Search

As large Pre-trained Language Models (PLMs) trained on large amounts of data in an unsupervised manner become more ubiquitous, identifying various types of bias in the text has come into sharp focus. Existing "Stereotype Detection" datasets mainly adopt a diagnostic approach toward large PLMs. Blodgett et. al (2021a) show that there are significant reliability issues with the existing benchmark datasets. Annotating a reliable dataset requires a precise understanding of the subtle nuances of how stereotypes manifest in text. In this paper, we annotate a focused evaluation set for "Stereotype Detection" that addresses those pitfalls by de-constructing various ways in which stereotypes manifest in text. Further, we present a multi-task model that leverages the abundance of data-rich neighboring tasks such as hate speech detection, offensive language detection, misogyny detection, etc., to improve the empirical performance on "Stereotype Detection". We then propose a reinforcement-learning agent that guides the multi-task learning model by learning to identify the training examples from the neighboring tasks that help the target task the most. We show that the proposed models achieve significant empirical gains over existing baselines on all the tasks., Long paper at ACL 2022 main conference

A long-term goal of AI research is to build intelligent agents that can communicate with humans in natural language, perceive the environment, and perform real-world tasks. Vision-and-Language Navigation (VLN) is a fundamental and interdisciplinary research topic towards this goal, and receives increasing attention from natural language processing, computer vision, robotics, and machine learning communities. In this paper, we review contemporary studies in the emerging field of VLN, covering tasks, evaluation metrics, methods, etc. Through structured analysis of current progress and challenges, we highlight the limitations of current VLN and opportunities for future work. This paper serves as a thorough reference for the VLN research community., Comment: 19 pages. Accepted to ACL 2022

This paper presents a production Semi-Supervised Learning (SSL) pipeline based on the student-teacher framework, which leverages millions of unlabeled examples to improve Natural Language Understanding (NLU) tasks. We investigate two questions related to the use of unlabeled data in production SSL context: 1) how to select samples from a huge unlabeled data pool that are beneficial for SSL training, and 2) how do the selected data affect the performance of different state-of-the-art SSL techniques. We compare four widely used SSL techniques, Pseudo-Label (PL), Knowledge Distillation (KD), Virtual Adversarial Training (VAT) and Cross-View Training (CVT) in conjunction with two data selection methods including committee-based selection and submodular optimization based selection. We further examine the benefits and drawbacks of these techniques when applied to intent classification (IC) and named entity recognition (NER) tasks, and provide guidelines specifying when each of these methods might be beneficial to improve large scale NLU systems., NAACL 2021 Industry track

Out-of-scope intent detection is of practical importance in task-oriented dialogue systems. Since the distribution of outlier utterances is arbitrary and unknown in the training stage, existing methods commonly rely on strong assumptions on data distribution such as mixture of Gaussians to make inference, resulting in either complex multi-step training procedures or hand-crafted rules such as confidence threshold selection for outlier detection. In this paper, we propose a simple yet effective method to train an out-of-scope intent classifier in a fully end-to-end manner by simulating the test scenario in training, which requires no assumption on data distribution and no additional post-processing or threshold setting. Specifically, we construct a set of pseudo outliers in the training stage, by generating synthetic outliers using inliner features via self-supervision and sampling out-of-scope sentences from easily available open-domain datasets. The pseudo outliers are used to train a discriminative classifier that can be directly applied to and generalize well on the test task. We evaluate our method extensively on four benchmark dialogue datasets and observe significant improvements over state-of-the-art approaches. Our code has been released at https://github.com/liam0949/DCLOOS., Published as long oral paper in ACL 2021

In this paper we implement and compare 7 different data augmentation strategies for the task of automatic scoring of children's ability to understand others' thoughts, feelings, and desires (or "mindreading"). We recruit in-domain experts to re-annotate augmented samples and determine to what extent each strategy preserves the original rating. We also carry out multiple experiments to measure how much each augmentation strategy improves the performance of automatic scoring systems. To determine the capabilities of automatic systems to generalize to unseen data, we create UK-MIND-20 - a new corpus of children's performance on tests of mindreading, consisting of 10,320 question-answer pairs. We obtain a new state-of-the-art performance on the MIND-CA corpus, improving macro-F1-score by 6 points. Results indicate that both the number of training examples and the quality of the augmentation strategies affect the performance of the systems. The task-specific augmentations generally outperform task-agnostic augmentations. Automatic augmentations based on vectors (GloVe, FastText) perform the worst. We find that systems trained on MIND-CA generalize well to UK-MIND-20. We demonstrate that data augmentation strategies also improve the performance on unseen data., The paper will be presented at ACL-IJCNLP 2021

Distributional semantics has had enormous empirical success in Computational Linguistics and Cognitive Science in modeling various semantic phenomena, such as semantic similarity, and distributional models are widely used in state-of-the-art Natural Language Processing systems. However, the theoretical status of distributional semantics within a broader theory of language and cognition is still unclear: What does distributional semantics model? Can it be, on its own, a fully adequate model of the meanings of linguistic expressions? The standard answer is that distributional semantics is not fully adequate in this regard, because it falls short on some of the central aspects of formal semantic approaches: truth conditions, entailment, reference, and certain aspects of compositionality. We argue that this standard answer rests on a misconception: These aspects do not belong in a theory of expression meaning, they are instead aspects of speaker meaning, i.e., communicative intentions in a particular context. In a slogan: words do not refer, speakers do. Clearing this up enables us to argue that distributional semantics on its own is an adequate model of expression meaning. Our proposal sheds light on the role of distributional semantics in a broader theory of language and cognition, its relationship to formal semantics, and its place in computational models., Comment: To appear in Proceedings of the 13th International Conference on Computational Semantics (IWCS 2019), Gothenburg, Sweden

Training the deep neural networks that dominate NLP requires large datasets. These are often collected automatically or via crowdsourcing, and may exhibit systematic biases or annotation artifacts. By the latter we mean spurious correlations between inputs and outputs that do not represent a generally held causal relationship between features and classes; models that exploit such correlations may appear to perform a given task well, but fail on out of sample data. In this paper we evaluate use of different attribution methods for aiding identification of training data artifacts. We propose new hybrid approaches that combine saliency maps (which highlight important input features) with instance attribution methods (which retrieve training samples influential to a given prediction). We show that this proposed training-feature attribution can be used to efficiently uncover artifacts in training data when a challenging validation set is available. We also carry out a small user study to evaluate whether these methods are useful to NLP researchers in practice, with promising results. We make code for all methods and experiments in this paper available., Comment: ACL Findings 2022

This paper investigates the effectiveness of sentence-level transformers for zero-shot offensive span identification on a code-mixed Tamil dataset. More specifically, we evaluate rationale extraction methods of Local Interpretable Model Agnostic Explanations (LIME) \cite{DBLP:conf/kdd/Ribeiro0G16} and Integrated Gradients (IG) \cite{DBLP:conf/icml/SundararajanTY17} for adapting transformer based offensive language classification models for zero-shot offensive span identification. To this end, we find that LIME and IG show baseline $F_{1}$ of 26.35\% and 44.83\%, respectively. Besides, we study the effect of data set size and training process on the overall accuracy of span identification. As a result, we find both LIME and IG to show significant improvement with Masked Data Augmentation and Multilabel Training, with $F_{1}$ of 50.23\% and 47.38\% respectively. \textit{Disclaimer : This paper contains examples that may be considered profane, vulgar, or offensive. The examples do not represent the views of the authors or their employers/graduate schools towards any person(s), group(s), practice(s), or entity/entities. Instead they are used to emphasize only the linguistic research challenges.}, Comment: Submission to https://dravidianlangtech.github.io/2022/

A well-calibrated neural model produces confidence (probability outputs) closely approximated by the expected accuracy. While prior studies have shown that mixup training as a data augmentation technique can improve model calibration on image classification tasks, little is known about using mixup for model calibration on natural language understanding (NLU) tasks. In this paper, we explore mixup for model calibration on several NLU tasks and propose a novel mixup strategy for pre-trained language models that improves model calibration further. Our proposed mixup is guided by both the Area Under the Margin (AUM) statistic (Pleiss et al., 2020) and the saliency map of each sample (Simonyan et al.,2013). Moreover, we combine our mixup strategy with model miscalibration correction techniques (i.e., label smoothing and temperature scaling) and provide detailed analyses of their impact on our proposed mixup. We focus on systematically designing experiments on three NLU tasks: natural language inference, paraphrase detection, and commonsense reasoning. Our method achieves the lowest expected calibration error compared to strong baselines on both in-domain and out-of-domain test samples while maintaining competitive accuracy., Comment: Accepted at ACL 2022 main conference

Tables store rich numerical data, but numerical reasoning over tables is still a challenge. In this paper, we find that the spreadsheet formula, which performs calculations on numerical values in tables, is naturally a strong supervision of numerical reasoning. More importantly, large amounts of spreadsheets with expert-made formulae are available on the web and can be obtained easily. FORTAP is the first method for numerical-reasoning-aware table pretraining by leveraging large corpus of spreadsheet formulae. We design two formula pretraining tasks to explicitly guide FORTAP to learn numerical reference and calculation in semi-structured tables. FORTAP achieves state-of-the-art results on two representative downstream tasks, cell type classification and formula prediction, showing great potential of numerical-reasoning-aware pretraining., Comment: Accepted by ACL'22 main track

Question answering over temporal knowledge graphs (KGs) efficiently uses facts contained in a temporal KG, which records entity relations and when they occur in time, to answer natural language questions (e.g., "Who was the president of the US before Obama?"). These questions often involve three time-related challenges that previous work fail to adequately address: 1) questions often do not specify exact timestamps of interest (e.g., "Obama" instead of 2000); 2) subtle lexical differences in time relations (e.g., "before" vs "after"); 3) off-the-shelf temporal KG embeddings that previous work builds on ignore the temporal order of timestamps, which is crucial for answering temporal-order related questions. In this paper, we propose a time-sensitive question answering (TSQA) framework to tackle these problems. TSQA features a timestamp estimation module to infer the unwritten timestamp from the question. We also employ a time-sensitive KG encoder to inject ordering information into the temporal KG embeddings that TSQA is based on. With the help of techniques to reduce the search space for potential answers, TSQA significantly outperforms the previous state of the art on a new benchmark for question answering over temporal KGs, especially achieving a 32% (absolute) error reduction on complex questions that require multiple steps of reasoning over facts in the temporal KG., Comment: 10 pages, 2 figures

Neural named entity recognition (NER) models may easily encounter the over-confidence issue, which degrades the performance and calibration. Inspired by label smoothing and driven by the ambiguity of boundary annotation in NER engineering, we propose boundary smoothing as a regularization technique for span-based neural NER models. It re-assigns entity probabilities from annotated spans to the surrounding ones. Built on a simple but strong baseline, our model achieves results better than or competitive with previous state-of-the-art systems on eight well-known NER benchmarks. Further empirical analysis suggests that boundary smoothing effectively mitigates over-confidence, improves model calibration, and brings flatter neural minima and more smoothed loss landscapes., Paper accepted to ACL 2022 main conference

Existing continual relation learning (CRL) methods rely on plenty of labeled training data for learning a new task, which can be hard to acquire in real scenario as getting large and representative labeled data is often expensive and time-consuming. It is therefore necessary for the model to learn novel relational patterns with very few labeled data while avoiding catastrophic forgetting of previous task knowledge. In this paper, we formulate this challenging yet practical problem as continual few-shot relation learning (CFRL). Based on the finding that learning for new emerging few-shot tasks often results in feature distributions that are incompatible with previous tasks' learned distributions, we propose a novel method based on embedding space regularization and data augmentation. Our method generalizes to new few-shot tasks and avoids catastrophic forgetting of previous tasks by enforcing extra constraints on the relational embeddings and by adding extra {relevant} data in a self-supervised manner. With extensive experiments we demonstrate that our method can significantly outperform previous state-of-the-art methods in CFRL task settings., Comment: ACL 2022

Building huge and highly capable language models has been a trend in the past years. Despite their great performance, they incur high computational cost. A common solution is to apply model compression or choose light-weight architectures, which often need a separate fixed-size model for each desirable computational budget, and may lose performance in case of heavy compression. This paper proposes an effective dynamic inference approach, called E-LANG, which distributes the inference between large accurate Super-models and light-weight Swift models. To this end, a decision making module routes the inputs to Super or Swift models based on the energy characteristics of the representations in the latent space. This method is easily adoptable and architecture agnostic. As such, it can be applied to black-box pre-trained models without a need for architectural manipulations, reassembling of modules, or re-training. Unlike existing methods that are only applicable to encoder-only backbones and classification tasks, our method also works for encoder-decoder structures and sequence-to-sequence tasks such as translation. The E-LANG performance is verified through a set of experiments with T5 and BERT backbones on GLUE, SuperGLUE, and WMT. In particular, we outperform T5-11B with an average computations speed-up of 3.3$\times$ on GLUE and 2.9$\times$ on SuperGLUE. We also achieve BERT-based SOTA on GLUE with 3.2$\times$ less computations. Code and demo are available in the supplementary materials., ACL 2022

We consider the problem of collectively detecting multiple events, particularly in cross-sentence settings. The key to dealing with the problem is to encode semantic information and model event inter-dependency at a document-level. In this paper, we reformulate it as a Seq2Seq task and propose a Multi-Layer Bidirectional Network (MLBiNet) to capture the document-level association of events and semantic information simultaneously. Specifically, a bidirectional decoder is firstly devised to model event inter-dependency within a sentence when decoding the event tag vector sequence. Secondly, an information aggregation module is employed to aggregate sentence-level semantic and event tag information. Finally, we stack multiple bidirectional decoders and feed cross-sentence information, forming a multi-layer bidirectional tagging architecture to iteratively propagate information across sentences. We show that our approach provides significant improvement in performance compared to the current state-of-the-art results., Accepted by ACL 2021

Pre-trained language models (PLMs) have achieved great success in natural language processing. Most of PLMs follow the default setting of architecture hyper-parameters (e.g., the hidden dimension is a quarter of the intermediate dimension in feed-forward sub-networks) in BERT (Devlin et al., 2019). Few studies have been conducted to explore the design of architecture hyper-parameters in BERT, especially for the more efficient PLMs with tiny sizes, which are essential for practical deployment on resource-constrained devices. In this paper, we adopt the one-shot Neural Architecture Search (NAS) to automatically search architecture hyper-parameters. Specifically, we carefully design the techniques of one-shot learning and the search space to provide an adaptive and efficient development way of tiny PLMs for various latency constraints. We name our method AutoTinyBERT and evaluate its effectiveness on the GLUE and SQuAD benchmarks. The extensive experiments show that our method outperforms both the SOTA search-based baseline (NAS-BERT) and the SOTA distillation-based methods (such as DistilBERT, TinyBERT, MiniLM and MobileBERT). In addition, based on the obtained architectures, we propose a more efficient development method that is even faster than the development of a single PLM., ACL 2021. The code and models are released at https://github.com/huawei-noah/Pretrained-Language-Model/tree/master/AutoTinyBERT

Compositional generalization is the ability to generalize systematically to a new data distribution by combining known components. Although humans seem to have a great ability to generalize compositionally, state-of-the-art neural models struggle to do so. In this work, we study compositional generalization in classification tasks and present two main contributions. First, we study ways to convert a natural language sequence-to-sequence dataset to a classification dataset that also requires compositional generalization. Second, we show that providing structural hints (specifically, providing parse trees and entity links as attention masks for a Transformer model) helps compositional generalization., Comment: Accepted as a short paper at ACL 2021

Mental health conditions remain underdiagnosed even in countries with common access to advanced medical care. The ability to accurately and efficiently predict mood from easily collectible data has several important implications for the early detection, intervention, and treatment of mental health disorders. One promising data source to help monitor human behavior is daily smartphone usage. However, care must be taken to summarize behaviors without identifying the user through personal (e.g., personally identifiable information) or protected (e.g., race, gender) attributes. In this paper, we study behavioral markers of daily mood using a recent dataset of mobile behaviors from adolescent populations at high risk of suicidal behaviors. Using computational models, we find that language and multimodal representations of mobile typed text (spanning typed characters, words, keystroke timings, and app usage) are predictive of daily mood. However, we find that models trained to predict mood often also capture private user identities in their intermediate representations. To tackle this problem, we evaluate approaches that obfuscate user identity while remaining predictive. By combining multimodal representations with privacy-preserving learning, we are able to push forward the performance-privacy frontier., Comment: ACL 2021. arXiv admin note: substantial text overlap with arXiv:2012.02359

In adversarial data collection (ADC), a human workforce interacts with a model in real time, attempting to produce examples that elicit incorrect predictions. Researchers hope that models trained on these more challenging datasets will rely less on superficial patterns, and thus be less brittle. However, despite ADC's intuitive appeal, it remains unclear when training on adversarial datasets produces more robust models. In this paper, we conduct a large-scale controlled study focused on question answering, assigning workers at random to compose questions either (i) adversarially (with a model in the loop); or (ii) in the standard fashion (without a model). Across a variety of models and datasets, we find that models trained on adversarial data usually perform better on other adversarial datasets but worse on a diverse collection of out-of-domain evaluation sets. Finally, we provide a qualitative analysis of adversarial (vs standard) data, identifying key differences and offering guidance for future research., Comment: Accepted at ACL-IJCNLP 2021

In open-domain question answering, questions are highly likely to be ambiguous because users may not know the scope of relevant topics when formulating them. Therefore, a system needs to find possible interpretations of the question, and predict one or multiple plausible answers. When multiple plausible answers are found, the system should rewrite the question for each answer to resolve the ambiguity. In this paper, we present a model that aggregates and combines evidence from multiple passages to adaptively predict a single answer or a set of question-answer pairs for ambiguous questions. In addition, we propose a novel round-trip prediction approach to iteratively generate additional interpretations that our model fails to find in the first pass, and then verify and filter out the incorrect question-answer pairs to arrive at the final disambiguated output. Our model, named Refuel, achieves a new state-of-the-art performance on the AmbigQA dataset, and shows competitive performance on NQ-Open and TriviaQA. The proposed round-trip prediction is a model-agnostic general approach for answering ambiguous open-domain questions, which improves our Refuel as well as several baseline models. We release source code for our models and experiments at https://github.com/amzn/refuel-open-domain-qa., ACL 2021 main conference, 14 pages, 7 figures. Code will be released at https://github.com/amzn/refuel-open-domain-qa

Transformer networks have revolutionized NLP representation learning since they were introduced. Though a great effort has been made to explain the representation in transformers, it is widely recognized that our understanding is not sufficient. One important reason is that there lack enough visualization tools for detailed analysis. In this paper, we propose to use dictionary learning to open up these "black boxes" as linear superpositions of transformer factors. Through visualization, we demonstrate the hierarchical semantic structures captured by the transformer factors, e.g., word-level polysemy disambiguation, sentence-level pattern formation, and long-range dependency. While some of these patterns confirm the conventional prior linguistic knowledge, the rest are relatively unexpected, which may provide new insights. We hope this visualization tool can bring further knowledge and a better understanding of how transformer networks work. The code is available at https://github.com/zeyuyun1/TransformerVis, Comment: This paper is published at DeeLIO Workshop@NAACL 2021

Human activities can be seen as sequences of events, which are crucial to understanding societies. Disproportional event distribution for different demographic groups can manifest and amplify social stereotypes, and potentially jeopardize the ability of members in some groups to pursue certain goals. In this paper, we present the first event-centric study of gender biases in a Wikipedia corpus. To facilitate the study, we curate a corpus of career and personal life descriptions with demographic information consisting of 7,854 fragments from 10,412 celebrities. Then we detect events with a state-of-the-art event detection model, calibrate the results using strategically generated templates, and extract events that have asymmetric associations with genders. Our study discovers that the Wikipedia pages tend to intermingle personal life events with professional events for females but not for males, which calls for the awareness of the Wikipedia community to formalize guidelines and train the editors to mind the implicit biases that contributors carry. Our work also lays the foundation for future works on quantifying and discovering event biases at the corpus level., Comment: ACL 2021

Accurate detection and classification of online hate is a difficult task. Implicit hate is particularly challenging as such content tends to have unusual syntax, polysemic words, and fewer markers of prejudice (e.g., slurs). This problem is heightened with multimodal content, such as memes (combinations of text and images), as they are often harder to decipher than unimodal content (e.g., text alone). This paper evaluates the role of semantic and multimodal context for detecting implicit and explicit hate. We show that both text- and visual- enrichment improves model performance, with the multimodal model (0.771) outperforming other models' F1 scores (0.544, 0.737, and 0.754). While the unimodal-text context-aware (transformer) model was the most accurate on the subtask of implicit hate detection, the multimodal model outperformed it overall because of a lower propensity towards false positives. We find that all models perform better on content with full annotator agreement and that multimodal models are best at classifying the content where annotators disagree. To conduct these investigations, we undertook high-quality annotation of a sample of 5,000 multimodal entries. Tweets were annotated for primary category, modality, and strategy. We make this corpus, along with the codebook, code, and final model, freely available., Comment: Please note the paper contains examples of hateful content

Structured information is an important knowledge source for automatic verification of factual claims. Nevertheless, the majority of existing research into this task has focused on textual data, and the few recent inquiries into structured data have been for the closed-domain setting where appropriate evidence for each claim is assumed to have already been retrieved. In this paper, we investigate verification over structured data in the open-domain setting, introducing a joint reranking-and-verification model which fuses evidence documents in the verification component. Our open-domain model achieves performance comparable to the closed-domain state-of-the-art on the TabFact dataset, and demonstrates performance gains from the inclusion of multiple tables as well as a significant improvement over a heuristic retrieval baseline.

Public trust in science depends on honest and factual communication of scientific papers. However, recent studies have demonstrated a tendency of news media to misrepresent scientific papers by exaggerating their findings. Given this, we present a formalization of and study into the problem of exaggeration detection in science communication. While there are an abundance of scientific papers and popular media articles written about them, very rarely do the articles include a direct link to the original paper, making data collection challenging. We address this by curating a set of labeled press release/abstract pairs from existing expert annotated studies on exaggeration in press releases of scientific papers suitable for benchmarking the performance of machine learning models on the task. Using limited data from this and previous studies on exaggeration detection in science, we introduce MT-PET, a multi-task version of Pattern Exploiting Training (PET), which leverages knowledge from complementary cloze-style QA tasks to improve few-shot learning. We demonstrate that MT-PET outperforms PET and supervised learning both when data is limited, as well as when there is an abundance of data for the main task., Accepted to EMNLP 2021; 13 pages, 6 figures, 9 tables

Recent investigations into the inner-workings of state-of-the-art large-scale pre-trained Transformer-based Natural Language Understanding (NLU) models indicate that they appear to know humanlike syntax, at least to some extent. We provide novel evidence that complicates this claim: we find that state-of-the-art Natural Language Inference (NLI) models assign the same labels to permuted examples as they do to the original, i.e. they are largely invariant to random word-order permutations. This behavior notably differs from that of humans; we struggle with ungrammatical sentences. To measure the severity of this issue, we propose a suite of metrics and investigate which properties of particular permutations lead models to be word-order invariant. In the MNLI dataset, for example, we find almost all (98.7%) examples contain at least one permutation which elicits the gold label. Models are sometimes even able to assign gold labels to permutations that they originally failed to predict correctly. We provide a comprehensive empirical evaluation of this phenomenon, and further show that this issue exists for both Transformers and pre-Transformer RNN / ConvNet based encoders, as well as across multiple languages (English and Mandarin Chinese). Our code and data are available at https://github.com/facebookresearch/unlu., Accepted at ACL 2021 (Long Paper), 9 pages + Appendix

Post-hoc explanation methods are an important class of approaches that help understand the rationale underlying a trained model's decision. But how useful are they for an end-user towards accomplishing a given task? In this vision paper, we argue the need for a benchmark to facilitate evaluations of the utility of post-hoc explanation methods. As a first step to this end, we enumerate desirable properties that such a benchmark should possess for the task of debugging text classifiers. Additionally, we highlight that such a benchmark facilitates not only assessing the effectiveness of explanations but also their efficiency., Short paper, to appear at TrustNLP @ NAACL 2021

Pretrained using large amount of data, autoregressive language models are able to generate high quality sequences. However, these models do not perform well under hard lexical constraints as they lack fine control of content generation process. Progressive insertion-based transformers can overcome the above limitation and efficiently generate a sequence in parallel given some input tokens as constraint. These transformers however may fail to support hard lexical constraints as their generation process is more likely to terminate prematurely. The paper analyses such early termination problems and proposes the Entity-constrained insertion transformer (ENCONTER), a new insertion transformer that addresses the above pitfall without compromising much generation efficiency. We introduce a new training strategy that considers predefined hard lexical constraints (e.g., entities to be included in the generated sequence). Our experiments show that ENCONTER outperforms other baseline models in several performance metrics rendering it more suitable in practical applications. Our code is available at https://github.com/LARC-CMU-SMU/Enconter, EACL 2021 long paper

The research community has proposed copious modifications to the Transformer architecture since it was introduced over three years ago, relatively few of which have seen widespread adoption. In this paper, we comprehensively evaluate many of these modifications in a shared experimental setting that covers most of the common uses of the Transformer in natural language processing. Surprisingly, we find that most modifications do not meaningfully improve performance. Furthermore, most of the Transformer variants we found beneficial were either developed in the same codebase that we used or are relatively minor changes. We conjecture that performance improvements may strongly depend on implementation details and correspondingly make some recommendations for improving the generality of experimental results., Comment: To appear at EMNLP 2021 as a conference paper

There is an increasing interest in continuous learning (CL), as data privacy is becoming a priority for real-world machine learning applications. Meanwhile, there is still a lack of academic NLP benchmarks that are applicable for realistic CL settings, which is a major challenge for the advancement of the field. In this paper we discuss some of the unrealistic data characteristics of public datasets, study the challenges of realistic single-task continuous learning as well as the effectiveness of data rehearsal as a way to mitigate accuracy loss. We construct a CL NER dataset from an existing publicly available dataset and release it along with the code to the research community., Comment: 11 pages, 2 figures, Findings of the 2021 Conference on Empirical Methods in Natural Language Processing (EMNLP) (short paper), November 2021

Crowdsourcing is regarded as one prospective solution for effective supervised learning, aiming to build large-scale annotated training data by crowd workers. Previous studies focus on reducing the influences from the noises of the crowdsourced annotations for supervised models. We take a different point in this work, regarding all crowdsourced annotations as gold-standard with respect to the individual annotators. In this way, we find that crowdsourcing could be highly similar to domain adaptation, and then the recent advances of cross-domain methods can be almost directly applied to crowdsourcing. Here we take named entity recognition (NER) as a study case, suggesting an annotator-aware representation learning model that inspired by the domain adaptation methods which attempt to capture effective domain-aware features. We investigate both unsupervised and supervised crowdsourcing learning, assuming that no or only small-scale expert annotations are available. Experimental results on a benchmark crowdsourced NER dataset show that our method is highly effective, leading to a new state-of-the-art performance. In addition, under the supervised setting, we can achieve impressive performance gains with only a very small scale of expert annotations., ACL-IJCNLP 2021 main conf, long paper; corrected the wrong reference for "argument retrieval" in first paragraph of Introduction

This paper presents a novel pre-trained language models (PLM) compression approach based on the matrix product operator (short as MPO) from quantum many-body physics. It can decompose an original matrix into central tensors (containing the core information) and auxiliary tensors (with only a small proportion of parameters). With the decomposed MPO structure, we propose a novel fine-tuning strategy by only updating the parameters from the auxiliary tensors, and design an optimization algorithm for MPO-based approximation over stacked network architectures. Our approach can be applied to the original or the compressed PLMs in a general way, which derives a lighter network and significantly reduces the parameters to be fine-tuned. Extensive experiments have demonstrated the effectiveness of the proposed approach in model compression, especially the reduction in finetuning parameters (91% reduction on average)., Comment: Accepted by ACL 2021 main conference

Spoken Language Understanding (SLU) systems parse speech into semantic structures like dialog acts and slots. This involves the use of an Automatic Speech Recognizer (ASR) to transcribe speech into multiple text alternatives (hypotheses). Transcription errors, common in ASRs, impact downstream SLU performance negatively. Approaches to mitigate such errors involve using richer information from the ASR, either in form of N-best hypotheses or word-lattices. We hypothesize that transformer models learn better with a simpler utterance representation using the concatenation of the N-best ASR alternatives, where each alternative is separated by a special delimiter [SEP]. In our work, we test our hypothesis by using concatenated N-best ASR alternatives as the input to transformer encoder models, namely BERT and XLM-RoBERTa, and achieve performance equivalent to the prior state-of-the-art model on DSTC2 dataset. We also show that our approach significantly outperforms the prior state-of-the-art when subjected to the low data regime. Additionally, this methodology is accessible to users of third-party ASR APIs which do not provide word-lattice information., Comment: 6 pages, 3 figures, Accepted at ACL 2021 as a main conference paper

Pre-trained language models have shown remarkable success in improving various downstream NLP tasks due to their ability to capture dependencies in textual data and generate natural responses. In this paper, we leverage the power of pre-trained language models for improving video-grounded dialogue, which is very challenging and involves complex features of different dynamics: (1) Video features which can extend across both spatial and temporal dimensions; and (2) Dialogue features which involve semantic dependencies over multiple dialogue turns. We propose a framework by extending GPT-2 models to tackle these challenges by formulating video-grounded dialogue tasks as a sequence-to-sequence task, combining both visual and textual representation into a structured sequence, and fine-tuning a large pre-trained GPT-2 network. Our framework allows fine-tuning language models to capture dependencies across multiple modalities over different levels of information: spatio-temporal level in video and token-sentence level in dialogue context. We achieve promising improvement on the Audio-Visual Scene-Aware Dialogues (AVSD) benchmark from DSTC7, which supports a potential direction in this line of research., Comment: Accepted at ACL 2020 (Short Paper)

Expert search aims to find and rank experts based on a user's query. In academia, retrieving experts is an efficient way to navigate through a large amount of academic knowledge. Here, we study how different distributed representations of academic papers (i.e. embeddings) impact academic expert retrieval. We use the Microsoft Academic Graph dataset and experiment with different configurations of a document-centric voting model for retrieval. In particular, we explore the impact of the use of contextualized embeddings on search performance. We also present results for paper embeddings that incorporate citation information through retrofitting. Additionally, experiments are conducted using different techniques for assigning author weights based on author order. We observe that using contextual embeddings produced by a transformer model trained for sentence similarity tasks produces the most effective paper representations for document-centric expert retrieval. However, retrofitting the paper embeddings and using elaborate author contribution weighting strategies did not improve retrieval performance., Comment: To be published in the Scholarly Document Processing 2020 Workshop @ EMNLP 2020 proceedings

Coreference resolution (CR) is an essential part of discourse analysis. Most recently, neural approaches have been proposed to improve over SOTA models from earlier paradigms. So far none of the published neural models leverage external semantic knowledge such as type information. This paper offers the first such model and evaluation, demonstrating modest gains in accuracy by introducing either gold standard or predicted types. In the proposed approach, type information serves both to (1) improve mention representation and (2) create a soft type consistency check between coreference candidate mentions. Our evaluation covers two different grain sizes of types over four different benchmark corpora., Comment: Accepted as Long Paper at CODI workshop EMNLP 2020

Training recurrent neural networks on long texts, in particular scholarly documents, causes problems for learning. While hierarchical attention networks (HANs) are effective in solving these problems, they still lose important information about the structure of the text. To tackle these problems, we propose the use of HANs combined with structure-tags which mark the role of sentences in the document. Adding tags to sentences, marking them as corresponding to title, abstract or main body text, yields improvements over the state-of-the-art for scholarly document quality prediction. The proposed system is applied to the task of accept/reject prediction on the PeerRead dataset and compared against a recent BiLSTM-based model and joint textual+visual model as well as against plain HANs. Compared to plain HANs, accuracy increases on all three domains. On the computation and language domain our new model works best overall, and increases accuracy 4.7% over the best literature result. We also obtain improvements when introducing the tags for prediction of the number of citations for 88k scientific publications that we compiled from the Allen AI S2ORC dataset. For our HAN-system with structure-tags we reach 28.5% explained variance, an improvement of 1.8% over our reimplementation of the BiLSTM-based model as well as 1.0% improvement over plain HANs., Comment: This new version of the paper brings the paper up-to-date with the improved paper, published at the First Workshop on Scholarly Document Processing, at EMNLP 2020. .Additionally, minor corrections were made including addition of color to Figures 1,2. The changes in comparison to the first arXiv version are substantial, including various additional results, and substantial improvements to the text

Document-level relation extraction aims to extract relations among entities within a document. Different from sentence-level relation extraction, it requires reasoning over multiple sentences across a document. In this paper, we propose Graph Aggregation-and-Inference Network (GAIN) featuring double graphs. GAIN first constructs a heterogeneous mention-level graph (hMG) to model complex interaction among different mentions across the document. It also constructs an entity-level graph (EG), based on which we propose a novel path reasoning mechanism to infer relations between entities. Experiments on the public dataset, DocRED, show GAIN achieves a significant performance improvement (2.85 on F1) over the previous state-of-the-art. Our code is available at https://github.com/DreamInvoker/GAIN ., Accepted as long paper to appear at the EMNLP 2020 main conference, 11 pages, 3 figures

The COVID-19 pandemic has sparked unprecedented mobilization of scientists, generating a deluge of papers that makes it hard for researchers to keep track and explore new directions. Search engines are designed for targeted queries, not for discovery of connections across a corpus. In this paper, we present SciSight, a system for exploratory search of COVID-19 research integrating two key capabilities: first, exploring associations between biomedical facets automatically extracted from papers (e.g., genes, drugs, diseases, patient outcomes); second, combining textual and network information to search and visualize groups of researchers and their ties. SciSight has so far served over $15K$ users with over $42K$ page views and $13\%$ returns., Comment: Accepted to EMNLP 2020

Automatic text summarization has been widely studied as an important task in natural language processing. Traditionally, various feature engineering and machine learning based systems have been proposed for extractive as well as abstractive text summarization. Recently, deep learning based, specifically Transformer-based systems have been immensely popular. Summarization is a cognitively challenging task - extracting summary worthy sentences is laborious, and expressing semantics in brief when doing abstractive summarization is complicated. In this paper, we specifically look at the problem of summarizing scientific research papers from multiple domains. We differentiate between two types of summaries, namely, (a) LaySumm: A very short summary that captures the essence of the research paper in layman terms restricting overtly specific technical jargon and (b) LongSumm: A much longer detailed summary aimed at providing specific insights into various ideas touched upon in the paper. While leveraging latest Transformer-based models, our systems are simple, intuitive and based on how specific paper sections contribute to human summaries of the two types described above. Evaluations against gold standard summaries using ROUGE metrics prove the effectiveness of our approach. On blind test corpora, our system ranks first and third for the LongSumm and LaySumm tasks respectively., Comment: Proceedings of the First Workshop on Scholarly Document Processing (SDP) at EMNLP 2020

Performing fact verification based on structured data is important for many real-life applications and is a challenging research problem, particularly when it involves both symbolic operations and informal inference based on language understanding. In this paper, we present a Program-enhanced Verbalization and Graph Attention Network (ProgVGAT) to integrate programs and execution into textual inference models. Specifically, a verbalization with program execution model is proposed to accumulate evidences that are embedded in operations over the tables. Built on that, we construct the graph attention verification networks, which are designed to fuse different sources of evidences from verbalized program execution, program structures, and the original statements and tables, to make the final verification decision. To support the above framework, we propose a program selection module optimized with a new training strategy based on margin loss, to produce more accurate programs, which is shown to be effective in enhancing the final verification results. Experimental results show that the proposed framework achieves the new state-of-the-art performance, a 74.4% accuracy, on the benchmark dataset TABFACT., Comment: 16 pages (Accepted by EMNLP 2020 as a long paper)

Many NLP tasks have benefited from transferring knowledge from contextualized word embeddings, however the picture of what type of knowledge is transferred is incomplete. This paper studies the types of linguistic phenomena accounted for by language models in the context of a Conversational Question Answering (CoQA) task. We identify the problematic areas for the finetuned RoBERTa, BERT and DistilBERT models through systematic error analysis - basic arithmetic (counting phrases), compositional semantics (negation and Semantic Role Labeling), and lexical semantics (surprisal and antonymy). When enhanced with the relevant linguistic knowledge through multitask learning, the models improve in performance. Ensembles of the enhanced models yield a boost between 2.2 and 2.7 points in F1 score overall, and up to 42.1 points in F1 on the hardest question classes. The results show differences in ability to represent compositional and lexical information between RoBERTa, BERT and DistilBERT., 8 pages + 2 pages references, 3 tables, 1 figure. Accepted as long paper in EMNLP2020 main conference

Disfluency detection is usually an intermediate step between an automatic speech recognition (ASR) system and a downstream task. By contrast, this paper aims to investigate the task of end-to-end speech recognition and disfluency removal. We specifically explore whether it is possible to train an ASR model to directly map disfluent speech into fluent transcripts, without relying on a separate disfluency detection model. We show that end-to-end models do learn to directly generate fluent transcripts; however, their performance is slightly worse than a baseline pipeline approach consisting of an ASR system and a specialized disfluency detection model. We also propose two new metrics for evaluating integrated ASR and disfluency removal models. The findings of this paper can serve as a benchmark for further research on the task of end-to-end speech recognition and disfluency removal in the future.

Decisions of complex language understanding models can be rationalized by limiting their inputs to a relevant subsequence of the original text. A rationale should be as concise as possible without significantly degrading task performance, but this balance can be difficult to achieve in practice. In this paper, we show that it is possible to better manage this trade-off by optimizing a bound on the Information Bottleneck (IB) objective. Our fully unsupervised approach jointly learns an explainer that predicts sparse binary masks over sentences, and an end-task predictor that considers only the extracted rationale. Using IB, we derive a learning objective that allows direct control of mask sparsity levels through a tunable sparse prior. Experiments on ERASER benchmark tasks demonstrate significant gains over norm-minimization techniques for both task performance and agreement with human rationales. Furthermore, we find that in the semi-supervised setting, a modest amount of gold rationales (25% of training examples) closes the gap with a model that uses the full input., Comment: EMNLP 2020 main track accepted paper

Adversarial attacks are carried out to reveal the vulnerability of deep neural networks. Textual adversarial attacking is challenging because text is discrete and a small perturbation can bring significant change to the original input. Word-level attacking, which can be regarded as a combinatorial optimization problem, is a well-studied class of textual attack methods. However, existing word-level attack models are far from perfect, largely because unsuitable search space reduction methods and inefficient optimization algorithms are employed. In this paper, we propose a novel attack model, which incorporates the sememe-based word substitution method and particle swarm optimization-based search algorithm to solve the two problems separately. We conduct exhaustive experiments to evaluate our attack model by attacking BiLSTM and BERT on three benchmark datasets. Experimental results demonstrate that our model consistently achieves much higher attack success rates and crafts more high-quality adversarial examples as compared to baseline methods. Also, further experiments show our model has higher transferability and can bring more robustness enhancement to victim models by adversarial training. All the code and data of this paper can be obtained on https://github.com/thunlp/SememePSO-Attack., Comment: Accepted at ACL 2020 as a long paper (a typo is corrected as compared with the official conference camera-ready version). 16 pages, 3 figures

Pre-trained language models have achieved huge improvement on many NLP tasks. However, these methods are usually designed for written text, so they do not consider the properties of spoken language. Therefore, this paper aims at generalizing the idea of language model pre-training to lattices generated by recognition systems. We propose a framework that trains neural lattice language models to provide contextualized representations for spoken language understanding tasks. The proposed two-stage pre-training approach reduces the demands of speech data and has better efficiency. Experiments on intent detection and dialogue act recognition datasets demonstrate that our proposed method consistently outperforms strong baselines when evaluated on spoken inputs. The code is available at https://github.com/MiuLab/Lattice-ELMo., Published in ACL 2020 as a short paper

Building equitable and inclusive NLP technologies demands consideration of whether and how social attitudes are represented in ML models. In particular, representations encoded in models often inadvertently perpetuate undesirable social biases from the data on which they are trained. In this paper, we present evidence of such undesirable biases towards mentions of disability in two different English language models: toxicity prediction and sentiment analysis. Next, we demonstrate that the neural embeddings that are the critical first step in most NLP pipelines similarly contain undesirable biases towards mentions of disability. We end by highlighting topical biases in the discourse about disability which may contribute to the observed model biases; for instance, gun violence, homelessness, and drug addiction are over-represented in texts discussing mental illness., ACL 2020 short paper. 5 pages

Automated evaluation metrics as a stand-in for manual evaluation are an essential part of the development of text-generation tasks such as text summarization. However, while the field has progressed, our standard metrics have not -- for nearly 20 years ROUGE has been the standard evaluation in most summarization papers. In this paper, we make an attempt to re-evaluate the evaluation method for text summarization: assessing the reliability of automatic metrics using top-scoring system outputs, both abstractive and extractive, on recently popular datasets for both system-level and summary-level evaluation settings. We find that conclusions about evaluation metrics on older datasets do not necessarily hold on modern datasets and systems., Accepted at EMNLP 2020

The recognition of pharmacological substances, compounds and proteins is an essential preliminary work for the recognition of relations between chemicals and other biomedically relevant units. In this paper, we describe an approach to Task 1 of the PharmaCoNER Challenge, which involves the recognition of mentions of chemicals and drugs in Spanish medical texts. We train a state-of-the-art BiLSTM-CRF sequence tagger with stacked Pooled Contextualized Embeddings, word and sub-word embeddings using the open-source framework FLAIR. We present a new corpus composed of articles and papers from Spanish health science journals, termed the Spanish Health Corpus, and use it to train domain-specific embeddings which we incorporate in our model training. We achieve a result of 89.76% F1-score using pre-trained embeddings and are able to improve these results to 90.52% F1-score using specialized embeddings., Comment: EMNLP-IJCNLP 2019: International Workshop on BioNLP Open Shared Tasks 2019, 5, pages, 1 figure

This paper addresses the challenging task of video captioning which aims to generate descriptions for video data. Recently, the attention-based encoder-decoder structures have been widely used in video captioning. In existing literature, the attention weights are often built from the information of an individual modality, while, the association relationships between multiple modalities are neglected. Motivated by this observation, we propose a video captioning model with High-Order Cross-Modal Attention (HOCA) where the attention weights are calculated based on the high-order correlation tensor to capture the frame-level cross-modal interaction of different modalities sufficiently. Furthermore, we novelly introduce Low-Rank HOCA which adopts tensor decomposition to reduce the extremely large space requirement of HOCA, leading to a practical and efficient implementation in real-world applications. Experimental results on two benchmark datasets, MSVD and MSR-VTT, show that Low-rank HOCA establishes a new state-of-the-art., Comment: Accepted as a long paper at EMNLP 2019