To democratize large language models (LLMs) to most natural languages, it is imperative to make these models capable of understanding and generating texts in many languages, in particular low-resource ones. While recent multilingual LLMs demonstrate remarkable performance in such capabilities, these LLMs still support a limited number of human languages due to the lack of training data for low resource languages. Moreover, these LLMs are not yet aligned with human preference for downstream tasks, which is crucial for the success of LLMs in English. In this paper, we introduce xLLaMA-100 and xBLOOM-100 (collectively xLLMs-100), which scale the multilingual capabilities of LLaMA and BLOOM to 100 languages. To do so, we construct two datasets: a multilingual instruction dataset including 100 languages, which represents the largest language coverage to date, and a cross-lingual human feedback dataset encompassing 30 languages. We perform multilingual instruction tuning on the constructed instruction data and further align the LLMs with human feedback using the DPO algorithm on our cross-lingual human feedback dataset. We evaluate the multilingual understanding and generating capabilities of xLLMs-100 on five multilingual benchmarks. Experimental results show that xLLMs-100 consistently outperforms its peers across the benchmarks by considerable margins, defining a new state-of-the-art multilingual LLM that supports 100 languages.

Table Question Answering (TQA) aims at composing an answer to a question based on tabular data. While prior research has shown that TQA models lack robustness, understanding the underlying cause and nature of this issue remains predominantly unclear, posing a significant obstacle to the development of robust TQA systems. In this paper, we formalize three major desiderata for a fine-grained evaluation of robustness of TQA systems. They should (i) answer questions regardless of alterations in table structure, (ii) base their responses on the content of relevant cells rather than on biases, and (iii) demonstrate robust numerical reasoning capabilities. To investigate these aspects, we create and publish a novel TQA evaluation benchmark in English. Our extensive experimental analysis reveals that none of the examined state-of-the-art TQA systems consistently excels in these three aspects. Our benchmark is a crucial instrument for monitoring the behavior of TQA systems and paves the way for the development of robust TQA systems. We release our benchmark publicly.

Generative language models have recently shown remarkable success in generating answers to questions in a given textual context. However, these answers may suffer from hallucination, wrongly cite evidence, and spread misleading information. In this work, we address this problem by employing ChatGPT, a state-of-the-art generative model, as a machine-reading system. We ask it to retrieve answers to lexically varied and open-ended questions from trustworthy instructive texts. We introduce WHERE (WikiHow Evidence REtrieval), a new high-quality evaluation benchmark of a set of WikiHow articles exhaustively annotated with evidence sentences to questions that comes with a special challenge: All questions are about the article’s topic, but not all can be answered using the provided context. We interestingly find that when using a regular question-answering prompt, ChatGPT neglects to detect the unanswerable cases. When provided with a few examples, it learns to better judge whether a text provides answer evidence or not. Alongside this important finding, our dataset defines a new benchmark for evidence retrieval in question answering, which we argue is one of the necessary next steps for making large language models more trustworthy.

In task-oriented dialog (ToD) new intents emerge on regular basis, with a handful of available utterances at best. This renders effective Few-Shot Intent Classification (FSIC) a central challenge for modular ToD systems. Recent FSIC methods appear to be similar: they use pretrained language models (PLMs) to encode utterances and predominantly resort to nearest-neighbor-based inference. However, they also differ in major components: they start from different PLMs, use different encoding architectures and utterance similarity functions, and adopt different training regimes. Coupling of these vital components together with the lack of informative ablations prevents the identification of factors that drive the (reported) FSIC performance. We propose a unified framework to evaluate these components along the following key dimensions:(1) Encoding architectures: Cross-Encoder vs Bi-Encoders;(2) Similarity function: Parameterized (i.e., trainable) vs non-parameterized; (3) Training regimes: Episodic meta-learning vs conventional (i.e., non-episodic) training. Our experimental results on seven FSIC benchmarks reveal three new important findings. First, the unexplored combination of cross-encoder architecture and episodic meta-learning consistently yields the best FSIC performance. Second, episodic training substantially outperforms its non-episodic counterpart. Finally, we show that splitting episodes into support and query sets has a limited and inconsistent effect on performance. Our findings show the importance of ablations and fair comparisons in FSIC. We publicly release our code and data.

With recent advances in question-answering models, various datasets have been collected to improve and study the effectiveness of these models on scientific texts. Questions and answers in these datasets explore a scientific paper by seeking factual information from the paper’s content. However, these datasets do not tackle the argumentative content of scientific papers, which is of huge importance in persuasiveness of a scientific discussion. We introduce ArgSciChat, a dataset of 41 argumentative dialogues between scientists on 20 NLP papers. The unique property of our dataset is that it includes both exploratory and argumentative questions and answers in a dialogue discourse on a scientific paper. Moreover, the size of ArgSciChat demonstrates the difficulties in collecting dialogues for specialized domains. Thus, our dataset is a challenging resource to evaluate dialogue agents in low-resource domains, in which collecting training data is costly. We annotate all sentences of dialogues in ArgSciChat and analyze them extensively. The results confirm that dialogues in ArgSciChat include exploratory and argumentative interactions. Furthermore, we use our dataset to fine-tune and evaluate a pre-trained document-grounded dialogue agent. The agent achieves a low performance on our dataset, motivating a need for dialogue agents with a capability to reason and argue about their answers. We publicly release ArgSciChat.

Code generation from text requires understanding the user’s intent from a natural languagedescription and generating an executable code snippet that satisfies this intent. While recent pretrained language models demonstrate remarkable performance for this task, these models fail when the given natural language description is under-specified. In this work, we introduce a novel and more realistic setup for this task. We hypothesize that the under-specification of a natural language description can be resolved by asking clarification questions. Therefore, we collect and introduce a new dataset named CodeClarQA containing pairs of natural language descriptions and code with created synthetic clarification questions and answers. The empirical results of our evaluation of pretrained language model performance on code generation show that clarifications result in more precisely generated code, as shown by the substantial improvement of model performance in all evaluation metrics. Alongside this, our task and dataset introduce new challenges to the community, including when and what clarification questions should be asked. Our code and dataset are available on GitHub.

Neural models for response generation produce responses that are semantically plausible but not necessarily factually consistent with facts describing the speaker’s persona. These models are trained with fully supervised learning where the objective function barely captures factual consistency. We propose to fine-tune these models by reinforcement learning and an efficient reward function that explicitly captures the consistency between a response and persona facts as well as semantic plausibility. Our automatic and human evaluations on the PersonaChat corpus confirm that our approach increases the rate of responses that are factually consistent with persona facts over its supervised counterpart while retains the language quality of responses.

Entity grids and entity graphs are two frameworks for modeling local coherence. These frameworks represent entity relations between sentences and then extract features from such representations to encode coherence. The benefits of convolutional neural models for extracting informative features from entity grids have been recently studied. In this work, we study the benefits of Relational Graph Convolutional Networks (RGCN) to encode entity graphs for measuring local coherence. We evaluate our neural graph-based model for two benchmark coherence evaluation tasks: sentence ordering (SO) and summary coherence rating (SCR). The results show that our neural graph-based model consistently outperforms the neural grid-based model for both tasks. Our model performs competitively with a strong baseline coherence model, while our model uses 50% fewer parameters. Our work defines a new, efficient, and effective baseline for local coherence modeling.

Recent dialogue coherence models use the coherence features designed for monologue texts, e.g. nominal entities, to represent utterances and then explicitly augment them with dialogue-relevant features, e.g., dialogue act labels. It indicates two drawbacks, (a) semantics of utterances are limited to entity mentions, and (b) the performance of coherence models strongly relies on the quality of the input dialogue act labels. We address these issues by introducing a novel approach to dialogue coherence assessment. We use dialogue act prediction as an auxiliary task in a multi-task learning scenario to obtain informative utterance representations for coherence assessment. Our approach alleviates the need for explicit dialogue act labels during evaluation. The results of our experiments show that our model substantially (more than 20 accuracy points) outperforms its strong competitors on the DailyDialogue corpus, and performs on par with them on the SwitchBoard corpus for ranking dialogues concerning their coherence. We release our source code.

Visual modifications to text are often used to obfuscate offensive comments in social media (e.g., “!d10t”) or as a writing style (“1337” in “leet speak”), among other scenarios. We consider this as a new type of adversarial attack in NLP, a setting to which humans are very robust, as our experiments with both simple and more difficult visual perturbations demonstrate. We investigate the impact of visual adversarial attacks on current NLP systems on character-, word-, and sentence-level tasks, showing that both neural and non-neural models are, in contrast to humans, extremely sensitive to such attacks, suffering performance decreases of up to 82%. We then explore three shielding methods—visual character embeddings, adversarial training, and rule-based recovery—which substantially improve the robustness of the models. However, the shielding methods still fall behind performances achieved in non-attack scenarios, which demonstrates the difficulty of dealing with visual attacks.

We propose a local coherence model that captures the flow of what semantically connects adjacent sentences in a text. We represent the semantics of a sentence by a vector and capture its state at each word of the sentence. We model what relates two adjacent sentences based on the two most similar semantic states, each of which is in one of the sentences. We encode the perceived coherence of a text by a vector, which represents patterns of changes in salient information that relates adjacent sentences. Our experiments demonstrate that our approach is beneficial for two downstream tasks: Readability assessment, in which our model achieves new state-of-the-art results; and essay scoring, in which the combination of our coherence vectors and other task-dependent features significantly improves the performance of a strong essay scorer.

Although coherence is an important aspect of any text generation system, it has received little attention in the context of machine translation (MT) so far. We hypothesize that the quality of document-level translation can be improved if MT models take into account the semantic relations among sentences during translation. We integrate the graph-based coherence model proposed by Mesgar and Strube, (2016) with Docent (Hardmeier et al., 2012, Hardmeier, 2014) a document-level machine translation system. The application of this graph-based coherence modeling approach is novel in the context of machine translation. We evaluate the coherence model and its effects on the quality of the machine translation. The result of our experiments shows that our coherence model slightly improves the quality of translation in terms of the average Meteor score.