We propose a touch-based editing method for translation, which is more flexible than traditional keyboard-mouse-based translation postediting. This approach relies on touch actions that users perform to indicate translation errors. We present a dual-encoder model to handle the actions and generate refined translations. To mimic the user feedback, we adopt the TER algorithm comparing between draft translations and references to automatically extract the simulated actions for training data construction. Experiments on translation datasets with simulated editing actions show that our method significantly improves original translation of Transformer (up to 25.31 BLEU) and outperforms existing interactive translation methods (up to 16.64 BLEU). We also conduct experiments on post-editing dataset to further prove the robustness and effectiveness of our method.
Multilingual pretrained language models (such as multilingual BERT) have achieved impressive results for cross-lingual transfer. However, due to the constant model capacity, multilingual pre-training usually lags behind the monolingual competitors. In this work, we present two approaches to improve zero-shot cross-lingual classification, by transferring the knowledge from monolingual pretrained models to multilingual ones. Experimental results on two cross-lingual classification benchmarks show that our methods outperform vanilla multilingual fine-tuning.
Social media platforms such as Twitter have become a breeding ground for unverified information or rumors. These rumors can threaten people’s health, endanger the economy, and affect the stability of a country. Many researchers have developed models to classify rumors using traditional machine learning or vanilla deep learning models. However, previous studies on rumor detection have achieved low precision and are time consuming. Inspired by the hierarchical model and multitask learning, a multiloss hierarchical BiLSTM model with an attenuation factor is proposed in this paper. The model is divided into two BiLSTM modules: post level and event level. By means of this hierarchical structure, the model can extract deep information from limited quantities of text. Each module has a loss function that helps to learn bilateral features and reduce the training time. An attenuation factor is added at the post level to increase the accuracy. The results on two rumor datasets demonstrate that our model achieves better performance than that of state-of-the-art machine learning and vanilla deep learning models.
Aspect-level sentiment analysis(ASC) predicts each specific aspect term’s sentiment polarity in a given text or review. Recent studies used attention-based methods that can effectively improve the performance of aspect-level sentiment analysis. These methods ignored the syntactic relationship between the aspect and its corresponding context words, leading the model to focus on syntactically unrelated words mistakenly. One proposed solution, the graph convolutional network (GCN), cannot completely avoid the problem. While it does incorporate useful information about syntax, it assigns equal weight to all the edges between connected words. It may still incorrectly associate unrelated words to the target aspect through the iterations of graph convolutional propagation. In this study, a graph attention network with memory fusion is proposed to extend GCN’s idea by assigning different weights to edges. Syntactic constraints can be imposed to block the graph convolutional propagation of unrelated words. A convolutional layer and a memory fusion were applied to learn and exploit multiword relations and draw different weights of words to improve performance further. Experimental results on five datasets show that the proposed method yields better performance than existing methods.
Unlike non-conversation scenes, emotion recognition in dialogues (ERD) poses more complicated challenges due to its interactive nature and intricate contextual information. All present methods model historical utterances without considering the content of the target utterance. However, different parts of a historical utterance may contribute differently to emotion inference of different target utterances. Therefore we propose Fine-grained Extraction and Reasoning Network (FERNet) to generate target-specific historical utterance representations. The reasoning module effectively handles both local and global sequential dependencies to reason over context, and updates target utterance representations to more informed vectors. Experiments on two benchmarks show that our method achieves competitive performance compared with previous methods.
Personalized news recommendation is important for online news services. Many news recommendation methods recommend news based on their relevance to users’ historical browsed news, and the recommended news usually have similar sentiment with browsed news. However, if browsed news is dominated by certain kinds of sentiment, the model may intensively recommend news with the same sentiment orientation, making it difficult for users to receive diverse opinions and news events. In this paper, we propose a sentiment diversity-aware neural news recommendation approach, which can recommend news with more diverse sentiment. In our approach, we propose a sentiment-aware news encoder, which is jointly trained with an auxiliary sentiment prediction task, to learn sentiment-aware news representations. We learn user representations from browsed news representations, and compute click scores based on user and candidate news representations. In addition, we propose a sentiment diversity regularization method to penalize the model by combining the overall sentiment orientation of browsed news as well as the click and sentiment scores of candidate news. Extensive experiments on real-world dataset show that our approach can effectively improve the sentiment diversity in news recommendation without performance sacrifice.
Similarity search is to find the most similar items for a certain target item. The ability of similarity search at large scale plays a significant role in many information retrieval applications, and thus has received much attention. Text hashing is a promising strategy, which utilizes binary encoding to represent documents, obtaining attractive performance. This paper makes the first attempt to utilize Bayesian Clustering for Text Hashing, dubbed as BCTH. Specifically, BCTH is able to map documents to binary codes by utilizing multiple Bayesian Clusterings in parallel, where each Bayesian Clustering is responsible for one bit. Our approach employs the bit-balanced constraint to maximize the amount of information in each bit. Meanwhile, the bit-uncorrected constraint is adopted to keep the independence among all bits. The time complexity of BCTH is linear, where the hash codes and hash function are jointly learned. The experimental results, based on four widely-used datasets, demonstrate that BCTH is competitive, compared with currently competitive baselines in the perspective of both precision and training speed.
Large and complex models have recently been developed that require many parameters and much time to solve various problems in natural language processing. This paper explores an efficient way to avoid models being too complicated and ensure nearly equal performance to models showing the state-of-the-art. We propose a single convolutional neural network (CNN) using the sinusoidal positional encoding (SPE) in text classification. The SPE provides useful position information of a word and can construct a more efficient model architecture than before in a CNN-based approach. Our model can significantly reduce the parameter size (at least 67%) and training time (up to 85%) while maintaining similar performance to the CNN-based approach on multiple benchmark datasets.
Recently, BERT has become an essential ingredient of various NLP deep models due to its effectiveness and universal-usability. However, the online deployment of BERT is often blocked by its large-scale parameters and high computational cost. There are plenty of studies showing that the knowledge distillation is efficient in transferring the knowledge from BERT into the model with a smaller size of parameters. Nevertheless, current BERT distillation approaches mainly focus on task-specified distillation, such methodologies lead to the loss of the general semantic knowledge of BERT for universal-usability. In this paper, we propose a sentence representation approximating oriented distillation framework that can distill the pre-trained BERT into a simple LSTM based model without specifying tasks. Consistent with BERT, our distilled model is able to perform transfer learning via fine-tuning to adapt to any sentence-level downstream task. Besides, our model can further cooperate with task-specific distillation procedures. The experimental results on multiple NLP tasks from the GLUE benchmark show that our approach outperforms other task-specific distillation methods or even much larger models, i.e., ELMO, with efficiency well-improved.
We propose a simple and effective method for incorporating word clusters into the Continuous Bag-of-Words (CBOW) model. Specifically, we propose to replace infrequent input and output words in CBOW model with their clusters. The resulting cluster-incorporated CBOW model produces embeddings of frequent words and a small amount of cluster embeddings, which will be fine-tuned in downstream tasks. We empirically show our replacing method works well on several downstream tasks. Through our analysis, we show that our method might be also useful for other similar models which produce word embeddings.
The recently introduced pre-trained language model BERT advances the state-of-the-art on many NLP tasks through the fine-tuning approach, but few studies investigate how the fine-tuning process improves the model performance on downstream tasks. In this paper, we inspect the learning dynamics of BERT fine-tuning with two indicators. We use JS divergence to detect the change of the attention mode and use SVCCA distance to examine the change to the feature extraction mode during BERT fine-tuning. We conclude that BERT fine-tuning mainly changes the attention mode of the last layers and modifies the feature extraction mode of the intermediate and last layers. Moreover, we analyze the consistency of BERT fine-tuning between different random seeds and different datasets. In summary, we provide a distinctive understanding of the learning dynamics of BERT fine-tuning, which sheds some light on improving the fine-tuning results.
In this paper, we propose second-order graph-based neural dependency parsing using message passing and end-to-end neural networks. We empirically show that our approaches match the accuracy of very recent state-of-the-art second-order graph-based neural dependency parsers and have significantly faster speed in both training and testing. We also empirically show the advantage of second-order parsing over first-order parsing and observe that the usefulness of the head-selection structured constraint vanishes when using BERT embedding.
Current end-to-end semantic role labeling is mostly accomplished via graph-based neural models. However, these all are first-order models, where each decision for detecting any predicate-argument pair is made in isolation with local features. In this paper, we present a high-order refining mechanism to perform interaction between all predicate-argument pairs. Based on the baseline graph model, our high-order refining module learns higher-order features between all candidate pairs via attention calculation, which are later used to update the original token representations. After several iterations of refinement, the underlying token representations can be enriched with globally interacted features. Our high-order model achieves state-of-the-art results on Chinese SRL data, including CoNLL09 and Universal Proposition Bank, meanwhile relieving the long-range dependency issues.
Answer selection (AS) is an important subtask of document-based question answering (DQA). In this task, the candidate answers come from the same document, and each answer sentence is semantically related to the given question, which makes it more challenging to select the true answer. WordNet provides powerful knowledge about concepts and their semantic relations so we employ WordNet to enrich the abilities of paraphrasing and reasoning of the network-based question answering model. Specifically, we exploit the synset and hypernym concepts to enrich the word representation and incorporate the similarity scores of two concepts that share the synset or hypernym relations into the attention mechanism. The proposed WordNet-enhanced hierarchical model (WEHM) consists of four modules, including WordNet-enhanced word representation, sentence encoding, WordNet-enhanced attention mechanism, and hierarchical document encoding. Extensive experiments on the public WikiQA and SelQA datasets demonstrate that our proposed model significantly improves the baseline system and outperforms all existing state-of-the-art methods by a large margin.
In this paper, we propose a simple method to predict salient locations from news article text using a knowledge base (KB). The proposed method uses a dictionary of locations created from the KB to identify occurrences of locations in the text and uses the hierarchical information between entities in the KB for assigning appropriate saliency scores to regions. It allows prediction at arbitrary region units and has only a few hyperparameters that need to be tuned. We show using manually annotated news articles that the proposed method improves the f-measure by > 0.12 compared to multiple baselines.
Most existing approaches for goal-oriented dialogue policy learning used reinforcement learning, which focuses on the target agent policy and simply treats the opposite agent policy as part of the environment. While in real-world scenarios, the behavior of an opposite agent often exhibits certain patterns or underlies hidden policies, which can be inferred and utilized by the target agent to facilitate its own decision making. This strategy is common in human mental simulation by first imaging a specific action and the probable results before really acting it. We therefore propose an opposite behavior aware framework for policy learning in goal-oriented dialogues. We estimate the opposite agent’s policy from its behavior and use this estimation to improve the target agent by regarding it as part of the target policy. We evaluate our model on both cooperative and competitive dialogue tasks, showing superior performance over state-of-the-art baselines.
Typically, tokenization is the very first step in most text processing works. As a token serves as an atomic unit that embeds the contextual information of text, how to define a token plays a decisive role in the performance of a model. Even though Byte Pair Encoding (BPE) has been considered the de facto standard tokenization method due to its simplicity and universality, it still remains unclear whether BPE works best across all languages and tasks. In this paper, we test several tokenization strategies in order to answer our primary research question, that is, “What is the best tokenization strategy for Korean NLP tasks?” Experimental results demonstrate that a hybrid approach of morphological segmentation followed by BPE works best in Korean to/from English machine translation and natural language understanding tasks such as KorNLI, KorSTS, NSMC, and PAWS-X. As an exception, for KorQuAD, the Korean extension of SQuAD, BPE segmentation turns out to be the most effective. Our code and pre-trained models are publicly available at https://github.com/kakaobrain/kortok.
We propose a novel, accurate, and explainable recommender model (BENEFICT) that addresses two drawbacks that most review-based recommender systems face. First is their utilization of traditional word embeddings that could influence prediction performance due to their inability to model the word semantics’ dynamic characteristic. Second is their black-box nature that makes the explanations behind every prediction obscure. Our model uniquely integrates three key elements: BERT, multilayer perceptron, and maximum subarray problem to derive contextualized review features, model user-item interactions, and generate explanations, respectively. Our experiments show that BENEFICT consistently outperforms other state-of-the-art models by an average improvement gain of nearly 7%. Based on the human judges’ assessment, the BENEFICT-produced explanations can capture the essence of the customer’s preference and help future customers make purchasing decisions. To the best of our knowledge, our model is one of the first recommender models to utilize BERT for neural collaborative filtering.
By predicting chemical compound structures from their names, we can better comprehend chemical compounds written in text and identify the same chemical compound given different notations for database creation. Previous methods have predicted the chemical compound structures from their names and represented them by Simplified Molecular Input Line Entry System (SMILES) strings. However, these methods mainly apply handcrafted rules, and cannot predict the structures of chemical compound names not covered by the rules. Instead of handcrafted rules, we propose Transformer-based models that predict SMILES strings from chemical compound names. We improve the conventional Transformer-based model by introducing two features: (1) a loss function that constrains the number of atoms of each element in the structure, and (2) a multi-task learning approach that predicts both SMILES strings and InChI strings (another string representation of chemical compound structures). In evaluation experiments, our methods achieved higher F-measures than previous rule-based approaches (Open Parser for Systematic IUPAC Nomenclature and two commercially used products), and the conventional Transformer-based model. We release the dataset used in this paper as a benchmark for the future research.
In recent years, pre-trained models have been extensively studied, and several downstream tasks have benefited from their utilization. In this study, we verify the effectiveness of two methods that incorporate a pre-trained model into an encoder-decoder model on Chinese grammatical error correction tasks. We also analyze the error type and conclude that sentence-level errors are yet to be addressed.
Event Argument Extraction (EAE) aims at predicting event argument roles of entities in text, which is a crucial subtask and bottleneck of event extraction. Existing EAE methods either extract each event argument roles independently or sequentially, which cannot adequately model the joint probability distribution among event arguments and their roles. In this paper, we propose a Bayesian model named Neural Gibbs Sampling (NGS) to jointly extract event arguments. Specifically, we train two neural networks to model the prior distribution and conditional distribution over event arguments respectively and then use Gibbs sampling to approximate the joint distribution with the learned distributions. For overcoming the shortcoming of the high complexity of the original Gibbs sampling algorithm, we further apply simulated annealing to efficiently estimate the joint probability distribution over event arguments and make predictions. We conduct experiments on the two widely-used benchmark datasets ACE 2005 and TAC KBP 2016. The Experimental results show that our NGS model can achieve comparable results to existing state-of-the-art EAE methods. The source code can be obtained from https://github.com/THU-KEG/NGS.
Named entity recognition is a critical task in the natural language processing field. Most existing methods for this task can only exploit contextual information within a sentence. However, their performance on recognizing entities in limited or ambiguous sentence-level contexts is usually unsatisfactory. Fortunately, other sentences in the same document can provide supplementary document-level contexts to help recognize these entities. In addition, words themselves contain word-level contextual information since they usually have different preferences of entity type and relative position from named entities. In this paper, we propose a unified framework to incorporate multi-level contexts for named entity recognition. We use TagLM as our basic model to capture sentence-level contexts. To incorporate document-level contexts, we propose to capture interactions between sentences via a multi-head self attention network. To mine word-level contexts, we propose an auxiliary task to predict the type of each word to capture its type preference. We jointly train our model in entity recognition and the auxiliary classification task via multi-task learning. The experimental results on several benchmark datasets validate the effectiveness of our method.
Despite the success of neural machine translation (NMT), simultaneous neural machine translation (SNMT), the task of translating in real time before a full sentence has been observed, remains challenging due to the syntactic structure difference and simultaneity requirements. In this paper, we propose a general framework for adapting neural machine translation to translate simultaneously. Our framework contains two parts: prefix translation that utilizes a consecutive NMT model to translate source prefixes and a stopping criterion that determines when to stop the prefix translation. Experiments on three translation corpora and two language pairs show the efficacy of the proposed framework on balancing the quality and latency in adapting NMT to perform simultaneous translation.
Chinese and Japanese share many characters with similar surface morphology. To better utilize the shared knowledge across the languages, we propose UnihanLM, a self-supervised Chinese-Japanese pretrained masked language model (MLM) with a novel two-stage coarse-to-fine training approach. We exploit Unihan, a ready-made database constructed by linguistic experts to first merge morphologically similar characters into clusters. The resulting clusters are used to replace the original characters in sentences for the coarse-grained pretraining of the MLM. Then, we restore the clusters back to the original characters in sentences for the fine-grained pretraining to learn the representation of the specific characters. We conduct extensive experiments on a variety of Chinese and Japanese NLP benchmarks, showing that our proposed UnihanLM is effective on both mono- and cross-lingual Chinese and Japanese tasks, shedding light on a new path to exploit the homology of languages.
In order to combat overfitting and in pursuit of better generalization, label smoothing is widely applied in modern neural machine translation systems. The core idea is to penalize over-confident outputs and regularize the model so that its outputs do not diverge too much from some prior distribution. While training perplexity generally gets worse, label smoothing is found to consistently improve test performance. In this work, we aim to better understand label smoothing in the context of neural machine translation. Theoretically, we derive and explain exactly what label smoothing is optimizing for. Practically, we conduct extensive experiments by varying which tokens to smooth, tuning the probability mass to be deducted from the true targets and considering different prior distributions. We show that label smoothing is theoretically well-motivated, and by carefully choosing hyperparameters, the practical performance of strong neural machine translation systems can be further improved.
In linguistics and cognitive science, Logical metonymies are defined as type clashes between an event-selecting verb and an entity-denoting noun (e.g. The editor finished the article), which are typically interpreted by inferring a hidden event (e.g. reading) on the basis of contextual cues. This paper tackles the problem of logical metonymy interpretation, that is, the retrieval of the covert event via computational methods. We compare different types of models, including the probabilistic and the distributional ones previously introduced in the literature on the topic. For the first time, we also tested on this task some of the recent Transformer-based models, such as BERT, RoBERTa, XLNet, and GPT-2. Our results show a complex scenario, in which the best Transformer-based models and some traditional distributional models perform very similarly. However, the low performance on some of the testing datasets suggests that logical metonymy is still a challenging phenomenon for computational modeling.
Structured semantic sentence representations such as Abstract Meaning Representations (AMRs) are potentially useful in various NLP tasks. However, the quality of automatic parses can vary greatly and jeopardizes their usefulness. This can be mitigated by models that can accurately rate AMR quality in the absence of costly gold data, allowing us to inform downstream systems about an incorporated parse’s trustworthiness or select among different candidate parses. In this work, we propose to transfer the AMR graph to the domain of images. This allows us to create a simple convolutional neural network (CNN) that imitates a human judge tasked with rating graph quality. Our experiments show that the method can rate quality more accurately than strong baselines, in several quality dimensions. Moreover, the method proves to be efficient and reduces the incurred energy consumption.
Commonsense explanation generation aims to empower the machine’s sense-making capability by generating plausible explanations to statements against commonsense. While this task is easy to human, the machine still struggles to generate reasonable and informative explanations. In this work, we propose a method that first extracts the underlying concepts which are served as bridges in the reasoning chain and then integrates these concepts to generate the final explanation. To facilitate the reasoning process, we utilize external commonsense knowledge to build the connection between a statement and the bridge concepts by extracting and pruning multi-hop paths to build a subgraph. We design a bridge concept extraction model that first scores the triples, routes the paths in the subgraph, and further selects bridge concepts with weak supervision at both the triple level and the concept level. We conduct experiments on the commonsense explanation generation task and our model outperforms the state-of-the-art baselines in both automatic and human evaluation.
KB-to-text task aims at generating texts based on the given KB triples. Traditional methods usually map KB triples to sentences via a supervised seq-to-seq model. However, existing annotated datasets are very limited and human labeling is very expensive. In this paper, we propose a method which trains the generation model in a completely unsupervised way with unaligned raw text data and KB triples. Our method exploits a novel dual training framework which leverages the inverse relationship between the KB-to-text generation task and an auxiliary triple extraction task. In our architecture, we reconstruct KB triples or texts via a closed-loop framework via linking a generator and an extractor. Therefore the loss function that accounts for the reconstruction error of KB triples and texts can be used to train the generator and extractor. To resolve the cold start problem in training, we propose a method using a pseudo data generator which generates pseudo texts and KB triples for learning an initial model. To resolve the multiple-triple problem, we design an allocated reinforcement learning component to optimize the reconstruction loss. The experimental results demonstrate that our model can outperform other unsupervised generation methods and close to the bound of supervised methods.
Despite the recent achievements made in the multi-modal emotion recognition task, two problems still exist and have not been well investigated: 1) the relationship between different emotion categories are not utilized, which leads to sub-optimal performance; and 2) current models fail to cope well with low-resource emotions, especially for unseen emotions. In this paper, we propose a modality-transferable model with emotion embeddings to tackle the aforementioned issues. We use pre-trained word embeddings to represent emotion categories for textual data. Then, two mapping functions are learned to transfer these embeddings into visual and acoustic spaces. For each modality, the model calculates the representation distance between the input sequence and target emotions and makes predictions based on the distances. By doing so, our model can directly adapt to the unseen emotions in any modality since we have their pre-trained embeddings and modality mapping functions. Experiments show that our model achieves state-of-the-art performance on most of the emotion categories. Besides, our model also outperforms existing baselines in the zero-shot and few-shot scenarios for unseen emotions.
In this paper, we aim at learning the relationships and similarities of a variety of tasks, such as humour detection, sarcasm detection, offensive content detection, motivational content detection and sentiment analysis on a somewhat complicated form of information, i.e., memes. We propose a multi-task, multi-modal deep learning framework to solve multiple tasks simultaneously. For multi-tasking, we propose two attention-like mechanisms viz., Inter-task Relationship Module (iTRM) and Inter-class Relationship Module (iCRM). The main motivation of iTRM is to learn the relationship between the tasks to realize how they help each other. In contrast, iCRM develops relations between the different classes of tasks. Finally, representations from both the attentions are concatenated and shared across the five tasks (i.e., humour, sarcasm, offensive, motivational, and sentiment) for multi-tasking. We use the recently released dataset in the Memotion Analysis task @ SemEval 2020, which consists of memes annotated for the classes as mentioned above. Empirical results on Memotion dataset show the efficacy of our proposed approach over the existing state-of-the-art systems (Baseline and SemEval 2020 winner). The evaluation also indicates that the proposed multi-task framework yields better performance over the single-task learning.
We propose Implicit Quote Extractor, an end-to-end unsupervised extractive neural summarization model for conversational texts. When we reply to posts, quotes are used to highlight important part of texts. We aim to extract quoted sentences as summaries. Most replies do not explicitly include quotes, so it is difficult to use quotes as supervision. However, even if it is not explicitly shown, replies always refer to certain parts of texts; we call them implicit quotes. Implicit Quote Extractor aims to extract implicit quotes as summaries. The training task of the model is to predict whether a reply candidate is a true reply to a post. For prediction, the model has to choose a few sentences from the post. To predict accurately, the model learns to extract sentences that replies frequently refer to. We evaluate our model on two email datasets and one social media dataset, and confirm that our model is useful for extractive summarization. We further discuss two topics; one is whether quote extraction is an important factor for summarization, and the other is whether our model can capture salient sentences that conventional methods cannot.
Unsupervised style transfer in text has previously been explored through the sentiment transfer task. The task entails inverting the overall sentiment polarity in a given input sentence, while preserving its content. From the Aspect-Based Sentiment Analysis (ABSA) task, we know that multiple sentiment polarities can often be present together in a sentence with multiple aspects. In this paper, the task of aspect-level sentiment controllable style transfer is introduced, where each of the aspect-level sentiments can individually be controlled at the output. To achieve this goal, a BERT-based encoder-decoder architecture with saliency weighted polarity injection is proposed, with unsupervised training strategies, such as ABSA masked-language-modelling. Through both automatic and manual evaluation, we show that the system is successful in controlling aspect-level sentiments.
Abstractive community detection is an important spoken language understanding task, whose goal is to group utterances in a conversation according to whether they can be jointly summarized by a common abstractive sentence. This paper provides a novel approach to this task. We first introduce a neural contextual utterance encoder featuring three types of self-attention mechanisms. We then train it using the siamese and triplet energy-based meta-architectures. Experiments on the AMI corpus show that our system outperforms multiple energy-based and non-energy based baselines from the state-of-the-art. Code and data are publicly available.
Modern task-oriented dialog systems need to reliably understand users’ intents. Intent detection is even more challenging when moving to new domains or new languages, since there is little annotated data. To address this challenge, we present a suite of pretrained intent detection models which can predict a broad range of intended goals from many actions because they are trained on wikiHow, a comprehensive instructional website. Our models achieve state-of-the-art results on the Snips dataset, the Schema-Guided Dialogue dataset, and all 3 languages of the Facebook multilingual dialog datasets. Our models also demonstrate strong zero- and few-shot performance, reaching over 75% accuracy using only 100 training examples in all datasets.
This work studies the widely adopted ancestral sampling algorithms for auto-regressive language models. We use the quality-diversity (Q-D) trade-off to investigate three popular sampling methods (top-k, nucleus and tempered sampling). We focus on the task of open-ended language generation, and first show that the existing sampling algorithms have similar performance. By carefully inspecting the transformations defined by different sampling algorithms, we identify three key properties that are shared among them: entropy reduction, order preservation, and slope preservation. To validate the importance of the identified properties, we design two sets of new sampling methods: one set in which each algorithm satisfies all three properties, and one set in which each algorithm violates at least one of the properties. We compare their performance with existing algorithms, and find that violating the identified properties could lead to drastic performance degradation, as measured by the Q-D trade-off. On the other hand, we find that the set of sampling algorithms that satisfy these properties performs on par with the existing sampling algorithms.
In this paper, we analyse the challenges of Chinese content scoring in comparison to English. As a review of prior work for Chinese content scoring shows a lack of open-access data in the field, we present two short-answer data sets for Chinese. The Chinese Educational Short Answers data set (CESA) contains 1800 student answers for five science-related questions. As a second data set, we collected ASAP-ZH with 942 answers by re-using three existing prompts from the ASAP data set. We adapt a state-of-the-art content scoring system for Chinese and evaluate it in several settings on these data sets. Results show that features on lower segmentation levels such as character n-grams tend to have better performance than features on token level.
This paper presents a new dataset, B-SHARP, that can be used to develop NLP models for the detection of Mild Cognitive Impairment (MCI) known as an early sign of Alzheimer’s disease. Our dataset contains 1-2 min speech segments from 326 human subjects for 3 topics, (1) daily activity, (2) room environment, and (3) picture description, and their transcripts so that a total of 650 speech segments are collected. Given the B-SHARP dataset, several hierarchical text classification models are developed that jointly learn combinatory features across all 3 topics. The best performance of 74.1% is achieved by an ensemble model that adapts 3 types of transformer encoders. To the best of our knowledge, this is the first work that builds deep learning-based text classification models on multiple contents for the detection of MCI.
Predicting the quality of machine translation has traditionally been addressed with language-specific models, under the assumption that the quality label distribution or linguistic features exhibit traits that are not shared across languages. An obvious disadvantage of this approach is the need for labelled data for each given language pair. We challenge this assumption by exploring different approaches to multilingual Quality Estimation (QE), including using scores from translation models. We show that these outperform single-language models, particularly in less balanced quality label distributions and low-resource settings. In the extreme case of zero-shot QE, we show that it is possible to accurately predict quality for any given new language from models trained on other languages. Our findings indicate that state-of-the-art neural QE models based on powerful pre-trained representations generalise well across languages, making them more applicable in real-world settings.
Approaching named entities transliteration as a Neural Machine Translation (NMT) problem is common practice. While many have applied various NMT techniques to enhance machine transliteration models, few focus on the linguistic features particular to the relevant languages. In this paper, we investigate the effect of incorporating phonetic features for English-to-Chinese transliteration under the multi-task learning (MTL) setting—where we define a phonetic auxiliary task aimed to improve the generalization performance of the main transliteration task. In addition to our system, we also release a new English-to-Chinese dataset and propose a novel evaluation metric which considers multiple possible transliterations given a source name. Our results show that the multi-task model achieves similar performance as the previous state of the art with a model of a much smaller size.
Attention-based encoder-decoder models have achieved great success in neural machine translation tasks. However, the lengths of the target sequences are not explicitly predicted in these models. This work proposes length prediction as an auxiliary task and set up a sub-network to obtain the length information from the encoder. Experimental results show that the length prediction sub-network brings improvements over the strong baseline system and that the predicted length can be used as an alternative to length normalization during decoding.
Recent work in unsupervised parsing has tried to incorporate visual information into learning, but results suggest that these models need linguistic bias to compete against models that only rely on text. This work proposes grammar induction models which use visual information from images for labeled parsing, and achieve state-of-the-art results on grounded grammar induction on several languages. Results indicate that visual information is especially helpful in languages where high frequency words are more broadly distributed. Comparison between models with and without visual information shows that the grounded models are able to use visual information for proposing noun phrases, gathering useful information from images for unknown words, and achieving better performance at prepositional phrase attachment prediction.
Transformer-based pre-trained language models (PLMs) have dramatically improved the state of the art in NLP across many tasks. This has led to substantial interest in analyzing the syntactic knowledge PLMs learn. Previous approaches to this question have been limited, mostly using test suites or probes. Here, we propose a novel fully unsupervised parsing approach that extracts constituency trees from PLM attention heads. We rank transformer attention heads based on their inherent properties, and create an ensemble of high-ranking heads to produce the final tree. Our method is adaptable to low-resource languages, as it does not rely on development sets, which can be expensive to annotate. Our experiments show that the proposed method often outperform existing approaches if there is no development set present. Our unsupervised parser can also be used as a tool to analyze the grammars PLMs learn implicitly. For this, we use the parse trees induced by our method to train a neural PCFG and compare it to a grammar derived from a human-annotated treebank.
Word embedding methods have become the de-facto way to represent words, having been successfully applied to a wide array of natural language processing tasks. In this paper, we explore the hypothesis that embedding methods can also be effectively used to represent spatial locations. Using a new dataset consisting of the location trajectories of 729 students over a seven month period and text data related to those locations, we implement several strategies to create location embeddings, which we then use to create embeddings of the sequences of locations a student has visited. To identify the surface level properties captured in the representations, we propose a number of probing tasks such as the presence of a specific location in a sequence or the type of activities that take place at a location. We then leverage the representations we generated and employ them in more complex downstream tasks ranging from predicting a student’s area of study to a student’s depression level, showing the effectiveness of these location embeddings.
Pairwise data automatically constructed from weakly supervised signals has been widely used for training deep learning models. Pairwise datasets such as parallel texts can have uneven quality levels overall, but usually contain data subsets that are more useful as learning examples. We present two methods to refine data that are aimed to obtain that kind of subsets in a self-supervised way. Our methods are based on iteratively training dual-encoder models to compute similarity scores. We evaluate our methods on de-noising parallel texts and training neural machine translation models. We find that: (i) The self-supervised refinement achieves most machine translation gains in the first iteration, but following iterations further improve its intrinsic evaluation. (ii) Machine translations can improve the de-noising performance when combined with selection steps. (iii) Our methods are able to reach the performance of a supervised method. Being entirely self-supervised, our methods are well-suited to handle pairwise data without the need of prior knowledge or human annotations.
Recent years have seen important advances in the quality of state-of-the-art models, but this has come at the expense of models becoming less interpretable. This survey presents an overview of the current state of Explainable AI (XAI), considered within the domain of Natural Language Processing (NLP). We discuss the main categorization of explanations, as well as the various ways explanations can be arrived at and visualized. We detail the operations and explainability techniques currently available for generating explanations for NLP model predictions, to serve as a resource for model developers in the community. Finally, we point out the current gaps and encourage directions for future work in this important research area.
Fine-tuning (FT) pre-trained sentence embedding models on small datasets has been shown to have limitations. In this paper we show that concatenating the embeddings from the pre-trained model with those from a simple sentence embedding model trained only on the target data, can improve over the performance of FT for few-sample tasks. To this end, a linear classifier is trained on the combined embeddings, either by freezing the embedding model weights or training the classifier and embedding models end-to-end. We perform evaluation on seven small datasets from NLP tasks and show that our approach with end-to-end training outperforms FT with negligible computational overhead. Further, we also show that sophisticated combination techniques like CCA and KCCA do not work as well in practice as concatenation. We provide theoretical analysis to explain this empirical observation.
Learning specific hands-on skills such as cooking, car maintenance, and home repairs increasingly happens via instructional videos. The user experience with such videos is known to be improved by meta-information such as time-stamped annotations for the main steps involved. Generating such annotations automatically is challenging, and we describe here two relevant contributions. First, we construct and release a new dense video captioning dataset, Video Timeline Tags (ViTT), featuring a variety of instructional videos together with time-stamped annotations. Second, we explore several multimodal sequence-to-sequence pretraining strategies that leverage large unsupervised datasets of videos and caption-like texts. We pretrain and subsequently finetune dense video captioning models using both YouCook2 and ViTT. We show that such models generalize well and are robust over a wide variety of instructional videos.
Systematic Generalization refers to a learning algorithm’s ability to extrapolate learned behavior to unseen situations that are distinct but semantically similar to its training data. As shown in recent work, state-of-the-art deep learning models fail dramatically even on tasks for which they are designed when the test set is systematically different from the training data. We hypothesize that explicitly modeling the relations between objects in their contexts while learning their representations will help achieve systematic generalization. Therefore, we propose a novel method that learns objects’ contextualized embeddings with dynamic message passing conditioned on the input natural language and end-to-end trainable with other downstream deep learning modules. To our knowledge, this model is the first one that significantly outperforms the provided baseline and reaches state-of-the-art performance on grounded SCAN (gSCAN), a grounded natural language navigation dataset designed to require systematic generalization in its test splits.
Many top-performing image captioning models rely solely on object features computed with an object detection model to generate image descriptions. However, recent studies propose to directly use scene graphs to introduce information about object relations into captioning, hoping to better describe interactions between objects. In this work, we thoroughly investigate the use of scene graphs in image captioning. We empirically study whether using additional scene graph encoders can lead to better image descriptions and propose a conditional graph attention network (C-GAT), where the image captioning decoder state is used to condition the graph updates. Finally, we determine to what extent noise in the predicted scene graphs influence caption quality. Overall, we find no significant difference between models that use scene graph features and models that only use object detection features across different captioning metrics, which suggests that existing scene graph generation models are still too noisy to be useful in image captioning. Moreover, although the quality of predicted scene graphs is very low in general, when using high quality scene graphs we obtain gains of up to 3.3 CIDEr compared to a strong Bottom-Up Top-Down baseline.
Our analysis of large summarization datasets indicates that redundancy is a very serious problem when summarizing long documents. Yet, redundancy reduction has not been thoroughly investigated in neural summarization. In this work, we systematically explore and compare different ways to deal with redundancy when summarizing long documents. Specifically, we organize existing methods into categories based on when and how the redundancy is considered. Then, in the context of these categories, we propose three additional methods balancing non-redundancy and importance in a general and flexible way. In a series of experiments, we show that our proposed methods achieve the state-of-the-art with respect to ROUGE scores on two scientific paper datasets, Pubmed and arXiv, while reducing redundancy significantly.
We present an empirical study in favor of a cascade architecture to neural text summarization. Summarization practices vary widely but few other than news summarization can provide a sufficient amount of training data enough to meet the requirement of end-to-end neural abstractive systems which perform content selection and surface realization jointly to generate abstracts. Such systems also pose a challenge to summarization evaluation, as they force content selection to be evaluated along with text generation, yet evaluation of the latter remains an unsolved problem. In this paper, we present empirical results showing that the performance of a cascaded pipeline that separately identifies important content pieces and stitches them together into a coherent text is comparable to or outranks that of end-to-end systems, whereas a pipeline architecture allows for flexible content selection. We finally discuss how we can take advantage of a cascaded pipeline in neural text summarization and shed light on important directions for future research.
Cross-lingual Summarization (CLS) aims at producing a summary in the target language for an article in the source language. Traditional solutions employ a two-step approach, i.e. translate -> summarize or summarize -> translate. Recently, end-to-end models have achieved better results, but these approaches are mostly limited by their dependence on large-scale labeled data. We propose a solution based on mixed-lingual pre-training that leverages both cross-lingual tasks such as translation and monolingual tasks like masked language models. Thus, our model can leverage the massive monolingual data to enhance its modeling of language. Moreover, the architecture has no task-specific components, which saves memory and increases optimization efficiency. We show in experiments that this pre-training scheme can effectively boost the performance of cross-lingual summarization. In NCLS dataset, our model achieves an improvement of 2.82 (English to Chinese) and 1.15 (Chinese to English) ROUGE-1 scores over state-of-the-art results.
Point-of-Interest (POI) oriented question answering (QA) aims to return a list of POIs given a question issued by a user. Recent advances in intelligent virtual assistants have opened the possibility of engaging the client software more actively in the provision of location-based services, thereby showing great promise for automatic POI retrieval. Some existing QA methods can be adopted on this task such as QA similarity calculation and semantic parsing using pre-defined rules. The returned results, however, are subject to inherent limitations due to the lack of the ability for handling some important POI related information, including tags, location entities, and proximity-related terms (e.g. “nearby”, “close”). In this paper, we present a novel deep learning framework integrated with joint inference to capture both tag semantic and geographic correlation between question and POIs. One characteristic of our model is to propose a special cross attention question embedding neural network structure to obtain question-to-POI and POI-to-question information. Besides, we utilize a skewed distribution to simulate the spatial relationship between questions and POIs. By measuring the results offered by the model against existing methods, we demonstrate its robustness and practicability, and supplement our conclusions with empirical evidence.
We show that leveraging metadata information from web pages can improve the performance of models for answer passage selection/reranking. We propose a neural passage selection model that leverages metadata information with a fine-grained encoding strategy, which learns the representation for metadata predicates in a hierarchical way. The models are evaluated on the MS MARCO (Nguyen et al., 2016) and Recipe-MARCO datasets. Results show that our models significantly outperform baseline models, which do not incorporate metadata. We also show that the fine-grained encoding’s advantage over other strategies for encoding the metadata.
Intermediate-task training—fine-tuning a pretrained model on an intermediate task before fine-tuning again on the target task—often improves model performance substantially on language understanding tasks in monolingual English settings. We investigate whether English intermediate-task training is still helpful on non-English target tasks. Using nine intermediate language-understanding tasks, we evaluate intermediate-task transfer in a zero-shot cross-lingual setting on the XTREME benchmark. We see large improvements from intermediate training on the BUCC and Tatoeba sentence retrieval tasks and moderate improvements on question-answering target tasks. MNLI, SQuAD and HellaSwag achieve the best overall results as intermediate tasks, while multi-task intermediate offers small additional improvements. Using our best intermediate-task models for each target task, we obtain a 5.4 point improvement over XLM-R Large on the XTREME benchmark, setting the state of the art as of June 2020. We also investigate continuing multilingual MLM during intermediate-task training and using machine-translated intermediate-task data, but neither consistently outperforms simply performing English intermediate-task training.
Slot-filling, Translation, Intent classification, and Language identification, or STIL, is a newly-proposed task for multilingual Natural Language Understanding (NLU). By performing simultaneous slot filling and translation into a single output language (English in this case), some portion of downstream system components can be monolingual, reducing development and maintenance cost. Results are given using the multilingual BART model (Liu et al., 2020) fine-tuned on 7 languages using the MultiATIS++ dataset. When no translation is performed, mBART’s performance is comparable to the current state of the art system (Cross-Lingual BERT by Xu et al. (2020)) for the languages tested, with better average intent classification accuracy (96.07% versus 95.50%) but worse average slot F1 (89.87% versus 90.81%). When simultaneous translation is performed, average intent classification accuracy degrades by only 1.7% relative and average slot F1 degrades by only 1.2% relative.
We investigate how to adapt simultaneous text translation methods such as wait-k and monotonic multihead attention to end-to-end simultaneous speech translation by introducing a pre-decision module. A detailed analysis is provided on the latency-quality trade-offs of combining fixed and flexible pre-decision with fixed and flexible policies. We also design a novel computation-aware latency metric, adapted from Average Lagging.
Automatically generating stories is a challenging problem that requires producing causally related and logical sequences of events about a topic. Previous approaches in this domain have focused largely on one-shot generation, where a language model outputs a complete story based on limited initial input from a user. Here, we instead focus on the task of interactive story generation, where the user provides the model mid-level sentence abstractions in the form of cue phrases during the generation process. This provides an interface for human users to guide the story generation. We present two content-inducing approaches to effectively incorporate this additional information. Experimental results from both automatic and human evaluations show that these methods produce more topically coherent and personalized stories compared to baseline methods.
In this paper, we introduce a large-scale Indonesian summarization dataset. We harvest articles from Liputan6.com, an online news portal, and obtain 215,827 document–summary pairs. We leverage pre-trained language models to develop benchmark extractive and abstractive summarization methods over the dataset with multilingual and monolingual BERT-based models. We include a thorough error analysis by examining machine-generated summaries that have low ROUGE scores, and expose both issues with ROUGE itself, as well as with extractive and abstractive summarization models.
Sports game summarization focuses on generating news articles from live commentaries. Unlike traditional summarization tasks, the source documents and the target summaries for sports game summarization tasks are written in quite different writing styles. In addition, live commentaries usually contain many named entities, which makes summarizing sports games precisely very challenging. To deeply study this task, we present SportsSum, a Chinese sports game summarization dataset which contains 5,428 soccer games of live commentaries and the corresponding news articles. Additionally, we propose a two-step summarization model consisting of a selector and a rewriter for SportsSum. To evaluate the correctness of generated sports summaries, we design two novel score metrics: name matching score and event matching score. Experimental results show that our model performs better than other summarization baselines on ROUGE scores as well as the two designed scores.
Document alignment aims to identify pairs of documents in two distinct languages that are of comparable content or translations of each other. Such aligned data can be used for a variety of NLP tasks from training cross-lingual representations to mining parallel data for machine translation. In this paper we develop an unsupervised scoring function that leverages cross-lingual sentence embeddings to compute the semantic distance between documents in different languages. These semantic distances are then used to guide a document alignment algorithm to properly pair cross-lingual web documents across a variety of low, mid, and high-resource language pairs. Recognizing that our proposed scoring function and other state of the art methods are computationally intractable for long web documents, we utilize a more tractable greedy algorithm that performs comparably. We experimentally demonstrate that our distance metric performs better alignment than current baselines outperforming them by 7% on high-resource language pairs, 15% on mid-resource language pairs, and 22% on low-resource language pairs.
Topic segmentation is critical in key NLP tasks and recent works favor highly effective neural supervised approaches. However, current neural solutions are arguably limited in how they model context. In this paper, we enhance a segmenter based on a hierarchical attention BiLSTM network to better model context, by adding a coherence-related auxiliary task and restricted self-attention. Our optimized segmenter outperforms SOTA approaches when trained and tested on three datasets. We also the robustness of our proposed model in domain transfer setting by training a model on a large-scale dataset and testing it on four challenging real-world benchmarks. Furthermore, we apply our proposed strategy to two other languages (German and Chinese), and show its effectiveness in multilingual scenarios.
We propose an entity linking (EL) model that jointly learns mention detection (MD) and entity disambiguation (ED). Our model applies task-specific heads on top of shared BERT contextualized embeddings. We achieve state-of-the-art results across a standard EL dataset using our model; we also study our model’s performance under the setting when hand-crafted entity candidate sets are not available and find that the model performs well under such a setting too.
Research in building intelligent agents have emphasized the need for understanding characteristic behavior of people. In order to reflect human-like behavior, agents require the capability to comprehend the context, infer individualized persona patterns and incrementally learn from experience. In this paper, we present a model called DAPPER that can learn to embed persona from natural language and alleviate task or domain-specific data sparsity issues related to personas. To this end, we implement a text encoding strategy that leverages a pretrained language model and an external memory to produce domain-adapted persona representations. Further, we evaluate the transferability of these embeddings by simulating low-resource scenarios. Our comparative study demonstrates the capability of our method over other approaches towards learning rich transferable persona embeddings. Empirical evidence suggests that the learnt persona embeddings can be effective in downstream tasks like hate speech detection.
We present two extensions to a state-of-theart joint model for event coreference resolution, which involve incorporating (1) a supervised topic model for improving trigger detection by providing global context, and (2) a preprocessing module that seeks to improve event coreference by discarding unlikely candidate antecedents of an event mention using discourse contexts computed based on salient entities. The resulting model yields the best results reported to date on the KBP 2017 English and Chinese datasets.
This paper evaluates the utility of Rhetorical Structure Theory (RST) trees and relations in discourse coherence evaluation. We show that incorporating silver-standard RST features can increase accuracy when classifying coherence. We demonstrate this through our tree-recursive neural model, namely RST-Recursive, which takes advantage of the text’s RST features produced by a state of the art RST parser. We evaluate our approach on the Grammarly Corpus for Discourse Coherence (GCDC) and show that when ensembled with the current state of the art, we can achieve the new state of the art accuracy on this benchmark. Furthermore, when deployed alone, RST-Recursive achieves competitive accuracy while having 62% fewer parameters.
Large-scale natural language inference (NLI) datasets such as SNLI or MNLI have been created by asking crowdworkers to read a premise and write three new hypotheses, one for each possible semantic relationships (entailment, contradiction, and neutral). While this protocol has been used to create useful benchmark data, it remains unclear whether the writing-based annotation protocol is optimal for any purpose, since it has not been evaluated directly. Furthermore, there is ample evidence that crowdworker writing can introduce artifacts in the data. We investigate two alternative protocols which automatically create candidate (premise, hypothesis) pairs for annotators to label. Using these protocols and a writing-based baseline, we collect several new English NLI datasets of over 3k examples each, each using a fixed amount of annotator time, but a varying number of examples to fit that time budget. Our experiments on NLI and transfer learning show negative results: None of the alternative protocols outperforms the baseline in evaluations of generalization within NLI or on transfer to outside target tasks. We conclude that crowdworker writing still the best known option for entailment data, highlighting the need for further data collection work to focus on improving writing-based annotation processes.
Span extraction is an essential problem in machine reading comprehension. Most of the existing algorithms predict the start and end positions of an answer span in the given corresponding context by generating two probability vectors. In this paper, we propose a novel approach that extends the probability vector to a probability matrix. Such a matrix can cover more start-end position pairs. Precisely, to each possible start index, the method always generates an end probability vector. Besides, we propose a sampling-based training strategy to address the computational cost and memory issue in the matrix training phase. We evaluate our method on SQuAD 1.1 and three other question answering benchmarks. Leveraging the most competitive models BERT and BiDAF as the backbone, our proposed approach can get consistent improvements in all datasets, demonstrating the effectiveness of the proposed method.
Providing instant response for product-related questions in E-commerce question answering platforms can greatly improve users’ online shopping experience. However, existing product question answering (PQA) methods only consider a single information source such as user reviews and/or require large amounts of labeled data. In this paper, we propose a novel framework to tackle the PQA task via exploiting heterogeneous information including natural language text and attribute-value pairs from two information sources of the concerned product, namely product details and user reviews. A heterogeneous information encoding component is then designed for obtaining unified representations of information with different formats. The sources of the candidate snippets are also incorporated when measuring the question-snippet relevance. Moreover, the framework is trained with a specifically designed weak supervision paradigm making use of available answers in the training phase. Experiments on a real-world dataset show that our proposed framework achieves superior performance over state-of-the-art models.
An NLP model’s ability to reason should be independent of language. Previous works utilize Natural Language Inference (NLI) to understand the reasoning ability of models, mostly focusing on high resource languages like English. To address scarcity of data in low-resource languages such as Hindi, we use data recasting to create NLI datasets for four existing text classification datasets. Through experiments, we show that our recasted dataset is devoid of statistical irregularities and spurious patterns. We further study the consistency in predictions of the textual entailment models and propose a consistency regulariser to remove pairwise-inconsistencies in predictions. We propose a novel two-step classification method which uses textual-entailment predictions for classification task. We further improve the performance by using a joint-objective for classification and textual entailment. We therefore highlight the benefits of data recasting and improvements on classification performance using our approach with supporting experimental results.
Disentangled representations have attracted increasing attention recently. However, how to transfer the desired properties of disentanglement to word representations is unclear. In this work, we propose to transform typical dense word vectors into disentangled embeddings featuring improved interpretability via encoding polysemous semantics separately. We also found the modular structure of our disentangled word embeddings helps generate more efficient and effective features for natural language processing tasks.
Most previous work on knowledge graph completion conducted single-view prediction or calculation for candidate triple evaluation, based only on the content information of the candidate triples. This paper describes a novel multi-view classification model for knowledge graph completion, where multiple classification views are performed based on both content and context information for candidate triple evaluation. Each classification view evaluates the validity of a candidate triple from a specific viewpoint, based on the content information inside the candidate triple and the context information nearby the triple. These classification views are implemented by a unified neural network and the classification predictions are weightedly integrated to obtain the final evaluation. Experiments show that, the multi-view model brings very significant improvements over previous methods, and achieves the new state-of-the-art on two representative datasets. We believe that, the flexibility and the scalability of the multi-view classification model facilitates the introduction of additional information and resources for better performance.
Named entity disambiguation is an important task that plays the role of bridge between text and knowledge. However, the performance of existing methods drops dramatically for short text, which is widely used in actual application scenarios, such as information retrieval and question answering. In this work, we propose a novel knowledge-enhanced method for named entity disambiguation. Considering the problem of information ambiguity and incompleteness for short text, two kinds of knowledge, factual knowledge graph and conceptual knowledge graph, are introduced to provide additional knowledge for the semantic matching between candidate entity and mention context. Our proposed method achieves significant improvement over previous methods on a large manually annotated short-text dataset, and also achieves the state-of-the-art on three standard datasets. The short-text dataset and the proposed model will be publicly available for research use.
Relational facts are an important component of human knowledge, which are hidden in vast amounts of text. In order to extract these facts from text, people have been working on relation extraction (RE) for years. From early pattern matching to current neural networks, existing RE methods have achieved significant progress. Yet with explosion of Web text and emergence of new relations, human knowledge is increasing drastically, and we thus require “more” from RE: a more powerful RE system that can robustly utilize more data, efficiently learn more relations, easily handle more complicated context, and flexibly generalize to more open domains. In this paper, we look back at existing RE methods, analyze key challenges we are facing nowadays, and show promising directions towards more powerful RE. We hope our view can advance this field and inspire more efforts in the community.
It has been shown that word embeddings can exhibit gender bias, and various methods have been proposed to quantify this. However, the extent to which the methods are capturing social stereotypes inherited from the data has been debated. Bias is a complex concept and there exist multiple ways to define it. Previous work has leveraged gender word pairs to measure bias and extract biased analogies. We show that the reliance on these gendered pairs has strong limitations: bias measures based off of them are not robust and cannot identify common types of real-world bias, whilst analogies utilising them are unsuitable indicators of bias. In particular, the well-known analogy “man is to computer-programmer as woman is to homemaker” is due to word similarity rather than bias. This has important implications for work on measuring bias in embeddings and related work debiasing embeddings.
Within the prosperity of Massive Open Online Courses (MOOCs), the education applications that automatically provide extracurricular knowledge for MOOC users become rising research topics. However, MOOC courses’ diversity and rapid updates make it more challenging to find suitable new knowledge for students. In this paper, we present ExpanRL, an end-to-end hierarchical reinforcement learning (HRL) model for concept expansion in MOOCs. Employing a two-level HRL mechanism of seed selection and concept expansion, ExpanRL is more feasible to adjust the expansion strategy to find new concepts based on the students’ feedback on expansion results. Our experiments on nine novel datasets from real MOOCs show that ExpanRL achieves significant improvements over existing methods and maintain competitive performance under different settings.
Question generation (QG) has recently attracted considerable attention. Most of the current neural models take as input only one or two sentences, and perform poorly when multiple sentences or complete paragraphs are given as input. However, in real-world scenarios it is very important to be able to generate high-quality questions from complete paragraphs. In this paper, we present a simple yet effective technique for answer-aware question generation from paragraphs. We augment a basic sequence-to-sequence QG model with dynamic, paragraph-specific dictionary and copy attention that is persistent across the corpus, without requiring features generated by sophisticated NLP pipelines or handcrafted rules. Our evaluation on SQuAD shows that our model significantly outperforms current state-of-the-art systems in question generation from paragraphs in both automatic and human evaluation. We achieve a 6-point improvement over the best system on BLEU-4, from 16.38 to 22.62.
Adversarial attacks are label-preserving modifications to inputs of machine learning classifiers designed to fool machines but not humans. Natural Language Processing (NLP) has mostly focused on high-level attack scenarios such as paraphrasing input texts. We argue that these are less realistic in typical application scenarios such as in social media, and instead focus on low-level attacks on the character-level. Guided by human cognitive abilities and human robustness, we propose the first large-scale catalogue and benchmark of low-level adversarial attacks, which we dub Zéroe, encompassing nine different attack modes including visual and phonetic adversaries. We show that RoBERTa, NLP’s current workhorse, fails on our attacks. Our dataset provides a benchmark for testing robustness of future more human-like NLP models.
Physical places help shape how we perceive the experiences we have there. We study the relationship between social media text and the type of the place from where it was posted, whether a park, restaurant, or someplace else. To facilitate this, we introduce a novel data set of ~200,000 English tweets published from 2,761 different points-of-interest in the U.S., enriched with place type information. We train classifiers to predict the type of the location a tweet was sent from that reach a macro F1 of 43.67 across eight classes and uncover the linguistic markers associated with each type of place. The ability to predict semantic place information from a tweet has applications in recommendation systems, personalization services and cultural geography.
Event information is usually scattered across multiple sentences within a document. The local sentence-level event extractors often yield many noisy event role filler extractions in the absence of a broader view of the document-level context. Filtering spurious extractions and aggregating event information in a document remains a challenging problem. Following the observation that a document has several relevant event regions densely populated with event role fillers, we build graphs with candidate role filler extractions enriched by sentential embeddings as nodes, and use graph attention networks to identify event regions in a document and aggregate event information. We characterize edges between candidate extractions in a graph into rich vector representations to facilitate event region identification. The experimental results on two datasets of two languages show that our approach yields new state-of-the-art performance for the challenging event extraction task.
We propose a newly annotated dataset for information extraction on recipes. Unlike previous approaches to machine comprehension of procedural texts, we avoid a priori pre-defining domain-specific predicates to recognize (e.g., the primitive instructionsin MILK) and focus on basic understanding of the expressed semantics rather than directly reduce them to a simplified state representation (e.g., ProPara). We thus frame the semantic comprehension of procedural text such as recipes, as fairly generic NLP subtasks, covering (i) entity recognition (ingredients, tools and actions), (ii) relation extraction (what ingredients and tools are involved in the actions), and (iii) zero anaphora resolution (link actions to implicit arguments, e.g., results from previous recipe steps). Further, our Recipe Instruction Semantic Corpus (RISeC) dataset includes textual descriptions for the zero anaphora, to facilitate language generation thereof. Besides the dataset itself, we contribute a pipeline neural architecture that addresses entity and relation extractionas well an identification of zero anaphora. These basic building blocks can facilitate more advanced downstream applications (e.g., question answering, conversational agents).
Studies on grammatical error correction (GEC) have reported on the effectiveness of pretraining a Seq2Seq model with a large amount of pseudodata. However, this approach requires time-consuming pretraining of GEC because of the size of the pseudodata. In this study, we explored the utility of bidirectional and auto-regressive transformers (BART) as a generic pretrained encoder-decoder model for GEC. With the use of this generic pretrained model for GEC, the time-consuming pretraining can be eliminated. We find that monolingual and multilingual BART models achieve high performance in GEC, with one of the results being comparable to the current strong results in English GEC.
Mandarin Alphabetical Word (MAW) is one indispensable component of Modern Chinese that demonstrates unique code-mixing idiosyncrasies influenced by language exchanges. Yet, this interesting phenomenon has not been properly addressed and is mostly excluded from the Chinese language system. This paper addresses the core problem of MAW identification and proposes to construct a large collection of MAWs from Sina Weibo (SMAW) using an automatic web-based technique which includes rule-based identification, informatics-based extraction, as well as Baidu search engine validation. A collection of 16,207 qualified SMAWs are obtained using this technique along with an annotated corpus of more than 200,000 sentences for linguistic research and applicable inquiries.
Although Indonesian is known to be the fourth most frequently used language over the internet, the research progress on this language in natural language processing (NLP) is slow-moving due to a lack of available resources. In response, we introduce the first-ever vast resource for training, evaluation, and benchmarking on Indonesian natural language understanding (IndoNLU) tasks. IndoNLU includes twelve tasks, ranging from single sentence classification to pair-sentences sequence labeling with different levels of complexity. The datasets for the tasks lie in different domains and styles to ensure task diversity. We also provide a set of Indonesian pre-trained models (IndoBERT) trained from a large and clean Indonesian dataset (Indo4B) collected from publicly available sources such as social media texts, blogs, news, and websites. We release baseline models for all twelve tasks, as well as the framework for benchmark evaluation, thus enabling everyone to benchmark their system performances.
The gaze behaviour of a reader is helpful in solving several NLP tasks such as automatic essay grading. However, collecting gaze behaviour from readers is costly in terms of time and money. In this paper, we propose a way to improve automatic essay grading using gaze behaviour, which is learnt at run time using a multi-task learning framework. To demonstrate the efficacy of this multi-task learning based approach to automatic essay grading, we collect gaze behaviour for 48 essays across 4 essay sets, and learn gaze behaviour for the rest of the essays, numbering over 7000 essays. Using the learnt gaze behaviour, we can achieve a statistically significant improvement in performance over the state-of-the-art system for the essay sets where we have gaze data. We also achieve a statistically significant improvement for 4 other essay sets, numbering about 6000 essays, where we have no gaze behaviour data available. Our approach establishes that learning gaze behaviour improves automatic essay grading.
We model source-selection in multi-source Unsupervised Domain Adaptation (UDA) as an attention-learning problem, where we learn attention over the sources per given target instance. We first independently learn source-specific classification models, and a relatedness map between sources and target domains using pseudo-labelled target domain instances. Next, we learn domain-attention scores over the sources for aggregating the predictions of the source-specific models. Experimental results on two cross-domain sentiment classification datasets show that the proposed method reports consistently good performance across domains, and at times outperforming more complex prior proposals. Moreover, the computed domain-attention scores enable us to find explanations for the predictions made by the proposed method.
Large pre-trained language models reach state-of-the-art results on many different NLP tasks when fine-tuned individually; They also come with a significant memory and computational requirements, calling for methods to reduce model sizes (green AI). We propose a two-stage model-compression method to reduce a model’s inference time cost. We first decompose the matrices in the model into smaller matrices and then perform feature distillation on the internal representation to recover from the decomposition. This approach has the benefit of reducing the number of parameters while preserving much of the information within the model. We experimented on BERT-base model with the GLUE benchmark dataset and show that we can reduce the number of parameters by a factor of 0.4x, and increase inference speed by a factor of 1.45x, while maintaining a minimal loss in metric performance.
Explainable recommendation is a good way to improve user satisfaction. However, explainable recommendation in dialogue is challenging since it has to handle natural language as both input and output. To tackle the challenge, this paper proposes a novel and practical task to explain evidences in recommending hotels given vague requests expressed freely in natural language. We decompose the process into two subtasks on hotel reviews: Evidence Identification and Evidence Explanation. The former predicts whether or not a sentence contains evidence that expresses why a given request is satisfied. The latter generates a recommendation sentence given a request and an evidence sentence. In order to address these subtasks, we build an Evidence-based Explanation dataset, which is the largest dataset for explaining evidences in recommending hotels for vague requests. The experimental results demonstrate that the BERT model can find evidence sentences with respect to various vague requests and that the LSTM-based model can generate recommendation sentences.
In this paper, we propose an effective deep learning framework for multilingual and code- mixed visual question answering. The pro- posed model is capable of predicting answers from the questions in Hindi, English or Code- mixed (Hinglish: Hindi-English) languages. The majority of the existing techniques on Vi- sual Question Answering (VQA) focus on En- glish questions only. However, many applica- tions such as medical imaging, tourism, visual assistants require a multilinguality-enabled module for their widespread usages. As there is no available dataset in English-Hindi VQA, we firstly create Hindi and Code-mixed VQA datasets by exploiting the linguistic properties of these languages. We propose a robust tech- nique capable of handling the multilingual and code-mixed question to provide the answer against the visual information (image). To better encode the multilingual and code-mixed questions, we introduce a hierarchy of shared layers. We control the behaviour of these shared layers by an attention-based soft layer sharing mechanism, which learns how shared layers are applied in different ways for the dif- ferent languages of the question. Further, our model uses bi-linear attention with a residual connection to fuse the language and image fea- tures. We perform extensive evaluation and ablation studies for English, Hindi and Code- mixed VQA. The evaluation shows that the proposed multilingual model achieves state-of- the-art performance in all these settings.
Hate speech and toxic comments are a common concern of social media platform users. Although these comments are, fortunately, the minority in these platforms, they are still capable of causing harm. Therefore, identifying these comments is an important task for studying and preventing the proliferation of toxicity in social media. Previous work in automatically detecting toxic comments focus mainly in English, with very few work in languages like Brazilian Portuguese. In this paper, we propose a new large-scale dataset for Brazilian Portuguese with tweets annotated as either toxic or non-toxic or in different types of toxicity. We present our dataset collection and annotation process, where we aimed to select candidates covering multiple demographic groups. State-of-the-art BERT models were able to achieve 76% macro-F1 score using monolingual data in the binary case. We also show that large-scale monolingual data is still needed to create more accurate models, despite recent advances in multilingual approaches. An error analysis and experiments with multi-label classification show the difficulty of classifying certain types of toxic comments that appear less frequently in our data and highlights the need to develop models that are aware of different categories of toxicity.
Stance classification can be a powerful tool for understanding whether and which users believe in online rumours. The task aims to automatically predict the stance of replies towards a given rumour, namely support, deny, question, or comment. Numerous methods have been proposed and their performance compared in the RumourEval shared tasks in 2017 and 2019. Results demonstrated that this is a challenging problem since naturally occurring rumour stance data is highly imbalanced. This paper specifically questions the evaluation metrics used in these shared tasks. We re-evaluate the systems submitted to the two RumourEval tasks and show that the two widely adopted metrics – accuracy and macro-F1 – are not robust for the four-class imbalanced task of rumour stance classification, as they wrongly favour systems with highly skewed accuracy towards the majority class. To overcome this problem, we propose new evaluation metrics for rumour stance detection. These are not only robust to imbalanced data but also score higher systems that are capable of recognising the two most informative minority classes (support and deny).
We propose a new character-based text classification framework for non-alphabetic languages, such as Chinese and Japanese. Our framework consists of a variational character encoder (VCE) and character-level text classifier. The VCE is composed of a β-variational auto-encoder (β -VAE) that learns the proposed glyph-aware disentangled character embedding (GDCE). Since our GDCE provides zero-mean unit-variance character embeddings that are dimensionally independent, it is applicable for our interpretable data augmentation, namely, semantic sub-character augmentation (SSA). In this paper, we evaluated our framework using Japanese text classification tasks at the document- and sentence-level. We confirmed that our GDCE and SSA not only provided embedding interpretability but also improved the classification performance. Our proposal achieved a competitive result to the state-of-the-art model while also providing model interpretability.
This paper investigates a new co-attention mechanism in neural transduction models for machine translation tasks. We propose a paradigm, termed Two-Headed Monster (THM), which consists of two symmetric encoder modules and one decoder module connected with co-attention. As a specific and concrete implementation of THM, Crossed Co-Attention Networks (CCNs) are designed based on the Transformer model. We test CCNs on WMT 2014 EN-DE and WMT 2016 EN-FI translation tasks and show both advantages and disadvantages of the proposed method. Our model outperforms the strong Transformer baseline by 0.51 (big) and 0.74 (base) BLEU points on EN-DE and by 0.17 (big) and 0.47 (base) BLEU points on EN-FI but the epoch time increases by circa 75%.
Curriculum learning, a training strategy where training data are ordered based on their difficulty, has been shown to improve performance and reduce training time on various NLP tasks. While much work over the years has developed novel approaches for generating curricula, these strategies are typically only suited for the task they were designed for. This work explores developing a task-agnostic model for problem difficulty and applying it to the Stanford Natural Language Inference (SNLI) dataset. Using the human responses that come with the dev set of SNLI, we train both regression and classification models to predict how many annotators will answer a question correctly and then project the difficulty estimates onto the full SNLI train set to create the curriculum. We argue that our curriculum is effectively capturing difficulty for this task through various analyses of both the model and the predicted difficulty scores.
This paper presents a deep neural architecture which applies the siamese convolutional neural network sharing model parameters for learning a semantic similarity metric between two sentences. In addition, two different similarity metrics (i.e., the Cosine Similarity and Manhattan similarity) are compared based on this architecture. Our experiments in binary similarity classification for Chinese sentence pairs show that the proposed siamese convolutional architecture with Manhattan similarity outperforms the baselines (i.e., the siamese Long Short-Term Memory architecture and the siamese Bidirectional Long Short-Term Memory architecture) by 8.7 points in accuracy.
Obtaining social media demographic information using machine learning is important for efficient computational social science research. Automatic age classification has been accomplished with relative success and allows for the study of youth populations, but student classification—determining which users are currently attending an academic institution—has not been thoroughly studied. Previous work (He et al., 2016) proposes a model which utilizes 3 tweet-content features to classify users as students or non-students. This model achieves an accuracy of 84%, but is restrictive and time intensive because it requires accessing and processing many user tweets. In this study, we propose classification models which use 7 numerical features and 10 text-based features drawn from simple profile information. These profile-based features allow for faster, more accessible data collection and enable the classification of users without needing access to their tweets. Compared to previous models, our models identify students with greater accuracy; our best model obtains an accuracy of 88.1% and an F1 score of .704. This improved student identification tool has the potential to facilitate research on topics ranging from professional networking to the impact of education on Twitter behaviors.
Code-switching is a commonly observed communicative phenomenon denoting a shift from one language to another within the same speech exchange. The analysis of code-switched data often becomes an assiduous task, owing to the limited availability of data. In this work, we propose converting code-switched data into its constituent high resource languages for exploiting both monolingual and cross-lingual settings. This conversion allows us to utilize the higher resource availability for its constituent languages for multiple downstream tasks. We perform experiments for two downstream tasks, sarcasm detection and hate speech detection in the English-Hindi code-switched setting. These experiments show an increase in 22% and 42.5% in F1-score for sarcasm detection and hate speech detection, respectively, compared to the state-of-the-art.
In this work, the task of extractive single document summarization applied to an education setting to generate summaries of chapters from grade 10 Hindi history textbooks is undertaken. Unsupervised approaches to extract summaries are employed and evaluated. TextRank, LexRank, Luhn and KLSum are used to extract summaries. When evaluated intrinsically, Luhn and TextRank summaries have the highest ROUGE scores. When evaluated extrinsically, the effective measure of a summary in answering exam questions, TextRank summaries performs the best.
User-generated contents’ score-based prediction and item recommendation has become an inseparable part of the online recommendation systems. The ratings allow people to express their opinions and may affect the market value of items and consumer confidence in e-commerce decisions. A major problem with the models designed for user review prediction is that they unknowingly neglect the rating bias occurring due to personal user bias preferences. We propose a tendency-based approach that models the user and item tendency for score prediction along with text review analysis with respect to ratings.
This research paper reports on the generation of the first Drenjongke corpus based on texts taken from a phrase book for beginners, written in the Tibetan script. A corpus of sentences was created after correcting errors in the text scanned through optical character reading (OCR). A total of 34 Part-of-Speech (PoS) tags were defined based on manual annotation performed by the three authors, one of whom is a native speaker of Drenjongke. The first corpus of the Drenjongke language comprises 275 sentences and 1379 tokens, which we plan to expand with other materials to promote further studies of this language.
In recent years, named entity recognition (NER) tasks in the Indonesian language have undergone extensive development. There are only a few corpora for Indonesian NER; hence, recent Indonesian NER studies have used diverse datasets. Although an open dataset is available, it includes only approximately 2,000 sentences and contains inconsistent annotations, thereby preventing accurate training of NER models without reliance on pre-trained models. Therefore, we re-annotated the dataset and compared the two annotations’ performance using the Bidirectional Long Short-Term Memory and Conditional Random Field (BiLSTM-CRF) approach. Fixing the annotation yielded a more consistent result for the organization tag and improved the prediction score by a large margin. Moreover, to take full advantage of pre-trained models, we compared different feature embeddings to determine their impact on the NER task for the Indonesian language.
The paper discusses the syntax of the primary statements of the Sanskritam, a programming language specification based on natural Sanskrit under a doctoral thesis. By a statement, we mean a syntactic unit regardless of its computational operations of variable declarations, program executions or evaluations of Boolean expressions etc. We have selected six common primary statements of declaration, assignment, inline initialization, if-then-else, for loop and while loop. The specification partly overlaps the ideas of natural language programming, Controlled Natural Language (Kunh, 2013), and Natural Language subset. The practice and application of structured natural language set in a discourse are deeply rooted in the theoretical text tradition of Sanskrit, like the sūtra-based disciplines and Navya-Nyāya (NN) formal language, etc. The effort is a kind of continuation and application of such traditions and their techniques in the modern field of Sanskrit NLP.
Along with the rise of people generated content on social sites, sentiment analysis has gained more importance. Aspect Based Sentiment Analysis (ABSA) is a task of identifying the sentiment at aspect level. It has more importance than sentiment analysis from commercial point of view. To the best of our knowledge, there is very few work on ABSA in Urdu language. Recent work on ABSA has limitations. Only predefined aspects are identified in a specific domain. So our focus is on the creation and evaluation of dataset for ABSA in Urdu language which will support multiple aspects. This dataset will provide a baseline evaluation for ABSA systems.
Explicit mechanisms for copying have improved the performance of neural models for sequence-to-sequence tasks in the low-resource setting. However, they rely on an overlap between source and target vocabularies. Here, we propose a model that does not: a pointer-generator transformer for disjoint vocabularies. We apply our model to a low-resource version of the grapheme-to-phoneme conversion (G2P) task, and show that it outperforms a standard transformer by an average of 5.1 WER over 15 languages. While our model does not beat the the best performing baseline, we demonstrate that it provides complementary information to it: an oracle that combines the best outputs of the two models improves over the strongest baseline by 7.7 WER on average in the low-resource setting. In the high-resource setting, our model performs comparably to a standard transformer.
Can we trust that the attention heatmaps produced by a neural machine translation (NMT) model reflect its true internal reasoning? We isolate and examine in detail the notion of faithfulness in NMT models. We provide a measure of faithfulness for NMT based on a variety of stress tests where model parameters are perturbed and measuring faithfulness based on how often the model output changes. We show that our proposed faithfulness measure for NMT models can be improved using a novel differentiable objective that rewards faithful behaviour by the model through probability divergence. Our experimental results on multiple language pairs show that our objective function is effective in increasing faithfulness and can lead to a useful analysis of NMT model behaviour and more trustworthy attention heatmaps. Our proposed objective improves faithfulness without reducing the translation quality and it also seems to have a useful regularization effect on the NMT model and can even improve translation quality in some cases.
Large-scale pre-trained representations such as BERT have been widely used in many natural language understanding tasks. The methods of incorporating BERT into document-level machine translation are still being explored. BERT is able to understand sentence relationships since BERT is pre-trained using the next sentence prediction task. In our work, we leverage this property to improve document-level machine translation. In our proposed model, BERT performs as a context encoder to achieve document-level contextual information, which is then integrated into both the encoder and decoder. Experiment results show that our proposed method can significantly outperform strong document-level machine translation baselines on BLEU score. Moreover, the ablation study shows our method can capture document-level context information to boost translation performance.
WikiSQL and Spider, the large-scale cross-domain text-to-SQL datasets, have attracted much attention from the research community. The leaderboards of WikiSQL and Spider show that many researchers propose their models trying to solve the text-to-SQL problem. This paper first divides the top models in these two leaderboards into two paradigms. We then present details not mentioned in their original paper by evaluating the key components, including schema linking, pretrained word embeddings, and reasoning assistance modules. Based on the analysis of these models, we want to promote understanding of the text-to-SQL field and find out some interesting future works, for example, it is worth studying the text-to-SQL problem in an environment where it is more challenging to build schema linking and also worth studying combing the advantage of each model toward text-to-SQL.
Automated Essay Scoring (AES) is a process that aims to alleviate the workload of graders and improve the feedback cycle in educational systems. Multi-task learning models, one of the deep learning techniques that have recently been applied to many NLP tasks, demonstrate the vast potential for AES. In this work, we present an approach for combining two tasks, sentiment analysis, and AES by utilizing multi-task learning. The model is based on a hierarchical neural network that learns to predict a holistic score at the document-level along with sentiment classes at the word-level and sentence-level. The sentiment features extracted from opinion expressions can enhance a vanilla holistic essay scoring, which mainly focuses on lexicon and text semantics. Our approach demonstrates that sentiment features are beneficial for some essay prompts, and the performance is competitive to other deep learning models on the Automated StudentAssessment Prize (ASAP) benchmark. TheQuadratic Weighted Kappa (QWK) is used to measure the agreement between the human grader’s score and the model’s prediction. Ourmodel produces a QWK of 0.763.
Aspect extraction is a widely researched field of natural language processing in which aspects are identified from the text as a means for information. For example, in aspect-based sentiment analysis (ABSA), aspects need to be first identified. Previous studies have introduced various approaches to increasing accuracy, although leaving room for further improvement. In a practical situation where the examined dataset is lacking labels, to fine-tune the process a novel unsupervised approach is proposed, combining a lexical rule-based approach with coreference resolution. The model increases accuracy through the recognition and removal of coreferring aspects. Experimental evaluations are performed on two benchmark datasets, demonstrating the greater performance of our approach to extracting coherent aspects through outperforming the baseline approaches.
In this work, we introduce a GRU-based architecture called GRUBERT that learns to map the different BERT hidden layers to fused embeddings with the aim of achieving high accuracy on the Twitter sentiment analysis task. Tweets are known for their highly diverse language, and by exploiting different linguistic information present across BERT hidden layers, we can capture the full extent of this language at the embedding level. Our method can be easily adapted to other embeddings capturing different linguistic information. We show that our method outperforms well-known heuristics of using BERT (e.g. using only the last layer) and other embeddings such as ELMo. We observe potential label noise resulting from the data acquisition process and employ early stopping as well as a voting classifier to overcome it.
Computational approaches to noun ellipsis resolution has been sparse, with only a naive rule-based approach that uses syntactic feature constraints for marking noun ellipsis licensors and selecting their antecedents. In this paper, we further the ellipsis research by exploring several statistical and neural models for both the subtasks involved in the ellipsis resolution process and addressing the representation and contribution of manual features proposed in previous research. Using the best performing models, we build an end-to-end supervised Machine Learning (ML) framework for this task that improves the existing F1 score by 16.55% for the detection and 14.97% for the resolution subtask. Our experiments demonstrate robust scores through pretrained BERT (Bidirectional Encoder Representations from Transformers) embeddings for word representation, and more so the importance of manual features– once again highlighting the syntactic and semantic characteristics of the ellipsis phenomenon. For the classification decision, we notice that a simple Multilayar Perceptron (MLP) works well for the detection of ellipsis; however, Recurrent Neural Networks (RNN) are a better choice for the much harder resolution step.
Models developed for Machine Reading Comprehension (MRC) are asked to predict an answer from a question and its related context. However, there exist cases that can be correctly answered by an MRC model using BERT, where only the context is provided without including the question. In this paper, these types of examples are referred to as “easy to answer”, while others are as “hard to answer”, i.e., unanswerable by an MRC model using BERT without being provided the question. Based on classifying examples as answerable or unanswerable by BERT without the given question, we propose a method based on BERT that splits the training examples from the MRC dataset SQuAD1.1 into those that are “easy to answer” or “hard to answer”. Experimental evaluation from a comparison of two models, one trained only with “easy to answer” examples and the other with “hard to answer” examples demonstrates that the latter outperforms the former.
We optimize rewards of reinforcement learning in text simplification using metrics that are highly correlated with human-perspectives. To address problems of exposure bias and loss-evaluation mismatch, text-to-text generation tasks employ reinforcement learning that rewards task-specific metrics. Previous studies in text simplification employ the weighted sum of sub-rewards from three perspectives: grammaticality, meaning preservation, and simplicity. However, the previous rewards do not align with human-perspectives for these perspectives. In this study, we propose to use BERT regressors fine-tuned for grammaticality, meaning preservation, and simplicity as reward estimators to achieve text simplification conforming to human-perspectives. Experimental results show that reinforcement learning with our rewards balances meaning preservation and simplicity. Additionally, human evaluation confirmed that simplified texts by our method are preferred by humans compared to previous studies.
Several recent state-of-the-art transfer learning methods model classification tasks as text generation, where labels are represented as strings for the model to generate. We investigate the effect that the choice of strings used to represent labels has on how effectively the model learns the task. For four standard text classification tasks, we design a diverse set of possible string representations for labels, ranging from canonical label definitions to random strings. We experiment with T5 on these tasks, varying the label representations as well as the amount of training data. We find that, in the low data setting, label representation impacts task performance on some tasks, with task-related labels being most effective, but fails to have an impact on others. In the full data setting, our results are largely negative: Different label representations do not affect overall task performance.
Automated grammatical error correction has been explored as an important research problem within NLP, with the majority of the work being done on English and similar resource-rich languages. Grammar correction using neural networks is a data-heavy task, with the recent state of the art models requiring datasets with millions of annotated sentences for proper training. It is difficult to find such resources for Indic languages due to their relative lack of digitized content and complex morphology, compared to English. We address this problem by generating a large corpus of artificial inflectional errors for training GEC models. Moreover, to evaluate the performance of models trained on this dataset, we create a corpus of real Hindi errors extracted from Wikipedia edits. Analyzing this dataset with a modified version of the ERRANT error annotation toolkit, we find that inflectional errors are very common in this language. Finally, we produce the initial baseline results using state of the art methods developed for English.
This article presents the AMesure platform, which aims to assist writers of French administrative texts in simplifying their writing. This platform includes a readability formula specialized for administrative texts and it also uses various natural language processing (NLP) tools to analyze texts and highlight a number of linguistic phenomena considered difficult to read. Finally, based on the difficulties identified, it offers pieces of advice coming from official plain language guides to users. This paper describes the different components of the system and reports an evaluation of these components.
With the renaissance of deep learning, neural networks have achieved promising results on many natural language understanding (NLU) tasks. Even though the source codes of many neural network models are publicly available, there is still a large gap from open-sourced models to solving real-world problems in enterprises. Therefore, to fill this gap, we introduce AutoNLU, an on-demand cloud-based system with an easy-to-use interface that covers all common use-cases and steps in developing an NLU model. AutoNLU has supported many product teams within Adobe with different use-cases and datasets, quickly delivering them working models. To demonstrate the effectiveness of AutoNLU, we present two case studies. i) We build a practical NLU model for handling various image-editing requests in Photoshop. ii) We build powerful keyphrase extraction models that achieve state-of-the-art results on two public benchmarks. In both cases, end users only need to write a small amount of code to convert their datasets into a common format used by AutoNLU.
Despite the growth of e-commerce, brick-and-mortar stores are still the preferred destinations for many people. In this paper, we present ISA, a mobile-based intelligent shopping assistant that is designed to improve shopping experience in physical stores. ISA assists users by leveraging advanced techniques in computer vision, speech processing, and natural language processing. An in-store user only needs to take a picture or scan the barcode of the product of interest, and then the user can talk to the assistant about the product. The assistant can also guide the user through the purchase process or recommend other similar products to the user. We take a data-driven approach in building the engines of ISA’s natural language processing component, and the engines achieve good performance.
Creating high-quality annotated dialogue corpora is challenging. It is essential to develop practical annotation tools to support humans in this time-consuming and error-prone task. We present metaCAT, which is an open-source web-based annotation tool designed specifically for developing task-oriented dialogue data. To the best of our knowledge, metaCAT is the first annotation tool that provides comprehensive metadata annotation coverage to the domain, intent, and span information. The data annotation quality is enhanced by a real-time annotation constraint-checking mechanism. An Automatic Speech Recognition (ASR) function is implemented to allow users to paraphrase and create more diversified annotated utterances. metaCAT is publicly available for the community.
Processing maintenance logbook records is an important step in the development of predictive maintenance systems. Logbooks often include free text fields with domain specific terms, abbreviations, and non-standard spelling posing challenges to off-the-shelf NLP pipelines trained on standard contemporary corpora. Despite the importance of this data type, processing predictive maintenance data is still an under-explored topic in NLP. With the goal of providing more datasets and resources to the community, in this paper we present a number of new resources available in MaintNet, a collaborative open-source library and data repository of predictive maintenance language datasets. We describe novel annotated datasets from multiple domains such as aviation, automotive, and facility maintenance domains and new tools for segmentation, spell checking, POS tagging, clustering, and classification.
We introduce fairseq S2T, a fairseq extension for speech-to-text (S2T) modeling tasks such as end-to-end speech recognition and speech-to-text translation. It follows fairseq’s careful design for scalability and extensibility. We provide end-to-end workflows from data pre-processing, model training to offline (online) inference. We implement state-of-the-art RNN-based as well as Transformer-based models and open-source detailed training recipes. Fairseq’s machine translation models and language models can be seamlessly integrated into S2T workflows for multi-task learning or transfer learning. Fairseq S2T is available at https://github.com/pytorch/fairseq/tree/master/examples/speech_to_text.
Statistical significance testing centered on p-values is commonly used to compare NLP system performance, but p-values alone are insufficient because statistical significance differs from practical significance. The latter can be measured by estimating effect size. In this pa-per, we propose a three-stage procedure for comparing NLP system performance and provide a toolkit, NLPStatTest, that automates the process. Users can upload NLP system evaluation scores and the toolkit will analyze these scores, run appropriate significance tests, estimate effect size, and conduct power analysis to estimate Type II error. The toolkit provides a convenient and systematic way to compare NLP system performance that goes beyond statistical significance testing.
Discourse parsing aims to comprehensively acquire the logical structure of the whole text which may be helpful to some downstream applications such as summarization, reading comprehension, QA and so on. One important issue behind discourse parsing is the representation of discourse structure. Up to now, many discourse structures have been proposed, and the correponding parsing methods are designed, promoting the development of discourse research. In this paper, we mainly introduce our recent discourse research and its preliminary application from the dependency view.
Machine translation (MT) models usually translate a text at sentence level by considering isolated sentences, which is based on a strict assumption that the sentences in a text are independent of one another. However, the fact is that the texts at discourse level have properties going beyond individual sentences. These properties reveal texts in the frequency and distribution of words, word senses, referential forms and syntactic structures. Dissregarding dependencies across sentences will harm translation quality especially in terms of coherence, cohesion, and consistency. To solve these problems, several approaches have previously been investigated for conventional statistical machine translation (SMT). With the fast growth of neural machine translation (NMT), discourse-level NMT has drawn increasing attention from researchers. In this work, we review major works on addressing discourse related problems for both SMT and NMT models with a survey of recent trends in the fields.
The need to evaluate the ability of context-aware neural machine translation (NMT) models in dealing with specific discourse phenomena arises in document-level NMT. However, test sets that satisfy this need are rare. In this paper, we propose a test suite to evaluate three common discourse phenomena in English-Chinese translation: pronoun, discourse connective and ellipsis where discourse divergences lie across the two languages. The test suite contains 1,200 instances, 400 for each type of discourse phenomena. We perform both automatic and human evaluation with three state-of-the-art context-aware NMT models on the proposed test suite. Results suggest that our test suite can be used as a challenging benchmark test bed for evaluating document-level NMT. The test suite will be publicly available soon.
In recent years, attention mechanism has been widely used in various neural machine translation tasks based on encoder decoder. This paper focuses on the performance of encoder decoder attention mechanism in word sense disambiguation task with different text length, trying to find out the influence of context marker on attention mechanism in word sense disambiguation task. We hypothesize that attention mechanisms have similar performance when translating texts of different lengths. Our conclusion is that the alignment effect of attention mechanism is magnified in short text translation tasks with ambiguous nouns, while the effect of attention mechanism is far less than expected in long-text tasks, which means that attention mechanism is not the main mechanism for NMT model to feed WSD to integrate context information. This may mean that attention mechanism pays more attention to ambiguous nouns than context markers. The experimental results show that with the increase of text length, the performance of NMT model using attention mechanism will gradually decline.
Previous neural approaches achieve significant progress for Chinese word segmentation (CWS) as a sentence-level task, but it suffers from limitations on real-world scenario. In this paper, we address this issue with a context-aware method and optimize the solution at document-level. This paper proposes a three-step strategy to improve the performance for discourse CWS. First, the method utilizes an auxiliary segmenter to remedy the limitation on pre-segmenter. Then the context-aware algorithm computes the confidence of each split. The maximum probability path is reconstructed via this algorithm. Besides, in order to evaluate the performance in discourse, we build a new benchmark consisting of the latest news and Chinese medical articles. Extensive experiments on this benchmark show that our proposed method achieves a competitive performance on a document-level real-world scenario for CWS.
With regards to WikiSum (CITATION) that empowers applicative explorations of Neural Multi-Document Summarization (MDS) to learn from large scale dataset, this study develops two hierarchical Transformers (HT) that describe both the cross-token and cross-document dependencies, at the same time allow extended length of input documents. By incorporating word- and paragraph-level multi-head attentions in the decoder based on the parallel and vertical architectures, the proposed parallel and vertical hierarchical Transformers (PHT &VHT) generate summaries utilizing context-aware word embeddings together with static and dynamics paragraph embeddings, respectively. A comprehensive evaluation is conducted on WikiSum to compare PHT &VHT with established models and to answer the question whether hierarchical structures offer more promising performances than flat structures in the MDS task. The results suggest that our hierarchical models generate summaries of higher quality by better capturing cross-document relationships, and save more memory spaces in comparison to flat-structure models. Moreover, we recommend PHT given its practical value of higher inference speed and greater memory-saving capacity.
In this paper, we explore a new approach based on discourse analysis for the task of intent segmentation. Our target texts are user queries from a real-world chatbot. Our results show the feasibility of our approach with an F1-score of 82.97 points, and some advantages and disadvantages compared to two machine learning baselines: BERT and LSTM+CRF.
In human question-answering (QA), questions are often expressed in the form of multiple sentences. One can see this in both spoken QA interactions, when one person asks a question of another, and written QA, such as are found on-line in FAQs and in what are called ”Community Question-Answering Forums”. Computer-based QA has taken the challenge of these ”multi-sentence questions” to be that of breaking them into an appropriately ordered sequence of separate questions, with both the previous questions and their answers serving as context for the next question. This can be seen, for example, in two recent workshops at AAAI called ”Reasoning for Complex QA” [https://rcqa-ws.github.io/program/]. We claim that, while appropriate for some types of ”multi-sentence questions” (MSQs), it is not appropriate for all, because they are essentially different types of discourse. To support this claim, we need to provide evidence that: • different types of MSQs are answered differently in written or spoken QA between people; • people can (and do) distinguish these different types of MSQs; • systems can be made to both distinguish different types of MSQs and provide appropriate answers.
NT Clause Complex Framework defines a clause complex as a combination of NT clauses through component sharing and logic-semantic relationship. This paper clarifies the existence of component sharing mechanism in both English and Chinese clause complexes, illustrates the differences in component sharing between the two languages, and introduces a formal annotation scheme to represent clause-complex level structural transformations. Under the guidance of the annotation scheme, the English-Chinese Clause Alignment Corpus is built. It is believed that this corpus will aid comparative linguistic studies, translation studies and machine translation studies by providing abundant formal and computable samples for English-Chinese structural transformations on the clause complex level.
Connected texts are characterised by the presence of linguistic elements relating to shared referents throughout the text. These elements together form a structure that lends cohesion to the text. The realisation of those cohesive structures is subject to different constraints and varying preferences in different languages. We regularly observe mismatches of cohesive structures across languages in parallel texts. This can be a result of either a divergence of language-internal constraints or of effects of the translation process. As fully automatic high-quality MT is starting to look achievable, the question arises how cohesive elements should be handled in MT evaluation, since the common assumption of 1:1 correspondence between referring expressions is a poor match for what we find in corpus data. Focusing on the translation of pronouns, I discuss different approaches to evaluating a particular type of cohesive elements in MT output and the trade-offs they make between evaluation cost, validity, specificity and coverage. I suggest that a meaningful evaluation of cohesive structures in translation is difficult to achieve simply by appealing to the intuition of human annotators, but requires a more structured approach that forces us to make up our minds about the standards we expect the translation output to adhere to.
The coronavirus disease (COVID-19) has claimed the lives of over one million people and infected more than thirty-five million people worldwide. Several search engines have surfaced to provide researchers with additional tools to find and retrieve information from the rapidly growing corpora on COVID19. These engines lack extraction and visualization tools necessary to retrieve and interpret complex relations inherent to scientific literature. Moreover, because these engines mainly rely upon semantic information, their ability to capture complex global relationships across documents is limited, which reduces the quality of similarity-based article recommendations for users. In this work, we present the COVID-19 Knowledge Graph (CKG), a heterogeneous graph for extracting and visualizing complex relationships between COVID-19 scientific articles. The CKG combines semantic information with document topological information for the application of similar document retrieval. The CKG is constructed using the latent schema of the data, and then enriched with biomedical entity information extracted from the unstructured text of articles using scalable AWS technologies to form relations in the graph. Finally, we propose a document similarity engine that leverages low-dimensional graph embeddings from the CKG with semantic embeddings for similar article retrieval. Analysis demonstrates the quality of relationships in the CKG and shows that it can be used to uncover meaningful information in COVID-19 scientific articles. The CKG helps power www.cord19.aws and is publicly available.
The Differentiable Neural Computer (DNC), a neural network model with an addressable external memory, can solve algorithmic and question answering tasks. There are various improved versions of DNC, such as rsDNC and DNC-DMS. However, how to integrate structured knowledge into these DNC models remains a challenging research question. We incorporate an architecture for knowledge into such DNC models, i.e. DNC, rsDNC and DNC-DMS, to improve the ability to generate correct responses using both contextual information and structured knowledge. Our improved rsDNC model improves the mean accuracy by approximately 20% to the original rsDNC on tasks requiring knowledge in the dialog bAbI tasks. In addition, our improved rsDNC and DNC-DMS models also yield better performance than their original models in the Movie Dialog dataset.
Pattisapu et al. (2020) formulate medical concept normalization (MCN) as text similarity problem and propose a model based on RoBERTa and graph embedding based target concept vectors. However, graph embedding techniques ignore valuable information available in the clinical ontology like concept description and synonyms. In this work, we enhance the model of Pattisapu et al. (2020) with two novel changes. First, we use retrofitted target concept vectors instead of graph embedding based vectors. It is the first work to leverage both concept description and synonyms to represent concepts in the form of retrofitted target concept vectors in text similarity framework based social media MCN. Second, we generate both concept and concept mention vectors with same size which eliminates the need of dense layers to project concept mention vectors into the target concept embedding space. Our model outperforms existing methods with improvements up to 3.75% on two standard datasets. Further when trained only on mapping lexicon synonyms, our model outperforms existing methods with significant improvements up to 14.61%. We attribute these significant improvements to the two novel changes introduced.
Traditional biomedical version of embeddings obtained from pre-trained language models have recently shown state-of-the-art results for relation extraction (RE) tasks in the medical domain. In this paper, we explore how to incorporate domain knowledge, available in the form of molecular structure of drugs, for predicting Drug-Drug Interaction from textual corpus. We propose a method, BERTChem-DDI, to efficiently combine drug embeddings obtained from the rich chemical structure of drugs (encoded in SMILES) along with off-the-shelf domain-specific BioBERT embedding-based RE architecture. Experiments conducted on the DDIExtraction 2013 corpus clearly indicate that this strategy improves other strong baselines architectures by 3.4% macro F1-score.
Recently, several studies have investigated active learning (AL) for natural language processing tasks to alleviate data dependency. However, for query selection, most of these studies mainly rely on uncertainty-based sampling, which generally does not exploit the structural information of the unlabeled data. This leads to a sampling bias in the batch active learning setting, which selects several samples at once. In this work, we demonstrate that the amount of labeled training data can be reduced using active learning when it incorporates both uncertainty and diversity in the sequence labeling task. We examined the effects of our sequence-based approach by selecting weighted diverse in the gradient embedding approach across multiple tasks, datasets, models, and consistently outperform classic uncertainty-based sampling and diversity-based sampling.
When training speech recognition systems, one often faces the situation that sufficient amounts of training data for the language in question are available but only small amounts of data for the domain in question. This problem is even bigger for end-to-end speech recognition systems that only accept transcribed speech as training data, which is harder and more expensive to obtain than text data. In this paper we present experiments in adapting end-to-end speech recognition systems by a method which is called batch-weighting and which we contrast against regular fine-tuning, i.e., to continue to train existing neural speech recognition models on adaptation data. We perform experiments using theses techniques in adapting to topic, accent and vocabulary, showing that batch-weighting consistently outperforms fine-tuning. In order to show the generalization capabilities of batch-weighting we perform experiments in several languages, i.e., Arabic, English and German. Due to its relatively small computational requirements batch-weighting is a suitable technique for supervised life-long learning during the life-time of a speech recognition system, e.g., from user corrections.
Language model based pre-trained models such as BERT have provided significant gains across different NLP tasks. In this paper, we study different types of transformer based pre-trained models such as auto-regressive models (GPT-2), auto-encoder models (BERT), and seq2seq models (BART) for conditional data augmentation. We show that prepending the class labels to text sequences provides a simple yet effective way to condition the pre-trained models for data augmentation. Additionally, on three classification benchmarks, pre-trained Seq2Seq model outperforms other data augmentation methods in a low-resource setting. Further, we explore how different pre-trained model based data augmentation differs in-terms of data diversity, and how well such methods preserve the class-label information.
Languages can be considered endangered for many reasons. One of the principal reasons for endangerment is the disappearance of its speakers. Another, more identifiable reason, is the lack of written resources. We present an automated sub-segmentation system called AshMorph that deals with the morphology of an Amazonian tribal language called Ashaninka which is at risk of being endangered due to the lack of availability (or resistance) of native speakers and the absence of written resources. We show that by the use of a cross-lingual lexicon and finite state transducers we can increase accuracy by more than 30% when compared to other modern sub-segmentation tools. Our results, made freely available on-line, are verified by an Ashaninka speaker and perform well in two distinct domains, everyday literary articles and the bible. This research serves as a first step in helping to preserve Ashaninka by offering a sub-segmentation process that can be used to normalize any Ashaninka text which will serve as input to a machine translation system for translation into other high-resource languages spoken by higher populated locations like Spanish and Portuguese in the case of Peru and Brazil where Ashaninka is mostly spoken.
The Philippines is home to more than 150 languages that is considered to be low-resourced even on its major languages. This results into a lack of pursuit in developing a translation system for the underrepresented languages. To simplify the process of developing translation system for multiple languages, and to aid in improving the translation quality of zero to low-resource languages, multilingual NMT became an active area of research. However, existing works in multilingual NMT disregards the analysis of a multilingual model on a closely related and low-resource language group in the context of pivot-based translation and zero-shot translation. In this paper, we benchmarked translation for several Philippine Languages, provided an analysis of a multilingual NMT system for morphologically rich and low-resource languages in terms of its effectiveness in translating zero-resource languages with zero-shot translations. To further evaluate the capability of the multilingual NMT model in translating unseen language pairs in training, we tested the model to translate between Tagalog and Cebuano and compared its performance with a simple NMT model that is directly trained on a parallel Tagalog and Cebuano data in which we showed that zero-shot translation outperforms a directly trained model in some instances, while utilizing English as a pivot language in translating outperform both approaches.
Low-resource languages present unique challenges to (neural) machine translation. We discuss the case of Bambara, a Mande language for which training data is scarce and requires significant amounts of pre-processing. More than the linguistic situation of Bambara itself, the socio-cultural context within which Bambara speakers live poses challenges for automated processing of this language. In this paper, we present the first parallel data set for machine translation of Bambara into and from English and French and the first benchmark results on machine translation to and from Bambara. We discuss challenges in working with low-resource languages and propose strategies to cope with data scarcity in low-resource machine translation (MT).
This paper presents the findings of the LoResMT 2020 Shared Task on zero-shot translation for low resource languages. This task was organised as part of the 3rd Workshop on Technologies for MT of Low Resource Languages (LoResMT) at AACL-IJCNLP 2020. The focus was on the zero-shot approach as a notable development in Neural Machine Translation to build MT systems for language pairs where parallel corpora are small or even non-existent. The shared task experience suggests that back-translation and domain adaptation methods result in better accuracy for small-size datasets. We further noted that, although translation between similar languages is no cakewalk, linguistically distinct languages require more data to give better results.
Neural machine translation (NMT) is a widely accepted approach in the machine translation (MT) community, translating from one natural language to another natural language. Although, NMT shows remarkable performance in both high and low resource languages, it needs sufficient training corpus. The availability of a parallel corpus in low resource language pairs is one of the challenging tasks in MT. To mitigate this issue, NMT attempts to utilize a monolingual corpus to get better at translation for low resource language pairs. Workshop on Technologies for MT of Low Resource Languages (LoResMT 2020) organized shared tasks of low resource language pair translation using zero-shot NMT. Here, the parallel corpus is not used and only monolingual corpora is allowed. We have participated in the same shared task with our team name CNLP-NITS for the Russian-Hindi language pair. We have used masked sequence to sequence pre-training for language generation (MASS) with only monolingual corpus following the unsupervised NMT architecture. The evaluated results are declared at the LoResMT 2020 shared task, which reports that our system achieves the bilingual evaluation understudy (BLEU) score of 0.59, precision score of 3.43, recall score of 5.48, F-measure score of 4.22, and rank-based intuitive bilingual evaluation score (RIBES) of 0.180147 in Russian to Hindi translation. And for Hindi to Russian translation, we have achieved BLEU, precision, recall, F-measure, and RIBES score of 1.11, 4.72, 4.41, 4.56, and 0.026842 respectively.
This paper reports a Machine Translation (MT) system submitted by the NLPRL team for the Bhojpuri–Hindi and Magahi–Hindi language pairs at LoResMT 2020 shared task. We used an unsupervised domain adaptation approach that gives promising results for zero or extremely low resource languages. Task organizers provide the development and the test sets for evaluation and the monolingual data for training. Our approach is a hybrid approach of domain adaptation and back-translation. Metrics used to evaluate the trained model are BLEU, RIBES, Precision, Recall and F-measure. Our approach gives relatively promising results, with a wide range, of 19.5, 13.71, 2.54, and 3.16 BLEU points for Bhojpuri to Hindi, Magahi to Hindi, Hindi to Bhojpuri and Hindi to Magahi language pairs, respectively.
Standard neural machine translation (NMT) allows a model to perform translation between a pair of languages. Multilingual neural machine translation (NMT), on the other hand, allows a model to perform translation between several language pairs, even between language pairs for which no sentences pair has been seen during training (zero-shot translation). This paper presents experiments with zero-shot translation on low resource Indian languages with a very small amount of data for each language pair. We first report results on balanced data over all considered language pairs. We then expand our experiments for additional three rounds by increasing the training data with 2,000 sentence pairs in each round for some of the language pairs. We obtain an increase in translation accuracy with its balanced data settings score multiplied by 7 for Manipuri to Hindi during Round-III of zero-shot translation.
Prior works have demonstrated that a low-resource language pair can benefit from multilingual machine translation (MT) systems, which rely on many language pairs’ joint training. This paper proposes two simple strategies to address the rare word issue in multilingual MT systems for two low-resource language pairs: French-Vietnamese and English-Vietnamese. The first strategy is about dynamical learning word similarity of tokens in the shared space among source languages while another one attempts to augment the translation ability of rare words through updating their embeddings during the training. Besides, we leverage monolingual data for multilingual MT systems to increase the amount of synthetic parallel corpora while dealing with the data sparsity problem. We have shown significant improvements of up to +1.62 and +2.54 BLEU points over the bilingual baseline systems for both language pairs and released our datasets for the research community.
The corpus preparation is one of the important challenging task for the domain of machine translation especially in low resource language scenarios. Country like India where multiple languages exists, machine translation attempts to minimize the communication gap among people with different linguistic backgrounds. Although Google Translation covers automatic translation of various languages all over the world but it lags in some languages including Assamese. In this paper, we have developed EnAsCorp1.0, corpus of English-Assamese low resource pair where parallel and monolingual data are collected from various online sources. We have also implemented baseline systems with statistical machine translation and neural machine translation approaches for the same corpus.
Availability of bitext dataset has been a key challenge in the conventional machine translation system which requires surplus amount of parallel data. In this work, we devise an unsupervised neural machine translation (UNMT) system consisting of a transformer based shared encoder and language specific decoders using denoising autoencoder and backtranslation with an additional Manipuri side multiple test reference. We report our work on low resource setting for English (en) - Manipuri (mni) language pair and attain a BLEU score of 3.1 for en-mni and 2.7 for mni-en respectively. Subjective evaluation on translated output gives encouraging findings.
In this paper, we evaluate LSTM, biLSTM, GRU, and Transformer architectures for the task of name transliteration in a many-to-one multilingual paradigm, transliterating from 590 languages to English. We experiment with different encoder-decoder combinations and evaluate them using accuracy, character error rate, and an F-measure based on longest continuous subsequences. We find that using a Transformer for the encoder and decoder performs best, improving accuracy by over 4 points compared to previous work. We explore whether manipulating the source text by adding macrolanguage flag tokens or pre-romanizing source strings can improve performance and find that neither manipulation has a positive effect. Finally, we analyze performance differences between the LSTM and Transformer encoders when using a Transformer decoder and find that the Transformer encoder is better able to handle insertions and substitutions when transliterating.
Machine translation is the task of translating texts from one language to another using computers. It has been one of the major tasks in natural language processing and computational linguistics and has been motivating to facilitate human communication. Kurdish, an Indo-European language, has received little attention in this realm due to the language being less-resourced. Therefore, in this paper, we are addressing the main issues in creating a machine translation system for the Kurdish language, with a focus on the Sorani dialect. We describe the available scarce parallel data suitable for training a neural machine translation model for Sorani Kurdish-English translation. We also discuss some of the major challenges in Kurdish language translation and demonstrate how fundamental text processing tasks, such as tokenization, can improve translation performance.
In this paper we present a new ensemble method, Continuous Bag-of-Skip-grams (CBOS), that produces high-quality word representations putting emphasis on the Greek language. The CBOS method combines the pioneering approaches for learning word representations: Continuous Bag-of-Words (CBOW) and Continuous Skip-gram. These methods are compared through intrinsic and extrinsic evaluation tasks on three different sources of data: the English Wikipedia corpus, the Greek Wikipedia corpus, and the Greek Web Content corpus. By comparing these methods across different tasks and datasets, it is evident that the CBOS method achieves state-of-the-art performance.
Building Machine Translation (MT) systems for low-resource languages remains challenging. For many language pairs, parallel data are not widely available, and in such cases MT models do not achieve results comparable to those seen with high-resource languages. When data are scarce, it is of paramount importance to make optimal use of the limited material available. To that end, in this paper we propose employing the same parallel sentences multiple times, only changing the way the words are split each time. For this purpose we use several Byte Pair Encoding models, with various merge operations used in their configuration. In our experiments, we use this technique to expand the available data and improve an MT system involving a low-resource language pair, namely English-Esperanto. As an additional contribution, we made available a set of English-Esperanto parallel data in the literary domain.
Statistical machine translation (SMT) which was the dominant paradigm in machine translation (MT) research for nearly three decades has recently been superseded by the end-to-end deep learning approaches to MT. Although deep neural models produce state-of-the-art results in many translation tasks, they are found to under-perform on resource-poor scenarios. Despite some success, none of the present-day benchmarks that have tried to overcome this problem can be regarded as a universal solution to the problem of translation of many low-resource languages. In this work, we investigate the performance of phrase-based SMT (PB-SMT) and neural MT (NMT) on a rarely-tested low-resource language-pair, English-to-Tamil, taking a specialised data domain (software localisation) into consideration. In particular, we produce rankings of our MT systems via a social media platform-based human evaluation scheme, and demonstrate our findings in the low-resource domain-specific text translation task.
There are several problems in applying grammatical error correction (GEC) to a writing support system. One of them is the handling of sentences in the middle of the input. Till date, the performance of GEC for incomplete sentences is not well-known. Hence, we analyze the performance of each model for incomplete sentences. Another problem is the correction speed. When the speed is slow, the usability of the system is limited, and the user experience is degraded. Therefore, in this study, we also focus on the non-autoregressive (NAR) model, which is a widely studied fast decoding method. We perform GEC in Japanese with traditional autoregressive and recent NAR models and analyze their accuracy and speed.
Traditional statistical approaches to spelling correction usually consist of two consecutive processes — error detection and correction — and they are generally computationally intensive. Current state-of-the-art neural spelling correction models usually attempt to correct spelling errors directly over an entire sentence, which, as a consequence, lacks control of the process, e.g. they are prone to overcorrection. In recent years, recurrent neural networks (RNNs), in particular long short-term memory (LSTM) hidden units, have proven increasingly popular and powerful models for many natural language processing (NLP) problems. Accordingly, we made use of a bidirectional LSTM language model (LM) for our context-sensitive spelling detection and correction model which is shown to have much control over the correction process. While the use of LMs for spelling checking and correction is not new to this line of NLP research, our proposed approach makes better use of the rich neighbouring context, not only from before the word to be corrected, but also after it, via a dual-input deep LSTM network. Although in theory our proposed approach can be applied to any language, we carried out our experiments on Arabic, which we believe adds additional value given the fact that there are limited linguistic resources readily available in Arabic in comparison to many languages. Our experimental results demonstrate that the proposed methods are effective in both improving the quality of correction suggestions and minimising overcorrection.
Developing a text readability assessment model specifically for texts in a foreign English Language Training (ELT) curriculum has never had much attention in the field of Natural Language Processing. Hence, most developed models show extremely low accuracy for L2 English texts, up to the point where not many even serve as a fair comparison. In this paper, we investigate a text readability assessment model for L2 English learners in Korea. In accordance, we improve and expand the Text Corpus of the Korean ELT curriculum (CoKEC-text). Each text is labeled with its target grade level. We train our model with CoKEC-text and significantly improve the accuracy of readability assessment for texts in the Korean ELT curriculum.
This paper presents the NLPTEA 2020 shared task for Chinese Grammatical Error Diagnosis (CGED) which seeks to identify grammatical error types, their range of occurrence and recommended corrections within sentences written by learners of Chinese as a foreign language. We describe the task definition, data preparation, performance metrics, and evaluation results. Of the 30 teams registered for this shared task, 17 teams developed the system and submitted a total of 43 runs. System performances achieved a significant progress, reaching F1 of 91% in detection level, 40% in position level and 28% in correction level. All data sets with gold standards and scoring scripts are made publicly available to researchers.
Grammatical error diagnosis is an important task in natural language processing. This paper introduces our system at NLPTEA-2020 Task: Chinese Grammatical Error Diagnosis (CGED). CGED aims to diagnose four types of grammatical errors which are missing words (M), redundant words (R), bad word selection (S) and disordered words (W). Our system is built on the model of multi-layer bidirectional transformer encoder and ResNet is integrated into the encoder to improve the performance. We also explore two ensemble strategies including weighted averaging and stepwise ensemble selection from libraries of models to improve the performance of single model. In official evaluation, our system obtains the highest F1 scores at identification level and position level. We also recommend error corrections for specific error types and achieve the second highest F1 score at correction level.
This paper describes our participating system on the Chinese Grammatical Error Diagnosis (CGED) 2020 shared task. For the detection subtask, we propose two BERT-based approaches 1) with syntactic dependency trees enhancing the model performance and 2) under the multi-task learning framework to combine the sequence labeling and the sequence-to-sequence (seq2seq) models. For the correction subtask, we utilize the masked language model, the seq2seq model and the spelling check model to generate corrections based on the detection results. Finally, our system achieves the highest recall rate on the top-3 correction and the second best F1 score on identification level and position level.
This paper describes our proposed model for the Chinese Grammatical Error Diagnosis (CGED) task in NLPTEA2020. The goal of CGED is to use natural language processing techniques to automatically diagnose Chinese grammatical errors in sentences. To this end, we design and implement a CGED model named BERT with Score-feature Gates Error Diagnoser (BSGED), which is based on the BERT model, Bidirectional Long Short-Term Memory (BiLSTM) and conditional random field (CRF). In order to address the problem of losing partial-order relationships when embedding continuous feature items as with previous works, we propose a gating mechanism for integrating continuous feature items, which effectively retains the partial-order relationships between feature items. We perform LSTM processing on the encoding result of the BERT model, and further extract the sequence features. In the final test-set evaluation, we obtained the highest F1 score at the detection level and are among the top 3 F1 scores at the identification level.
This paper presents the UNIPUS-Flaubert team’s hybrid system for the NLPTEA 2020 shared task of Chinese Grammatical Error Diagnosis (CGED). As a challenging NLP task, CGED has attracted increasing attention recently and has not yet fully benefited from the powerful pre-trained BERT-based models. We explore this by experimenting with three types of models. The position-tagging models and correction-tagging models are sequence tagging models fine-tuned on pre-trained BERT-based models, where the former focuses on detecting, positioning and classifying errors, and the latter aims at correcting errors. We also utilize rich representations from BERT-based models by transferring the BERT-fused models to the correction task, and further improve the performance by pre-training on a vast size of unsupervised synthetic data. To the best of our knowledge, we are the first to introduce and transfer the BERT-fused NMT model and sequence tagging model into the Chinese Grammatical Error Correction field. Our work achieved the second highest F1 score at the detecting errors, the best F1 score at correction top1 subtask and the second highest F1 score at correction top3 subtask.
This paper introduces our system at NLPTEA2020 shared task for CGED, which is able to detect, locate, identify and correct grammatical errors in Chinese writings. The system consists of three components: GED, GEC, and post processing. GED is an ensemble of multiple BERT-based sequence labeling models for handling GED tasks. GEC performs error correction. We exploit a collection of heterogenous models, including Seq2Seq, GECToR and a candidate generation module to obtain correction candidates. Finally in the post processing stage, results from GED and GEC are fused to form the final outputs. We tune our models to lean towards optimizing precision, which we believe is more crucial in practice. As a result, among the six tracks in the shared task, our system performs well in the correction tracks: measured in F1 score, we rank first, with the highest precision, in the TOP3 correction track and third in the TOP1 correction track, also with the highest precision. Ours are among the top 4 to 6 in other tracks, except for FPR where we rank 12. And our system achieves the highest precisions among the top 10 submissions at IDENTIFICATION and POSITION tracks.
A better Chinese Grammatical Error Diagnosis (CGED) system for automatic Grammatical Error Correction (GEC) can benefit foreign Chinese learners and lower Chinese learning barriers. In this paper, we introduce our solution to the CGED2020 Shared Task Grammatical Error Correction in detail. The task aims to detect and correct grammatical errors that occur in essays written by foreign Chinese learners. Our solution combined data augmentation methods, spelling check methods, and generative grammatical correction methods, and achieved the best recall score in the Top 1 Correction track. Our final result ranked fourth among the participants.
In this paper, we introduce our system for NLPTEA 2020 shared task of Chinese Grammatical Error Diagnosis (CGED). In recent years, pre-trained models have been extensively studied, and several downstream tasks have benefited from their utilization. In this study, we treat the grammar error diagnosis (GED) task as a grammatical error correction (GEC) problem and propose a method that incorporates a pre-trained model into an encoder-decoder model to solve this problem.
This paper reports our Chinese Grammatical Error Diagnosis system in the NLPTEA-2020 CGED shared task. In 2020, we sent two runs with two approaches. The first one is a combination of conditional random fields (CRF) and a BERT model deep-learning approach. The second one is a BERT model deep-learning approach. The official results shows that our run1 achieved the highest precision rate 0.9875 with the lowest false positive rate 0.0163 on detection, while run2 gives a more balanced performance.
Chinese Grammatical Error Diagnosis (CGED) is a natural language processing task for the NLPTEA6 workshop. The goal of this task is to automatically diagnose grammatical errors in Chinese sentences written by L2 learners. This paper proposes a RoBERTa-BiLSTM-CRF model to detect grammatical errors in sentences. Firstly, RoBERTa model is used to obtain word vectors. Secondly, word vectors are input into BiLSTM layer to learn context features. Last, CRF layer without hand-craft features work for processing the output by BiLSTM. The optimal global sequences are obtained according to state transition matrix of CRF and adjacent labels of training data. In experiments, the result of RoBERTa-CRF model and ERNIE-BiLSTM-CRF model are compared, and the impacts of parameters of the models and the testing datasets are analyzed. In terms of evaluation results, our recall score of RoBERTa-BiLSTM-CRF ranks fourth at the detection level.
In the process of learning Chinese, second language learners may have various grammatical errors due to the negative transfer of native language. This paper describes our submission to the NLPTEA 2020 shared task on CGED. We present a hybrid system that utilizes both detection and correction stages. The detection stage is a sequential labelling model based on BiLSTM-CRF and BERT contextual word representation. The correction stage is a hybrid model based on the n-gram and Seq2Seq. Without adding additional features and external data, the BERT contextual word representation can effectively improve the performance metrics of Chinese grammatical error detection and correction.
Automatic grammatical error correction is of great value in assisting second language writing. In 2020, the shared task for Chinese grammatical error diagnosis(CGED) was held in NLP-TEA. As the LDU team, we participated the competition and submitted the final results. Our work mainly focused on grammatical error detection, that is, to judge whether a sentence contains grammatical errors. We used the BERT pre-trained model for binary classification, and we achieve 0.0391 in FPR track, ranking the second in all teams. In error detection track, the accuracy, recall and F-1 of our submitted result are 0.9851, 0.7496 and 0.8514 respectively.
Due to the general availability, relative abundance and wide diversity of opinions, news Media texts are very good sources for sentiment analysis. However, the major challenge with such texts is the difficulty in aligning the expressed opinions to the concerned political leaders as this entails a non-trivial task of named-entity recognition and anaphora resolution. In this work, our primary focus is on developing a Natural Language Processing (NLP) pipeline involving a robust Named-Entity Recognition followed by Anaphora Resolution and then after alignment of the recognized and resolved named-entities, in this case, political leaders to the correct class of opinions as expressed in the texts. We visualize the popularity of the politicians via the time series graph of positive and negative sentiments as an outcome of the pipeline. We have achieved the performance metrics of the individual components of the pipeline as follows: Part of speech tagging – 93.06% (F1-score), Named-Entity Recognition – 86% (F1-score), Anaphora Resolution – 87.45% (Accuracy), Sentiment Analysis – 80.2% (F1-score).
Text simplification is an important branch of natural language processing. At present, methods used to evaluate the semantic retention of text simplification are mostly based on string matching. We propose the SEMA (text Simplification Evaluation Measure through Semantic Alignment), which is based on semantic alignment. Semantic alignments include complete alignment, partial alignment and hyponymy alignment. Our experiments show that the evaluation results of SEMA have a high consistency with human evaluation for the simplified corpus of Chinese and English news texts.
Language and music are closely related. Regarding the linguistic feature richness, pop songs are probably suitable to be used as extracurricular materials in language teaching. In order to prove this point, this paper presents the Contemporary Chinese Pop Lyrics (CCPL) corpus. Based on that, we investigated and evaluated the appropriateness of pop songs for Teaching Chinese as a Second Language (TCSL) with the assistance of Natural Language Processing methods from the perspective of Chinese character coverage, lexical coverage and the addressed topic similarity. Some suggestions in Chinese teaching with the aid of pop lyrics are provided.
This paper presents the results of the shared tasks from the 7th workshop on Asian translation (WAT2020). For the WAT2020, 20 teams participated in the shared tasks and 14 teams submitted their translation results for the human evaluation. We also received 12 research paper submissions out of which 7 were accepted. About 500 translation results were submitted to the automatic evaluation server, and selected submissions were manually evaluated.
In this paper, we propose a useful optimization method for low-resource Neural Machine Translation (NMT) by investigating the effectiveness of multiple neural network optimization algorithms. Our results confirm that applying the proposed optimization method on English-Persian translation can exceed translation quality compared to the English-Persian Statistical Machine Translation (SMT) paradigm.
This paper presents a simple method that extends a standard Transformer-based autoregressive decoder, to speed up decoding. The proposed method generates a token from the head and tail of a sentence (two tokens in total) in each step. By simultaneously generating multiple tokens that rarely depend on each other, the decoding speed is increased while the degradation in translation quality is minimized. In our experiments, the proposed method increased the translation speed by around 113%-155% in comparison with a standard autoregressive decoder, while degrading the BLEU scores by no more than 1.03. It was faster than an iterative non-autoregressive decoder in many conditions.
New things are being created and new words are constantly being added to languages worldwide. However, it is not practical to translate them all manually into a new foreign language. When translating from an alphabetic language such as English to Chinese, appropriate Chinese characters must be assigned, which is particularly costly compared to other language pairs. Therefore, we propose a task of generating and evaluating new translations from English to Chinese focusing on named entities. We defined three criteria for human evaluation—fluency, adequacy of pronunciation, and adequacy of meaning—and constructed evaluation data based on these definitions. In addition, we built a baseline system and analyzed the output of the system.
This paper describes the Japanese-Chinese Neural Machine Translation (NMT) system submitted by the joint team of Kyoto University and East China Normal University (Kyoto-U+ECNU) to WAT 2020 (Nakazawa et al.,2020). We participate in APSEC Japanese-Chinese translation task. We revisit several techniques for NMT including various architectures, different data selection and augmentation methods, denoising pre-training, and also some specific tricks for Japanese-Chinese translation. We eventually perform a meta ensemble to combine all of the models into a single model. BLEU results of this meta ensembled model rank the first both on 2 directions of ASPEC Japanese-Chinese translation.
This paper describes the system of the NHK-NES team for the WAT 2020 Japanese–English newswire task. There are two main problems in Japanese-English news translation: translation of dropped subjects and compatibility between equivalent translations and English news-style outputs. We address these problems by extracting subjects from the context based on predicate-argument structures and using them as additional inputs, and constructing parallel Japanese-English news sentences equivalently translated from English news sentences. The evaluation results confirm the effectiveness of our context-utilization method.
We introduce our TMU system submitted to the Japanese<->English Multimodal Task (constrained) for WAT 2020 (Nakazawa et al., 2020). This task aims to improve translation performance with the help of another modality (images) associated with the input sentences. In a multimodal translation task, the dataset is, by its nature, a low-resource one. Our method used herein augments the data by generating noisy translations and adding noise to existing training images. Subsequently, we pretrain a translation model on the augmented noisy data, and then fine-tune it on the clean data. We also examine the probabilistic dropping of either the textual or visual context vector in the decoder. This aims to regularize the network to make use of both features while training. The experimental results indicate that translation performance can be improved using our method of textual data augmentation with noising on the target side and probabilistic dropping of either context vector.
This paper describes our work in the WAT 2020 Indic Multilingual Translation Task. We participated in all 7 language pairs (En<->Bn/Hi/Gu/Ml/Mr/Ta/Te) in both directions under the constrained condition—using only the officially provided data. Using transformer as a baseline, our Multi->En and En->Multi translation systems achieve the best performances. Detailed data filtering and data domain selection are the keys to performance enhancement in our experiment, with an average improvement of 2.6 BLEU scores for each language pair in the En->Multi system and an average improvement of 4.6 BLEU scores regarding the Multi->En. In addition, we employed language independent adapter to further improve the system performances. Our submission obtains competitive results in the final evaluation.
In this paper we describe our team‘s (NICT-5) Neural Machine Translation (NMT) models whose translations were submitted to shared tasks of the 7th Workshop on Asian Translation. We participated in the Indic language multilingual sub-task as well as the NICT-SAP multilingual multi-domain sub-task. We focused on naive many-to-many NMT models which gave reasonable translation quality despite their simplicity. Our observations are twofold: (a.) Many-to-many models suffer from a lack of consistency where the translation quality for some language pairs is very good but for some others it is terrible when compared against one-to-many and many-to-one baselines. (b.) Oversampling smaller corpora does not necessarily give the best translation quality for the language pair associated with that pair.
This paper describes the ODIANLP submission to WAT 2020. We have participated in the English-Hindi Multimodal task and Indic task. We have used the state-of-the-art Transformer model for the translation task and InceptionResNetV2 for the Hindi Image Captioning task. Our submission tops in English->Hindi Multimodal task in its track and Odia<->English translation tasks. Also, our submissions performed well in the Indic Multilingual tasks.
Machine translation (MT) focuses on the automatic translation of text from one natural language to another natural language. Neural machine translation (NMT) achieves state-of-the-art results in the task of machine translation because of utilizing advanced deep learning techniques and handles issues like long-term dependency, and context-analysis. Nevertheless, NMT still suffers low translation quality for low resource languages. To encounter this challenge, the multi-modal concept comes in. The multi-modal concept combines textual and visual features to improve the translation quality of low resource languages. Moreover, the utilization of monolingual data in the pre-training step can improve the performance of the system for low resource language translations. Workshop on Asian Translation 2020 (WAT2020) organized a translation task for multimodal translation in English to Hindi. We have participated in the same in two-track submission, namely text-only and multi-modal translation with team name CNLP-NITS. The evaluated results are declared at the WAT2020 translation task, which reports that our multi-modal NMT system attained higher scores than our text-only NMT on both challenge and evaluation test set. For the challenge test data, our multi-modal neural machine translation system achieves Bilingual Evaluation Understudy (BLEU) score of 33.57, Rank-based Intuitive Bilingual Evaluation Score (RIBES) 0.754141, Adequacy-Fluency Metrics (AMFM) score 0.787320 and for evaluation test data, BLEU, RIBES, and, AMFM score of 40.51, 0.803208, and 0.820980 for English to Hindi translation respectively.
This paper describes the ADAPT Centre sub-missions to WAT 2020 for the English-to-Odia translation task. We present the approaches that we followed to try to build competitive machine translation (MT) systems for English-to-Odia. Our approaches include monolingual data selection for creating synthetic data and identifying optimal sets of hyperparameters for the Transformer in a low-resource scenario. Our best MT system produces 4.96BLEU points on the evaluation test set in the English-to-Odia translation task.
In this manuscript, we (team name is NLPRL) describe systems description that was submitted to the translation shared tasks at WAT 2020. We describe our model as transformer based NMT by using byte-level based BPE (BBPE). We used the OdiEnCorp 2.0 parallel corpus provided by the shared task organizer where the training, validation, and test data contain 69370, 13544, and 14344 lines of parallel sentences, respectively. The evaluation results show the BLEU score of English-to-Oria below the Organizer (1.34) and Oria-to-English direction shows above the Organizer (11.33).
In this paper we present an English–Hindi and Hindi–English neural machine translation (NMT) system, submitted to the Translation shared Task organized at WAT 2020. We trained a multilingual NMT system based on transformer architecture. In this paper we show: (i) how effective pre-processing helps to improve performance, (ii) how synthetic data through back-translation from available monolingual data can help in overall translation performance, (iii) how language similarity can aid more onto it. Our submissions ranked 1st in both English to Hindi and Hindi to English translation achieving BLEU 20.80 and 29.59 respectively.
In this paper, we describe our TMU neural machine translation (NMT) system submitted for the Patent task (Korean→Japanese) of the 7th Workshop on Asian Translation (WAT 2020, Nakazawa et al., 2020). We propose a novel method to train a Korean-to-Japanese translation model. Specifically, we focus on the vocabulary overlap of Korean Hanja words and Japanese Kanji words, and propose strategies to leverage Hanja information. Our experiment shows that Hanja information is effective within a specific domain, leading to an improvement in the BLEU scores by +1.09 points compared to the baseline.
This paper introduces our neural machine translation systems’ participation in the WAT 2020 (team ID: goku20). We participated in the (i) Patent, (ii) Business Scene Dialogue (BSD) document-level translation, (iii) Mixed-domain tasks. Regardless of simplicity, standard Transformer models have been proven to be very effective in many machine translation systems. Recently, some advanced pre-training generative models have been proposed on the basis of encoder-decoder framework. Our main focus of this work is to explore how robust Transformer models perform in translation from sentence-level to document-level, from resource-rich to low-resource languages. Additionally, we also investigated the improvement that fine-tuning on the top of pre-trained transformer-based models can achieve on various tasks.
In this paper we describe the ADAPT Centre’s submissions to the WAT 2020 document-level Business Scene Dialogue (BSD) Translation task. We only consider translating from Japanese to English for this task and we use the MarianNMT toolkit to train Transformer models. In order to improve the translation quality, we made use of both in-domain and out-of-domain data for training our Machine Translation (MT) systems, as well as various data augmentation techniques for fine-tuning the model parameters. This paper outlines the experiments we ran to train our systems and report the accuracy achieved through these various experiments.
The paper describes the development process of the The University of Tokyo’s NMT systems that were submitted to the WAT 2020 Document-level Business Scene Dialogue Translation sub-task. We describe the data processing workflow, NMT system training architectures, and automatic evaluation results. For the WAT 2020 shared task, we submitted 12 systems (both constrained and unconstrained) for English-Japanese and Japanese-English translation directions. The submitted systems were trained using Transformer models and one was a SMT baseline.
Document-Level Machine Translation (MT) has become an active research area among the NLP community in recent years. Unlike sentence-level MT, which translates the sentences independently, document-level MT aims to utilize contextual information while translating a given source sentence. This paper demonstrates our submission (Team ID - DEEPNLP) to the Document-Level Translation task organized by WAT 2020. This task focuses on translating texts from a business dialog corpus while optionally utilizing the context present in the dialog. In our proposed approach, we utilize publicly available parallel corpus from different domains to train an open domain base NMT model. We then use monolingual target data to create filtered pseudo parallel data and employ Back-Translation to fine-tune the base model. This is further followed by fine-tuning on the domain-specific corpus. We also ensemble various models to improvise the translation performance. Our best models achieve a BLEU score of 26.59 and 22.83 in an unconstrained setting and 15.10 and 10.91 in the constrained settings for En->Ja & Ja->En direction, respectively.
This paper accompanies the software documentation data set for machine translation, a parallel evaluation data set of data originating from the SAP Help Portal, that we released to the machine translation community for research purposes. It offers the possibility to tune and evaluate machine translation systems in the domain of corporate software documentation and contributes to the availability of a wider range of evaluation scenarios. The data set comprises of the language pairs English to Hindi, Indonesian, Malay and Thai, and thus also increases the test coverage for the many low-resource language pairs. Unlike most evaluation data sets that consist of plain parallel text, the segments in this data set come with additional metadata that describes structural information of the document context. We provide insights into the origin and creation, the particularities and characteristics of the data set as well as machine translation results.
Neural Machine Translation (NMT) on logographic source languages struggles when translating ‘unseen’ characters, which never appear in the training data. One possible approach to this problem uses sub-character decomposition for training and test sentences. However, this approach involves complete retraining, and its effectiveness for unseen character translation to non-logographic languages has not been fully explored. We investigate existing ideograph-based sub-character decomposition approaches for Chinese-to-English and Japanese-to-English NMT, for both high-resource and low-resource domains. For each language pair and domain we construct a test set where all source sentences contain at least one unseen logographic character. We find that complete sub-character decomposition often harms unseen character translation, and gives inconsistent results generally. We offer a simple alternative based on decomposition before inference for unseen characters only. Our approach allows flexible application, achieving translation adequacy improvements and requiring no additional models or training.
Statistical machine translation (SMT) was the state-of-the-art in machine translation (MT) research for more than two decades, but has since been superseded by neural MT (NMT). Despite producing state-of-the-art results in many translation tasks, neural models underperform in resource-poor scenarios. Despite some success, none of the present-day benchmarks that have tried to overcome this problem can be regarded as a universal solution to the problem of translation of many low-resource languages. In this work, we investigate the performance of phrase-based SMT (PB-SMT) and NMT on two rarely-tested low-resource language-pairs, English-to-Tamil and Hindi-to-Tamil, taking a specialised data domain (software localisation) into consideration. This paper demonstrates our findings including the identification of several issues of the current neural approaches to low-resource domain-specific text translation.