in-domain and zero-shot settings, with average improvements of +0.5 and +12.6 from using a T5-large backbone, respectively.

pdf abs
D-CALM: A Dynamic Clustering-based Active Learning Approach for Mitigating Bias
Sabit Hassan | Malihe Alikhani

Despite recent advancements, NLP models continue to be vulnerable to bias. This bias often originates from the uneven distribution of real-world data and can propagate through the annotation process. Escalated integration of these models in our lives calls for methods to mitigate bias without overbearing annotation costs. While active learning (AL) has shown promise in training models with a small amount of annotated data, AL’s reliance on the model’s behavior for selective sampling can lead to an accumulation of unwanted bias rather than bias mitigation. However, infusing clustering with AL can overcome the bias issue of both AL and traditional annotation methods while exploiting AL’s annotation efficiency. In this paper, we propose a novel adaptive clustering-based active learning algorithm, D-CALM, that dynamically adjusts clustering and annotation efforts in response to an estimated classifier error-rate. Experiments on eight datasets for a diverse set of text classification tasks, including emotion, hatespeech, dialog act, and book type detection, demonstrate that our proposed algorithm significantly outperforms baseline AL approaches with both pretrained transformers and traditional Support Vector Machines. D-CALM showcases robustness against different measures of information gain and, as evident from our analysis of label and error distribution, can significantly reduce unwanted model bias.

pdf abs
Language Anisotropic Cross-Lingual Model Editing
Yang Xu | Yutai Hou | Wanxiang Che | Min Zhang

Multilingual pre-trained language models can learn task-specific abilities or memorize facts across multiple languages but inevitably make undesired predictions with specific inputs. Under similar observation, model editing aims to post-hoc calibrate a model targeted to specific inputs with keeping the model’s raw behavior. However, existing work only studies the monolingual scenario, which lacks the cross-lingual transferability to perform editing simultaneously across languages. In this work, we focus on cross-lingual model editing. Firstly, we define the cross-lingual model editing task and corresponding metrics, where an edit in one language propagates to the others. Next, we propose a framework to naturally adapt monolingual model editing approaches to the cross-lingual scenario using parallel corpus. Further, we propose language anisotropic editing to improve cross-lingual editing by amplifying different subsets of parameters for each language. On the newly defined cross-lingual model editing task, we empirically demonstrate the failure of monolingual baselines in propagating the edit to multiple languages and the effectiveness of the proposed language anisotropic model editing. Our code is publicly available at https://github.com/franklear/LiME.

pdf abs
Diverse Retrieval-Augmented In-Context Learning for Dialogue State Tracking
Brendan King | Jeffrey Flanigan

There has been significant interest in zero and few-shot learning for dialogue state tracking (DST) due to the high cost of collecting and annotating task-oriented dialogues. Recent work has demonstrated that in-context learning requires very little data and zero parameter updates, and even outperforms trained methods in the few-shot setting. We propose RefPyDST, which advances the state of the art with three advancements to in-context learning for DST.First, we formulate DST as a Python programming task, explicitly modeling language coreference as variable reference in Python. Second, since in-context learning depends highly on the context examples, we propose a method to retrieve a diverse set of relevant examples to improve performance. Finally, we introduce a novel re-weighting method during decoding that takes into account probabilities of competing surface forms, and produces a more accurate dialogue state prediction. We evaluate our approach using MultiWOZ and achieve state-of-the-art multi-domain joint-goal accuracy in zero and few-shot settings.

pdf abs
Pre-Trained Language-Meaning Models for Multilingual Parsing and Generation
Chunliu Wang | Huiyuan Lai | Malvina Nissim | Johan Bos

Pre-trained language models (PLMs) have achieved great success in NLP and have recently been used for tasks in computational semantics. However, these tasks do not fully benefit from PLMs since meaning representations are not explicitly included. We introduce multilingual pre-trained language-meaning models based on Discourse Representation Structures (DRSs), including meaning representations besides natural language texts in the same model, and design a new strategy to reduce the gap between the pre-training and fine-tuning objectives. Since DRSs are language neutral, cross-lingual transfer learning is adopted to further improve the performance of non-English tasks. Automatic evaluation results show that our approach achieves the best performance on both the multilingual DRS parsing and DRS-to-text generation tasks. Correlation analysis between automatic metrics and human judgements on the generation task further validates the effectiveness of our model. Human inspection reveals that out-of-vocabulary tokens are the main cause of erroneous results.

pdf abs
Multi-modal Sarcasm Generation: Dataset and Solution
Wenye Zhao | Qingbao Huang | Dongsheng Xu | Peizhi Zhao

As an interesting and challenging task, sarcasm generation has attracted widespread attention. Although very recent studies have made promising progress, none of them considers generating a sarcastic description for a given image - as what people are doing on Twitter. In this paper, we present a Multi-modal Sarcasm Generation (MSG) task: Given an image with hashtags that provide the sarcastic target, MSG aims to generate sarcastic descriptions like humans. Different from textual sarcasm generation, MSG is more challenging as it is difficult to accurately capture the key information from images, hashtags, and OCR tokens and exploit multi-modal incongruity to generate sarcastic descriptions. To support the research on MSG, we develop MuSG, a new dataset with 5000 images and related Twitter text. We also propose a multi-modal Transformer-based method as a solution to this MSG task. The input features are embedded in the common space and passed through the multi-modal Transformer layers to generate the sarcastic descriptions by the auto-regressive paradigm. Both automatic and manual evaluations demonstrate the superiority of our method. The dataset and code will be available soon.

Semi-supervised learning (SSL) is a popular setting aiming to effectively utilize unlabelled data to improve model performance in downstream natural language processing (NLP) tasks. Currently, there are two popular approaches to make use of the unlabelled data: Self-training (ST) and Task-adaptive pre-training (TAPT). ST uses a teacher model to assign pseudo-labels to the unlabelled data, while TAPT continues pre-training on the unlabelled data before fine-tuning. To the best of our knowledge, the effectiveness of TAPT in SSL tasks has not been systematically studied, and no previous work has directly compared TAPT and ST in terms of their ability to utilize the pool of unlabelled data. In this paper, we provide an extensive empirical study comparing five state-of-the-art ST approaches and TAPT across various NLP tasks and data sizes, including in- and out-of domain settings. Surprisingly, we find that TAPT is a strong and more robust SSL learner, even when using just a few hundred unlabelled samples or in the presence of domain shifts, compared to more sophisticated ST approaches, and tends to bring greater improvements in SSL than in fully-supervised settings. Our further analysis demonstrates the risks of using ST approaches when the size of labelled or unlabelled data is small or when domain shifts exist, and highlights TAPT as a potential solution.

pdf abs
Retrieval-Based Transformer for Table Augmentation
Michael Glass | Xueqing Wu | Ankita Rajaram Naik | Gaetano Rossiello | Alfio Gliozzo

Data preparation, also called data wrangling, is considered one of the most expensive and time-consuming steps when performing analytics or building machine learning models. Preparing data typically involves collecting and merging data from complex heterogeneous, and often large-scale data sources, such as data lakes. In this paper, we introduce a novel approach toward automatic data wrangling in an attempt to alleviate the effort of end-users, e.g. data analysts, in structuring dynamic views from data lakes in the form of tabular data. Given a corpus of tables, we propose a retrieval augmented transformer model that is self-trained for the table augmentation tasks of row/column population and data imputation. Our self-learning strategy consists in randomly ablating tables from the corpus and training the retrieval-based model with the objective of reconstructing the partial tables given as input with the original values or headers. We adopt this strategy to first train the dense neural retrieval model encoding portions of tables to vectors, and then the end-to-end model trained to perform table augmentation tasks. We test on EntiTables, the standard benchmark for table augmentation, as well as introduce a new benchmark to advance further research: WebTables. Our model consistently and substantially outperforms both supervised statistical methods and the current state-of-the-art transformer-based models.

pdf abs
ECG-QALM: Entity-Controlled Synthetic Text Generation using Contextual Q&A for NER
Karan Aggarwal | Henry Jin | Aitzaz Ahmad

Named Entity Recognition (NER) state-of-the-art methods requires high-quality labeled datasets. Issues such as scarcity of labeled data, under-representation of entities, and privacy concerns with using sensitive data for training, can be significant barriers. Generating synthetic data to train models is a promising solution to mitigate these problems. We propose ECG-QALM, a contextual question and answering approach using pre-trained language models to synthetically generate entity-controlled text. Generated text is then used to augment small labeled datasets for downstream NER tasks. We evaluate our method on two publicly available datasets. We find ECG-QALM is capable of producing full text samples with desired entities appearing in a controllable way, while retaining sentence coherence closest to the real world data. Evaluations on NER tasks show significant improvements (75% - 140%) in low-labeled data regimes.

pdf abs
Tokenization Impacts Multilingual Language Modeling: Assessing Vocabulary Allocation and Overlap Across Languages
Tomasz Limisiewicz | Jiří Balhar | David Mareček

Multilingual language models have recently gained attention as a promising solution for representing multiple languages in a single model. In this paper, we propose new criteria to evaluate the quality of lexical representation and vocabulary overlap observed in sub-word tokenizers.Our findings show that the overlap of vocabulary across languages can be actually detrimental to certain downstream tasks (POS, dependency tree labeling). In contrast, NER and sentence-level tasks (cross-lingual retrieval, NLI) benefit from sharing vocabulary. We also observe that the coverage of the language-specific tokens in the multilingual vocabulary significantly impacts the word-level tasks. Our study offers a deeper understanding of the role of tokenizers in multilingual language models and guidelines for future model developers to choose the most suitable tokenizer for their specific application before undertaking costly model pre-training.

In a real-world dialogue system, generated text must be truthful and informative while remaining fluent and adhering to a prescribed style. Satisfying these constraints simultaneously isdifficult for the two predominant paradigms in language generation: neural language modeling and rule-based generation. We describe a hybrid architecture for dialogue response generation that combines the strengths of both paradigms. The first component of this architecture is a rule-based content selection model defined using a new formal framework called dataflow transduction, which uses declarative rules to transduce a dialogue agent’s actions and their results (represented as dataflow graphs) into context-free grammars representing the space of contextually acceptable responses. The second component is a constrained decoding procedure that uses these grammars to constrain the output of a neural language model, which selects fluent utterances. Our experiments show that this system outperforms both rule-based and learned approaches in human evaluations of fluency, relevance, and truthfulness.

pdf abs
Know What I don’t Know: Handling Ambiguous and Unknown Questions for Text-to-SQL
Bing Wang | Yan Gao | Zhoujun Li | Jian-Guang Lou

The task of text-to-SQL aims to convert a natural language question into its corresponding SQL query within the context of relational tables. Existing text-to-SQL parsers generate a plausible SQL query for an arbitrary user question, thereby failing to correctly handle problematic user questions. To formalize this problem, we conduct a preliminary study on the observed ambiguous and unanswerable cases in text-to-SQL and summarize them into 6 feature categories. Correspondingly, we identify the causes behind each category and propose requirements for handling ambiguous and unanswerable questions. Following this study, we propose a simple yet effective counterfactual example generation approach that automatically produces ambiguous and unanswerable text-to-SQL examples. Furthermore, we propose a weakly supervised DTE (Detecting-Then-Explaining) model for error detection, localization, and explanation. Experimental results show that our model achieves the best result on both real-world examples and generated examples compared with various baselines. We release our data and code at: https://github.com/wbbeyourself/DTE.

pdf abs
Rethinking Document-Level Relation Extraction: A Reality Check
Jing Li | Yequan Wang | Shuai Zhang | Min Zhang

Recently, numerous efforts have continued to push up performance boundaries of document-level relation extraction (DocRE) and have claimed significant progress in DocRE. In this paper, we do not aim at proposing a novel model for DocRE. Instead, we take a closer look at the field to see if these performance gains are actually true. By taking a comprehensive literature review and a thorough examination of popular DocRE datasets, we find that these performance gains are achieved upon a strong or even untenable assumption in common: all named entities are perfectly localized, normalized, and typed in advance. Next, we construct four types of entity mention attacks to examine the robustness of typical DocRE models by behavioral probing. We also have a close check on model usability in a more realistic setting. Our findings reveal that most of current DocRE models are vulnerable to entity mention attacks and difficult to be deployed in real-world end-user NLP applications. Our study calls more attentions for future research to stop simplifying problem setups, and to model DocRE in the wild rather than in an unrealistic Utopian world.

pdf abs
Optimizing Test-Time Query Representations for Dense Retrieval
Mujeen Sung | Jungsoo Park | Jaewoo Kang | Danqi Chen | Jinhyuk Lee

Recent developments of dense retrieval rely on quality representations of queries and contexts from pre-trained query and context encoders. In this paper, we introduce TOUR (Test-Time Optimization of Query Representations), which further optimizes instance-level query representations guided by signals from test-time retrieval results. We leverage a cross-encoder re-ranker to provide fine-grained pseudo labels over retrieval results and iteratively optimize query representations with gradient descent. Our theoretical analysis reveals that TOUR can be viewed as a generalization of the classical Rocchio algorithm for pseudo relevance feedback, and we present two variants that leverage pseudo-labels as hard binary or soft continuous labels. We first apply TOUR on phrase retrieval with our proposed phrase re-ranker, and also evaluate its effectiveness on passage retrieval with an off-the-shelf re-ranker. TOUR greatly improves end-to-end open-domain question answering accuracy, as well as passage retrieval performance. TOUR also consistently improves direct re-ranking by up to 2.0% while running 1.3–2.4x faster with an efficient implementation.

As privacy issues are receiving increasing attention within the Natural Language Processing (NLP) community, numerous methods have been proposed to sanitize texts subject to differential privacy. However, the state-of-the-art text sanitization mechanisms based on a relaxed notion of metric local differential privacy (MLDP) do not apply to non-metric semantic similarity measures and cannot achieve good privacy-utility trade-offs. To address these limitations, we propose a novel Customized Text sanitization (CusText) mechanism based on the original 𝜖-differential privacy (DP) definition, which is compatible with any similarity measure.Moreover, CusText assigns each input token a customized output set to provide more advanced privacy protection at the token level.Extensive experiments on several benchmark datasets show that CusText achieves a better trade-off between privacy and utility than existing mechanisms.The code is available at https://github.com/sai4july/CusText.

pdf abs
LABO: Towards Learning Optimal Label Regularization via Bi-level Optimization
Peng Lu | Ahmad Rashid | Ivan Kobyzev | Mehdi Rezagholizadeh | Phillippe Langlais

Regularization techniques are crucial to improving the generalization performance and training efficiency of deep neural networks. Many deep learning algorithms rely on weight decay, dropout, batch/layer normalization to converge faster and generalize. Label Smoothing (LS) is another simple, versatile and efficient regularization which can be applied to various supervised classification tasks. Conventional LS, however, regardless of the training instance assumes that each non-target class is equally likely. In this work, we present a general framework for training with label regularization, which includes conventional LS but can also model instance-specific variants. Based on this formulation, we propose an efficient way of learning LAbel regularization by devising a Bi-level Optimization (LABO) problem. We derive a deterministic and interpretable solution of the inner loop as the optimal label smoothing without the need to store the parameters or the output of a trained model. Finally, we conduct extensive experiments and demonstrate our LABO consistently yields improvement over conventional label regularization on various fields, including seven machine translation and three image classification tasks across various neural network architectures while maintaining training efficiency.

pdf abs
Frustratingly Easy Label Projection for Cross-lingual Transfer
Yang Chen | Chao Jiang | Alan Ritter | Wei Xu

Translating training data into many languages has emerged as a practical solution for improving cross-lingual transfer. For tasks that involve span-level annotations, such as information extraction or question answering, an additional label projection step is required to map annotated spans onto the translated texts. Recently, a few efforts have utilized a simple mark-then-translate method to jointly perform translation and projection by inserting special markers around the labeled spans in the original sentence. However, as far as we are aware, no empirical analysis has been conducted on how this approach compares to traditional annotation projection based on word alignment. In this paper, we present an extensive empirical study across 57 languages and three tasks (QA, NER, and Event Extraction) to evaluate the effectiveness and limitations of both methods, filling an important gap in the literature. Experimental results show that our optimized version of mark-then-translate, which we call EasyProject, is easily applied to many languages and works surprisingly well, outperforming the more complex word alignment-based methods. We analyze several key factors that affect the end-task performance, and show EasyProject works well because it can accurately preserve label span boundaries after translation. We will publicly release all our code and data.

Hierarchical Text Classification (HTC) is an essential and challenging subtask of multi-label text classification with a taxonomic hierarchy. Recent advances in deep learning and pre-trained language models have led to significant breakthroughs in the HTC problem. However, despite their effectiveness, these methods are often restricted by a lack of domain knowledge, which leads them to make mistakes in a variety of situations. Generally, when manually classifying a specific document to the taxonomic hierarchy, experts make inference based on their prior knowledge and experience. For machines to achieve this capability, we propose a novel Knowledge-enabled Hierarchical Text Classification model (K-HTC), which incorporates knowledge graphs into HTC. Specifically, K-HTC innovatively integrates knowledge into both the text representation and hierarchical label learning process, addressing the knowledge limitations of traditional methods. Additionally, a novel knowledge-aware contrastive learning strategy is proposed to further exploit the information inherent in the data. Extensive experiments on two publicly available HTC datasets show the efficacy of our proposed method, and indicate the necessity of incorporating knowledge graphs in HTC tasks.

The multi-answer phenomenon, where a question may have multiple answers scattered in the document, can be well handled by humans but is challenging enough for machine reading comprehension (MRC) systems. Despite recent progress in multi-answer MRC, there lacks a systematic analysis of how this phenomenon arises and how to better address it. In this work, we design a taxonomy to categorize commonly-seen multi-answer MRC instances, with which we inspect three multi-answer datasets and analyze where the multi-answer challenge comes from. We further analyze how well different paradigms of current multi-answer MRC models deal with different types of multi-answer instances. We find that some paradigms capture well the key information in the questions while others better model the relation between questions and contexts. We thus explore strategies to make the best of the strengths of different paradigms. Experiments show that generation models can be a promising platform to incorporate different paradigms. Our annotations and code are released for further research.

pdf abs
An Exploration of Encoder-Decoder Approaches to Multi-Label Classification for Legal and Biomedical Text
Yova Kementchedjhieva | Ilias Chalkidis

Standard methods for multi-label text classification largely rely on encoder-only pre-trained language models, whereas encoder-decoder models have proven more effective in other classification tasks. In this study, we compare four methods for multi-label classification, two based on an encoder only, and two based on an encoder-decoder. We carry out experiments on four datasets—two in the legal domain and two in the biomedical domain, each with two levels of label granularity— and always depart from the same pre-trained model, T5. Our results show that encoder-decoder methods outperform encoder-only methods, with a growing advantage on more complex datasets and labeling schemes of finer granularity. Using encoder-decoder models in a non-autoregressive fashion, in particular, yields the best performance overall, so we further study this approach through ablations to better understand its strengths.

Lifelong learning (LL) is an important ability for NLP models to learn new tasks continuously. Architecture-based approaches are reported to be effective implementations for LL models. However, it is non-trivial to extend previous approaches to domain incremental LL scenarios since they either require access to task identities in the testing phase or cannot handle samples from unseen tasks. In this paper, we propose Diana: a dynamic architecture-based lifelong learning model that tries to learn a sequence of tasks with a prompt-enhanced language model. Four types of hierarchically organized prompts are used in Diana to capture knowledge from different granularities. Specifically, we dedicate task-level prompts to capture task-specific knowledge to retain high LL performances and maintain instance-level prompts to learn knowledge shared across input samples to improve the model’s generalization performance. Moreover, we dedicate separate prompts to explicitly model unseen tasks and introduce a set of prompt key vectors to facilitate knowledge sharing between tasks. Extensive experiments demonstrate that Diana outperforms state-of-the-art LL models, especially in handling unseen tasks.

Contrastive learning has recently shown great potential to improve text-based knowledge graph completion (KGC). In this paper, we propose to learn a more semantically structured entity representation space in text-based KGC via hard negatives mining. Specifically, we novelly leverage a sequence-to-sequence architecture to generate high-quality hard negatives. These negatives are sampled from the same decoding distributions as the anchor (or correct entity), inherently being semantically close to the anchor and thus enjoying good hardness. A self-information-enhanced contrasting strategy is further incorporated into the Seq2Seq generator to systematically diversify the produced negatives. Extensive experiments on three KGC benchmarks demonstrate the sound hardness and diversity of our generated negatives and the resulting performance superiority on KGC.

pdf abs
Visually-Enhanced Phrase Understanding
Tsu-Yuan Hsu | Chen-An Li | Chao-Wei Huang | Yun-Nung Chen

Large-scale vision-language pre-training has exhibited strong performance in various visual and textual understanding tasks. Recently, the textual encoders of multi-modal pre-trained models have been shown to generate high-quality textual representations, which often outperform models that are purely text-based, such as BERT. In this study, our objective is to utilize both textual and visual encoders of multi-modal pre-trained models to enhance language understanding tasks. We achieve this by generating an image associated with a textual prompt, thus enriching the representation of a phrase for downstream tasks. Results from experiments conducted on four benchmark datasets demonstrate that our proposed method, which leverages visually-enhanced text representations, significantly improves performance in the entity clustering task.

pdf abs
Reasoning in Large Language Models Through Symbolic Math Word Problems
Vedant Gaur | Nikunj Saunshi

Large language models (LLMs) have revolutionized NLP by solving downstream tasks with little to no labeled data. Despite their versatile abilities, the larger question of their ability to reason remains ill-understood. This paper addresses reasoning in math word problems (MWPs) by studying symbolic versions of the numeric problems, since a symbolic expression is a “concise explanation” of the numeric answer. We create and use a symbolic version of the SVAMP dataset and find that GPT-3’s davinci-002 model also has good zero-shot accuracy on symbolic MWPs. To evaluate the faithfulness of the model’s reasoning, we go beyond accuracy and additionally evaluate the alignment between the final answer and the outputted reasoning, which correspond to numeric and symbolic answers respectively for MWPs. We explore a self-prompting approach to encourage the symbolic reasoning to align with the numeric answer, thus equipping the LLM with the ability to provide a concise and verifiable reasoning and making it more interpretable. Surprisingly, self-prompting also improves the symbolic accuracy to be higher than both the numeric and symbolic accuracies, thus providing an ensembling effect. The SVAMP-Sym dataset will be released for future research on symbolic math problems.

pdf abs
It’s not Sexually Suggestive; It’s Educative | Separating Sex Education from Suggestive Content on TikTok videos
Enfa George | Mihai Surdeanu

We introduce SexTok, a multi-modal dataset composed of TikTok videos labeled as sexually suggestive (from the annotator’s point of view), sex-educational content, or neither. Such a dataset is necessary to address the challenge of distinguishing between sexually suggestive content and virtual sex education videos on TikTok. Children’s exposure to sexually suggestive videos has been shown to have adversarial effects on their development (Collins et al. 2017). Meanwhile, virtual sex education, especially on subjects that are more relevant to the LGBTQIA+ community, is very valuable (Mitchell et al. 2014). The platform’s current system removes/punishes some of both types of videos, even though they serve different purposes. Our dataset contains video URLs, and it is also audio transcribed. To validate its importance, we explore two transformer-based models for classifying the videos. Our preliminary results suggest that the task of distinguishing between these types of videos is learnable but challenging. These experiments suggest that this dataset is meaningful and invites further study on the subject.

pdf abs
Dynamic Structured Neural Topic Model with Self-Attention Mechanism
Nozomu Miyamoto | Masaru Isonuma | Sho Takase | Junichiro Mori | Ichiro Sakata

This study presents a dynamic structured neural topic model, which can handle the time-series development of topics while capturing their dependencies. Our model captures the topic branching and merging processes by modeling topic dependencies based on a self-attention mechanism. Additionally, we introduce citation regularization, which induces attention weights to represent citation relations by modeling text and citations jointly. Our model outperforms a prior dynamic embedded topic model regarding perplexity and coherence, while maintaining sufficient diversity across topics. Furthermore, we confirm that our model can potentially predict emerging topics from academic literature.

pdf abs
Hybrid-Regressive Paradigm for Accurate and Speed-Robust Neural Machine Translation
Qiang Wang | Xinhui Hu | Ming Chen

This work empirically confirms that non-autoregressive translation (NAT) is less robust in decoding batch size and hardware settings than autoregressive translation (AT). To address this issue, we demonstrate that prompting a small number of AT predictions can significantly reduce the performance gap between AT and NAT through synthetic experiments. Following this line, we propose hybrid-regressive translation (HRT), a two-stage translation prototype that combines the strengths of AT and NAT. Specifically, HRT first generates discontinuous sequences via autoregression (e.g., make a prediction for every k tokens, k>1) and then fills in all previously skipped tokens at once in a non-autoregressive manner. Experiments on five translation tasks show that HRT achieves comparable translation quality with AT while having at least 1.5x faster inference regardless of batch size and device. Additionally, HRT successfully inherits the sound characteristics of AT in the deep-encoder-shallow-decoder architecture, allowing for further speedup without BLEU loss.

pdf abs
Commonsense Knowledge Transfer for Pre-trained Language Models
Wangchunshu Zhou | Ronan Le Bras | Yejin Choi

Despite serving as the foundation models for a wide range of NLP benchmarks, pre-trained language models have shown limited capabilities of acquiring implicit commonsense knowledge from self-supervision alone, compared to learning linguistic and factual knowledge that appear more explicitly in the surface patterns in text. In this work, we introduce commonsense knowledge transfer, a framework to transfer the commonsense knowledge stored in a neural commonsense knowledge model to a general-purpose pre-trained language model. It first exploits general texts to form queries for extracting commonsense knowledge from the neural commonsense knowledge model and then refines the language model with two self-supervised objectives: commonsense mask infilling and commonsense relation prediction, which align human language with the underlying commonsense knowledge. Empirical results show that our approach consistently improves the model’s performance on downstream tasks that require commonsense reasoning. Moreover, we find that the improvement is more significant in the few-shot setting. This suggests that our approach helps language models better transfer to downstream tasks without extensive supervision by injecting commonsense knowledge into their parameters.

pdf abs
Shielded Representations: Protecting Sensitive Attributes Through Iterative Gradient-Based Projection
Shadi Iskander | Kira Radinsky | Yonatan Belinkov

Natural language processing models tend to learn and encode social biases present in the data. One popular approach for addressing such biases is to eliminate encoded information from the model’s representations. However, current methods are restricted to removing only linearly encoded information. In this work, we propose Iterative Gradient-Based Projection (IGBP), a novel method for removing non-linear encoded concepts from neural representations. Our method consists of iteratively training neural classifiers to predict a particular attribute we seek to eliminate, followed by a projection of the representation on a hypersurface, such that the classifiers become oblivious to the target attribute. We evaluate the effectiveness of our method on the task of removing gender and race information as sensitive attributes. Our results demonstrate that IGBP is effective in mitigating bias through intrinsic and extrinsic evaluations, with minimal impact on downstream task accuracy.

pdf abs
Focal Training and Tagger Decouple for Grammatical Error Correction
Minghuan Tan | Min Yang | Ruifeng Xu

In this paper, we investigate how to improve tagging-based Grammatical Error Correction models. We address two issues of current tagging-based approaches, label imbalance issue, and tagging entanglement issue. Then we propose to down-weight the loss of well-classified labels using Focal Loss and decouple the error detection layer from the label tagging layer through an extra self-attention-based matching module. Experiments over three latest Chinese Grammatical Error Correction datasets show that our proposed methods are effective. We further analyze choices of hyper-parameters for Focal Loss and inference tweaking.

Grammatical error correction (GEC) aims to correct errors in given sentences and is significant to many downstream natural language understanding tasks. Recent work introduces the idea of grammatical error detection (GED) to improve the GEC task performance. In contrast, these explicit multi-stage works propagate and amplify the problem of misclassification of the GED module. To introduce more convincing error type information, we propose an end-to-end framework in this paper, which Leverages Error Type (LET) information in the generation process. First, the input text is fed into a classification module to obtain the error type corresponding to each token. Then, we introduce the category information into the decoder’s input and cross-attention module in two ways, respectively. Experiments on various datasets show that our proposed method outperforms existing methods by a clear margin.

pdf abs
On the Role of Parallel Data in Cross-lingual Transfer Learning
Machel Reid | Mikel Artetxe

While prior work has established that the use of parallel data is conducive for cross-lingual learning, it is unclear if the improvements come from the data itself, or if it is the modeling of parallel interactions that matters. Exploring this, we examine the usage of unsupervised machine translation to generate synthetic parallel data, and compare it to supervised machine translation and gold parallel data. We find that even model generated parallel data can be useful for downstream tasks, in both a general setting (continued pretraining) as well as the task-specific setting (translate-train), although our best results are still obtained using real parallel data. Our findings suggest that existing multilingual models do not exploit the full potential of monolingual data, and prompt the community to reconsider the traditional categorization of cross-lingual learning approaches.

Attribute Value Extraction (AVE) aims to automatically obtain attribute value pairs from product descriptions to aid e-commerce. Despite the progressive performance of existing approaches in e-commerce platforms, they still suffer from two challenges: 1) difficulty in identifying values at different scales simultaneously; 2) easy confusion by some highly similar fine-grained attributes. This paper proposes a pre-training technique for AVE to address these issues. In particular, we first improve the conventional token-level masking strategy, guiding the language model to understand multi-scale values by recovering spans at the phrase and sentence level. Second, we apply clustering to build a challenging negative set for each example and design a pre-training objective based on contrastive learning to force the model to discriminate similar attributes. Comprehensive experiments show that our solution provides a significant improvement over traditional pre-trained models in the AVE task, and achieves state-of-the-art on four benchmarks.

pdf abs
Phrase Retrieval for Open Domain Conversational Question Answering with Conversational Dependency Modeling via Contrastive Learning
Soyeong Jeong | Jinheon Baek | Sung Ju Hwang | Jong Park

Open-Domain Conversational Question Answering (ODConvQA) aims at answering questions through a multi-turn conversation based on a retriever-reader pipeline, which retrieves passages and then predicts answers with them. However, such a pipeline approach not only makes the reader vulnerable to the errors propagated from the retriever, but also demands additional effort to develop both the retriever and the reader, which further makes it slower since they are not runnable in parallel. In this work, we propose a method to directly predict answers with a phrase retrieval scheme for a sequence of words, reducing the conventional two distinct subtasks into a single one. Also, for the first time, we study its capability for ODConvQA tasks. However, simply adopting it is largely problematic, due to the dependencies between previous and current turns in a conversation. To address this problem, we further introduce a novel contrastive learning strategy, making sure to reflect previous turns when retrieving the phrase for the current context, by maximizing representational similarities of consecutive turns in a conversation while minimizing irrelevant conversational contexts. We validate our model on two ODConvQA datasets, whose experimental results show that it substantially outperforms the relevant baselines with the retriever-reader. Code is available at: https://github.com/starsuzi/PRO-ConvQA.

pdf abs
Unlearning Bias in Language Models by Partitioning Gradients
Charles Yu | Sullam Jeoung | Anish Kasi | Pengfei Yu | Heng Ji

Recent research has shown that large-scale pretrained language models, specifically transformers, tend to exhibit issues relating to racism, sexism, religion bias, and toxicity in general. Unfortunately, these pretrained language models are used almost universally in downstream tasks, and natural language processing is often applied to make real-world predictions. Thus, debiasing these language models as early in development as possible is increasingly crucial for preventing unintentional harms caused by natural language systems. To this end, we propose a new technique called partitioned contrastive gradient unlearning (PCGU), a gray-box method for debiasing pretrained masked language models. PCGU aims to optimize only the weights that contribute most to a specific domain of bias, doing so by computing a first-order approximation based on the gradients of contrastive sentence pairs. Our experiments show that PCGU is both low-cost and seems particularly effective at pinpointing the sources of implicit social bias in large pretrained transformers. Although we train using PCGU in the gender-profession domain only, we find that doing so can also partially mitigate bias across other domains. All code for our implementation and experiments can be found at https://github.com/CharlesYu2000/PCGU-UnlearningBias.

pdf abs
Meta-training with Demonstration Retrieval for Efficient Few-shot Learning
Aaron Mueller | Kanika Narang | Lambert Mathias | Qifan Wang | Hamed Firooz

Large language models show impressive results on few-shot NLP tasks. However, these models are memory and computation-intensive. Meta-training allows one to leverage smaller models for few-shot generalization in a domain-general and task-agnostic manner; however, these methods alone results in models that may not have sufficient parameterization or knowledge to adapt quickly to a large variety of tasks. To overcome this issue, we propose meta-training with demonstration retrieval, where we use a dense passage retriever to retrieve semantically similar labeled demonstrations to each example for more varied supervision. By separating external knowledge from model parameters, we can use meta-training to train parameter-efficient models that generalize well on a larger variety of tasks. We construct a meta-training set from UnifiedQA and CrossFit, and propose a demonstration bank based on UnifiedQA tasks. To our knowledge, our work is the first to combine retrieval with meta-training, to use DPR models to retrieve demonstrations, and to leverage demonstrations from many tasks simultaneously, rather than randomly sampling demonstrations from the training set of the target task. Our approach outperforms a variety of targeted parameter-efficient and retrieval-augmented few-shot methods on QA, NLI, and text classification tasks (including SQuAD, QNLI, and TREC). Our approach can be meta-trained and fine-tuned quickly on a single GPU.

Compared to news and chat summarization, the development of meeting summarization is hugely decelerated by the limited data. To this end, we introduce a versatile Chinese meeting summarization dataset, dubbed VCSum, consisting of 239 real-life meetings, with a total duration of over 230 hours. We claim our dataset is versatile because we provide the annotations of topic segmentation, headlines, segmentation summaries, overall meeting summaries, and salient sentences for each meeting transcript. As such, the dataset can adapt to various summarization tasks or methods, including segmentation-based summarization, multi-granularity summarization and retrieval-then-generate summarization. Our analysis confirms the effectiveness and robustness of VCSum. We also provide a set of benchmark models regarding different downstream summarization tasks on VCSum to facilitate further research.

Collaboration increasingly happens online. This is especially true for large groups working on global tasks, with collaborators all around the globe. The size and distributed nature of such groups makes decision-making challenging. This paper proposes a set of dialog acts for the study of decision-making mechanisms in such groups, and provides a new annotated dataset based on real-world data from the public mail-archives of one such organisation – the Internet Engineering Task Force (IETF). We provide an initial data analysis showing that this dataset can be used to better understand decision-making in such organisations. Finally, we experiment with a preliminary transformer-based dialog act tagging model.

pdf abs
Negation Scope Refinement via Boundary Shift Loss
Yin Wu | Aixin Sun

Negation in natural language may affect many NLP applications, e.g., information extraction and sentiment analysis. The key sub-task of negation detection is negation scope resolution which aims to extract the portion of a sentence that is being negated by a negation cue (e.g., keyword “not” and never”) in the sentence. Due to the long spans, existing methods tend to make wrong predictions around the scope boundaries. In this paper, we propose a simple yet effective model named R-BSL which engages the Boundary Shift Loss to refine the predicted boundary. On multiple benchmark datasets, we show that the extremely simple R-BSL achieves best results.

Recent advances in QA pair generation (QAG) have raised interest in applying this technique to the educational field. However, the diversity of QA types remains a challenge despite its contributions to comprehensive learning and assessment of children. In this paper, we propose a QAG framework that enhances QA type diversity by producing different interrogative sentences and implicit/explicit answers. Our framework comprises a QFS-based answer generator, an iterative QA generator, and a relevancy-aware ranker. The two generators aim to expand the number of candidates while covering various types. The ranker trained on the in-context negative samples clarifies the top-N outputs based on the ranking score. Extensive evaluations and detailed analyses demonstrate that our approach outperforms previous state-of-the-art results by significant margins, achieving improved diversity and quality. Our task-oriented processes are consistent with real-world demand, which highlights our system’s high applicability.

pdf abs
Pulling Out All The Full Stops: Punctuation Sensitivity in Neural Machine Translation and Evaluation
Prathyusha Jwalapuram

Much of the work testing machine translation systems for robustness and sensitivity has been adversarial or tended towards testing noisy input such as spelling errors, or non-standard input such as dialects. In this work, we take a step back to investigate a sensitivity problem that can seem trivial and is often overlooked: punctuation. We perform basic sentence-final insertion and deletion perturbation tests with full stops, exclamation and questions marks across source languages and demonstrate a concerning finding: commercial, production-level machine translation systems are vulnerable to mere single punctuation insertion or deletion, resulting in unreliable translations. Moreover, we demonstrate that both string-based and model-based evaluation metrics also suffer from this vulnerability, producing significantly different scores when translations only differ in a single punctuation, with model-based metrics penalizing each punctuation differently. Our work calls into question the reliability of machine translation systems and their evaluation metrics, particularly for real-world use cases, where inconsistent punctuation is often the most common and the least disruptive noise.

We present a reality check on large language models and inspect the promise of retrieval-augmented language models in comparison. Such language models are semi-parametric, where models integrate model parameters and knowledge from external data sources to make their predictions, as opposed to the parametric nature of vanilla large language models. We give initial experimental findings that semi-parametric architectures can be enhanced with views, a query analyzer/planner, and provenance to make a significantly more powerful system for question answering in terms of accuracy and efficiency, and potentially for other NLP tasks.

pdf abs
Numeric Magnitude Comparison Effects in Large Language Models
Raj Shah | Vijay Marupudi | Reba Koenen | Khushi Bhardwaj | Sashank Varma

Large Language Models (LLMs) do not differentially represent numbers, which are pervasive in text. In contrast, neuroscience research has identified distinct neural representations for numbers and words. In this work, we investigate how well popular LLMs capture the magnitudes of numbers (e.g., that 4<5) from a behavioral lens. Prior research on the representational capabilities of LLMs evaluates whether they show human-level performance, for instance, high overall accuracy on standard benchmarks. Here, we ask a different question, one inspired by cognitive science: How closely do the number representations of LLMscorrespond to those of human language users, who typically demonstrate the distance, size, and ratio effects? We depend on a linking hypothesis to map the similarities among the model embeddings of number words and digits to human response times. The results reveal surprisingly human-like representations across language models of different architectures, despite the absence of the neural circuitry that directly supports these representations in the human brain. This research shows the utility of understanding LLMs using behavioral benchmarks and points the way to future work on the number of representations of LLMs and their cognitive plausibility.

pdf abs
Multi-Relational Probabilistic Event Representation Learning via Projected Gaussian Embedding
Linhai Zhang | Congzhi Zhang | Deyu Zhou

Event representation learning has been shown beneficial in various downstream tasks. Current event representation learning methods, which mainly focus on capturing the semantics of events via deterministic vector embeddings, have made notable progress. However, they ignore two important properties: the multiple relations between events and the uncertainty within events. In this paper, we propose a novel approach to learning multi-relational probabilistic event embeddings based on contrastive learning. Specifically, the proposed method consists of three major modules, a multi-relational event generation module to automatically generate multi-relational training data, a probabilistic event encoding module to model uncertainty of events by Gaussian density embeddings, and a relation-aware projection module to adapt unseen relations by projecting Gaussian embeddings into relation-aware subspaces. Moreover, a novel contrastive learning loss is elaborately designed for learning the multi-relational probabilistic embeddings. Since the existing benchmarks for event representation learning ignore relations and uncertainty of events, a novel dataset named MRPES is constructed to investigate whether multiple relations between events and uncertainty within events are learned. Experimental results show that the proposed approach outperforms other state-of-the-art baselines on both existing and newly constructed datasets.

pdf abs
PragmatiCQA: A Dataset for Pragmatic Question Answering in Conversations
Peng Qi | Nina Du | Christopher Manning | Jing Huang

Pragmatic reasoning about another speaker’s unspoken intent and state of mind is crucial to efficient and effective human communication. It is virtually omnipresent in conversations between humans, e.g., when someone asks “do you have a minute?”, instead of interpreting it literally as a query about your schedule, you understand that the speaker might have requests that take time, and respond accordingly. In this paper, we present PragmatiCQA, the first large-scale open-domain question answering (QA) dataset featuring 6873 QA pairs that explores pragmatic reasoning in conversations over a diverse set of topics. We designed innovative crowdsourcing mechanisms for interest-based and task-driven data collection to address the common issue of incentive misalignment between crowdworkers and potential users. To compare computational models’ capability at pragmatic reasoning, we also propose several quantitative metrics to evaluate question answering systems on PragmatiCQA. We find that state-of-the-art systems still struggle to perform human-like pragmatic reasoning, and highlight their limitations for future research.

pdf abs
Modular and On-demand Bias Mitigation with Attribute-Removal Subnetworks
Lukas Hauzenberger | Shahed Masoudian | Deepak Kumar | Markus Schedl | Navid Rekabsaz

Societal biases are reflected in large pre-trained language models and their fine-tuned versions on downstream tasks. Common in-processing bias mitigation approaches, such as adversarial training and mutual information removal, introduce additional optimization criteria, and update the model to reach a new debiased state. However, in practice, end-users and practitioners might prefer to switch back to the original model, or apply debiasing only on a specific subset of protected attributes. To enable this, we propose a novel modular bias mitigation approach, consisting of stand-alone highly sparse debiasing subnetworks, where each debiasing module can be integrated into the core model on-demand at inference time. Our approach draws from the concept of diff pruning, and proposes a novel training regime adaptable to various representation disentanglement optimizations. We conduct experiments on three classification tasks with gender, race, and age as protected attributes. The results show that our modular approach, while maintaining task performance, improves (or at least remains on-par with) the effectiveness of bias mitigation in comparison with baseline finetuning. Particularly on a two-attribute dataset, our approach with separately learned debiasing subnetworks shows effective utilization of either or both the subnetworks for selective bias mitigation.

pdf abs
Scientific Fact-Checking: A Survey of Resources and Approaches
Juraj Vladika | Florian Matthes

The task of fact-checking deals with assessing the veracity of factual claims based on credible evidence and background knowledge. In particular, scientific fact-checking is the variation of the task concerned with verifying claims rooted in scientific knowledge. This task has received significant attention due to the growing importance of scientific and health discussions on online platforms. Automated scientific fact-checking methods based on NLP can help combat the spread of misinformation, assist researchers in knowledge discovery, and help individuals understand new scientific breakthroughs. In this paper, we present a comprehensive survey of existing research in this emerging field and its related tasks. We provide a task description, discuss the construction process of existing datasets, and analyze proposed models and approaches. Based on our findings, we identify intriguing challenges and outline potential future directions to advance the field.

Sample-and-rank is a key decoding strategy for modern generation-based dialogue systems. It helps achieve diverse and high-quality responses by selecting an answer from a small pool of generated candidates. The current state-of-the-art ranking methods mainly use an encoding paradigm called Cross-Encoder, which separately encodes each context-candidate pair and ranks the candidates according to their fitness scores. However, Cross-Encoder repeatedly encodes the same lengthy context for each candidate, resulting in high computational costs. Poly-Encoder addresses the above problems by reducing the interaction between context and candidates, but with a price of performance drop. In this work, we develop a new paradigm called Uni-Encoder, that keeps the full attention over each pair as in Cross-Encoder while only encoding the context once, as in Poly-Encoder. Uni-Encoder encodes all the candidates with the context in one forward pass. We use the same positional embedding for all candidates to ensure they are treated equally and design a new attention mechanism to avoid confusion. Our Uni-Encoder can simulate other ranking paradigms using different attention and response concatenation methods. Extensive experiments show that our proposed paradigm achieves new state-of-the-art results on four benchmark datasets with high computational efficiency. For instance, it improves R10@1 by 2.9% with an approximately 4X faster inference speed on the Ubuntu V2 dataset.

pdf abs
DLAMA: A Framework for Curating Culturally Diverse Facts for Probing the Knowledge of Pretrained Language Models
Amr Keleg | Walid Magdy

A few benchmarking datasets have been released to evaluate the factual knowledge of pretrained language models. These benchmarks (e.g., LAMA, and ParaRel) are mainly developed in English and later are translated to form new multilingual versions (e.g., mLAMA, and mParaRel). Results on these multilingual benchmarks suggest that using English prompts to recall the facts from multilingual models usually yields significantly better and more consistent performance than using non-English prompts. Our analysis shows that mLAMA is biased toward facts from Western countries, which might affect the fairness of probing models. We propose a new framework for curating factual triples from Wikidata that are culturally diverse. A new benchmark DLAMA-v1 is built of factual triples from three pairs of contrasting cultures having a total of 78,259 triples from 20 relation predicates. The three pairs comprise facts representing the (Arab and Western), (Asian and Western), and (South American and Western) countries respectively. Having a more balanced benchmark (DLAMA-v1) supports that mBERT performs better on Western facts than non-Western ones, while monolingual Arabic, English, and Korean models tend to perform better on their culturally proximate facts. Moreover, both monolingual and multilingual models tend to make a prediction that is culturally or geographically relevant to the correct label, even if the prediction is wrong.

The multimodal emotion recognition in conversation task aims to predict the emotion label for a given utterance with its context and multiple modalities. Existing approaches achieve good results but also suffer from the following two limitations: 1) lacking modeling of diverse dependency ranges, i.e., long, short, and independent context-specific representations and without consideration of the different recognition difficulty for each utterance; 2) consistent treatment of the contribution for various modalities. To address the above challenges, we propose the Self-adaptive Context and Modal-interaction Modeling (SCMM) framework. We first design the context representation module, which consists of three submodules to model multiple contextual representations. Thereafter, we propose the modal-interaction module, including three interaction submodules to make full use of each modality. Finally, we come up with a self-adaptive path selection module to select an appropriate path in each module and integrate the features to obtain the final representation. Extensive experiments under four settings on three multimodal datasets, including IEMOCAP, MELD, and MOSEI, demonstrate that our proposed method outperforms the state-of-the-art approaches.

Conditional question answering on long documents aims to find probable answers and identify conditions that need to be satisfied to make the answers correct over long documents. Existing approaches solve this task by segmenting long documents into multiple sections, and attending information at global and local tokens to predict the answers and corresponding conditions. However, the natural structure of the document and discourse relations between sentences in each document section are ignored, which are crucial for condition retrieving across sections, as well as logical interaction over the question and conditions. To address this issue, this paper constructs a Structure-Discourse Hierarchical Graph (SDHG) and conducts bottom-up information propagation. Firstly we build the sentence-level discourse graphs for each section and encode the discourse relations by graph attention. Secondly, we construct a section-level structure graph based on natural structures, and conduct interactions over the question and contexts. Finally different levels of representations are integrated into jointly answer and condition decoding. The experiments on the benchmark ConditionalQA shows our approach gains over the prior state-of-the-art, by 3.0 EM score and 2.4 F1 score on answer measuring, as well as 2.2 EM score and 1.9 F1 score on jointly answer and condition measuring.

Warning: This paper contains content that may be offensive or upsetting. Understanding the harms and offensiveness of statements requires reasoning about the social and situational context in which statements are made. For example, the utterance “your English is very good” may implicitly signal an insult when uttered by a white man to a non-white colleague, but uttered by an ESL teacher to their student would be interpreted as a genuine compliment. Such contextual factors have been largely ignored by previous approaches to toxic language detection. We introduce COBRA frames, the first context-aware formalism for explaining the intents, reactions, and harms of offensive or biased statements grounded in their social and situational context. We create COBRACORPUS, a dataset of 33k potentially offensive statements paired with machine-generated contexts and free-text explanations of offensiveness, implied biases, speaker intents, and listener reactions. To study the contextual dynamics of offensiveness, we train models to generate COBRA explanations, with and without access to the context. We find that explanations by context-agnostic models are significantly worse than by context-aware ones, especially in situations where the context inverts the statement’s offensiveness (29% accuracy drop). Our work highlights the importance and feasibility of contextualized NLP by modeling social factors.

pdf abs
Distilling Calibrated Knowledge for Stance Detection
Yingjie Li | Cornelia Caragea

Stance detection aims to determine the position of an author toward a target and provides insights into people’s views on controversial topics such as marijuana legalization. Despite recent progress in this task, most existing approaches use hard labels (one-hot vectors) during training, which ignores meaningful signals among categories offered by soft labels. In this work, we explore knowledge distillation for stance detection and present a comprehensive analysis. Our contributions are: 1) we propose to use knowledge distillation over multiple generations in which a student is taken as a new teacher to transfer knowledge to a new fresh student; 2) we propose a novel dynamic temperature scaling for knowledge distillation to calibrate teacher predictions in each generation step. Extensive results on three stance detection datasets show that knowledge distillation benefits stance detection and a teacher is able to transfer knowledge to a student more smoothly via calibrated guiding signals. We publicly release our code to facilitate future research.

pdf abs
PTCSpell: Pre-trained Corrector Based on Character Shape and Pinyin for Chinese Spelling Correction
Xiao Wei | Jianbao Huang | Hang Yu | Qian Liu

Chinese spelling correction (CSC) is a challenging task with the goal of correcting each wrong character in Chinese texts. Incorrect characters in a Chinese text are mainly due to the similar shape and similar pronunciation of Chinese characters. Recently, the paradigm of pre-training and fine-tuning has achieved remarkable success in natural language processing. However, the pre-training objectives in existing methods are not tailored for the CSC task since they neglect the visual and phonetic properties of characters, resulting in suboptimal spelling correction. In this work, we propose to pre-train a new corrector named PTCSpell for the CSC task under the detector-corrector architecture. The corrector we propose has the following two improvements. First, we design two novel pre-training objectives to capture pronunciation and shape information in Chinese characters. Second, we propose a new strategy to tackle the issue that the detector’s prediction results mislead the corrector by balancing the loss of wrong characters and correct characters. Experiments on three benchmarks (i.e., SIGHAN 2013, 2014, and 2015) show that our model achieves an average of 5.8% F1 improvements at the correction level over state-of-the-art methods, verifying its effectiveness.

pdf abs
Disentangling Text Representation With Counter-Template For Unsupervised Opinion Summarization
Yanyue Zhang | Deyu Zhou

Approaches for unsupervised opinion summarization are generally based on the reconstruction model and generate a summary by decoding the aggregated representation of inputs. Recent work has shown that aggregating via simple average leads to vector degeneration, generating the generic summary. To tackle the challenge, some approaches select the inputs before aggregating. However, we argue that the selection is too coarse as not all information in each input is equally essential for the summary. For example, the content information such as “great coffee maker, easy to set up” is more valuable than the pattern such as “this is a great product”. Therefore, we propose a novel framework for unsupervised opinion summarization based on text representation disentanglement with counter-template. In specific, a disentangling module is added to the encoder-decoder architecture which decouples the input text representation into two parts: content and pattern. To capture the pattern information, a counter-template is utilized as supervision, which is automatically generated based on contrastive learning. Experimental results on two benchmark datasets show that the proposed approach outperforms the state-of-the-art baselines on both quality and stability.

Question generation (QG) is the task of generating a valid and fluent question based on a given context and the target answer. According to various purposes, even given the same context, instructors can ask questions about different concepts, and even the same concept can be written in different ways. However, the evaluation for QG usually depends on single reference-based similarity metrics, such as n-gram-based metric or learned metric, which is not sufficient to fully evaluate the potential of QG methods. To this end, we propose to paraphrase the reference question for a more robust QG evaluation. Using large language models such as GPT-3, we created semantically and syntactically diverse questions, then adopt the simple aggregation of the popular evaluation metrics as the final scores. Through our experiments, we found that using multiple (pseudo) references is more effective for QG evaluation while showing a higher correlation with human evaluations than evaluation with a single reference.

Recently, Contrastive Visual-Language Pre-training (CLIP) has demonstrated remarkable capability in various Visual Language Understanding (VLU) tasks. Yet, most CLIP-based methods require tasks-specific designs and sufficient training data. In this paper, we introduce a simple yet efficient paradigm for low-resource VLU named XtremeCLIP, which involves very few trainable parameters to improve the generalization ability of the trained models. In our XtremeCLIP framework, we reformulate a series of VLU tasks as a unified open-book affinity-matching problem. Furthermore, to handle the insufficient supervised signals in small datasets, we adopt contrastive learning to utilize the implicit sorting information of ground-truth labels to provide more supervised cues. Extensive experiments over multiple datasets on visual entailment, visual question answering, and image classification show that XtremeCLIP consistently outperforms existing baselines in low-resource settings.

Textual scene graph parsing has become increasingly important in various vision-language applications, including image caption evaluation and image retrieval. However, existing scene graph parsers that convert image captions into scene graphs often suffer from two types of errors. First, the generated scene graphs fail to capture the true semantics of the captions or the corresponding images, resulting in a lack of faithfulness. Second, the generated scene graphs have high inconsistency, with the same semantics represented by different annotations. To address these challenges, we propose a novel dataset, which involves re-annotating the captions in Visual Genome (VG) using a new intermediate representation called FACTUAL-MR. FACTUAL-MR can be directly converted into faithful and consistent scene graph annotations. Our experimental results clearly demonstrate that the parser trained on our dataset outperforms existing approaches in terms of faithfulness and consistency. This improvement leads to a significant performance boost in both image caption evaluation and zero-shot image retrieval tasks. Furthermore, we introduce a novel metric for measuring scene graph similarity, which, when combined with the improved scene graph parser, achieves state-of-the-art (SOTA) results on multiple benchmark datasets for the aforementioned tasks.

pdf abs
Target-Oriented Relation Alignment for Cross-Lingual Stance Detection
Ruike Zhang | Nan Xu | Hanxuan Yang | Yuan Tian | Wenji Mao

Stance detection is an important task in text mining and social media analytics, aiming to automatically identify the user’s attitude toward a specific target from text, and has wide applications in a variety of domains. Previous work on stance detection has mainly focused on monolingual setting. To address the problem of imbalanced language resources, cross-lingual stance detection is proposed to transfer the knowledge learned from a high-resource (source) language (typically English) to another low-resource (target) language. However, existing research on cross-lingual stance detection has ignored the inconsistency in the occurrences and distributions of targets between languages, which consequently degrades the performance of stance detection in low-resource languages. In this paper, we first identify the target inconsistency issue in cross-lingual stance detection, and propose a fine-grained Target-oriented Relation Alignment (TaRA) method for the task, which considers both target-level associations and language-level alignments. Specifically, we propose the Target Relation Graph to learn the in-language and cross-language target associations. We further devise the relation alignment strategy to enable knowledge transfer between semantically correlated targets across languages. Experimental results on the representative datasets demonstrate the effectiveness of our method compared to competitive methods under variant settings.

pdf abs
NonFactS: NonFactual Summary Generation for Factuality Evaluation in Document Summarization
Amir Soleimani | Christof Monz | Marcel Worring

Pre-trained abstractive summarization models can generate fluent summaries and achieve high ROUGE scores. Previous research has found that these models often generate summaries that are inconsistent with their context document and contain nonfactual information. To evaluate factuality in document summarization, a document-level Natural Language Inference (NLI) classifier can be used. However, training such a classifier requires large-scale high-quality factual and nonfactual samples. To that end, we introduce NonFactS, a data generation model, to synthesize nonfactual summaries given a context document and a human-annotated (reference) factual summary. Compared to previous methods, our nonfactual samples are more abstractive and more similar to their corresponding factual samples, resulting in state-of-the-art performance on two factuality evaluation benchmarks, FALSESUM and SUMMAC. Our experiments demonstrate that even without human-annotated summaries, NonFactS can use random sentences to generate nonfactual summaries and a classifier trained on these samples generalizes to out-of-domain documents.

pdf abs
When to Read Documents or QA History: On Unified and Selective Open-domain QA
Kyungjae Lee | Sang-eun Han | Seung-won Hwang | Moontae Lee

This paper studies the problem of open-domain question answering, with the aim of answering a diverse range of questions leveraging knowledge resources. Two types of sources, QA-pair and document corpora, have been actively leveraged with the following complementary strength. The former is highly precise when the paraphrase of given question q was seen and answered during training, often posed as a retrieval problem, while the latter generalizes better for unseen questions. A natural follow-up is thus leveraging both models, while a naive pipelining or integration approaches have failed to bring additional gains over either model alone. Our distinction is interpreting the problem as calibration, which estimates the confidence of predicted answers as an indicator to decide when to use a document or QA-pair corpus. The effectiveness of our method was validated on widely adopted benchmarks such as Natural Questions and TriviaQA.

pdf abs
Interpretable Automatic Fine-grained Inconsistency Detection in Text Summarization
Hou Pong Chan | Qi Zeng | Heng Ji

Existing factual consistency evaluation approaches for text summarization provide binary predictions and limited insights into the weakness of summarization systems. Therefore, we propose the task of fine-grained inconsistency detection, the goal of which is to predict the fine-grained types of factual errors in a summary. Motivated by how humans inspect factual inconsistency in summaries, we propose an interpretable fine-grained inconsistency detection model, FineGrainFact, which explicitly represents the facts in the documents and summaries with semantic frames extracted by semantic role labeling, and highlights the related semantic frames to predict inconsistency. The highlighted semantic frames help verify predicted error types and correct inconsistent summaries. Experiment results demonstrate that our model outperforms strong baselines and provides evidence to support or refute the summary.

pdf abs
A Multi-dimensional study on Bias in Vision-Language models
Gabriele Ruggeri | Debora Nozza

In recent years, joint Vision-Language (VL) models have increased in popularity and capability. Very few studies have attempted to investigate bias in VL models, even though it is a well-known issue in both individual modalities. This paper presents the first multi-dimensional analysis of bias in English VL models, focusing on gender, ethnicity, and age as dimensions. When subjects are input as images, pre-trained VL models complete a neutral template with a hurtful word 5% of the time, with higher percentages for female and young subjects. Bias presence in downstream models has been tested on Visual Question Answering. We developed a novel bias metric called the Vision-Language Association Test based on questions designed to elicit biased associations between stereotypical concepts and targets. Our findings demonstrate that pre-trained VL models contain biases that are perpetuated in downstream tasks.

pdf abs
Correction of Errors in Preference Ratings from Automated Metrics for Text Generation
Jan Deriu | Pius von Däniken | Don Tuggener | Mark Cieliebak

A major challenge in the field of Text Generation is evaluation: Human evaluations are cost-intensive, and automated metrics often display considerable disagreements with human judgments. In this paper, we propose to apply automated metrics for Text Generation in a preference-based evaluation protocol. The protocol features a statistical model that incorporates various levels of uncertainty to account for the error-proneness of the metrics. We show that existing metrics are generally over-confident in assigning significant differences between systems. As a remedy, the model allows to combine human ratings with automated ratings. We show that it can reduce the required amounts of human ratings to arrive at robust and statistically significant results by more than 50%, while yielding the same evaluation outcome as the pure human evaluation in 95% of cases. We showcase the benefits of the evaluation protocol for three text generation tasks: dialogue systems, machine translation, and text summarization.

pdf abs
PEER: Pre-training ELECTRA Extended by Ranking
Ru He | Wei Wang | Songfang Huang | Fei Huang

The BERT model and its variants have made great achievements in many downstream natural language processing tasks. The achievements of these models, however, demand highly expensive pre-training computation cost. To address this pre-training efficiency issue, the ELECTRA model is proposed to use a discriminator to perform replaced token detection (RTD) task, that is, to classify whether each input token is original or replaced by a generator. The RTD task performed by the ELECTRA accelerates pre-training so substantially, such that it is very challenging to further improve the pre-training efficiency established by the ELECTRA by using or adding other pre-training tasks, as the recent comprehensive study of Bajaj et al. (2022) summarizes. To further advance this pre-training efficiency frontier, in this paper we propose to extend the RTD task into a task of ranking input tokens according to K different quality levels. Essentially, we generalize the binary classifier in the ELECTRA into a K-level ranker to undertake a more precise task with negligible additional computation cost. Our extensive experiments show that our proposed method is able to outperform the state-of-the-art pre-training efficient models including ELECTRA in downstream GLUE tasks given the same computation cost.

Spoken language understanding (SLU) is a fundamental task in the task-oriented dialogue systems. However, the inevitable errors from automatic speech recognition (ASR) usually impair the understanding performance and lead to error propagation. Although there are some attempts to address this problem through contrastive learning, they (1) treat clean manual transcripts and ASR transcripts equally without discrimination in fine-tuning; (2) neglect the fact that the semantically similar pairs are still pushed away when applying contrastive learning; (3) suffer from the problem of Kullback–Leibler (KL) vanishing. In this paper, we propose Mutual Learning and Large-Margin Contrastive Learning (ML-LMCL), a novel framework for improving ASR robustness in SLU. Specifically, in fine-tuning, we apply mutual learning and train two SLU models on the manual transcripts and the ASR transcripts, respectively, aiming to iteratively share knowledge between these two models. We also introduce a distance polarization regularizer to avoid pushing away the intra-cluster pairs as much as possible. Moreover, we use a cyclical annealing schedule to mitigate KL vanishing issue. Experiments on three datasets show that ML-LMCL outperforms existing models and achieves new state-of-the-art performance.

In the target-oriented dialogue, the representation and achievement of targets are two interrelated essential issues. In current approaches, the target is typically supposed to be a single object represented as a word, which makes it relatively easy to achieve the target through dialogue with the help of a knowledge graph (KG). However, when the target has complex semantics, the existing knowledge graph is often incomplete in tracking complex semantic relations. This paper studies target-oriented dialog where the target is a topic sentence. We combine the methods of knowledge retrieval and relationship prediction to construct a context-related dynamic KG. On dynamic KG, we can track the implicit semantic paths in the speaker’s mind that may not exist in the existing KGs. In addition, we also designed a novel metric to evaluate the tracked path automatically. The experimental results show that our method can control the agent more logically and smoothly toward the complex target.

pdf abs
Chain of Thought Prompting Elicits Knowledge Augmentation
Dingjun Wu | Jing Zhang | Xinmei Huang

The knowledge-augmented deep learning paradigm refers to a paradigm in which domain knowledge is identified and integrated into deep models. Conventional methods typically employ task-specific approaches to gather external knowledge from various sources. In contrast, large language models are extensively pre-trained and can serve as a comprehensive source of external knowledge. In this paper, we propose CoT-KA, a Chain-of-Thought-based method that augments knowledge for deep learning. CoT-KA avoids the need for additional knowledge retrieval or knowledge reasoning models, as required in conventional augmentation methods. Our results demonstrate that CoT-KA outperforms both pure CoT-based methods and the non-augmented method across the majority of eleven publicly available benchmarks for various reasoning tasks.

Hybrid Question-Answering (HQA), which targets reasoning over tables and passages linked from table cells, has witnessed significant research in recent years. A common challenge in HQA and other passage-table QA datasets is that it is generally unrealistic to iterate over all table rows, columns, and linked passages to retrieve evidence. Such a challenge made it difficult for previous studies to show their reasoning ability in retrieving answers. To bridge this gap, we propose a novel Table-alignment-based Cell-selection and Reasoning model (TACR) for hybrid text and table QA, evaluated on the HybridQA and WikiTableQuestions datasets. In evidence retrieval, we design a table-question-alignment enhanced cell-selection method to retrieve fine-grained evidence. In answer reasoning, we incorporate a QA module that treats the row containing selected cells as context. Experimental results over the HybridQA and WikiTableQuestions (WTQ) datasets show that TACR achieves state-of-the-art results on cell selection and outperforms fine-grained evidence retrieval baselines on HybridQA, while achieving competitive performance on WTQ. We also conducted a detailed analysis to demonstrate that being able to align questions to tables in the cell-selection stage can result in important gains from experiments of over 90% table row and column selection accuracy, meanwhile also improving output explainability.

Pragmatic reference enables efficient interpersonal communication. Prior work uses simple reference games to test models of pragmatic reasoning, often with unidentified speakers and listeners. In practice, however, speakers’ sociocultural background shapes their pragmatic assumptions. For example, readers of this paper assume NLP refers to Natural Language Processing, and not “Neuro-linguistic Programming.” This work introduces the Cultural Codes dataset, which operationalizes sociocultural pragmatic inference in a simple word reference game. Cultural Codes is based on the multi-turn collaborative two-player game, Codenames Duet. Our dataset consists of 794 games with 7,703 turns, distributed across 153 unique players. Alongside gameplay, we collect information about players’ personalities, values, and demographics. Utilizing theories of communication and pragmatics, we predict each player’s actions via joint modeling of their sociocultural priors and the game context. Our experiments show that accounting for background characteristics significantly improves model performance for tasks related to both clue-giving and guessing, indicating that sociocultural priors play a vital role in gameplay decisions.

Persuasion modeling is a key building block for conversational agents. Existing works in this direction are limited to analyzing textual dialogue corpus. We argue that visual signals also play an important role in understanding human persuasive behaviors. In this paper, we introduce the first multimodal dataset for modeling persuasion behaviors. Our dataset includes 199 dialogue transcriptions and videos captured in a multi-player social deduction game setting, 26,647 utterance level annotations of persuasion strategy, and game level annotations of deduction game outcomes. We provide extensive experiments to show how dialogue context and visual signals benefit persuasion strategy prediction. We also explore the generalization ability of language models for persuasion modeling and the role of persuasion strategies in predicting social deduction game outcomes. Our dataset can be found at https://persuasion-deductiongame. socialai-data.org. The codes and models are available at https://github.com/SALT-NLP/PersuationGames.

pdf abs
Long to reign over us: A Case Study of Machine Translation and a New Monarch
Rebecca Knowles | Samuel Larkin

Novel terminology and changes in terminology are often a challenge for machine translation systems. The passing of Queen Elizabeth II and the accession of King Charles III provide a striking example of translation shift in the real world, particularly in translation contexts that have ambiguity. Examining translation between French and English, we present a focused case-study of translations about King Charles III as produced both by publicly-available MT systems and by a neural machine translation system trained specifically on Canadian parliamentary text. We find that even in cases where human translators would have adequate context to disambiguate terms from the source language, machine translation systems do not always produce the expected output. Where we are able to analyze the training data, we note that this may represent artifacts in the data, raising important questions about machine translation updates in light of real world events.

pdf abs
A Unified Generative Approach to Product Attribute-Value Identification
Keiji Shinzato | Naoki Yoshinaga | Yandi Xia | Wei-Te Chen

Product attribute-value identification (PAVI) has been studied to link products on e-commerce sites with their attribute values (e.g., ⟨Material, Cotton⟩) using product text as clues. Technical demands from real-world e-commerce platforms require PAVI methods to handle unseen values, multi-attribute values, and canonicalized values, which are only partly addressed in existing extraction- and classification-based approaches. Motivated by this, we explore a generative approach to the PAVI task. We finetune a pre-trained generative model, T5, to decode a set of attribute-value pairs as a target sequence from the given product text. Since the attribute value pairs are unordered set elements, how to linearize them will matter; we, thus, explore methods of composing an attribute-value pair and ordering the pairs for the task. Experimental results confirm that our generation-based approach outperforms the existing extraction and classification-based methods on large-scale real-world datasets meant for those methods.

We present in this work a new Universal Morphology dataset for Korean. Previously, the Korean language has been underrepresented in the field of morphological paradigms amongst hundreds of diverse world languages. Hence, we propose this Universal Morphological paradigms for the Korean language that preserve its distinct characteristics. For our K-UniMorph dataset, we outline each grammatical criterion in detail for the verbal endings, clarify how to extract inflected forms, and demonstrate how we generate the morphological schemata. This dataset adopts morphological feature schema from CITATION and CITATION for the Korean language as we extract inflected verb forms from the Sejong morphologically analyzed corpus that is one of the largest annotated corpora for Korean. During the data creation, our methodology also includes investigating the correctness of the conversion from the Sejong corpus. Furthermore, we carry out the inflection task using three different Korean word forms: letters, syllables and morphemes. Finally, we discuss and describe future perspectives on Korean morphological paradigms and the dataset.

pdf abs
How does the brain process syntactic structure while listening?
Subba Reddy Oota | Mounika Marreddy | Manish Gupta | Raju Bapi

Syntactic parsing is the task of assigning a syntactic structure to a sentence. There are two popular syntactic parsing methods: constituency and dependency parsing. Recent works have used syntactic embeddings based on constituency trees, incremental top-down parsing, and other word syntactic features for brain activity prediction given the text stimuli to study how the syntax structure is represented in the brain’s language network. However, the effectiveness of dependency parse trees or the relative predictive power of the various syntax parsers across brain areas, especially for the listening task, is yet unexplored. In this study, we investigate the predictive power of the brain encoding models in three settings: (i) individual performance of the constituency and dependency syntactic parsing based embedding methods, (ii) efficacy of these syntactic parsing based embedding methods when controlling for basic syntactic signals, (iii) relative effectiveness of each of the syntactic embedding methods when controlling for the other. Further, we explore the relative importance of syntactic information (from these syntactic embedding methods) versus semantic information using BERT embeddings. We find that constituency parsers help explain activations in the temporal lobe and middle-frontal gyrus, while dependency parsers better encode syntactic structure in the angular gyrus and posterior cingulate cortex. Although semantic signals from BERT are more effective compared to any of the syntactic features or embedding methods, syntactic embedding methods explain additional variance for a few brain regions.

pdf abs
Towards Imperceptible Document Manipulations against Neural Ranking Models
Xuanang Chen | Ben He | Zheng Ye | Le Sun | Yingfei Sun

Adversarial attacks have gained traction in order to identify vulnerabilities in neural ranking models (NRMs), but current attack methods often introduce noticeable errors. Moreover, current methods rely heavily on using a well-imitated surrogate NRM to guarantee the attack effect, making them difficult to use in practice. This paper proposes a framework called Imperceptible DocumEnt Manipulation (IDEM) to produce adversarial documents that are less noticeable to both algorithms and humans. IDEM instructs a well-established generative language model like BART to generate error-free connection sentences, and employs a separate position-wise merging strategy to balance between relevance and coherence of the perturbed text. Evaluation results on the MS MARCO benchmark demonstrate that IDEM outperforms strong baselines while preserving fluency and correctness of the target documents. Furthermore, the separation of adversarial text generation from the surrogate NRM makes IDEM more robust and less affected by the quality of the surrogate NRM.

pdf abs
Ask an Expert: Leveraging Language Models to Improve Strategic Reasoning in Goal-Oriented Dialogue Models
Qiang Zhang | Jason Naradowsky | Yusuke Miyao

Existing dialogue models may encounter scenarios which are not well-represented in the training data, and as a result generate responses that are unnatural, inappropriate, or unhelpful. We propose the “Ask an Expert” framework in which the model is trained with access to an “expert” which it can consult at each turn. Advice is solicited via a structured dialogue with the expert, and the model is optimized to selectively utilize (or ignore) it given the context and dialogue history. In this work the expert takes the form of an LLM.We evaluate this framework in a mental health support domain, where the structure of the expert conversation is outlined by pre-specified prompts which reflect a reasoning strategy taught to practitioners in the field. Blenderbot models utilizing “Ask an Expert” show quality improvements across all expert sizes, including those with fewer parameters than the dialogue model itself. Our best model provides a ~10% improvement over baselines, approaching human-level scores on “engingingness” and “helpfulness” metrics.

pdf abs
SciReviewGen: A Large-scale Dataset for Automatic Literature Review Generation
Tetsu Kasanishi | Masaru Isonuma | Junichiro Mori | Ichiro Sakata

Automatic literature review generation is one of the most challenging tasks in natural language processing. Although large language models have tackled literature review generation, the absence of large-scale datasets has been a stumbling block to the progress. We release SciReviewGen, consisting of over 10,000 literature reviews and 690,000 papers cited in the reviews. Based on the dataset, we evaluate recent transformer-based summarization models on the literature review generation task, including Fusion-in-Decoder extended for literature review generation. Human evaluation results show that some machine-generated summaries are comparable to human-written reviews, while revealing the challenges of automatic literature review generation such as hallucinations and a lack of detailed information. Our dataset and code are available at [https://github.com/tetsu9923/SciReviewGen](https://github.com/tetsu9923/SciReviewGen).

Knowing exactly how many data points need to be labeled to achieve a certain model performance is a hugely beneficial step towards reducing the overall budgets for annotation. It pertains to both active learning and traditional data annotation, and is particularly beneficial for low resource scenarios. Nevertheless, it remains a largely under-explored area of research in NLP. We therefore explored various techniques for estimating the training sample size necessary to achieve a targeted performance value. We derived a simple yet effective approach to predict the maximum achievable model performance based on small amount of training samples – which serves as an early indicator during data annotation for data quality and sample size determination. We performed ablation studies on four language understanding tasks, and showed that the proposed approach allows us to forecast model performance within a small margin of mean absolute error (~0.9%) with only 10% data.

pdf abs
TransESC: Smoothing Emotional Support Conversation via Turn-Level State Transition
Weixiang Zhao | Yanyan Zhao | Shilong Wang | Bing Qin

Emotion Support Conversation (ESC) is an emerging and challenging task with the goal of reducing the emotional distress of people. Previous attempts fail to maintain smooth transitions between utterances in ESC because they ignoring to grasp the fine-grained transition information at each dialogue turn. To solve this problem, we propose to take into account turn-level state Transitions of ESC (TransESC) from three perspectives, including semantics transition, strategy transition and emotion transition, to drive the conversation in a smooth and natural way. Specifically, we construct the state transition graph with a two-step way, named transit-then-interact, to grasp such three types of turn-level transition information. Finally, they are injected into the transition aware decoder to generate more engaging responses. Both automatic and human evaluations on the benchmark dataset demonstrate the superiority of TransESC to generate more smooth and effective supportive responses. Our source code will be publicly available.

Prompt tuning is one of the successful approaches for parameter-efficient tuning of pre-trained language models. Despite being arguably the most parameter-efficient (tuned soft prompts constitute <0.1% of total parameters), it typically performs worse than other efficient tuning methods and is quite sensitive to hyper-parameters. In this work, we introduce Residual Prompt Tuning - a simple and efficient method that significantly improves the performance and stability of prompt tuning. We propose to reparameterize soft prompt embeddings using a shallow network with a residual connection. Our experiments show that Residual Prompt Tuning significantly outperforms prompt tuning across T5-Large, T5-Base and BERT-Base models. Notably, our method reaches +7 points improvement over prompt tuning on SuperGLUE benchmark with T5-Base model and allows to reduce the prompt length by 10 times without hurting performance. In addition, we show that our approach is robust to the choice of learning rate and prompt initialization, and is effective in few-shot settings.

pdf abs
Attend, Select and Eliminate: Accelerating Multi-turn Response Selection with Dual-attention-based Content Elimination
Jianxin Liang | Chang Liu | Chongyang Tao | Jiazhan Feng | Dongyan Zhao

Although the incorporation of pre-trained language models (PLMs) significantly pushes the research frontier of multi-turn response selection, it brings a new issue of heavy computation costs. To alleviate this problem and make the PLM-based response selection model both effective and efficient, we propose an inference framework together with a post-training strategy that builds upon any pre-trained transformer-based response selection models to accelerate inference by progressively selecting and eliminating unimportant content under the guidance of context-response dual-attention. Specifically, at each transformer layer, we first identify the importance of each word based on context-to-response and response-to-context attention, then select a number of unimportant words to be eliminated following a retention configuration derived from evolutionary search while passing the rest of the representations into deeper layers. To mitigate the training-inference gap posed by content elimination, we introduce a post-training strategy where we use knowledge distillation to force the model with progressively eliminated content to mimic the predictions of the original model with no content elimination. Experiments on three benchmarks indicate that our method can effectively speeds-up SOTA models without much performance degradation and shows a better trade-off between speed and performance than previous methods.

pdf abs
Medical Dialogue Generation via Dual Flow Modeling
Kaishuai Xu | Wenjun Hou | Yi Cheng | Jian Wang | Wenjie Li

Medical dialogue systems (MDS) aim to provide patients with medical services, such as diagnosis and prescription. Since most patients cannot precisely describe their symptoms, dialogue understanding is challenging for MDS. Previous studies mainly addressed this by extracting the mentioned medical entities as critical dialogue history information. In this work, we argue that it is also essential to capture the transitions of the medical entities and the doctor’s dialogue acts in each turn, as they help the understanding of how the dialogue flows and enhance the prediction of the entities and dialogue acts to be adopted in the following turn. Correspondingly, we propose a Dual Flow enhanced Medical (DFMed) dialogue generation framework. It extracts the medical entities and dialogue acts used in the dialogue history and models their transitions with an entity-centric graph flow and a sequential act flow, respectively. We employ two sequential models to encode them and devise an interweaving component to enhance their interactions. Experiments on two datasets demonstrate that our method exceeds baselines in both automatic and manual evaluations.

Existing supervised sign language recognition systems rely on an abundance of well-annotated data. Instead, an unsupervised speech-to-sign language recognition (SSR-U) system learns to translate between spoken and sign languages by observing only non-parallel speech and sign-language corpora. We propose speech2sign-U, a neural network-based approach capable of both character-level and word-level SSR-U. Our approach significantly outperforms baselines directly adapted from unsupervised speech recognition (ASR-U) models by as much as 50% recall@10 on several challenging American sign language corpora with various levels of sample sizes, vocabulary sizes, and audio and visual variability. The code is available at https://github.com/cactuswiththoughts/UnsupSpeech2Sign.gitcactuswiththoughts/UnsupSpeech2Sign.git.

pdf abs
Distinguishing Address vs. Reference Mentions of Personal Names in Text
Vinodkumar Prabhakaran | Aida Mostafazadeh Davani | Melissa Ferguson | Stav Atir

Detecting named entities in text has long been a core NLP task. However, not much work has gone into distinguishing whether an entity mention is addressing the entity vs. referring to the entity; e.g., John, would you turn the light off? vs. John turned the light off. While this distinction is marked by a vocative case marker in some languages, many modern Indo-European languages such as English do not use such explicit vocative markers, and the distinction is left to be interpreted in context. In this paper, we present a new annotated dataset that captures the address vs. reference distinction in English, an automatic tagger that performs at 85% accuracy in making this distinction, and demonstrate how this distinction is important in NLP and computational social science applications in English language.

Deep neural networks (DNNs) are often used for text classification due to their high accuracy. However, DNNs can be computationally intensive, requiring millions of parameters and large amounts of labeled data, which can make them expensive to use, to optimize, and to transfer to out-of-distribution (OOD) cases in practice. In this paper, we propose a non-parametric alternative to DNNs that’s easy, lightweight, and universal in text classification: a combination of a simple compressor like gzip with a k-nearest-neighbor classifier. Without any training parameters, our method achieves results that are competitive with non-pretrained deep learning methods on six in-distribution datasets.It even outperforms BERT on all five OOD datasets, including four low-resource languages. Our method also excels in the few-shot setting, where labeled data are too scarce to train DNNs effectively.

pdf abs
LR-Sum: Summarization for Less-Resourced Languages
Chester Palen-Michel | Constantine Lignos

We introduce LR-Sum, a new permissively-licensed dataset created with the goal of enabling further research in automatic summarization for less-resourced languages.LR-Sum contains human-written summaries for 40 languages, many of which are less-resourced. We describe our process for extracting and filtering the dataset from the Multilingual Open Text corpus (Palen-Michel et al., 2022).The source data is public domain newswire collected from from Voice of America websites, and LR-Sum is released under a Creative Commons license (CC BY 4.0), making it one of the most openly-licensed multilingual summarization datasets. We describe abstractive and extractive summarization experiments to establish baselines and discuss the limitations of this dataset.

Existing metrics for evaluating the quality of automatically generated questions such as BLEU, ROUGE, BERTScore, and BLEURT compare the reference and predicted questions, providing a high score when there is a considerable lexical overlap or semantic similarity between the candidate and the reference questions. This approach has two major shortcomings. First, we need expensive human-provided reference questions. Second, it penalises valid questions that may not have high lexical or semantic similarity to the reference questions. In this paper, we propose a new metric, RQUGE, based on the answerability of the candidate question given the context. The metric consists of a question-answering and a span scorer modules, using pre-trained models from existing literature, thus it can be used without any further training. We demonstrate that RQUGE has a higher correlation with human judgment without relying on the reference question. Additionally, RQUGE is shown to be more robust to several adversarial corruptions. Furthermore, we illustrate that we can significantly improve the performance of QA models on out-of-domain datasets by fine-tuning on synthetic data generated by a question generation model and reranked by RQUGE.

pdf abs
Unsupervised Semantic Variation Prediction using the Distribution of Sibling Embeddings
Taichi Aida | Danushka Bollegala

Languages are dynamic entities, where the meanings associated with words constantly change with time. Detecting the semantic variation of words is an important task for various NLP applications that must make time-sensitive predictions. Existing work on semantic variation prediction have predominantly focused on comparing some form of an averaged contextualised representation of a target word computed from a given corpus. However, some of the previously associated meanings of a target word can become obsolete over time (e.g. meaning of gay as happy), while novel usages of existing words are observed (e.g. meaning of cell as a mobile phone).We argue that mean representations alone cannot accurately capture such semantic variations and propose a method that uses the entire cohort of the contextualised embeddings of the target word, which we refer to as the sibling distribution. Experimental results on SemEval-2020 Task 1 benchmark dataset for semantic variation prediction show that our method outperforms prior work that consider only the mean embeddings, and is comparable to the current state-of-the-art. Moreover, a qualitative analysis shows that our method detects important semantic changes in words that are not captured by the existing methods.

Learning multiscale Transformer models has been evidenced as a viable approach to augmenting machine translation systems. Prior research has primarily focused on treating subwords as basic units in developing such systems. However, the incorporation of fine-grained character-level features into multiscale Transformer has not yet been explored. In this work, we present a Slow-Fast two-stream learning model, referred to as TranSFormer, which utilizes a “slow” branch to deal with subword sequences and a “fast” branch to deal with longer character sequences. This model is efficient since the fast branch is very lightweight by reducing the model width, and yet provides useful fine-grained features for the slow branch. Our TranSFormer shows consistent BLEU improvements (larger than 1 BLEU point) on several machine translation benchmarks.

pdf abs
Mitigating the Learning Bias towards Repetition by Self-Contrastive Training for Open-Ended Generation
Jian Guan | Minlie Huang

Despite the huge progress in myriad generation tasks, pretrained language models (LMs) such as GPT2 still tend to generate repetitive texts with maximization-based decoding algorithms for open-ended generation. We attribute their overestimation of token-level repetition probabilities to the learning bias: LMs capture simple repetitive patterns faster with the MLE loss. We propose self-contrastive training to penalize the output of a premature checkpoint of the same model when it incorrectly predicts repetition, which is shown to mitigate repetition effectively while maintaining fluency on two datasets. Furthermore, we find that LMs use longer-range dependencies to predict repetitive tokens than non-repetitive ones, which may be the cause of sentence-level repetition loops.

pdf abs
Digging out Discrimination Information from Generated Samples for Robust Visual Question Answering
Zhiquan Wen | Yaowei Wang | Mingkui Tan | Qingyao Wu | Qi Wu

Visual Question Answering (VQA) aims to answer a textual question based on a given image. Nevertheless, recent studies have shown that VQA models tend to capture the biases to answer the question, instead of using the reasoning ability, resulting in poor generalisation ability. To alleviate the issue, some existing methods consider the natural distribution of the data, and construct samples to balance the dataset, achieving remarkable performance. However, these methods may encounter some limitations: 1) rely on additional annotations, 2) the generated samples may be inaccurate, e.g., assigned wrong answers, and 3) ignore the power of positive samples. In this paper, we propose a method to Dig out Discrimination information from Generated samples (DDG) to address the above limitations. Specifically, we first construct positive and negative samples in vision and language modalities, without using additional annotations. Then, we introduce a knowledge distillation mechanism to promote the learning of the original samples by the positive samples. Moreover, we impel the VQA models to focus on vision and language modalities using the negative samples. Experimental results on the VQA-CP v2 and VQA v2 datasets show the effectiveness of our DDG.

pdf abs
Words as Gatekeepers: Measuring Discipline-specific Terms and Meanings in Scholarly Publications
Li Lucy | Jesse Dodge | David Bamman | Katherine Keith

Scholarly text is often laden with jargon, or specialized language that can facilitate efficient in-group communication within fields but hinder understanding for out-groups. In this work, we develop and validate an interpretable approach for measuring scholarly jargon from text. Expanding the scope of prior work which focuses on word types, we use word sense induction to also identify words that are widespread but overloaded with different meanings across fields. We then estimate the prevalence of these discipline-specific words and senses across hundreds of subfields, and show that word senses provide a complementary, yet unique view of jargon alongside word types. We demonstrate the utility of our metrics for science of science and computational sociolinguistics by highlighting two key social implications. First, though most fields reduce their use of jargon when writing for general-purpose venues, and some fields (e.g., biological sciences) do so less than others. Second, the direction of correlation between jargon and citation rates varies among fields, but jargon is nearly always negatively correlated with interdisciplinary impact. Broadly, our findings suggest that though multidisciplinary venues intend to cater to more general audiences, some fields’ writing norms may act as barriers rather than bridges, and thus impede the dispersion of scholarly ideas.

pdf abs
Trade-Offs Between Fairness and Privacy in Language Modeling
Cleo Matzken | Steffen Eger | Ivan Habernal

Protecting privacy in contemporary NLP models is gaining in importance. So does the need to mitigate social biases of such models. But can we have both at the same time? Existing research suggests that privacy preservation comes at the price of worsening biases in classification tasks. In this paper, we explore the extent to which this tradeoff really holds when we incorporate both privacy preservation and de-biasing techniques into training text generation models. How does improving the model along one dimension affect the other dimension as well as the utility of the model? We conduct an extensive set of experiments that include bias detection, privacy attacks, language modeling, and performance on downstream tasks.

The cross-domain text-to-SQL task aims to build a system that can parse user questions into SQL on complete unseen databases, and the single-domain text-to-SQL task evaluates the performance on identical databases. Both of these setups confront unavoidable difficulties in real-world applications. To this end, we introduce the cross-schema text-to-SQL task, where the databases of evaluation data are different from that in the training data but come from the same domain. Furthermore, we present CSS, a large-scale CrosS-Schema Chinese text-to-SQL dataset, to carry on corresponding studies. CSS originally consisted of 4,340 question/SQL pairs across 2 databases. In order to generalize models to different medical systems, we extend CSS and create 19 new databases along with 29,280 corresponding dataset examples. Moreover, CSS is also a large corpus for single-domain Chinese text-to-SQL studies. We present the data collection approach and a series of analyses of the data statistics. To show the potential and usefulness of CSS, benchmarking baselines have been conducted and reported. Our dataset is publicly available at https://huggingface.co/datasets/zhanghanchong/css.

pdf abs
Silver Syntax Pre-training for Cross-Domain Relation Extraction
Elisa Bassignana | Filip Ginter | Sampo Pyysalo | Rob van der Goot | Barbara Plank

Relation Extraction (RE) remains a challenging task, especially when considering realistic out-of-domain evaluations. One of the main reasons for this is the limited training size of current RE datasets: obtaining high-quality (manually annotated) data is extremely expensive and cannot realistically be repeated for each new domain. An intermediate training step on data from related tasks has shown to be beneficial across many NLP tasks. However, this setup still requires supplementary annotated data, which is often not available. In this paper, we investigate intermediate pre-training specifically for RE. We exploit the affinity between syntactic structure and semantic RE, and identify the syntactic relations which are closely related to RE by being on the shortest dependency path between two entities. We then take advantage of the high accuracy of current syntactic parsers in order to automatically obtain large amounts of low-cost pre-training data. By pre-training our RE model on the relevant syntactic relations, we are able to outperform the baseline in five out of six cross-domain setups, without any additional annotated data.

Generative adversarial networks (GANs) and denoising diffusion probabilistic models (DDPMs) have recently achieved impressive performances in image and audio synthesis. After revisiting their success in conditional speech synthesis, we find that 1) GANs sacrifice sample diversity for quality and speed, 2) diffusion models exhibit outperformed sample quality and diversity at a high computational cost, where achieving high-quality, fast, and diverse speech synthesis challenges all neural synthesizers. In this work, we propose to converge advantages from GANs and diffusion models by incorporating both classes, introducing dual-empowered modeling perspectives: 1) FastDiff 2 (DiffGAN), a diffusion model whose denoising process is parametrized by conditional GANs, and the non-Gaussian denoising distribution makes it much more stable to implement the reverse process with large steps sizes; and 2) FastDiff 2 (GANDiff), a generative adversarial network whose forward process is constructed by multiple denoising diffusion iterations, which exhibits better sample diversity than traditional GANs. Experimental results show that both variants enjoy an efficient 4-step sampling process and demonstrate superior sample quality and diversity. Audio samples are available at https://RevisitSpeech.github.io/

pdf abs
Uncovering Hidden Consequences of Pre-training Objectives in Sequence-to-Sequence Models
Tannon Kew | Rico Sennrich

Some variants of self-supervised denoising objectives for pre-training encoder-decoder language models have been reported to have a negligible impact on downstream performance. Yet the design of these pre-training objectives leads to behavioural differences that can be uncovered with specific manipulations. We reproduce a recently proposed zero-shot control method and find that it is only successful on a subset of models. To understand what causes the difference in its effectiveness, we perform a set of controlled experiments, varying only the pre-training objective, and find unexpected interactions between the pre-training method and downstream controllability of models after fine-tuning. Our results show that different pre-training objectives have consequences that may not be visible in standard downstream evaluation, but which should be taken into account when developing models with controllability in mind.

pdf abs
Exploring Anisotropy and Outliers in Multilingual Language Models for Cross-Lingual Semantic Sentence Similarity
Katharina Haemmerl | Alina Fastowski | Jindřich Libovický | Alexander Fraser

Previous work has shown that the representations output by contextual language models are more anisotropic than static type embeddings, and typically display outlier dimensions. This seems to be true for both monolingual and multilingual models, although much less work has been done on the multilingual context. Why these outliers occur and how they affect the representations is still an active area of research. We investigate outlier dimensions and their relationship to anisotropy in multiple pre-trained multilingual language models. We focus on cross-lingual semantic similarity tasks, as these are natural tasks for evaluating multilingual representations. Specifically, we examine sentence representations. Sentence transformers which are fine-tuned on parallel resources (that are not always available) perform better on this task, and we show that their representations are more isotropic. However, we aim to improve multilingual representations in general. We investigate how much of the performance difference can be made up by only transforming the embedding space without fine-tuning, and visualise the resulting spaces. We test different operations: Removing individual outlier dimensions, cluster-based isotropy enhancement, and ZCA whitening. We publish our code for reproducibility.

Tasks involving text generation based on multiple input texts, such as multi-document summarization, long-form question answering and contemporary dialogue applications, challenge models for their ability to properly consolidate partly-overlapping multi-text information. However, these tasks entangle the consolidation phase with the often subjective and ill-defined content selection requirement, impeding proper assessment of models’ consolidation capabilities. In this paper, we suggest revisiting the sentence union generation task as an effective well-defined testbed for assessing text consolidation capabilities, decoupling the consolidation challenge from subjective content selection. To support research on this task, we present refined annotation methodology and tools for crowdsourcing sentence union, create the largest union dataset to date and provide an analysis of its rich coverage of various consolidation aspects. We then propose a comprehensive evaluation protocol for union generation, including both human and automatic evaluation. Finally, as baselines, we evaluate state-of-the-art language models on the task, along with a detailed analysis of their capacity to address multi-text consolidation challenges and their limitations.

pdf abs
Distilling Reasoning Capabilities into Smaller Language Models
Kumar Shridhar | Alessandro Stolfo | Mrinmaya Sachan

Step-by-step reasoning approaches like chain of thought (CoT) have proved to be very effective in inducing reasoning capabilities in large language models. However, the success of the CoT approach is fundamentally tied to the model size, and billion parameter-scale models are often needed to get CoT to work. In this paper, we propose a knowledge distillation approach that leverages the step-by-step CoT reasoning capabilities of larger models and distills these abilities into smaller models. In this work, we propose an alternative reasoning scheme, Socratic CoT that learns a decomposition of the original problem into a sequence of subproblems and uses it to guide the intermediate reasoning steps. We use Socratic CoT to train a combination of two small distilled models: a problem decomposer and a subproblem solver. In practice, given a new problem, the two distilled models work in sync to decompose and solve complex problems. On multiple reasoning datasets (GSM8K, StrategyQA, and SVAMP), our proposed distillation strategies boosts the performance of smaller models over 70% compared to the baselines. Finally, we investigate when Socratic CoT is an effective alternative to CoT, demonstrating cases where a much smaller model (GPT-2 large) can outperform a 10X larger model (GPT-3 6B). Our code is available: https://github.com/kumar-shridhar/Distiiling-LM.

pdf abs
AlignSTS: Speech-to-Singing Conversion via Cross-Modal Alignment
Ruiqi Li | Rongjie Huang | Lichao Zhang | Jinglin Liu | Zhou Zhao

The speech-to-singing (STS) voice conversion task aims to generate singing samples corresponding to speech recordings while facing a major challenge: the alignment between the target (singing) pitch contour and the source (speech) content is difficult to learn in a text-free situation. This paper proposes AlignSTS, an STS model based on explicit cross-modal alignment, which views speech variance such as pitch and content as different modalities. Inspired by the mechanism of how humans will sing the lyrics to the melody, AlignSTS: 1) adopts a novel rhythm adaptor to predict the target rhythm representation to bridge the modality gap between content and pitch, where the rhythm representation is computed in a simple yet effective way and is quantized into a discrete space; and 2) uses the predicted rhythm representation to re-align the content based on cross-attention and conducts a cross-modal fusion for re-synthesize. Extensive experiments show that AlignSTS achieves superior performance in terms of both objective and subjective metrics. Audio samples are available at https://alignsts.github.io.

pdf abs
A New Task and Dataset on Detecting Attacks on Human Rights Defenders
Shihao Ran | Di Lu | Aoife Cahill | Joel Tetreault | Alejandro Jaimes

The ability to conduct retrospective analyses of attacks on human rights defenders over time and by location is important for humanitarian organizations to better understand historical or ongoing human rights violations and thus better manage the global impact of such events. We hypothesize that NLP can support such efforts by quickly processing large collections of news articles to detect and summarize the characteristics of attacks on human rights defenders. To that end, we propose a new dataset for detecting Attacks on Human Rights Defenders (HRDsAttack) consisting of crowdsourced annotations on 500 online news articles. The annotations include fine-grained information about the type and location of the attacks, as well as information about the victim(s). We demonstrate the usefulness of the dataset by using it to train and evaluate baseline models on several sub-tasks to predict the annotated characteristics.

pdf abs
Improving Language Model Integration for Neural Machine Translation
Christian Herold | Yingbo Gao | Mohammad Zeineldeen | Hermann Ney

The integration of language models for neural machine translation has been extensively studied in the past. It has been shown that an external language model, trained on additional target-side monolingual data, can help improve translation quality. However, there has always been the assumption that the translation model also learns an implicit target-side language model during training, which interferes with the external language model at decoding time. Recently, some works on automatic speech recognition have demonstrated that, if the implicit language model is neutralized in decoding, further improvements can be gained when integrating an external language model. In this work, we transfer this concept to the task of machine translation and compare with the most prominent way of including additional monolingual data - namely back-translation. We find that accounting for the implicit language model significantly boosts the performance of language model fusion, although this approach is still outperformed by back-translation.

We introduce the task of correcting named entity recognition (NER) errors without re-training model. After an NER model is trained and deployed in production,it makes prediction errors, which usually need to be fixed quickly. To address this problem, we firstly construct a gazetteer containing named entities and corresponding possible entity types. And then, we propose type enhanced BERT (TyBERT),a method that integrates the named entity’s type information into BERT by an adapter layer. When errors are identified, we can repair the model by updating the gazetteer. In other words, the gazetteer becomes a trigger to control NER model’s output. The experiment results in multiple corpus show the effectiveness of our method, which outperforms strong baselines.x

pdf abs
Bridge the Gap Between CV and NLP! A Gradient-based Textual Adversarial Attack Framework
Lifan Yuan | YiChi Zhang | Yangyi Chen | Wei Wei

Despite recent success on various tasks, deep learning techniques still perform poorly on adversarial examples with small perturbations. While optimization-based methods for adversarial attacks are well-explored in the field of computer vision, it is impractical to directly apply them in natural language processing due to the discrete nature of the text. To address the problem, we propose a unified framework to extend the existing optimization-based adversarial attack methods in the vision domain to craft textual adversarial samples. In this framework, continuously optimized perturbations are added to the embedding layer and amplified in the forward propagation process. Then the final perturbed latent representations are decoded with a masked language model head to obtain potential adversarial samples. In this paper, we instantiate our framework with an attack algorithm named Textual Projected Gradient Descent (T-PGD). We find our algorithm effective even using proxy gradient information. Therefore, we perform the more challenging transfer black-box attack and conduct comprehensive experiments to evaluate our attack algorithm with several models on three benchmark datasets. Experimental results demonstrate that our method achieves overall better performance and produces more fluent and grammatical adversarial samples compared to strong baseline methods. The code and data are available at https://github.com/Phantivia/T-PGD.

pdf abs
DUB: Discrete Unit Back-translation for Speech Translation
Dong Zhang | Rong Ye | Tom Ko | Mingxuan Wang | Yaqian Zhou

How can speech-to-text translation (ST) perform as well as machine translation (MT)? The key point is to bridge the modality gap between speech and text so that useful MT techniques can be applied to ST.Recently, the approach of representing speech with unsupervised discrete units yields a new way to ease the modality problem. This motivates us to propose Discrete Unit Back-translation(DUB) to answer two questions (1) Is it better to represent speech with discrete units than with continuous features in direct ST? (2) How much benefit can useful MT techniques bring to ST? With DUB, the back-translation technique can successfully be applied on direct ST and obtains an average boost of 5.5 BLEU on MuST-C En-De/Fr/Es. In the low-resource language scenario, our method achieves comparable performance to existing methods that rely on large-scale external data. Code and models are available at https://anonymous.4open.science/r/DUB/.

Knowledge graphs mostly exhibit a mixture of branching relations, e.g., hasFriend, and complex structures, e.g., hierarchy and loop. Most knowledge graph embeddings have problems expressing them, because they model a specific relation r from a head h to tails by starting at the node embedding of h and transitioning deterministically to exactly one other point in the embedding space. We overcome this issue in our novel framework ItCAREToE by modeling relations between nodes by relation-specific, stochastic transitions. Our framework is based on stochastic ItCARETo processes, which operate on low-dimensional manifolds. ItCAREToE is highly expressive and generic subsuming various state-of-the-art models operating on different, also non-Euclidean, manifolds. Experimental results show the superiority of ItCAREToE over other deterministic embedding models with regard to the KG completion task.

pdf abs
Leveraging Denoised Abstract Meaning Representation for Grammatical Error Correction
Hejing Cao | Dongyan Zhao

Grammatical Error Correction (GEC) is the task of correcting errorful sentences into grammatically correct, semantically consistent, and coherent sentences. Popular GEC models either use large-scale synthetic corpora or use a large number of human-designed rules. The former is costly to train, while the latter requires quite a lot of human expertise. In recent years, AMR, a semantic representation framework, has been widely used by many natural language tasks due to its completeness and flexibility. A non-negligible concern is that AMRs of grammatically incorrect sentences may not be exactly reliable. In this paper, we propose the AMR-GEC, a seq-to-seq model that incorporates denoised AMR as additional knowledge. Specifically, We design a semantic aggregated GEC model and explore denoising methods to get AMRs more reliable. Experiments on the BEA-2019 shared task and the CoNLL-2014 shared task have shown that AMR-GEC performs comparably to a set of strong baselines with a large number of synthetic data. Compared with the T5 model with synthetic data, AMR-GEC can reduce the training time by 32% while inference time is comparable. To the best of our knowledge, we are the first to incorporate AMR for grammatical error correction.

pdf abs
Prediction and Calibration: Complex Reasoning over Knowledge Graph with Bi-directional Directed Acyclic Graph Neural Network
Yao Xu | Shizhu He | Li Cai | Kang Liu | Jun Zhao

Answering complex logical queries is a challenging task for knowledge graph (KG) reasoning. Recently, query embedding (QE) has been proposed to encode queries and entities into the same vector space, and obtain answers based on numerical computation. However, such models obtain the node representations of a query only based on its predecessor nodes, which ignore the information contained in successor nodes. In this paper, we proposed a Bi-directional Directed Acyclic Graph neural network (BiDAG) that splits the reasoning process into prediction and calibration. The joint probability of all nodes is considered by applying a graph neural network (GNN) to the query graph in the calibration process. By the prediction in the first layer and the calibration in deep layers of GNN, BiDAG can outperform previous QE based methods on FB15k, FB15k-237, and NELL995.

pdf abs
Prompt-Based Metric Learning for Few-Shot NER
Yanru Chen | Yanan Zheng | Zhilin Yang

Few-shot named entity recognition (NER) targets generalizing to unseen labels and/or domains with few labeled examples. Existing metric learning methods compute token-level similarities between query and support sets, but are not able to fully incorporate label semantics into modeling. To address this issue, we propose a simple method to largely improve metric learning for NER: 1) multiple prompt schemas are designed to enhance label semantics; 2) we propose a novel architecture to effectively combine multiple prompt-based representations. Empirically, our method achieves new state-of-the-art (SOTA) results under 16 of the 18 considered settings, substantially outperforming the previous SOTA by an average of 9.12% and a maximum of 34.51% in relative gains of micro F1.

pdf abs
OpenPI-C: A Better Benchmark and Stronger Baseline for Open-Vocabulary State Tracking
Xueqing Wu | Sha Li | Heng Ji

Open-vocabulary state tracking is a more practical version of state tracking that aims to track state changes of entities throughout a process without restricting the state space and entity space. OpenPI (Tandon et al., 2020) is to date the only dataset annotated for open-vocabulary state tracking. However, we identify issues with the dataset quality and evaluation metric. For the dataset, we categorize 3 types of problems on the procedure level, step level and state change level respectively, and build a clean dataset OpenPI-C using multiple rounds of human judgment. For the evaluation metric, we propose a cluster-based metric to fix the original metric’s preference for repetition. Model-wise, we enhance the seq2seq generation baseline by reinstating two key properties for state tracking: temporal dependency and entity awareness. The state of the world after an action is inherently dependent on the previous state. We model this dependency through a dynamic memory bank and allow the model to attend to the memory slots during decoding. On the other hand, the state of the world is naturally a union of the states of involved entities. Since the entities are unknown in the open-vocabulary setting, we propose a two-stage model that refines the state change prediction conditioned on entities predicted from the first stage. Empirical results show the effectiveness of our proposed model, especially on the cleaned dataset and the cluster-based metric. The code and data are released at https://github.com/shirley-wu/openpi-c

A simile is a figure of speech that compares two different things (called the tenor and the vehicle) via shared properties. The tenor and the vehicle are usually connected with comparator words such as “like” or “as”. The simile phenomena are unique and complex in a real-life dialogue scene where the tenor and the vehicle can be verbal phrases or sentences, mentioned by different speakers, exist in different sentences, or occur in reversed order. However, the current simile research usually focuses on similes in a triplet tuple (tenor, property, vehicle) or a single sentence where the tenor and vehicle are usually entities or noun phrases, which could not reflect complex simile phenomena in real scenarios. In this paper, we propose a novel and high-quality multilingual simile dialogue (MSD) dataset to facilitate the study of complex simile phenomena. The MSD is the largest manually annotated simile data (~21K) and it contains both English and Chinese data. Meanwhile, the MSD data can also be used on dialogue tasks to test the ability of dialogue systems when using similes. We design 3 simile tasks (recognition, interpretation, and generation) and 2 dialogue tasks (retrieval and generation) with MSD. For each task, we provide experimental results from strong pre-trained or state-of-the-art models. The experiments demonstrate the challenge of MSD and we will release the data/code on GitHub.

pdf abs
Controllable Conversation Generation with Conversation Structures via Diffusion Models
Jiaao Chen | Diyi Yang

Generating coherent conversation is an important and challenging long text generation task, as it has various applications such as daily entertainment, children education or building conversational AI to facilitate human-computer interaction. However, current generation models often fail to effectively utilize rich linguistic and world knowledge to generate conversations just like human. In this work, we introduce a novel conversation generation framework to effectively incorporate human knowledge and conversation structures with both controllability and interpretability for better conversation generation. Specifically, we first generate the prototype conversations from short descriptions. We then gradually and strategically incorporate different levels of conversation structures including the action triples, dialogue acts and discourse relations via diffusion models to directly edit the prototype conversations. We demonstrate the effectiveness of our framework through experiments on two datasets by comparing our method with the state-of-the-art baseline models.

pdf abs
Few-shot Low-resource Knowledge Graph Completion with Reinforced Task Generation
Shichao Pei | Qiannan Zhang | Xiangliang Zhang

Despite becoming a prevailing paradigm for organizing knowledge, most knowledge graphs (KGs) suffer from the low-resource issue due to the deficiency of data sources. The enrichment of KGs by automatic knowledge graph completion is impeded by the intrinsic long-tail property of KGs. In spite of their prosperity, existing few-shot learning-based models have difficulty alleviating the impact of the long-tail issue on low-resource KGs because of the lack of training tasks. To tackle the challenging long-tail issue on low-resource KG completion, in this paper, we propose a novel few-shot low-resource knowledge graph completion framework, which is composed of three components, i.e., few-shot learner, task generator, and task selector. The key idea is to generate and then select the beneficial few-shot tasks that complement the current tasks and enable the optimization of the few-shot learner using the selected few-shot tasks. Extensive experiments conducted on several real-world knowledge graphs validate the effectiveness of our proposed method.

pdf abs
Incomplete Utterance Rewriting as Sequential Greedy Tagging
Yunshan Chen

The task of incomplete utterance rewriting has recently gotten much attention. Previous models struggled to extract information from the dialogue context, as evidenced by the low restoration scores. To address this issue, we propose a novel sequence tagging-based model, which is more adept at extracting information from context. Meanwhile, we introduce speaker-aware embedding to model speaker variation. Experiments on multiple public datasets show that our model achieves optimal results on all nine restoration scores while having other metric scores comparable to previous state-of-the-art models. Furthermore, benefitting from the model’s simplicity, our approach outperforms most previous models on inference speed.

pdf abs
Exploiting Commonsense Knowledge about Objects for Visual Activity Recognition
Tianyu Jiang | Ellen Riloff

Situation recognition is the task of recognizing the activity depictedin an image, including the people and objects involved. Previousmodels for this task typically train a classifier to identify theactivity using a backbone image feature extractor. We propose thatcommonsense knowledge about the objects depicted in an image can alsobe a valuable source of information for activity identification. Previous NLP research has argued that knowledge about the prototypicalfunctions of physical objects is important for language understanding,and NLP techniques have been developed to acquire this knowledge. Our work investigates whether this prototypical function knowledgecan also be beneficial for visual situation recognition. Webuild a framework that incorporates this type of commonsense knowledgein a transformer-based model that is trained to predict the actionverb for situation recognition. Our experimental results show thatadding prototypical function knowledge about physical objects doesimprove performance for the visual activity recognition task.

pdf abs
Tucker Decomposition with Frequency Attention for Temporal Knowledge Graph Completion
Likang Xiao | Richong Zhang | Zijie Chen | Junfan Chen

Temporal Knowledge Graph Completion aims to complete missing entities or relations under temporal constraints. Previous tensor decomposition-based models for TKGC only independently consider the combination of one single relation with one single timestamp, ignoring the global nature of the embedding. We propose a Frequency Attention (FA) model to capture the global temporal dependencies between one relation and the entire timestamp. Specifically, we use Discrete Cosine Transform (DCT) to capture the frequency of the timestamp embedding and further compute the frequency attention weight to scale embedding. Meanwhile, the previous temporal tucker decomposition method uses a simple norm regularization to constrain the core tensor, which limits the optimization performance. Thus, we propose Orthogonal Regularization (OR) variants for the core tensor, which can limit the non-superdiagonal elements of the 3-rd core tensor. Experiments on three standard TKGC datasets demonstrate that our method outperforms the state-of-the-art results on several metrics. The results suggest that the direct-current component is not the best feature for TKG representation learning. Additional analysis shows the effectiveness of our FA and OR models, even with smaller embedding dimensions.

pdf abs
Another Dead End for Morphological Tags? Perturbed Inputs and Parsing
Alberto Muñoz-Ortiz | David Vilares

The usefulness of part-of-speech tags for parsing has been heavily questioned due to the success of word-contextualized parsers. Yet, most studies are limited to coarse-grained tags and high quality written content; while we know little about their influence when it comes to models in production that face lexical errors. We expand these setups and design an adversarial attack to verify if the use of morphological information by parsers: (i) contributes to error propagation or (ii) if on the other hand it can play a role to correct mistakes that word-only neural parsers make. The results on 14 diverse UD treebanks show that under such attacks, for transition- and graph-based models their use contributes to degrade the performance even faster, while for the (lower-performing) sequence labeling parsers they are helpful. We also show that if morphological tags were utopically robust against lexical perturbations, they would be able to correct parsing mistakes.

The task of textual geolocation — retrieving the coordinates of a place based on a free-form language description — calls for not only grounding but also natural language understanding and geospatial reasoning. Even though there are quite a few datasets in English used for geolocation, they are currently based on open-source data (Wikipedia and Twitter), where the location of the described place is mostly implicit, such that the location retrieval resolution is limited. Furthermore, there are no datasets available for addressing the problem of textual geolocation in morphologically rich and resource-poor languages, such as Hebrew. In this paper, we present the Hebrew Geo-Location (HeGeL) corpus, designed to collect literal place descriptions and analyze lingual geospatial reasoning. We crowdsourced 5,649 literal Hebrew place descriptions of various place types in three cities in Israel. Qualitative and empirical analysis show that the data exhibits abundant use of geospatial reasoning and requires a novel environmental representation.

pdf abs
Modeling Adversarial Attack on Pre-trained Language Models as Sequential Decision Making
Xuanjie Fang | Sijie Cheng | Yang Liu | Wei Wang

Pre-trained language models (PLMs) have been widely used to underpin various downstream tasks. However, the adversarial attack task has found that PLMs are vulnerable to small perturbations. Mainstream methods adopt a detached two-stage framework to attack without considering the subsequent influence of substitution at each step. In this paper, we formally model the adversarial attack task on PLMs as a sequential decision-making problem, where the whole attack process is sequential with two decision-making problems, i.e., word finder and word substitution. Considering the attack process can only receive the final state without any direct intermediate signals, we propose to use reinforcement learning to find an appropriate sequential attack path to generate adversaries, named SDM-ATTACK. Our experimental results show that SDM-ATTACK achieves the highest attack success rate with a comparable modification rate and semantic similarity to attack fine-tuned BERT. Furthermore, our analyses demonstrate the generalization and transferability of SDM-ATTACK.Resources of this work will be released after this paper’s publication.

Generating persona consistent dialogue response is important for developing an intelligent conversational agent. Recent works typically fine-tune large-scale pre-trained models on this task by concatenating persona texts and dialogue history as a single input sequence to generate the target response. While simple and effective, our analysis shows that this popular practice is seriously affected by order sensitivity where different input orders of persona sentences significantly impact the quality and consistency of generated response, resulting in severe performance fluctuations (i.e., 29.4% on GPT2 and 83.2% on BART). To mitigate the order sensitivity problem, we propose a model-agnostic framework, ORder Insensitive Generation (ORIG), which enables dialogue models to learn robust representation under different persona orders and improve the consistency of response generation. Experiments on the Persona-Chat dataset justify the effectiveness and superiority of our method with two dominant pre-trained models (GPT2 and BART).

pdf abs
Cost-effective Distillation of Large Language Models
Sayantan Dasgupta | Trevor Cohn | Timothy Baldwin

Knowledge distillation (KD) involves training a small “student” model to replicate the strong performance of a high-capacity “teacher” model, enabling efficient deployment in resource-constrained settings. Top-performing methods tend to be task- or architecture-specific and lack generalizability. Several existing approaches require pretraining of the teacher on task-specific datasets, which can be costly for large and unstable for small datasets. Here we propose an approach for improving KD through a novel distillation loss agnostic to the task and model architecture. We successfully apply our method to the distillation of the BERT-base and achieve highly competitive results from the distilled student across a range of GLUE tasks, especially for tasks with smaller datasets.

pdf abs
Task-Optimized Adapters for an End-to-End Task-Oriented Dialogue System
Namo Bang | Jeehyun Lee | Myoung-Wan Koo

Task-Oriented Dialogue (TOD) systems are designed to carry out specific tasks by tracking dialogue states and generating appropriate responses to help users achieve defined goals. Recently, end-to-end dialogue models pre-trained based on large datasets have shown promising performance in the conversational system. However, they share the same parameters to train tasks of the dialogue system (NLU, DST, NLG), so debugging each task is challenging. Also, they require a lot of effort to fine-tune large parameters to create a task-oriented chatbot, making it difficult for non-experts to handle. Therefore, we intend to train relatively lightweight and fast models compared to PLM. In this paper, we propose an End-to-end TOD system with Task-Optimized Adapters which learn independently per task, adding only small number of parameters after fixed layers of pre-trained network. We also enhance the performance of the DST and NLG modules through reinforcement learning, overcoming the learning curve that has lacked at the adapter learning and enabling the natural and consistent response generation that is appropriate for the goal. Our method is a model-agnostic approach and does not require prompt-tuning as only input data without a prompt. As results of the experiment, our method shows competitive performance on the MultiWOZ benchmark compared to the existing end-to-end models. In particular, we attain state-of-the-art performance on the DST task of 2.2 dataset.

Visual metaphors are powerful rhetorical devices used to persuade or communicate creative ideas through images. Similar to linguistic metaphors, they convey meaning implicitly through symbolism and juxtaposition of the symbols. We propose a new task of generating visual metaphors from linguistic metaphors. This is a challenging task for diffusion-based text-to-image models, such as DALL⋅E 2, since it requires the ability to model implicit meaning and compositionality. We propose to solve the task through the collaboration between Large Language Models (LLMs) and Diffusion Models: Instruct GPT-3 (davinci-002) with Chain-of-Thought prompting generates text that represents a visual elaboration of the linguistic metaphor containing the implicit meaning and relevant objects, which is then used as input to the diffusion-based text-to-image models. Using a human-AI collaboration framework, where humans interact both with the LLM and the top-performing diffusion model, we create a high-quality dataset containing 6,476 visual metaphors for 1,540 linguistic metaphors and their associated visual elaborations. Evaluation by professional illustrators shows the promise of LLM-Diffusion Model collaboration for this task.To evaluate the utility of our Human-AI collaboration framework and the quality of our dataset, we perform both an intrinsic human-based evaluation and an extrinsic evaluation using visual entailment as a downstream task.

pdf abs
Text Augmentation Using Dataset Reconstruction for Low-Resource Classification
Adir Rahamim | Guy Uziel | Esther Goldbraich | Ateret Anaby Tavor

In the deployment of real-world text classification models, label scarcity is a common problem and as the number of classes increases, this problem becomes even more complex. An approach to addressing this problem is by applying text augmentation methods. One of the more prominent methods involves using the text-generation capabilities of language models. In this paper, we propose Text AUgmentation by Dataset Reconstruction (TAU-DR), a novel method of data augmentation for text classification. We conduct experiments on several multi-class datasets, showing that our approach improves the current state-of-the-art techniques for data augmentation.

pdf abs
LaSQuE: Improved Zero-Shot Classification from Explanations Through Quantifier Modeling and Curriculum Learning
Sayan Ghosh | Rakesh R. Menon | Shashank Srivastava

A hallmark of human intelligence is the ability to learn new concepts purely from language. Several recent approaches have explored training machine learning models via natural language supervision. However, these approaches fall short in leveraging linguistic quantifiers (such as ‘always’ or ‘rarely’) and mimicking humans in compositionally learning complex tasks. Here, we present LaSQuE, a method that can learn zero-shot classifiers from language explanations by using three new strategies - (1) modeling the semantics of linguistic quantifiers in explanations (including exploiting ordinal strength relationships, such as ‘always’ > ‘likely’), (2) aggregating information from multiple explanations using an attention-based mechanism, and (3) model training via curriculum learning. With these strategies, LaSQuE outperforms prior work, showing an absolute gain of up to 7% in generalizing to unseen real-world classification tasks.

pdf abs
Learned Adapters Are Better Than Manually Designed Adapters
Yuming Zhang | Peng Wang | Ming Tan | Wei Zhu

Recently, a series of works have looked into further improving the adapter-based tuning by manually designing better adapter architectures. Understandably, these manually designed solutions are sub-optimal. In this work, we propose the Learned Adapter framework to automatically learn the optimal adapter architectures for better task adaptation of pre-trained models (PTMs). First, we construct a unified search space for adapter architecture designs. In terms of the optimization method on the search space, we propose a simple-yet-effective method, GDNAS for better architecture optimization. Extensive experiments show that our Learned Adapter framework can outperform the previous parameter-efficient tuning (PETuning) baselines while tuning comparable or fewer parameters. Moreover: (a) the learned adapter architectures are explainable and transferable across tasks. (b) We demonstrate that our architecture search space design is valid.

pdf abs
Automatic Identification of Code-Switching Functions in Speech Transcripts
Ritu Belani | Jeffrey Flanigan

Code-switching, or switching between languages, occurs for many reasons and has important linguistic, sociological, and cultural implications. Multilingual speakers code-switch for a variety of communicative functions, such as expressing emotions, borrowing terms, making jokes, introducing a new topic, etc. The function of code-switching may be quite useful for the analysis of linguists, cognitive scientists, speech therapists, and others, but is not readily apparent. To remedy this situation, we annotate and release a new dataset of functions of code-switching in Spanish-English. We build the first system (to our knowledge) to automatically identify a wide range of functions for which speakers code-switch in everyday speech, achieving an accuracy of 75% across all functions.

Federated learning involves collaborative training with private data from multiple platforms, while not violating data privacy. We study the problem of federated domain adaptation for Named Entity Recognition (NER), where we seek to transfer knowledge across different platforms with data of multiple domains. In addition, we consider a practical and challenging scenario, where NER datasets of different platforms of federated learning are annotated with heterogeneous tag sets, i.e., different sets of entity types. The goal is to train a global model with federated learning, such that it can predict with a complete tag set, i.e., with all the occurring entity types for data across all platforms. To cope with the heterogeneous tag sets in a multi-domain setting, we propose a distillation approach along with a mechanism of instance weighting to facilitate knowledge transfer across platforms. Besides, we release two re-annotated clinic NER datasets, for testing the proposed method in the clinic domain. Our method shows superior empirical performance for NER with federated learning.

As the key to sentiment analysis, sentiment composition considers the classification of a constituent via classifications of its contained sub-constituents and rules operated on them. Such compositionality has been widely studied previously in the form of hierarchical trees including untagged and sentiment ones, which are intrinsically suboptimal in our view. To address this, we propose semantic tree, a new tree form capable of interpreting the sentiment composition in a principled way. Semantic tree is a derivation of a context-free grammar (CFG) describing the specific composition rules on difference semantic roles, which is designed carefully following previous linguistic conclusions. However, semantic tree is a latent variable since there is no its annotation in regular datasets. Thus, in our method, it is marginalized out via inside algorithm and learned to optimize the classification performance. Quantitative and qualitative results demonstrate that our method not only achieves better or competitive results compared to baselines in the setting of regular and domain adaptation classification, and also generates plausible tree explanations.

Language models have been shown to exhibit positive scaling, where performance improves as models are scaled up in terms of size, compute, or data. In this work, we introduce NeQA, a dataset consisting of questions with negation in which language models do not exhibit straightforward positive scaling. We show that this task can exhibit inverse scaling, U-shaped scaling, or positive scaling, and the three scaling trends shift in this order as we use more powerful prompting methods or model families. We hypothesize that solving NeQA depends on two subtasks: question answering (task 1) and negation understanding (task 2). We find that task 1 has linear scaling, while task 2 has sigmoid-shaped scaling with an emergent transition point, and composing these two scaling trends yields the final scaling trend of NeQA. Our work reveals and provides a way to analyze the complex scaling trends of language models.

We investigate semi-structured document classification in a zero-shot setting. Classification of semi-structured documents is more challenging than that of standard unstructured documents, as positional, layout, and style information play a vital role in interpreting such documents. The standard classification setting where categories are fixed during both training and testing falls short in dynamic environments where new classification categories could potentially emerge. We focus exclusively on the zero-shot learning setting where inference is done on new unseen classes. To address this task, we propose a matching-based approach that relies on a pairwise contrastive objective for both pretraining and fine-tuning. Our results show a significant boost in Macro F1 from the proposed pretraining step and comparable performance of the contrastive fine-tuning to a standard prediction objective in both supervised and unsupervised zero-shot settings.

pdf abs
Extracting Shopping Interest-Related Product Types from the Web
Yinghao Li | Colin Lockard | Prashant Shiralkar | Chao Zhang

Recommending a diversity of product types (PTs) is important for a good shopping experience when customers are looking for products around their high-level shopping interests (SIs) such as hiking. However, the SI-PT connection is typically absent in e-commerce product catalogs and expensive to construct manually due to the volume of potential SIs, which prevents us from establishing a recommender with easily accessible knowledge systems. To establish such connections, we propose to extract PTs from the Web pages containing hand-crafted PT recommendations for SIs. The extraction task is formulated as binary HTML node classification given the general observation that an HTML node in our target Web pages can present one and only one PT phrase. Accordingly, we introduce TrENC, which stands for Tree-Transformer Encoders for Node Classification. It improves the inter-node dependency modeling with modified attention mechanisms that preserve the long-term sibling and ancestor-descendant relations. TrENC also injects SI into node features for better semantic representation. Trained on pages regarding limited SIs, TrEnc is ready to be applied to other unobserved interests. Experiments on our manually constructed dataset, WebPT, show that TrENC outperforms the best baseline model by 2.37 F1 points in the zero-shot setup. The performance indicates the feasibility of constructing SI-PT relations and using them to power downstream applications such as search and recommendation.

pdf abs
Multilingual Pre-training with Self-supervision from Global Co-occurrence Information
Xi Ai | Bin Fang

Global co-occurrence information is the primary source of structural information on multilingual corpora, and we find that analogical/parallel compound words across languages have similar co-occurrence counts/frequencies (normalized) giving weak but stable self-supervision for cross-lingual transfer. Following the observation, we aim at associating contextualized representations with relevant (contextualized) representations across languages with the help of co-occurrence counts. The result is MLM-GC (MLM with Global Co-occurrence) pre-training that the model learns local bidirectional information from MLM and global co-occurrence information from a log-bilinear regression. Experiments show that MLM-GC pre-training substantially outperforms MLM pre-training for 4 downstream cross-lingual tasks and 1 additional monolingual task, showing the advantages of forming isomorphic spaces across languages.

pdf abs
Low-Rank Updates of pre-trained Weights for Multi-Task Learning
Alexandre Audibert | Massih R Amini | Konstantin Usevich | Marianne Clausel

Multi-Task Learning used with pre-trained models has been quite popular in the field of Natural Language Processing in recent years. This framework remains still challenging due to the complexity of the tasks and the challenges associated with fine-tuning large pre-trained models. In this paper, we propose a new approach for Multi-task learning which is based on stacking the weights of Neural Networks as a tensor. We show that low-rank updates in the canonical polyadic tensor decomposition of this tensor of weights lead to a simple, yet efficient algorithm, which without loss of performance allows to reduce considerably the model parameters. We investigate the interactions between tasks inside the model as well as the inclusion of sparsity to find the best tensor rank and to increase the compression rate. Our strategy is consistent with recent efforts that attempt to use constraints to fine-tune some model components. More precisely, we achieve equivalent performance as the state-of-the-art on the General Language Understanding Evaluation benchmark by training only 0.3 of the parameters per task while not modifying the baseline weights.

pdf abs
Sequential Integrated Gradients: a simple but effective method for explaining language models
Joseph Enguehard

Several explanation methods such as Integrated Gradients (IG) can be characterised as path-based methods, as they rely on a straight line between the data and an uninformative baseline. However, when applied to language models, these methods produce a path for each word of a sentence simultaneously, which could lead to creating sentences from interpolated words either having no clear meaning, or having a significantly different meaning compared to the original sentence. In order to keep the meaning of these sentences as close as possible to the original one, we propose Sequential Integrated Gradients (SIG), which computes the importance of each word in a sentence by keeping fixed every other words, only creating interpolations between the baseline and the word of interest. Moreover, inspired by the training procedure of language models, we also propose to replace the baseline token “pad” with the trained token “mask”. While being a simple improvement over the original IG method, we show on various models and datasets that SIG proves to be a very effective method for explaining language models.

Text detoxification is a conditional text generation task aiming to remove offensive content from toxic text. It is highly useful for online forums and social media, where offensive content is frequently encountered. Intuitively, there are diverse ways to detoxify sentences while preserving their meanings, and we can select from detoxified sentences before displaying text to users. Conditional diffusion models are particularly suitable for this task given their demonstrated higher generative diversity than existing conditional text generation models based on language models. Nonetheless, text fluency declines when they are trained with insufficient data, which is the case for this task. In this work, we propose DiffuDetox, a mixed conditional and unconditional diffusion model for text detoxification. The conditional model takes toxic text as the condition and reduces its toxicity, yielding a diverse set of detoxified sentences. The unconditional model is trained to recover the input text, which allows the introduction of additional fluent text for training and thus ensures text fluency. Extensive experimental results and in-depth analysis demonstrate the effectiveness of our proposed DiffuDetox.

pdf abs
Separating Context and Pattern: Learning Disentangled Sentence Representations for Low-Resource Extractive Summarization
Ruifeng Yuan | Shichao Sun | Zili Wang | Ziqiang Cao | Wenjie Li

Extractive summarization aims to select a set of salient sentences from the source document to form a summary. Context information has been considered one of the key factors for this task. Meanwhile, there also exist other pattern factors that can identify sentence importance, such as sentence position or certain n-gram tokens. However, such pattern information is only effective in specific datasets or domains and can not be generalized like the context information when there only exists limited data. In this case, current extractive summarization models may suffer from a performance drop when transferring to a new dataset. In this paper, we attempt to apply disentangled representation learning on extractive summarization, and separate the two key factors for the task, context and pattern, for a better generalization ability in the low-resource setting. To achieve this, we propose two groups of losses for encoding and disentangling sentence representations into context representations and pattern representations. In this case, we can either use only the context information in the zero-shot setting or fine-tune the pattern information in the few-shot setting. Experimental results on three summarization datasets from different domains show the effectiveness of our proposed approach.

This paper presents ReasonFormer, a unified reasoning framework for mirroring the modular and compositional reasoning process of humans in complex decision-making. Inspired by dual-process theory in cognitive science, the representation module (automatic thinking) and reasoning modules (controlled thinking) are decoupled to capture different levels of cognition. Upon the top of the representation module, the pre-trained reasoning modules are modular and professional in specific and fundamental reasoning skills (e.g., logic, simple QA, etc). To mimic the controlled compositional thinking process, different reasoning modules are dynamically activated and composed in both parallel and cascaded manners to control what reasoning skills are activated and how deep the reasoning process will be reached to solve the current problems. The unified reasoning framework solves multiple tasks with a single model, and is trained and inferred in an end-to-end manner. Evaluated on 11 datasets requiring different reasoning skills and complexity, ReasonFormer demonstrates substantial performance boosts, revealing the compositional reasoning ability. Few-shot experiments exhibit better generalization ability by learning to compose pre-trained skills for new tasks with limited data, and decoupling the representation module and the reasoning modules. Further analysis shows the modularity of reasoning modules as different tasks activate distinct reasoning skills at different reasoning depths.

pdf abs
Towards Argument-Aware Abstractive Summarization of Long Legal Opinions with Summary Reranking
Mohamed Elaraby | Yang Zhong | Diane Litman

We propose a simple approach for the abstractive summarization of long legal opinions that takes into account the argument structure of the document. Legal opinions often contain complex and nuanced argumentation, making it challenging to generate a concise summary that accurately captures the main points of the legal opinion. Our approach involves using argument role information to generate multiple candidate summaries, then reranking these candidates based on alignment with the document’s argument structure. We demonstrate the effectiveness of our approach on a dataset of long legal opinions and show that it outperforms several strong baselines.

pdf abs
Probabilistic Transformer: A Probabilistic Dependency Model for Contextual Word Representation
Haoyi Wu | Kewei Tu

Syntactic structures used to play a vital role in natural language processing (NLP), but since the deep learning revolution, NLP has been gradually dominated by neural models that do not consider syntactic structures in their design. One vastly successful class of neural models is transformers. When used as an encoder, a transformer produces contextual representation of words in the input sentence. In this work, we propose a new model of contextual word representation, not from a neural perspective, but from a purely syntactic and probabilistic perspective. Specifically, we design a conditional random field that models discrete latent representations of all words in a sentence as well as dependency arcs between them; and we use mean field variational inference for approximate inference. Strikingly, we find that the computation graph of our model resembles transformers, with correspondences between dependencies and self-attention and between distributions over latent representations and contextual embeddings of words. Experiments show that our model performs competitively to transformers on small to medium sized datasets. We hope that our work could help bridge the gap between traditional syntactic and probabilistic approaches and cutting-edge neural approaches to NLP, and inspire more linguistically-principled neural approaches in the future.

pdf abs
Joint Speech Transcription and Translation: Pseudo-Labeling with Out-of-Distribution Data
Mozhdeh Gheini | Tatiana Likhomanenko | Matthias Sperber | Hendra Setiawan

Self-training has been shown to be helpful in addressing data scarcity for many domains, including vision, speech, and language. Specifically, self-training, or pseudo-labeling, labels unsupervised data and adds that to the training pool. In this work, we investigate and use pseudo-labeling for a recently proposed novel setup: joint transcription and translation of speech, which suffers from an absence of sufficient parallel data resources. We show that under such data-deficient circumstances, the unlabeled data can significantly vary in domain from the supervised data, which results in pseudo-label quality degradation. We investigate two categories of remedies that require no additional supervision and target the domain mismatch: pseudo-label filtering and data augmentation. We show that pseudo-label analysis and processing in this way results in additional gains on top of the vanilla pseudo-labeling setup providing a total improvement of up to 0.4% absolute WER and 2.1 BLEU points for En–De and 0.6% absolute WER and 2.2 BLEU points for En–Zh.

pdf abs
Word-level Prefix/Suffix Sense Detection: A Case Study on Negation Sense with Few-shot Learning
Yameng Li | Zicheng Li | Ying Chen | Shoushan Li

Morphological analysis is an important research issue in the field of natural language processing. In this study, we propose a context-free morphological analysis task, namely word-level prefix/suffix sense detection, which deals with the ambiguity of sense expressed by prefix/suffix. To research this novel task, we first annotate a corpus with prefixes/suffixes expressing negation (e.g., il-, un-, -less) and then propose a novel few-shot learning approach that applies an input-augmentation prompt to a token-replaced detection pre-training model. Empirical studies demonstrate the effectiveness of the proposed approach to word-level prefix/suffix negation sense detection.

pdf abs
End-to-End Simultaneous Speech Translation with Differentiable Segmentation
Shaolei Zhang | Yang Feng

End-to-end simultaneous speech translation (SimulST) outputs translation while receiving the streaming speech inputs (a.k.a. streaming speech translation), and hence needs to segment the speech inputs and then translate based on the current received speech. However, segmenting the speech inputs at unfavorable moments can disrupt the acoustic integrity and adversely affect the performance of the translation model. Therefore, learning to segment the speech inputs at those moments that are beneficial for the translation model to produce high-quality translation is the key to SimulST. Existing SimulST methods, either using the fixed-length segmentation or external segmentation model, always separate segmentation from the underlying translation model, where the gap results in segmentation outcomes that are not necessarily beneficial for the translation process. In this paper, we propose Differentiable Segmentation (DiSeg) for SimulST to directly learn segmentation from the underlying translation model. DiSeg turns hard segmentation into differentiable through the proposed expectation training, enabling it to be jointly trained with the translation model and thereby learn translation-beneficial segmentation. Experimental results demonstrate that DiSeg achieves state-of-the-art performance and exhibits superior segmentation capability.

Generate-then-rank is a widely used mechanism for text generation, where a generator produces multiple text candidates and a ranker chooses the best one among the text candidates. However, existing methods usually train the generator and the ranker individually, neglecting the mutual feedback that could further enhance the generation quality. To tackle this limitation, we propose JGR, a novel joint training algorithm that integrates the generator and the ranker in a single framework. JGR optimizes the generator with a hybrid objective that combines data likelihood and ranker reward, and trains the ranker with a contrastive loss that compares the generator outputs. By iteratively updating the generator and the ranker, JGR can effectively harmonize their learning and enhance their quality jointly. We evaluate JGR on various text generation tasks and demonstrate that it surpasses existing methods on four public datasets across three common generation scenarios. Our code and models are publicly available at https://github.com/microsoft/ProphetNet/tree/master/JGR.

pdf abs
Multilingual Sequence-to-Sequence Models for Hebrew NLP
Matan Eyal | Hila Noga | Roee Aharoni | Idan Szpektor | Reut Tsarfaty

Recent work attributes progress in NLP to large language models (LMs) with increased model size and large quantities of pretraining data. Despite this, current state-of-the-art LMs for Hebrew are both under-parameterized and under-trained compared to LMs in other languages. Additionally, previous work on pretrained Hebrew LMs focused on encoder-only models. While the encoder-only architecture is beneficial for classification tasks, it does not cater well for sub-word prediction tasks, such as Named Entity Recognition, when considering the morphologically rich nature of Hebrew. In this paper we argue that sequence-to-sequence generative architectures are more suitable for large LMs in morphologically rich languages (MRLs) such as Hebrew. We demonstrate this by casting tasks in the Hebrew NLP pipeline as text-to-text tasks, for which we can leverage powerful multilingual, pretrained sequence-to-sequence models as mT5, eliminating the need for a separate, specialized, morpheme-based, decoder. Using this approach, our experiments show substantial improvements over previously published results on all existing Hebrew NLP benchmarks. These results suggest that multilingual sequence-to-sequence models present a promising building block for NLP for MRLs.

Multilingual Knowledge Graph Completion (mKGC) aim at solving queries in different languages by reasoning a tail entity thus improving multilingual knowledge graphs. Previous studies leverage multilingual pretrained language models (PLMs) and the generative paradigm to achieve mKGC. Although multilingual pretrained language models contain extensive knowledge of different languages, its pretraining tasks cannot be directly aligned with the mKGC tasks. Moreover, the majority of KGs and PLMs currently available exhibit a pronounced English-centric bias. This makes it difficult for mKGC to achieve good results, particularly in the context of low-resource languages. To overcome previous problems, this paper introduces global and local knowledge constraints for mKGC. The former is used to constrain the reasoning of answer entities , while the latter is used to enhance the representation of query contexts. The proposed method makes the pretrained model better adapt to the mKGC task. Experimental results on public datasets demonstrate that our method outperforms the previous SOTA on Hits@1 and Hits@10 by an average of 12.32% and 16.03%, which indicates that our proposed method has significant enhancement on mKGC.

Hierarchical text classification (HTC) focuses on classifying one text into multiple labels, which are organized as a hierarchical taxonomy. Due to its wide involution in realistic scenarios, HTC attracts long-term attention from both industry and academia. However, the high cost of hierarchical multi-label annotation makes HTC suffer from the data scarcity problem. In view of the difficulty in balancing the controllability of multiple structural labels and text diversity, automatically generating high-quality data for HTC is challenging and under-explored. To fill this blank, we propose a novel data generation framework tailored for HTC, which can achieve both label controllability and text diversity by extracting high-quality semantic-level and phrase-level hierarchical label information. Experimental results on three benchmarks demonstrate that, compared with existing data augmentation methods, the data generated from our method can bring the most significant performance improvements of several strong HTC models. Extensive analysis confirms that the improvements yielded by our proposed method do correlate to the enhancement of label controllability and text diversity.

pdf abs
History repeats: Overcoming catastrophic forgetting for event-centric temporal knowledge graph completion
Mehrnoosh Mirtaheri | Mohammad Rostami | Aram Galstyan

Temporal knowledge graph (TKG) completion models typically rely on having access to the entire graph during training. However, in real-world scenarios, TKG data is often received incrementally as events unfold, leading to a dynamic non-stationary data distribution over time. While one could incorporate fine-tuning to existing methods to allow them to adapt to evolving TKG data, this can lead to forgetting previously learned patterns. Alternatively, retraining the model with the entire updated TKG can mitigate forgetting but is computationally burdensome. To address these challenges, we propose a general continual training framework that is applicable to any TKG completion method, and leverages two key ideas: (i) a temporal regularization that encourages repurposing of less important model parameters for learning new knowledge, and (ii) a clustering-based experience replay that reinforces the past knowledge by selectively preserving only a small portion of the past data. Our experimental results on widely used event-centric TKG datasets demonstrate the effectiveness of our proposed continual training framework in adapting to new events while reducing catastrophic forgetting. Further, we perform ablation studies to show the effectiveness of each component of our proposed framework. Finally, we investigate the relation between the memory dedicated to experience replay and the benefit gained from our clustering-based sampling strategy.

pdf abs
Multi-Agent Language Learning: Symbolic Mapping
Yicheng Feng | Zongqing Lu

The study of emergent communication has long been devoted to coax neural network agents to learn a language sharing similar properties with human language. In this paper, we try to find a ‘natural’ way to help agents learn a compositional and symmetric language in complex settings like dialog games. Inspired by the theory that human language was originated from simple interactions, we hypothesize that language may evolve from simple tasks to difficult tasks. We propose a curriculum learning method called task transfer, and propose a novel architecture called symbolic mapping. We find that task transfer distinctly helps language learning in difficult tasks, and symbolic mapping promotes the effect. Further, we explore vocabulary expansion, and show that with the help of symbolic mapping, agents can easily learn to use new symbols when the environment becomes more complex. All in all, we find that a process from simplicity to complexity can serve as a natural way to help multi-agent language learning, and the proposed symbolic mapping is effective for this process.

pdf abs
Scaling Laws for BERT in Low-Resource Settings
Gorka Urbizu | Iñaki San Vicente | Xabier Saralegi | Rodrigo Agerri | Aitor Soroa

Large language models are very resource intensive, both financially and environmentally, and require an amount of training data which is simply unobtainable for the majority of NLP practitioners. Previous work has researched the scaling laws of such models, but optimal ratios of model parameters, dataset size, and computation costs focused on the large scale. In contrast, we analyze the effect those variables have on the performance of language models in constrained settings, by building three lightweight BERT models (16M/51M/124M parameters) trained over a set of small corpora (5M/25M/125M words).We experiment on four languages of different linguistic characteristics (Basque, Spanish, Swahili and Finnish), and evaluate the models on MLM and several NLU tasks. We conclude that the power laws for parameters, data and compute for low-resource settings differ from the optimal scaling laws previously inferred, and data requirements should be higher. Our insights are consistent across all the languages we study, as well as across the MLM and downstream tasks. Furthermore, we experimentally establish when the cost of using a Transformer-based approach is worth taking, instead of favouring other computationally lighter solutions.

pdf abs
Pre-trained Language Model with Prompts for Temporal Knowledge Graph Completion
Wenjie Xu | Ben Liu | Miao Peng | Xu Jia | Min Peng

Temporal Knowledge graph completion (TKGC) is a crucial task that involves reasoning at known timestamps to complete the missing part of facts and has attracted more and more attention in recent years. Most existing methods focus on learning representations based on graph neural networks while inaccurately extracting information from timestamps and insufficiently utilizing the implied information in relations. To address these problems, we propose a novel TKGC model, namely Pre-trained Language Model with Prompts for TKGC (PPT). We convert a series of sampled quadruples into pre-trained language model inputs and convert intervals between timestamps into different prompts to make coherent sentences with implicit semantic information. We train our model with a masking strategy to convert TKGC task into a masked token prediction task, which can leverage the semantic information in pre-trained language models. Experiments on three benchmark datasets and extensive analysis demonstrate that our model has great competitiveness compared to other models with four metrics. Our model can effectively incorporate information from temporal knowledge graphs into the language models.

pdf abs
Is Continuous Prompt a Combination of Discrete Prompts? Towards a Novel View for Interpreting Continuous Prompts
Tianjie Ju | Yubin Zheng | Hanyi Wang | Haodong Zhao | Gongshen Liu

The broad adoption of continuous prompts has brought state-of-the-art results on a diverse array of downstream natural language processing (NLP) tasks. Nonetheless, little attention has been paid to the interpretability and transferability of continuous prompts. Faced with the challenges, we investigate the feasibility of interpreting continuous prompts as the weighting of discrete prompts by jointly optimizing prompt fidelity and downstream fidelity. Our experiments show that: (1) one can always find a combination of discrete prompts as the replacement of continuous prompts that performs well on downstream tasks; (2) our interpretable framework faithfully reflects the reasoning process of source prompts; (3) our interpretations provide effective readability and plausibility, which is helpful to understand the decision-making of continuous prompts and discover potential shortcuts. Moreover, through the bridge constructed between continuous prompts and discrete prompts using our interpretations, it is promising to implement the cross-model transfer of continuous prompts without extra training signals. We hope this work will lead to a novel perspective on the interpretations of continuous prompts.

pdf abs
Putting Natural in Natural Language Processing
Grzegorz Chrupała

Human language is firstly spoken and only secondarily written. Text, however, is a very convenient and efficient representation oflanguage, and modern civilization has made it ubiquitous. Thus the field of NLP has overwhelmingly focused on processing written ratherthan spoken language. Work on spoken language, on the other hand, has been siloed off within the largely separate speech processingcommunity which has been inordinately preoccupied with transcribing speech into text. Recent advances in deep learning have led to afortuitous convergence in methods between speech processing and mainstream NLP. Arguably, the time is ripe for a unification of thesetwo fields, and for starting to take spoken language seriously as the primary mode of human communication. Truly natural language processingcould lead to better integration with the rest of language science and could lead to systems which are more data-efficient and morehuman-like, and which can communicate beyond the textual modality.

pdf abs
Impact of Adversarial Training on Robustness and Generalizability of Language Models
Enes Altinisik | Hassan Sajjad | Husrev Sencar | Safa Messaoud | Sanjay Chawla

Adversarial training is widely acknowledged as the most effective defense against adversarial attacks. However, it is also well established that achieving both robustness and generalization in adversarially trained models involves a trade-off. The goal of this work is to provide an in depth comparison of different approaches for adversarial training in language models. Specifically, we study the effect of pre-training data augmentation as well as training time input perturbations vs. embedding space perturbations on the robustness and generalization of transformer-based language models. Our findings suggest that better robustness can be achieved by pre-training data augmentation or by training with input space perturbation. However, training with embedding space perturbation significantly improves generalization. A linguistic correlation analysis of neurons of the learned models reveal that the improved generalization is due to ‘more specialized’ neurons. To the best of our knowledge, this is the first work to carry out a deep qualitative analysis of different methods of generating adversarial examples in adversarial training of language models.

Entities can be expressed in diverse formats, such as texts, images, or column names and cell values in tables. While existing entity linking (EL) models work well on per modality configuration, such as text-only EL, visual grounding or schema linking, it is more challenging to design a unified model for diverse modality configurations. To bring various modality configurations together, we constructed a benchmark for diverse-modal EL (DMEL) from existing EL datasets, covering all three modalities including text, image and table. To approach the DMEL task, we proposed a generative diverse-modal model (GDMM) following a multimodal-encoder-decoder paradigm. Pre-training GDMM with rich corpora builds a solid foundation for DMEL without storing the entire KB for inference. Fine-tuning GDMM builds a stronger DMEL baseline, outperforming state-of-the-art task-specific EL models by 8.51 F1 score on average. Additionally, extensive error analyses are conducted to highlight the challenge of DMEL, facilitating future researches on this task.

In empathetic conversations, individuals express their empathy towards others. Previous work has mainly focused on generating empathetic responses by utilizing the speaker’s emotion. Besides, external commonsense knowledge has been applied to enhance the system’s understandings of the speaker’s situation. However, given an event, commonsense knowledge base contains various relations, potentially leading to confusion for the dialogue system. Consequently, inconsistencies arise among the emotion, generated response and speaker’s contextual information. To this end, we propose a novel approach for empathetic response generation, which incorporates an adaptive module for commonsense knowledge selection to ensure consistency between the generated empathetic responses and the speaker’s situation. This selected knowledge is used to refine the commonsense cognition and empathy expression for generated responses. Experimental results show that our approach significantly outperforms baseline models in both automatic and human evaluations, exhibiting the generation of more coherent and empathetic responses. Moreover, case studies highlight the interpretability of knowledge selection in the responses and the effectiveness of adaptive module in our model. Code: https://github.com/Hanscal/DCKS.

pdf abs
Additive manifesto decomposition: A policy domain aware method for understanding party positioning
Tanise Ceron | Dmitry Nikolaev | Sebastian Padó

Automatic extraction of party (dis)similarities from texts such as party election manifestos or parliamentary speeches plays an increasing role in computational political science. However, existing approaches are fundamentally limited to targeting only global party (dis)-similarity: they condense the relationship between a pair of parties into a single figure, their similarity. In aggregating over all policy domains (e.g., health or foreign policy), they do not provide any qualitative insights into which domains parties agree or disagree on. This paper proposes a workflow for estimating policy domain aware party similarity that overcomes this limitation. The workflow covers (a) definition of suitable policy domains; (b) automatic labeling of domains, if no manual labels are available; (c) computation of domain-level similarities and aggregation at a global level; (d) extraction of interpretable party positions on major policy axes via multidimensional scaling. We evaluate our workflow on manifestos from the German federal elections. We find that our method (a) yields high correlation when predicting party similarity at a global level and (b) provides accurate party-specific positions, even with automatically labelled policy domains.

Neural language models are vulnerable to word-level adversarial text attacks, which generate adversarial examples by directly substituting discrete input words. Previous search methods for word-level attacks assume that the information in the important words is more influential on prediction than unimportant words. In this paper, motivated by this assumption, we propose a self-supervised regularization method for Similarizing the Influence of Words with Contrastive Learning (SIWCon) that encourages the model to learn sentence representations in which words of varying importance have a more uniform influence on prediction. Experiments show that SIWCon is compatible with various training methods and effectively improves model robustness against various unforeseen adversarial attacks. The effectiveness of SIWCon is also intuitively shown through qualitative analysis and visualization of the loss landscape, sentence representation, and changes in model confidence.

pdf abs
Responsibility Perspective Transfer for Italian Femicide News
Gosse Minnema | Huiyuan Lai | Benedetta Muscato | Malvina Nissim

Different ways of linguistically expressing the same real-world event can lead to different perceptions of what happened. Previous work has shown that different descriptions of gender-based violence (GBV) influence the reader’s perception of who is to blame for the violence, possibly reinforcing stereotypes which see the victim as partly responsible, too. As a contribution to raise awareness on perspective-based writing, and to facilitate access to alternative perspectives, we introduce the novel task of automatically rewriting GBV descriptions as a means to alter the perceived level of blame on the perpetrator. We present a quasi-parallel dataset of sentences with low and high perceived responsibility levels for the perpetrator, and experiment with unsupervised (mBART-based), zero-shot and few-shot (GPT3-based) methods for rewriting sentences. We evaluate our models using a questionnaire study and a suite of automatic metrics.

pdf abs
Stereotypes and Smut: The (Mis)representation of Non-cisgender Identities by Text-to-Image Models
Eddie Ungless | Bjorn Ross | Anne Lauscher

Cutting-edge image generation has been praised for producing high-quality images, suggesting a ubiquitous future in a variety of applications. However, initial studies have pointed to the potential for harm due to predictive bias, reflecting and potentially reinforcing cultural stereotypes. In this work, we are the first to investigate how multimodal models handle diverse gender identities. Concretely, we conduct a thorough analysis in which we compare the output of three image generation models for prompts containing cisgender vs. non-cisgender identity terms. Our findings demonstrate that certain non-cisgender identities are consistently (mis)represented as less human, more stereotyped and more sexualised. We complement our experimental analysis with (a) a survey among non-cisgender individuals and (b) a series of interviews, to establish which harms affected individuals anticipate, and how they would like to be represented. We find respondents are particularly concerned about misrepresentation, and the potential to drive harmful behaviours and beliefs. Simple heuristics to limit offensive content are widely rejected, and instead respondents call for community involvement, curated training data and the ability to customise. These improvements could pave the way for a future where change is led by the affected community, and technology is used to positively ”[portray] queerness in ways that we haven’t even thought of”’ rather than reproducing stale, offensive stereotypes.

The Wav2Vec and its variants have achieved unprecedented success in computational auditory and speech processing. Meanwhile, neural encoding studies that integrate the superb representation capability of Wav2Vec and link those representations to brain activities have provided novel insights into a fundamental question of how auditory and speech processing unfold in the human brain. Without an explicit definition, most existing studies treat each transformer encoding layer in Wav2Vec as a single artificial neuron (AN). That is, the layer-level embeddings are used to predict neural responses. However, the comprehensive layer-level embedding aggregates multiple types of contextual attention captured by multi-head self-attention (MSA) modules. Thus, the layer-level ANs lack fine-granularity for neural encoding. To address this limitation, we define the elementary units, i.e., each hidden dimension, as neuron-level ANs in Wav2Vec2.0, quantify their temporal responses, and couple those ANs with their biological-neuron (BN) counterparts in the human brain. Our experimental results demonstrated that: 1) The proposed neuron-level ANs carry meaningful neurolinguistic information; 2) Those ANs anchor to their BN signatures; 3) The AN-BN anchoring patterns are interpretable from a neurolinguistic perspective. More importantly, our results suggest an intermediate stage in both the computational representation in Wav2Vec2.0 and the cortical representation in the brain. Our study validates the fine-grained ANs in Wav2Vec2.0, which may serve as a novel and general strategy to link transformer-based deep learning models to neural responses for probing the sensory processing in the brain.

Recent advancements in multimodal foundation models (e.g., CLIP) have excelled in zero-shot generalization. Prompt tuning involved in the knowledge transfer from foundation models to downstream tasks has gained significant attention recently. Existing prompt-tuning methods in cross-modal learning, however, either solely focus on language branch, or learn vision-language interaction in a shallow mechanism. In this context, we propose a Deeply coupled Cross-modal Prompt learning (DCP) method based on CLIP. DCP flexibly accommodates the interplay between vision and language with a Cross-Modal Prompt Attention (CMPA) mechanism, which enables the mutual exchange of respective representation through a well-connected multi-head attention progressively and strongly. We then conduct comprehensive few-shot learning experiments on 11 image classification datasets and analyze the robustness to domain shift as well. Thorough experimental analysis evidently demonstrates the superb few-shot generalization and compelling domain adaption capacity of a well-executed DCP.

pdf abs
Opinion Tree Parsing for Aspect-based Sentiment Analysis
Xiaoyi Bao | Xiaotong Jiang | Zhongqing Wang | Yue Zhang | Guodong Zhou

Extracting sentiment elements using pre-trained generative models has recently led to large improvements in aspect-based sentiment analysis benchmarks. These models avoid explicit modeling of structure between sentiment elements, which are succinct yet lack desirable properties such as structure well-formedness guarantees or built-in elements alignments. In this study, we propose an opinion tree parsing model, aiming to parse all the sentiment elements from an opinion tree, which can explicitly reveal a more comprehensive and complete aspect-level sentiment structure. In particular, we first introduce a novel context-free opinion grammar to normalize the sentiment structure. We then employ a neural chart-based opinion tree parser to fully explore the correlations among sentiment elements and parse them in the opinion tree form. Extensive experiments show the superiority of our proposed model and the capacity of the opinion tree parser with the proposed context-free opinion grammar. More importantly, our model is much faster than previous models.

pdf abs
CoMix: Guide Transformers to Code-Mix using POS structure and Phonetics
Gaurav Arora | Srujana Merugu | Vivek Sembium

Code-mixing is ubiquitous in multilingual societies, which makes it vital to build models for code-mixed data to power human language interfaces. Existing multilingual transformer models trained on pure corpora lack the ability to intermix words of one language into the structure of another. These models are also not robust to orthographic variations. We propose CoMixCoMix is not a trademark and only used to refer to our models for code-mixed data for presentational brevity., a pretraining approach to improve representation of code-mixed data in transformer models by incorporating phonetic signals, a modified attention mechanism, and weak supervision guided generation by parts-of-speech constraints. We show that CoMix improves performance across four code-mixed tasks: machine translation, sequence classification, named entity recognition (NER), and abstractive summarization. It also achieves the new SOTA performance for English-Hinglish translation and NER on LINCE Leaderboard and provides better generalization on out-of-domain translation. Motivated by variations in human annotations, we also propose a new family of metrics based on phonetics and demonstrate that the phonetic variant of BLEU correlates better with human judgement than BLEU on code-mixed text.

Deploying large language models (LLMs) is challenging because they are memory inefficient and compute-intensive for practical applications. In reaction, researchers train smaller task-specific models by either finetuning with human labels or distilling using LLM-generated labels. However, finetuning and distillation require large amounts of training data to achieve comparable performance to LLMs. We introduce Distilling step-by-step, a new mechanism that (a) trains smaller models that outperform LLMs, and (b) achieves so by leveraging less training data needed by finetuning or distillation. Our method extracts LLM rationales as additional supervision for training small models within a multi-task framework. We present three findings across 4 NLP benchmarks: First, compared to both finetuning and distillation, our mechanism achieves better performance with much fewer labeled/unlabeled training examples. Second, compared to few-shot prompted LLMs, we achieve better performance using substantially smaller model sizes. Third, we reduce both the model size and the amount of data required to outperform LLMs; our finetuned 770M T5 model outperforms the few-shot prompted 540B PaLM model using only 80% of available data on a benchmark, whereas standard finetuning the same T5 model struggles to match even by using 100% of the dataset.

pdf abs
Prosody-TTS: Improving Prosody with Masked Autoencoder and Conditional Diffusion Model For Expressive Text-to-Speech
Rongjie Huang | Chunlei Zhang | Yi Ren | Zhou Zhao | Dong Yu

Expressive text-to-speech aims to generate high-quality samples with rich and diverse prosody, which is hampered by dual challenges: 1) prosodic attributes in highly dynamic voices are difficult to capture and model without intonation; and 2) highly multimodal prosodic representations cannot be well learned by simple regression (e.g., MSE) objectives, which causes blurry and over-smoothing predictions. This paper proposes Prosody-TTS, a two-stage pipeline that enhances prosody modeling and sampling by introducing several components: 1) a self-supervised masked autoencoder to model the prosodic representation without relying on text transcriptions or local prosody attributes, which ensures to cover diverse speaking voices with superior generalization; and 2) a diffusion model to sample diverse prosodic patterns within the latent space, which prevents TTS models from generating samples with dull prosodic performance. Experimental results show that Prosody-TTS achieves new state-of-the-art in text-to-speech with natural and expressive synthesis. Both subjective and objective evaluation demonstrate that it exhibits superior audio quality and prosody naturalness with rich and diverse prosodic attributes. Audio samples are available at https://improved_prosody.github.io

pdf abs
Duplex Diffusion Models Improve Speech-to-Speech Translation
Xianchao Wu

Speech-to-speech translation is a typical sequence-to-sequence learning task that naturally has two directions. How to effectively leverage bidirectional supervision signals to produce high-fidelity audio for both directions? Existing approaches either train two separate models or a multitask-learned model with low efficiency and inferior performance. In this paper, we propose a duplex diffusion model that applies diffusion probabilistic models to both sides of a reversible duplex Conformer, so that either end can simultaneously input and output a distinct language’s speech. Our model enables reversible speech translation by simply flipping the input and output ends. Experiments show that our model achieves the first success of reversible speech translation with significant improvements of ASR-BLEU scores compared with a list of state-of-the-art baselines.

pdf abs
Global and Local Hierarchy-aware Contrastive Framework for Implicit Discourse Relation Recognition
Yuxin Jiang | Linhan Zhang | Wei Wang

Due to the absence of explicit connectives, implicit discourse relation recognition (IDRR) remains a challenging task in discourse analysis. The critical step for IDRR is to learn high-quality discourse relation representations between two arguments. Recent methods tend to integrate the whole hierarchical information of senses into discourse relation representations for multi-level sense recognition. Nevertheless, they insufficiently incorporate the static hierarchical structure containing all senses (defined as global hierarchy), and ignore the hierarchical sense label sequence corresponding to each instance (defined as local hierarchy). For the purpose of sufficiently exploiting global and local hierarchies of senses to learn better discourse relation representations, we propose a novel GlObal and Local Hierarchy-aware Contrastive Framework (GOLF), to model two kinds of hierarchies with the aid of multi-task learning and contrastive learning. Experimental results on PDTB 2.0 and PDTB 3.0 datasets demonstrate that our method remarkably outperforms current state-of-the-art models at all hierarchical levels.

While transformer-based pre-trained language models (PLMs) have dominated a number of NLP applications, these models are heavy to deploy and expensive to use. Therefore, effectively compressing large-scale PLMs becomes an increasingly important problem. Quantization, which represents high-precision tensors with low-bit fix-point format, is a viable solution. However, most existing quantization methods are task-specific, requiring customized training and quantization with a large number of trainable parameters on each individual task. Inspired by the observation that the over-parameterization nature of PLMs makes it possible to freeze most of the parameters during the fine-tuning stage, in this work, we propose a novel “quantize before fine-tuning” framework, PreQuant, that differs from both quantization-aware training and post-training quantization. {pasted macro ‘OUR’} is compatible with various quantization strategies, with outlier-aware parameter-efficient fine-tuning incorporated to correct the induced quantization error. We demonstrate the effectiveness of PreQuant on the GLUE benchmark using BERT, RoBERTa, and T5. We also provide an empirical investigation into the workflow of PreQuant, which sheds light on its efficacy.

pdf abs
Synthetic Pre-Training Tasks for Neural Machine Translation
Zexue He | Graeme Blackwood | Rameswar Panda | Julian McAuley | Rogerio Feris

Pre-training models with large crawled corpora can lead to issues such as toxicity and bias, as well as copyright and privacy concerns. A promising way of alleviating such concerns is to conduct pre-training with synthetic tasks and data, since no real-world information is ingested by the model. Our goal in this paper is to understand the factors that contribute to the effectiveness of pre-training models when using synthetic resources, particularly in the context of neural machine translation. We propose several novel approaches to pre-training translation models that involve different levels of lexical and structural knowledge, including: 1) generating obfuscated data from a large parallel corpus 2) concatenating phrase pairs extracted from a small word-aligned corpus, and 3) generating synthetic parallel data without real human language corpora. Our experiments on multiple language pairs reveal that pre-training benefits can be realized even with high levels of obfuscation or purely synthetic parallel data. We hope the findings from our comprehensive empirical analysis will shed light on understanding what matters for NMT pre-training, as well as pave the way for the development of more efficient and less toxic models.

pdf abs
IDOL: Indicator-oriented Logic Pre-training for Logical Reasoning
Zihang Xu | Ziqing Yang | Yiming Cui | Shijin Wang

In the field of machine reading comprehension (MRC), existing systems have surpassed the average performance of human beings in many tasks like SQuAD. However, there is still a long way to go when it comes to logical reasoning. Although some methods for it have been put forward, they either are designed in a quite complicated way or rely too much on external structures. In this paper, we proposed IDOL (InDicator-Oriented Logic Pre-training), an easy-to-understand but highly effective further pre-training task which logically strengthens the pre-trained models with the help of 6 types of logical indicators and a logically rich dataset LoGic Pre-training (LGP). IDOL achieves state-of-the-art performance on ReClor and LogiQA, the two most representative benchmarks in logical reasoning MRC, and is proven to be capable of generalizing to different pre-trained models and other types of MRC benchmarks like RACE and SQuAD 2.0 while keeping competitive general language understanding ability through testing on tasks in GLUE. Besides, at the beginning of the era of large language models, we take several of them like ChatGPT into comparison and find that IDOL still shows its advantage.

pdf abs
Adversarial Training for Low-Resource Disfluency Correction
Vineet Bhat | Preethi Jyothi | Pushpak Bhattacharyya

Disfluencies commonly occur in conversational speech. Speech with disfluencies can result in noisy Automatic Speech Recognition (ASR) transcripts, which affects downstream tasks like machine translation. In this paper, we propose an adversarially-trained sequence-tagging model for Disfluency Correction (DC) that utilizes a small amount of labeled real disfluent data in conjunction with a large amount of unlabeled data. We show the benefit of our proposed technique, which crucially depends on synthetically generated disfluent data, by evaluating it for DC in three Indian languages- Bengali, Hindi, and Marathi (all from the Indo-Aryan family). Our technique also performs well in removing stuttering disfluencies in ASR transcripts introduced by speech impairments. We achieve an average 6.15 points improvement in F1-score over competitive baselines across all three languages mentioned. To the best of our knowledge, we are the first to utilize adversarial training for DC and use it to correct stuttering disfluencies in English, establishing a new benchmark for this task.

pdf abs
Computer says “No”: The Case Against Empathetic Conversational AI
Alba Cercas Curry | Amanda Cercas Curry

Emotions are an integral part of human cognition and they guide not only our understanding of the world but also our actions within it. As such, whether we soothe or flame an emotion is not inconsequential. Recent work in conversational AI has focused on responding empathetically to users, validating and soothing their emotions without a real basis. This AI-aided emotional regulation can have negative consequences for users and society, tending towards a one-noted happiness defined as only the absence of “negative” emotions. We argue that we must carefully consider whether and how to respond to users’ emotions.

pdf abs
Stubborn Lexical Bias in Data and Models
Sofia Serrano | Jesse Dodge | Noah A. Smith

In NLP, recent work has seen increased focus on spurious correlations between various features and labels in training data, and how these influence model behavior. However, the presence and effect of such correlations are typically examined feature by feature. We investigate the cumulative impact on a model of many such intersecting features. Using a new statistical method, we examine whether such spurious patterns in data appear in models trained on the data. We select two tasks— natural language inference and duplicate-question detection— for which any unigram feature on its own should ideally be uninformative, which gives us a large pool of automatically extracted features with which to experiment. The large size of this pool allows us to investigate the intersection of features spuriously associated with (potentially different) labels. We then apply an optimization approach to *reweight* the training data, reducing thousands of spurious correlations, and examine how doing so affects models trained on the reweighted data. Surprisingly, though this method can successfully reduce lexical biases in the training data, we still find strong evidence of corresponding bias in the trained models, including worsened bias for slightly more complex features (bigrams). We close with discussion about the implications of our results on what it means to “debias” training data, and how issues of data quality can affect model bias.

pdf abs
Distilling Efficient Language-Specific Models for Cross-Lingual Transfer
Alan Ansell | Edoardo Maria Ponti | Anna Korhonen | Ivan Vulić

Massively multilingual Transformers (MMTs), such as mBERT and XLM-R, are widely used for cross-lingual transfer learning. While these are pretrained to represent hundreds of languages, end users of NLP systems are often interested only in individual languages. For such purposes, the MMTs’ language coverage makes them unnecessarily expensive to deploy in terms of model size, inference time, energy, and hardware cost. We thus propose to extract compressed, language-specific models from MMTs which retain the capacity of the original MMTs for cross-lingual transfer. This is achieved by distilling the MMT *bilingually*, i.e., using data from only the source and target language of interest. Specifically, we use a two-phase distillation approach, termed BiStil: (i) the first phase distils a general bilingual model from the MMT, while (ii) the second, task-specific phase sparsely fine-tunes the bilingual “student” model using a task-tuned variant of the original MMT as its “teacher”. We evaluate this distillation technique in zero-shot cross-lingual transfer across a number of standard cross-lingual benchmarks. The key results indicate that the distilled models exhibit minimal degradation in target language performance relative to the base MMT despite being significantly smaller and faster. Furthermore, we find that they outperform multilingually distilled models such as DistilmBERT and MiniLMv2 while having a very modest training budget in comparison, even on a per-language basis. We also show that bilingual models distilled from MMTs greatly outperform bilingual models trained from scratch.

pdf abs
An Extensive Exploration of Back-Translation in 60 Languages
Paul McNamee | Kevin Duh

Back-translation is a data augmentation technique that has been shown to improve model quality through the creation of synthetic training bitext. Early studies showed the promise of the technique and follow on studies have produced additional refinements. We have undertaken a broad investigation using back-translation to train models from 60 languages into English; the majority of these languages are considered moderate- or low-resource languages. We observed consistent gains, though compared to prior work we saw conspicuous gains in quite a number of lower-resourced languages. We analyzed differences in translations between baseline and back-translation models, and observed many indications of improved translation quality. Translation of both rare and common terms is improved, and these improvements occur despite the less natural synthetic source-language text used in training.

Multimodal aspect-based sentiment analysis (MABSA) aims to extract aspects from text-image pairs and recognize their sentiments. Existing methods make great efforts to align the whole image to corresponding aspects. However, different regions of the image may relate to different aspects in the same sentence, and coarsely establishing image-aspect alignment will introduce noise to aspect-based sentiment analysis (i.e., visual noise). Besides, the sentiment of a specific aspect can also be interfered by descriptions of other aspects (i.e., textual noise). Considering the aforementioned noises, this paper proposes an Aspect-oriented Method (AoM) to detect aspect-relevant semantic and sentiment information. Specifically, an aspect-aware attention module is designed to simultaneously select textual tokens and image blocks that are semantically related to the aspects. To accurately aggregate sentiment information, we explicitly introduce sentiment embedding into AoM, and use a graph convolutional network to model the vision-text and text-text interaction. Extensive experiments demonstrate the superiority of AoM to existing methods.

pdf abs
Forecasting Earnings Surprises from Conference Call Transcripts
Ross Koval | Nicholas Andrews | Xifeng Yan

There is a multitude of textual data relevant to the financial markets, spanning genres such as financial news, earnings conference calls, and social media posts. Earnings conference calls are one of the most important to information flow as they reflect a direct communication between company executives, financial analysts, and large shareholders. Since these calls contain content that is forward-looking in nature, they can be used to forecast the future performance of the company relative to market expectations. However, they typically contain over 5,000 words of text and large amounts of industry jargon. This length and domain-specific language present problems for many generic pretrained language models. In this work, we introduce a novel task of predicting earnings surprises from earnings call transcripts and contribute a new long document dataset that tests financial understanding with complex signals. We explore a variety of approaches for this long document classification task and establish some strong baselines. Furthermore, we demonstrate that it is possible to predict companies’ future earnings surprises from solely the text of their conference calls with reasonable accuracy. Finally, we probe the models through different interpretability methods and reveal some intuitive explanations of the linguistic features captured that go beyond traditional sentiment analysis.

pdf abs
MTCue: Learning Zero-Shot Control of Extra-Textual Attributes by Leveraging Unstructured Context in Neural Machine Translation
Sebastian Vincent | Robert Flynn | Carolina Scarton

Efficient utilisation of both intra- and extra-textual context remains one of the critical gaps between machine and human translation. Existing research has primarily focused on providing individual, well-defined types of context in translation, such as the surrounding text or discrete external variables like the speaker’s gender. This work introduces MTCue, a novel neural machine translation (NMT) framework that interprets all context (including discrete variables) as text. MTCue learns an abstract representation of context, enabling transferability across different data settings and leveraging similar attributes in low-resource scenarios. With a focus on a dialogue domain with access to document and metadata context, we extensively evaluate MTCue in four language pairs in both translation directions. Our framework demonstrates significant improvements in translation quality over a parameter-matched non-contextual baseline, as measured by BLEU (+0.88) and Comet (+1.58). Moreover, MTCue significantly outperforms a “tagging” baseline at translating English text. Analysis reveals that the context encoder of MTCue learns a representation space that organises context based on specific attributes, such as formality, enabling effective zero-shot control. Pre-training on context embeddings also improves MTCue’s few-shot performance compared to the “tagging” baseline. Finally, an ablation study conducted on model components and contextual variables further supports the robustness of MTCue for context-based NMT.

pdf abs
Evaluation for Change
Rishi Bommasani

Evaluation is the central means for assessing, understanding, and communicating about NLP models. In this position paper, we argue evaluation should be more than that: it is a force for driving change, carrying a sociological and political character beyond its technical dimensions. As a force, evaluation’s power arises from its adoption: under our view, evaluation succeeds when it achieves the desired change in the field. Further, by framing evaluation as a force, we consider how it competes with other forces. Under our analysis, we conjecture that the current trajectory of NLP suggests evaluation’s power is waning, in spite of its potential for realizing more pluralistic ambitions in the field. We conclude by discussing the legitimacy of this power, who acquires this power and how it distributes. Ultimately, we hope the research community will more aggressively harness evaluation to drive change.

pdf abs
Reconstruction Probing
Najoung Kim | Jatin Khilnani | Alex Warstadt | Abdelrahim Qaddoumi

We propose reconstruction probing, a new analysis method for contextualized representations based on reconstruction probabilities in masked language models (MLMs). This method relies on comparing the reconstruction probabilities of tokens in a given sequence when conditioned on the representation of a single token that has been fully contextualized and when conditioned on only the decontextualized lexical prior of the model. This comparison can be understood as quantifying the contribution of contextualization towards reconstruction—the difference in the reconstruction probabilities can only be attributed to the representational change of the single token induced by contextualization. We apply this analysis to three MLMs and find that contextualization boosts reconstructability of tokens that are close to the token being reconstructed in terms of linear and syntactic distance. Furthermore, we extend our analysis to finer-grained decomposition of contextualized representations, and we find that these boosts are largely attributable to static and positional embeddings at the input layer.

pdf abs
Towards Distribution-shift Robust Text Classification of Emotional Content
Luana Bulla | Aldo Gangemi | Misael Mongiovi’

Supervised models based on Transformers have been shown to achieve impressive performances in many natural language processing tasks. However, besides requiring a large amount of costly manually annotated data, supervised models tend to adapt to the characteristics of the training dataset, which are usually created ad-hoc and whose data distribution often differs from the one in real applications, showing significant performance degradation in real-world scenarios. We perform an extensive assessment of the out-of-distribution performances of supervised models for classification in the emotion and hate-speech detection tasks and show that NLI-based zero-shot models often outperform them, making task-specific annotation useless when the characteristics of final-user data are not known in advance. To benefit from both supervised and zero-shot approaches, we propose to fine-tune an NLI-based model on the task-specific dataset. The resulting model often outperforms all available supervised models both in distribution and out of distribution, with only a few thousand training samples.

Figurative language permeates human communication, but at the same time is relatively understudied in NLP. Datasets have been created in English to accelerate progress towards measuring and improving figurative language processing in language models (LMs). However, the use of figurative language is an expression of our cultural and societal experiences, making it difficult for these phrases to be universally applicable. In this work, we create a figurative language inference dataset, {pasted macro ‘DATASETNAME’}, for seven diverse languages associated with a variety of cultures: Hindi, Indonesian, Javanese, Kannada, Sundanese, Swahili and Yoruba. Our dataset reveals that each language relies on cultural and regional concepts for figurative expressions, with the highest overlap between languages originating from the same region. We assess multilingual LMs’ abilities to interpret figurative language in zero-shot and few-shot settings. All languages exhibit a significant deficiency compared to English, with variations in performance reflecting the availability of pre-training and fine-tuning data, emphasizing the need for LMs to be exposed to a broader range of linguistic and cultural variation during training. Data and code is released at https://anonymous.4open.science/r/Multilingual-Fig-QA-7B03/

pdf abs
Open-WikiTable : Dataset for Open Domain Question Answering with Complex Reasoning over Table
Sunjun Kweon | Yeonsu Kwon | Seonhee Cho | Yohan Jo | Edward Choi

Despite recent interest in open domain question answering (ODQA) over tables, many studies still rely on datasets that are not truly optimal for the task with respect to utilizing structural nature of table. These datasets assume answers reside as a single cell value and do not necessitate exploring over multiple cells such as aggregation, comparison, and sorting. Thus, we release Open-WikiTable, the first ODQA dataset that requires complex reasoning over tables. Open-WikiTable is built upon WikiSQL and WikiTableQuestions to be applicable in the open-domain setting. As each question is coupled with both textual answers and SQL queries, Open-WikiTable opens up a wide range of possibilities for future research, as both reader and parser methods can be applied. The dataset is publicly available.

pdf abs
What In-Context Learning “Learns” In-Context: Disentangling Task Recognition and Task Learning
Jane Pan | Tianyu Gao | Howard Chen | Danqi Chen

Large language models (LLMs) exploit in-context learning (ICL) to solve tasks with only a few demonstrations, but its mechanisms are not yet well-understood. Some works suggest that LLMs only recall already learned concepts from pre-training, while others hint that ICL performs implicit learning over demonstrations. We characterize two ways through which ICL leverages demonstrations. Task recognition (TR) captures the extent to which LLMs can recognize a task through demonstrations – even without ground-truth labels – and apply their pre-trained priors, whereas task learning (TL) is the ability to capture new input-label mappings unseen in pre-training. Using a wide range of classification datasets and three LLM families (GPT-3, LLaMA and OPT), we design controlled experiments to disentangle the roles of TR and TL in ICL. We show that (1) models can achieve non-trivial performance with only TR, and TR does not further improve with larger models or more demonstrations; (2) LLMs acquire TL as the model scales, and TL’s performance consistently improves with more demonstrations in context. Our findings unravel two different forces behind ICL and we advocate for discriminating them in future ICL research due to their distinct nature.

pdf abs
Cross-Lingual Retrieval Augmented Prompt for Low-Resource Languages
Ercong Nie | Sheng Liang | Helmut Schmid | Hinrich Schütze

Multilingual Pretrained Language Models (MPLMs) perform strongly in cross-lingual transfer. We propose Prompts Augmented by Retrieval Crosslingually (PARC) to improve zero-shot performance on low-resource languages (LRLs) by augmenting the context with prompts consisting of semantically similar sentences retrieved from a high-resource language (HRL). PARC improves zero-shot performance on three downstream tasks (sentiment classification, topic categorization, natural language inference) with multilingual parallel test sets across 10 LRLs covering 6 language families in unlabeled (+5.1%) and labeled settings (+16.3%). PARC also outperforms finetuning by 3.7%. We find a significant positive correlation between cross-lingual transfer performance on one side, and the similarity between high- and low-resource languages as well as the amount of low-resource pretraining data on the other side. A robustness analysis suggests that PARC has the potential to achieve even stronger performance with more powerful MPLMs.

pdf abs
Unsupervised Summarization Re-ranking
Mathieu Ravaut | Shafiq Joty | Nancy Chen

With the rise of task-specific pre-training objectives, abstractive summarization models like PEGASUS offer appealing zero-shot performance on downstream summarization tasks. However, the performance of such unsupervised models still lags significantly behind their supervised counterparts. Similarly to the supervised setup, we notice a very high variance in quality among summary candidates from these models while only one candidate is kept as the summary output. In this paper, we propose to re-rank summary candidates in an unsupervised manner, aiming to close the performance gap between unsupervised and supervised models. Our approach improves the unsupervised PEGASUS by up to 7.27% and ChatGPT by up to 6.86% relative mean ROUGE across four widely-adopted summarization benchmarks ; and achieves relative gains of 7.51% (up to 23.73% from XSum to WikiHow) averaged over 30 zero-shot transfer setups (finetuning on a dataset, evaluating on another).

Attribute-based generation methods are of growing significance in controlling the generation of large pre-trained language models (PLMs). Existing studies control the generation by (1) finetuning the model with attributes or (2) guiding the inference processing toward control signals while freezing the PLM. However, finetuning approaches infuse domain bias into generation, making it hard to generate out-of-domain texts. Besides, many methods guide the inference in its word-by-word generation, pushing the word probability to the target attributes, resulting in less fluent sentences. We argue that distilling controlling information from natural texts can produce fluent sentences while maintaining high controllability. In this paper, we propose GRAdient-guided Controllable rEtrieval (GRACE), a retrieval-augmented generation framework to facilitate the generation of fluent sentences with high attribute relevance. GRACE memorizes the semantic and attribute information from unlabeled corpora and applies a controllable retrieval to obtain desired information. For the generation, we design techniques to eliminate the domain bias from the retrieval results and integrate it into the generation model. Additionally, we propose a gradient-guided generation scheme that iteratively steers generation toward higher attribute relevance. Experimental results and quantities of examples verify the effectiveness of our method.

pdf abs
So many design choices: Improving and interpreting neural agent communication in signaling games
Timothée Bernard | Timothee Mickus

Emergent language games are experimental protocols designed to model how communication may arise among a group of agents. In this paper, we focus on how to improve performances of neural agents playing a signaling game: a sender is exposed to an image and generates a sequence of symbols that is transmitted to a receiver, which uses it to distinguish between two images, one that is semantically related to the original image, and one that is not. We consider multiple design choices, such as pretraining the visual components of the agents, introducing regularization terms, how to sample training items from the dataset, and we study how these different choices impact the behavior and performances of the agents. To that end, we introduce a number of automated metrics to measure the properties of the emergent language. We find that some implementation choices are always beneficial, and that the information that is conveyed by the agents’ messages is shaped not only by the game, but also by the overall design of the agents as well as seemingly unrelated implementation choices.

pdf abs
Constructing Word-Context-Coupled Space Aligned with Associative Knowledge Relations for Interpretable Language Modeling
Fanyu Wang | Zhenping Xie

As the foundation of current natural language processing methods, pre-trained language model has achieved excellent performance. However, the black-box structure of the deep neural network in pre-trained language models seriously limits the interpretability of the language modeling process. After revisiting the coupled requirement of deep neural representation and semantics logic of language modeling, a Word-Context-Coupled Space (W2CSpace) is proposed by introducing the alignment processing between uninterpretable neural representation and interpretable statistical logic. Moreover, a clustering process is also designed to connect the word- and context-level semantics. Specifically, an associative knowledge network (AKN), considered interpretable statistical logic, is introduced in the alignment process for word-level semantics. Furthermore, the context-relative distance is employed as the semantic feature for the downstream classifier, which is greatly different from the current uninterpretable semantic representations of pre-trained models. Our experiments for performance evaluation and interpretable analysis are executed on several types of datasets, including SIGHAN, Weibo, and ChnSenti. Wherein a novel evaluation strategy for the interpretability of machine learning models is first proposed. According to the experimental results, our language model can achieve better performance and highly credible interpretable ability compared to related state-of-the-art methods.

pdf abs
Fixed Input Parameterization for Efficient Prompting
Eunbi Choi | Yongrae Jo | Joel Jang | Joonwon Jang | Minjoon Seo

Recent works have shown that attaching prompts to the input is effective at conditioning Language Models (LM) to perform specific tasks. However, prompts are always included in the input text during inference, even when they are fixed, thus incurring substantial computational and memory overhead. Also, there is currently no straightforward method of utilizing prompts that are longer than the maximum input length of the LMs without incurring additional costs during inference. We formally define Fixed Input Parameterization (FIP) problem that focuses on injecting the fixed prompt into the parameters of an LM to be an efficient alternative to attaching fixed prompts to the input. We show that in scenarios with long fixed prompts, FIP can be up to 280 times more efficient in terms of total FLOPs than previous approaches. We further explore methodologies for FIP and show promising results in persona-dependent conversation, semantic parsing, and zero-shot learning with task instructions. Through these explorations, we show that FIP can be a promising direction for conditioning language models, in scenarios with long and fixed prompts.

pdf abs
Data Augmentation for Low-Resource Keyphrase Generation
Krishna Garg | Jishnu Ray Chowdhury | Cornelia Caragea

Keyphrase generation is the task of summarizing the contents of any given article into a few salient phrases (or keyphrases). Existing works for the task mostly rely on large-scale annotated datasets, which are not easy to acquire. Very few works address the problem of keyphrase generation in low-resource settings, but they still rely on a lot of additional unlabeled data for pretraining and on automatic methods for pseudo-annotations. In this paper, we present data augmentation strategies specifically to address keyphrase generation in purely resource-constrained domains. We design techniques that use the full text of the articles to improve both present and absent keyphrase generation. We test our approach comprehensively on three datasets and show that the data augmentation strategies consistently improve the state-of-the-art performance. We release our source code at https://github.com/kgarg8/kpgen-lowres-data-aug.

We present a large-scale video subtitle translation dataset, *BigVideo*, to facilitate the study of multi-modality machine translation. Compared with the widely used *How2* and *VaTeX* datasets, *BigVideo* is more than 10 times larger, consisting of 4.5 million sentence pairs and 9,981 hours of videos. We also introduce two deliberately designed test sets to verify the necessity of visual information: *Ambiguous* with the presence of ambiguous words, and *Unambiguous* in which the text context is self-contained for translation. To better model the common semantics shared across texts and videos, we introduce a contrastive learning method in the cross-modal encoder. Extensive experiments on the *BigVideo* shows that: a) Visual information consistently improves the NMT model in terms of BLEU, BLEURT and COMET on both Ambiguous and Unambiguous test sets. b) Visual information helps disambiguation, compared to the strong text baseline on terminology-targeted scores and human evaluation.

pdf abs
Constructing Procedural Graphs with Multiple Dependency Relations: A New Dataset and Baseline
Haopeng Ren | Yushi Zeng | Yi Cai | Bihan Zhou | Zetao Lian

Current structured and semi-structured knowledge bases mainly focus on representing descriptive knowledge but ignore another commonsense knowledge (Procedural Knowledge). To structure the procedural knowledge, existing methods are proposed to automatically generate flow graphs from procedural documents. They focus on extracting sequential dependency between sentences but neglect another two important dependencies (i.e., inclusion dependency and constraint dependency) in procedural documents. In our paper, we explore a problem of automatically generating procedural graph with multiple dependency relations to extend the flow graph constructed by existing methods and propose a procedural graph construction method with syntactic information and discourse structures. A new dataset (WHPG) is built and extensive experiments are conducted to evaluate the effectiveness of our proposed model.

Evaluation of natural language generation (NLG) is complex and multi-dimensional. Generated text can be evaluated for fluency, coherence, factuality, or any other dimensions of interest. Most frameworks that perform such multi-dimensional evaluation require training on large manually or synthetically generated datasets. In this paper, we study the efficacy of large language models as multi-dimensional evaluators using in-context learning, obviating the need for large training datasets. Our experiments show that in-context learning-based evaluators are competitive with learned evaluation frameworks for the task of text summarization, establishing state-of-the-art on dimensions such as relevance and factual consistency. We then analyze the effects of factors such as the selection and number of in-context examples on performance. Finally, we study the efficacy of in-context learning-based evaluators in evaluating zero-shot summaries written by large language models such as GPT-3.

pdf abs
Learning to Rank Utterances for Query-Focused Meeting Summarization
Xingxian Liu | Yajing Xu

Query-focused meeting summarization(QFMS) aims to generate a specific summary for the given query according to the meeting transcripts. Due to the conflict between long meetings and limited input size, previous works mainly adopt extract-then-summarize methods, which use extractors to simulate binary labels or ROUGE scores to extract utterances related to the query and then generate a summary. However, the previous approach fails to fully use the comparison between utterances. To the extractor, comparison orders are more important than specific scores. In this paper, we propose a Ranker-Generator framework. It learns to rank the utterances by comparing them in pairs and learning from the global orders, then uses top utterances as the generator’s input. We show that learning to rank utterances helps to select utterances related to the query effectively, and the summarizer can benefit from it. Experimental results on QMSum show that the proposed model outperforms all existing multi-stage models with fewer parameters.

pdf abs
Neural Architecture Search for Parameter-Efficient Fine-tuning of Large Pre-trained Language Models
Neal Lawton | Anoop Kumar | Govind Thattai | Aram Galstyan | Greg Ver Steeg

Parameter-efficient tuning (PET) methods fit pre-trained language models (PLMs) to downstream tasks by either computing a small compressed update for a subset of model parameters, or appending and fine-tuning a small number of new model parameters to the pre-trained network. Hand-designed PET architectures from the literature perform well in practice, but have the potential to be improved via automated neural architecture search (NAS). We propose an efficient NAS method for learning PET architectures via structured and unstructured pruning. We present experiments on GLUE demonstrating the effectiveness of our algorithm and discuss how PET architectural design choices affect performance in practice.

Large language models have demonstrated great potential to assist programmers in generating code. For such human-AI pair programming scenarios, we empirically demonstrate that while generated code are most often evaluated in terms of their functional correctness (i.e., whether generations pass available unit tests), correctness does not fully capture (e.g., may underestimate) the productivity gains these models may provide. Through a user study with N=49 experienced programmers, we show that while correctness captures high-value generations, programmers still rate code that fails unit tests as valuable if it reduces the overall effort needed to complete a coding task. Finally, we propose a hybrid metric that combines functional correctness and syntactic similarity and show that it achieves a 14% stronger correlation with value and can therefore better represent real-world gains when evaluating and comparing models.

pdf abs
Do transformer models do phonology like a linguist?
Saliha Muradoglu | Mans Hulden

Neural sequence-to-sequence models have been very successful at tasks in phonology and morphology that seemingly require a capacity for intricate linguistic generalisations. In this paper, we perform a detailed breakdown of the power of such models to capture various phonological generalisations and to benefit from exposure to one phonological rule to infer the behaviour of another similar rule. We present two types of experiments, one of which establishes the efficacy of the transformer model on 29 different processes. The second experiment type follows a priming and held-out case split where our model is exposed to two (or more) phenomena; one which is used as a primer to make the model aware of a linguistic category (e.g. voiceless stops) and a second one which contains a rule with a withheld case that the model is expected to infer (e.g. word-final devoicing with a missing training example such as b→p) results show that the transformer model can successfully model all 29 phonological phenomena considered, regardless of perceived process difficulty. We also show that the model can generalise linguistic categories and structures, such as vowels and syllables, through priming processes.

pdf abs
DiMS: Distilling Multiple Steps of Iterative Non-Autoregressive Transformers for Machine Translation
Sajad Norouzi | Rasa Hosseinzadeh | Felipe Perez | Maksims Volkovs

The computational benefits of iterative non-autoregressive transformers decrease as the number of decoding steps increases. As a remedy, we introduce Distill Multiple Steps (DiMS), a simple yet effective distillation technique to decrease the number of required steps to reach a certain translation quality. The distilled model enjoys the computational benefits of early iterations while preserving the enhancements from several iterative steps. DiMS relies on two models namely student and teacher. The student is optimized to predict the output of the teacher after multiple decoding steps while the teacher follows the student via a slow-moving average. The moving average keeps the teacher’s knowledge updated and enhances the quality of the labels provided by the teacher. During inference, the student is used for translation and no additional computation is added. We verify the effectiveness of DiMS on various models obtaining 7.8 and 12.9 BLEU points improvements in single-step translation accuracy on distilled and raw versions of WMT’14 De-En.Full code for this work is available here: https://github.com/layer6ai-labs/DiMS

Instructional videos make learning knowledge more efficient, by providing a detailed multimodal context of each procedure in instruction.A unique challenge posed by instructional videos is key-object degeneracy, where any single modality fails to sufficiently capture the key objects referred to in the procedure. For machine systems, such degeneracy can disturb the performance of a downstream task such as dense video captioning, leading to the generation of incorrect captions omitting key objects. To repair degeneracy, we propose a retrieval-based framework to augment the model representations in the presence of such key-object degeneracy. We validate the effectiveness and generalizability of our proposed framework over baselines using modalities with key-object degeneracy.

pdf abs
Bi-level Finetuning with Task-dependent Similarity Structure for Low-resource Training
Sai Ashish Somayajula | Lifeng Jin | Linfeng Song | Haitao Mi | Dong Yu

Training a large language model in low-resource settings is challenging since they are susceptible to overfitting with limited generalization abilities. Previous work addresses this issue by approaches such as tunable parameters reduction or data augmentation. However, they either limit the trained models’ expressiveness or rely on task-independent knowledge. In this paper, we propose the Bi-level Finetuning with Task-dependent Similarity Structure framework where all parameters, including the embeddings for unseen tokens, are finetuned with task-dependent information from the training data only. In this framework, a task-dependent similarity structure is learned in a data-driven fashion, which in turn is used to compose soft embeddings from conventional embeddings to be used in training to update all parameters. In order to learn the similarity structure and model parameters, we propose a bi-level optimization algorithm with two stages—search and finetune—to ensure successful learning. Results of experiments on several classification datasets in low-resource scenarios demonstrate that models trained with our method outperform strong baselines. Ablation experiments further support the effectiveness of different components in our framework. Code is available at https://github.com/Sai-Ashish/BFTSS.

pdf abs
Kanbun-LM: Reading and Translating Classical Chinese in Japanese Methods by Language Models
Hao Wang | Hirofumi Shimizu | Daisuke Kawahara

Recent studies in natural language processing (NLP) have focused on modern languages and achieved state-of-the-art results in many tasks. Meanwhile, little attention has been paid to ancient texts and related tasks. Classical Chinese first came to Japan approximately 2,000 years ago. It was gradually adapted to a Japanese form called Kanbun-Kundoku (Kanbun) in Japanese reading and translating methods, which has significantly impacted Japanese literature. However, compared to the rich resources of ancient texts in mainland China, Kanbun resources remain scarce in Japan.To solve this problem, we construct the first Classical-Chinese-to-Kanbun dataset in the world. Furthermore, we introduce two tasks, character reordering and machine translation, both of which play a significant role in Kanbun comprehension. We also test the current language models on these tasks and discuss the best evaluation method by comparing the results with human scores. We release our code and dataset on GitHub.

pdf abs
Adaptive Attention for Sparse-based Long-sequence Transformer
Xuanyu Zhang | Zhepeng Lv | Qing Yang

Recently, Transformers have been widely used in various fields and have achieved remarkable results. But it is still difficult for Transformer-based models to process longer sequences because self-attention in them scales quadratically with the sequence length. Although some models attempt to use sparse attention to reduce computational complexity, hand-crafted attention patterns are unable to select useful tokens adaptively according to the context. Thus, in this paper, we propose a novel efficient Transformer model with adaptive attention, A2-Former, for long sequence modeling. It can select useful tokens automatically in sparse attention by learnable position vectors, which consist of meta position and offset position vectors. Because the learnable offset position is not an integer vector, we utilize the interpolation technique to gather corresponding vectors from the input embedding matrix by discrete indexes. Experiments on Long Range Arena (LRA), a systematic and unified benchmark with different tasks, show that our model has achieved further improvement in performance compared with other sparse-based Transformers.

pdf abs
Sentiment Analysis using the Relationship between Users and Products
Natthawut Kertkeidkachorn | Kiyoaki Shirai

In product reviews, user and product aspects are useful in sentiment analysis. Nevertheless, previous studies mainly focus on modeling user and product aspects without considering the relationship between users and products. The relationship between users and products is typically helpful in estimating the bias of a user toward a product. In this paper, we, therefore, introduce the Graph Neural Network-based model with the pre-trained Language Model (GNNLM), where the relationship between users and products is incorporated. We conducted experiments on three well-known benchmarks for sentiment classification with the user and product information. The experimental results show that the relationship between users and products improves the performance of sentiment analysis. Furthermore, GNNLM achieves state-of-the-art results on yelp-2013 and yelp-2014 datasets.

pdf abs
Entropy-guided Vocabulary Augmentation of Multilingual Language Models for Low-resource Tasks
Arijit Nag | Bidisha Samanta | Animesh Mukherjee | Niloy Ganguly | Soumen Chakrabarti

Multilingual language models (MLLMs) like mBERTpromise to extend the benefits of NLP research to low-resource languages (LRLs). However, LRL words are under-represented in the wordpiece/subword vocabularies of MLLMs. This leads to many LRL words getting replaced by UNK, or concatenated from morphologically unrelated wordpieces, leading to low task accuracy. (Pre)-training MLLMs after including LRL documents is resource-intensive in terms of both human inputs and computational resources. In response, we propose EVALM (entropy-based vocabulary augmented language model), which uses a new task-cognizant measurement to detect the most vulnerable LRL words, whose wordpiece segmentations are undesirable. EVALM then provides reasonable initializations of their embeddings, followed by limited fine-tuning using the small LRL task corpus. Our experiments show significant performance improvements and also some surprising limits to such vocabulary augmentation strategies in various classification tasks for multiple diverse LRLs, as well as code-mixed texts. We will release the code and data to enable further research.

pdf abs
Class-Adaptive Self-Training for Relation Extraction with Incompletely Annotated Training Data
Qingyu Tan | Lu Xu | Lidong Bing | Hwee Tou Ng

Relation extraction (RE) aims to extract relations from sentences and documents. Existing relation extraction models typically rely on supervised machine learning. However, recent studies showed that many RE datasets are incompletely annotated. This is known as the false negative problem in which valid relations are falsely annotated as ‘no_relation’. Models trained with such data inevitably make similar mistakes during the inference stage. Self-training has been proven effective in alleviating the false negative problem. However, traditional self-training is vulnerable to confirmation bias and exhibits poor performance in minority classes. To overcome this limitation, we proposed a novel class-adaptive re-sampling self-training framework. Specifically, we re-sampled the pseudo-labels for each class by precision and recall scores. Our re-sampling strategy favored the pseudo-labels of classes with high precision and low recall, which improved the overall recall without significantly compromising precision. We conducted experiments on document-level and biomedical relation extraction datasets, and the results showed that our proposed self-training framework consistently outperforms existing competitive methods on the Re-DocRED and ChemDisgene datasets when the training data are incompletely annotated.

pdf abs
Solving Cosine Similarity Underestimation between High Frequency Words by ℓ₂ Norm Discounting
Saeth Wannasuphoprasit | Yi Zhou | Danushka Bollegala

Cosine similarity between two words, computed using their contextualised token embeddings obtained from masked language models (MLMs) such as BERT has shown to underestimate the actual similarity between those words CITATION.This similarity underestimation problem is particularly severe for high frequent words. Although this problem has been noted in prior work, no solution has been proposed thus far. We observe that the ℓ₂ norm of contextualised embeddings of a word correlates with its log-frequency in the pretraining corpus.Consequently, the larger ℓ₂ norms associated with the high frequent words reduce the cosine similarity values measured between them, thus underestimating the similarity scores.To solve this issue, we propose a method to discount the ℓ₂ norm of a contextualised word embedding by the frequency of that word in a corpus when measuring the cosine similarities between words.We show that the so called stop words behave differently from the rest of the words, which require special consideration during their discounting process.Experimental results on a contextualised word similarity dataset show that our proposed discounting method accurately solves the similarity underestimation problem.An anonymized version of the source code of our proposed method is submitted to the reviewing system.

Large language models (LLMs) have a wealth of knowledge that allows them to excel in various Natural Language Processing (NLP) tasks. Current research focuses on enhancing their performance within their existing knowledge. Despite their vast knowledge, LLMs are still limited by the amount of information they can accommodate and comprehend. Therefore, the ability to understand their own limitations on the unknows, referred to as self-knowledge, is of paramount importance. This study aims to evaluate LLMs’ self-knowledge by assessing their ability to identify unanswerable or unknowable questions. We introduce an automated methodology to detect uncertainty in the responses of these models, providing a novel measure of their self-knowledge. We further introduce a unique dataset, SelfAware, consisting of unanswerable questions from five diverse categories and their answerable counterparts. Our extensive analysis, involving 20 LLMs including GPT-3, InstructGPT, and LLaMA, discovering an intrinsic capacity for self-knowledge within these models. Moreover, we demonstrate that in-context learning and instruction tuning can further enhance this self-knowledge. Despite this promising insight, our findings also highlight a considerable gap between the capabilities of these models and human proficiency in recognizing the limits of their knowledge.

pdf abs
AltCLIP: Altering the Language Encoder in CLIP for Extended Language Capabilities
Zhongzhi Chen | Guang Liu | Bo-Wen Zhang | Qinghong Yang | Ledell Wu

CLIP (Contrastive Language–Image Pretraining) is an English multimodal representation model learned from a massive amount of English text-image pairs and has achieved great success in various downstream tasks, including image classification, text-to-image retrieval, and image generation. When extending CLIP to other languages, the major problem is the lack of good-quality text-image pairs. In this work, we present AltCLIP, a simple and low-resource method to build a strong multilingual multimodal representation model. Instead of training a model from scratch on multilingual text-image pairs, we take the original CLIP model trained on English text-image pairs and alter its text encoder with a pre-trained multilingual text encoder (XLM-R). We then align text and image representations by a two-stage training schema consisting of teacher learning and contrastive learning. Our method utilizes the existence of rich parallel text data and pre-trained multilingual language models. We present extensive experimental evaluations to demonstrate the effectiveness of our proposed method. Our model sets new state-of-the-art zero-shot performances on a wide range of tasks in multilingual multimodal benchmarks, including ImageNet-CN/IT/JA/KO serials, Flicker30k-CN, COCO-CN, Multi30k, and XTD. Further, our model outperforms the original CLIP model on zero-shot cross-modal retrieval, Image Classification in the Wild (ICinW) tasks, and CLIP Benchmark. We plan to open-source our code, pre-trained model weights, and evaluation toolkits of multilingual multimodal tasks, to facilitate research on multilingual multimodal representation learning.

Entity Alignment, which aims to identify equivalent entities from various Knowledge Graphs (KGs), is a fundamental and crucial task in knowledge graph fusion. Existing methods typically use triple or neighbor information to represent entities, and then align those entities using similarity matching. Most of them, however, fail to account for the heterogeneity among KGs and the distinction between KG entities and relations. To better solve these problems, we propose a Relation-gated Heterogeneous Graph Network (RHGN) for entity alignment. Specifically, RHGN contains a relation-gated convolutional layer to distinguish relations and entities in the KG. In addition, RHGN adopts a cross-graph embedding exchange module and a soft relation alignment module to address the neighbor heterogeneity and relation heterogeneity between different KGs, respectively. Extensive experiments on four benchmark datasets demonstrate that RHGN is superior to existing state-of-the-art entity alignment methods.

pdf abs
Feature Interactions Reveal Linguistic Structure in Language Models
Jaap Jumelet | Willem Zuidema

We study feature interactions in the context of feature attribution methods for post-hoc interpretability. In interpretability research, getting to grips with feature interactions is increasingly recognised as an important challenge, because interacting features are key to the success of neural networks. Feature interactions allow a model to build up hierarchical representations for its input, and might provide an ideal starting point for the investigation into linguistic structure in language models. However, uncovering the exact role that these interactions play is also difficult, and a diverse range of interaction attribution methods has been proposed. In this paper, we focus on the question which of these methods most faithfully reflects the inner workings of the target models. We work out a grey box methodology, in which we train models to perfection on a formal language classification task, using PCFGs. We show that under specific configurations, some methods are indeed able to uncover the grammatical rules acquired by a model. Based on these findings we extend our evaluation to a case study on language models, providing novel insights into the linguistic structure that these models have acquired.

pdf abs
Clustering-Aware Negative Sampling for Unsupervised Sentence Representation
Jinghao Deng | Fanqi Wan | Tao Yang | Xiaojun Quan | Rui Wang

Contrastive learning has been widely studied in sentence representation learning. However, earlier works mainly focus on the construction of positive examples, while in-batch samples are often simply treated as negative examples. This approach overlooks the importance of selecting appropriate negative examples, potentially leading to a scarcity of hard negatives and the inclusion of false negatives. To address these issues, we propose ClusterNS (Clustering-aware Negative Sampling), a novel method that incorporates cluster information into contrastive learning for unsupervised sentence representation learning. We apply a modified K-means clustering algorithm to supply hard negatives and recognize in-batch false negatives during training, aiming to solve the two issues in one unified framework. Experiments on semantic textual similarity (STS) tasks demonstrate that our proposed ClusterNS compares favorably with baselines in unsupervised sentence representation learning. Our code has been made publicly available at github.com/djz233/ClusterNS.

pdf abs
An Effective Deployment of Contrastive Learning in Multi-label Text Classification
Nankai Lin | Guanqiu Qin | Gang Wang | Dong Zhou | Aimin Yang

The effectiveness of contrastive learning technology in natural language processing tasks is yet to be explored and analyzed. How to construct positive and negative samples correctly and reasonably is the core challenge of contrastive learning. It is even harder to discover contrastive objects in multi-label text classification tasks. There are very few contrastive losses proposed previously. In this paper, we investigate the problem from a different angle by proposing five novel contrastive losses for multi-label text classification tasks. These are Strict Contrastive Loss (SCL), Intra-label Contrastive Loss (ICL), Jaccard Similarity Contrastive Loss (JSCL), Jaccard Similarity Probability Contrastive Loss (JSPCL), and Stepwise Label Contrastive Loss (SLCL). We explore the effectiveness of contrastive learning for multi-label text classification tasks by the employment of these novel losses and provide a set of baseline models for deploying contrastive learning techniques on specific tasks. We further perform an interpretable analysis of our approach to show how different components of contrastive learning losses play their roles. The experimental results show that our proposed contrastive losses can bring improvement to multi-label text classification tasks. Our work also explores how contrastive learning should be adapted for multi-label text classification tasks.

Knowledge-based referring expression comprehension (KB-REC) aims to identify visual objects referred to by expressions that incorporate knowledge. Existing methods employ sentence-level retrieval and fusion methods, which may lead to issues of similarity bias and interference from irrelevant information in unstructured knowledge sentences. To address these limitations, we propose a segment-level and category-oriented network (SLCO). Our approach includes a segment-level and prompt-based knowledge retrieval method to mitigate the similarity bias problem and a category-based grounding method to alleviate interference from irrelevant information in knowledge sentences. Experimental results show that our SLCO can eliminate interference and improve the overall performance of the KB-REC task.

pdf abs
MVP: Multi-task Supervised Pre-training for Natural Language Generation
Tianyi Tang | Junyi Li | Wayne Xin Zhao | Ji-Rong Wen

Pre-trained language models (PLMs) have achieved remarkable success in natural language generation (NLG) tasks. Up to now, most NLG-oriented PLMs are pre-trained in an unsupervised manner using the large-scale general corpus. In the meanwhile, an increasing number of models pre-trained with labeled data (i.e. “supervised pre-training”) showcase superior performance compared to unsupervised pre-trained models. Motivated by the success of supervised pre-training, we propose Multi-task superVised Pre-training (MVP) for natural language generation. We collect a large-scale natural language generation corpus, MVPCorpus, from 77 datasets over 11 diverse NLG tasks. Then we unify these examples into a general text-to-text format to pre-train the text generation model MVP in a supervised manner. For each task, we further pre-train specific soft prompts to stimulate the model’s capacity to perform a specific task. Our MVP model can be seen as a practice that utilizes recent instruction tuning on relatively small PLMs. Extensive experiments have demonstrated the effectiveness and generality of our MVP model in a number of NLG tasks, which achieves state-of-the-art performance on 13 out of 17 datasets, outperforming BART by 9.3% and Flan-T5 by 5.8%.

Entity Alignment (EA) aims to find the equivalent entities between two Knowledge Graphs (KGs). Existing methods usually encode the triples of entities as embeddings and learn to align the embeddings, which prevents the direct interaction between the original information of the cross-KG entities. Moreover, they encode the relational triples and attribute triples of an entity in heterogeneous embedding spaces, which prevents them from helping each other. In this paper, we transform both triples into unified textual sequences, and model the EA task as a bi-directional textual entailment task between the sequences of cross-KG entities. Specifically, we feed the sequences of two entities simultaneously into a pre-trained language model (PLM) and propose two kinds of PLM-based entity aligners that model the entailment probability between sequences as the similarity between entities. Our approach captures the unified correlation pattern of two kinds of information between entities, and explicitly models the fine-grained interaction between original entity information. The experiments on five cross-lingual EA datasets show that our approach outperforms the state-of-the-art EA methods and enables the mutual enhancement of the heterogeneous information. Codes are available at https://github.com/OreOZhao/TEA.

pdf abs
It is a Bird Therefore it is a Robin: On BERT’s Internal Consistency Between Hypernym Knowledge and Logical Words
Nicolas Guerin | Emmanuel Chemla

The lexical knowledge of NLP systems shouldbe tested (i) for their internal consistency(avoiding groundedness issues) and (ii) bothfor content words and logical words. In thispaper we propose a new method to test the understandingof the hypernymy relationship bymeasuring its antisymmetry according to themodels. Previous studies often rely only on thedirect question (e.g., A robin is a ...), where weargue a correct answer could only rely on collocationalcues, rather than hierarchical cues. We show how to control for this, and how it isimportant. We develop a method to ask similarquestions about logical words that encode anentailment-like relation (e.g., because or therefore).Our results show important weaknessesof BERT-like models on these semantic tasks.

pdf abs
Defending against Insertion-based Textual Backdoor Attacks via Attribution
Jiazhao Li | Zhuofeng Wu | Wei Ping | Chaowei Xiao | V.G.Vinod Vydiswaran

Textual backdoor attack, as a novel attack model, has been shown to be effective in adding a backdoor to the model during training. Defending against such backdoor attacks has become urgent and important. In this paper, we propose AttDef, an efficient attribution-based pipeline to defend against two insertion-based poisoning attacks, BadNL and InSent. Specifically, we regard the tokens with larger attribution scores as potential triggers since larger attribution words contribute more to the false prediction results and therefore are more likely to be poison triggers. Additionally, we further utilize an external pre-trained language model to distinguish whether input is poisoned or not. We show that our proposed method can generalize sufficiently well in two common attack scenarios (poisoning training data and testing data), which consistently improves previous methods. For instance, AttDef can successfully mitigate both attacks with an average accuracy of 79.97% (56.59% up) and 48.34% (3.99% up) under pre-training and post-training attack defense respectively, achieving the new state-of-the-art performance on prediction recovery over four benchmark datasets.

pdf abs
ActiveAED: A Human in the Loop Improves Annotation Error Detection
Leon Weber | Barbara Plank

Manually annotated datasets are crucial for training and evaluating Natural Language Processing models. However, recent work has discovered that even widely-used benchmark datasets contain a substantial number of erroneous annotations. This problem has been addressed with Annotation Error Detection (AED) models, which can flag such errors for human re-annotation. However, even though many of these AED methods assume a final curation step in which a human annotator decides whether the annotation is erroneous, they have been developed as static models without any human-in-the-loop component. In this work, we propose ActiveAED, an AED method that can detect errors more accurately by repeatedly querying a human for error corrections in its prediction loop. We evaluate ActiveAED on eight datasets spanning five different tasks and find that it leads to improvements over the state of the art on seven of them, with gains of up to six percentage points in average precision.

pdf abs
Assessing Word Importance Using Models Trained for Semantic Tasks
Dávid Javorský | Ondřej Bojar | François Yvon

Many NLP tasks require to automatically identify the most significant words in a text. In this work, we derive word significance from models trained to solve semantic task: Natural Language Inference and Paraphrase Identification. Using an attribution method aimed to explain the predictions of these models, we derive importance scores for each input token. We evaluate their relevance using a so-called cross-task evaluation: Analyzing the performance of one model on an input masked according to the other model’s weight, we show that our method is robust with respect to the choice of the initial task. Additionally, we investigate the scores from the syntax point of view and observe interesting patterns, e.g. words closer to the root of a syntactic tree receive higher importance scores. Altogether, these observations suggest that our method can be used to identify important words in sentences without any explicit word importance labeling in training.

pdf abs
In-context Examples Selection for Machine Translation
Sweta Agrawal | Chunting Zhou | Mike Lewis | Luke Zettlemoyer | Marjan Ghazvininejad

Large-scale generative models show an impressive ability to perform a wide range of Natural Language Processing (NLP) tasks using in-context learning, where a few examples are used to describe a task to the model. For Machine Translation (MT), these examples are typically randomly sampled from the development dataset with a similar distribution as the evaluation set. However, it is unclear how the choice of these in context examples and their ordering impacts the output translation quality. In this work, we aim to understand the properties of good in-context examples for MT in both in-domain and out-of-domain settings. We show that the translation quality and the domain of the in-context examples matter and that 1-shot noisy unrelated examples can have a catastrophic impact on output quality. While concatenating multiple random examples reduces the effect of noise, a single good prompt optimized to maximize translation quality on the development dataset can elicit learned information from the pre-trained language model. Adding similar examples based on an n-gram overlap with the test source significantly and consistently improves the translation quality of the outputs, outperforming a strong kNN-MT baseline in 2 out of 4 out-of-domain datasets.

pdf abs
PropSegmEnt: A Large-Scale Corpus for Proposition-Level Segmentation and Entailment Recognition
Sihao Chen | Senaka Buthpitiya | Alex Fabrikant | Dan Roth | Tal Schuster

The widely studied task of Natural Language Inference (NLI) requires a system to recognize whether one piece of text is textually entailed by another, i.e. whether the entirety of its meaning can be inferred from the other. In current NLI datasets and models, textual entailment relations are typically defined on the sentence- or paragraph-level. However, even a simple sentence often contains multiple propositions, i.e. distinct units of meaning conveyed by the sentence. As these propositions can carry different truth values in the context of a given premise, we argue for the need to recognize the textual entailment relation of each proposition in a sentence individually. We propose PropSegmEnt, a corpus of over 45K propositions annotated by expert human raters. Our dataset structure resembles the tasks of (1) segmenting sentences within a document to the set of propositions, and (2) classifying the entailment relation of each proposition with respect to a different yet topically-aligned document, i.e. documents describing the same event or entity. We establish strong baselines for the segmentation and entailment tasks. Through case studies on summary hallucination detection and document-level NLI, we demonstrate that our conceptual framework is potentially useful for understanding and explaining the compositionality of NLI labels.

Speech or text representation generated by pre-trained models contains modal-specific information that could be combined for benefiting spoken language understanding (SLU) tasks. In this work, we propose a novel pre-training paradigm termed Continuous Integrate-and-Fire Pre-Training (CIF-PT). It relies on a simple but effective frame-to-token alignment: continuous integrate-and-fire (CIF) to bridge the representations between speech and text. It jointly performs speech-to-text training and language model distillation through CIF as the pre-training (PT). Evaluated on SLU benchmark SLURP dataset, CIF-PT outperforms the state-of-the-art model by 1.94% of accuracy and 2.71% of SLU-F1 on the tasks of intent classification and slot filling, respectively. We also observe the cross-modal representation extracted by CIF-PT obtains better performance than other neural interfaces for the tasks of SLU, including the dominant speech representation learned from self-supervised pre-training.

pdf abs
Improving Diachronic Word Sense Induction with a Nonparametric Bayesian method
Ashjan Alsulaimani | Erwan Moreau

Diachronic Word Sense Induction (DWSI) is the task of inducing the temporal representations of a word meaning from the context, as a set of senses and their prevalence over time. We introduce two new models for DWSI, based on topic modelling techniques: one is based on Hierarchical Dirichlet Processes (HDP), a nonparametric model; the other is based on the Dynamic Embedded Topic Model (DETM), a recent dynamic neural model. We evaluate these models against two state of the art DWSI models, using a time-stamped labelled dataset from the biomedical domain. We demonstrate that the two proposed models perform better than the state of the art. In particular, the HDP-based model drastically outperforms all the other models, including the dynamic neural model.

pdf abs
What to Fuse and How to Fuse: Exploring Emotion and Personality Fusion Strategies for Explainable Mental Disorder Detection
Sourabh Zanwar | Xiaofei Li | Daniel Wiechmann | Yu Qiao | Elma Kerz

Mental health disorders (MHD) are increasingly prevalent worldwide and constitute one of the greatest challenges facing our healthcare systems and modern societies in general. In response to this societal challenge, there has been a surge in digital mental health research geared towards the development of new techniques for unobtrusive and efficient automatic detection of MHD. Within this area of research, natural language processing techniques are playing an increasingly important role, showing promising detection results from a variety of textual data. Recently, there has been a growing interest in improving mental illness detection from textual data by way of leveraging emotions: ‘Emotion fusion’ refers to the process of integrating emotion information with general textual information to obtain enhanced information for decision-making. However, while the available research has shown that MHD prediction can be improved through a variety of different fusion strategies, previous works have been confined to a particular fusion strategy applied to a specific dataset, and so is limited by the lack of meaningful comparability. In this work, we integrate and extend this research by conducting extensive experiments with three types of deep learning-based fusion strategies: (i) feature-level fusion, where a pre-trained masked language model for mental health detection (MentalRoBERTa) was infused with a comprehensive set of engineered features, (ii) model fusion, where the MentalRoBERTa model was infused with hidden representations of other language models and (iii) task fusion, where a multi-task framework was leveraged to learn the features for auxiliary tasks. In addition to exploring the role of different fusion strategies, we expand on previous work by broadening the information infusion to include a second domain related to mental health, namely personality. We evaluate algorithm performance on data from two benchmark datasets, encompassing five mental health conditions: attention deficit hyperactivity disorder, anxiety, bipolar disorder, depression and psychological stress.

In this paper, we propose a new knowledge distillation approach called adaptive contrastive knowledge distillation (ACKD) for BERT compression. Different from existing knowledge distillation methods for BERT that implicitly learn discriminative student features by mimicking the teacher features, we first introduce a novel contrastive distillation loss (CDL) based on hidden state features in BERT as the explicit supervision to learn discriminative student features. We further observe sentences with similar features may have completely different meanings, which makes them hard to distinguish. Existing methods do not pay sufficient attention to these hard samples with less discriminative features. Therefore, we propose a new strategy called sample adaptive reweighting (SAR) to adaptively pay more attention to these hard samples and strengthen their discrimination abilities. We incorporate our SAR strategy into our CDL and form the adaptive contrastive distillation loss, based on which we construct our ACKD framework. Comprehensive experiments on multiple natural language processing tasks demonstrate the effectiveness of our ACKD framework.

The transformer model is known to be computationally demanding, and prohibitively costly for long sequences, as the self-attention module uses a quadratic time and space complexity with respect to sequence length. Many researchers have focused on designing new forms of self-attention or introducing new parameters to overcome this limitation, however a large portion of them prohibits the model to inherit weights from large pretrained models. In this work, the transformer’s inefficiency has been taken care of from another perspective. We propose Fourier Transformer, a simple yet effective approach by progressively removing redundancies in hidden sequence using the ready-made Fast Fourier Transform (FFT) operator to perform Discrete Cosine Transformation (DCT). Fourier Transformer is able to significantly reduce computational costs while retain the ability to inherit from various large pretrained models. Experiments show that our model achieves state-of-the-art performances among all transformer-based models on the long-range modeling benchmark LRA with significant improvement in both speed and space. For generative seq-to-seq tasks including CNN/DailyMail and ELI5, by inheriting the BART weights our model outperforms the standard BART and other efficient models. Our code will be publicly available at https://github.com/LUMIA-Group/FourierTransformer

pdf abs
Zero-Shot Classification by Logical Reasoning on Natural Language Explanations
Chi Han | Hengzhi Pei | Xinya Du | Heng Ji

Humans can classify data of an unseen category by reasoning on its language explanations. This ability is owing to the compositional nature of language: we can combine previously seen attributes to describe the new category. For example, we might describe a sage thrasher as “it has a slim straight relatively short bill, yellow eyes and a long tail”, so that others can use their knowledge of attributes “slim straight relatively short bill”, “yellow eyes” and “long tail” to recognize a sage thrasher. Inspired by this observation, in this work we tackle zero-shot classification task by logically parsing and reasoning on natural language explanations. To this end, we propose the framework CLORE (Classification by LOgical Reasoning on Explanations). While previous methods usually regard textual information as implicit features, CLORE parses explanations into logical structures and then explicitly reasons along this structure on the input to produce a classification score. Experimental results on explanation-based zero-shot classification benchmarks demonstrate that CLORE is superior to baselines, which we show is mainly due to higher scores on tasks requiring more logical reasoning. We also demonstrate that our framework can be extended to zero-shot classification on visual modality. Alongside classification decisions, CLORE can provide the logical parsing and reasoning process as a clear form of rationale. Through empirical analysis we demonstrate that CLORE is also less affected by linguistic biases than baselines.

Multi-Modal Relation Extraction (MMRE) aims at identifying the relation between two entities in texts that contain visual clues. Rich visual content is valuable for the MMRE task, but existing works cannot well model finer associations among different modalities, failing to capture the truly helpful visual information and thus limiting relation extraction performance. In this paper, we propose a novel MMRE framework to better capture the deeper correlations of text, entity pair, and image/objects, so as to mine more helpful information for the task, termed as DGF-PT. We first propose a prompt-based autoregressive encoder, which builds the associations of intra-modal and inter-modal features related to the task, respectively by entity-oriented and object-oriented prefixes. To better integrate helpful visual information, we design a dual-gated fusion module to distinguish the importance of image/objects and further enrich text representations. In addition, a generative decoder is introduced with entity type restriction on relations, better filtering out candidates. Extensive experiments conducted on the benchmark dataset show that our approach achieves excellent performance compared to strong competitors, even in the few-shot situation.

pdf abs
Pruning Pre-trained Language Models with Principled Importance and Self-regularization
Siyu Ren | Kenny Zhu

Iterative pruning is one of the most effective compression methods for pre-trained language models. We discovered that finding the optimal pruning decision is an equality-constrained 0-1 Integer Linear Programming problem. The solution to this optimization problem leads to a principled importance criterion which we use to rank parameters during iterative model pruning. To mitigate the poor generalization at high sparsity levels, we propose a self-regularization scheme where model prediction is regularized by the latest checkpoint with increasing sparsity throughout pruning. Our experiments on natural language understanding, question answering, named entity recognition, and data-to-text generation with various Transformer-based PLMs show the effectiveness of the approach at various sparsity levels.

pdf abs
The Magic of IF: Investigating Causal Reasoning Abilities in Large Language Models of Code
Xiao Liu | Da Yin | Chen Zhang | Yansong Feng | Dongyan Zhao

Causal reasoning, the ability to identify cause-and-effect relationship, is crucial in human thinking. Although large language models (LLMs) succeed in many NLP tasks, it is still challenging for them to conduct complex causal reasoning like abductive reasoning and counterfactual reasoning. Given the fact that programming code may express causal relations more often and explicitly with conditional statements like “if“, we want to explore whether Code-LLMs acquire better causal reasoning abilities. Our experiments show that compared to text-only LLMs, Code-LLMs with code prompts are better causal reasoners. We further intervene on the prompts from different aspects, and discover that the key point is the programming structure. Code and data are available at https://github.com/xxxiaol/magic-if.

pdf abs
Learning to Leverage High-Order Medical Knowledge Graph for Joint Entity and Relation Extraction
Zhe Yang | Yi Huang | Junlan Feng

Automatic medical entity and relation extraction is essential for daily electronic medical record (EMR) analysis, and has attracted a lot of academic attention. Tremendous progress has been made in recent years. However, medical terms are difficult to understand, and their relations are more complicated than general ones. Based on this situation, domain knowledge gives better background and contexts for medical terms. Despite the benefits of medical domain knowledge, the utilization way of it for joint entity and relation extraction is inadequate. To foster this line of research, in this work, we propose to leverage the medical knowledge graph for extracting entities and relations for Chinese Medical Texts in a collective way. Specifically, we propose to construct a high-order heterogeneous graph based on medical knowledge graph, which is linked to the entity mentions in the text. In this way, neighbors from the high-order heterogeneous graph can pass the message to each other for better global context representations. Our experiments on real Chinese Medical Texts show that our method is more effective than state-of-the-art methods.

pdf abs
Data-Efficient Finetuning Using Cross-Task Nearest Neighbors
Hamish Ivison | Noah A. Smith | Hannaneh Hajishirzi | Pradeep Dasigi

Obtaining labeled data to train a model for a task of interest is often expensive. Prior work shows training models on multitask data augmented with task descriptions (prompts) effectively transfers knowledge to new tasks. Towards efficiently building task-specific models, we assume access to a small number (32-1000) of unlabeled target-task examples and use those to retrieve the most similar labeled examples from a large pool of multitask data augmented with prompts. Compared to the current practice of finetuning models on uniformly sampled prompted multitask data (e.g.: FLAN, T0), our approach of finetuning on cross-task nearest neighbors is significantly more data-efficient. Using only 2% of the data from the P3 pool without any labeled target-task data, our models outperform strong baselines trained on all available data by 3-30% on 12 out of 14 datasets representing held-out tasks including legal and scientific document QA. Similarly, models trained on cross-task nearest neighbors from SuperNaturalInstructions, representing about 5% of the pool, obtain comparable performance to state-of-the-art models on 12 held-out tasks from that pool. Moreover, the models produced by our approach also provide a better initialization than single multitask finetuned models for few-shot finetuning on target-task data, as shown by a 2-23% relative improvement over few-shot finetuned T0-3B models on 8 datasets.

Collecting labeled data for Named Entity Recognition (NER) tasks is challenging due to the high cost of manual annotations. Instead, researchers have proposed few-shot self-training and rule-augmentation techniques to minimize the reliance on large datasets. However, inductive biases and restricted logical language lexicon, respectively, can limit the ability of these models to perform well. In this work, we propose CoAug, a co-augmentation framework that allows us to improve few-shot models and rule-augmentation models by bootstrapping predictions from each model. By leveraging rules and neural model predictions to train our models, we complement the benefits of each and achieve the best of both worlds. In our experiments, we show that our best CoAug model can outperform strong weak-supervision-based NER models at least by 6.5 F1 points.

Data augmentation techniques have been used to alleviate the problem of scarce labeled data in various NER tasks (flat, nested, and discontinuous NER tasks). Existing augmentation techniques either manipulate the words in the original text that break the semantic coherence of the text, or exploit generative models that ignore preserving entities in the original text, which impedes the use of augmentation techniques on nested and discontinuous NER tasks. In this work, we propose a novel Entity-to-Text based data augmentation technique named EnTDA to add, delete, replace or swap entities in the entity list of the original texts, and adopt these augmented entity lists to generate semantically coherent and entity preserving texts for various NER tasks. Furthermore, we introduce a diversity beam search to increase the diversity during the text generation process. Experiments on thirteen NER datasets across three tasks (flat, nested, and discontinuous NER tasks) and two settings (full data and low resource settings) show that EnTDA could bring more performance improvements compared to the baseline augmentation techniques.

pdf abs
World Models for Math Story Problems
Andreas Opedal | Niklas Stoehr | Abulhair Saparov | Mrinmaya Sachan

Solving math story problems is a complex task for students and NLP models alike, requiring them to understand the world as described in the story and reason over it to compute an answer. Recent years have seen impressive performance on automatically solving these problems with large pre-trained language models and innovative techniques to prompt them. However, it remains unclear if these models possess accurate representations of mathematical concepts. This leads to lack of interpretability and trustworthiness which impedes their usefulness in various applications. In this paper, we consolidate previous work on categorizing and representing math story problems and develop MathWorld, which is a graph-based semantic formalism specific for the domain of math story problems. With MathWorld, we can assign world models to math story problems which represent the situations and actions introduced in the text and their mathematical relationships. We combine math story problems from several existing datasets and annotate a corpus of 1,019 problems and 3,204 logical forms with MathWorld. Using this data, we demonstrate the following use cases of MathWorld: (1) prompting language models with synthetically generated question-answer pairs to probe their reasoning and world modeling abilities, and (2) generating new problems by using the world models as a design space.

Mixture-of-Expert (MoE) models have obtained state-of-the-art performance in Neural Machine Translation (NMT) tasks. Existing works in MoE mostly consider a homogeneous design where the same number of experts of the same size are placed uniformly throughout the network. Furthermore, existing MoE works do not consider computational constraints (e.g., FLOPs, latency) to guide their design. To this end, we develop AutoMoE – a framework for designing heterogeneous MoE’s under computational constraints. AutoMoE leverages Neural Architecture Search (NAS) to obtain efficient sparse MoE sub-transformers with 4x inference speedup (CPU) and FLOPs reduction over manually designed Transformers, with parity in BLEU score over dense Transformer and within 1 BLEU point of MoE SwitchTransformer, on aggregate over benchmark datasets for NMT.Heterogeneous search space with dense and sparsely activated Transformer modules (e.g., how many experts? where to place them? what should be their sizes?) allows for adaptive compute – where different amounts of computations are used for different tokens in the input. Adaptivity comes naturally from routing decisions which send tokens to experts of different sizes. AutoMoE code, data, and trained models are available at https://aka.ms/AutoMoE.

Multilingual information retrieval (IR) is challenging since annotated training data is costly to obtain in many languages. We present an effective method to train multilingual IR systems when only English IR training data and some parallel corpora between English and other languages are available. We leverage parallel and non-parallel corpora to improve the pretrained multilingual language models’ cross-lingual transfer ability. We design a semantic contrastive loss to align representations of parallel sentences that share the same semantics in different languages, and a new language contrastive loss to leverage parallel sentence pairs to remove language-specific information in sentence representations from non-parallel corpora. When trained on English IR data with these losses and evaluated zero-shot on non-English data, our model demonstrates significant improvement to prior work on retrieval performance, while it requires much less computational effort. We also demonstrate the value of our model for a practical setting when a parallel corpus is only available for a few languages, but a lack of parallel corpora resources persists for many other low-resource languages. Our model can work well even with a small number of parallel sentences, and be used as an add-on module to any backbones and other tasks.

pdf abs
Easy to Decide, Hard to Agree: Reducing Disagreements Between Saliency Methods
Josip Jukić | Martin Tutek | Jan Snajder

A popular approach to unveiling the black box of neural NLP models is to leverage saliency methods, which assign scalar importance scores to each input component. A common practice for evaluating whether an interpretability method is faithful has been to use evaluation-by-agreement – if multiple methods agree on an explanation, its credibility increases. However, recent work has found that saliency methods exhibit weak rank correlations even when applied to the same model instance and advocated for alternative diagnostic methods. In our work, we demonstrate that rank correlation is not a good fit for evaluating agreement and argue that Pearson-r is a better-suited alternative. We further show that regularization techniques that increase faithfulness of attention explanations also increase agreement between saliency methods. By connecting our findings to instance categories based on training dynamics, we show that the agreement of saliency method explanations is very low for easy-to-learn instances. Finally, we connect the improvement in agreement across instance categories to local representation space statistics of instances, paving the way for work on analyzing which intrinsic model properties improve their predisposition to interpretability methods.

pdf abs
Enhancing Cross-lingual Transfer via Phonemic Transcription Integration
Hoang Nguyen | Chenwei Zhang | Tao Zhang | Eugene Rohrbaugh | Philip Yu

Previous cross-lingual transfer methods are restricted to orthographic representation learning via textual scripts. This limitation hampers cross-lingual transfer and is biased towards languages sharing similar well-known scripts. To alleviate the gap between languages from different writing scripts, we propose PhoneXL, a framework incorporating phonemic transcriptions as an additional linguistic modality beyond the traditional orthographic transcriptions for cross-lingual transfer. Particularly, we propose unsupervised alignment objectives to capture (1) local one-to-one alignment between the two different modalities, (2) alignment via multi-modality contexts to leverage information from additional modalities, and (3) alignment via multilingual contexts where additional bilingual dictionaries are incorporated. We also release the first phonemic-orthographic alignment dataset on two token-level tasks (Named Entity Recognition and Part-of-Speech Tagging) among the understudied but interconnected Chinese-Japanese-Korean-Vietnamese (CJKV) languages. Our pilot study reveals phonemic transcription provides essential information beyond the orthography to enhance cross-lingual transfer and bridge the gap among CJKV languages, leading to consistent improvements on cross-lingual token-level tasks over orthographic-based multilingual PLMs.

pdf abs
Human-in-the-loop Abstractive Dialogue Summarization
Jiaao Chen | Mohan Dodda | Diyi Yang

Abstractive dialogue summarization has received increasing attention recently. Despite the fact that most of the current dialogue summarization systems are trained to maximize the likelihood of human-written summaries and have achieved significant results, there is still a huge gap in generating high-quality summaries as determined by humans, such as coherence and faithfulness, partly due to the misalignment in maximizing a single human-written summary. To this end, we propose to incorporate different levels of human feedback into the training process. This will enable us to guide the models to capture the behaviors humans care about for summaries. Specifically, we ask humans to highlight the salient information to be included in summaries to provide the local feedback, and to make overall comparisons among summaries in terms of coherence, accuracy, coverage, concise and overall quality, as the global feedback. We then combine both local and global feedback to fine-tune the dialog summarization policy with Reinforcement Learning. Experiments conducted on multiple datasets demonstrate the effectiveness and generalization of our methods over the state-of-the-art supervised baselines, especially in terms of human judgments.

Quality Estimation (QE) is the task of evaluating machine translation output in the absence of reference translation. Conventional approaches to QE involve training separate models at different levels of granularity viz., word-level, sentence-level, and document-level, which sometimes lead to inconsistent predictions for the same input. To overcome this limitation, we focus on jointly training a single model for sentence-level and word-level QE tasks in a multi-task learning framework. Using two multi-task learning-based QE approaches, we show that multi-task learning improves the performance of both tasks. We evaluate these approaches by performing experiments in different settings, viz., single-pair, multi-pair, and zero-shot. We compare the multi-task learning-based approach with baseline QE models trained on single tasks and observe an improvement of up to 4.28% in Pearson’s correlation (r) at sentence-level and 8.46% in F1-score at word-level, in the single-pair setting. In the multi-pair setting, we observe improvements of up to 3.04% at sentence-level and 13.74% at word-level; while in the zero-shot setting, we also observe improvements of up to 5.26% and 3.05%, respectively. We make the models proposed in this paper publically available.

Event extraction (EE) is a crucial task aiming at extracting events from texts, which includes two subtasks: event detection (ED) and event argument extraction (EAE). In this paper, we check the reliability of EE evaluations and identify three major pitfalls: (1) The data preprocessing discrepancy makes the evaluation results on the same dataset not directly comparable, but the data preprocessing details are not widely noted and specified in papers. (2) The output space discrepancy of different model paradigms makes different-paradigm EE models lack grounds for comparison and also leads to unclear mapping issues between predictions and annotations. (3) The absence of pipeline evaluation of many EAE-only works makes them hard to be directly compared with EE works and may not well reflect the model performance in real-world pipeline scenarios. We demonstrate the significant influence of these pitfalls through comprehensive meta-analyses of recent papers and empirical experiments. To avoid these pitfalls, we suggest a series of remedies, including specifying data preprocessing, standardizing outputs, and providing pipeline evaluation results. To help implement these remedies, we develop a consistent evaluation framework OmniEvent, which can be obtained from https://github.com/THU-KEG/OmniEvent.

pdf abs
Yes, this Way! Learning to Ground Referring Expressions into Actions with Intra-episodic Feedback from Supportive Teachers
Philipp Sadler | Sherzod Hakimov | David Schlangen

The ability to pick up on language signals in an ongoing interaction is crucial for future machine learning models to collaborate and interact with humans naturally. In this paper, we present an initial study that evaluates intra-episodic feedback given in a collaborative setting. We use a referential language game as a controllable example of a task-oriented collaborative joint activity. A teacher utters a referring expression generated by a well-known symbolic algorithm (the “Incremental Algorithm”) as an initial instruction and then monitors the follower’s actions to possibly intervene with intra-episodic feedback (which does not explicitly have to be requested). We frame this task as a reinforcement learning problem with sparse rewards and learn a follower policy for a heuristic teacher. Our results show that intra-episodic feedback allows the follower to generalize on aspects of scene complexity and performs better than providing only the initial statement.

pdf abs
Investigating Transformer-Guided Chaining for Interpretable Natural Logic Reasoning
Kanagasabai Rajaraman | Saravanan Rajamanickam | Wei Shi

Natural logic reasoning has received increasing attention lately, with several datasets and neural models proposed, though with limited success. More recently, a new class of works have emerged adopting a Neuro-Symbolic approach, called transformer guided chaining, whereby the idea is to iteratively perform 1-step neural inferences and chain together the results to generate a multi-step reasoning trace. Several works have adapted variants of this central idea and reported significantly high accuracies compared to vanilla LLM’s. In this paper, we perform a critical empirical investigation of the chaining approach on a multi-hop First-Order Logic (FOL) reasoning benchmark. In particular, we develop a reference implementation, called Chainformer, and conduct several experiments to analyze the accuracy, generalization, interpretability, and performance over FOLs. Our findings highlight key strengths and possible current limitations and suggest potential areas for future research in logic reasoning.

pdf abs
Multilingual Multi-Figurative Language Detection
Huiyuan Lai | Antonio Toral | Malvina Nissim

Figures of speech help people express abstract concepts and evoke stronger emotions than literal expressions, thereby making texts more creative and engaging. Due to its pervasive and fundamental character, figurative language understanding has been addressed in Natural Language Processing, but it’s highly understudied in a multilingual setting and when considering more than one figure of speech at the same time. To bridge this gap, we introduce multilingual multi-figurative language modelling, and provide a benchmark for sentence-level figurative language detection, covering three common figures of speech and seven languages. Specifically, we develop a framework for figurative language detection based on template-based prompt learning. In so doing, we unify multiple detection tasks that are interrelated across multiple figures of speech and languages, without requiring task- or language-specific modules. Experimental results show that our framework outperforms several strong baselines and may serve as a blueprint for the joint modelling of other interrelated tasks.

pdf abs
Zero-shot Visual Question Answering with Language Model Feedback
Yifan Du | Junyi Li | Tianyi Tang | Wayne Xin Zhao | Ji-Rong Wen

In this paper, we propose a novel language model guided captioning approach, LAMOC, for knowledge-based visual question answering (VQA). Our approach employs the generated captions by a captioning model as the context of an answer prediction model, which is a Pre-Trained Language model (PLM). As the major contribution, we leverage the guidance and feedback of the prediction model to improve the capability of the captioning model. In this way, the captioning model can become aware of the task goal and information need from the PLM. To develop our approach, we design two specific training stages, where the first stage adapts the captioning model to the prediction model (selecting more suitable caption propositions for training) and the second stage tunes the captioning model according to the task goal (learning from feedback of the PLM). Extensive experiments demonstrate the effectiveness of the proposed approach on the knowledge-based VQA task. Specifically, on the challenging A-OKVQA dataset, LAMOC outperforms several competitive zero-shot methods and even achieves comparable results to a fine-tuned VLP model. Our code is publicly available at https://github.com/RUCAIBox/LAMOC.

pdf abs
Prompted Opinion Summarization with GPT-3.5
Adithya Bhaskar | Alex Fabbri | Greg Durrett

Large language models have shown impressive performance across a wide variety of tasks, including text summarization. In this paper, we show that this strong performance extends to opinion summarization. We explore several pipeline methods for applying GPT-3.5 to summarize a large collection of user reviews in aprompted fashion. To handle arbitrarily large numbers of user reviews, we explore recursive summarization as well as methods for selecting salient content to summarize through supervised clustering or extraction. On two datasets, an aspect-oriented summarization dataset of hotel reviews (SPACE) and a generic summarization dataset of Amazon and Yelp reviews (FewSum), we show that GPT-3.5 models achieve very strong performance in human evaluation. We argue that standard evaluation metrics do not reflect this, and introduce three new metrics targeting faithfulness, factuality, and genericity to contrast these different methods.

pdf abs
Sentence Ordering with a Coherence Verifier
Sainan Jia | Wei Song | Jiefu Gong | Shijin Wang | Ting Liu

This paper presents a novel sentence ordering method by plugging a coherence verifier (CoVer) into pair-wise ranking-based and sequence generation-based methods. It does not change the model parameters of the baseline, and only verifies the coherence of candidate (partial) orders produced by the baseline and reranks them in beam search. We also propose a coherence model as CoVer with a novel graph formulation and a novel data construction strategy for contrastive pre-training independently of the sentence ordering task. Experimental results on four benchmarks demonstrate the effectiveness of our method with topological sorting-based and pointer network-based methods as the baselines. Detailed analyses illustrate how CoVer improves the baselines and confirm the importance of its graph formulation and training strategy. Our code is available at https://github.com/SN-Jia/SO_with_CoVer.

pdf abs
GUMSum: Multi-Genre Data and Evaluation for English Abstractive Summarization
Yang Janet Liu | Amir Zeldes

Automatic summarization with pre-trained language models has led to impressively fluent results, but is prone to ‘hallucinations’, low performance on non-news genres, and outputs which are not exactly summaries. Targeting ACL 2023’s ‘Reality Check’ theme, we present GUMSum, a small but carefully crafted dataset of English summaries in 12 written and spoken genres for evaluation of abstractive summarization. Summaries are highly constrained, focusing on substitutive potential, factuality, and faithfulness. We present guidelines and evaluate human agreement as well as subjective judgments on recent system outputs, comparing general-domain untuned approaches, a fine-tuned one, and a prompt-based approach, to human performance. Results show that while GPT3 achieves impressive scores, it still underperforms humans, with varying quality across genres. Human judgments reveal different types of errors in supervised, prompted, and human-generated summaries, shedding light on the challenges of producing a good summary.

Grammatical error correction (GEC) is a promising task aimed at correcting errors in a text. Many methods have been proposed to facilitate this task with remarkable results. However, most of them only focus on enhancing textual feature extraction without exploring the usage of other modalities’ information (e.g., speech), which can also provide valuable knowledge to help the model detect grammatical errors. To shore up this deficiency, we propose a novel framework that integrates both speech and text features to enhance GEC. In detail, we create new multimodal GEC datasets for English and German by generating audio from text using the advanced text-to-speech models. Subsequently, we extract acoustic and textual representations by a multimodal encoder that consists of a speech and a text encoder. A mixture-of-experts (MoE) layer is employed to selectively align representations from the two modalities, and then a dot attention mechanism is used to fuse them as final multimodal representations. Experimental results on CoNLL14, BEA19 English, and Falko-MERLIN German show that our multimodal GEC models achieve significant improvements over strong baselines and achieve a new state-of-the-art result on the Falko-MERLIN test set.

pdf abs
Teaching the Pre-trained Model to Generate Simple Texts for Text Simplification
Renliang Sun | Wei Xu | Xiaojun Wan

Randomly masking text spans in ordinary texts in the pre-training stage hardly allows models to acquire the ability to generate simple texts. It can hurt the performance of pre-trained models on text simplification tasks. In this paper, we propose a new continued pre-training strategy to teach the pre-trained model to generate simple texts. We continue pre-training BART, a representative model, to obtain SimpleBART. It consistently and significantly improves the results on lexical simplification, sentence simplification, and document-level simplification tasks over BART. At the end, we compare SimpleBART with several representative large language models (LLMs).

pdf abs
Acquiring Frame Element Knowledge with Deep Metric Learning for Semantic Frame Induction
Kosuke Yamada | Ryohei Sasano | Koichi Takeda

The semantic frame induction tasks are defined as a clustering of words into the frames that they evoke, and a clustering of their arguments according to the frame element roles that they should fill. In this paper, we address the latter task of argument clustering, which aims to acquire frame element knowledge, and propose a method that applies deep metric learning. In this method, a pre-trained language model is fine-tuned to be suitable for distinguishing frame element roles through the use of frame-annotated data, and argument clustering is performed with embeddings obtained from the fine-tuned model. Experimental results on FrameNet demonstrate that our method achieves substantially better performance than existing methods.

pdf abs
Leveraging Synthetic Targets for Machine Translation
Sarthak Mittal | Oleksii Hrinchuk | Oleksii Kuchaiev

In this work, we provide a recipe for training machine translation models in a limited resource setting by leveraging synthetic target data generated using a large pre-trained model. We show that consistently across different benchmarks in bilingual, multilingual, and speech translation setups, training models on synthetic targets outperforms training on the actual ground-truth data. This performance gap grows bigger with increasing limits on the amount of available resources in the form of the size of the dataset and the number of parameters in the model. We also provide preliminary analysis into whether this boost in performance is linked to ease of optimization or more deterministic nature of the predictions, and whether this paradigm leads to better out-of-distribution performance across different testing domains.

Pre-trained encoder-only and sequence-to-sequence (seq2seq) models each have advantages, however training both model types from scratch is computationally expensive. We explore recipes to improve pre-training efficiency by initializing one model from the other. (1) Extracting the encoder from a seq2seq model, we show it under-performs a Masked Language Modeling (MLM) encoder, particularly on sequence labeling tasks. Variations of masking during seq2seq training, reducing the decoder size, and continuing with a small amount of MLM training do not close the gap. (2) Conversely, using an encoder to warm-start seq2seq training, we show that by unfreezing the encoder partway through training, we can match task performance of a from-scratch seq2seq model. Overall, this two-stage approach is an efficient recipe to obtain both a multilingual encoder and a seq2seq model, matching the performance of training each model from scratch while reducing the total compute cost by 27%.

pdf abs
Constructing Code-mixed Universal Dependency Forest for Unbiased Cross-lingual Relation Extraction
Hao Fei | Meishan Zhang | Min Zhang | Tat-Seng Chua

Latest efforts on cross-lingual relation extraction (XRE) aggressively leverage the language-consistent structural features from the universal dependency (UD) resource, while they may largely suffer from biased transfer (e.g., either target-biased or source-biased) due to the inevitable linguistic disparity between languages. In this work, we investigate an unbiased UD- based XRE transfer by constructing a type of code-mixed UD forest. We first translate the sentence of the source language to the parallel target-side language, for both of which we parse the UD tree respectively. Then, we merge the source-/target-side UD structures as a unified code-mixed UD forest. With such forest features, the gaps of UD-based XRE between the training and predicting phases can be effectively closed. We conduct experiments on the ACE XRE benchmark datasets, where the results demonstrate that the proposed code-mixed UD forests help unbiased UD-based XRE transfer, with which we achieve significant XRE performance gains.

pdf abs
Spontaneous gestures encoded by hand positions improve language models: An Information-Theoretic motivated study
Yang Xu | Yang Cheng

The multi-modality nature of human communication has been utilized to enhance the performance of language modeling-related tasks. Driven by the development of large-scale end-to-end learning techniques and the availability of multi-modal data, it becomes possible to represent non-verbal communication behaviors through joint-learning, and directly study their interaction with verbal communication. However, there is still gaps in existing studies to better address the underlying mechanism of how non-verbal expression contributes to the overall communication purpose. Therefore, we explore two questions using mixed-modal language models trained against monologue video data: first, whether incorporating gesture representations can improve the language model’s performance (perplexity); second, whether spontaneous gestures demonstrate entropy rate constancy (ERC), which is an empirical pattern found in most verbal language data that supports the rational communication assumption from Information Theory. We have positive and interesting findings for both questions: speakers indeed use spontaneous gestures to convey “meaningful” information that enhances verbal communication, which can be captured with a simple spatial encoding scheme. More importantly, gestures are produced and organized rationally in a similar way as words, which optimizes the communication efficiency.

pdf abs
Progressive Translation: Improving Domain Robustness of Neural Machine Translation with Intermediate Sequences
Chaojun Wang | Yang Liu | Wai Lam

Previous studies show that intermediate supervision signals benefit various Natural Language Processing tasks. However, it is not clear whether there exist intermediate signals that benefit Neural Machine Translation (NMT). Borrowing techniques from Statistical Machine Translation, we propose intermediate signals which are intermediate sequences from the “source-like” structure to the “target-like” structure. Such intermediate sequences introduce an inductive bias that reflects a domain-agnostic principle of translation, which reduces spurious correlations that are harmful to out-of-domain generalisation. Furthermore, we introduce a full-permutation multi-task learning to alleviate the spurious causal relations from intermediate sequences to the target, which results from exposure bias. The Minimum Bayes Risk decoding algorithm is used to pick the best candidate translation from all permutations to further improve the performance. Experiments show that the introduced intermediate signals can effectively improve the domain robustness of NMT and reduces the amount of hallucinations on out-of-domain translation. Further analysis shows that our methods are especially promising in low-resource scenarios.

We propose CHRT (Control HiddenRepresentation Transformation) – a con-trolled language generation framework thatsteers large language models to generatetext pertaining to certain attributes (such astoxicity). CHRT gains attribute control bymodifying the hidden representation of thebase model through learned transformations. We employ a contrastive-learning frameworkto learn these transformations that can becombined to gain multi-attribute control. Theeffectiveness of CHRT is experimentallyshown by comparing it with seven baselinesover three attributes. CHRT outperforms all thebaselines in the task of detoxification, positivesentiment steering, and text simplificationwhile minimizing the loss in linguistic qualities. Further, our approach has the lowest inferencelatency of only 0.01 seconds more than thebase model, making it the most suitable forhigh-performance production environments. We open-source our code and release two noveldatasets to further propel controlled languagegeneration research

pdf abs
Visual Coherence Loss for Coherent and Visually Grounded Story Generation
Xudong Hong | Vera Demberg | Asad Sayeed | Qiankun Zheng | Bernt Schiele

Local coherence is essential for long-form text generation models. We identify two important aspects of local coherence within the visual storytelling task: (1) the model needs to represent re-occurrences of characters within the image sequence in order to mention them correctly in the story; (2) character representations should enable us to find instances of the same characters and distinguish different characters. In this paper, we propose a loss function inspired by a linguistic theory of coherence for self-supervised learning for image sequence representations. We further propose combining features from an object and a face detector to construct stronger character features. To evaluate input-output relevance that current reference-based metrics don’t measure, we propose a character matching metric to check whether the models generate referring expressions correctly for characters in input image sequences. Experiments on a visual story generation dataset show that our proposed features and loss function are effective for generating more coherent and visually grounded stories.

Tabular data analysis is performed everyday across various domains. It requires an accurate understanding of field semantics to correctly operate on table fields and find common patterns in daily analysis. In this paper, we introduce the AnaMeta dataset, a collection of 467k tables with derived supervision labels for four types of commonly used field metadata: measure/dimension dichotomy, common field roles, semantic field type, and default aggregation function. We evaluate a wide range of models for inferring metadata as the benchmark. We also propose a multi-encoder framework, called KDF, which improves the metadata understanding capability of tabular models by incorporating distribution and knowledge information. Furthermore, we propose four interfaces for incorporating field metadata into downstream analysis tasks.

pdf abs
Large Language Models Are Partially Primed in Pronoun Interpretation
Suet-Ying Lam | Qingcheng Zeng | Kexun Zhang | Chenyu You | Rob Voigt

While a large body of literature suggests that large language models (LLMs) acquire rich linguistic representations, little is known about whether they adapt to linguistic biases in a human-like way. The present study probes this question by asking whether LLMs display human-like referential biases using stimuli and procedures from real psycholinguistic experiments. Recent psycholinguistic studies suggest that humans adapt their referential biases with recent exposure to referential patterns; closely replicating three relevant psycholinguistic experiments from Johnson & Arnold (2022) in an in-context learning (ICL) framework, we found that InstructGPT adapts its pronominal interpretations in response to the frequency of referential patterns in the local discourse, though in a limited fashion: adaptation was only observed relative to syntactic but not semantic biases. By contrast, FLAN-UL2 fails to generate meaningful patterns. Our results provide further evidence that contemporary LLMs discourse representations are sensitive to syntactic patterns in the local context but less so to semantic patterns. Our data and code are available at https://github.com/zkx06111/llm_priming.

pdf abs
Counterfactuals of Counterfactuals: a back-translation-inspired approach to analyse counterfactual editors
George Filandrianos | Edmund Dervakos | Orfeas Menis Mastromichalakis | Chrysoula Zerva | Giorgos Stamou

In the wake of responsible AI, interpretability methods, which attempt to provide an explanation for the predictions of neural models have seen rapid progress. In this work, we are concerned with explanations that are applicable to natural language processing (NLP) models and tasks, and we focus specifically on the analysis of counterfactual, contrastive explanations. We note that while there have been several explainers proposed to produce counterfactual explanations, their behaviour can vary significantly and the lack of a universal ground truth for the counterfactual edits imposes an insuperable barrier on their evaluation. We propose a new back translation-inspired evaluation methodology that utilises earlier outputs of the explainer as ground truth proxies to investigate the consistency of explainers. We show that by iteratively feeding the counterfactual to the explainer we can obtain valuable insights into the behaviour of both the predictor and the explainer models, and infer patterns that would be otherwise obscured. Using this methodology, we conduct a thorough analysis and propose a novel metric to evaluate the consistency of counterfactual generation approaches with different characteristics across available performance indicators.

pdf abs
A Pilot Study on Dialogue-Level Dependency Parsing for Chinese
Gongyao Jiang | Shuang Liu | Meishan Zhang | Min Zhang

Dialogue-level dependency parsing has received insufficient attention, especially for Chinese. To this end, we draw on ideas from syntactic dependency and rhetorical structure theory (RST), developing a high-quality human-annotated corpus, which contains 850 dialogues and 199,803 dependencies. Considering that such tasks suffer from high annotation costs, we investigate zero-shot and few-shot scenarios. Based on an existing syntactic treebank, we adopt a signal-based method to transform seen syntactic dependencies into unseen ones between elementary discourse units (EDUs), where the signals are detected by masked language modeling. Besides, we apply single-view and multi-view data selection to access reliable pseudo-labeled instances. Experimental results show the effectiveness of these baselines. Moreover, we discuss several crucial points about our dataset and approach.

pdf abs
On the Off-Target Problem of Zero-Shot Multilingual Neural Machine Translation
Liang Chen | Shuming Ma | Dongdong Zhang | Furu Wei | Baobao Chang

While multilingual neural machine translation has achieved great success, it suffers from the off-target issue, where the translation is in the wrong language. This problem is more pronounced on zero-shot translation tasks. In this work, we find that failing in encoding discriminative target language signal will lead to off-target and a closer lexical distance (i.e., KL-divergence) between two languages’ vocabularies is related with a higher off-target rate. We also find that solely isolating the vocab of different languages in the decoder can alleviate the problem. Motivated by the findings, we propose Language Aware Vocabulary Sharing (LAVS), a simple and effective algorithm to construct the multilingual vocabulary, that greatly alleviates the off-target problem of the translation model by increasing the KL-divergence between languages. We conduct experiments on a multilingual machine translation benchmark in 11 languages. Experiments show that the off-target rate for 90 translation tasks is reduced from 29% to 8%, while the overall BLEU score is improved by an average of 1.9 points without extra training cost or sacrificing the supervised directions’ performance. We release the code at https://github.com/PKUnlp-icler/Off-Target-MNMT for reproduction.

pdf abs
ORCA: A Challenging Benchmark for Arabic Language Understanding
AbdelRahim Elmadany | ElMoatez Billah Nagoudi | Muhammad Abdul-Mageed

Due to the crucial role pretrained language models play in modern NLP, several benchmarks have been proposed to evaluate their performance. In spite of these efforts, no public benchmark of diverse nature currently exists for evaluating Arabic NLU. This makes it challenging to measure progress for both Arabic and multilingual language models. This challenge is compounded by the fact that any benchmark targeting Arabic needs to take into account the fact that Arabic is not a single language but rather a collection of languages and language varieties. In this work, we introduce a publicly available benchmark for Arabic language understanding evaluation dubbed ORCA. It is carefully constructed to cover diverse Arabic varieties and a wide range of challenging Arabic understanding tasks exploiting 60 different datasets (across seven NLU task clusters). To measure current progress in Arabic NLU, we use ORCA to offer a comprehensive comparison between 18 multilingual and Arabic language models. We also provide a public leaderboard with a unified single-number evaluation metric (ORCA score) to facilitate future research.

pdf abs
Delving into the Openness of CLIP
Shuhuai Ren | Lei Li | Xuancheng Ren | Guangxiang Zhao | Xu Sun

Contrastive Language-Image Pre-training (CLIP) formulates image classification as an image-to-text matching task, i.e., matching images to the corresponding natural language descriptions instead of discrete category IDs. This allows for open-vocabulary visual recognition, where the model can recognize images from an open class set (also known as an open vocabulary) in a zero-shot manner. However, evaluating the openness of CLIP-like models is challenging, as the models are open to arbitrary vocabulary in theory, but their accuracy varies in practice. To address this, we resort to an incremental perspective to assess the openness through vocabulary expansions, and define extensibility to measure a model’s ability to handle novel classes. Our evaluation shows that CLIP-like models are not truly open, and their performance deteriorates as the vocabulary expands. We further dissect the feature space of CLIP from the perspectives of representation alignment and uniformity. Our investigation reveals that the overestimation of openness is due to confusion among competing text features, rather than a failure to capture the similarity between image features and text features of novel classes. We hope that our investigation and analysis will facilitate future research on the CLIP openness issue.

Textual adversarial attacks can discover models’ weaknesses by adding semantic-preserved but misleading perturbations to the inputs. The long-lasting adversarial attack-and-defense arms race in Natural Language Processing (NLP) is algorithm-centric, providing valuable techniques for automatic robustness evaluation. However, the existing practice of robustness evaluation may exhibit issues of incomprehensive evaluation, impractical evaluation protocol, and invalid adversarial samples. In this paper, we aim to set up a unified automatic robustness evaluation framework, shifting towards model-centric evaluation to further exploit the advantages of adversarial attacks. To address the above challenges, we first determine robustness evaluation dimensions based on model capabilities and specify the reasonable algorithm to generate adversarial samples for each dimension. Then we establish the evaluation protocol, including evaluation settings and metrics, under realistic demands. Finally, we use the perturbation degree of adversarial samples to control the sample validity. We implement a toolkit RobTest that realizes our automatic robustness evaluation framework. In our experiments, we conduct a robustness evaluation of RoBERTa models to demonstrate the effectiveness of our evaluation framework, and further show the rationality of each component in the framework.

Aspect-Based Sentiment Analysis (ABSA) aims to recognize fine-grained opinions and sentiments of users, which is an important problem in sentiment analysis. Recent work has shown that Sentiment-enhanced Pre-Training (SPT) can substantially improve the performance of various ABSA tasks. However, there is currently a lack of comprehensive evaluation and fair comparison of existing SPT approaches. Therefore, this paper performs an empirical study to investigate the effectiveness of different SPT approaches. First, we develop an effective knowledge-mining method and leverage it to build a large-scale knowledge-annotated SPT corpus. Second, we systematically analyze the impact of integrating sentiment knowledge and other linguistic knowledge in pre-training. For each type of sentiment knowledge, we also examine and compare multiple integration methods. Finally, we conduct extensive experiments on a wide range of ABSA tasks to see how much SPT can facilitate the understanding of aspect-level sentiments.

Despite growing interest in applications based on natural customer support conversations,there exist remarkably few publicly available datasets that reflect the expected characteristics of conversations in these settings. Existing task-oriented dialogue datasets, which were collected to benchmark dialogue systems mainly in written human-to-bot settings, are not representative of real customer support conversations and do not provide realistic benchmarks for systems that are applied to natural data. To address this gap, we introduce NatCS, a multi-domain collection of spoken customer service conversations. We describe our process for collecting synthetic conversations between customers and agents based on natural language phenomena observed in real conversations. Compared to previous dialogue datasets, the conversations collected with our approach are more representative of real human-to-human conversations along multiple metrics. Finally, we demonstrate potential uses of NatCS, including dialogue act classification and intent induction from conversations as potential applications, showing that dialogue act annotations in NatCS provide more effective training data for modeling real conversations compared to existing synthetic written datasets. We publicly release NatCS to facilitate research in natural dialog systems

The large scale of pre-trained language models poses a challenge for their deployment on various devices, with a growing emphasis on methods to compress these models, particularly knowledge distillation. However, current knowledge distillation methods rely on the model’s intermediate layer features and the golden labels (also called hard labels), which usually require aligned model architecture and enough labeled data respectively. Moreover, the parameters of vocabulary are usually neglected in existing methods. To address these problems, we propose a general language model distillation (GLMD) method that performs two-stage word prediction distillation and vocabulary compression, which is simple and surprisingly shows extremely strong performance. Specifically, GLMD supports more general application scenarios by eliminating the constraints of dimension and structure between models and the need for labeled datasets through the absence of intermediate layers and golden labels. Meanwhile, based on the long-tailed distribution of word frequencies in the data, GLMD designs a strategy of vocabulary compression through decreasing vocabulary size instead of dimensionality. Experimental results show that our method outperforms 25 state-of-the-art methods on the SuperGLUE benchmark, achieving an average score that surpasses the best method by 3%.

Sequence-to-sequence state-of-the-art systems for dialogue state tracking (DST) use the full dialogue history as input, represent the current state as a list with all the slots, and generate the entire state from scratch at each dialogue turn. This approach is inefficient, especially when the number of slots is large and the conversation is long. We propose Diable, a new task formalisation that simplifies the design and implementation of efficient DST systems and allows one to easily plug and play large language models. We represent the dialogue state as a table and formalise DST as a table manipulation task. At each turn, the system updates the previous state by generating table operations based on the dialogue context. Extensive experimentation on the MultiWoz datasets demonstrates that Diable (i) outperforms strong efficient DST baselines, (ii) is 2.4x more time efficient than current state-of-the-art methods while retaining competitive Joint Goal Accuracy, and (iii) is robust to noisy data annotations due to the table operations approach.

pdf abs
Neural Topic Modeling based on Cycle Adversarial Training and Contrastive Learning
Boyu Wang | Linhai Zhang | Deyu Zhou | Yi Cao | Jiandong Ding

Neural topic models have been widely used to extract common topics across documents. Recently, contrastive learning has been applied to variational autoencoder-based neural topic models, achieving promising results. However, due to the limitation of the unidirectional structure of the variational autoencoder, the encoder is enhanced with the contrastive loss instead of the decoder, leading to a gap between model training and evaluation. To address the limitation, we propose a novel neural topic modeling framework based on cycle adversarial training and contrastive learning to apply contrastive learning on the generator directly. Specifically, a self-supervised contrastive loss is proposed to make the generator capture similar topic information, which leads to better topic-word distributions. Meanwhile, a discriminative contrastive loss is proposed to cooperate with the self-supervised contrastive loss to balance the generation and discrimination. Moreover, based on the reconstruction ability of the cycle generative adversarial network, a novel data augmentation strategy is designed and applied to the topic distribution directly. Experiments have been conducted on four benchmark datasets and results show that the proposed approach outperforms competitive baselines.

pdf abs
Alleviating Exposure Bias via Multi-level Contrastive Learning and Deviation Simulation in Abstractive Summarization
Jiawen Xie | Qi Su | Shaoting Zhang | Xiaofan Zhang

Most Transformer based abstractive summarization systems have a severe mismatch between training and inference, i.e., exposure bias. From diverse perspectives, we introduce a simple multi-level contrastive learning framework for abstractive summarization (SimMCS) and a tailored sparse decoder self-attention pattern (SDSA) to bridge the gap between training and inference to improve model performance. Compared with previous contrastive objectives focusing only on the relative order of probability mass assigned to non-gold summaries, SimMCS additionally takes their absolute positions into account, which guarantees that the relatively high-quality (positive) summaries among them could be properly assigned high probability mass, and further enhances the capability of discriminating summary quality beyond exploiting potential artifacts of specific metrics. SDSA simulates the possible inference scenarios of deviation in the training phase to get closer to the ideal paradigm. Our approaches outperform the previous state-of-the-art results on two summarization datasets while just adding fairly low overhead. Further empirical analysis shows our model preserves the advantages of prior contrastive methods and possesses strong few-shot learning ability.

pdf abs
Mapping Brains with Language Models: A Survey
Antonia Karamolegkou | Mostafa Abdou | Anders Søgaard

Over the years, many researchers have seemingly made the same observation: Brain and language model activations exhibit some structural similarities, enabling linear partial mappings between features extracted from neural recordings and computational language models. In an attempt to evaluate how much evidence has been accumulated for this observation, we survey over 30 studies spanning 10 datasets and 8 metrics. How much evidence has been accumulated, and what, if anything, is missing before we can draw conclusions? Our analysis of the evaluation methods used in the literature reveals that some of the metrics are less conservative. We also find that the accumulated evidence, for now, remains ambiguous, but correlations with model size and quality provide grounds for cautious optimism.

pdf abs
Parameter-Efficient Finetuning for Robust Continual Multilingual Learning
Kartikeya Badola | Shachi Dave | Partha Talukdar

We introduce and study the problem of Continual Multilingual Learning (CML) where a previously trained multilingual model is periodically updated using new data arriving in stages. If the new data is present only in a subset of languages, we find that the resulting model shows improved performance only on the languages included in the latest update (and a few closely related languages) while its performance on all the remaining languages degrade significantly. We address this challenge by proposing LAFT-URIEL, a parameter-efficient finetuning strategy which aims to increase the number of languages on which the model improves after an update, while reducing the magnitude of loss in performance for the remaining languages. LAFT-URIEL uses linguistic knowledge to balance overfitting and knowledge sharing across languages, allowing for an additional 25% of task languages to see an improvement in performance after an update, while also reducing the average magnitude of losses on the remaining languages by 78% relative.

pdf abs
Interpretable Multimodal Misinformation Detection with Logic Reasoning
Hui Liu | Wenya Wang | Haoliang Li

Multimodal misinformation on online social platforms is becoming a critical concern due to increasing credibility and easier dissemination brought by multimedia content, compared to traditional text-only information. While existing multimodal detection approaches have achieved high performance, the lack of interpretability hinders these systems’ reliability and practical deployment. Inspired by Neural-Symbolic AI which combines the learning ability of neural networks with the explainability of symbolic learning, we propose a novel logic-based neural model for multimodal misinformation detection which integrates interpretable logic clauses to express the reasoning process of the target task. To make learning effective, we parameterize the symbolic logical elements using neural representations, which facilitate the automatic generation and evaluation of meaningful logic clauses. Additionally, to make our framework generalizable across diverse misinformation sources, we introduce five meta-predicates that can be instantiated with different correlations. Results on three public datasets (Twitter, Weibo, and Sarcasm) demonstrate the feasibility and versatility of our model.

Visual segmentation from language queries has attracted significant research interest. Despite the effectiveness, existing works require expensive labeling and suffer severe degradation when deployed to an unseen domain. In this paper, we investigate a novel task Cross-domain Query-based Visual Segmentation (CQVS), aiming to adapt the segmentation model from a labeled domain to a new unlabeled domain. The challenges of CQVS stem from three domain discrepancies: (1) multi-modal content shift, (2) uni-modal feature gap and (3) cross-modal relation bias. Existing domain adaptation methods fail to address them comprehensively and precisely (e.g. at pixel level), thus being suboptimal for CQVS. To overcome this limitation, we propose Semantic-conditioned Dual Adaptation (SDA), a novel framework to achieve precise feature- and relation-invariant across domains via a universal semantic structure. The SDA consists of two key components: Content-aware Semantic Modeling (CSM) and Dual Adaptive Branches (DAB). First, CSM introduces a common semantic space across domains to provide uniform guidance. Then, DAB seamlessly leverages this semantic information to develop a contrastive feature branch for category-wise pixel alignment, and design a reciprocal relation branch for relation enhancement via two complementary masks. Extensive experiments on three video benchmarks and three image benchmarks evidence the superiority of our approach over the state-of-the-arts.

pdf abs
Figurative Language Processing: A Linguistically Informed Feature Analysis of the Behavior of Language Models and Humans
Hyewon Jang | Qi Yu | Diego Frassinelli

Recent years have witnessed a growing interest in investigating what Transformer-based language models (TLMs) actually learn from the training data. This is especially relevant for complex tasks such as the understanding of non-literal meaning. In this work, we probe the performance of three black-box TLMs and two intrinsically transparent white-box models on figurative language classification of sarcasm, similes, idioms, and metaphors. We conduct two studies on the classification results to provide insights into the inner workings of such models. With our first analysis on feature importance, we identify crucial differences in model behavior. With our second analysis using an online experiment with human participants, we inspect different linguistic characteristics of the four figurative language types.

pdf abs
Taxonomy of Problems in Lexical Semantics
Bradley Hauer | Grzegorz Kondrak

Semantic tasks are rarely formally defined, and the exact relationship between them is an open question. We introduce a taxonomy that elucidates the connection between several problems in lexical semantics, including monolingual and cross-lingual variants. Our theoretical framework is based on the hypothesis of the equivalence of concept and meaning distinctions. Using algorithmic problem reductions, we demonstrate that all problems in the taxonomy can be reduced to word sense disambiguation (WSD), and that WSD itself can be reduced to some problems, making them theoretically equivalent. In addition, we carry out experiments that strongly support the soundness of the concept-meaning hypothesis, and the correctness of our reductions.

pdf abs
Making Pre-trained Language Models both Task-solvers and Self-calibrators
Yangyi Chen | Xingyao Wang | Heng Ji

Pre-trained language models (PLMs) serve as backbones for various real-world systems. For high-stake applications, it’s equally essential to have reasonable confidence estimations in predictions. While the vanilla confidence scores of PLMs can already be effectively utilized, PLMs consistently become overconfident in their wrong predictions, which is not desirable in practice. Previous work shows that introducing an extra calibration task can mitigate this issue. The basic idea involves acquiring additional data to train models in predicting the confidence of their initial predictions. However, it only demonstrates the feasibility of this kind of method, assuming that there are abundant extra available samples for the introduced calibration task. In this work, we consider the practical scenario that we need to effectively utilize training samples to make PLMs both task-solvers and self-calibrators. Three challenges are presented, including limited training samples, data imbalance, and distribution shifts. We first conduct pilot experiments to quantify various decisive factors in the calibration task. Based on the empirical analysis results, we propose a training algorithm LM-TOAST to tackle the challenges. Experimental results show that LM-TOAST can effectively utilize the training data to make PLMs have reasonable confidence estimations while maintaining the original task performance. Further, we consider three downstream applications, namely selective classification, adversarial defense, and model cascading, to show the practical usefulness of LM-TOAST.

pdf abs
EmbedTextNet: Dimension Reduction with Weighted Reconstruction and Correlation Losses for Efficient Text Embedding
Dae Yon Hwang | Bilal Taha | Yaroslav Nechaev

The size of embeddings generated by large language models can negatively affect system latency and model size in certain downstream practical applications (e.g. KNN search). In this work, we propose EmbedTextNet, a light add-on network that can be appended to an arbitrary language model to generate a compact embedding without requiring any changes in its architecture or training procedure. Specifically, we use a correlation penalty added to the weighted reconstruction loss that better captures the informative features in the text embeddings, which improves the efficiency of the language models. We evaluated EmbedTextNet on three different downstream tasks: text similarity, language modelling, and text retrieval. Empirical results on diverse benchmark datasets demonstrate the effectiveness and superiority of EmbedTextNet compared to state-of-art methodologies in recent works, especially in extremely low dimensional embedding sizes. The developed code for reproducibility is included in the supplementary material.

pdf abs
Denoising Enhanced Distantly Supervised Ultrafine Entity Typing
Yue Zhang | Hongliang Fei | Ping Li

Recently, the task of distantly supervised (DS) ultra-fine entity typing has received significant attention. However, DS data is noisy and often suffers from missing or wrong labeling issues resulting in low precision and low recall. This paper proposes a novel ultra-fine entity typing model with denoising capability. Specifically, we build a noise model to estimate the unknown labeling noise distribution over input contexts and noisy type labels. With the noise model, more trustworthy labels can be recovered by subtracting the estimated noise from the input. Furthermore, we propose an entity typing model, which adopts a bi-encoder architecture, is trained on the denoised data. Finally, the noise model and entity typing model are trained iteratively to enhance each other. We conduct extensive experiments on the Ultra-Fine entity typing dataset as well as OntoNotes dataset and demonstrate that our approach significantly outperforms other baseline methods.

pdf abs
INTapt: Information-Theoretic Adversarial Prompt Tuning for Enhanced Non-Native Speech Recognition
Eunseop Yoon | Hee Suk Yoon | John Harvill | Mark Hasegawa-Johnson | Chang Yoo

Automatic Speech Recognition (ASR) systems have attained unprecedented performance with large speech models pre-trained based on self-supervised speech representation learning. However, these pre-trained speech models suffer from representational bias as they tend to better represent those prominent accents (i.e., native (L1) English accent) in the pre-training speech corpus than less represented accents, resulting in a deteriorated performance for non-native (L2) English accents. Although there have been some approaches to mitigate this issue, all of these methods require updating the pre-trained model weights. In this paper, we propose Information Theoretic Adversarial Prompt Tuning (INTapt), which introduces prompts concatenated to the original input that can re-modulate the attention of the pre-trained model such that the corresponding input resembles a native (L1) English speech without updating the backbone weights. INTapt is trained simultaneously in the following two manners: (1) adversarial training to reduce accent feature dependence between the original input and the prompt-concatenated input and (2) training to minimize CTC loss for improving ASR performance to a prompt-concatenated input. Experimental results show that INTapt improves the performance of L2 English and increases feature similarity between L2 and L1 accents.

Previous studies have constantly observed that a language model repeats itself, creating repetitions in an output sequence. To cope with the issue, stochastic decoding schemes have been the de facto approaches; the strategies add randomness in inference, hence avoiding the “self-loop”. However, the remedy comes at the cost of sacrificing output quality due to the randomness involved. In this work, we introduce a deterministic decoding scheme, local temperature beam search. This inference algorithm is an embarrassingly simple variant of beam search, yet it reduces repetition, whose level is superior to that of a sampling-based decoding algorithm, while maintaining the level of coherence as in beam search. Our idea is rooted in the concept of model calibration; we view a repetition as a casualty from overconfidence in a model. Therefore, our work mitigates the miscalibration present in the course of inference with a post-calibration approach applied in beam-specific manner. Our inference scheme is validated on text completion tasks, in which the repetition problem is seen most clearly, and is exhaustively compared with existing inference schemes.

pdf abs
Explanation Graph Generation via Generative Pre-training over Synthetic Graphs
Han Cui | Shangzhan Li | Yu Zhang | Qi Shi

The generation of explanation graphs is a significant task that aims to produce explanation graphs in response to user input, revealing the internal reasoning process. This task is challenging due to the significant discrepancy be- tween unstructured user queries and structured explanation graphs. Current research commonly fine-tunes a text-based pre-trained language model on a small downstream dataset that is annotated with labeled graphs. However, due to the limited scale of available datasets, this approach may prove to be insufficient in bridging the gap between natural language text and structured graphs. In this paper, to alleviate the above limitations, we propose a novel pre-trained framework EG3P(for Explanation Graph Generation via Generative Pre-training over synthetic graphs) for the explanation graph generation task. Specifically, we first propose a text-to-graph generative task to pre-train the model with the goal of bridging the text-graph gap. Additionally, we propose an automatic corpus synthesis strategy for synthesizing a large scale of high-quality corpus, reducing the reliance on costly manual annotation methods. Experimental results on ExplaGraphs show the effectiveness of EG3P that our model surpasses all baseline systems with remarkable margins. Besides, further analysis demonstrates that EG3P is able to generate better explanation graphs on actual reasoning tasks such as CommonsenseQA and OpenbookQA.

We introduce NaSGEC, a new dataset to facilitate research on Chinese grammatical error correction (CGEC) for native speaker texts from multiple domains. Previous CGEC research primarily focuses on correcting texts from a single domain, especially learner essays. To broaden the target domain, we annotate multiple references for 12,500 sentences from three native domains, i.e., social media, scientific writing, and examination. We provide solid benchmark results for NaSGEC by employing cutting-edge CGEC models and different training data. We further perform detailed analyses of the connections and gaps between our domains from both empirical and statistical views. We hope this work can inspire future studies on an important but under-explored direction–cross-domain GEC.

pdf abs
FORK: A Bite-Sized Test Set for Probing Culinary Cultural Biases in Commonsense Reasoning Models
Shramay Palta | Rachel Rudinger

It is common sense that one should prefer to eat a salad with a fork rather than with a chainsaw. However, for eating a bowl of rice, the choice between a fork and a pair of chopsticks is culturally relative. We introduce FORK, a small, manually-curated set of CommonsenseQA-style questions for probing cultural biases and assumptions present in commonsense reasoning systems, with a specific focus on food-related customs. We test several CommonsenseQA systems on FORK, and while we see high performance on questions about the US culture, the poor performance of these systems on questions about non-US cultures highlights systematic cultural assumptions aligned with US over non-US cultures.

With increasing concerns about data privacy, there is an increasing necessity of fine-tuning pre-trained language models (PLMs) for adapting to downstream tasks located in end-user devices or local clients without transmitting data to the central server. This urgent necessity therefore calls the research of investigating federated learning (FL) for PLMs. However, large PLMs bring the curse of prohibitive communication overhead and local model adaptation costs for the FL system. To this end, we investigate the parameter-efficient tuning (PETuning) of PLMs and develop a corresponding federated benchmark for four representative PETuning methods, dubbed FedPETuning. Specifically, FedPETuning provides the first holistic empirical study of representative PLMs tuning methods in FL, covering privacy attacks, performance comparisons, and resource-constrained analysis. Intensive experimental results have indicated that FedPETuning can efficiently defend against privacy attacks and maintains acceptable performance with reducing heavy resource consumption. The open-source code and data are available at https://github.com/SMILELab-FL/FedPETuning.

The task of product attribute value extraction is to identify values of an attribute from product information. Product attributes are important features, which help improve online shopping experience of customers, such as product search, recommendation and comparison. Most existing works only focus on extracting values for a set of known attributes with sufficient training data. However, with the emerging nature of e-commerce, new products with their unique set of new attributes are constantly generated from different retailers and merchants. Collecting a large number of annotations for every new attribute is costly and time consuming. Therefore, it is an important research problem for product attribute value extraction with limited data. In this work, we propose a novel prompt tuning approach with Mixed Prompts for few-shot Attribute Value Extraction, namely MixPAVE. Specifically, MixPAVE introduces only a small amount (< 1%) of trainable parameters, i.e., a mixture of two learnable prompts, while keeping the existing extraction model frozen. In this way, MixPAVE not only benefits from parameter-efficient training, but also avoids model overfitting on limited training examples. Experimental results on two product benchmarks demonstrate the superior performance of the proposed approach over several state-of-the-art baselines. A comprehensive set of ablation studies validate the effectiveness of the prompt design, as well as the efficiency of our approach.

pdf abs
SlowBERT: Slow-down Attacks on Input-adaptive Multi-exit BERT
Shengyao Zhang | Xudong Pan | Mi Zhang | Min Yang

For pretrained language models such as Google’s BERT, recent research designs several input-adaptive inference mechanisms to improve the efficiency on cloud and edge devices. In this paper, we reveal a new attack surface on input-adaptive multi-exit BERT, where the adversary imperceptibly modifies the input texts to drastically increase the average inference cost. Our proposed slow-down attack called SlowBERT integrates a new rank-and-substitute adversarial text generation algorithm to efficiently search for the perturbation which maximally delays the exiting time. With no direct access to the model internals, we further devise a time-based approximation algorithm to infer the exit position as the loss oracle. Our extensive evaluation on two popular instances of multi-exit BERT for GLUE classification tasks validates the effectiveness of SlowBERT. In the worst case, SlowBERT increases the inference cost by 4.57×, which would strongly hurt the service quality of multi-exit BERT in practice, e.g., increasing the real-time cloud services’ response times for online users.

Solving math word problem (MWP) remains a challenging task, as it requires to understand both the semantic meanings of the text and the mathematical logic among quantities, i.e., for both semantics modal and quantity modal learning. Current MWP encoders work in a uni-modal setting and map the given problem description to a latent representation, then for decoding. The generalizability of these MWP encoders is thus limited because some problems are semantics-demanding and others are quantity-demanding. To address this problem, we propose a Compositional Math Word Problem Solver (C-MWP) which works in a bi-modal setting encoding in an interactive way. Extensive experiments validate the effectiveness of C-MWP and show its superiority over state-of-the-art models on public benchmarks.

pdf abs
PREADD: Prefix-Adaptive Decoding for Controlled Text Generation
Jonathan Pei | Kevin Yang | Dan Klein

We propose Prefix-Adaptive Decoding (PREADD), a flexible method for controlled text generation. Unlike existing methods that use auxiliary expert models to control for attributes, PREADD does not require an external model, instead relying on linearly combining output logits from multiple prompts. Specifically, PREADD contrasts the output logits generated using a raw prompt against those generated using a prefix-prepended prompt, enabling both positive and negative control with respect to any attribute encapsulated by the prefix. We evaluate PREADD on three tasks—toxic output mitigation, gender bias reduction, and sentiment control—and find that PREADD outperforms not only prompting baselines, but also an auxiliary-expert control method, by 12% or more in relative gain on our main metrics for each task.

Event ontology provides a shared and formal specification about what happens in the real world and can benefit many natural language understanding tasks. However, the independent development of event ontologies often results in heterogeneous representations that raise the need for establishing alignments between semantically related events. There exists a series of works about ontology alignment (OA), but they only focus on the entity-based OA, and neglect the event-based OA. To fill the gap, we construct an Event Ontology Alignment (EventOA) dataset based on FrameNet and Wikidata, which consists of 900+ event type alignments and 8,000+ event argument alignments. Furthermore, we propose a multi-view event ontology alignment (MEOA) method, which utilizes description information (i.e., name, alias and definition) and neighbor information (i.e., subclass and superclass) to obtain richer representation of the event ontologies. Extensive experiments show that our MEOA outperforms the existing entity-based OA methods and can serve as a strong baseline for EventOA research.

Continual relation extraction (CRE) models aim at handling emerging new relations while avoiding catastrophically forgetting old ones in the streaming data. Though improvements have been shown by previous CRE studies, most of them only adopt a vanilla strategy when models first learn representations of new relations. In this work, we point out that there exist two typical biases after training of this vanilla strategy: classifier bias and representation bias, which causes the previous knowledge that the model learned to be shaded. To alleviate those biases, we propose a simple yet effective classifier decomposition framework that splits the last FFN layer into separated previous and current classifiers, so as to maintain previous knowledge and encourage the model to learn more robust representations at this training stage. Experimental results on two standard benchmarks show that our proposed framework consistently outperforms the state-of-the-art CRE models, which indicates that the importance of the first training stage to CRE models may be underestimated. Our code will be released upon acceptance.

pdf abs
A Comparative Analysis of the Effectiveness of Rare Tokens on Creative Expression using ramBERT
Youbin Lee | Deokgi Kim | Byung-Won On | Ingyu Lee

Until now, few studies have been explored on Automated Creative Essay Scoring (ACES), in which a pre-trained model automatically labels an essay as a creative or a non-creative. Since the creativity evaluation of essays is very subjective, each evaluator often has his or her own criteria for creativity. For this reason, quantifying creativity in essays is very challenging. In this work, as one of preliminary studies in developing a novel model for ACES, we deeply investigate the correlation between creative essays and expressiveness. Specifically, we explore how rare tokens affect the evaluation of creativity for essays. For such a journey, we present five distinct methods to extract rare tokens, and conduct a comparative study on the correlation between rare tokens and creative essay evaluation results using BERT. Our experimental results showed clear correlation between rare tokens and creative essays. In all test sets, accuracies of our rare token masking-based BERT (ramBERT) model were improved over the existing BERT model up to 14%.

A long-standing difficulty in AI is the introduction of human-like reasoning in machine reading comprehension. Since algorithmic models can already perform as well as humans on simple quality assurance tasks thanks to the development of deep learning techniques, more difficult reasoning datasets have been presented. However, these datasets mainly focus on a single type of reasoning. There are still significant gaps in the studies when compared to the complex reasoning used in daily life. In this work, we introduce a brand-new dataset, named MTR. There are two parts to it: the first combines deductive and inductive reasoning, and the second does the same with inductive and defeasible reasoning. It consists of more than 30k QA instances, inferring relations between characters in short stories. Results show that state-of-the-art neural models do noticeably worse than expected. Our empirical results highlight the gap in the models’ ability to handle sophisticated inference.

pdf abs
NewsMet : A ‘do it all’ Dataset of Contemporary Metaphors in News Headlines
Rohan Joseph | Timothy Liu | Aik Beng Ng | Simon See | Sunny Rai

Metaphors are highly creative constructs of human language that grow old and eventually die. Popular datasets used for metaphor processing tasks were constructed from dated source texts. In this paper, we propose NewsMet, a large high-quality contemporary dataset of news headlines hand-annotated with metaphorical verbs. The dataset comprises headlines from various sources including political, satirical, reliable and fake. Our dataset serves the purpose of evaluation for the tasks of metaphor interpretation and generation. The experiments reveal several insights and limitations of using LLMs to automate metaphor processing tasks as frequently seen in the recent literature. The dataset is publicly available for research purposes https://github.com/AxleBlaze3/NewsMet_Metaphor_Dataset.

Knowledge graph embeddings (KGE) have been extensively studied to embed large-scale relational data for many real-world applications. Existing methods have long ignored the fact many KGs contain two fundamentally different views: high-level ontology-view concepts and fine-grained instance-view entities. They usually embed all nodes as vectors in one latent space. However, a single geometric representation fails to capture the structural differences between two views and lacks probabilistic semantics towards concepts’ granularity. We propose Concept2Box, a novel approach that jointly embeds the two views of a KG using dual geometric representations. We model concepts with box embeddings, which learn the hierarchy structure and complex relations such as overlap and disjoint among them. Box volumes can be interpreted as concepts’ granularity. Different from concepts, we model entities as vectors. To bridge the gap between concept box embeddings and entity vector embeddings, we propose a novel vector-to-box distance metric and learn both embeddings jointly. Experiments on both the public DBpedia KG and a newly-created industrial KG showed the effectiveness of Concept2Box.

pdf abs
Noise-Robust Training with Dynamic Loss and Contrastive Learning for Distantly-Supervised Named Entity Recognition
Zhiyuan Ma | Jintao Du | Shuheng Zhou

Distantly-supervised named entity recognition (NER) aims at training networks with distantly-labeled data, which is automatically obtained by matching entity mentions in the raw text with entity types in a knowledge base. Distant supervision may induce incomplete and noisy labels, so recent state-of-the-art methods employ sample selection mechanism to separate clean data from noisy data based on the model’s prediction scores. However, they ignore the noise distribution change caused by data selection, and they simply excludes noisy data during training, resulting in information loss. We propose to (1) use a dynamic loss function to better adapt to the changing noise during the training process, and (2) incorporate token level contrastive learning to fully utilize the noisy data as well as facilitate feature learning without relying on labels. Our method achieves superior performance on three benchmark datasets, outperforming existing distantly supervised NER models by significant margins.

pdf abs
Take a Break in the Middle: Investigating Subgoals towards Hierarchical Script Generation
Xinze Li | Yixin Cao | Muhao Chen | Aixin Sun

Goal-oriented Script Generation is a new task of generating a list of steps that can fulfill the given goal. In this paper, we propose to extend the task from the perspective of cognitive theory. Instead of a simple flat structure, the steps are typically organized hierarchically — Human often decompose a complex task into subgoals, where each subgoal can be further decomposed into steps. To establish the benchmark, we contribute a new dataset, propose several baseline methods, and set up evaluation metrics. Both automatic and human evaluation verify the high-quality of dataset, as well as the effectiveness of incorporating subgoals into hierarchical script generation. Furthermore, We also design and evaluate the model to discover subgoal, and find that it is a bit more difficult to decompose the goals than summarizing from segmented steps.

pdf abs
End-to-End Task-Oriented Dialogue Systems Based on Schema
Wiradee Imrattanatrai | Ken Fukuda

This paper presents a schema-aware end-to-end neural network model for handling task-oriented dialogues based on a dynamic set of slots within a schema. Contrary to existing studies that proposed end-to-end approaches for task-oriented dialogue systems by relying on a unified schema across domains, we design our approach to support a domain covering multiple services where diverse schemas are available. To enable better generalizability among services and domains with different schemas, we supply the schema’s context information including slot descriptions and value constraints to the model. The experimental results on a well-known Schema-Guided Dialogue (SGD) dataset demonstrated the performance improvement by the proposed model compared to state-of-the-art baselines in terms of end-to-end modeling, dialogue state tracking task, and generalization on new services and domains using a limited number of dialogues.

This paper presents “HaVQA”, the first multimodal dataset for visual question answering (VQA) tasks in the Hausa language. The dataset was created by manually translating 6,022 English question-answer pairs, which are associated with 1,555 unique images from the Visual Genome dataset. As a result, the dataset provides 12,044 gold standard English-Hausa parallel sentences that were translated in a fashion that guarantees their semantic match with the corresponding visual information. We conducted several baseline experiments on the dataset, including visual question answering, visual question elicitation, text-only and multimodal machine translation.

pdf abs
Claim-Dissector: An Interpretable Fact-Checking System with Joint Re-ranking and Veracity Prediction
Martin Fajcik | Petr Motlicek | Pavel Smrz

We present Claim-Dissector: a novel latent variable model for fact-checking and analysis, which given a claim and a set of retrieved evidence jointly learns to identify: (i) the relevant evidences to the given claim (ii) the veracity of the claim. We propose to disentangle the per-evidence relevance probability and its contribution to the final veracity probability in an interpretable way — the final veracity probability is proportional to a linear ensemble of per-evidence relevance probabilities. In this way, the individual contributions of evidences towards the final predicted probability can be identified. In per-evidence relevance probability, our model can further distinguish whether each relevant evidence is supporting (S) or refuting (R) the claim. This allows to quantify how much the S/R probability contributes to final verdict or to detect disagreeing evidence. Despite its interpretable nature, our system achieves results competetive with state-of-the-art on the FEVER dataset, as compared to typical two-stage system pipelines, while using significantly fewer parameters. Furthermore, our analysis shows that our model can learn fine-grained relevance cues while using coarse-grained supervision and we demonstrate it in 2 ways. (i) We show that our model can achieve competitive sentence recall while using only paragraph-level relevance supervision. (ii) Traversing towards the finest granularity of relevance, we show that our model is capable of identifying relevance at the token level. To do this, we present a new benchmark TLR-FEVER focusing on token-level interpretability — humans annotate tokens in relevant evidences they considered essential when making their judgment. Then we measure how similar are these annotations to the tokens our model is focusing on.

pdf abs
StructSP: Efficient Fine-tuning of Task-Oriented Dialog System by Using Structure-aware Boosting and Grammar Constraints
Truong Do | Phuong Nguyen | Minh Nguyen

We have investigated methods utilizing hierarchical structure information representation in the semantic parsing task and have devised a method that reinforces the semantic awareness of a pre-trained language model via a two-step fine-tuning mechanism: hierarchical structure information strengthening and a final specific task. The model used is better than existing ones at learning the contextual representations of utterances embedded within its hierarchical semantic structure and thereby improves system performance. In addition, we created a mechanism using inductive grammar to dynamically prune the unpromising directions in the semantic structure parsing process. Finally, through experimentsOur code will be published when this paper is accepted. on the TOP and TOPv2 (low-resource setting) datasets, we achieved state-of-the-art (SOTA) performance, confirming the effectiveness of our proposed model.

Relation extraction (RE) tasks show promising performance in extracting relations from two entities mentioned in sentences, given sufficient annotations available during training. Such annotations would be labor-intensive to obtain in practice. Existing work adopts data augmentation techniques to generate pseudo-annotated sentences beyond limited annotations. These techniques neither preserve the semantic consistency of the original sentences when rule-based augmentations are adopted, nor preserve the syntax structure of sentences when expressing relations using seq2seq models, resulting in less diverse augmentations. In this work, we propose a dedicated augmentation technique for relational texts, named GDA, which uses two complementary modules to preserve both semantic consistency and syntax structures. We adopt a generative formulation and design a multi-tasking solution to achieve synergies. Furthermore, GDA adopts entity hints as the prior knowledge of the generative model to augment diverse sentences. Experimental results in three datasets under a low-resource setting showed that GDA could bring 2.0% F1 improvements compared with no augmentation technique.

Web documents have become rich data resources in current era, and understanding their discourse structure will potentially benefit various downstream document processing applications. Unfortunately, current discourse analysis and document intelligence research mostly focus on either discourse structure of plain text or superficial visual structures in document, which cannot accurately describe discourse structure of highly free-styled and semi-structured web documents. To promote discourse studies on web documents, in this paper we introduced a benchmark – WebDP, orienting a new task named Web Document Discourse Parsing. Specifically, a web document discourse structure representation schema is proposed by extending classical discourse theories and adding special features to well represent discourse characteristics of web documents. Then, a manually annotated web document dataset – WEBDOCS is developed to facilitate the study of this parsing task. We compared current neural models on WEBDOCS and experimental results show that WebDP is feasible but also challenging for current models.

pdf abs
Tab-CoT: Zero-shot Tabular Chain of Thought
Jin Ziqi | Wei Lu

The chain-of-though (CoT) prompting methods were successful in various natural language processing (NLP) tasks thanks to their ability to unveil the underlying complex reasoning processes. Such reasoning processes typically exhibit highly structured steps. Recent efforts also started investigating methods to encourage more structured reasoning procedures to be captured (cite least to most).In this work, we propose Tab-CoT, a novel tabular-format CoT prompting method, which allows the complex reasoning process to be explicitly modeled in a highly structured manner. Despite its simplicity, we show that our approach is capable of performing reasoning across multiple dimensions (i.e., both rows and columns).We demonstrate our approach’s strong zero-shot and few-shot capabilities through extensive experiments on a range of reasoning tasks.

pdf abs
KNSE: A Knowledge-aware Natural Language Inference Framework for Dialogue Symptom Status Recognition
Wei Chen | Shiqi Wei | Zhongyu Wei | Xuanjing Huang

Symptom diagnosis in medical conversations aims to correctly extract both symptom entities and their status from the doctor-patient dialogue. In this paper, we propose a novel framework called KNSE for symptom status recognition (SSR), where the SSR is formulated as a natural language inference (NLI) task. For each mentioned symptom in a dialogue window, we first generate knowledge about the symptom and hypothesis about status of the symptom, to form a (premise, knowledge, hypothesis) triplet. The BERT model is then used to encode the triplet, which is further processed by modules including utterance aggregation, self-attention, cross-attention, and GRU to predict the symptom status. Benefiting from the NLI formalization, the proposed framework can encode more informative prior knowledge to better localize and track symptom status, which can effectively improve the performance of symptom status recognition. Preliminary experiments on Chinese medical dialogue datasets show that KNSE outperforms previous competitive baselines and has advantages in cross-disease and cross-symptom scenarios.

Using translation memories (TMs) as prompts is a promising approach to in-context learning of machine translation models. In this work, we take a step towards prompting large language models (LLMs) with TMs and making them better translators. We find that the ability of LLMs to “understand” prompts is indeed helpful for making better use of TMs. Experiments show that the results of a pre-trained LLM translator can be greatly improved by using high-quality TM-based prompts. These results are even comparable to those of the state-of-the-art NMT systems which have access to large-scale in-domain bilingual data and are well tuned on the downstream tasks.

pdf abs
Character Coreference Resolution in Movie Screenplays
Sabyasachee Baruah | Shrikanth Narayanan

Movie screenplays have a distinct narrative structure. It segments the story into scenes containing interleaving descriptions of actions, locations, and character dialogues.A typical screenplay spans several scenes and can include long-range dependencies between characters and events.A holistic document-level understanding of the screenplay requires several natural language processing capabilities, such as parsing, character identification, coreference resolution, action recognition, summarization, and attribute discovery. In this work, we develop scalable and robust methods to extract the structural information and character coreference clusters from full-length movie screenplays. We curate two datasets for screenplay parsing and character coreference — MovieParse and MovieCoref, respectively.We build a robust screenplay parser to handle inconsistencies in screenplay formatting and leverage the parsed output to link co-referring character mentions.Our coreference models can scale to long screenplay documents without drastically increasing their memory footprints.

Most existing event causality identification (ECI) methods rarely consider the event causal label information and the interaction information between event pairs. In this paper, we propose a framework to enrich the representation of event pairs by introducing the event causal label information and the event pair interaction information. In particular, 1) we design an event-causal-label-aware module to model the event causal label information, in which we design the event causal label prediction task as an auxiliary task of ECI, aiming to predict which events are involved in the causal relationship (we call them causality-related events) by mining the dependencies between events. 2) We further design an event pair interaction graph module to model the interaction information between event pairs, in which we construct the interaction graph with event pairs as nodes and leverage graph attention mechanism to model the degree of dependency between event pairs. The experimental results show that our approach outperforms previous state-of-the-art methods on two benchmark datasets EventStoryLine and Causal-TimeBank.

pdf abs
LightFormer: Light-weight Transformer Using SVD-based Weight Transfer and Parameter Sharing
Xiuqing Lv | Peng Zhang | Sunzhu Li | Guobing Gan | Yueheng Sun

Transformer has become an important technique for natural language processing tasks with great success. However, it usually requires huge storage space and computational cost, making it difficult to be deployed on resource-constrained edge devices. To compress and accelerate Transformer, we propose LightFormer, which adopts a low-rank factorization initialized by SVD-based weight transfer and parameter sharing. The SVD-based weight transfer can effectively utilize the well-trained Transformer parameter knowledge to speed up the model convergence, and effectively alleviate the low-rank bottleneck problem combined with parameter sharing. We validate our method on machine translation, text summarization and text classification tasks. Experiments show that on IWSLT’14 De-En and WMT’14 En-De, LightFormer achieves similar performance to the baseline Transformer with 3.8 times and 1.8 times fewer parameters, and achieves 2.3 times speedup and 1.5 times speedup respectively, generally outperforming recent light-weight Transformers.

pdf abs
Multi-hop Evidence Retrieval for Cross-document Relation Extraction
Keming Lu | I-Hung Hsu | Wenxuan Zhou | Mingyu Derek Ma | Muhao Chen

Relation Extraction (RE) has been extended to cross-document scenarios because many relations are not simply described in a single document. This inevitably brings the challenge of efficient open-space evidence retrieval to support the inference of cross-document relations,along with the challenge of multi-hop reasoning on top of entities and evidence scattered in an open set of documents. To combat these challenges, we propose Mr.Cod (Multi-hop evidence retrieval for Cross-document relation extraction), which is a multi-hop evidence retrieval method based on evidence path mining and ranking. We explore multiple variants of retrievers to show evidence retrieval is essential in cross-document RE.We also propose a contextual dense retriever for this setting. Experiments on CodRED show that evidence retrieval with Mr.Cod effectively acquires cross-document evidence and boosts end-to-end RE performance in both closed and open settings.

pdf abs
Which Examples Should be Multiply Annotated? Active Learning When Annotators May Disagree
Connor Baumler | Anna Sotnikova | Hal Daumé III

Linguistic annotations, especially for controversial topics like hate speech detection, are frequently contested due to annotator backgrounds and positionalities. In such situations, preserving this disagreement through the machine learning pipeline can be important for downstream use cases. However, capturing disagreement can increase annotation time and expense. Fortunately, for many tasks, not all examples are equally controversial; we develop an active learning approach, Disagreement Aware Active Learning (DAAL) that concentrates annotations on examples where model entropy and annotator entropy are the most different. Because we cannot know the true entropy of annotations on unlabeled examples, we estimate a model that predicts annotator entropy trained using very few multiply-labeled examples. We find that traditional uncertainty-based active learning underperforms simple passive learning on tasks with high levels of disagreement, but that our active learning approach is able to successfully improve on passive and active baselines, reducing the number of annotations required by at least 24% on average across several datasets.

pdf abs
PIP: Parse-Instructed Prefix for Syntactically Controlled Paraphrase Generation
Yixin Wan | Kuan-Hao Huang | Kai-Wei Chang

Syntactically controlled paraphrase generation requires language models to generate paraphrases for sentences according to specific syntactic structures. Existing fine-tuning methods on this task is costly, as all parameters of the model need to be updated during the training process. Inspired by recent studies on parameter-efficient learning, we propose Parse-Instructed Prefix (PIP), a novel adaptation of prefix-tuning to tune large pre-trained language models on syntactically controlled paraphrase generation task in a low-data setting with significantly less training cost. We introduce two methods to instruct a model’s encoder prefix to capture syntax-related knowledge: direct initiation (PIP-Direct) and indirect optimization (PIP-Indirect). Comparing to traditional fine-tuning methods for this task, PIP is a compute-efficient alternative with 10 times less learnable parameters. Comparing to existing prefix-tuning methods, PIP excels at capturing syntax control information, achieving significantly higher performance at the same level of learnable parameter count.

Visual language such as charts and plots is ubiquitous in the human world. Comprehending plots and charts requires strong reasoning skills. Prior state-of-the-art (SOTA) models require at least tens of thousands of training examples and their reasoning capabilities are still much limited, especially on complex human-written queries. This paper presents the first one-shot solution to visual language reasoning. We decompose the challenge of visual language reasoning into two steps: (1) plot-to-text translation, and (2) reasoning over the translated text. The key in this method is a modality conversion module, named as DePlot, which translates the image of a plot or chart to a linearized table. The output of DePlot can then be directly used to prompt a pretrained large language model (LLM), exploiting the few-shot reasoning capabilities of LLMs. To obtain DePlot, we standardize the plot-to-table task by establishing unified task formats and metrics, and train DePlot end-to-end on this task. DePlot can then be used off-the-shelf together with LLMs in a plug-and-play fashion. Compared with a SOTA model finetuned on more than thousands of data points, DePlot+LLM with just one-shot prompting achieves a 29.4% improvement over finetuned SOTA on human-written queries from the task of chart QA.

Parameter-efficient tuning methods (PETs) have achieved promising results in tuning large pre-trained language models (PLMs). By formalizing frozen PLMs and additional tunable parameters as systems and controls respectively, PETs can be theoretically grounded to optimal control and further viewed as optimizing the terminal cost and running cost in the optimal control literature. Despite the elegance of this theoretical grounding, in practice, existing PETs often ignore the running cost and only optimize the terminal cost, i.e., focus on optimizing the loss function of the output state, regardless of the running cost that depends on the intermediate states. Since it is non-trivial to directly model the intermediate states and design a running cost function, we propose to use latent stochastic bridges to regularize the intermediate states and use the regularization as the running cost of PETs. As the first work to propose regularized PETs that use stochastic bridges as the regularizers (running costs) for the intermediate states, we show the effectiveness and generality of this regularization across different tasks, PLMs and PETs. In view of the great potential and capacity, we believe more sophisticated regularizers can be designed for PETs and better performance can be achieved in the future.

pdf abs
Learning from a Friend: Improving Event Extraction via Self-Training with Feedback from Abstract Meaning Representation
Zhiyang Xu | Jay Yoon Lee | Lifu Huang

Data scarcity has been the main factor that hinders the progress of event extraction. To overcome this issue, we propose a Self-Training with Feedback (STF) framework that leverages the large-scale unlabeled data and acquires feedback for each new event prediction from the unlabeled data by comparing it to the Abstract Meaning Representation (AMR) graph of the same sentence. Specifically, STF consists of (1) a base event extraction model trained on existing event annotations and then applied to large-scale unlabeled corpora to predict new event mentions as pseudo training samples, and (2) a novel scoring model that takes in each new predicted event trigger, an argument, its argument role, as well as their paths in the AMR graph to estimate a compatibility score indicating the correctness of the pseudo label. The compatibility scores further act as feedback to encourage or discourage the model learning on the pseudo labels during self-training. Experimental results on three benchmark datasets, including ACE05-E, ACE05-E+, and ERE, demonstrate the effectiveness of the STF framework on event extraction, especially event argument extraction, with significant performance gain over the base event extraction models and strong baselines. Our experimental analysis further shows that STF is a generic framework as it can be applied to improve most, if not all, event extraction models by leveraging large-scale unlabeled data, even when high-quality AMR graph annotations are not available.

pdf abs
How Well Do Large Language Models Perform on Faux Pas Tests?
Natalie Shapira | Guy Zwirn | Yoav Goldberg

Motivated by the question of the extent to which large language models “understand” social intelligence, we investigate the ability of such models to generate correct responses to questions involving descriptions of faux pas situations. The faux pas test is a test used in clinical psychology, which is known to be more challenging for children than individual tests of theory-of-mind or social intelligence. Our results demonstrate that, while the models seem to sometimes offer correct responses, they in fact struggle with this task, and that many of the seemingly correct responses can be attributed to over-interpretation by the human reader (“the ELIZA effect”). An additional phenomenon observed is the failure of most models to generate a correct response to presupposition questions. Finally, in an experiment in which the models are tasked with generating original faux pas stories, we find that while some models are capable of generating novel faux pas stories, the stories are all explicit, as the models are limited in their abilities to describe situations in an implicit manner.

pdf abs
Modular Transformers: Compressing Transformers into Modularized Layers for Flexible Efficient Inference
Wangchunshu Zhou | Ronan Le Bras | Yejin Choi

Pre-trained Transformer models like T5 and BART have advanced the state of the art on a wide range of text generation tasks. Compressing these models into smaller ones has become critically important for practical use. Common neural network compression techniques such as knowledge distillation or quantization are limited to static compression where the compression ratio is fixed. In this paper, we introduce Modular Transformers, a modularized encoder-decoder framework for flexible sequence-to-sequence model compression. Modular Transformers trains modularized layers that have the same function of two or more consecutive layers in the original model via module replacing and knowledge distillation. After training, the modularized layers can be flexibly assembled into sequence-to-sequence models that meet different performance-efficiency trade-offs. Experimental results show that after a single training phase, by simply varying the assemble strategy, Modular Transformers can achieve flexible compression ratios from 1.1x to 6x with little to moderate relative performance drop.

pdf abs
ISLTranslate: Dataset for Translating Indian Sign Language
Abhinav Joshi | Susmit Agrawal | Ashutosh Modi

Sign languages are the primary means of communication for many hard-of-hearing people worldwide. Recently, to bridge the communication gap between the hard-of-hearing community and the rest of the population, several sign language translation datasets have been proposed to enable the development of statistical sign language translation systems. However, there is a dearth of sign language resources for the Indian sign language. This resource paper introduces ISLTranslate, a translation dataset for continuous Indian Sign Language (ISL) consisting of 31k ISL-English sentence/phrase pairs. To the best of our knowledge, it is the largest translation dataset for continuous Indian Sign Language. We provide a detailed analysis of the dataset. To validate the performance of existing end-to-end Sign language to spoken language translation systems, we benchmark the created dataset with a transformer-based model for ISL translation.

pdf abs
LMentry: A Language Model Benchmark of Elementary Language Tasks
Avia Efrat | Or Honovich | Omer Levy

As the performance of large language models rapidly improves, benchmarks are getting larger and more complex as well. We present LMentry, a benchmark that avoids this “arms race” by focusing on a compact set of tasks that are trivial to humans, e.g. writing a sentence containing a specific word, identifying which words in a list belong to a specific category, or choosing which of two words is longer.LMentry is specifically designed to provide quick and interpretable insights into the capabilities and robustness of large language models. Our experiments reveal a wide variety of failure cases that, while immediately obvious to humans, pose a considerable challenge for large language models, including OpenAI’s latest 175B-parameter instruction-tuned model, TextDavinci002.LMentry complements contemporary evaluation approaches of large language models, providing a quick, automatic, and easy-to-run “unit test”, without resorting to large benchmark suites of complex tasks.

pdf abs
Differentiable Instruction Optimization for Cross-Task Generalization
Masaru Isonuma | Junichiro Mori | Ichiro Sakata

Instruction tuning has been attracting much attention to achieve generalization ability across a wide variety of tasks. Although various types of instructions have been manually created for instruction tuning, it is still unclear what kind of instruction is optimal to obtain cross-task generalization ability. This work presents instruction optimization, which optimizes training instructions with respect to generalization ability. Rather than manually tuning instructions, we introduce learnable instructions and optimize them with gradient descent by leveraging bilevel optimization. Experimental results show that the learned instruction enhances the diversity of instructions and improves the generalization ability compared to using only manually created instructions.

pdf abs
Leveraging Training Data in Few-Shot Prompting for Numerical Reasoning
Zhanming Jie | Wei Lu

Chain-of-thought (CoT) prompting with large language models has proven effective in numerous natural language process tasks, but designing prompts that generalize well to diverse problem types can be challenging CITATION, especially in the context of math word problem solving. Additionally, it is common to have a large amount of training data that have a better diversity coverage but CoT annotations are not available, which limits the use of supervised learning techniques. To address these issues, we investigate two approaches to leverage the training data in few-shot prompting scenario: dynamic program prompting and program distillation.Our approach is largely inspired by CITATION where they proposed to replace the CoT with the programs as the intermediate reasoning step. Such a prompting strategy allows us to accurately verify the answer correctness through program execution in MWP solving.Our dynamic program prompting involves annotating the training data by sampling correct programs from a large language model, while program distillation involves adapting a smaller model to the program-annotated training data.Our experiments on three standard MWP datasets demonstrate the effectiveness of these approaches, yielding significant improvements over previous baselines for prompting and fine-tuning.Our results suggest that leveraging a large amount of training data can improve the generalization ability of prompts and boost the performance of fine-tuned smaller models in MWP solving.

pdf abs
How does the task complexity of masked pretraining objectives affect downstream performance?
Atsuki Yamaguchi | Hiroaki Ozaki | Terufumi Morishita | Gaku Morio | Yasuhiro Sogawa

Masked language modeling (MLM) is a widely used self-supervised pretraining objective, where a model needs to predict an original token that is replaced with a mask given contexts. Although simpler and computationally efficient pretraining objectives, e.g., predicting the first character of a masked token, have recently shown comparable results to MLM, no objectives with a masking scheme actually outperform it in downstream tasks. Motivated by the assumption that their lack of complexity plays a vital role in the degradation, we validate whether more complex masked objectives can achieve better results and investigate how much complexity they should have to perform comparably to MLM. Our results using GLUE, SQuAD, and Universal Dependencies benchmarks demonstrate that more complicated objectives tend to show better downstream results with at least half of the MLM complexity needed to perform comparably to MLM. Finally, we discuss how we should pretrain a model using a masked objective from the task complexity perspective.

High-quality data is essential for conversational recommendation systems and serves as the cornerstone of the network architecture development and training strategy design. Existing works contribute heavy human efforts to manually labeling or designing and extending recommender dialogue templates. However, they suffer from: (i) the limited number of human annotators results in datasets can hardly capture rich and large-scale cases in the real world, (ii) the limited experience and knowledge of annotators accounts for the uninformative corpus and inappropriate recommendations. In this paper, we propose a novel automatic dataset synthesis approach that can generate large-scale and high-quality recommendation dialogues through a data2text generation process, where unstructured recommendation conversations are generated from structured graphs based on user-item information from the real world. In doing so, we comprehensively exploit: (i) rich personalized user profiles from traditional recommendation datasets, (ii) rich external knowledge from knowledge graphs, and (iii) the conversation ability contained in human-to-human conversational recommendation datasets. Extensive experiments validate the benefit brought by the automatically synthesized data under low-resource scenarios, and demonstrate the promising potential to facilitate developing a more effective conversational recommendation system.

The machine reading comprehension (MRC) of user manuals has huge potential in customer service. However, current methods have trouble answering complex questions. Therefore, we introduce the knowing-how & knowing-that task that requires the model to answer factoid-style, procedure-style, and inconsistent questions about user manuals. We resolve this task by jointly representing the sTeps and fActs in a gRAh (TARA), which supports a unified inference of various questions. Towards a systematical benchmarking study, we design a heuristic method to automatically parse user manuals into TARAs and build an annotated dataset to test the model’s ability in answering real-world questions. Empirical results demonstrate that representing user manuals as TARAs is a desired solution for the MRC of user manuals. An in-depth investigation of TARA further sheds light on the issues and broader impacts of future representations of user manuals. We hope our work can move the MRC of user manuals to a more complex and realistic stage.

pdf abs
Deep Span Representations for Named Entity Recognition
Enwei Zhu | Yiyang Liu | Jinpeng Li

Span-based models are one of the most straightforward methods for named entity recognition (NER). Existing span-based NER systems shallowly aggregate the token representations to span representations. However, this typically results in significant ineffectiveness for long entities, a coupling between the representations of overlapping spans, and ultimately a performance degradation. In this study, we propose DSpERT (Deep Span Encoder Representations from Transformers), which comprises a standard Transformer and a span Transformer. The latter uses low-layered span representations as queries, and aggregates the token representations as keys and values, layer by layer from bottom to top. Thus, DSpERT produces span representations of deep semantics. With weight initialization from pretrained language models, DSpERT achieves performance higher than or competitive with recent state-of-the-art systems on six NER benchmarks. Experimental results verify the importance of the depth for span representations, and show that DSpERT performs particularly well on long-span entities and nested structures. Further, the deep span representations are well structured and easily separable in the feature space.

Lexically constrained neural machine translation (LCNMT), which controls the translation generation with pre-specified constraints, is important in many practical applications. Current approaches to LCNMT typically assume that the pre-specified lexicon constraints are contextually appropriate. This assumption limits their application to real-world scenarios where a source lexicon may have multiple target constraints, and disambiguation is needed to select the most suitable one. In this paper, we propose disambiguated LCNMT (D-LCNMT) to solve the problem. D-LCNMT is a robust and effective two-stage framework that disambiguates the constraints based on contexts at first, then integrates the disambiguated constraints into LCNMT. Experimental results show that our approach outperforms strong baselines including existing data argumentation based approaches on benchmark datasets, and comprehensive experiments in scenarios where a source lexicon corresponds to multiple target constraints demonstrate the constraint disambiguation superiority of our approach.

pdf abs
Curating Datasets for Better Performance with Example Training Dynamics
Aviad Sar-Shalom | Roy Schwartz

The landscape of NLP research is dominated by large-scale models training on colossal datasets, relying on data quantity rather than quality. As an alternative to this landscape, we propose a method for weighing the relative importance of examples in a dataset based on their Example Training dynamics (swayamdipta et al., 2020) — a set of metrics computed during training. We propose a new way of computing the ETD of a dataset, and show that they can be used to improve performance in both in-distribution and out-of-distribution testing. We show that ETD can be transferable, i.e., they can be computed once and used for training different models, effectively reducing their computation cost. Finally, we suggest an active learning approach for computing ETD during training rather than as a preprocessing step — an approach that is not as effective, but dramatically reduces the extra computational costs.

pdf abs
Multi-armed bandits for resource efficient, online optimization of language model pre-training: the use case of dynamic masking
Inigo Urteaga | Moulay Zaidane Draidia | Tomer Lancewicki | Shahram Khadivi

We design and evaluate a Bayesian optimization framework for resource efficient pre-training of Transformer-based language models (TLMs). TLM pre-training requires high computational resources and introduces many unresolved design choices, such as selecting its pre-training hyperparameters.We propose a multi-armed bandit framework for the sequential selection of pre-training hyperparameters, aimed at optimizing language model performance, in a resource efficient manner. We design a Thompson sampling algorithm, with a surrogate Gaussian process reward model of the Masked Language Model (MLM) pre-training objective, for its sequential minimization. Instead of MLM pre-training with fixed masking probabilities, the proposed Gaussian process-based Thompson sampling (GP-TS) accelerates pre-training by sequentially selecting masking hyperparameters that improve performance. We empirically demonstrate how GP-TS pre-trains language models efficiently, i.e., it achieves lower MLM loss in fewer epochs, across a variety of settings. In addition, GP-TS pre-trained TLMs attain competitive downstream performance, while avoiding expensive hyperparameter grid search. GP-TS provides an interactive framework for efficient and optimized TLM pre-training that, by circumventing costly hyperparameter selection, enables substantial computational savings.

Software engineers working with the same programming language (PL) may speak different natural languages (NLs) and vice versa, erecting huge barriers to communication and working efficiency. Recent studies have demonstrated the effectiveness of generative pre-training in computer programs, yet they are always English-centric. In this work, we step towards bridging the gap between multilingual NLs and multilingual PLs for large language models (LLMs). We release ERNIE-Code, a unified pre-trained language model for 116 NLs and 6 PLs. We employ two methods for universal cross-lingual pre-training: span-corruption language modeling that learns patterns from monolingual NL or PL; and pivot-based translation language modeling that relies on parallel data of many NLs and PLs. Extensive results show that ERNIE-Code outperforms previous multilingual LLMs for PL or NL across a wide range of end tasks of code intelligence, including multilingual code-to-text, text-to-code, code-to-code, and text-to-text generation. We further show its advantage of zero-shot prompting on multilingual code summarization and text-to-text translation. We release our code and pre-trained checkpoints.

pdf abs
PromptAttack: Probing Dialogue State Trackers with Adversarial Prompts
Xiangjue Dong | Yun He | Ziwei Zhu | James Caverlee

A key component of modern conversational systems is the Dialogue State Tracker (or DST), which models a user’s goals and needs. Toward building more robust and reliable DSTs, we introduce a prompt-based learning approach to automatically generate effective adversarial examples to probe DST models. Two key characteristics of this approach are: (i) it only needs the output of the DST with no need for model parameters, and (ii) it can learn to generate natural language utterances that can target any DST. Through experiments over state-of-the-art DSTs, the proposed framework leads to the greatest reduction in accuracy and the best attack success rate while maintaining good fluency and a low perturbation ratio. We also show how much the generated adversarial examples can bolster a DST through adversarial training. These results indicate the strength of prompt-based attacks on DSTs and leave open avenues for continued refinement.

pdf abs
Understanding Programs by Exploiting (Fuzzing) Test Cases
Jianyu Zhao | Yuyang Rong | Yiwen Guo | Yifeng He | Hao Chen

Semantic understanding of programs has attracted great attention in the community. Inspired by recent successes of large language models (LLMs) in natural language understanding, tremendous progress has been made by treating programming language as another sort of natural language and training LLMs on corpora of program code. However, programs are essentially different from texts after all, in a sense that they are normally heavily structured and syntax-strict. In particular, programs and their basic units (i.e., functions and subroutines) are designed to demonstrate a variety of behaviors and/or provide possible outputs, given different inputs. The relationship between inputs and possible outputs/behaviors represents the functions/subroutines and profiles the program as a whole. Hence, we propose to incorporate such a relationship into learning, for achieving a deeper semantic understanding of programs. To obtain inputs that are representative enough to trigger the execution of most part of the code, we resort to fuzz testing and propose fuzz tuning to boost the performance of program understanding and code representation learning, given a pre-trained LLM. The effectiveness of the proposed method is verified on two program understanding tasks including code clone detection and code classification, and it outperforms current state-of-the-arts by large margins. Code is available at https://github.com/rabbitjy/FuzzTuning.

Retrieval accuracy is crucial to the performance of open-domain question answering (ODQA) systems. Recent work has demonstrated that dense hierarchical retrieval (DHR), which retrieves document candidates first and then relevant passages from the refined document set, can significantly outperform the single stage dense passage retriever (DPR). While effective, this approach requires document structure information to learn document representation and is hard to adopt to other domains without this information. Additionally, the dense retrievers tend to generalize poorly on out-of-domain data comparing with sparse retrievers such as BM25. In this paper, we propose Hybrid Hierarchical Retrieval (HHR) to address the existing limitations. Instead of relying solely on dense retrievers, we can apply sparse retriever, dense retriever, and a combination of them in both stages of document and passage retrieval. We perform extensive experiments on ODQA benchmarks and observe that our framework not only brings in-domain gains, but also generalizes better to zero-shot TriviaQA and Web Questions datasets with an average of 4.69% improvement on recall@100 over DHR. We also offer practical insights to trade off between retrieval accuracy, latency, and storage cost. The code is available on github.

pdf abs
Coherent or Not? Stressing a Neural Language Model for Discourse Coherence in Multiple Languages
Dominique Brunato | Felice Dell’Orletta | Irene Dini | Andrea Amelio Ravelli

In this study, we investigate the capability of a Neural Language Model (NLM) to distinguish between coherent and incoherent text, where the latter has been artificially created to gradually undermine local coherence within text. While previous research on coherence assessment using NLMs has primarily focused on English, we extend our investigation to multiple languages. We employ a consistent evaluation framework to compare the performance of monolingual and multilingual models in both in-domain and out-domain settings. Additionally, we explore the model’s performance in a cross-language scenario.

pdf abs
Understanding Differential Search Index for Text Retrieval
Xiaoyang Chen | Yanjiang Liu | Ben He | Le Sun | Yingfei Sun

The Differentiable Search Index (DSI) is a novel information retrieval (IR) framework that utilizes a differentiable function to generate a sorted list of document identifiers in response to a given query. However, due to the black-box nature of the end-to-end neural architecture, it remains to be understood to what extent DSI possesses the basic indexing and retrieval abilities. To mitigate this gap, in this study, we define and examine three important abilities that a functioning IR framework should possess, namely, exclusivity, completeness, and relevance ordering. Our analytical experimentation shows that while DSI demonstrates proficiency in memorizing the unidirectional mapping from pseudo queries to document identifiers, it falls short in distinguishing relevant documents from random ones, thereby negatively impacting its retrieval effectiveness. To address this issue, we propose a multi-task distillation approach to enhance the retrieval quality without altering the structure of the model and successfully endow it with improved indexing abilities. Through experiments conducted on various datasets, we demonstrate that our proposed method outperforms previous DSI baselinesThe code and data for this work can be found at https://github.com/VerdureChen/Understang_DSI.

Masked Language Models (MLMs) have proven to be effective for second-pass rescoring in Automatic Speech Recognition (ASR) systems. In this work, we propose Masked Audio Text Encoder (MATE), a multi-modal masked language model rescorer which incorporates acoustic representations into the input space of MLM. We adopt contrastive learning for effectively aligning the modalities by learning shared representations. We show that using a multi-modal rescorer is beneficial for domain generalization of the ASR system when target domain data is unavailable. MATE reduces word error rate (WER) by 4%-16% on in-domain, and 3%-7% on out-of-domain datasets, over the text-only baseline. Additionally, with very limited amount of training data (0.8 hours) MATE achieves a WER reduction of 8%-23% over the first-pass baseline.

pdf abs
Replace and Report: NLP Assisted Radiology Report Generation
Kaveri Kale | Pushpak Bhattacharyya | Kshitij Jadhav

Clinical practice frequently uses medical imaging for diagnosis and treatment. A significant challenge for automatic radiology report generation is that the radiology reports are long narratives consisting of multiple sentences for both abnormal and normal findings. Therefore, applying conventional image captioning approaches to generate the whole report proves to be insufficient, as these are designed to briefly describe images with short sentences. We propose a template-based approach to generate radiology reports from radiographs. Our approach involves the following: i) using a multilabel image classifier, produce the tags for the input radiograph; ii) using a transformer-based model, generate pathological descriptions (a description of abnormal findings seen on radiographs) from the tags generated in step (i); iii) using a BERT-based multi-label text classifier, find the spans in the normal report template to replace with the generated pathological descriptions; and iv) using a rule-based system, replace the identified span with the generated pathological description. We performed experiments with the two most popular radiology report datasets, IU Chest X-ray and MIMIC-CXR and demonstrated that the BLEU-1, ROUGE-L, METEOR, and CIDEr scores are better than the State-of-the-Art models by 25%, 36%, 44% and 48% respectively, on the IU X-RAY dataset. To the best of our knowledge, this is the first attempt to generate chest X-ray radiology reports by first creating small sentences for abnormal findings and then replacing them in the normal report template.

pdf abs
Pre-trained Personalized Review Summarization with Effective Salience Estimation
Hongyan Xu | Hongtao Liu | Zhepeng Lv | Qing Yang | Wenjun Wang

Personalized review summarization in recommender systems is a challenging task of generating condensed summaries for product reviews while preserving the salient content of reviews. Recently, Pretrained Language Models (PLMs) have become a new paradigm in text generation for the strong ability of natural language comprehension. However, it is nontrivial to apply PLMs in personalized review summarization directly since there are rich personalized information (e.g., user preferences and product characteristics) to be considered, which is crucial to the salience estimation of input review. In this paper, we propose a pre-trained personalized review summarization method, which aims to effectively incorporate the personalized information of users and products into the salience estimation of the input reviews. We design a personalized encoder that could identify the salient contents of the input sequence by jointly considering the semantic and personalized information respectively (i.e., ratings, user and product IDs, and linguistic features), yielding personalized representations for the input reviews and history summaries separately. Moreover, we design an interactive information selection mechanism that further identifies the salient contents of the input reviews and selects relative information from the history summaries. The results on real-world datasets show that our method performs better than the state-of-the-art baselines and could generate more readable summaries.

Hallucination is a known issue for neural abstractive summarization models. Recent work suggests that the degree of hallucination may depend on factual errors in the training data. In this work, we propose a new method called Contrastive Parameter Ensembling (CaPE) to use training data more effectively, utilizing variations in noise in training samples to reduce hallucination. Starting with a base model fine-tuned on an entire dataset, we additionally train expert and anti-expert models on clean and noisy subsets of the data, respectively. We then adjust the parameters of the base model by adding (subtracting) the parameters of the expert (anti-expert), advancing the recent work on additive parameter ensembling approaches. Trained on a much smaller data subset, expert and anti-expert models only fractionally (<14%) increases the total training time. Further, CaPE uses parameter ensembling and does not increase the inference time. Experimental results show that CaPE improves performance across different automatic factual metrics and human evaluation, with a maximum improvement of 16.69% and 15.38% on summary-level dependency-arc entailment accuracy for the XSUM and CNN/DM datasets. The CaPE model performs comparably to the base model on metrics of informativeness such as ROUGE.

pdf abs
OpineSum: Entailment-based self-training for abstractive opinion summarization
Annie Louis | Joshua Maynez

A typical product or place often has hundreds of reviews, and summarization of these texts is an important and challenging problem. Recent progress on abstractive summarization in domains such as news has been driven by supervised systems trained on hundreds of thousands of news articles paired with human-written summaries. However for opinion texts, such large scale datasets are rarely available. Unsupervised methods, self-training, and few-shot learning approaches bridge that gap. In this work, we present a novel self-training approach, OpineSum for abstractive opinion summarization. The self-training summaries in this approach are built automatically using a novel application of textual entailment and capture the consensus of opinions across the various reviews for an item. This method can be used to obtain silver-standard summaries on a large scale and train both unsupervised and few-shot abstractive summarization systems. OpineSum outperforms strong peer systems in both settings.

pdf abs
A Call for Standardization and Validation of Text Style Transfer Evaluation
Phil Ostheimer | Mayank Kumar Nagda | Marius Kloft | Sophie Fellenz

Text Style Transfer (TST) evaluation is, in practice, inconsistent. Therefore, we conduct a meta-analysis on human and automated TST evaluation and experimentation that thoroughly examines existing literature in the field. The meta-analysis reveals a substantial standardization gap in human and automated evaluation. In addition, we also find a validation gap: only few automated metrics have been validated using human experiments. To this end, we thoroughly scrutinize both the standardization and validation gap and reveal the resulting pitfalls. This work also paves the way to close the standardization and validation gap in TST evaluation by calling out requirements to be met by future research.

While Transformer has become the de-facto standard for speech, modeling upon the fine-grained frame-level features remains an open challenge of capturing long-distance dependencies and distributing the attention weights. We propose Progressive Down-Sampling (PDS) which gradually compresses the acoustic features into coarser-grained units containing more complete semantic information, like text-level representation. In addition, we develop a representation fusion method to alleviate information loss that occurs inevitably during high compression. In this way, we compress the acoustic features into 1/32 of the initial length while achieving better or comparable performances on the speech recognition task. And as a bonus, it yields inference speedups ranging from 1.20x to 1.47x.By reducing the modeling burden, we also achieve competitive results when training on the more challenging speech translation task.

Multi-modal sarcasm detection has attracted much recent attention. Nevertheless, the existing benchmark (MMSD) has some shortcomings that hinder the development of reliable multi-modal sarcasm detection system: (1) There are some spurious cues in MMSD, leading to the model bias learning; (2) The negative samples in MMSD are not always reasonable. To solve the aforementioned issues, we introduce MMSD2.0, a correction dataset that fixes the shortcomings of MMSD, by removing the spurious cues and re-annotating the unreasonable samples. Meanwhile, we present a novel framework called multi-view CLIP that is capable of leveraging multi-grained cues from multiple perspectives (i.e., text, image, and text-image interaction view) for multi-modal sarcasm detection. Extensive experiments show that MMSD2.0 is a valuable benchmark for building reliable multi-modal sarcasm detection systems and multi-view CLIP can significantly outperform the previous best baselines.

Entity linking models have achieved significant success via utilizing pretrained language models to capture semantic features. However, the NIL prediction problem, which aims to identify mentions without a corresponding entity in the knowledge base, has received insufficient attention. We categorize mentions linking to NIL into Missing Entity and Non-Entity Phrase, and propose an entity linking dataset NEL that focuses on the NIL prediction problem.NEL takes ambiguous entities as seeds, collects relevant mention context in the Wikipedia corpus, and ensures the presence of mentions linking to NIL by human annotation and entity masking. We conduct a series of experiments with the widely used bi-encoder and cross-encoder entity linking models, results show that both types of NIL mentions in training data have a significant influence on the accuracy of NIL prediction. Our code and dataset can be accessed at https://github.com/solitaryzero/NIL_EL.

Text-based personality computing (TPC) has gained many research interests in NLP. In this paper, we describe 15 challenges that we consider deserving the attention of the NLP research community. These challenges are organized by the following topics: personality taxonomies, measurement quality, datasets, performance evaluation, modelling choices, as well as ethics and fairness. When addressing each challenge, not only do we combine perspectives from both NLP and social sciences, but also offer concrete suggestions. We hope to inspire more valid and reliable TPC research.

The increasing sizes of large generative Pre-trained Language Models (PLMs) hinder their deploymentin real-world applications. To obtain efficient PLMs, previous studies mostly focus on pruning the attention heads and feed-forward networks (FFNs) of the Transformer. Nevertheless, we find that in generative PLMs, the hidden dimension shared by many other modules (e.g., embedding layer and layer normalization) contains persistent outliers regardless of the network input. This study comprehensively investigates the structured pruning of generative PLMs with all the above compressible components. To identify redundant network structures, we assign learnable masks over compressible components followed by sparse training. Various sizes of PLMs can be flexibly extracted via different thresholds, and are then task-specifically fine-tuned for further improvement. Extensive experiments on language modeling, summarization and machine translation validate the effectiveness of the proposed method. For example, the pruned BART brings 1.51x/6.96x inference speedup on GPU/CPU with 67% size reduction, and can be further combined with quantization for more than 25× compression.

pdf abs
Prompt-Guided Retrieval Augmentation for Non-Knowledge-Intensive Tasks
Zhicheng Guo | Sijie Cheng | Yile Wang | Peng Li | Yang Liu

Retrieval-augmented methods have received increasing attention to support downstream tasks by leveraging useful information from external resources. Recent studies mainly focus on exploring retrieval to solve knowledge-intensive (KI) tasks. However, the potential of retrieval for most non-knowledge-intensive (NKI) tasks remains under-explored. There are two main challenges to leveraging retrieval-augmented methods for NKI tasks: 1) the demand for diverse relevance score functions and 2) the dilemma between training cost and task performance. To address these challenges, we propose a two-stage framework for NKI tasks, named PGRA. In the first stage, we adopt a task-agnostic retriever to build a shared static index and select candidate evidence efficiently. In the second stage, we design a prompt-guided reranker to rerank the nearest evidence according to task-specific relevance for the reader. Experimental results show that PGRA outperforms other state-of-the-art retrieval-augmented methods. Our analyses further investigate the influence factors to model performance and demonstrate the generality of PGRA. The code and model will be released for further research.

pdf abs
Contextualized Semantic Distance between Highly Overlapped Texts
Letian Peng | Zuchao Li | Hai Zhao

Overlapping frequently occurs in paired texts in natural language processing tasks like text editing and semantic similarity evaluation. Better evaluation of the semantic distance between the overlapped sentences benefits the language system’s understanding and guides the generation. Since conventional semantic metrics are based on word representations, they are vulnerable to the disturbance of overlapped components with similar representations. This paper aims to address the issue with a mask-and-predict strategy. We take the words in the longest common sequence (LCS) as neighboring words and use masked language modeling (MLM) from pre-trained language models (PLMs) to predict the distributions in their positions. Our metric, Neighboring Distribution Divergence (NDD), represents the semantic distance by calculating the divergence between distributions in the overlapped parts. Experiments on Semantic Textual Similarity show NDD to be more sensitive to various semantic differences, especially on highly overlapped paired texts. Based on the discovery, we further implement an unsupervised and training-free method for text compression, leading to a significant improvement on the previous perplexity-based method. The high compression rate controlling ability of our method even enables NDD to outperform the supervised state-of-the-art in domain adaption by a huge margin. Further experiments on syntax and semantics analyses verify the awareness of internal sentence structures, indicating the high potential of NDD for further studies.

Dense retrievers have achieved impressive performance, but their demand for abundant training data limits their application scenarios. Contrastive pre-training, which constructs pseudo-positive examples from unlabeled data, has shown great potential to solve this problem. However, the pseudo-positive examples crafted by data augmentations can be irrelevant. To this end, we propose relevance-aware contrastive learning. It takes the intermediate-trained model itself as an imperfect oracle to estimate the relevance of positive pairs and adaptively weighs the contrastive loss of different pairs according to the estimated relevance. Our method consistently improves the SOTA unsupervised Contriever model on the BEIR and open-domain QA retrieval benchmarks. Further exploration shows that our method can not only beat BM25 after further pre-training on the target corpus but also serves as a good few-shot learner. Our code is publicly available at https://github.com/Yibin-Lei/ReContriever.

pdf abs
Verifying Annotation Agreement without Multiple Experts: A Case Study with Gujarati SNACS
Maitrey Mehta | Vivek Srikumar

Good datasets are a foundation of NLP research, and form the basis for training and evaluating models of language use. While creating datasets, the standard practice is to verify the annotation consistency using a committee of human annotators. This norm assumes that multiple annotators are available, which is not the case for highly specialized tasks or low-resource languages. In this paper, we ask: Can we evaluate the quality of a dataset constructed by a single human annotator? To address this question, we propose four weak verifiers to help estimate dataset quality, and outline when each may be employed. We instantiate these strategies for the task of semantic analysis of adpositions in Gujarati, a low-resource language, and show that our weak verifiers concur with a double-annotation study. As an added contribution, we also release the first dataset with semantic annotations in Gujarati along with several model baselines.

pdf abs
Reinforced Active Learning for Low-Resource, Domain-Specific, Multi-Label Text Classification
Lukas Wertz | Jasmina Bogojeska | Katsiaryna Mirylenka | Jonas Kuhn

Text classification datasets from specialised or technical domains are in high demand, especially in industrial applications. However, due to the high cost of annotation such datasets are usually expensive to create. While Active Learning (AL) can reduce the labeling cost, required AL strategies are often only tested on general knowledge domains and tend to use information sources that are not consistent across tasks. We propose Reinforced Active Learning (RAL) to train a Reinforcement Learning policy that utilizes many different aspects of the data and the task in order to select the most informative unlabeled subset dynamically over the course of the AL procedure. We demonstrate the superior performance of the proposed RAL framework compared to strong AL baselines across four intricate multi-class, multi-label text classification datasets taken from specialised domains. In addition, we experiment with a unique data augmentation approach to further reduce the number of samples RAL needs to annotate.

pdf abs
Improving Classroom Dialogue Act Recognition from Limited Labeled Data with Self-Supervised Contrastive Learning Classifiers
Vikram Kumaran | Jonathan Rowe | Bradford Mott | Snigdha Chaturvedi | James Lester

Recognizing classroom dialogue acts has significant promise for yielding insight into teaching, student learning, and classroom dynamics. However, obtaining K-12 classroom dialogue data with labels is a significant challenge, and therefore, developing data-efficient methods for classroom dialogue act recognition is essential. This work addresses the challenge of classroom dialogue act recognition from limited labeled data using a contrastive learning-based self-supervised approach (SSCon). SSCon uses two independent models that iteratively improve each other’s performance by increasing the accuracy of dialogue act recognition and minimizing the embedding distance between the same dialogue acts. We evaluate the approach on three complementary dialogue act recognition datasets: the TalkMoves dataset (annotated K-12 mathematics lesson transcripts), the DailyDialog dataset (multi-turn daily conversation dialogues), and the Dialogue State Tracking Challenge 2 (DSTC2) dataset (restaurant reservation dialogues). Results indicate that our self-supervised contrastive learning-based model outperforms competitive baseline models when trained with limited examples per dialogue act. Furthermore, SSCon outperforms other few-shot models that require considerably more labeled data.

Visual temporal-aligned translation aims to transform the visual sequence into natural words, including important applicable tasks such as lipreading and fingerspelling recognition. However, various performance habits of specific words by different speakers or signers can lead to visual ambiguity, which has become a major obstacle to the development of current methods. Considering the constraints above, the generalization ability of the translation system is supposed to be further explored through the evaluation results on unseen performers. In this paper, we develop a novel generalizable framework named Contrastive Token-Wise Meta-learning (CtoML), which strives to transfer recognition skills to unseen performers. To the best of our knowledge, employing meta-learning methods directly in the image domain poses two main challenges, and we propose corresponding strategies. First, sequence prediction in visual temporal-aligned translation, which aims to generate multiple words autoregressively, is different from the vanilla classification. Thus, we devise the token-wise diversity-aware weights for the meta-train stage, which encourages the model to make efforts on those ambiguously recognized tokens. Second, considering the consistency of word-visual prototypes across different domains, we develop two complementary global and local contrastive losses to maintain inter-class relationships and promote domain-independent. We conduct extensive experiments on the widely-used lipreading dataset GRID and the fingerspelling dataset ChicagoFSWild, and the experimental results show the effectiveness of our proposed CtoML over existing state-of-the-art methods.

pdf abs
Enhancing Cross-lingual Prompting with Dual Prompt Augmentation
Meng Zhou | Xin Li | Yue Jiang | Lidong Bing

Prompting shows promising results in few-shot scenarios. However, its strength for multilingual/cross-lingual problems has not been fully exploited. hao and Schütze (2021) made initial explorations in this direction by presenting that cross-lingual prompting outperforms cross-lingual finetuning. In this paper, we conduct an empirical exploration on the effect of each component in cross-lingual prompting and derive Universal Prompting, which helps alleviate the discrepancies between source-language training and target-language inference. Based on this, we propose DPA, a dual prompt augmentation framework, aiming at relieving the data scarcity issue in few-shot cross-lingual prompting. Notably, for XNLI, our method achieves 46.54% with only 16 English training examples per class, significantly better than 34.99% of fine-tuning. Our code is available at https://github.com/DAMO-NLP-SG/DPA.

pdf abs
Foveate, Attribute, and Rationalize: Towards Physically Safe and Trustworthy AI
Alex Mei | Sharon Levy | William Yang Wang

Users’ physical safety is an increasing concern as the market for intelligent systems continues to grow, where unconstrained systems may recommend users dangerous actions that can lead to serious injury. Covertly unsafe text is an area of particular interest, as such text may arise from everyday scenarios and are challenging to detect as harmful. We propose FARM, a novel framework leveraging external knowledge for trustworthy rationale generation in the context of safety. In particular, FARM foveates on missing knowledge to qualify the information required to reason in specific scenarios and retrieves this information with attribution to trustworthy sources. This knowledge is used to both classify the safety of the original text and generate human-interpretable rationales, shedding light on the risk of systems to specific user groups and helping both stakeholders manage the risks of their systems and policymakers to provide concrete safeguards for consumer safety. Our experiments show that FARM obtains state-of-the-art results on the SafeText dataset, showing absolute improvement in safety classification accuracy by 5.9%.

pdf abs
Multijugate Dual Learning for Low-Resource Task-Oriented Dialogue System
Shimin Li | Xiaotian Zhang | Yanjun Zheng | Linyang Li | Xipeng Qiu

Dialogue data in real scenarios tend to be sparsely available, rendering data-starved end-to-end dialogue systems trained inadequately. We discover that data utilization efficiency in low-resource scenarios can be enhanced by mining alignment information uncertain utterance and deterministic dialogue state. Therefore, we innovatively implement dual learning in task-oriented dialogues to exploit the correlation of heterogeneous data. In addition, the one-to-one duality is converted into a multijugate duality to reduce the influence of spurious correlations in dual training for generalization. Without introducing additional parameters, our method could be implemented in arbitrary networks. Extensive empirical analyses demonstrate that our proposed method improves the effectiveness of end-to-end task-oriented dialogue systems under multiple benchmarks and obtains state-of-the-art results in low-resource scenarios.

pdf abs
A Class-Rebalancing Self-Training Framework for Distantly-Supervised Named Entity Recognition
Qi Li | Tingyu Xie | Peng Peng | Hongwei Wang | Gaoang Wang

Distant supervision reduces the reliance on human annotation in the named entity recognition tasks. The class-level imbalanced distant annotation is a realistic and unexplored problem, and the popular method of self-training can not handle class-level imbalanced learning. More importantly, self-training is dominated by the high-performance class in selecting candidates, and deteriorates the low-performance class with the bias of generated pseudo label. To address the class-level imbalance performance, we propose a class-rebalancing self-training framework for improving the distantly-supervised named entity recognition. In candidate selection, a class-wise flexible threshold is designed to fully explore other classes besides the high-performance class. In label generation, injecting the distant label, a hybrid pseudo label is adopted to provide straight semantic information for the low-performance class. Experiments on five flat and two nested datasets show that our model achieves state-of-the-art results. We also conduct extensive research to analyze the effectiveness of the flexible threshold and the hybrid pseudo label.

pdf abs
MURMUR: Modular Multi-Step Reasoning for Semi-Structured Data-to-Text Generation
Swarnadeep Saha | Xinyan Yu | Mohit Bansal | Ramakanth Pasunuru | Asli Celikyilmaz

Prompting large language models has enabled significant recent progress in multi-step reasoning over text. However, when applied to text generation from semi-structured data (e.g., graphs or tables), these methods typically suffer from low semantic coverage, hallucination, and logical inconsistency. We propose MURMUR a neuro-symbolic modular approach to text generation from semi-structured data with multi-step reasoning. MURMUR is a best-first search method that generates reasoning paths using: (1) neural and symbolic modules with specific linguistic and logical skills, (2) a grammar whose production rules define valid compositions of modules, and (3) value functions that assess the quality of each reasoning step. We conduct experiments on two diverse data-to-text generation tasks like WebNLG and LogicNLG. The tasks differ in their data representations (graphs and tables) and span multiple linguistic and logical skills. MURMUR obtains significant improvements over recent few-shot baselines like direct prompting and chain-of-thought prompting, while also achieving comparable performance to fine-tuned GPT-2 on out-of-domain data. Moreover, human evaluation shows that MURMUR generates highly faithful and correct reasoning paths that lead to 26% more logically consistent summaries on LogicNLG, compared to direct prompting.

Solving math word problem (MWP) with AI techniques has recently made great progress with the success of deep neural networks (DNN), but it is far from being solved. We argue that the ability of learning by analogy is essential for an MWP solver to better understand same problems which may typically be formulated in diverse ways. However most existing works exploit the shortcut learning to train MWP solvers simply based on samples with a single question. In lack of diverse questions, these methods merely learn shallow heuristics. In this paper, we make a first attempt to solve MWPs by generating diverse yet consistent questions/equations. Given a typical MWP including the scenario description, question, and equation (i.e., answer), we first generate multiple consistent equations via a group of heuristic rules. We then feed them to a question generator together with the scenario to obtain the corresponding diverse questions, forming a new MWP with a variety of questions and equations. Finally we engage a data filter to remove those unreasonable MWPs, keeping the high-quality augmented ones. To evaluate the ability of learning by analogy for an MWP solver, we generate a new MWP dataset (called DiverseMath23K) with diverse questions by extending the current benchmark Math23K. Extensive experimental results demonstrate that our proposed method can generate high-quality diverse questions with corresponding equations, further leading to performance improvement on Diverse-Math23K. The code and dataset is available at: https://github.com/zhouzihao501/DiverseMWP.

pdf abs
Revisit Few-shot Intent Classification with PLMs: Direct Fine-tuning vs. Continual Pre-training
Haode Zhang | Haowen Liang | Li-Ming Zhan | Xiao-Ming Wu | Albert Y.S. Lam

We consider the task of few-shot intent detection, which involves training a deep learning model to classify utterances based on their underlying intents using only a small amount of labeled data. The current approach to address this problem is through continual pre-training, i.e., fine-tuning pre-trained language models (PLMs) on external resources (e.g., conversational corpora, public intent detection datasets, or natural language understanding datasets) before using them as utterance encoders for training an intent classifier. In this paper, we show that continual pre-training may not be essential, since the overfitting problem of PLMs on this task may not be as serious as expected. Specifically, we find that directly fine-tuning PLMs on only a handful of labeled examples already yields decent results compared to methods that employ continual pre-training, and the performance gap diminishes rapidly as the number of labeled data increases. To maximize the utilization of the limited available data, we propose a context augmentation method and leverage sequential self-distillation to boost performance. Comprehensive experiments on real-world benchmarks show that given only two or more labeled samples per class, direct fine-tuning outperforms many strong baselines that utilize external data sources for continual pre-training. The code can be found at https://github.com/hdzhang-code/DFTPlus.

pdf abs
Improving Contrastive Learning of Sentence Embeddings from AI Feedback
Qinyuan Cheng | Xiaogui Yang | Tianxiang Sun | Linyang Li | Xipeng Qiu

Contrastive learning has become a popular approach in natural language processing, particularly for the learning of sentence embeddings.However, the discrete nature of natural language makes it difficult to ensure the quality of positive and negative sample pairs generated through data augmentation methods. Although supervised contrastive learning can produce more accurate sample pairs with human feedback labels, it still lacks fine-grained training signals. In this paper, we propose to improve Contrastive Learning of sentence embeddings from AI Feedback (CLAIF).Our method utilizes AI feedback from large pre-trained language models (LLMs) to construct sample pairs with fine-grained sample similarity scores to improve contrastive learning. Besides, we combine human feedback and AI feedback to provide better supervision signals for supervised contrastive learning of sentence embeddings.Experimental results show that our method achieves state-of-the-art performance on several semantic textual similarity (STS) and transfer learning tasks compared to other unsupervised and supervised contrastive learning methods.

pdf abs
Mars: Modeling Context & State Representations with Contrastive Learning for End-to-End Task-Oriented Dialog
Haipeng Sun | Junwei Bao | Youzheng Wu | Xiaodong He

Traditional end-to-end task-oriented dialog systems first convert dialog context into belief state and action state before generating the system response. The system response performance is significantly affected by the quality of the belief state and action state. We first explore what dialog context representation is beneficial to improving the quality of the belief state and action state, which further enhances the generated response quality. To tackle our exploration, we propose Mars, an end-to-end task-oriented dialog system with two contrastive learning strategies to model the relationship between dialog context and belief/action state representations. Empirical results show dialog context representations, which are more different from semantic state representations, are more conducive to multi-turn task-oriented dialog. Moreover, our proposed Mars achieves state-of-the-art performance on the MultiWOZ 2.0, CamRest676, and CrossWOZ.

The mission of open knowledge graph (KG) completion is to draw new findings from known facts. Existing works that augment KG completion require either (1) factual triples to enlarge the graph reasoning space or (2) manually designed prompts to extract knowledge from a pre-trained language model (PLM), exhibiting limited performance and requiring expensive efforts from experts. To this end, we propose TagReal that automatically generates quality query prompts and retrieves support information from large text corpora to probe knowledge from PLM for KG completion. The results show that TagReal achieves state-of-the-art performance on two benchmark datasets. We find that TagReal has superb performance even with limited training data, outperforming existing embedding-based, graph-based, and PLM-based methods.

Automatic discourse processing is bottlenecked by data: current discourse formalisms pose highly demanding annotation tasks involving large taxonomies of discourse relations, making them inaccessible to lay annotators. This work instead adopts the linguistic framework of Questions Under Discussion (QUD) for discourse analysis and seeks to derive QUD structures automatically. QUD views each sentence as an answer to a question triggered in prior context; thus, we characterize relationships between sentences as free-form questions, in contrast to exhaustive fine-grained taxonomies. We develop the first-of-its-kind QUD parser that derives a dependency structure of questions over full documents, trained using a large, crowdsourced question-answering dataset DCQA (Ko et al., 2022). Human evaluation results show that QUD dependency parsing is possible for language models trained with this crowdsourced, generalizable annotation scheme. We illustrate how our QUD structure is distinct from RST trees, and demonstrate the utility of QUD analysis in the context of document simplification. Our findings show that QUD parsing is an appealing alternative for automatic discourse processing.

pdf abs
An Integrated Approach for Political Bias Prediction and Explanation Based on Discursive Structure
Nicolas Devatine | Philippe Muller | Chloé Braud

One crucial aspect of democracy is fair information sharing. While it is hard to prevent biases in news, they should be identified for better transparency. We propose an approach to automatically characterize biases that takes into account structural differences and that is efficient for long texts. This yields new ways to provide explanations for a textual classifier, going beyond mere lexical cues. We show that: (i) the use of discourse-based structure-aware document representations compare well to local, computationally heavy, or domain-specific models on classification tasks that deal with textual bias (ii) our approach based on different levels of granularity allows for the generation of better explanations of model decisions, both at the lexical and structural level, while addressing the challenge posed by long texts.

Enhancing word usage is a desired feature for writing assistance. To further advance research in this area, this paper introduces “Smart Word Suggestions” (SWS) task and benchmark. Unlike other works, SWS emphasizes end-to-end evaluation and presents a more realistic writing assistance scenario. This task involves identifying words or phrases that require improvement and providing substitution suggestions. The benchmark includes human-labeled data for testing, a large distantly supervised dataset for training, and the framework for evaluation. The test data includes 1,000 sentences written by English learners, accompanied by over 16,000 substitution suggestions annotated by 10 native speakers. The training dataset comprises over 3.7 million sentences and 12.7 million suggestions generated through rules. Our experiments with seven baselines demonstrate that SWS is a challenging task. Based on experimental analysis, we suggest potential directions for future research on SWS. The dataset and related codes will be available for research purposes.

Commonsense reasoning simulates the human ability to make presumptions about our physical world, and it is an essential cornerstone in building general AI systems. We proposea new commonsense reasoning dataset based on human’s Interactive Fiction (IF) gameplaywalkthroughs as human players demonstrate plentiful and diverse commonsense reasoning. The new dataset provides a natural mixture of various reasoning types and requires multi-hopreasoning. Moreover, the IF game-based construction procedure requires much less humaninterventions than previous ones. Different from existing benchmarks, our dataset focuseson the assessment of functional commonsense knowledge rules rather than factual knowledge. Hence, in order to achieve higher performance on our tasks, models need to effectively uti-lize such functional knowledge to infer the outcomes of actions, rather than relying solely onmemorizing facts. Experiments show that the introduced dataset is challenging to previousmachine reading models as well as the new large language models with a significant 20%performance gap compared to human experts.

Distillation from Weak Teacher (DWT) is a method of transferring knowledge from a smaller, weaker teacher model to a larger student model to improve its performance. Previous studies have shown that DWT can be effective in the vision domain and natural language processing (NLP) pre-training stage. Specifically, DWT shows promise in practical scenarios, such as enhancing new generation or larger models using pre-trained yet older or smaller models and lacking a resource budget. However, the optimal conditions for using DWT have yet to be fully investigated in NLP pre-training. Therefore, this study examines three key factors to optimize DWT, distinct from those used in the vision domain or traditional knowledge distillation. These factors are:(i) the impact of teacher model quality on DWT effectiveness, (ii) guidelines for adjusting the weighting value for DWT loss, and (iii) the impact of parameter remapping as a student model initialization technique for DWT.

Logical data-to-text generation is a representative task in measuring the capabilities of both language generation and complex reasoning. Despite the introduction of reasoning skills in generation, existing works still rely on neural language models to output the final table description. However, due to the inefficacy of neural language models in complex reasoning, these methods inevitably have difficulty working out key entities in the description and might produce unfaithful descriptions. To alleviate these issues, we propose a dependency-aware symbolic reasoning framework that reasons out each entity in the table description with our designed table-compatible programming language. To figure out the dependency relationship among entities, we devise an entity scheduling mechanism to determine the order of programme synthesis such that the reasoning of an entity only relies on other “resolved” entities. Experiments on three datasets and three backbones show that ours outperforms previous methods not only in surface-level fidelity but also in logical fidelity. Notably, the proposed framework enhances GPT-2, BART and T5 with an absolute improvement of 5.7%~11.5% on SP-Acc.

Event extraction aims to recognize pre-defined event triggers and arguments from texts, which suffer from the lack of high-quality annotations. In most NLP applications, involving a large scale of synthetic training data is a practical and effective approach to alleviate the problem of data scarcity. However, when applying to the task of event extraction, recent data augmentation methods often neglect the problem of grammatical incorrectness, structure misalignment, and semantic drifting, leading to unsatisfactory performances. In order to solve these problems, we propose a denoised structure-to-text augmentation framework for event extraction (DAEE), which generates additional training data through the knowledge-based structure-to-text generation model and selects the effective subset from the generated data iteratively with a deep reinforcement learning agent. Experimental results on several datasets demonstrate that the proposed method generates more diverse text representations for event extraction and achieves comparable results with the state-of-the-art.

Deep neural networks (DNNs) have been proven to be sensitive towards perturbations on input samples, and previous works highlight that adversarial samples are even more vulnerable than normal ones. In this work, this phenomenon is illustrated frWe first show that adversarial samples locate in steep and narrow local minima of the loss landscape (high sharpness) while normal samples, which differs distinctly from adversarial ones, reside in the loss surface that is more flatter (low sharpness).om the perspective of sharpness via visualizing the input loss landscape of models. Based on this, we propose a simple and effective sharpness-based detector to distinct adversarial samples by maximizing the loss increment within the region where the inference sample is located. Considering that the notion of sharpness of a loss landscape is relative, we further propose an adaptive optimization strategy in an attempt to fairly compare the relative sharpness among different samples. Experimental results show that our approach can outperform previous detection methods by large margins (average +6.6 F1 score) for four advanced attack strategies considered in this paper across three text classification tasks.

pdf abs
A Simple, Yet Effective Approach to Finding Biases in Code Generation
Spyridon Mouselinos | Mateusz Malinowski | Henryk Michalewski

Recently, high-performing code generation systems based on large language models have surfaced. They are trained on massive corpora containing much more natural text than actual executable computer code. This work shows that current code generation systems exhibit undesired biases inherited from their large language model backbones, which can reduce the quality of the generated code under specific circumstances. To investigate the effect, we propose the “block of influence” concept, which enables a modular decomposition and analysis of the coding challenges. We introduce an automated intervention mechanism reminiscent of adversarial testing that exposes undesired biases through the failure modes of the models under test. Finally, we demonstrate how our framework can be used as a data transformation technique during fine-tuning, acting as a mitigation strategy for these biases.

Membership Inference attacks (MIAs) aim to predict whether a data sample was present in the training data of a machine learning model or not, and are widely used for assessing the privacy risks of language models. Most existing attacks rely on the observation that models tend toassign higher probabilities to their training samples than non-training points. However, simple thresholding of the model score in isolation tends to lead to high false-positive rates as it does not account for the intrinsic complexity of a sample. Recent work has demonstrated that reference-based attacks which compare model scores to those obtained from a reference model trained on similar data can substantially improve the performance of MIAs.However, in order to train reference models, attacks of this kind make the strong and arguably unrealistic assumption that an adversary has access to samples closely resembling the original training data. Therefore, we investigate their performance in more realistic scenarios and find that they are highly fragile in relation to the data distribution used to train reference models. To investigate whether this fragility provides a layer of safety, we propose and evaluate neighbourhood attacks, which compare model scores for a given sample to scores of synthetically generated neighbour texts and therefore eliminate the need for access to the training data distribution. We show that, in addition to being competitive with reference-based attacks that have perfect knowledge about the training data distribution, our attack clearly outperforms existing reference-free attacks as well as reference-based attacks with imperfect knowledge, which demonstrates the need for a reevaluation of the threat model of adversarial attacks.

pdf abs
CFL: Causally Fair Language Models Through Token-level Attribute Controlled Generation
Rahul Madhavan | Rishabh Garg | Kahini Wadhawan | Sameep Mehta

We propose a method to control the attributes of Language Models (LMs) for the text generation task using Causal Average Treatment Effect (ATE) scores and counterfactual augmentation. We explore this method, in the context of LM detoxification, and propose the Causally Fair Language (CFL) architecture for detoxifying pre-trained LMs in a plug-and-play manner. Our architecture is based on a Structural Causal Model (SCM) that is mathematically transparent and computationally efficient as compared with many existing detoxification techniques. We also propose several new metrics that aim to better understand the behaviour of LMs in the context of toxic text generation. Further, we achieve state of the art performance for toxic degeneration, which are computed using Real Toxicity Prompts. Our experiments show that CFL achieves such a detoxification without much impact on the model perplexity. We also show that CFL mitigates the unintended bias problem through experiments on the BOLD dataset.

Diffusion models have been successfully adapted to text generation tasks by mapping the discrete text into the continuous space. However, there exist nonnegligible gaps between training and inference, owing to the absence of the forward process during inference. Thus, the model only predicts based on the previously generated reverse noise rather than the noise computed by the forward process. Besides, the widely-used downsampling strategy in speeding up the inference will cause the mismatch of diffusion trajectories between training and inference. To understand and mitigate the above two types of training-inference discrepancies, we launch a thorough preliminary study. Based on our observations, we propose two simple yet effective methods to bridge the gaps mentioned above, named Distance Penalty and Adaptive Decay Sampling. Extensive experiments on 6 generation tasks confirm the superiority of our methods, which can achieve 100× → 200× speedup with better performance. Our code will be released at https://github.com/CODINNLG/Bridge_Gap_Diffusion.

pdf abs
Topic-Guided Self-Introduction Generation for Social Media Users
Chunpu Xu | Jing Li | Piji Li | Min Yang

Millions of users are active on social media. To allow users to better showcase themselves and network with others, we explore the auto-generation of social media self-introduction, a short sentence outlining a user’s personal interests. While most prior work profiling users with tags (e.g., ages), we investigate sentence-level self-introductions to provide a more natural and engaging way for users to know each other. Here we exploit a user’s tweeting history to generate their self-introduction. The task is non-trivial because the history content may be lengthy, noisy, and exhibit various personal interests. To address this challenge, we propose a novel unified topic-guided encoder-decoder (UTGED) framework; it models latent topics to reflect salient user interest, whose topic mixture then guides encoding a user’s history and topic words control decoding their self-introduction. For experiments, we collect a large-scale Twitter dataset, and extensive results show the superiority of our UTGED to the advanced encoder-decoder models without topic modeling.

Continual pre-training is the paradigm where pre-trained language models (PLMs) continually acquire fresh knowledge from growing data and gradually get upgraded. Before an upgraded PLM is released, we may have tuned the original PLM for various tasks and stored the adapted weights. However, when tuning the upgraded PLM, these outdated adapted weights will typically be ignored and discarded, causing a potential waste of resources. We bring this issue to the forefront and contend that proper algorithms for recycling outdated adapted weights should be developed. To this end, we formulate the task of recyclable tuning for continual pre-training. In pilot studies, we find that after continual pre-training, the upgraded PLM remains compatible with the outdated adapted weights to some extent. Motivated by this finding, we analyze the connection between continually pre-trained PLMs from two novel aspects, i.e., mode connectivity, and functional similarity. Based on the corresponding findings, we propose both an initialization-based method and a distillation-based method for our task. We demonstrate their feasibility in improving the convergence and performance for tuning the upgraded PLM. We also show that both methods can be combined to achieve better performance.

pdf abs
BLOCSUM: Block Scope-based Source Code Summarization via Shared Block Representation
YunSeok Choi | Hyojun Kim | Jee-Hyong Lee

Code summarization, which aims to automatically generate natural language descriptions from source code, has become an essential task in software development for better program understanding. Abstract Syntax Tree (AST), which represents the syntax structure of the source code, is helpful when utilized together with the sequence of code tokens to improve the quality of code summaries. Recent works on code summarization attempted to capture the sequential and structural information of the source code, but they considered less the property that source code consists of multiple code blocks. In this paper, we propose BLOCSUM, BLOck scope-based source Code SUMmarization via shared block representation that utilizes block-scope information by representing various structures of the code block. We propose a shared block position embedding to effectively represent the structure of code blocks and merge both code and AST.Furthermore, we develop variant ASTs to learn rich information such as block and global dependencies of the source code. To prove our approach, we perform experiments on two real-world datasets, the Java dataset and the Python dataset. We demonstrate the effectiveness of BLOCSUM through various experiments, including ablation studies and a human evaluation.

With the scale and capacity of pretrained models growing rapidly, parameter-efficient language model tuning has emerged as a popular paradigm for solving various NLP and Vision-and-Language (V&L) tasks. In this paper, we design a unified parameter-efficient multitask learning framework that works effectively on both NLP and V&L tasks. In particular, we use a shared hypernetwork that takes trainable hyper-embeddings and visual modality as input, and outputs weights for different modules in a pretrained language model, such as the parameters inserted into multi-head attention blocks (i.e., prefix-tuning) and feed-forward blocks (i.e., adapter-tuning.). Our proposed framework adds fewer trainable parameters in multi-task learning while achieving superior performances and transfer ability compared to state-of-the-art methods. Empirical results on the GLUE benchmark and multiple V&L tasks confirm the effectiveness of our framework.

We extend a non-parametric Bayesian model of (Titov and Klementiev, 2011) to deal with homonymy and polysemy by leveraging distributed contextual word and phrase representations pre-trained on a large collection of unlabelled texts. Then, unsupervised semantic parsing is performed by decomposing sentences into fragments, clustering the fragments to abstract away syntactic variations of the same meaning, and predicting predicate-argument relations between the fragments. To better model the statistical dependencies between predicates and their arguments, we further conduct a hierarchical Pitman-Yor process. An improved Metropolis-Hastings merge-split sampler is proposed to speed up the mixing and convergence of Markov chains by leveraging pre-trained distributed representations. The experimental results show that the models achieve better accuracy on both question-answering and relation extraction tasks.

pdf abs
Generating Labeled Data for Relation Extraction: A Meta Learning Approach with Joint GPT-2 Training
Amir Pouran Ben Veyseh | Franck Dernoncourt | Bonan Min | Thien Nguyen

Relation Extraction (RE) is the task of identifying semantic relation between real-world entities mentioned in text. Despite significant progress in RE research, a remaining challenge for RE concerns the lack of training data for data-hungry deep learning models. Cost of annotation and difficulty of the task are among hindrance to collect a large-scale RE dataset in different domains. To address this limitation, we propose a novel framework to automatically generate labeled data for RE. Our framework presents the pre-trained language model GPT-2 for data generation. In addition, to optimize the generated samples for an RE model, we introduce a meta learning approach to allow the GPT-2 model to be updated during the training process for RE. In particular, to leverage the feedback from the RE model to improve the data generation from GPT-2, we propose a novel reward function to update the GPT-2 model with REINFORCE, seeking to promote the similarity of the RE loss function’s gradients computed for generated data and a meta development set. We conduct extensive experiments on two benchmark datasets to produce state-of-the-art performance for RE.

pdf abs
Disfluency Generation for More Robust Dialogue Systems
Benjamin Marie

Disfluencies in user utterances can trigger a chain of errors impacting all the modules of a dialogue system: natural language understanding, dialogue state tracking, and response generation. In this work, we first analyze existing dialogue datasets commonly used in research and show that they only contain a marginal number of disfluent utterances. Due to this relative absence of disfluencies in their training data, dialogue systems may then critically fail when exposed to disfluent utterances. Following this observation, we propose to augment existing datasets with disfluent user utterances by paraphrasing fluent utterances into disfluent ones. Relying on a pre-trained language model, our few-shot disfluent paraphraser guided by a disfluency classifier can generate useful disfluent utterances for training better dialogue systems. We report on improvements for both dialogue state tracking and response generation when the dialogue systems are trained on datasets augmented with our disfluent utterances.

pdf abs
Dipping PLMs Sauce: Bridging Structure and Text for Effective Knowledge Graph Completion via Conditional Soft Prompting
Chen Chen | Yufei Wang | Aixin Sun | Bing Li | Kwok-Yan Lam

Knowledge Graph Completion (KGC) often requires both KG structural and textual information to be effective. Pre-trained Language Models (PLMs) have been used to learn the textual information, usually under the fine-tune paradigm for the KGC task. However, the fine-tuned PLMs often overwhelmingly focus on the textual information and overlook structural knowledge. To tackle this issue, this paper proposes CSProm-KG (Conditional Soft Prompts for KGC) which maintains a balance between structural information and textual knowledge. CSProm-KG only tunes the parameters of Conditional Soft Prompts that are generated by the entities and relations representations. We verify the effectiveness of CSProm-KG on three popular static KGC benchmarks WN18RR, FB15K-237 and Wikidata5M, and two temporal KGC benchmarks ICEWS14 and ICEWS05-15. CSProm-KG outperforms competitive baseline models and sets new state-of-the-art on these benchmarks. We conduct further analysis to show (i) the effectiveness of our proposed components, (ii) the efficiency of CSProm-KG, and (iii) the flexibility of CSProm-KG.

pdf abs
Revisiting Pathologies of Neural Models under Input Reduction
Canasai Kruengkrai | Junichi Yamagishi

We revisit the question of why neural models tend to produce high-confidence predictions on inputs that appear nonsensical to humans. Previous work has suggested that the models fail to assign low probabilities to such inputs due to model overconfidence. We evaluate various regularization methods on fact verification benchmarks and find that this problem persists even with well-calibrated or underconfident models, suggesting that overconfidence is not the only underlying cause. We also find that regularizing the models with reduced examples helps improve interpretability but comes with the cost of miscalibration. We show that although these reduced examples are incomprehensible to humans, they can contain valid statistical patterns in the dataset utilized by the model.

Multilingual neural machine translation (MNMT) aims to build a unified model for many language directions. Existing monolithic models for MNMT encounter two challenges: parameter interference among languages and inefficient inference for large models. In this paper, we revisit the classic multi-way structures and develop a detachable model by assigning each language (or group of languages) to an individual branch that supports plug-and-play training and inference. To address the needs of learning representations for all languages in a unified space, we propose a novel efficient training recipe, upon which we build an effective detachable model, Lego-MT.For a fair comparison, we collect data from OPUS and build a translation benchmark covering 433 languages and 1.3B parallel data. Experiments show that Lego-MT with 1.2B parameters brings an average gain of 3.2 spBLEU. It even outperforms M2M-100 with 12B parameters. The proposed training recipe brings a 28.2× speedup over the conventional multi-way training method.code and data repo: https://github.com/CONE-MT/Lego-MT.git.

Fusion-in-Decoder (FiD) is a powerful retrieval-augmented language model that sets the state-of-the-art on many knowledge-intensive NLP tasks. However, the architecture used for FiD was chosen by making minimal modifications to a standard T5 model, which our analysis shows to be highly suboptimal for a retrieval-augmented model. In particular, FiD allocates the bulk of FLOPs to the encoder, while the majority of inference time results from memory bandwidth constraints in the decoder. We propose two simple changes to the FiD architecture to alleviate memory bandwidth constraints, and speed up inference by 7x. This allows us to use a much larger decoder at modest cost. We denote FiD with the above modifications as FiDO, and show that it strongly improves performance over existing FiD models for a wide range of inference budgets. For example, FiDO-Large-XXL performs faster inference than FiD-Base and achieves better performance than FiD-Large.

pdf abs
Detecting Edit Failures In Large Language Models: An Improved Specificity Benchmark
Jason Hoelscher-Obermaier | Julia Persson | Esben Kran | Ioannis Konstas | Fazl Barez

Recent model editing techniques promise to mitigate the problem of memorizing false or outdated associations during LLM training. However, we show that these techniques can introduce large unwanted side effects which are not detected by existing specificity benchmarks. We extend the existing CounterFact benchmark to include a dynamic component and dub our benchmark CounterFact+. Additionally, we extend the metrics used for measuring specificity by a principled KL divergence-based metric. We use this improved benchmark to evaluate recent model editing techniques and find that they suffer from low specificity. Our findings highlight the need for improved specificity benchmarks that identify and prevent unwanted side effects.

This paper presents Structure Aware Dense Retrieval (SANTA) model, which encodes user queries and structured data in one universal embedding space for retrieving structured data. SANTA proposes two pretraining methods to make language models structure-aware and learn effective representations for structured data: 1) Structured Data Alignment, which utilizes the natural alignment relations between structured data and unstructured data for structure-aware pretraining. It contrastively trains language models to represent multi-modal text data and teaches models to distinguish matched structured data for unstructured texts. 2) Masked Entity Prediction, which designs an entity-oriented mask strategy and asks language models to fill in the masked entities. Our experiments show that SANTA achieves state-of-the-art on code search and product search and conducts convincing results in the zero-shot setting. SANTA learns tailored representations for multi-modal text data by aligning structured and unstructured data pairs and capturing structural semantics by masking and predicting entities in the structured data. All codes are available at https://github.com/OpenMatch/OpenMatch.

We have witnessed the rapid proliferation of multimodal data on numerous social media platforms. Conventional studies typically require massive labeled data to train models for Multimodal Aspect-Based Sentiment Analysis (MABSA). However, collecting and annotating fine-grained multimodal data for MABSA is tough. To alleviate the above issue, we perform three MABSA-related tasks with quite a small number of labeled multimodal samples. We first build diverse and comprehensive multimodal few-shot datasets according to the data distribution. To capture the specific prompt for each aspect term in a few-shot scenario, we propose a novel Generative Multimodal Prompt (GMP) model for MABSA, which includes the Multimodal Encoder module and the N-Stream Decoders module. We further introduce a subtask to predict the number of aspect terms in each instance to construct the multimodal prompt. Extensive experiments on two datasets demonstrate that our approach outperforms strong baselines on two MABSA-related tasks in the few-shot setting.

pdf abs
Predicting Human Translation Difficulty Using Automatic Word Alignment
Zheng Wei Lim | Trevor Cohn | Charles Kemp | Ekaterina Vylomova

Translation difficulty arises when translators are required to resolve translation ambiguity from multiple possible translations. Translation difficulty can be measured by recording the diversity of responses provided by human translators and the time taken to provide these responses, but these behavioral measures are costly and do not scale. In this work, we use word alignments computed over large scale bilingual corpora to develop predictors of lexical translation difficulty. We evaluate our approach using behavioural data from translations provided both in and out of context, and report results that improve on a previous embedding-based approach (Thompson et al., 2020). Our work can therefore contribute to a deeper understanding of cross-lingual differences and of causes of translation difficulty.

pdf abs
Know Where You’re Going: Meta-Learning for Parameter-Efficient Fine-Tuning
Mozhdeh Gheini | Xuezhe Ma | Jonathan May

A recent family of techniques, dubbed lightweight fine-tuning methods, facilitates parameter-efficient transfer by updating only a small set of additional parameters while keeping the parameters of the original model frozen. While proven to be an effective approach, there are no existing studies on if and how such knowledge of the downstream fine-tuning approach calls for complementary measures after pre-training and before fine-tuning. In this work, we show that taking the ultimate choice of fine-tuning into consideration boosts the performance of parameter-efficient fine-tuning. By relying on optimization-based meta-learning using MAML with certain modifications for our distinct purpose, we prime the pre-trained model specifically for parameter-efficient fine-tuning, resulting in gains of up to 4.96 points on cross-lingual NER fine-tuning. Our ablation settings and analyses further reveal that the specific approach we take to meta-learning is crucial for the attained gains.

Neural information retrieval (IR) systems have progressed rapidly in recent years, in large part due to the release of publicly available benchmarking tasks. Unfortunately, some dimensions of this progress are illusory: the majority of the popular IR benchmarks today focus exclusively on downstream task accuracy and thus conceal the costs incurred by systems that trade away efficiency for quality. Latency, hardware cost, and other efficiency considerations are paramount to the deployment of IR systems in user-facing settings. We propose that IR benchmarks structure their evaluation methodology to include not only metrics of accuracy, but also efficiency considerations such as a query latency and the corresponding cost budget for a reproducible hardware setting. For the popular IR benchmarks MS MARCO and XOR-TyDi, we show how the best choice of IR system varies according to how these efficiency considerations are chosen and weighed. We hope that future benchmarks will adopt these guidelines toward more holistic IR evaluation.

pdf abs
AxomiyaBERTa: A Phonologically-aware Transformer Model for Assamese
Abhijnan Nath | Sheikh Mannan | Nikhil Krishnaswamy

Despite their successes in NLP, Transformer-based language models still require extensive computing resources and suffer in low-resource or low-compute settings. In this paper, we present AxomiyaBERTa, a novel BERT model for Assamese, a morphologically-rich low-resource language (LRL) of Eastern India. AxomiyaBERTa is trained only on the masked language modeling (MLM) task, without the typical additional next sentence prediction (NSP) objective, and our results show that in resource-scarce settings for very low-resource languages like Assamese, MLM alone can be successfully leveraged for a range of tasks. AxomiyaBERTa achieves SOTA on token-level tasks like Named Entity Recognition and also performs well on “longer-context” tasks like Cloze-style QA and Wiki Title Prediction, with the assistance of a novel embedding disperser and phonological signals respectively. Moreover, we show that AxomiyaBERTa can leverage phonological signals for even more challenging tasks, such as a novel cross-document coreference task on a translated version of the ECB+ corpus, where we present a new SOTA result for an LRL. Our source code and evaluation scripts may be found at https://github.com/csu-signal/axomiyaberta.

Text-to-SQL translates user queries into SQL statements that can retrieve relevant answers from relational databases. Recent approaches to Text-to-SQL rely on pre-trained language models that are computationally expensive and technically challenging to deploy in real-world applications that require real-time or on-device processing capabilities. In this paper, we perform a focused study on the feasibility of applying recent model compression techniques to sketch-based and sequence-to-sequence Text-to-SQL models. Our results reveal that sketch-based Text-to-SQL models generally have higher inference efficiency and respond better to model compression than sequence-to-sequence models, making them ideal for real-world deployments, especially in use cases with simple SQL statements.

Stutter removal is an essential scenario in the field of speech editing. However, when the speech recording contains stutters, the existing text-based speech editing approaches still suffer from: 1) the over-smoothing problem in the edited speech; 2) lack of robustness due to the noise introduced by stutter; 3) to remove the stutters, users are required to determine the edited region manually. To tackle the challenges in stutter removal, we propose FluentSpeech, a stutter-oriented automatic speech editing model. Specifically, 1) we propose a context-aware diffusion model that iteratively refines the modified mel-spectrogram with the guidance of context features; 2) we introduce a stutter predictor module to inject the stutter information into the hidden sequence; 3) we also propose a stutter-oriented automatic speech editing (SASE) dataset that contains spontaneous speech recordings with time-aligned stutter labels to train the automatic stutter localization model. Experimental results on VCTK and LibriTTS datasets demonstrate that our model achieves state-of-the-art performance on speech editing. Further experiments on our SASE dataset show that FluentSpeech can effectively improve the fluency of stuttering speech in terms of objective and subjective metrics. Code and audio samples can be found at https://github.com/Zain-Jiang/Speech-Editing-Toolkit.

Hierarchical Topic Models (HTMs) are useful for discovering topic hierarchies in a collection of documents. However, traditional HTMs often produce hierarchies where lower-level topics are unrelated and not specific enough to their higher-level topics. Additionally, these methods can be computationally expensive. We present HyHTM - a Hyperbolic geometry-based Hierarchical Topic Model - that addresses these limitations by incorporating hierarchical information from hyperbolic geometry to explicitly model hierarchies in topic models. Experimental results with four baselines show that HyHTM can better attend to parent-child relationships among topics. HyHTM produces coherent topic hierarchies that specialize in granularity from generic higher-level topics to specific lower-level topics. Further, our model is significantly faster and leaves a much smaller memory footprint than our best-performing baseline. We have made the source code for our algorithm publicly accessible.

Deep text understanding, which requires the connections between a given document and prior knowledge beyond its text, has been highlighted by many benchmarks in recent years. However, these benchmarks have encountered two major limitations. On the one hand, most of them require human annotation of knowledge, which leads to limited knowledge coverage. On the other hand, they usually use choices or spans in the texts as the answers, which results in narrow answer space. To overcome these limitations, we build a new challenging benchmark named KoRC in this paper. Compared with previous benchmarks, KoRC has two advantages, i.e., broad knowledge coverage and flexible answer format. Specifically, we utilize massive knowledge bases to guide annotators or large language models (LLMs) to construct knowledgable questions. Moreover, we use labels in knowledge bases rather than spans or choices as the final answers. We test state-of-the-art models on KoRC and the experimental results show that the strongest baseline only achieves 68.3% and 30.0% F1 measure in the IID and OOD test set, respectively. These results indicate that deep text understanding is still an unsolved challenge. We will release our dataset and baseline methods upon acceptance.

pdf abs
DKAF: KB Arbitration for Learning Task-Oriented Dialog Systems with Dialog-KB Inconsistencies
Vishal Saley | Rocktim Das | Dinesh Raghu | Mausam

Task-oriented dialog (TOD) agents often ground their responses on external knowledge bases (KBs). These KBs can be dynamic and may be updated frequently. Existing approaches for learning TOD agents assume the KB snapshot contemporary to each individual dialog is available during training. However, in real-world scenarios, only the latest KB snapshot is available during training and as a result, the train dialogs may contain facts conflicting with the latest KB. These dialog-KB inconsistencies in the training data may potentially confuse the TOD agent learning algorithm. In this work, we define the novel problem of learning a TOD agent with dialog-KB inconsistencies in the training data. We propose a Dialog-KB Arbitration Framework (DKAF) which reduces the dialog-KB inconsistencies by predicting the contemporary KB snapshot for each train dialog. These predicted KB snapshots are then used for training downstream TOD agents. As there are no existing datasets with dialog-KB inconsistencies, we systematically introduce inconsistencies in two publicly available dialog datasets. We show that TOD agents trained with DKAF perform better than existing baselines on both these datasets.

pdf abs
Scale-Invariant Infinite Hierarchical Topic Model
Shusei Eshima | Daichi Mochihashi

Hierarchical topic models have been employed to organize a large number of diverse topics from corpora into a latent tree structure. However, existing models yield fragmented topics with overlapping themes whose expected probability becomes exponentially smaller along the depth of the tree. To solve this intrinsic problem, we propose a scale-invariant infinite hierarchical topic model (ihLDA). The ihLDA adaptively adjusts the topic creation to make the expected topic probability decay considerably slower than that in existing models. Thus, it facilitates the estimation of deeper topic structures encompassing diverse topics in a corpus. Furthermore, the ihLDA extends a widely used tree-structured prior (Adams et al., 2010) in a hierarchical Bayesian way, which enables drawing an infinite topic tree from the base tree while efficiently sampling the topic assignments for the words. Experiments demonstrate that the ihLDA has better topic uniqueness and hierarchical diversity thanexisting approaches, including state-of-the-art neural models.

Multilingual vision-language (V&L) pre-training has achieved remarkable progress in learning universal representations across different modalities and languages. In spite of recent success, there still remain challenges limiting further improvements of V&L pre-trained models in multilingual settings. Particularly, current V&L pre-training methods rely heavily on strictly-aligned multilingual image-text pairs generated from English-centric datasets through machine translation. However, the cost of collecting and translating such strictly-aligned datasets is usually unbearable. In this paper, we propose Regularized Contrastive Cross-lingual Cross-modal (RC3) pre-training, which further exploits more abundant weakly-aligned multilingual image-text pairs. Specifically, we design a regularized cross-lingual visio-textual contrastive learning objective that constrains the representation proximity of weakly-aligned visio-textual inputs according to textual relevance. Besides, existing V&L pre-training approaches mainly deal with visual inputs by either region-of-interest (ROI) features or patch embeddings. We flexibly integrate the two forms of visual features into our model for pre-training and downstream multi-modal tasks. Extensive experiments on 5 downstream multi-modal tasks across 6 languages demonstrate the effectiveness of our proposed method over competitive contrast models with strong zero-shot capability.

pdf abs
Deep Equilibrium Non-Autoregressive Sequence Learning
Zaixiang Zheng | Yi Zhou | Hao Zhou

In this work, we argue that non-autoregressive (NAR) sequence generative models can equivalently be regarded as an iterative refinement process towards the target sequence, implying an underlying dynamical system of NAR model: z = f (z, x) → y. In such a way, the optimal prediction of a NAR model should be the equilibrium state of its dynamics if given infinitely many iterations. However, this is infeasible in practice due to limited computational and memory budgets. To this end, we propose DEQNAR to directly solve for the equilibrium state of NAR models based on deep equilibrium networks (Bai et al., 2019) with black-box root-finding solvers and back-propagate through the equilibrium point via implicit differentiation with constant memory. We conduct extensive experiments on four WMT machine translation benchmarks. Our main findings show that DEQNAR can indeed converge to a more accurate prediction and is a general-purpose framework that consistently helps yield substantial improvement for several strong NAR backbones.

With the development of large language models (LLMs), zero-shot learning has attracted much attention for various NLP tasks. Different from prior works that generate training data with billion-scale natural language generation (NLG) models, we propose a retrieval-enhanced framework to create training data from a general-domain unlabeled corpus. To realize this, we first conduct contrastive pretraining to learn an unsupervised dense retriever for extracting the most relevant documents using class-descriptive verbalizers. We then further pro- pose two simple strategies, namely Verbalizer Augmentation with Demonstrations and Self- consistency Guided Filtering to improve the topic coverage of the dataset while removing noisy examples. Experiments on nine datasets demonstrate that ReGen achieves 4.3% gain over the strongest baselines and saves around 70% of the time when compared with baselines using large NLG models. Besides, REGEN can be naturally integrated with recently proposed large language models to boost performance.

pdf abs
Race, Gender, and Age Biases in Biomedical Masked Language Models
Michelle Kim | Junghwan Kim | Kristen Johnson

Biases cause discrepancies in healthcare services. Race, gender, and age of a patient affect interactions with physicians and the medical treatments one receives. These biases in clinical practices can be amplified following the release of pre-trained language models trained on biomedical corpora. To bring awareness to such repercussions, we examine social biases present in the biomedical masked language models. We curate prompts based on evidence-based practice and compare generated diagnoses based on biases. For a case study, we measure bias in diagnosing coronary artery disease and using cardiovascular procedures based on bias. Our study demonstrates that biomedical models are less biased than BERT in gender, while the opposite is true for race and age.

pdf abs
Neighboring Words Affect Human Interpretation of Saliency Explanations
Alon Jacovi | Hendrik Schuff | Heike Adel | Ngoc Thang Vu | Yoav Goldberg

Word-level saliency explanations (“heat maps over words”) are often used to communicate feature-attribution in text-based models. Recent studies found that superficial factors such as word length can distort human interpretation of the communicated saliency scores. We conduct a user study to investigate how the marking of a word’s *neighboring words* affect the explainee’s perception of the word’s importance in the context of a saliency explanation. We find that neighboring words have significant effects on the word’s importance rating. Concretely, we identify that the influence changes based on neighboring direction (left vs. right) and a-priori linguistic and computational measures of phrases and collocations (vs. unrelated neighboring words).Our results question whether text-based saliency explanations should be continued to be communicated at word level, and inform future research on alternative saliency explanation methods.

pdf abs
HELP ME THINK: A Simple Prompting Strategy for Non-experts to Create Customized Content with Models
Swaroop Mishra | Elnaz Nouri

Controlling the text generated by language models and customizing the content has been a long-standing challenge. Existing prompting techniques proposed in pursuit of providing control are task-specific and lack generality; this provides overwhelming choices for non-expert users to find a suitable method for their task. The effort associated with those techniques, such as in writing examples, explanations, instructions, etc. further limits their adoption among non-expert users. In this paper, we propose a simple prompting strategy Help Me Think where we encourage largelanguage models (such as GPT3 and ChatGPT) to help non-expert users by asking a set of relevant questions and leveraging user answers to execute the task. We demonstrate the efficacy of our technique Help Me Think on a variety of tasks. Specifically, we focus on tasks that are hard for average humans and require significant thinking to perform. We hope our work will encourage the development of unconventional ways to harness the power of large language models.

pdf abs
Decker: Double Check with Heterogeneous Knowledge for Commonsense Fact Verification
Anni Zou | Zhuosheng Zhang | Hai Zhao

Commonsense fact verification, as a challenging branch of commonsense question-answering (QA), aims to verify through facts whether a given commonsense claim is correct or not. Answering commonsense questions necessitates a combination of knowledge from various levels. However, existing studies primarily rest on grasping either unstructured evidence or potential reasoning paths from structured knowledge bases, yet failing to exploit the benefits of heterogeneous knowledge simultaneously. In light of this, we propose Decker, a commonsense fact verification model that is capable of bridging heterogeneous knowledge by uncovering latent relationships between structured and unstructured knowledge. Experimental results on two commonsense fact verification benchmark datasets, CSQA2.0 and CREAK demonstrate the effectiveness of our Decker and further analysis verifies its capability to seize more precious information through reasoning. The official implementation of Decker is available at https://github.com/Anni-Zou/Decker.

Recently, binaural audio synthesis (BAS) has emerged as a promising research field for its applications in augmented and virtual realities. Binaural audio helps ususers orient themselves and establish immersion by providing the brain with interaural time differences reflecting spatial information. However, existing BAS methods are limited in terms of phase estimation, which is crucial for spatial hearing. In this paper, we propose the DopplerBAS method to explicitly address the Doppler effect of the moving sound source. Specifically, we calculate the radial relative velocity of the moving speaker in spherical coordinates, which further guides the synthesis of binaural audio. This simple method introduces no additional hyper-parameters and does not modify the loss functions, and is plug-and-play: it scales well to different types of backbones. DopperBAS distinctly improves the representative WarpNet and BinauralGrad backbones in the phase error metric and reaches a new state of the art (SOTA): 0.780 (versus the current SOTA 0.807). Experiments and ablation studies demonstrate the effectiveness of our method.

pdf abs
Easy-to-Hard Learning for Information Extraction
Chang Gao | Wenxuan Zhang | Wai Lam | Lidong Bing

Information extraction (IE) systems aim to automatically extract structured information, such as named entities, relations between entities, and events, from unstructured texts. While most existing work addresses a particular IE task, universally modeling various IE tasks with one model has achieved great success recently. Despite their success, they employ a one-stage learning strategy, i.e., directly learning to extract the target structure given the input text, which contradicts the human learning process. In this paper, we propose a unified easy-to-hard learning framework consisting of three stages, i.e., the easy stage, the hard stage, and the main stage, for IE by mimicking the human learning process. By breaking down the learning process into multiple stages, our framework facilitates the model to acquire general IE task knowledge and improve its generalization ability. Extensive experiments across four IE tasks demonstrate the effectiveness of our framework. We achieve new state-of-the-art results on 13 out of 17 datasets.

pdf abs
SConE: Simplified Cone Embeddings with Symbolic Operators for Complex Logical Queries
Chau Nguyen | Tim French | Wei Liu | Michael Stewart

Geometric representation of query embeddings (using points, particles, rectangles and cones) can effectively achieve the task of answering complex logical queries expressed in first-order logic (FOL) form over knowledge graphs, allowing intuitive encodings. However, current geometric-based methods depend on the neural approach to model FOL operators (conjunction, disjunction and negation), which are not easily explainable with considerable computation cost. We overcome this challenge by introducing a symbolic modeling approach for the FOL operators, emphasizing the direct calculation of the intersection between geometric shapes, particularly sector-cones in the embedding space, to model the conjunction operator. This approach reduces the computation cost as a non-neural approach is involved in the core logic operators. Moreover, we propose to accelerate the learning in the relation projection operator using the neural approach to emphasize the essential role of this operator in all query structures. Although empirical evidence for explainability is challenging, our approach demonstrates a significant improvement in answering complex logical queries (both non-negative and negative FOL forms) over previous geometric-based models.

The prevalence of short video platforms has spawned a lot of fake news videos, which have stronger propagation ability than textual fake news. Thus, automatically detecting fake news videos has been an important countermeasure in practice. Previous works commonly verify each news video individually with multimodal information. Nevertheless, news videos from different perspectives regarding the same event are commonly posted together, which contain complementary or contradictory information and thus can be used to evaluate each other mutually. To this end, we introduce a new and practical paradigm, i.e., cross-sample fake news video detection, and propose a novel framework, Neighbor-Enhanced fakE news video Detection (NEED), which integrates the neighborhood relationship of new videos belonging to the same event. NEED can be readily combined with existing single-sample detectors and further enhance their performances with the proposed graph aggregation (GA) and debunking rectification (DR) modules. Specifically, given the feature representations obtained from single-sample detectors, GA aggregates the neighborhood information with the dynamic graph to enrich the features of independent samples. After that, DR explicitly leverages the relationship between debunking videos and fake news videos to refute the candidate videos via textual and visual consistency. Extensive experiments on the public benchmark demonstrate that NEED greatly improves the performance of both single-modal (up to 8.34% in accuracy) and multimodal (up to 4.97% in accuracy) base detectors.

pdf abs
An Annotated Dataset for Explainable Interpersonal Risk Factors of Mental Disturbance in Social Media Posts
Muskan Garg | Amirmohammad Shahbandegan | Amrit Chadha | Vijay Mago

With a surge in identifying suicidal risk and its severity in social media posts, we argue that a more consequential and explainable research is required for optimal impact on clinical psychology practice and personalized mental healthcare. The success of computational intelligence techniques for inferring mental illness from social media resources, points to natural language processing as a lens for determining Interpersonal Risk Factors (IRF) in human writings. Motivated with limited availability of datasets for social NLP research community, we construct and release a new annotated dataset with human-labelled explanations and classification of IRF affecting mental disturbance on social media: (i) Thwarted Belongingness (TBe), and (ii) Perceived Burdensomeness (PBu). We establish baseline models on our dataset facilitating future research directions to develop real-time personalized AI models by detecting patterns of TBe and PBu in emotional spectrum of user’s historical social media profile.

pdf abs
Nano: Nested Human-in-the-Loop Reward Learning for Few-shot Language Model Control
Xiang Fan | Yiwei Lyu | Paul Pu Liang | Ruslan Salakhutdinov | Louis-Philippe Morency

Pretrained language models have demonstrated extraordinary capabilities in language generation. However, real-world tasks often require controlling the distribution of generated text in order to mitigate bias, promote fairness, and achieve personalization. Existing techniques for controlling the distribution of generated text only work with quantified distributions, which require pre-defined categories, proportions of the distribution, or an existing corpus following the desired distributions. However, many important distributions, such as personal preferences, are unquantified. In this work, we tackle the problem of generating text following arbitrary distributions (quantified and unquantified) by proposing NANO, a few-shot human-in-the-loop training algorithm that continuously learns from human feedback. NANO achieves state-of-the-art results on single topic/attribute as well as quantified distribution control compared to previous works. We also show that NANO is able to learn unquantified distributions, achieves personalization, and captures differences between different individuals’ personal preferences with high sample efficiency.

Task embeddings are task-specific vectors designed to construct a semantic space of tasks, which can be used to predict the most transferable source task for a given target task via the similarity between task embeddings. However, existing methods use optimized parameters and representations as task embeddings, resulting in substantial computational complexity and storage requirements. In this work, we draw inspiration from the operating mechanism of deep neural networks (DNNs) and biological brains, where neuronal activations are sparse and task-specific, and we use the connectivity patterns of neurons as a unique identifier associated with the task. The proposed method learns to assign importance masks for sub-structures of DNNs, and accordingly indicate the task-specific connectivity patterns. In addition to the storage advantages brought by the binary masking mechanism and structured sparsity, the early-bird nature of the sparse optimization process can deliver an efficient computation advantage. Experiments show that our method consistently outperforms other baselines in predicting inter-task transferability across data regimes and transfer settings, while keeping high efficiency in computation and storage.

Grammatical Error Correction (GEC) aims to correct grammatical errors in sentences. We find that autoregressive models tend to assign low probabilities to tokens that need corrections. Here we introduce additional signals to the training of GEC models so that these systems can learn to better predict at ambiguous positions. To do this, we use a non-autoregressive model as an auxiliary model, and develop a new regularization term of training by considering the difference in predictions between the autoregressive and non-autoregressive models. We experiment with this method on both English and Chinese GEC tasks. Experimental results show that our GEC system outperforms the baselines on all the data sets significantly.

Dual encoders have been used for retrieval tasks and representation learning with good results. A standard way to train dual encoders is using a contrastive loss with in-batch negatives. In this work, we propose an improved contrastive learning objective by adding queries or documents from the same encoder towers to the negatives, for which we name it as “contrastive loss with SAMe TOwer NEgatives” (SamToNe). By evaluating on question answering retrieval benchmarks from MS MARCO and MultiReQA, and heterogenous zero-shot information retrieval benchmarks (BEIR), we demonstrate that SamToNe can effectively improve the retrieval quality for both symmetric and asymmetric dual encoders. By directly probing the embedding spaces of the two encoding towers via the t-SNE algorithm (van der Maaten and Hinton, 2008), we observe that SamToNe ensures the alignment between the embedding spaces from the two encoder towers. Based on the analysis of the embedding distance distributions of the top-1 retrieved results, we further explain the efficacy of the method from the perspective of regularisation.

pdf abs
On the Strength of Sequence Labeling and Generative Models for Aspect Sentiment Triplet Extraction
Shen Zhou | Tieyun Qian

Generative models have achieved great success in aspect sentiment triplet extraction tasks. However, existing methods ignore the mutual informative clues between aspect and opinion terms and may generate false paired triplets. Furthermore, the inherent limitations of generative models, i.e., the token-by-token decoding and the simple structured prompt, prevent models from handling complex structures especially multi-word terms and multi-triplet sentences. To address these issues, we propose a sequence labeling enhanced generative model. Firstly, we encode the dependency between aspect and opinion into two bidirectional templates to avoid false paired triplets. Secondly, we introduce a marker-oriented sequence labeling module to improve generative models’ ability of tackling complex structures. Specifically, this module enables the generative model to capture the boundary information of aspect/opinion spans and provides hints to decode multiple triplets with the shared marker. Experimental results on four datasets prove that our model yields a new state-of-art performance. Our code and data are available at https://github.com/NLPWM-WHU/SLGM.

pdf abs
Revisiting Non-Autoregressive Translation at Scale
Zhihao Wang | Longyue Wang | Jinsong Su | Junfeng Yao | Zhaopeng Tu

In real-world systems, scaling has been critical for improving the translation quality in autoregressive translation (AT), which however has not been well studied for non-autoregressive translation (NAT). In this work, we bridge the gap by systematically studying the impact of scaling on NAT behaviors. Extensive experiments on six WMT benchmarks over two advanced NAT models show that scaling can alleviate the commonly-cited weaknesses of NAT models, resulting in better translation performance. To reduce the side-effect of scaling on decoding speed, we empirically investigate the impact of NAT encoder and decoder on the translation performance. Experimental results on the large-scale WMT20 En-De show that the asymmetric architecture (e.g. bigger encoder and smaller decoder) can achieve comparable performance with the scaling model, while maintaining the superiority of decoding speed with standard NAT models. To this end, we establish a new benchmark by validating scaled NAT models on the scaled dataset, which can be regarded as a strong baseline for future works. We release code and system outputs at https://github.com/DeepLearnXMU/Scaling4NAT.

pdf abs
Improving Radiology Summarization with Radiograph and Anatomy Prompts
Jinpeng Hu | Zhihong Chen | Yang Liu | Xiang Wan | Tsung-Hui Chang

The impression is crucial for the referring physicians to grasp key information since it is concluded from the findings and reasoning of radiologists. To alleviate the workload of radiologists and reduce repetitive human labor in impression writing, many researchers have focused on automatic impression generation. However, recent works on this task mainly summarize the corresponding findings and pay less attention to the radiology images. In clinical, radiographs can provide more detailed valuable observations to enhance radiologists’ impression writing, especially for complicated cases. Besides, each sentence in findings usually focuses on single anatomy, such that they only need to be matched to corresponding anatomical regions instead of the whole image, which is beneficial for textual and visual features alignment. Therefore, we propose a novel anatomy-enhanced multimodal model to promote impression generation. In detail, we first construct a set of rules to extract anatomies and put these prompts into each sentence to highlight anatomy characteristics. Then, two separate encoders are applied to extract features from the radiograph and findings. Afterward, we utilize a contrastive learning module to align these two representations at the overall level and use a co-attention to fuse them at the sentence level with the help of anatomy-enhanced sentence representation. The experimental results on two benchmark datasets confirm the effectiveness of the proposed method, which achieves state-of-the-art results.

pdf abs
Explanation Regeneration via Information Bottleneck
Qintong Li | Zhiyong Wu | Lingpeng Kong | Wei Bi

Explaining the black-box predictions of NLP models naturally and accurately is an important open problem in natural language generation. These free-text explanations are expected to contain sufficient and carefully-selected evidence to form supportive arguments for predictions. Thanks to the superior generative capacity of large pretrained language models (PLM), recent work built on prompt engineering enables explanations generated without specific training. However, explanations generated through single-pass prompting often lack sufficiency and conciseness, due to the prompt complexity and hallucination issues. To discard the dross and take the essence of current PLM’s results, we propose to produce sufficient and concise explanations via the information bottleneck (EIB) theory. EIB regenerates explanations by polishing the single-pass output of PLM but retaining the information that supports the contents being explained by balancing two information bottleneck objectives. Experiments on two different tasks verify the effectiveness of EIB through automatic evaluation and thoroughly-conducted human evaluation.

pdf abs
Improving Zero-shot Multilingual Neural Machine Translation by Leveraging Cross-lingual Consistency Regularization
Pengzhi Gao | Liwen Zhang | Zhongjun He | Hua Wu | Haifeng Wang

The multilingual neural machine translation (NMT) model has a promising capability of zero-shot translation, where it could directly translate between language pairs unseen during training. For good transfer performance from supervised directions to zero-shot directions, the multilingual NMT model is expected to learn universal representations across different languages. This paper introduces a cross-lingual consistency regularization, CrossConST, to bridge the representation gap among different languages and boost zero-shot translation performance. The theoretical analysis shows that CrossConST implicitly maximizes the probability distribution for zero-shot translation, and the experimental results on both low-resource and high-resource benchmarks show that CrossConST consistently improves the translation performance. The experimental analysis also proves that CrossConST could close the sentence representation gap and better align the representation space. Given the universality and simplicity of CrossConST, we believe it can serve as a strong baseline for future multilingual NMT research.

Structured chemical reaction information plays a vital role for chemists engaged in laboratory work and advanced endeavors such as computer-aided drug design. Despite the importance of extracting structured reactions from scientific literature, data annotation for this purpose is cost-prohibitive due to the significant labor required from domain experts. Consequently, the scarcity of sufficient training data poses an obstacle to the progress of related models in this domain. In this paper, we propose ReactIE, which combines two weakly supervised approaches for pre-training. Our method utilizes frequent patterns within the text as linguistic cues to identify specific characteristics of chemical reactions. Additionally, we adopt synthetic data from patent records as distant supervision to incorporate domain knowledge into the model. Experiments demonstrate that ReactIE achieves substantial improvements and outperforms all existing baselines.

pdf abs
Expand, Rerank, and Retrieve: Query Reranking for Open-Domain Question Answering
Yung-Sung Chuang | Wei Fang | Shang-Wen Li | Wen-tau Yih | James Glass

We propose EAR, a query Expansion And Reranking approach for improving passage retrieval, with the application to open-domain question answering. EAR first applies a query expansion model to generate a diverse set of queries, and then uses a query reranker to select the ones that could lead to better retrieval results. Motivated by the observation that the best query expansion often is not picked by greedy decoding, EAR trains its reranker to predict the rank orders of the gold passages when issuing the expanded queries to a given retriever. By connecting better the query expansion model and retriever, EAR significantly enhances a traditional sparse retrieval method, BM25. Empirically, EAR improves top-5/20 accuracy by 3-8 and 5-10 points in in-domain and out-of-domain settings, respectively, when compared to a vanilla query expansion model, GAR, and a dense retrieval model, DPR.

pdf abs
Neural Networks Against (and For) Self-Training: Classification with Small Labeled and Large Unlabeled Sets
Payam Karisani

We propose a semi-supervised text classifier based on self-training using one positive and one negative property of neural networks. One of the weaknesses of self-training is the semantic drift problem, where noisy pseudo-labels accumulate over iterations and consequently the error rate soars. In order to tackle this challenge, we reshape the role of pseudo-labels and create a hierarchical order of information. In addition, a crucial step in self-training is to use the classifier confidence prediction to select the best candidate pseudo-labels. This step cannot be efficiently done by neural networks, because it is known that their output is poorly calibrated. To overcome this challenge, we propose a hybrid metric to replace the plain confidence measurement. Our metric takes into account the prediction uncertainty via a subsampling technique. We evaluate our model in a set of five standard benchmarks, and show that it significantly outperforms a set of ten diverse baseline models. Furthermore, we show that the improvement achieved by our model is additive to language model pretraining, which is a widely used technique for using unlabeled documents.

pdf abs
Inducing Character-level Structure in Subword-based Language Models with Type-level Interchange Intervention Training
Jing Huang | Zhengxuan Wu | Kyle Mahowald | Christopher Potts

Language tasks involving character-level manipulations (e.g., spelling corrections, arithmetic operations, word games) are challenging for models operating on subword units. To address this, we develop a causal intervention framework to learn robust and interpretable character representations inside subword-based language models. Our method treats each character as a typed variable in a causal model and learns such causal structures by adapting the interchange intervention training method of Geiger et al. (2021). We additionally introduce a suite of character-level tasks that systematically vary in their dependence on meaning and sequence-level context. While character-level models still perform best on purely form-based tasks like string reversal, our method outperforms character-level models on more complex tasks that blend form, meaning, and context, such as spelling correction in context and word search games. Compared with standard subword-based models, our approach also significantly improves robustness on unseen token sequences and leads to human-interpretable internal representations of characters.

pdf abs
Efficient Document Embeddings via Self-Contrastive Bregman Divergence Learning
Daniel Saggau | Mina Rezaei | Bernd Bischl | Ilias Chalkidis

Learning quality document embeddings is a fundamental problem in natural language processing (NLP), information retrieval (IR), recommendation systems, and search engines. Despite recent advances in the development of transformer-based models that produce sentence embeddings with self-contrastive learning, the encoding of long documents (Ks of words) is still challenging with respect to both efficiency and quality considerations. Therefore, we train Longfomer-based document encoders using a state-of-the-art unsupervised contrastive learning method (SimCSE). Further on, we complement the baseline method -siamese neural network- with additional convex neural networks based on functional Bregman divergence aiming to enhance the quality of the output document representations. We show that overall the combination of a self-contrastive siamese network and our proposed neural Bregman network outperforms the baselines in two linear classification settings on three long document topic classification tasks from the legal and biomedical domains.

Multimodal emotion recognition for video has gained considerable attention in recent years, in which three modalities (i.e., textual, visual and acoustic) are involved. Due to the diverse levels of informational content related to emotion, three modalities typically possess varying degrees of contribution to emotion recognition. More seriously, there might be inconsistencies between the emotion of individual modality and the video. The challenges mentioned above are caused by the inherent uncertainty of emotion. Inspired by the recent advances of quantum theory in modeling uncertainty, we make an initial attempt to design a quantum-inspired adaptive-priority-learning model (QAP) to address the challenges. Specifically, the quantum state is introduced to model modal features, which allows each modality to retain all emotional tendencies until the final classification. Additionally, we design Q-attention to orderly integrate three modalities, and then QAP learns modal priority adaptively so that modalities can provide different amounts of information based on priority. Experimental results on the IEMOCAP and MOSEI datasets show that QAP establishes new state-of-the-art results.

pdf abs
Language acquisition: do children and language models follow similar learning stages?
Linnea Evanson | Yair Lakretz | Jean Rémi King

During language acquisition, children follow a typical sequence of learning stages, whereby they first learn to categorize phonemes before they develop their lexicon and eventually master increasingly complex syntactic structures. However, the computational principles that lead to this learning trajectory remain largely unknown. To investigate this, we here compare the learning trajectories of deep language models to those of human children. Specifically, we test whether, during its training, GPT-2 exhibits stages of language acquisition comparable to those observed in children aged between 18 months and 6 years. For this, we train 48 GPT-2 models from scratch and evaluate their syntactic and semantic abilities at each training step, using 96 probes curated from the BLiMP, Zorro and BIG-Bench benchmarks. We then compare these evaluations with the behavior of 54 children during language production. Our analyses reveal three main findings. First, similarly to children, the language models tend to learn linguistic skills in a systematic order. Second, this learning scheme is parallel: the language tasks that are learned last improve from the very first training steps. Third, some – but not all – learning stages are shared between children and these language models. Overall, these results shed new light on the principles of language acquisition, and highlight important divergences in how humans and modern algorithms learn to process natural language.

pdf abs
The Role of Output Vocabulary in T2T LMs for SPARQL Semantic Parsing
Debayan Banerjee | Pranav Nair | Ricardo Usbeck | Chris Biemann

In this work, we analyse the role of output vocabulary for text-to-text (T2T) models on the task of SPARQL semantic parsing. We perform experiments within the the context of knowledge graph question answering (KGQA), where the task is to convert questions in natural language to the SPARQL query language. We observe that the query vocabulary is distinct from human vocabulary. Language Models (LMs) are pre-dominantly trained for human language tasks, and hence, if the query vocabulary is replaced with a vocabulary more attuned to the LM tokenizer, the performance of models may improve. We carry out carefully selected vocabulary substitutions on the queries and find absolute gains in the range of 17% on the GrailQA dataset.

pdf abs
UniCOQE: Unified Comparative Opinion Quintuple Extraction As A Set
Zinong Yang | Feng Xu | Jianfei Yu | Rui Xia

Comparative Opinion Quintuple Extraction (COQE) aims to identify comparative opinion sentences in product reviews, extract comparative opinion elements in the sentences, and then incorporate them into quintuples. Existing methods decompose the COQE task into multiple primary subtasks and then solve them in a pipeline manner. However, these approaches ignore the intrinsic connection between subtasks and the error propagation among stages. This paper proposes a unified generative model, UniCOQE, to solve the COQE task in one shot. We design a generative template where all the comparative tuples are concatenated as the target output sequence. However, the multiple tuples are inherently not an ordered sequence but an unordered set. The pre-defined order will force the generative model to learn a false order bias and hinge the model’s training. To alleviate this bias, we introduce a new “predict-and-assign” training paradigm that models the golden tuples as a set. Specifically, we utilize a set-matching strategy to find the optimal order of tuples. The experimental results on multiple benchmarks show that our unified generative model significantly outperforms the SOTA method, and ablation experiments prove the effectiveness of the set-matching strategy.

pdf abs
Response-conditioned Turn-taking Prediction
Bing’er Jiang | Erik Ekstedt | Gabriel Skantze

Previous approaches to turn-taking and response generation in conversational systems have treated it as a two-stage process: First, the end of a turn is detected (based on conversation history), then the system generates an appropriate response. Humans, however, do not take the turn just because it is likely, but also consider whether what they want to say fits the position. In this paper, we present a model (an extension of TurnGPT) that conditions the end-of-turn prediction on both conversation history and what the next speaker wants to say. We found that our model consistently outperforms the baseline model in a variety of metrics. The improvement is most prominent in two scenarios where turn predictions can be ambiguous solely from the conversation history: 1) when the current utterance contains a statement followed by a question; 2) when the end of the current utterance semantically matches the response. Treating the turn-prediction and response-ranking as a one-stage process, our findings suggest that our model can be used as an incremental response ranker, which can be applied in various settings.

pdf abs
A Unified One-Step Solution for Aspect Sentiment Quad Prediction
Junxian Zhou | Haiqin Yang | Yuxuan He | Hao Mou | JunBo Yang

Aspect sentiment quad prediction (ASQP) is a challenging yet significant subtask in aspectbased sentiment analysis as it provides a complete aspect-level sentiment structure. However, existing ASQP datasets are usually small and low-density, hindering technical advancement. To expand the capacity, in this paper, we release two new datasets for ASQP, which contain the following characteristics: larger size, more words per sample, and higher density. With such datasets, we unveil the shortcomings of existing strong ASQP baselines and therefore propose a unified one-step solution for ASQP, namely One-ASQP, to detect the aspect categories and to identify the aspectopinion-sentiment (AOS) triplets simultaneously. Our One-ASQP holds several unique advantages: (1) by separating ASQP into two subtasks and solving them independently and simultaneously, we can avoid error propagation in pipeline-based methods and overcome slow training and inference in generation-based methods; (2) by introducing sentiment-specific horns tagging schema in a token-pair-based two-dimensional matrix, we can exploit deeper interactions between sentiment elements and efficiently decode the AOS triplets; (3) we design "[NULL]” token can help us effectively identify the implicit aspects or opinions. Experiments on two benchmark datasets and our released two datasets demonstrate the advantages of our One-ASQP. The two new datasets are publicly released at https://www.github.com/Datastory-CN/ASQP-Datasets.

pdf abs
On Isotropy, Contextualization and Learning Dynamics of Contrastive-based Sentence Representation Learning
Chenghao Xiao | Yang Long | Noura Al Moubayed

Incorporating contrastive learning objectives in sentence representation learning (SRL) has yielded significant improvements on many sentence-level NLP tasks. However, it is not well understood why contrastive learning works for learning sentence-level semantics. In this paper, we aim to help guide future designs of sentence representation learning methods by taking a closer look at contrastive SRL through the lens of isotropy, contextualization and learning dynamics. We interpret its successes through the geometry of the representation shifts and show that contrastive learning brings isotropy, and drives high intra-sentence similarity: when in the same sentence, tokens converge to similar positions in the semantic space. We also find that what we formalize as “spurious contextualization” is mitigated for semantically meaningful tokens, while augmented for functional ones. We find that the embedding space is directed towards the origin during training, with more areas now better defined. We ablate these findings by observing the learning dynamics with different training temperatures, batch sizes and pooling methods.

pdf abs
Few-shot Fine-tuning vs. In-context Learning: A Fair Comparison and Evaluation
Marius Mosbach | Tiago Pimentel | Shauli Ravfogel | Dietrich Klakow | Yanai Elazar

Few-shot fine-tuning and in-context learning are two alternative strategies for task adaptation of pre-trained language models. Recently, in-context learning has gained popularity over fine-tuning due to its simplicity and improved out-of-domain generalization, and because extensive evidence shows that fine-tuned models pick up on spurious correlations. Unfortunately, previous comparisons of the two approaches were done using models of different sizes. This raises the question of whether the observed weaker out-of-domain generalization of fine-tuned models is an inherent property of fine-tuning or a limitation of the experimental setup. In this paper, we compare the generalization of few-shot fine-tuning and in-context learning to challenge datasets, while controlling for the models used, the number of examples, and the number of parameters, ranging from 125M to 30B. Our results show that fine-tuned language models can in fact generalize well out-of-domain. We find that both approaches generalize similarly; they exhibit large variation and depend on properties such as model size and the number of examples, highlighting that robust task adaptation remains a challenge.

Metrics for Inter-Annotator Agreement (IAA), like Cohen’s Kappa, are crucial for validating annotated datasets. Although high agreement is often used to show the reliability of annotation procedures, it is insufficient to ensure or reproducibility. While researchers are encouraged to increase annotator agreement, this can lead to specific and tailored annotation guidelines. We hypothesize that this may result in diverging annotations from different groups. To study this, we first propose the Lee et al. Protocol (LEAP), a standardized and codified annotation protocol. LEAP strictly enforces transparency in the annotation process, which ensures reproducibility of annotation guidelines. Using LEAP to annotate a dialog dataset, we empirically show that while research groups may create reliable guidelines by raising agreement, this can cause divergent annotations across different research groups, thus questioning the validity of the annotations. Therefore, we caution NLP researchers against using reliability as a proxy for reproducibility and validity.

pdf abs
HuaSLIM: Human Attention Motivated Shortcut Learning Identification and Mitigation for Large Language models
Yuqi Ren | Deyi Xiong

Large language models have made remarkable progress on a variety of NLP tasks. However, it has been found that they tend to rely on shortcut features that spuriously correlate with labels for prediction, which weakens their generalization on out-of-distribution samples. In this paper, we propose a human attention guided approach to identifying and mitigating shortcut learning, which encourages the LLM-based target model to learn relevant features. We define an attention-based measurement to capture both model and data bias and identify shortcut tokens by exploring both human and neural attention. In a self-distillation framework, we mitigate shortcut learning by dynamically adjusting the distillation temperature according to the detected shortcut tokens and estimated shortcut degree. Additionally, we utilize human attention as a supervisory signal to constrain large language models to pay more attention to relevant tokens. Experimental results on multiple NLP tasks show that our proposed method can effectively identify shortcut tokens, and significantly improve the robustness of large language models on OOD samples, while not undermining the performance on IID data.

pdf abs
PMI-Align: Word Alignment With Point-Wise Mutual Information Without Requiring Parallel Training Data
Fatemeh Azadi | Heshaam Faili | Mohammad Javad Dousti

Word alignment has many applications including cross-lingual annotation projection, bilingual lexicon extraction, and the evaluation or analysis of translation outputs. Recent studies show that using contextualized embeddings from pre-trained multilingual language models could give us high quality word alignments without the need of parallel training data. In this work, we propose PMI-Align which computes and uses the point-wise mutual information between source and target tokens to extract word alignments, instead of the cosine similarity or dot product which is mostly used in recent approaches. Our experiments show that our proposed PMI-Align approach could outperform the rival methods on five out of six language pairs. Although our approach requires no parallel training data, we show that this method could also benefit the approaches using parallel data to fine-tune pre-trained language models on word alignments. Our code and data are publicly available.

pdf abs
Exploring Non-Verbal Predicates in Semantic Role Labeling: Challenges and Opportunities
Riccardo Orlando | Simone Conia | Roberto Navigli

Although we have witnessed impressive progress in Semantic Role Labeling (SRL), most of the research in the area is carried out assuming that the majority of predicates are verbs. Conversely, predicates can also be expressed using other parts of speech, e.g., nouns and adjectives. However, non-verbal predicates appear in the benchmarks we commonly use to measure progress in SRL less frequently than in some real-world settings – newspaper headlines, dialogues, and tweets, among others. In this paper, we put forward a new PropBank dataset which boasts wide coverage of multiple predicate types. Thanks to it, we demonstrate empirically that standard benchmarks do not provide an accurate picture of the current situation in SRL and that state-of-the-art systems are still incapable of transferring knowledge across different predicate types. Having observed these issues, we also present a novel, manually-annotated challenge set designed to give equal importance to verbal, nominal, and adjectival predicate-argument structures. We use such dataset to investigate whether we can leverage different linguistic resources to promote knowledge transfer. In conclusion, we claim that SRL is far from “solved”, and its integration with other semantic tasks might enable significant improvements in the future, especially for the long tail of non-verbal predicates, thereby facilitating further research on SRL for non-verbal predicates. We release our software and datasets at https://github.com/sapienzanlp/exploring-srl.

pdf abs
DSPM-NLG: A Dual Supervised Pre-trained Model for Few-shot Natural Language Generation in Task-oriented Dialogue System
Yufan Wang | Bowei Zou | Rui Fan | Ai Ti Aw | Tingting He

In few-shot settings, fully conveying the semantic information of the dialogue act is a crucial challenge for Natural Language Generation (NLG) in the task-oriented dialogue system. An interesting fact is that NLG and Spoken Language Understanding (SLU) are a natural dual problem pair. Suppose the response generated by the NLG module can be restored to the corresponding dialogue act by the SLU module, which reflects that the generated response fully conveys the semantic information of the dialogue act. Based on this idea, a novel Dual Supervised Pre-trained Model for a few-shot Natural Language Generation (DSPM-NLG) is proposed to regularize the pre-training process. We adopt a joint model with a dual supervised framework to learn the dual correlation between NLG and SLU from the perspective of probability. In addition, a slot-masked strategy is designed to enable the model to focus better on the key slot-value pairs. DSPM-NLG is continuously trained on existing public large-scale annotated data, which thoroughly learns the duality between two tasks to enhance the semantically controlling and generalization abilities of the pre-trained model. Experiments demonstrate that our proposed model performs outstandingly on the few-shot benchmark dataset and outperforms the previous SOTA results.

pdf abs
TEPrompt: Task Enlightenment Prompt Learning for Implicit Discourse Relation Recognition
Wei Xiang | Chao Liang | Bang Wang

Implicit Discourse Relation Recognition (IDRR) aims at classifying the relation sense between two arguments without an explicit connective. Recently, the ConnPrompt (Xiang et al., 2022) has leveraged the powerful prompt learning for IDRR based on the fusion of multi-prompt decisions from three different yet much similar connective prediction templates. Instead of multi-prompt ensembling, we propose to design auxiliary tasks with enlightened prompt learning for the IDRR task. Although an auxiliary task is not used to directly output final prediction, we argue that during the joint training some of its learned features can be useful to boost the main task. In light of such motivations, we propose a task enlightenment prompt learning model, called TEPrompt, to fuse learned features from three related tasks for IDRR. In particular, the TEPrompt contains three tasks, viz., Discourse Relation Recognition (DRR), Sense Semantics Classification (SSC) and Annotated Connective Prediction (ACP), each with a unique prompt template and an answer space. In the training phase, we jointly train three prompt learning tasks with shared argument representation. In the testing phase, we only take the DRR output with fused features as the final IDRR decision. Experiments with the same conditions have shown that the proposed TEPrompt outperforms the ConnPrompt. This can be attributed to the promoted decision features and language models benefited from joint-training of auxiliary tasks.

pdf abs
Evaluating Factuality in Cross-lingual Summarization
Mingqi Gao | Wenqing Wang | Xiaojun Wan | Yuemei Xu

Cross-lingual summarization aims to help people efficiently grasp the core idea of the document written in a foreign language. Modern text summarization models generate highly fluent but often factually inconsistent outputs, which has received heightened attention in recent research. However, the factual consistency of cross-lingual summarization has not been investigated yet. In this paper, we propose a cross-lingual factuality dataset by collecting human annotations of reference summaries as well as generated summaries from models at both summary level and sentence level. Furthermore, we perform the fine-grained analysis and observe that over 50% of generated summaries and over 27% of reference summaries contain factual errors with characteristics different from monolingual summarization. Existing evaluation metrics for monolingual summarization require translation to evaluate the factuality of cross-lingual summarization and perform differently at different tasks and levels. Finally, we adapt the monolingual factuality metrics as an initial step towards the automatic evaluation of summarization factuality in cross-lingual settings. Our dataset and code are available at https://github.com/kite99520/Fact_CLS.

pdf abs
On the Correspondence between Compositionality and Imitation in Emergent Neural Communication
Emily Cheng | Mathieu Rita | Thierry Poibeau

Compositionality is a hallmark of human language that not only enables linguistic generalization, but also potentially facilitates acquisition. When simulating language emergence with neural networks, compositionality has been shown to improve communication performance; however, its impact on imitation learning has yet to be investigated. Our work explores the link between compositionality and imitation in a Lewis game played by deep neural agents. Our contributions are twofold: first, we show that the learning algorithm used to imitate is crucial: supervised learning tends to produce more average languages, while reinforcement learning introduces a selection pressure toward more compositional languages. Second, our study reveals that compositional languages are easier to imitate, which may induce the pressure toward compositional languages in RL imitation settings.

We demonstrate that coreference resolution in procedural texts is significantly improved when performing transformation-based entity linking prior to coreference relation identification. When events in the text introduce changes to the state of participating entities, it is often impossible to accurately link entities in anaphoric and coreference relations without an understanding of the transformations those entities undergo. We show how adding event semantics helps to better model entity coreference. We argue that all transformation predicates, not just creation verbs, introduce a new entity into the discourse, as a kind of generalized Result Role, which is typically not textually mentioned. This allows us to model procedural texts as process graphs and to compute the coreference type for any two entities in the recipe. We present our annotation methodology and the corpus generated as well as describe experiments on coreference resolution of entity mentions under a process-oriented model of events.

pdf abs
Why Does Zero-Shot Cross-Lingual Generation Fail? An Explanation and a Solution
Tianjian Li | Kenton Murray

Zero-shot cross-lingual transfer is when a multilingual model is trained to perform a task in one language and then is applied to another language. Although the zero-shot cross-lingual transfer approach has achieved success in various classification tasks, its performance on natural language generation tasks falls short in quality and sometimes outputs an incorrect language. In our study, we show that the fine-tuning process learns language invariant representations, which is beneficial for classification tasks but harmful for generation tasks. Motivated by this, we propose a simple method to regularize the model from learning language invariant representations and a method to select model checkpoints without a development set in the target language, both resulting in better generation quality. Experiments on three semantically diverse generation tasks show that our method reduces the accidental translation problem by 68% and improves the ROUGE-L score by 1.5 on average.

In this paper, we address the task of cloze-style multiple choice question (MCQs) distractor generation. Our study is featured by the following designs. First, we propose to formulate the cloze distractor generation as a Text2Text task. Second, we propose pseudo Kullback-Leibler Divergence for regulating the generation to consider the item discrimination index in education evaluation. Third, we explore the candidate augmentation strategy and multi-tasking training with cloze-related tasks to further boost the generation performance. Through experiments with benchmarking datasets, our best perfomring model advances the state-of-the-art result from 10.81 to 22.00 (p@1 score).

pdf abs
Lexical Translation Inconsistency-Aware Document-Level Translation Repair
Zhen Zhang | Junhui Li | Shimin Tao | Hao Yang

Following the idea of “one translation per discourse”, in this paper we aim to improve translation consistency via document-level translation repair (DocRepair), i.e., automatic post-editing on translations of documents. To this end, we propose a lexical translation inconsistency-aware DocRepair to explicitly model translation inconsistency. First we locate the inconsistency in automatic translation. Then we provide translation candidates for those inconsistency. Finally, we propose lattice-like input to properly model inconsistent tokens and phrases and their candidates. Experimental results on three document-level translation datasets show that based on G-Transformer, a state-of-the-art document-to-document (Doc2Doc) translation model, our Doc2Doc DocRepair achieves significant improvement on translation quality in BLEU scores, but also greatly improves lexical translation consistency.

Despite their widespread adoption, neural conversation models have yet to exhibit natural chat capabilities with humans. In this research, we examine user utterances as causes and generated responses as effects, recognizing that changes in a cause should produce a different effect. To further explore this concept, we have compiled and expanded upon a new dataset called CausalDialogue through crowd-sourcing. This dataset includes multiple cause-effect pairs within a directed acyclic graph (DAG) structure. Our analysis reveals that traditional loss functions struggle to effectively incorporate the DAG structure, leading us to propose a causality-enhanced method called Exponential Maximum Average Treatment Effect (ExMATE) to enhance the impact of causality at the utterance level in training neural conversation models. To evaluate the needs of considering causality in dialogue generation, we built a comprehensive benchmark on CausalDialogue dataset using different models, inference, and training methods. Through experiments, we find that a causality-inspired loss like ExMATE can improve the diversity and agility of conventional loss function and there is still room for improvement to reach human-level quality on this new dataset.

pdf abs
Towards Unified Spoken Language Understanding Decoding via Label-aware Compact Linguistics Representations
Zhihong Zhu | Xuxin Cheng | Zhiqi Huang | Dongsheng Chen | Yuexian Zou

Joint intent detection and slot filling models have shown promising success in recent years due to the high correlations between the two tasks. However, previous works independently decode the two tasks, which could result in misaligned predictions for both tasks. To address this shortcoming, we propose a novel method named Label-aware Compact Linguistics Representation (LCLR), which leverages label embeddings to jointly guide the decoding process. Concretely, LCLR projects both task-specific hidden states into a joint label latent space, where both task-specific hidden states could be concisely represented as linear combinations of label embeddings. Such feature decomposition of task-specific hidden states increases the representing power for the linguistics of utterance. Extensive experiments on two single- and multi-intent SLU benchmarks prove that LCLR can learn more discriminative label information than previous separate decoders, and consistently outperform previous state-of-the-art methods across all metrics. More encouragingly, LCLR can be applied to boost the performance of existing approaches, making it easy to be incorporated into any existing SLU models.

A human decision-maker benefits the most from an AI assistant that corrects for their biases. For problems such as generating interpretation of a radiology report given findings, a system predicting only highly likely outcomes may be less useful, where such outcomes are already obvious to the user. To alleviate biases in human decision-making, it is worth considering a broad differential diagnosis, going beyond the most likely options. We introduce a new task, “less likely brainstorming,” that asks a model to generate outputs that humans think are relevant but less likely to happen. We explore the task in two settings: a brain MRI interpretation generation setting and an everyday commonsense reasoning setting. We found that a baseline approach of training with less likely hypotheses as targets generates outputs that humans evaluate as either likely or irrelevant nearly half of the time; standard MLE training is not effective. To tackle this problem, we propose a controlled text generation method that uses a novel contrastive learning strategy to encourage models to differentiate between generating likely and less likely outputs according to humans. We compare our method with several state-of-the-art controlled text generation models via automatic and human evaluations and show that our models’ capability of generating less likely outputs is improved.

pdf abs
Language Modeling with Latent Situations
Belinda Z. Li | Maxwell Nye | Jacob Andreas

Language models (LMs) often generate incoherent outputs: they refer to events and entity states that are incompatible with the state of the world described in inputs. We introduce SITUATIONSUPERVISION, a family of approaches for improving coherence in LMs by training them to construct and condition on explicit representations of entities and their states. SITUATIONSUPERVISION has two components: an *auxiliary situation modeling* task that trains models to predict entity state representations in context, and a *latent state inference* procedure that imputes these states from partially annotated training data. SITUATIONSUPERVISION can be applied via fine-tuning (by supervising LMs to encode state variables in their hidden representations) and prompting (by inducing LMs to interleave textual descriptions of entity states with output text). In both cases, it requires only a small number of state annotations to produce substantial coherence improvements (up to an 16% reduction in errors), showing that standard LMs can be efficiently adapted to explicitly model language and aspects of its meaning.

pdf abs
Can Cross-Lingual Transferability of Multilingual Transformers Be Activated Without End-Task Data?
Zewen Chi | Heyan Huang | Xian-Ling Mao

Pretrained multilingual Transformers have achieved great success in cross-lingual transfer learning. Current methods typically activate the cross-lingual transferability of multilingual Transformers by fine-tuning them on end-task data. However, the methods cannot perform cross-lingual transfer when end-task data are unavailable. In this work, we explore whether the cross-lingual transferability can be activated without end-task data. We propose a cross-lingual transfer method, named PlugIn-X. PlugIn-X disassembles monolingual and multilingual Transformers into sub-modules, and reassembles them to be the multilingual end-task model. After representation adaptation, PlugIn-X finally performs cross-lingual transfer in a plug-and-play style. Experimental results show that PlugIn-X successfully activates the cross-lingual transferability of multilingual Transformers without accessing end-task data. Moreover, we analyze how the cross-model representation alignment affects the cross-lingual transferability.

Inquiry conversation is a common form of conversation that aims to complete the investigation (e.g., court hearing, medical consultation and police interrogation) during which a series of focus shifts occurs. While many models have been proposed to generate a smooth response to a given conversation history, neglecting the focus can limit performance in inquiry conversation where the order of the focuses plays there a key role. In this paper, we investigate the problem of response generation in inquiry conversation by taking the focus into consideration. We propose a novel Focus-aware Response Generation (FRG) method by jointly optimizing a multi-level encoder and a set of focal decoders to generate several candidate responses that correspond to different focuses. Additionally, a focus ranking module is proposed to predict the next focus and rank the candidate responses. Experiments on two orthogonal inquiry conversation datasets (judicial, medical domain) demonstrate that our method generates results significantly better in automatic metrics and human evaluation compared to the state-of-the-art approaches.

pdf abs
A Hierarchical Explanation Generation Method Based on Feature Interaction Detection
Yiming Ju | Yuanzhe Zhang | Kang Liu | Jun Zhao

The opaqueness of deep NLP models has motivated efforts to explain how deep models predict. Recently, work has introduced hierarchical attribution explanations, which calculate attribution scores for compositional text hierarchically to capture compositional semantics. Existing work on hierarchical attributions tends to limit the text groups to a continuous text span, which we call the connecting rule. While easy for humans to read, limiting the attribution unit to a continuous span might lose important long-distance feature interactions for reflecting model predictions. In this work, we introduce a novel strategy for capturing feature interactions and employ it to build hierarchical explanations without the connecting rule. The proposed method can convert ubiquitous non-hierarchical explanations (e.g., LIME) into their corresponding hierarchical versions. Experimental results show the effectiveness of our approach in building high-quality hierarchical explanations.

pdf abs
Jointly Reparametrized Multi-Layer Adaptation for Efficient and Private Tuning
Umang Gupta | Aram Galstyan | Greg Ver Steeg

Efficient finetuning of pretrained language transformers is becoming increasingly prevalent for solving natural language processing tasks. While effective, it can still require a large number of tunable parameters. This can be a drawback for low-resource applications and training with differential-privacy constraints, where excessive noise may be introduced during finetuning. To this end, we propose a novel language transformer finetuning strategy that introduces task-specific parameters in multiple transformer layers. These parameters are derived from fixed random projections of a single trainable vector, enabling finetuning with significantly fewer parameters while maintaining performance. We achieve within 5% of full finetuning performance on GLUE tasks with as few as 4,100 parameters per task, outperforming other parameter-efficient finetuning approaches that use a similar number of per-task parameters. Besides, the random projections can be precomputed at inference, avoiding additional computational latency. All these make our method particularly appealing for low-resource applications. Finally, our method achieves the best or comparable utility compared to several recent finetuning methods when training with the same privacy constraints, underscoring its effectiveness and potential real-world impact.

Event skeleton generation, aiming to induce an event schema skeleton graph with abstracted event nodes and their temporal relations from a set of event instance graphs, is a critical step in the temporal complex event schema induction task. Existing methods effectively address this task from a graph generation perspective but suffer from noise-sensitive and error accumulation, e.g., the inability to correct errors while generating schema. We, therefore, propose a novel Diffusion Event Graph Model (DEGM) to address these issues. Our DEGM is the first workable diffusion model for event skeleton generation, where the embedding and rounding techniques with a custom edge-based loss are introduced to transform a discrete event graph into learnable latent representations. Furthermore, we propose a denoising training process to maintain the model’s robustness. Consequently, DEGM derives the final schema, where error correction is guaranteed by iteratively refining the latent representations during the schema generation process. Experimental results on three IED bombing datasets demonstrate that our DEGM achieves better results than other state-of-the-art baselines. Our code and data are available at https://github.com/zhufq00/EventSkeletonGeneration.

The generative retrieval model depends solely on the information encoded in its model parameters without external memory, its information capacity is limited and fixed. To overcome the limitation, we propose Nonparametric Decoding (Np Decoding) which can be applied to existing generative retrieval models. Np Decoding uses nonparametric contextualized vocab embeddings (external memory) rather than vanilla vocab embeddings as decoder vocab embeddings. By leveraging the contextualized vocab embeddings, the generative retrieval model is able to utilize both the parametric and nonparametric space. Evaluation over 9 datasets (8 single-hop and 1 multi-hop) in the document retrieval task shows that applying Np Decoding to generative retrieval models significantly improves the performance. We also show that Np Decoding is data- and parameter-efficient, and shows high performance in the zero-shot setting.

pdf abs
Aspect-aware Unsupervised Extractive Opinion Summarization
Haoyuan Li | Somnath Basu Roy Chowdhury | Snigdha Chaturvedi

Extractive opinion summarization extracts sentences from users’ reviews to represent the prevalent opinions about a product or service. However, the extracted sentences can be redundant and may miss some important aspects, especially for centroid-based extractive summarization models (Radev et al., 2004). To alleviate these issues, we introduce TokenCluster– a method for unsupervised extractive opinion summarization that automatically identifies the aspects described in the review sentences and then extracts sentences based on their aspects. It identifies the underlying aspects of the review sentences using roots of noun phrases and adjectives appearing in them. Empirical evaluation shows that TokenCluster improves aspect coverage in summaries and achieves strong performance on multiple opinion summarization datasets, for both general and aspect-specific summarization. We also perform extensive ablation and human evaluation studies to validate the design choices of our method. The implementation of our work is available at https://github.com/leehaoyuan/TokenCluster

To better handle long-tail cases in the sequence labeling (SL) task, in this work, we introduce graph neural networks sequence labeling (GNN-SL), which augments the vanilla SL model output with similar tagging examples retrieved from the whole training set. Since not all the retrieved tagging examples benefit the model prediction, we construct a heterogeneous graph, and leverage graph neural networks (GNNs) to transfer information between the retrieved tagging examples and the input word sequence. The augmented node which aggregates information from neighbors is used to do prediction. This strategy enables the model to directly acquire similar tagging examples and improves the general quality of predictions. We conduct a variety of experiments on three typical sequence labeling tasks: Named Entity Recognition (NER), Part of Speech Tagging (POS), and Chinese Word Segmentation (CWS) to show the significant performance of our GNN-SL. Notably, GNN-SL achieves SOTA results of 96.9 (+0.2) on PKU, 98.3 (+0.4) on CITYU, 98.5 (+0.2) on MSR, and 96.9 (+0.2) on AS for the CWS task, and resultscomparable to SOTA performances on NER datasets, and POS datasets.

pdf abs
Serial Contrastive Knowledge Distillation for Continual Few-shot Relation Extraction
Xinyi Wang | Zitao Wang | Wei Hu

Continual few-shot relation extraction (RE) aims to continuously train a model for new relations with few labeled training data, of which the major challenges are the catastrophic forgetting of old relations and the overfitting caused by data sparsity. In this paper, we propose a new model, namely SCKD, to accomplish the continual few-shot RE task. Specifically, we design serial knowledge distillation to preserve the prior knowledge from previous models and conduct contrastive learning with pseudo samples to keep the representations of samples in different relations sufficiently distinguishable. Our experiments on two benchmark datasets validate the effectiveness of SCKD for continual few-shot RE and its superiority in knowledge transfer and memory utilization over state-of-the-art models.

pdf abs
Revisiting the Architectures like Pointer Networks to Efficiently Improve the Next Word Distribution, Summarization Factuality, and Beyond
Haw-Shiuan Chang | Zonghai Yao | Alolika Gon | Hong Yu | Andrew McCallum

Is the output softmax layer, which is adopted by most language models (LMs), always the best way to compute the next word probability? Given so many attention layers in a modern transformer-based LM, are the pointer networks redundant nowadays? In this study, we discover that the answers to both questions are no. This is because the softmax bottleneck sometimes prevents the LMs from predicting the desired distribution and the pointer networks can be used to break the bottleneck efficiently. Based on the finding, we propose several softmax alternatives by simplifying the pointer networks and accelerating the word-by-word rerankers. In GPT-2, our proposals are significantly better and more efficient than mixture of softmax, a state-of-the-art softmax alternative. In summarization experiments, without very significantly decreasing its training/testing speed, our best method based on T5-Small improves factCC score by 2 points in CNN/DM and XSUM dataset, and improves MAUVE scores by 30% in BookSum paragraph-level dataset.

Pre-trained language models (PLMs) are known to improve the generalization performance of natural language understanding models by leveraging large amounts of data during the pre-training phase. However, the out-of-distribution (OOD) generalization problem remains a challenge in many NLP tasks, limiting the real-world deployment of these methods. This paper presents the first attempt at creating a unified benchmark named GLUE-X for evaluating OOD robustness in NLP models, highlighting the importance of OOD robustness and providing insights on how to measure the robustness of a model and how to improve it. The benchmark includes 13 publicly available datasets for OOD testing, and evaluations are conducted on 8 classic NLP tasks over 21 popularly used PLMs. Our findings confirm the need for improved OOD accuracy in NLP tasks, as significant performance degradation was observed in all settings compared to in-distribution (ID) accuracy.

pdf abs
Investigating the Saliency of Sentiment Expressions in Aspect-Based Sentiment Analysis
Joachim Wagner | Jennifer Foster

We examine the behaviour of an aspect-based sentiment classifier built by fine-tuning the BERT BASE model on the SemEval 2016 English dataset. In a set of masking experiments, we examine the extent to which the tokens identified as salient by LIME and a gradient-based method are being used by the classifier. We find that both methods are able to produce faithful rationales, with LIME outperforming the gradient-based method. We also identify a set of manually annotated sentiment expressions for this dataset, and carry out more masking experiments with these as human rationales. The enhanced performance of a classifier that only sees the relevant sentiment expressions suggests that they are not being used to their full potential. A comparison of the LIME and gradient rationales with the sentiment expressions reveals only a moderate level of agreement. Some disagreements are related to the fixed length of the rationales and the tendency of the rationales to contain content words related to the aspect itself.

pdf abs
DMLM: Descriptive Masked Language Modeling
Edoardo Barba | Niccolò Campolungo | Roberto Navigli

Over the last few years, Masked Language Modeling (MLM) pre-training has resulted in remarkable advancements in many Natural Language Understanding (NLU) tasks, which sparked an interest in researching alternatives and extensions to the MLM objective. In this paper, we tackle the absence of explicit semantic grounding in MLM and propose Descriptive Masked Language Modeling (DMLM), a knowledge-enhanced reading comprehension objective, where the model is required to predict the most likely word in a context, being provided with the word’s definition. For instance, given the sentence “I was going to the _”, if we provided as definition “financial institution”, the model would have to predict the word “bank”; if, instead, we provided “sandy seashore”, the model should predict “beach”. Our evaluation highlights the effectiveness of DMLM in comparison with standard MLM, showing improvements on a number of well-established NLU benchmarks, as well as other semantics-focused tasks, e.g., Semantic Role Labeling. Furthermore, we demonstrate how it is possible to take full advantage of DMLM to embed explicit semantics in downstream tasks, explore several properties of DMLM-based contextual representations and suggest a number of future directions to investigate.

pdf abs
Reproducibility in NLP: What Have We Learned from the Checklist?
Ian Magnusson | Noah A. Smith | Jesse Dodge

Scientific progress in NLP rests on the reproducibility of researchers’ claims. The *CL conferences created the NLP Reproducibility Checklist in 2020 to be completed by authors at submission to remind them of key information to include. We provide the first analysis of the Checklist by examining 10,405 anonymous responses to it. First, we find evidence of an increase in reporting of information on efficiency, validation performance, summary statistics, and hyperparameters after the Checklist’s introduction. Further, we show acceptance rate grows for submissions with more Yes responses. We find that the 44% of submissions that gather new data are 5% less likely to be accepted than those that did not; the average reviewer-rated reproducibility of these submissions is also 2% lower relative to the rest. We find that only 46% of submissions claim to open-source their code, though submissions that do have 8% higher reproducibility score relative to those that do not, the most for any item. We discuss what can be inferred about the state of reproducibility in NLP, and provide a set of recommendations for future conferences, including: a) allowing submitting code and appendices one week after the deadline, and b) measuring dataset reproducibility by a checklist of data collection practices.

pdf abs
Domain Generalization via Switch Knowledge Distillation for Robust Review Representation
You Zhang | Jin Wang | Liang-Chih Yu | Dan Xu | Xuejie Zhang

Applying neural models injected with in-domain user and product information to learn review representations of unseen or anonymous users incurs an obvious obstacle in content-based recommender systems. For the generalization of the in-domain classifier, most existing models train an extra plain-text model for the unseen domain. Without incorporating historical user and product information, such a schema makes unseen and anonymous users dissociate from the recommender system. To simultaneously learn the review representation of both existing and unseen users, this study proposed a switch knowledge distillation for domain generalization. A generalization-switch (GSwitch) model was initially applied to inject user and product information by flexibly encoding both domain-invariant and domain-specific features. By turning the status ON or OFF, the model introduced a switch knowledge distillation to learn a robust review representation that performed well for either existing or anonymous unseen users. The empirical experiments were conducted on IMDB, Yelp-2013, and Yelp-2014 by masking out users in test data as unseen and anonymous users. The comparative results indicate that the proposed method enhances the generalization capability of several existing baseline models. For reproducibility, the code for this paper is available at: https://github.com/yoyo-yun/DG_RRR.

pdf abs
On Search Strategies for Document-Level Neural Machine Translation
Christian Herold | Hermann Ney

Compared to sentence-level systems, document-level neural machine translation (NMT) models produce a more consistent output across a document and are able to better resolve ambiguities within the input. There are many works on document-level NMT, mostly focusing on modifying the model architecture or training strategy to better accommodate the additional context-input. On the other hand, in most works, the question on how to perform search with the trained model is scarcely discussed, sometimes not mentioned at all. In this work, we aim to answer the question how to best utilize a context-aware translation model in decoding. We start with the most popular document-level NMT approach and compare different decoding schemes, some from the literature and others proposed by us. In the comparison, we are using both, standard automatic metrics, as well as specific linguistic phenomena on three standard document-level translation benchmarks. We find that most commonly used decoding strategies perform similar to each other and that higher quality context information has the potential to further improve the translation.

pdf abs
Causal Intervention for Mitigating Name Bias in Machine Reading Comprehension
Jiazheng Zhu | Shaojuan Wu | Xiaowang Zhang | Yuexian Hou | Zhiyong Feng

Machine Reading Comprehension (MRC) is to answer questions based on a given passage, which has made great achievements using pre-trained Language Models (LMs). We study the robustness of MRC models to names which is flexible and repeatability. MRC models based on LMs may overuse the name information to make predictions, which causes the representation of names to be non-interchangeable, called name bias. In this paper, we propose a novel Causal Interventional paradigm for MRC (CI4MRC) to mitigate name bias. Specifically, we uncover that the pre-trained knowledge concerning names is indeed a confounder by analyzing the causalities among the pre-trained knowledge, context representation and answers based on a Structural Causal Model (SCM). We develop effective CI4MRC algorithmic implementations to constrain the confounder based on the neuron-wise and token-wise adjustments. Experiments demonstrate that our proposed CI4MRC effectively mitigates the name bias and achieves competitive performance on the original SQuAD. Moreover, our method is general to various pre-trained LMs and performs robustly on the adversarial datasets.

pdf abs
Counterfactual Probing for the Influence of Affect and Specificity on Intergroup Bias
Venkata Subrahmanyan Govindarajan | David Beaver | Kyle Mahowald | Junyi Jessy Li

While existing work on studying bias in NLP focues on negative or pejorative language use, Govindarajan et al. (2023) offer a revised framing of bias in terms of intergroup social context, and its effects on language behavior. In this paper, we investigate if two pragmatic features (specificity and affect) systematically vary in different intergroup contexts — thus connecting this new framing of bias to language output. Preliminary analysis finds modest correlations between specificity and affect of tweets with supervised intergroup relationship (IGR) labels. Counterfactual probing further reveals that while neural models finetuned for predicting IGR reliably use affect in classification, the model’s usage of specificity is inconclusive.

pdf abs
SongRewriter: A Chinese Song Rewriting System with Controllable Content and Rhyme Scheme
Yusen Sun | Liangyou Li | Qun Liu | Dit-Yan Yeung

Although lyrics generation has achieved significant progress in recent years, it has limited practical applications because the generated lyrics cannot be performed without composing compatible melodies. In this work, we bridge this practical gap by proposing a song rewriting system which rewrites the lyrics of an existing song such that the generated lyrics are compatible with the rhythm of the existing melody and thus singable. In particular, we propose SongRewriter, a controllable Chinese lyric generation and editing system which assists users without prior knowledge of melody composition. The system is trained by a randomized multi-level masking strategy which produces a unified model for generating entirely new lyrics or editing a few fragments. To improve the controllabiliy of the generation process, we further incorporate a keyword prompt to control the lexical choices of the content and propose novel decoding constraints and a vowel modeling task to enable flexible end and internal rhyme schemes. While prior rhyming metrics are mainly for rap lyrics, we propose three novel rhyming evaluation metrics for song lyrics. Both automatic and human evaluations show that the proposed model performs better than the state-of-the-art models in both contents and rhyming quality.

pdf abs
Triplet-Free Knowledge-Guided Response Generation
Dongming Li | Jianfeng Liu | Baoyuan Wang

Generating vivid and informative responses (e.g., comments for social posts and utterances for dialogues) is challenging without giving relevant knowledge. Prior works focus on constructing the ”latent” knowledge first and then learning how to ”ground” it based on pseudo (context, knowledge, response) triplets. However, the retrieval between real responses and their latent knowledge is difficult in nature. In this paper, instead of focusing on how to ground knowledge given the responses, we take a different perspective to optimize the final responses for given guided knowledge directly. This allows us to re-formulate the entire problem in a simplified yet more scalable way. Specifically, we pretrain a response language model (LM) to measure the relevance and consistency between any context and response, then use search engines to collect the top-ranked passages to serve as the guiding knowledge without explicitly optimizing the ‘‘best” latent knowledge that corresponds to a given response. The final response generation model is trained through reinforcement learning by taking both the response LM prior and knowledge-injection rate as rewards. For better evaluations, we construct a new Chinese benchmark, ”IceKC”, using fresh multimodal online social posts. Both automatic evaluations and human evaluations show our zero-resource approach performs significantly better than prior works.

pdf abs
Implicit Memory Transformer for Computationally Efficient Simultaneous Speech Translation
Matthew Raffel | Lizhong Chen

Simultaneous speech translation is an essential communication task difficult for humans whereby a translation is generated concurrently with oncoming speech inputs. For such a streaming task, transformers using block processing to break an input sequence into segments have achieved state-of-the-art performance at a reduced cost. Current methods to allow information to propagate across segments, including left context and memory banks, have faltered as they are both insufficient representations and unnecessarily expensive to compute. In this paper, we propose an Implicit Memory Transformer that implicitly retains memory through a new left context method, removing the need to explicitly represent memory with memory banks. We generate the left context from the attention output of the previous segment and include it in the keys and values of the current segment’s attention calculation. Experiments on the MuST-C dataset show that the Implicit Memory Transformer provides a substantial speedup on the encoder forward pass with nearly identical translation quality when compared with the state-of-the-art approach that employs both left context and memory banks.

pdf abs
Enhancing Document-level Event Argument Extraction with Contextual Clues and Role Relevance
Wanlong Liu | Shaohuan Cheng | Dingyi Zeng | Qu Hong

Document-level event argument extraction poses new challenges of long input and cross-sentence inference compared to its sentence-level counterpart. However, most prior works focus on capturing the relations between candidate arguments and the event trigger in each event, ignoring two crucial points: a) non-argument contextual clue information; b) the relevance among argument roles. In this paper, we propose a SCPRG (Span-trigger-based Contextual Pooling and latent Role Guidance) model, which contains two novel and effective modules for the above problem. The Span-Trigger-based Contextual Pooling (STCP) adaptively selects and aggregates the information of non-argument clue words based on the context attention weights of specific argument-trigger pairs from pre-trained model. The Role-based Latent Information Guidance (RLIG) module constructs latent role representations, makes them interact through role-interactive encoding to capture semantic relevance, and merges them into candidate arguments. Both STCP and RLIG introduce no more than 1% new parameters compared with the base model and can be easily applied to other event extraction models, which are compact and transplantable. Experiments on two public datasets show that our SCPRG outperforms previous state-of-the-art methods, with 1.13 F1 and 2.64 F1 improvements on RAMS and WikiEvents respectively. Further analyses illustrate the interpretability of our model.

pdf abs
Exploring the Impact of Vision Features in News Image Captioning
Junzhe Zhang | Xiaojun Wan

The task of news image captioning aims to generate a detailed caption which describes the specific information of an image in a news article. However, we find that recent state-of-art models can achieve competitive performance even without vision features. To resolve the impact of vision features in the news image captioning task, we conduct extensive experiments with mainstream models based on encoder-decoder framework. From our exploration, we find 1) vision features do contribute to the generation of news image captions; 2) vision features can assist models to better generate entities of captions when the entity information is sufficient in the input textual context of the given article; 3) Regions of specific objects in images contribute to the generation of related entities in captions.

pdf abs
Using Collostructional Analysis to evaluate BERT’s representation of linguistic constructions
Tim Veenboer | Jelke Bloem

Collostructional analysis is a technique devised to find correlations between particular words and linguistic constructions in order to analyse meaning associations of these constructions. Contrasting collostructional analysis results with output from BERT might provide insights into the way BERT represents the meaning of linguistic constructions. This study tests to what extent English BERT’s meaning representations correspond to known constructions from the linguistics literature by means of two tasks that we propose. Firstly, by predicting the words that can be used in open slots of constructions, the meaning associations of more lexicalized constructions can be observed. Secondly, by finding similar sequences using BERT’s output embeddings and manually reviewing the resulting sentences, we can observe whether instances of less lexicalized constructions are clustered together in semantic space. These two methods show that BERT represents constructional meaning to a certain extent, but does not separate instances of a construction from a near-synonymous construction that has a different form.

Large Language Models (LLMs) have in recent years demonstrated impressive prowess in natural language generation. A common practice to improve generation diversity is to sample multiple outputs from the model. However, partly due to the inaccessibility of LLMs, there lacks a simple and robust way of selecting the best output from these stochastic samples. As a case study framed in the context of question generation, we propose two prompt-based approaches, namely round-trip and prompt-based score, to selecting high-quality questions from a set of LLM-generated candidates. Our method works without the need to modify the underlying model, nor does it rely on human-annotated references — both of which are realistic constraints for real-world deployment of LLMs. With automatic as well as human evaluations, we empirically demonstrate that our approach can effectively select questions of higher qualities than greedy generation.

pdf abs
Sentiment Knowledge Enhanced Self-supervised Learning for Multimodal Sentiment Analysis
Fan Qian | Jiqing Han | Yongjun He | Tieran Zheng | Guibin Zheng

Multimodal Sentiment Analysis (MSA) has made great progress that benefits from extraordinary fusion scheme. However, there is a lack of labeled data, resulting in severe overfitting and poor generalization for supervised models applied in this field. In this paper, we propose Sentiment Knowledge Enhanced Self-supervised Learning (SKESL) to capture common sentimental patterns in unlabeled videos, which facilitates further learning on limited labeled data. Specifically, with the help of sentiment knowledge and non-verbal behavior, SKESL conducts sentiment word masking and predicts fine-grained word sentiment intensity, so as to embed sentiment information at the word level into pre-trained multimodal representation. In addition, a non-verbal injection method is also proposed to integrate non-verbal information into the word semantics. Experiments on two standard benchmarks of MSA clearly show that SKESL significantly outperforms the baseline, and achieves new State-Of-The-Art (SOTA) results.

pdf abs
Theory of Mind in Freely-Told Children’s Narratives: A Classification Approach
Bram van Dijk | Marco Spruit | Max van Duijn

Children are the focal point for studying the link between language and Theory of Mind (ToM) competence. Language and ToM are often studied with younger children and standardized tests, but as both are social competences, data and methods with higher ecological validity are critical. We leverage a corpus of 442 freely-told stories by Dutch children aged 4-12, recorded in their everyday classroom environments, to study language and ToM with NLP-tools. We labelled stories according to the mental depth of story characters children create, as a proxy for their ToM competence ‘in action’, and built a classifier with features encoding linguistic competences identified in existing work as predictive of ToM.We obtain good and fairly robust results (F1-macro = .71), relative to the complexity of the task for humans. Our results are explainable in that we link specific linguistic features such as lexical complexity and sentential complementation, that are relatively independent of children’s ages, to higher levels of character depth. This confirms and extends earlier work, as our study includes older children and socially embedded data from a different domain. Overall, our results support the idea that language and ToM are strongly interlinked, and that in narratives the former can scaffold the latter.

pdf abs
Better Language Models of Code through Self-Improvement
Hung To | Nghi Bui | Jin L.C. Guo | Tien Nguyen

Pre-trained language models for code (PLMCs) have gained attention in recent research. These models are pre-trained on large-scale datasets using multi-modal objectives. However, fine-tuning them requires extensive supervision and is limited by the size of the dataset provided. We aim to improve this issue by proposing a data augmentation framework using knowledge distillation. Our framework utilizes knowledge gained during the pre-training and fine-tuning stage to augment training data, which is then used for the next step. We incorporate this framework into the state-of-the-art language models, such as CodeT5, CodeBERT, and UnixCoder. The results show that our framework significantly improves PLMCs’ performance in sequence-generation tasks, such as code summarization and code generation in the CodeXGLUE benchmark.

BIG-Bench (Srivastava et al., 2022) is a diverse evaluation suite that focuses on tasks believed to be beyond the capabilities of current language models. Language models have already made good progress on this benchmark, with the best model in the BIG-Bench paper outperforming average reported human-rater results on 65% of the BIG-Bench tasks via few-shot prompting. But on what tasks do language models fall short of average human-rater performance, and are those tasks actually unsolvable by current language models? In this work, we focus on a suite of 23 challenging BIG-Bench tasks which we call BIG-Bench Hard (BBH). These are the tasks for which prior language model evaluations did not outperform the average human-rater. We find that applying chain-of-thought (CoT) prompting to BBH tasks enables PaLM to surpass the average human-rater performance on 10 of the 23 tasks, and Codex (code-davinci-002) to surpass the average human-rater performance on 17 of the 23 tasks. Since many tasks in BBH require multi-step reasoning, few-shot prompting without CoT, as done in the BIG-Bench evaluations (Srivastava et al., 2022), substantially underestimates the best performance and capabilities of language models, which is better captured via CoT prompting. As further analysis, we explore the interaction between CoT and model scale on BBH, finding that CoT enables emergent task performance on several BBH tasks with otherwise flat scaling curves.

pdf abs
Score It All Together: A Multi-Task Learning Study on Automatic Scoring of Argumentative Essays
Yuning Ding | Marie Bexte | Andrea Horbach

When scoring argumentative essays in an educational context, not only the presence or absence of certain argumentative elements but also their quality is important. On the recently published student essay dataset PERSUADE, we first show that the automatic scoring of argument quality benefits from additional information about context, writing prompt and argument type. We then explore the different combinations of three tasks: automated span detection, type and quality prediction. Results show that a multi-task learning approach combining the three tasks outperforms sequential approaches that first learn to segment and then predict the quality/type of a segment.

pdf abs
Data Sampling and (In)stability in Machine Translation Evaluation
Chi-kiu Lo | Rebecca Knowles

We analyze the different data sampling approaches used in selecting data for human evaluation and ranking of machine translation systems at the highly influential Conference on Machine Translation (WMT). By using automatic evaluation metrics, we are able to focus on the impact of the data sampling procedure as separate from questions about human annotator consistency. We provide evidence that the latest data sampling approach used at WMT skews the annotated data toward shorter documents, not necessarily representative of the full test set. Lastly, we examine a new data sampling method that uses the available labour budget to sample data in a more representative manner, with the goals of improving representation of various document lengths in the sample and producing more stable rankings of system translation quality.

Argument pair extraction (APE) aims to extract interactive argument pairs from two passages within a discussion. The key challenge of APE is to effectively capture the complex context-aware interactive relations of arguments between the two passages. In this paper, we elicit relational semantic knowledge from large-scale pre-trained language models (PLMs) via a probing technique. The induced sentence-level relational probing graph can help capture rich explicit interactive relations between argument pairs effectively. Since the relevance score of a sentence pair within a passage is generally larger than that of the sentence pair from different passages, each sentence would prefer to propagate information within the same passage and under-explore the interactive relations between two passages. To tackle this issue, we propose a graph decomposition method to decompose the probing graph into four sub-graphs from intra- and inter-passage perspectives, where the intra-passage graphs can help detect argument spans within each passage and the inter-passage graphs can help identify the argument pairs between the review and rebuttal passages. Experimental results on two benchmark datasets show that our method achieves substantial improvements over strong baselines for APE.

pdf abs
DiffuSum: Generation Enhanced Extractive Summarization with Diffusion
Haopeng Zhang | Xiao Liu | Jiawei Zhang

Extractive summarization aims to form a summary by directly extracting sentences from the source document. Existing works mostly formulate it as a sequence labeling problem by making individual sentence label predictions. This paper proposes DiffuSum, a novel paradigm for extractive summarization, by directly generating the desired summary sentence representations with diffusion models and extracting sentences based on sentence representation matching. In addition, DiffuSum jointly optimizes a contrastive sentence encoder with a matching loss for sentence representation alignment and a multi-class contrastive loss for representation diversity. Experimental results show that DiffuSum achieves the new state-of-the-art extractive results on CNN/DailyMail with ROUGE scores of 44.83/22.56/40.56. Experiments on the other two datasets with different summary lengths and cross-dataset evaluation also demonstrate the effectiveness of DiffuSum. The strong performance of our framework shows the great potential of adapting generative models for extractive summarization.

This paper explores the task of Temporal Video Grounding (TVG) where, given an untrimmed video and a query sentence, the goal is to recognize and determine temporal boundaries of action instances in the video described by natural language queries. Recent works tackled this task by improving query inputs with large pre-trained language models (PLM), at the cost of more expensive training. However, the effects of this integration are unclear, as these works also propose improvements in the visual inputs. Therefore, this paper studies the role of query sentence representation with PLMs in TVG and assesses the applicability of parameter-efficient training with NLP adapters. We couple popular PLMs with a selection of existing approaches and test different adapters to reduce the impact of the additional parameters. Our results on three challenging datasets show that, with the same visual inputs, TVG models greatly benefited from the PLM integration and fine-tuning, stressing the importance of the text query representation in this task. Furthermore, adapters were an effective alternative to full fine-tuning, even though they are not tailored to our task, allowing PLM integration in larger TVG models and delivering results comparable to SOTA models. Finally, our results shed light on which adapters work best in different scenarios.

pdf abs
A Memory Model for Question Answering from Streaming Data Supported by Rehearsal and Anticipation of Coreference Information
Vladimir Araujo | Alvaro Soto | Marie-Francine Moens

Existing question answering methods often assume that the input content (e.g., documents or videos) is always accessible to solve the task. Alternatively, memory networks were introduced to mimic the human process of incremental comprehension and compression of the information in a fixed-capacity memory. However, these models only learn how to maintain memory by backpropagating errors in the answers through the entire network. Instead, it has been suggested that humans have effective mechanisms to boost their memorization capacities, such as rehearsal and anticipation. Drawing inspiration from these, we propose a memory model that performs rehearsal and anticipation while processing inputs to memorize important information for solving question answering tasks from streaming data. The proposed mechanisms are applied self-supervised during training through masked modeling tasks focused on coreference information. We validate our model on a short-sequence (bAbI) dataset as well as large-sequence textual (NarrativeQA) and video (ActivityNet-QA) question answering datasets, where it achieves substantial improvements over previous memory network approaches. Furthermore, our ablation study confirms the proposed mechanisms’ importance for memory models.

Attribute Value Extraction (AVE) boosts many e-commerce platform services such as targeted recommendation, product retrieval and question answering. Most previous studies adopt an extractive framework such as named entity recognition (NER) to capture subtokens in the product descriptions as the corresponding values of target attributes. However, in the real world scenario, there also exist implicit attribute values that are not mentioned explicitly but embedded in the image information and implied text meaning of products, for which the power of extractive methods is severely constrained. To address the above issues, we exploit a unified multi-modal AVE framework named DEFLATE (a multi-modal unifieD framEwork For impLicit And expliciT AVE) to acquire implicit attribute values in addition to the explicit ones. DEFLATE consists of a QA-based generation model to produce candidate attribute values from the product information of different modalities, and a discriminative model to ensure the credibility of the generated answers. Meanwhile, to provide a testbed that close to the real world, we collect and annotate a multi-modal dataset with parts of implicit attribute values. Extensive experiments conducted on multiple datasets demonstrate that DEFLATE significantly outperforms previous methods on the extraction of implicit attribute values, while achieving comparable performance for the explicit ones.

pdf abs
CoRRPUS: Code-based Structured Prompting for Neurosymbolic Story Understanding
Yijiang River Dong | Lara J. Martin | Chris Callison-Burch

Story generation and understanding—as with all NLG/NLU tasks—has seen a surge in neurosymbolic work. Researchers have recognized that, while large language models (LLMs) have tremendous utility, they can be augmented with symbolic means to be even better and to make up for many flaws that neural networks have. However, symbolic methods are extremely costly in terms of the amount of time and expertise needed to create them. In this work, we capitalize on state-of-the-art Code-LLMs, such as Codex, to bootstrap the use of symbolic methods for tracking the state of stories and aiding in story understanding. We show that our CoRRPUS system and abstracted prompting procedures can beat current state-of-the-art structured LLM techniques on pre-existing story understanding tasks (bAbI Task 2 and Re³) with minimal hand engineering. This work highlights the usefulness of code-based symbolic representations for enabling LLMs to better perform story reasoning tasks.

pdf abs
Fighting Bias With Bias: Promoting Model Robustness by Amplifying Dataset Biases
Yuval Reif | Roy Schwartz

NLP models often rely on superficial cues known as dataset biases to achieve impressive performance, and can fail on examples where these biases do not hold. Recent work sought to develop robust, unbiased models by filtering biased examples from training sets. In this work, we argue that such filtering can obscure the true capabilities of models to overcome biases, which might never be removed in full from the dataset. We suggest that in order to drive the development of models robust to subtle biases, dataset biases should be amplified in the training set. We introduce an evaluation framework defined by a bias-amplified training set and an anti-biased test set, both automatically extracted from existing datasets. Experiments across three notions of bias, four datasets and two models show that our framework is substantially more challenging for models than the original data splits, and even more challenging than hand-crafted challenge sets. Our evaluation framework can use any existing dataset, even those considered obsolete, to test model robustness. We hope our work will guide the development of robust models that do not rely on superficial biases and correlations. To this end, we publicly release our code and data.

pdf abs
Context-Aware Document Simplification
Liam Cripwell | Joël Legrand | Claire Gardent

To date, most work on text simplification has focused on sentence-level inputs. Early attempts at document simplification merely applied these approaches iteratively over the sentences of a document. However, this fails to coherently preserve the discourse structure, leading to suboptimal output quality. Recently, strategies from controllable simplification have been leveraged to achieve state-of-the-art results on document simplification by first generating a document-level plan (a sequence of sentence-level simplification operations) and using this plan to guide sentence-level simplification downstream. However, this is still limited in that the simplification model has no direct access to the local inter-sentence document context, likely having a negative impact on surface realisation. We explore various systems that use document context within the simplification process itself, either by iterating over larger text units or by extending the system architecture to attend over a high-level representation of document context. In doing so, we achieve state-of-the-art performance on the document simplification task, even when not relying on plan-guidance. Further, we investigate the performance and efficiency tradeoffs of system variants and make suggestions of when each should be preferred.

Existing knowledge-enhanced methods have achieved remarkable results in certain Q&A tasks via obtaining diverse knowledge from different knowledge bases. However, limited by the properties of retrieved knowledge, they still have trouble benefiting from both the knowledge relevance and distinguishment simultaneously. To address the challenge, we propose CPACE, a Concept-centric Prompt-bAsed Contrastive Explanation Generation model, which aims to convert obtained symbolic knowledge into the contrastive explanation for better distinguishing the differences among given candidates. Firstly, following previous works, we retrieve different types of symbolic knowledge with a concept-centric knowledge extraction module. After that, we generate corresponding contrastive explanation using acquired symbolic knowledge and prompt as guidance for better modeling the knowledge distinguishment and interpretability. Finally, we regard the generated contrastive explanation as external knowledge for downstream task enhancement. We conduct a series of experiments on three widely-used question-answering datasets: CSQA, QASC, and OBQA. Experimental results demonstrate that with the help of generated contrastive explanation, our CPACE model achieves new SOTA on CSQA (89.8% on the testing set, 0.9% higher than human performance), and gains impressive improvement on QASC and OBQA (4.2% and 3.5%, respectively).

pdf abs
Abstract then Play: A Skill-centric Reinforcement Learning Framework for Text-based Games
Anjie Zhu | Peng-Fei Zhang | Yi Zhang | Zi Huang | Jie Shao

Text-based games present an exciting test-bed for reinforcement learning algorithms in the natural language environment. In these adventure games, an agent must learn to interact with the environment through text in order to accomplish tasks, facing large and combinational action space as well as partial observability issues. However, existing solutions fail to decompose the task and abstract the action autonomously, which either pre-specify the subtasks or pre-train on the human gameplay dataset. In this work, we introduce a novel skill-centric reinforcement learning framework, which is capable of abstracting the action in an end-to-end manner. To learn a more disentangled skill, we focus on the informativeness and distinguishability of the skill in accordance with the information bottleneck principle. Specifically, we introduce a discriminator to enable the skill to reflect the trajectory and push their representations onto the unit hypersphere to distribute uniformly. Moreover, a self-predictive mechanism is employed to learn inverse and forward dynamics, and a self-recovery mechanism is leveraged to refine the action representation, thus resulting in a more comprehensive perception of dynamics and more effective representations of textual state and action. Empirical experiments are carried out on the Jericho environment and the results validate the superiority against state-of-the-art baselines.

Conversational search has been regarded as the next-generation search paradigm. Constrained by data scarcity, most existing methods distill the well-trained ad-hoc retriever to the conversational retriever. However, these methods, which usually initialize parameters by query reformulation to discover contextualized dependency, have trouble in understanding the dialogue structure information and struggle with contextual semantic vanishing. In this paper, we propose {pasted macro ‘FULLMODEL’} ({pasted macro ‘MODEL’}) which is a new post-training paradigm with three self-supervised tasks to efficiently initialize the conversational search model to enhance the dialogue structure and contextual semantic understanding. Furthermore, the {pasted macro ‘MODEL’} can be plugged into most of the existing conversational models to boost their performance. To verify the effectiveness of our proposed method, we apply the conversational encoder post-trained by {pasted macro ‘MODEL’} on the conversational search task using two benchmark datasets: CAsT-19 and CAsT-20.Extensive experiments that our {pasted macro ‘MODEL’} can boost the performance of several existing conversational search methods. Our source code is available at https://github.com/morecry/SSP.

pdf abs
Towards Reference-free Text Simplification Evaluation with a BERT Siamese Network Architecture
Xinran Zhao | Esin Durmus | Dit-Yan Yeung

Text simplification (TS) aims to modify sentences to make their both content and structure easier to understand. Traditional n-gram matching-based TS evaluation metrics heavily rely on the exact token match and human-annotated simplified sentences. In this paper, we present a novel neural-network-based reference-free TS metric BETS that leverages pre-trained contextualized language representation models and large-scale paraphrasing datasets to evaluate simplicity and meaning preservation. We show that our metric, without collecting any costly human simplification reference, correlates better than existing metrics with human judgments for the quality of both overall simplification (+7.7%) and its key aspects, i.e., comparative simplicity (+11.2%) and meaning preservation (+9.2%).

pdf abs
Causal interventions expose implicit situation models for commonsense language understanding
Takateru Yamakoshi | James McClelland | Adele Goldberg | Robert Hawkins

Accounts of human language processing have long appealed to implicit “situation models” that enrich comprehension with relevant but unstated world knowledge. Here, we apply causal intervention techniques to recent transformer models to analyze performance on the Winograd Schema Challenge (WSC), where a single context cue shifts interpretation of an ambiguous pronoun. We identify a relatively small circuit of attention heads that are responsible for propagating information from the context word that guides which of the candidate noun phrases the pronoun ultimately attends to. We then compare how this circuit behaves in a closely matched “syntactic” control where the situation model is not strictly necessary. These analyses suggest a distinct pathway through which implicit situation models may be constructed to guide pronoun resolution

Unsupervised domain adaptation of machine translation, which adapts a pre-trained translation model to a specific domain without in-domain parallel data, has drawn extensive attention in recent years. However, most existing methods focus on the fine-tuning based techniques, which is non-extensible. In this paper, we propose a new method to perform unsupervised domain adaptation in a non-parametric manner. Our method only resorts to in-domain monolingual data, and we jointly perform nearest neighbour inference on both forward and backward translation directions. The forward translation model creates nearest neighbour datastore for the backward direction, and vice versa, strengthening each other in an iterative style. Experiments on multi-domain datasets demonstrate that our method significantly improves the in-domain translation performance and achieves state-of-the-art results among non-parametric methods.

pdf abs
PruMUX: Augmenting Data Multiplexing with Model Compression
Yushan Su | Vishvak Murahari | Karthik Narasimhan | Kai Li

As language models increase in size by the day, methods for efficient inference are critical to leveraging their capabilities for various applications. Prior work has investigated techniques like model pruning, knowledge distillation, and data multiplexing to increase model throughput without sacrificing accuracy. In this paper, we combine two such methods – structured pruning and data multiplexing – to compound the speedup gains obtained by either method. Our approach, PruMUX, obtains up to 7.5-29.5X throughput improvement over BERT-base model with accuracy threshold from 80% to 74%. We further study various combinations of parameters (such as sparsity and multiplexing factor) in the two techniques to provide a comprehensive analysis of the tradeoff between accuracy and throughput in the resulting models. We then propose Auto-PruMUX, a meta-level model that can predict the high-performance parameters for pruning and multiplexing given a desired accuracy loss budget, providing a practical method to leverage the combination effectively.

pdf abs
With Prejudice to None: A Few-Shot, Multilingual Transfer Learning Approach to Detect Social Bias in Low Resource Languages
Nihar Sahoo | Niteesh Mallela | Pushpak Bhattacharyya

In this paper, we describe our work on social bias detection in a low-resource multilingual setting in which the languages are from two very divergent families- Indo-European (English, Hindi, and Italian) and Altaic (Korean). Currently, the majority of the social bias datasets available are in English and this inhibits progress on social bias detection in low-resource languages. To address this problem, we introduce a new dataset for social bias detection in Hindi and investigate multilingual transfer learning using publicly available English, Italian, and Korean datasets. The Hindi dataset contains 9k social media posts annotated for (i) binary bias labels (bias/neutral), (ii) binary labels for sentiment (positive/negative), (iii) target groups for each bias category, and (iv) rationale for annotated bias labels (a short piece of text). We benchmark our Hindi dataset using different multilingual models, with XLM-R achieving the best performance of 80.8 macro-F1 score. Our results show that the detection of social biases in resource-constrained languages such as Hindi and Korean may be improved with the use of a similar dataset in English. We also show that translating all datasets into English does not work effectively for detecting social bias, since the nuances of source language are lost in translation. All the scripts and datasets utilized in this study will be publicly available.

pdf abs
Don’t Lose Yourself! Empathetic Response Generation via Explicit Self-Other Awareness
Weixiang Zhao | Yanyan Zhao | Xin Lu | Bing Qin

As a critical step to achieve human-like chatbots, empathetic response generation has attained increasing interests. Previous attempts are incomplete and not sufficient enough to elicit empathy because they only stay on the initial stage of empathy to automatically sense and simulate the feelings and thoughts of others via other-awareness. However, they ignore to include self-awareness to consider the own views of the self in their responses, which is a crucial process to achieve the empathy. To this end, we propose to generate Empathetic response with explicit Self-Other Awareness (EmpSOA). Specifically, three stages, self-other differentiation, self-other modulation and self-other generation, are devised to clearly maintain, regulate and inject the self-other aware information into the process of empathetic response generation. Both automatic and human evaluations on the benchmark dataset demonstrate the superiority of EmpSOA to generate more empathetic responses. Our source code will be publicly available.

Recent work has shown that infusing layout features into language models (LMs) improves processing of visually-rich documents such as scientific papers. Layout-infused LMs are often evaluated on documents with familiar layout features (e.g., papers from the same publisher), but in practice models encounter documents with unfamiliar distributions of layout features, such as new combinations of text sizes and styles, or new spatial configurations of textual elements. In this work we test whether layout-infused LMs are robust to layout distribution shifts. As a case study we use the task of scientific document structure recovery, segmenting a scientific paper into its structural categories (e.g., “title”, “caption”, “reference”). To emulate distribution shifts that occur in practice we re-partition the GROTOAP2 dataset. We find that under layout distribution shifts model performance degrades by up to 20 F1. Simple training strategies, such as increasing training diversity, can reduce this degradation by over 35% relative F1; however, models fail to reach in-distribution performance in any tested out-of-distribution conditions. This work highlights the need to consider layout distribution shifts during model evaluation, and presents a methodology for conducting such evaluations.

Efforts have been made to apply topic seed words to improve the topic interpretability of topic models. However, due to the semantic diversity of natural language, supervisions from seed words could be ambiguous, making it hard to be incorporated into the current neural topic models. In this paper, we propose SeededNTM, a neural topic model enhanced with supervisions from seed words on both word and document levels. We introduce a context-dependency assumption to alleviate the ambiguities with context document information, and an auto-adaptation mechanism to automatically balance between multi-level information. Moreover, an intra-sample consistency regularizer is proposed to deal with noisy supervisions via encouraging perturbation and semantic consistency. Extensive experiments on multiple datasets show that SeededNTM can derive semantically meaningful topics and outperforms the state-of-the-art seeded topic models in terms of topic quality and classification accuracy.

pdf abs
Learning from Children: Improving Image-Caption Pretraining via Curriculum
Hammad Ayyubi | Rahul Lokesh | Alireza Zareian | Bo Wu | Shih-Fu Chang

Image-caption pretraining has been quite successfully used for downstream vision tasks like zero-shot image classification and object detection. However, image-caption pretraining is still a hard problem – it requires multiple concepts (nouns) from captions to be aligned to several objects in images. To tackle this problem, we go to the roots – the best learner, children. We take inspiration from cognitive science studies dealing with children’s language learning to propose a curriculum learning framework. The learning begins with easy-to-align image caption pairs containing one concept per caption. The difficulty is progressively increased with each new phase by adding one more concept per caption. Correspondingly, the knowledge acquired in each learning phase is utilized in subsequent phases to effectively constrain the learning problem to aligning one new concept-object pair in each phase. We show that this learning strategy improves over vanilla image-caption training in various settings – pretraining from scratch, using a pretrained image or/and pretrained text encoder, low data regime etc.

As language models (LMs) scale, they develop many novel behaviors, good and bad, exacerbating the need to evaluate how they behave. Prior work creates evaluations with crowdwork (which is time-consuming and expensive) or existing data sources (which are not always available). Here, we automatically generate evaluations with LMs. We explore approaches with varying amounts of human effort, from instructing LMs to write yes/no questions to making complex Winogender schemas with multiple stages of LM-based generation and filtering. Crowdworkers rate the examples as highly relevant and agree with 90-100% of labels, sometimes more so than corresponding human-written datasets. We generate 154 datasets and discover new cases of inverse scaling where LMs get worse with size. Larger LMs repeat back a dialog user’s preferred answer (“sycophancy”) and express greater desire to pursue concerning goals like resource acquisition and goal preservation. We also find some of the first examples of inverse scaling in RL from Human Feedback (RLHF), where more RLHF makes LMs worse. For example, RLHF makes LMs express stronger political views (on gun rights and immigration) and a greater desire to avoid shut down. Overall, LM-written evaluations are high-quality and let us quickly discover many novel LM behaviors.

pdf abs
Cross-Domain Argument Quality Estimation
Michael Fromm | Max Berrendorf | Evgeniy Faerman | Thomas Seidl

Argumentation is one of society’s foundational pillars, and, sparked by advances in NLP, and the vast availability of text data, automated mining of arguments receives increasing attention. A decisive property of arguments is their strength or quality. While there are works on the automated estimation of argument strength, their scope is narrow:They focus on isolated datasets and neglect the interactions with related argument-mining tasks, such as argument identification and evidence detection. In this work, we close this gap by approaching argument quality estimation from multiple different angles:Grounded on rich results from thorough empirical evaluations, we assess the generalization capabilities of argument quality estimation across diverse domains and the interplay with related argument mining tasks. We find that generalization depends on a sufficient representation of different domains in the training part. In zero-shot transfer and multi-task experiments, we reveal that argument quality is among the more challenging tasks but can improve others. We publish our code at https://github.com/fromm-m/acl-cross-domain-aq.

The rapid development of aspect-based sentiment analysis (ABSA) within recent decades shows great potential for real-world society. The current ABSA works, however, are mostly limited to the scenario of a single text piece, leaving the study in dialogue contexts unexplored. To bridge the gap between fine-grained sentiment analysis and conversational opinion mining, in this work, we introduce a novel task of conversational aspect-based sentiment quadruple analysis, namely DiaASQ, aiming to detect the quadruple of target-aspect-opinion-sentiment in a dialogue. We manually construct a large-scale high-quality DiaASQ dataset in both Chinese and English languages. We deliberately develop a neural model to benchmark the task, which advances in effectively performing end-to-end quadruple prediction, and manages to incorporate rich dialogue-specific and discourse feature representations for better cross-utterance quadruple extraction. We hope the new benchmark will spur more advancements in the sentiment analysis community.

pdf abs
GeoDRL: A Self-Learning Framework for Geometry Problem Solving using Reinforcement Learning in Deductive Reasoning
Shuai Peng | Di Fu | Yijun Liang | Liangcai Gao | Zhi Tang

Ensuring both interpretability and correctness is a great challenge in automated geometry problem solving (GPS), and the scarcity of labeled data hinders learning mathematical reasoning from samples. Therefore, we present GeoDRL, a self-learning geometry problem solving framework that integrates logic graph deduction and Deep Reinforcement Learning (DRL) to optimize geometry reasoning as a Markov Decision Process. GeoDRL employs a Graph Neural Network on a Geometry Logic Graph, updating the problem state using a symbolic system. Incorporating DRL into deductive reasoning enables GeoDRL to achieve unsupervised self-learning while maintaining correctness. GeoDRL, through unsupervised learning, exhibits enhanced accuracy in the Geometry3K dataset, improving by 11.1% over previous SOTA methods, and simultaneously boosts efficiency and interpretability.

Recently, aspect sentiment quad prediction has received widespread attention in the field of aspect-based sentiment analysis. Existing studies extract quadruplets via pre-trained generative language models to paraphrase the original sentence into a templated target sequence. However, previous works only focus on what to generate but ignore what not to generate. We argue that considering the negative samples also leads to potential benefits. In this work, we propose a template-agnostic method to control the token-level generation, which boosts original learning and reduces mistakes simultaneously. Specifically, we introduce Monte Carlo dropout to understand the built-in uncertainty of pre-trained language models, acquiring the noises and errors. We further propose marginalized unlikelihood learning to suppress the uncertainty-aware mistake tokens. Finally, we introduce minimization entropy to balance the effects of marginalized unlikelihood learning. Extensive experiments on four public datasets demonstrate the effectiveness of our approach on various generation templates.

Prompt-based fine-tuning has boosted the performance of Pre-trained Language Models(PLMs) on few-shot Natural Language Understanding (NLU) tasks by employing task-specific prompts. Yet, PLMsare unfamiliar with prompt-style expressionsduring pre-training, which limits the few-shotlearning performance on downstream tasks. It would be desirable if the models can stimulate prompting knowledge while adaptation to specific NLU tasks. We present the Adversarial Knowledge Stimulated Contrastive Prompting (AKSCP) framework, leading to better few-shot NLU tasks for language models by implicitly stimulate knowledge from pretrained language model. In AKSCP, a novel paradigm Cloze-driven prompt is proposed for joint prompt tuning across word cloze task and prompt-based learning, forcing PLMs to stimulate prompting knowledge. We further design an Adversarial Contrastive learning method to improve the generalization ability of PLM for different downstream tasks. Experiments over a variety of NLU tasks show that AKSCP consistently outperforms state-of-the-arts for prompt-based fine-tuning.

Most previous few-shot Spoken Language Understanding (SLU) models typically need to be trained on a set of data-rich source domains and adapt to the target domain with a few examples. In this paper, we explore a more practical scenario for few-shot SLU, in which we only assume access to a pre-trained language model and a few labeled examples without any other source domain data. We concentrate on understanding how far the few-shot SLU could be pushed in this setting. To this end, we develop a prompt-based intent detection model in few-shot settings, which leverages the BERT original pre-training next sentence prediction task and the prompt template to detect the user’s intent. For slot filling, we propose an approach of reconstructing slot labels, which reduces the training complexity by reducing the number of slot labels in few-shot settings. To evaluate the few-shot SLU for a more practical scenario, we present two benchmarks, FewShotATIS and FewShotSNIPS. And a dynamic sampling strategy is designed to construct the two datasets according to the learning difficulty of each intent and slot. Experiments on FewShotATIS and FewShotSNIPS demonstrate that our proposed model achieves state-of-the-art performance.

pdf abs
Typology Guided Multilingual Position Representations: Case on Dependency Parsing
Tao Ji | Yuanbin Wu | Xiaoling Wang

Recent multilingual models benefit from strong unified semantic representation models. However, due to conflict linguistic regularities, ignoring language-specific features during multilingual learning may suffer from negative transfer. In this work, we analyze the relationbetween a language’s position space and its typological characterization, and suggest deploying different position spaces for different languages. We develop a position generation network which combines prior knowledge from typology features and existing position vectors. Experiments on the multilingual dependency parsing task show that the learned position vectors exhibit meaningful hidden structures, and they can help achieving the best multilingual parsing results.

pdf abs
Learning Event-aware Measures for Event Coreference Resolution
Yao Yao | Zuchao Li | Hai Zhao

Researchers are witnessing knowledge-inspired natural language processing shifts the focus from entity-level to event-level, whereas event coreference resolution is one of the core challenges. This paper proposes a novel model for within-document event coreference resolution. On the basis of event but not entity as before, our model learns and integrates multiple representations from both event alone and event pair. For the former, we introduce multiple linguistics-motivated event alone features for more discriminative event representations. For the latter, we consider multiple similarity measures to capture the distinction of event pair. Our proposed model achieves new state-of-the-art on the ACE 2005 benchmark, demonstrating the effectiveness of our proposed framework.

pdf abs
Second Language Acquisition of Neural Language Models
Miyu Oba | Tatsuki Kuribayashi | Hiroki Ouchi | Taro Watanabe

With the success of neural language models (LMs), their language acquisition has gained much attention. This work sheds light on the second language (L2) acquisition of LMs, while previous work has typically explored their first language (L1) acquisition. Specifically, we trained bilingual LMs with a scenario similar to human L2 acquisition and analyzed their cross-lingual transfer from linguistic perspectives. Our exploratory experiments demonstrated that the L1 pretraining accelerated their linguistic generalization in L2, and language transfer configurations (e.g., the L1 choice, and presence of parallel texts) substantially affected their generalizations. These clarify their (non-)human-like L2 acquisition in particular aspects.

Detecting adversarial samples that are carefully crafted to fool the model is a critical step to socially-secure applications. However, existing adversarial detection methods require access to sufficient training data, which brings noteworthy concerns regarding privacy leakage and generalizability. In this work, we validate that the adversarial sample generated by attack algorithms is strongly related to a specific vector in the high-dimensional inputs. Such vectors, namely UAPs (Universal Adversarial Perturbations), can be calculated without original training data. Based on this discovery, we propose a data-agnostic adversarial detection framework, which induces different responses between normal and adversarial samples to UAPs. Experimental results show that our method achieves competitive detection performance on various text classification tasks, and maintains an equivalent time consumption to normal inference.

pdf abs
Exploring the Effectiveness of Prompt Engineering for Legal Reasoning Tasks
Fangyi Yu | Lee Quartey | Frank Schilder

The use of large language models (LLMs) for zero- or few-shot prompting in natural language processing has given rise to a new research area known as prompt engineering. Recent studies have demonstrated that Chain-of-Thought (CoT) prompts can lead to significant improvements in tasks such as arithmetic and common-sense reasoning. This paper explores the use of such approaches in legal reasoning tasks by conducting experiments on the COLIEE entailment task, which is based on the Japanese Bar exam. We evaluate zero-shot/few-shot and fine-tuning approaches with and without explanations, as well as various prompting strategies. Our results indicate that while CoT prompting and fine-tuning with explanations can improve performance, the best results are achieved with prompts derived from specific legal reasoning techniques, such as IRAC (Issue, Rule, Application, Conclusion). In addition, we observe that few-shot learning where the demonstrations are derived from clustering past training data consistently yields high performance on the COLIEE entailment task for both the years of the data that we tested. Through our experiments, we improve the previous best result on the 2021 COLIEE task from 0.7037 to 0.8025 and surpass the best system from 2022 with an accuracy of 0.789.

pdf abs
End-to-end Aspect-based Sentiment Analysis with Combinatory Categorial Grammar
Yuanhe Tian | Weidong Chen | Bo Hu | Yan Song | Fei Xia

End-to-end Aspect-based Sentiment Analysis (EASA) is a natural language processing (NLP) task that involves extracting aspect terms and identifying the sentiments for them, which provides a fine-grained level of text analysis and thus requires a deep understanding of the running text. Many previous studies leverage advanced text encoders to extract context information and use syntactic information, e.g., the dependency structure of the input sentence, to improve the model performance. However, such models may reach a bottleneck since the dependency structure is not designed to provide semantic information of the text, which is also important for identifying the sentiment and thus leave room for further improvement. Considering that combinatory categorial grammar (CCG) is a formalism that expresses both syntactic and semantic information of a sentence, it has the potential to be beneficial to EASA. In this paper, we propose a novel approach to improve EASA with CCG supertags, which carry the syntactic and semantic information of the associated words and serve as the most important part of the CCG derivation. Specifically, our approach proposes a CCG supertag decoding process to learn the syntactic and semantic information carried by CCG supertags and use the information to guide the attention over the input words so as to identify important contextual information for EASA. Furthermore, a gate mechanism is used in incorporating the weighted contextual information into the backbone EASA decoding process. We evaluate our approach on three publicly available English datasets for EASA, and show that it outperforms strong baselines and achieves state-of-the-art results on all datasets.

Multimodal Sentiment Analysis leverages multimodal signals to detect the sentiment of a speaker. Previous approaches concentrate on performing multimodal fusion and representation learning based on general knowledge obtained from pretrained models, which neglects the effect of domain-specific knowledge. In this paper, we propose Contrastive Knowledge Injection (ConKI) for multimodal sentiment analysis, where specific-knowledge representations for each modality can be learned together with general knowledge representations via knowledge injection based on an adapter architecture. In addition, ConKI uses a hierarchical contrastive learning procedure performed between knowledge types within every single modality, across modalities within each sample, and across samples to facilitate the effective learning of the proposed representations, hence improving multimodal sentiment predictions. The experiments on three popular multimodal sentiment analysis benchmarks show that ConKI outperforms all prior methods on a variety of performance metrics.

pdf abs
On Degrees of Freedom in Defining and Testing Natural Language Understanding
Saku Sugawara | Shun Tsugita

Natural language understanding (NLU) studies often exaggerate or underestimate the capabilities of systems, thereby limiting the reproducibility of their findings. These erroneous evaluations can be attributed to the difficulty of defining and testing NLU adequately. In this position paper, we reconsider this challenge by identifying two types of researcher degrees of freedom. We revisit Turing’s original interpretation of the Turing test and reveal that an effective test of NLU does not provide an operational definition; it merely provides inductive evidence that the test subject understands the language sufficiently well to meet stakeholder objectives. In other words, stakeholders are free to arbitrarily define NLU through their objectives. To use the test results as inductive evidence, stakeholders must carefully assess if the interpretation of test scores is valid or not. However, designing and using NLU tests involve other degrees of freedom, such as specifying target skills and defining evaluation metrics. As a result, achieving consensus among stakeholders becomes difficult. To resolve this issue, we propose a validity argument, which is a framework comprising a series of validation criteria across test components. By demonstrating that current practices in NLU studies can be associated with those criteria and organizing them into a comprehensive checklist, we prove that the validity argument can serve as a coherent guideline for designing credible test sets and facilitating scientific communication.

pdf abs
AttenWalker: Unsupervised Long-Document Question Answering via Attention-based Graph Walking
Yuxiang Nie | Heyan Huang | Wei Wei | Xian-Ling Mao

Annotating long-document question answering (long-document QA) pairs is time-consuming and expensive. To alleviate the problem, it might be possible to generate long-document QA pairs via unsupervised question answering (UQA) methods. However, existing UQA tasks are based on short documents, and can hardly incorporate long-range information. To tackle the problem, we propose a new task, named unsupervised long-document question answering (ULQA), aiming to generate high-quality long-document QA instances in an unsupervised manner. Besides, we propose AttenWalker, a novel unsupervised method to aggregate and generate answers with long-range dependency so as to construct long-document QA pairs. Specifically, AttenWalker is composed of three modules, i.e. span collector, span linker and answer aggregator. Firstly, the span collector takes advantage of constituent parsing and reconstruction loss to select informative candidate spans for constructing answers. Secondly, with the help of the attention graph of a pre-trained long-document model, potentially interrelated text spans (that might be far apart) could be linked together via an attention-walking algorithm. Thirdly, in the answer aggregator, linked spans are aggregated into the final answer via the mask-filling ability of a pre-trained model. Extensive experiments show that AttenWalker outperforms previous methods on NarrativeQA and Qasper. In addition, AttenWalker also shows strong performance in the few-shot learning setting.

pdf abs
Adaptive Ordered Information Extraction with Deep Reinforcement Learning
Wenhao Huang | Jiaqing Liang | Zhixu Li | Yanghua Xiao | Chuanjun Ji

Information extraction (IE) has been studied extensively. The existing methods always follow a fixed extraction order for complex IE tasks with multiple elements to be extracted in one instance such as event extraction. However, we conduct experiments on several complex IE datasets and observe that different extraction orders can significantly affect the extraction results for a great portion of instances, and the ratio of sentences that are sensitive to extraction orders increases dramatically with the complexity of the IE task. Therefore, this paper proposes a novel adaptive ordered IE paradigm to find the optimal element extraction order for different instances, so as to achieve the best extraction results. We also propose an reinforcement learning (RL) based framework to generate optimal extraction order for each instance dynamically. Additionally, we propose a co-training framework adapted to RL to mitigate the exposure bias during the extractor training phase. Extensive experiments conducted on several public datasets demonstrate that our proposed method can beat previous methods and effectively improve the performance of various IE tasks, especially for complex ones.

Answering complex queries on knowledge graphs is important but particularly challenging because of the data incompleteness. Query embedding methods address this issue by learningbased models and simulating logical reasoning with set operators. Previous works focus on specific forms of embeddings, but scoring functions between embeddings are underexplored. In contrast to existing scorning functions motivated by local comparison or global transport, this work investigates the local and global trade-off with unbalanced optimal transport theory. Specifically, we embed sets as bounded measures in R endowed with a scoring function motivated by the Wasserstein-Fisher-Rao metric. Such a design also facilitates closed-form set operators in the embedding space. Moreover, we introduce a convolution-based algorithm for linear time computation and a block diagonal kernel to enforce the trade-off. Results show that WFRE is capable of outperforming existing query embedding methods on standard datasets, evaluation sets with combinatorially complex queries, and hierarchical knowledge graphs. Ablation study shows that finding a better local and global trade-off is essential for performance improvement.

pdf abs
RISE: Leveraging Retrieval Techniques for Summarization Evaluation
David Uthus | Jianmo Ni

Evaluating automatically-generated text summaries is a challenging task. While there have been many interesting approaches, they still fall short of human evaluations. We present RISE, a new approach for evaluating summaries by leveraging techniques from information retrieval. RISE is first trained as a retrieval task using a dual-encoder retrieval setup, and can then be subsequently utilized for evaluating a generated summary given an input document, without gold reference summaries. RISE is especially well suited when working on new datasets where one may not have reference summaries available for evaluation. We conduct comprehensive experiments on the SummEval benchmark (Fabbri et al., 2021) and a long document summarization benchmark. The results show that RISE consistently achieves higher correlation with human evaluations compared to many past approaches to summarization evaluation. Furthermore, RISE also demonstrates data-efficiency and generalizability across languages.

pdf abs
On the Difference of BERT-style and CLIP-style Text Encoders
Zhihong Chen | Guiming Chen | Shizhe Diao | Xiang Wan | Benyou Wang

Masked language modeling (MLM) has been one of the most popular pretraining recipes in natural language processing, e.g., BERT, one of the representative models. Recently, contrastive language-image pretraining (CLIP) has also attracted attention, especially its vision models that achieve excellent performance on a broad range of vision tasks. However, few studies are dedicated to studying the text encoders learned by CLIP. In this paper, we analyze the difference between BERT-style and CLIP-style text encoders from three experiments: (i) general text understanding, (ii) vision-centric text understanding, and (iii) text-to-image generation. Experimental analyses show that although CLIP-style text encoders underperform BERT-style ones for general text understanding tasks, they are equipped with a unique ability, i.e., synesthesia, for the cross-modal association, which is more similar to the senses of humans.

pdf abs
Model Interpretability and Rationale Extraction by Input Mask Optimization
Marc Brinner | Sina Zarrieß

Concurrent with the rapid progress in neural network-based models in NLP, the need for creating explanations for the predictions of these black-box models has risen steadily. Yet, especially for complex inputs like texts or images, existing interpretability methods still struggle with deriving easily interpretable explanations that also accurately represent the basis for the model’s decision. To this end, we propose a new, model-agnostic method to generate extractive explanations for predictions made by neural networks, that is based on masking parts of the input which the model does not consider to be indicative of the respective class. The masking is done using gradient-based optimization combined with a new regularization scheme that enforces sufficiency, comprehensiveness, and compactness of the generated explanation. Our method achieves state-of-the-art results in a challenging paragraph-level rationale extraction task, showing that this task can be performed without training a specialized model. We further apply our method to image inputs and obtain high-quality explanations for image classifications, which indicates that the objectives for optimizing explanation masks in text generalize to inputs of other modalities.

We present NusaCrowd, a collaborative initiative to collect and unify existing resources for Indonesian languages, including opening access to previously non-public resources. Through this initiative, we have brought together 137 datasets and 118 standardized data loaders. The quality of the datasets has been assessed manually and automatically, and their value is demonstrated through multiple experiments.NusaCrowd’s data collection enables the creation of the first zero-shot benchmarks for natural language understanding and generation in Indonesian and the local languages of Indonesia. Furthermore, NusaCrowd brings the creation of the first multilingual automatic speech recognition benchmark in Indonesian and the local languages of Indonesia. Our work strives to advance natural language processing (NLP) research for languages that are under-represented despite being widely spoken.

pdf abs
Transcribing Vocal Communications of Domestic Shiba lnu Dogs
Jieyi Huang | Chunhao Zhang | Mengyue Wu | Kenny Zhu

How animals communicate and whether they have languages is a persistent curiosity of human beings. However, the study of animal communications has been largely restricted to data from field recordings or in a controlled environment, which is expensive and limited in scale and variety. In this paper, we take domestic Shiba Inu dogs as an example, and extract their vocal communications from large amount of YouTube videos of Shiba Inu dogs. We classify these clips into different scenarios and locations, and further transcribe the audio into phonetically symbolic scripts through a systematic process. We discover consistent phonetic symbols among their expressions, which indicates that Shiba Inu dogs can have systematic verbal communication patterns. This reusable framework produces the first-of-its-kind Shiba Inu vocal communication dataset that will be valuable to future research in both zoology and linguistics.

pdf abs
SkillQG: Learning to Generate Question for Reading Comprehension Assessment
Xiaoqiang Wang | Bang Liu | Siliang Tang | Lingfei Wu

We present SkillQG: a question generation framework with controllable comprehension types for assessing and improving machine reading comprehension models. Existing question generation systems widely differentiate questions by literal information such as question words and answer types to generate semantically relevant questions for a given context. However, they rarely consider the comprehension nature of questions, i.e., the different comprehension capabilities embodied by different questions. In comparison, our SkillQG is able to tailor a fine-grained assessment and improvement to the capabilities of questions answering models built on it. Specifically, we first frame the comprehension type of questions based on a hierarchical skill-based schema. We then formulate SkillQG as a skill-conditioned question generator. Furthermore, to improve the controllability of generation, we augment the input text with skill-specific question focus and knowledge, which are constructed by iteratively prompting the pre-trained language models. Empirical results demonstrate that SkillQG outperforms baselines in terms of quality, relevance, and skill-controllability while showing a promising performance boost in downstream question answering task.

pdf abs
Improving Long Dialogue Summarization with Semantic Graph Representation
Yilun Hua | Zhaoyuan Deng | Kathleen McKeown

Although Large Language Models (LLMs) are successful in abstractive summarization of short dialogues, summarization of long dialogues remains challenging. To address this challenge, we propose a novel algorithm that processes complete dialogues comprising thousands of tokens into topic-segment-level Abstract Meaning Representation (AMR) graphs, which explicitly capture the dialogue structure, highlight salient semantics, and preserve high-level information. We also develop a new text-graph attention to leverage both graph semantics and a pretrained LLM that exploits the text. Finally, we propose an AMR node selection loss used jointly with conventional cross-entropy loss, to create additional training signals that facilitate graph feature encoding and content selection. Experiments show that our system outperforms the state-of-the-art models on multiple long dialogue summarization datasets, especially in low-resource settings, and generalizes well to out-of-domain data.

In this study, we analyze the model intrinsic features of a summarization model by varying the fine-tuning objectives and datasets. We fine-tune BART models combining three fine-tuning objectives (negative log-likelihood, unlikelihood, and contrastive loss) and two datasets (CNN/DailyMail and XSum) and provide shuffled or aligned documents to observe changes in the model predictions and intrinsic features. We find that (i) the inductive bias for factual consistency during the fine-tuning procedure depends on both the objectives and datasets, and (ii) summarization models with relatively low factual consistency are more likely to model summaries that are not conditional to the documents. We demonstrate that splitting data based on the unconditional and conditional summary modeling difficulty affects the factual consistency and intrinsic features of the summarization models. Our experimental results highlight the importance of studying the inductive bias during fine-tuning for factual consistency.

pdf abs
EfficientVLM: Fast and Accurate Vision-Language Models via Knowledge Distillation and Modal-adaptive Pruning
Tiannan Wang | Wangchunshu Zhou | Yan Zeng | Xinsong Zhang

Pre-trained vision-language models (VLMs) have achieved impressive results in a range of vision-language tasks. However, popular VLMs usually consist of hundreds of millions of parameters which brings challenges for fine-tuning and deployment in real-world applications due to space, memory, and latency constraints. In this work, we introduce a distilling then pruning framework to compress large vision-language models into smaller, faster, and more accurate ones. We first shrink the size ofa pre-trained large VLM and apply knowledge distillation in the vision-language pre-training stage to obtain a task-agnostic compact VLM. Then we propose a modal-adaptive pruning algorithm to automatically infer the importance of vision and language modalities for different downstream tasks and adaptively remove redundant structures and neurons in different encoders with controllable target sparsity. We apply our framework to train EfficientVLM, a fast and accurate vision-language model consisting of 6 vision layers, 3 text layers, and 3 cross-modal fusion layers, accounting for only 93 million parameters in total, which is 44.3% of the teacher model. EfficientVLM retains 98.4% performance of the teacher model and accelerates its inference speed by 2.2×. EfficientVLM achieves a large absolute improvement over previous SoTA efficient VLMs of similar sizes by a large margin on various vision-language tasks, including VQAv2 (+4.9%), NLVR2 (+5.6%), ITR (R@1 on TR +17.2%, on IR + 15.6% ) and COCO caption generation (CIDEr +6.5), demonstrating a large potential on training lightweight VLMs.

pdf abs
DP-BART for Privatized Text Rewriting under Local Differential Privacy
Timour Igamberdiev | Ivan Habernal

Privatized text rewriting with local differential privacy (LDP) is a recent approach that enables sharing of sensitive textual documents while formally guaranteeing privacy protection to individuals. However, existing systems face several issues, such as formal mathematical flaws, unrealistic privacy guarantees, privatization of only individual words, as well as a lack of transparency and reproducibility. In this paper, we propose a new system ‘DP-BART’ that largely outperforms existing LDP systems. Our approach uses a novel clipping method, iterative pruning, and further training of internal representations which drastically reduces the amount of noise required for DP guarantees. We run experiments on five textual datasets of varying sizes, rewriting them at different privacy guarantees and evaluating the rewritten texts on downstream text classification tasks. Finally, we thoroughly discuss the privatized text rewriting approach and its limitations, including the problem of the strict text adjacency constraint in the LDP paradigm that leads to the high noise requirement.

Traditional supervised learning mostly works on individual tasks and requires training on a large set of task-specific examples. This paradigm seriously hinders the development of task generalization since preparing a task-specific example set is costly. To build a system that can quickly and easily generalize to new tasks, task instructions have been adopted as an emerging trend of supervision recently. These instructions give the model the definition of the task and allow the model to output the appropriate answer based on the instructions and inputs. However, task instructions are often expressed in different forms, which can be interpreted from two threads: first, some instructions are short sentences and are pretrained language model (PLM) oriented, such as prompts, while other instructions are paragraphs and are human-oriented, such as those in Amazon MTurk; second, different end-users very likely explain the same task with instructions of different textual expressions. A robust system for task generalization should be able to handle any new tasks regardless of the variability of instructions. However, the system robustness in dealing with instruction-driven task generalization is still unexplored. This work investigates the system robustness when the instructions of new tasks are (i) manipulated, (ii) paraphrased, or (iii) from different levels of conciseness. To our knowledge, this is the first work that systematically studies how robust a PLM is when it is supervised by instructions with different factors of variability.

pdf abs
Masked Latent Semantic Modeling: an Efficient Pre-training Alternative to Masked Language Modeling
Gábor Berend

In this paper, we propose an alternative to the classic masked language modeling (MLM) pre-training paradigm, where the objective is altered from the reconstruction of the exact identity of randomly selected masked subwords to the prediction of their latent semantic properties. We coin the proposed pre-training technique masked latent semantic modeling (MLSM for short). In order to make the contextualized determination of the latent semantic properties of the masked subwords possible, we rely on an unsupervised technique which uses sparse coding. Our experimental results reveal that the fine-tuned performance of those models that we pre-trained via MLSM is consistently and significantly better compared to the use of vanilla MLM pretraining and other strong baselines.

pdf abs
Detection and Mitigation of the Negative Impact of Dataset Extractivity on Abstractive Summarization
Yubin Ge | Sullam Jeoung | Ly Dinh | Jana Diesner

In text summarization, extractivity is defined as a measurement of the degree of overlap between a source document and its summary. Previous research has shown that the extractivity level of training data can influence both output extractivity and the amount of factual information (i.e. faithfulness) in outputs for abstractive summarization. However, it remains unclear if and how extractivity impacts the performance of abstractive models. In this work, we investigate the relationship between dataset extractivity and model performance by comparing the performance of trained models under different degrees of extractivity. We find that while low levels of extractivity can improve performance, as extractivity increases, performance is negatively impacted. Furthermore, through an analysis of the model’s copy continuity of content, we discover that higher extractivity leads to a greater tendency for the model to copy text continuously from the source document rather than identifying and summarizing important content that should be covered in the target summary. To address these issues, we propose a simple and effective method to design copy labels for fixing the model’s copying behaviors and train the model with a copy mechanism. The experimental results illustrate the effectiveness of our strategy in alleviating the negative impact on model performance resulting from high dataset extractivity, and that our method outperforms several competitive baselines.

pdf abs
Commonsense Knowledge Graph Completion Via Contrastive Pretraining and Node Clustering
Siwei Wu | Xiangqing Shen | Rui Xia

The nodes in the commonsense knowledge graph (CSKG) are normally represented by free-form short text (e.g., word or phrase). Different nodes may represent the same concept. This leads to the problems of edge sparsity and node redundancy, which challenges CSKG representation and completion. On the one hand, edge sparsity limits the performance of graph representation learning; On the other hand, node redundancy makes different nodes corresponding to the same concept have inconsistent relations with other nodes. To address the two problems, we propose a new CSKG completion framework based on Contrastive Pretraining and Node Clustering (CPNC). Contrastive Pretraining constructs positive and negative head-tail node pairs on CSKG and utilizes contrastive learning to obtain better semantic node representation. Node Clustering aggregates nodes with the same concept into a latent concept, assisting the task of CSKG completion. We evaluate our CPNC approach on two CSKG completion benchmarks (CN-100K and ATOMIC), where CPNC outperforms the state-of-the-art methods. Extensive experiments demonstrate that both Contrastive Pretraining and Node Clustering can significantly improve the performance of CSKG completion. The source code of CPNC is publicly available on https://github.com/NUSTM/CPNC.

pdf abs
Incorporating Factuality Inference to Identify Document-level Event Factuality
Heng Zhang | Peifeng Li | Zhong Qian | Xiaoxu Zhu

Document-level Event Factuality Identification (DEFI) refers to identifying the degree of certainty that a specific event occurs in a document. Previous studies on DEFI failed to link the document-level event factuality with various sentence-level factuality values in the same document. In this paper, we innovatively propose an event factuality inference task to bridge the sentence-level and the document-level event factuality semantically. Specifically, we present a Sentence-to-Document Inference Network (SDIN) that contains a multi-layer interaction module and a gated aggregation module to integrate the above two tasks, and employ a multi-task learning framework to improve the performance of DEFI. The experimental results on the public English and Chinese DLEF datasets show that our model outperforms the SOTA baselines significantly.

pdf abs
Hybrid and Collaborative Passage Reranking
Zongmeng Zhang | Wengang Zhou | Jiaxin Shi | Houqiang Li

In passage retrieval system, the initial passage retrieval results may be unsatisfactory, which can be refined by a reranking scheme. Existing solutions to passage reranking focus on enriching the interaction between query and each passage separately, neglecting the context among the top-ranked passages in the initial retrieval list. To tackle this problem, we propose a Hybrid and Collaborative Passage Reranking (HybRank) method, which leverages the substantial similarity measurements of upstream retrievers for passage collaboration and incorporates the lexical and semantic properties of sparse and dense retrievers for reranking. Besides, built on off-the-shelf retriever features, HybRank is a plug-in reranker capable of enhancing arbitrary passage lists including previously reranked ones. Extensive experiments demonstrate the stable improvements of performance over prevalent retrieval and reranking methods, and verify the effectiveness of the core components of HybRank.

pdf abs
Sentence Embedding Leaks More Information than You Expect: Generative Embedding Inversion Attack to Recover the Whole Sentence
Haoran Li | Mingshi Xu | Yangqiu Song

Sentence-level representations are beneficial for various natural language processing tasks. It is commonly believed that vector representations can capture rich linguistic properties. Currently, large language models (LMs) achieve state-of-the-art performance on sentence embedding. However, some recent works suggest that vector representations from LMs can cause information leakage. In this work, we further investigate the information leakage issue and propose a generative embedding inversion attack (GEIA) that aims to reconstruct input sequences based only on their sentence embeddings. Given the black-box access to a language model, we treat sentence embeddings as initial tokens’ representations and train or fine-tune a powerful decoder model to decode the whole sequences directly. We conduct extensive experiments to demonstrate that our generative inversion attack outperforms previous embedding inversion attacks in classification metrics and generates coherent and contextually similar sentences as the original inputs.

pdf abs
Learning Query Adaptive Anchor Representation for Inductive Relation Prediction
Zhiwen Xie | Yi Zhang | Jin Liu | Guangyou Zhou | Jimmy Huang

Relation prediction on knowledge graphs (KGs) attempts to infer the missing links between entities. Most previous studies are limited to the transductive setting where all entities must be seen during the training, making them unable to perform reasoning on emerging entities. Recently, the inductive setting is proposed to handle the entities in the test phase to be unseen during training, However, it suffers from the inefficient reasoning under the enclosing subgraph extraction issue and the lack of effective entity-independent feature modeling. To this end, we propose a novel Query Adaptive Anchor Representation (QAAR) model for inductive relation prediction. First, we extract one opening subgraph and perform reasoning by one time for all candidate triples, which is more efficient when the number of candidate triples is large. Second, we define some query adaptive anchors which are independent on any specific entity. Based on these anchors, we take advantage of the transferable entity-independent features (relation-aware, structure-aware and distance features) that can be used to produce entity embeddings for emerging unseen entities. Such entity-independent features is modeled by a query-aware graph attention network on the opening subgraph. Experimental results demonstrate that our proposed QAAR outperforms state-of-the-art baselines in inductive relation prediction task.

pdf abs
Context or Knowledge is Not Always Necessary: A Contrastive Learning Framework for Emotion Recognition in Conversations
Geng Tu | Bin Liang | Ruibin Mao | Min Yang | Ruifeng Xu

Emotion recognition in conversations (ERC) aims to detect the emotion of utterances in conversations. Existing efforts generally focus on modeling context- and knowledge-sensitive dependencies. However, it is observed that the emotions of many utterances can be correctly detected without context or external knowledge. In such cases, blindly leveraging the context and external knowledge may impede model training. Based on this, we propose a novel framework based on contrastive learning (CL), called CKCL (including the contrastive learning scenarios among Context and Knowledge), to distinguish the above utterances for better vector representations. The CKCL framework defines context- and knowledge-independent utterances, as the positive sample, whose predicted results are unchanged even masking context and knowledge representations, otherwise, the negative sample. This can obtain a latent feature reflecting the impact degree of context and external knowledge on predicted results, thus effectively denoising irrelevant context and knowledge during training. Experimental results on four datasets show the performance of CKCL-based models is significantly boosted and outperforms state-of-the-art methods.

Speaker diarization is a classic task in speech processing and is crucial in multi-party scenarios such as meetings and conversations. Current mainstream speaker diarization approaches consider acoustic information only, which result in performance degradation when encountering adverse acoustic environment. In this paper, we propose methods to extract speaker-related information from semantic content in multi-party meetings, which, as we will show, can further benefit speaker diarization. We introduce two sub-tasks, Dialogue Detection and Speaker-Turn Detection, in which we effectively extract speaker information from conversational semantics. We also propose a simple yet effective algorithm to jointly model acoustic and semantic information and obtain speaker-identified texts. Experiments on both AISHELL-4 and AliMeeting datasets show that our method achieves consistent improvements over acoustic-only speaker diarization systems.

pdf abs
Cross-Lingual Knowledge Distillation for Answer Sentence Selection in Low-Resource Languages
Shivanshu Gupta | Yoshitomo Matsubara | Ankit Chadha | Alessandro Moschitti

While impressive performance has been achieved on the task of Answer Sentence Selection (AS2) for English, the same does not hold for languages that lack large labeled datasets. In this work, we propose Cross-Lingual Knowledge Distillation (CLKD) from a strong English AS2 teacher as a method to train AS2 models for low-resource languages in the tasks without the need of labeled data for the target language. To evaluate our method, we introduce 1) Xtr-WikiQA, a translation-based WikiQA dataset for 9 additional languages, and 2) TyDi-AS2, a multilingual AS2 dataset with over 70K questions spanning 8 typologically diverse languages. We conduct extensive experiments on Xtr-WikiQA and TyDi-AS2 with multiple teachers, diverse monolingual and multilingual pretrained language models (PLMs) as students, and both monolingual and multilingual training. The results demonstrate that CLKD either outperforms or rivals even supervised fine-tuning with the same amount of labeled data and a combination of machine translation and the teacher model. Our method can potentially enable stronger AS2 models for low-resource languages, while TyDi-AS2 can serve as the largest multilingual AS2 dataset for further studies in the research community.

pdf abs
Run Like a Girl! Sport-Related Gender Bias in Language and Vision
Sophia Harrison | Eleonora Gualdoni | Gemma Boleda

Gender bias in Language and Vision datasets and models has the potential to perpetuate harmful stereotypes and discrimination. We analyze gender bias in two Language and Vision datasets. Consistent with prior work, we find that both datasets underrepresent women, which promotes their invisibilization. Moreover, we hypothesize and find that a bias affects human naming choices for people playing sports: speakers produce names indicating the sport (e.g. “tennis player” or “surfer”) more often when it is a man or a boy participating in the sport than when it is a woman or a girl, with an average of 46% vs. 35% of sports-related names for each gender. A computational model trained on these naming data reproduces thebias. We argue that both the data and the model result in representational harm against women.

pdf abs
People and Places of Historical Europe: Bootstrapping Annotation Pipeline and a New Corpus of Named Entities in Late Medieval Texts
Vit Novotny | Kristina Luger | Michal Štefánik | Tereza Vrabcova | Ales Horak

Although pre-trained named entity recognition (NER) models are highly accurate on modern corpora, they underperform on historical texts due to differences in language OCR errors. In this work, we develop a new NER corpus of 3.6M sentences from late medieval charters written mainly in Czech, Latin, and German.We show that we can start with a list of known historical figures and locations and an unannotated corpus of historical texts, and use information retrieval techniques to automatically bootstrap a NER-annotated corpus. Using our corpus, we train a NER model that achieves entity-level Precision of 72.81–93.98% with 58.14–81.77% Recall on a manually-annotated test dataset. Furthermore, we show that using a weighted loss function helps to combat class imbalance in token classification tasks. To make it easy for others to reproduce and build upon our work, we publicly release our corpus, models, and experimental code.

We present a new fact-checking benchmark, Check-COVID, that requires systems to verify claims about COVID-19 from news using evidence from scientific articles. This approach to fact-checking is particularly challenging as it requires checking internet text written in everyday language against evidence from journal articles written in formal academic language. Check-COVID contains 1, 504 expert-annotated news claims about the coronavirus paired with sentence-level evidence from scientific journal articles and veracity labels. It includes both extracted (journalist-written) and composed (annotator-written) claims. Experiments using both a fact-checking specific system and GPT-3.5, which respectively achieve F1 scores of 76.99 and 69.90 on this task, reveal the difficulty of automatically fact-checking both claim types and the importance of in-domain data for good performance. Our data and models are released publicly at https://github.com/posuer/Check-COVID.

Dynamic early exit has demonstrated great potential in coping with the sharply increasing number of pre-trained language model parameters, which can achieve a good trade-off between performance and efficiency. The existing early exit paradigm relies on training parametrical internal classifiers at each intermediate layer to complete specific tasks. Based on the predictions of these internal classifiers, different methods are designed to decide when to exit. Under this circumstance, each intermediate layer takes on both generic language representation learning and task-specific feature extraction, which makes each intermediate layer struggle to balance two types of backward loss signals during training. To break this dilemma, we propose an adapter method to decouple the two distinct types of representation and further introduce a non-parametric simplex equiangular tight frame classifier (ETF) for improvement. Extensive experiments on monolingual and multilingual tasks demonstrate that our method gains significant improvements over strong PLM backbones and early exit methods.

pdf abs
Tokenization with Factorized Subword Encoding
David Samuel | Lilja Øvrelid

In recent years, language models have become increasingly larger and more complex. However, the input representations for these models continue to rely on simple and greedy subword tokenization methods. In this paper, we propose a novel tokenization method that factorizes subwords onto discrete triplets using a VQ-VAE model. The effectiveness of the proposed tokenization method, referred to as the Factorizer, is evaluated on language modeling and morpho-syntactic tasks for 7 diverse languages. Results indicate that this method is more appropriate and robust for morphological tasks than the commonly used byte-pair encoding (BPE) tokenization algorithm.

pdf abs
Rarely a problem? Language models exhibit inverse scaling in their predictions following few-type quantifiers
James Michaelov | Benjamin Bergen

How well do language models deal with quantification? In this study, we focus on ‘few’-type quantifiers, as in ‘few children like toys’, which might pose a particular challenge for language models because the sentence components with out the quantifier are likely to co-occur, and ‘few’-type quantifiers are rare. We present 960 English sentence stimuli from two human neurolinguistic experiments to 22 autoregressive transformer models of differing sizes. Not only do all the models perform poorly on ‘few’-type quantifiers, but overall the larger the model, the worse its performance. This inverse scaling is consistent with previous work suggesting that larger models increasingly reflect online rather than offline human processing, and we argue that the decreasing performance of larger models may challenge uses of language models as the basis for natural language systems.

pdf abs
“A Little is Enough”: Few-Shot Quality Estimation based Corpus Filtering improves Machine Translation
Akshay Batheja | Pushpak Bhattacharyya

Quality Estimation (QE) is the task of evaluating the quality of a translation when reference translation is not available. The goal of QE aligns with the task of corpus filtering, where we assign the quality score to the sentence pairs present in the pseudo-parallel corpus. We propose a Quality Estimation based Filtering approach to extract high-quality parallel data from the pseudo-parallel corpus. To the best of our knowledge, this is a novel adaptation of QE framework to extracting quality parallel corpus from the pseudo-parallel corpus.. By training with this filtered corpus, we observe an improvement in the Machine Translation (MT) system’s performance by up to 1.8 BLEU points, for English-Marathi, Chinese-English, and Hindi-Bengali language pairs, over the baseline model. The baseline model is the one that is trained on the whole pseudo-parallel corpus. Our Few-shot QE model transfer learned from the English-Marathi QE model and fine-tuned on only 500 Hindi-Bengali training instances, shows an improvement of up to 0.6 BLEU points for Hindi-Bengali language pair, compared to the baseline model. This demonstrates the promise of transfer learning in the setting under discussion. QE systems typically require in the order of (7K-25K) of training data. Our Hindi-Bengali QE is trained on only 500 instances of training that is 1/40th of the normal requirement and achieves comparable performance. All the scripts and datasets utilized in this study will be publicly available.

pdf abs
How effective is machine translation on low-resource code-switching? A case study comparing human and automatic metrics
Li Nguyen | Christopher Bryant | Oliver Mayeux | Zheng Yuan

This paper presents an investigation into the differences between processing monolingual input and code-switching (CSW) input in the context of machine translation (MT). Specifically, we compare the performance of three MT systems (Google, mBART-50 and M2M-100-big) in terms of their ability to translate monolingual Vietnamese, a low-resource language, and Vietnamese-English CSW respectively. To our knowledge, this is the first study to systematically analyse what might happen when multilingual MT systems are exposed to CSW data using both automatic and human metrics. We find that state-of-the-art neural translation systems not only achieve higher scores on automatic metrics when processing CSW input (compared to monolingual input), but also produce translations that are consistently rated as more semantically faithful by humans. We further suggest that automatic evaluation alone is insufficient for evaluating the translation of CSW input. Our findings establish a new benchmark that offers insights into the relationship between MT and CSW.

pdf abs
Images in Language Space: Exploring the Suitability of Large Language Models for Vision & Language Tasks
Sherzod Hakimov | David Schlangen

Large language models have demonstrated robust performance on various language tasks using zero-shot or few-shot learning paradigms. While being actively researched, multimodal models that can additionally handle images as input have yet to catch up in size and generality with language-only models. In this work, we ask whether language-only models can be utilised for tasks that require visual input – but also, as we argue, often require a strong reasoning component. Similar to some recent related work, we make visual information accessible to the language model using separate verbalisation models. Specifically, we investigate the performance of open-source, open-access language models against GPT-3 on five vision-language tasks when given textually-encoded visual information. Our results suggest that language models are effective for solving vision-language tasks even with limited samples. This approach also enhances the interpretability of a model’s output by providing a means of tracing the output back through the verbalised image content.

pdf abs
On the Expressivity Role of LayerNorm in Transformers’ Attention
Shaked Brody | Uri Alon | Eran Yahav

Layer Normalization (LayerNorm) is an inherent component in all Transformer-based models. In this paper, we show that LayerNorm is crucial to the expressivity of the multi-head attention layer that follows it. This is in contrast to the common belief that LayerNorm’s only role is to normalize the activations during the forward pass, and their gradients during the backward pass. We consider a geometric interpretation of LayerNorm and show that it consists of two components: (a) projection of the input vectors to a d-1 space that is orthogonal to the [1,1,...,1] vector, and(b) scaling of all vectors to the same norm of d. We show that each of these components is important for the attention layer that follows it in Transformers:(a) projection allows the attention mechanism to create an attention query that attends to all keys equally, offloading the need to learn this operation in the attention; and(b) scaling allows each key to potentially receive the highest attention, and prevents keys from being “un-select-able”.We show empirically that Transformers do indeed benefit from these properties of LayeNorm in general language modeling and even in computing simple functions such as “majority”. Our code is available at https://github.com/tech-srl/layer_norm_expressivity_role .

pdf abs
DEnsity: Open-domain Dialogue Evaluation Metric using Density Estimation
ChaeHun Park | Seungil Lee | Daniel Rim | Jaegul Choo

Despite the recent advances in open-domain dialogue systems, building a reliable evaluation metric is still a challenging problem. Recent studies proposed learnable metrics based on classification models trained to distinguish the correct response. However, neural classifiers are known to make overly confident predictions for examples from unseen distributions. We propose DENSITY, which evaluates a response by utilizing density estimation on the feature space derived from a neural classifier. Our metric measures how likely a response would appear in the distribution of human conversations. Moreover, to improve the performance of DENSITY, we utilize contrastive learning to further compress the feature space. Experiments on multiple response evaluation datasets show that DENSITY correlates better with human evaluations than the existing metrics.

pdf abs
Fixing MoE Over-Fitting on Low-Resource Languages in Multilingual Machine Translation
Maha Elbayad | Anna Sun | Shruti Bhosale

Sparsely gated Mixture of Experts (MoE) models have been shown to be a compute-efficient method to scale model capacity for multilingual machine translation. However, for low-resource tasks, MoE models severely over-fit. We show effective regularization strategies, namely dropout techniques for MoE layers in EOM and FOM, Conditional MoE Routing and Curriculum Learning methods that prevent over-fitting and improve the performance of MoE models on low-resource tasks without adversely affecting high-resource tasks. On a massively multilingual machine translation benchmark, our strategies result in about +1 chrF++ improvement in very low resource language pairs. We perform an extensive analysis of the learned MoE routing to better understand the impact of our regularization methods and how we can improve them.

pdf abs
Intent Discovery with Frame-guided Semantic Regularization and Augmentation
Yajing Sun | Rui Zhang | Jingyuan Yang | Wei Peng

Most existing intent discovery methods leverage representation learning and clustering to transfer the prior knowledge of known intents to unknown ones. The learned representations are limited to the syntactic forms of sentences, therefore, fall short of recognizing adequate variations under the same meaning of unknown intents. This paper proposes an approach utilizing frame knowledge as conceptual semantic guidance to bridge the gap between known intents representation learning and unknown intents clustering. Specifically, we employ semantic regularization to minimize the bidirectional KL divergence between model predictions for frame-based and sentence-based samples. Moreover, we construct a frame-guided data augmenter to capture intent-friendly semantic information and implement contrastive clustering learning for unsupervised sentence embedding. Extensive experiments on two benchmark datasets show that our method achieves substantial improvements in accuracy (5%+) compared to solid baselines.

pdf abs
An Empirical Comparison of LM-based Question and Answer Generation Methods
Asahi Ushio | Fernando Alva-Manchego | Jose Camacho-Collados

Question and answer generation (QAG) consists of generating a set of question-answer pairs given a context (e.g. a paragraph). This task has a variety of applications, such as data augmentation for question answering (QA) models, information retrieval and education. In this paper, we establish baselines with three different QAG methodologies that leverage sequence-to-sequence language model (LM) fine-tuning. Experiments show that an end-to-end QAG model, which is computationally light at both training and inference times, is generally robust and outperforms other more convoluted approaches. However, there are differences depending on the underlying generative LM. Finally, our analysis shows that QA models fine-tuned solely on generated question-answer pairs can be competitive when compared to supervised QA models trained on human-labeled data.

With the evolution of Knowledge Graphs (KGs), new entities emerge which are not seen before. Representation learning of KGs in such an inductive setting aims to capture and transfer the structural patterns from existing entities to new entities. However, the performance of existing methods in inductive KGs are limited by sparsity and implicit transfer. In this paper, we propose VMCL, a Contrastive Learning (CL) framework with graph guided Variational autoencoder on Meta-KGs in the inductive setting. We first propose representation generation to capture the encoded and generated representations of entities, where the generated variations can densify representations with complementary features. Then, we design two CL objectives that work across entities and meta-KGs to simulate the transfer mode. With extensive experiments we demonstrate that our proposed VMCL can significantly outperform previous state-of-the-art baselines.

pdf abs
Decouple knowledge from paramters for plug-and-play language modeling
Xin Cheng | Yankai Lin | Xiuying Chen | Dongyan Zhao | Rui Yan

Pre-trained language models (PLM) have made impressive results in a wide range of NLP tasks and it has been revealed that one of the key factors to their success is the parameters of these models implicitly learn various types of knowledge in the pre-training corpus. However, encoding knowledge implicitly in the model parameters has two fundamental drawbacks. First, the knowledge is neither editable nor scalable once the model is trained, which is especially problematic in that knowledge is consistently evolving. Second, it lacks interpretability and prevents us from understanding what kind of knowledge PLM needs to solve a certain task. In this paper, we introduce {pasted macro ‘MODEL’}, a pre-training model with differentiable plug-in memory (DPM). The key intuition behind is to decouple the knowledge storage from model parameters with an editable and scalable key-value memory and leverage knowledge in an explainable manner by knowledge retrieval in the {pasted macro ‘MEMORY’}. We conduct extensive experiments under various settings to justify this design choice. In domain adaptation setting, {pasted macro ‘MODEL’} could be easily adapted to different domains with pluggable in-domain memory—obtaining 3.95 F1 improvements across four domains, without any in-domain training. {pasted macro ‘MODEL’} could also keep absorbing new knowledge after pre-training is done by knowledge updating operation in the {pasted macro ‘MEMORY’} without re-training. Finally, we show that by incorporating training samples into {pasted macro ‘MEMORY’} with knowledge prompting, {pasted macro ‘MODEL’} could further be improved by the instruction of in-task knowledge.