EMNLP 2021 Papers

Research in NLP is often supported by experimental results, and improved reporting of such results can lead to better understanding and more reproducible science. In this paper we analyze three statistical estimators for expected validation performance, a tool used for reporting performance (e.g., accuracy) as a function of computational budget (e.g., number of hyperparameter tuning experiments). Where previous work analyzing such estimators focused on the bias, we also examine the variance and mean squared error (MSE). In both synthetic and realistic scenarios, we evaluate three estimators and find the unbiased estimator has the highest variance, and the estimator with the smallest variance has the largest bias; the estimator with the smallest MSE strikes a balance between bias and variance, displaying a classic bias-variance tradeoff. We use expected validation performance to compare between different models, and analyze how frequently each estimator leads to drawing incorrect conclusions about which of two models performs best. We find that the two biased estimators lead to the fewest incorrect conclusions, which hints at the importance of minimizing variance and MSE.

End-to-end question answering using a differentiable knowledge graph is a promising technique that requires only weak supervision, produces interpretable results, and is fully differentiable. Previous implementations of this technique (Cohen et al, 2020) have focused on single-entity questions using a relation following operation. In this paper, we propose a model that explicitly handles multiple-entity questions by implementing a new intersection operation, which identifies the shared elements between two sets of entities. We find that introducing intersection improves performance over a baseline model on two datasets, WebQuestionsSP (69.6% to 73.3% Hits@1) and ComplexWebQuestions (39.8% to 48.7% Hits@1), and in particular, improves performance on questions with multiple entities by over 14% on WebQuestionsSP and by 19% on ComplexWebQuestions.

Resolving pronouns to their referents has long been studied as a fundamental natural language understanding problem. Previous works on pronoun coreference resolution (PCR) mostly focus on resolving pronouns to mentions in text while ignoring the exophoric scenario. Exophoric pronouns are common in daily communications, where speakers may directly use pronouns to refer to some objects present in the environment without introducing the objects first. Although such objects are not mentioned in the dialogue text, they can often be disambiguated by the general topics of the dialogue. Motivated by this, we propose to jointly leverage the local context and global topics of dialogues to solve the out-of-text PCR problem. Extensive experiments demonstrate the effectiveness of adding topic regularization for resolving exophoric pronouns.

Timely responses from policy makers to mitigate the impact of the COVID-19 pandemic rely on a comprehensive grasp of events, their causes, and their impacts. These events are reported at such a speed and scale as to be overwhelming. In this paper, we present ExcavatorCovid, a machine reading system that ingests open-source text documents (e.g., news and scientific publications), extracts COVID-19 related events and relations between them, and builds a Temporal and Causal Analysis Graph (TCAG). Excavator will help government agencies alleviate the information overload, understand likely downstream effects of political and economic decisions and events related to the pandemic, and respond in a timely manner to mitigate the impact of COVID-19. We expect the utility of Excavator to outlive the COVID-19 pandemic: analysts and decision makers will be empowered by Excavator to better understand and solve complex problems in the future. A demonstration video is available at https://vimeo.com/528619007.

State-of-the-art contextual embeddings are obtained from large language models available only for a few languages. For others, we need to learn representations using a multilingual model. There is an ongoing debate on whether multilingual embeddings can be aligned in a space shared across many languages. The novel Orthogonal Structural Probe (Limisiewicz and Mareček, 2021) allows us to answer this question for specific linguistic features and learn a projection based only on mono-lingual annotated datasets. We evaluate syntactic (UD) and lexical (WordNet) structural information encoded inmBERT’s contextual representations for nine diverse languages. We observe that for languages closely related to English, no transformation is needed. The evaluated information is encoded in a shared cross-lingual embedding space. For other languages, it is beneficial to apply orthogonal transformation learned separately for each language. We successfully apply our findings to zero-shot and few-shot cross-lingual parsing.

The task of verifying the truthfulness of claims in textual documents, or fact-checking, has received significant attention in recent years. Many existing evidence-based factchecking datasets contain synthetic claims and the models trained on these data might not be able to verify real-world claims. Particularly few studies addressed evidence-based fact-checking of health-related claims that require medical expertise or evidence from the scientific literature. In this paper, we introduce HEALTHVER, a new dataset for evidence-based fact-checking of health-related claims that allows to study the validity of real-world claims by evaluating their truthfulness against scientific articles. Using a three-step data creation method, we first retrieved real-world claims from snippets returned by a search engine for questions about COVID-19. Then we automatically retrieved and re-ranked relevant scientific papers using a T5 relevance-based model. Finally, the relations between each evidence statement and the associated claim were manually annotated as SUPPORT, REFUTE and NEUTRAL. To validate the created dataset of 14,330 evidence-claim pairs, we developed baseline models based on pretrained language models. Our experiments showed that training deep learning models on real-world medical claims greatly improves performance compared to models trained on synthetic and open-domain claims. Our results and manual analysis suggest that HEALTHVER provides a realistic and challenging dataset for future efforts on evidence-based fact-checking of health-related claims. The dataset, source code, and a leaderboard are available at https://github.com/sarrouti/healthver.

Zero-shot cross-lingual information extraction (IE) describes the construction of an IE model for some target language, given existing annotations exclusively in some other language, typically English. While the advance of pretrained multilingual encoders suggests an easy optimism of “train on English, run on any language”, we find through a thorough exploration and extension of techniques that a combination of approaches, both new and old, leads to better performance than any one cross-lingual strategy in particular. We explore techniques including data projection and self-training, and how different pretrained encoders impact them. We use English-to-Arabic IE as our initial example, demonstrating strong performance in this setting for event extraction, named entity recognition, part-of-speech tagging, and dependency parsing. We then apply data projection and self-training to three tasks across eight target languages. Because no single set of techniques performs the best across all tasks, we encourage practitioners to explore various configurations of the techniques described in this work when seeking to improve on zero-shot training.

Relations in most of the traditional knowledge graphs (KGs) only reflect static and factual connections, but fail to represent the dynamic activities and state changes about entities. In this paper, we emphasize the importance of incorporating events in KG representation learning, and propose an event-enhanced KG embedding model EventKE. Specifically, given the original KG, we first incorporate event nodes by building a heterogeneous network, where entity nodes and event nodes are distributed on the two sides of the network inter-connected by event argument links. We then use entity-entity relations from the original KG and event-event temporal links to inner-connect entity and event nodes respectively. We design a novel and effective attention-based message passing method, which is conducted on entity-entity, event-entity, and event-event relations to fuse the event information into KG embeddings. Experimental results on real-world datasets demonstrate that events can greatly improve the quality of the KG embeddings on multiple downstream tasks.

Sentence fusion is a conditional generation task that merges several related sentences into a coherent one, which can be deemed as a summary sentence. The importance of sentence fusion has long been recognized by communities in natural language generation, especially in text summarization. It remains challenging for a state-of-the-art neural abstractive summarization model to generate a well-integrated summary sentence. In this paper, we explore the effective sentence fusion method in the context of text summarization. We propose to build an event graph from the input sentences to effectively capture and organize related events in a structured way and use the constructed event graph to guide sentence fusion. In addition to make use of the attention over the content of sentences and graph nodes, we further develop a graph flow attention mechanism to control the fusion process via the graph structure. When evaluated on sentence fusion data built from two summarization datasets, CNN/DaliyMail and Multi-News, our model shows to achieve state-of-the-art performance in terms of Rouge and other metrics like fusion rate and faithfulness.

Cross-document event coreference resolution (CDCR) is the task of identifying which event mentions refer to the same events throughout a collection of documents. Annotating CDCR data is an arduous and expensive process, explaining why existing corpora are small and lack domain coverage. To overcome this bottleneck, we automatically extract event coreference data from hyperlinks in online news: When referring to a significant real-world event, writers often add a hyperlink to another article covering this event. We demonstrate that collecting hyperlinks which point to the same article(s) produces extensive and high-quality CDCR data and create a corpus of 2M documents and 2.7M silver-standard event mentions called HyperCoref. We evaluate a state-of-the-art system on three CDCR corpora and find that models trained on small subsets of HyperCoref are highly competitive, with performance similar to models trained on gold-standard data. With our work, we free CDCR research from depending on costly human-annotated training data and open up possibilities for research beyond English CDCR, as our data extraction approach can be easily adapted to other languages.

Question answering (QA) primarily descends from two branches of research: (1) Alan Turing’s investigation of machine intelligence at Manchester University and (2) Cyril Cleverdon’s comparison of library card catalog indices at Cranfield University. This position paper names and distinguishes these paradigms. Despite substantial overlap, subtle but significant distinctions exert an outsize influence on research. While one evaluation paradigm values creating more intelligent QA systems, the other paradigm values building QA systems that appeal to users. By better understanding the epistemic heritage of QA, researchers, academia, and industry can more effectively accelerate QA research.

High-performance neural language models have obtained state-of-the-art results on a wide range of Natural Language Processing (NLP) tasks. However, results for common benchmark datasets often do not reflect model reliability and robustness when applied to noisy, real-world data. In this study, we design and implement various types of character-level and word-level perturbation methods to simulate realistic scenarios in which input texts may be slightly noisy or different from the data distribution on which NLP systems were trained. Conducting comprehensive experiments on different NLP tasks, we investigate the ability of high-performance language models such as BERT, XLNet, RoBERTa, and ELMo in handling different types of input perturbations. The results suggest that language models are sensitive to input perturbations and their performance can decrease even when small changes are introduced. We highlight that models need to be further improved and that current benchmarks are not reflecting model robustness well. We argue that evaluations on perturbed inputs should routinely complement widely-used benchmarks in order to yield a more realistic understanding of NLP systems’ robustness.

Text generation systems are ubiquitous in natural language processing applications. However, evaluation of these systems remains a challenge, especially in multilingual settings. In this paper, we propose L’AMBRE – a metric to evaluate the morphosyntactic well-formedness of text using its dependency parse and morphosyntactic rules of the language. We present a way to automatically extract various rules governing morphosyntax directly from dependency treebanks. To tackle the noisy outputs from text generation systems, we propose a simple methodology to train robust parsers. We show the effectiveness of our metric on the task of machine translation through a diachronic study of systems translating into morphologically-rich languages.

While the field of style transfer (ST) has been growing rapidly, it has been hampered by a lack of standardized practices for automatic evaluation. In this paper, we evaluate leading automatic metrics on the oft-researched task of formality style transfer. Unlike previous evaluations, which focus solely on English, we expand our focus to Brazilian-Portuguese, French, and Italian, making this work the first multilingual evaluation of metrics in ST. We outline best practices for automatic evaluation in (formality) style transfer and identify several models that correlate well with human judgments and are robust across languages. We hope that this work will help accelerate development in ST, where human evaluation is often challenging to collect.

Quantitatively measuring the impact-related aspects of scientific, engineering, and technological (SET) innovations is a fundamental problem with broad applications. Traditional citation-based measures for assessing the impact of innovations and related entities do not take into account the content of the publications. This limits their ability to provide rigorous quality-related metrics because they cannot account for the reasons that led to a citation. We present approaches to estimate content-aware bibliometrics to quantitatively measure the scholarly impact of a publication. Our approaches assess the impact of a cited publication by the extent to which the cited publication informs the citing publication. We introduce a new metric, called “Content Informed Index” (CII), that uses the content of the paper as a source of distant-supervision, to quantify how much the cited-node informs the citing-node. We evaluate the weights estimated by our approach on three manually annotated datasets, where the annotations quantify the extent of information in the citation. Particularly, we evaluate how well the ranking imposed by our approach associates with the ranking imposed by the manual annotations. CII achieves up to 103% improvement in performance as compared to the second-best performing approach.

Bias is pervasive for NLP models, motivating the development of automatic debiasing techniques. Evaluation of NLP debiasing methods has largely been limited to binary attributes in isolation, e.g., debiasing with respect to binary gender or race, however many corpora involve multiple such attributes, possibly with higher cardinality. In this paper we argue that a truly fair model must consider ‘gerrymandering’ groups which comprise not only single attributes, but also intersectional groups. We evaluate a form of bias-constrained model which is new to NLP, as well an extension of the iterative nullspace projection technique which can handle multiple identities.

It is a well-known approach for fringe groups and organizations to use euphemisms—ordinary-sounding and innocent-looking words with a secret meaning—to conceal what they are discussing. For instance, drug dealers often use “pot” for marijuana and “avocado” for heroin. From a social media content moderation perspective, though recent advances in NLP have enabled the automatic detection of such single-word euphemisms, no existing work is capable of automatically detecting multi-word euphemisms, such as “blue dream” (marijuana) and “black tar” (heroin). Our paper tackles the problem of euphemistic phrase detection without human effort for the first time, as far as we are aware. We first perform phrase mining on a raw text corpus (e.g., social media posts) to extract quality phrases. Then, we utilize word embedding similarities to select a set of euphemistic phrase candidates. Finally, we rank those candidates by a masked language model—SpanBERT. Compared to strong baselines, we report 20-50% higher detection accuracies using our algorithm for detecting euphemistic phrases.

Multi-modal machine translation (MMT) aims at improving translation performance by incorporating visual information. Most of the studies leverage the visual information through integrating the global image features as auxiliary input or decoding by attending to relevant local regions of the image. However, this kind of usage of visual information makes it difficult to figure out how the visual modality helps and why it works. Inspired by the findings of (CITATION) that entities are most informative in the image, we propose an explicit entity-level cross-modal learning approach that aims to augment the entity representation. Specifically, the approach is framed as a reconstruction task that reconstructs the original textural input from multi-modal input in which entities are replaced with visual features. Then, a multi-task framework is employed to combine the translation task and the reconstruction task to make full use of cross-modal entity representation learning. The extensive experiments demonstrate that our approach can achieve comparable or even better performance than state-of-the-art models. Furthermore, our in-depth analysis shows how visual information improves translation.

While powerful pre-trained language models have improved the fluency of text generation models, semantic adequacy -the ability to generate text that is semantically faithful to the input- remains an unsolved issue. In this paper, we introduce a novel automatic evaluation metric, Entity-Based Semantic Adequacy, which can be used to assess to what extent generation models that verbalise RDF (Resource Description Framework) graphs produce text that contains mentions of the entities occurring in the RDF input. This is important as RDF subject and object entities make up 2/3 of the input. We use our metric to compare 25 models from the WebNLG Shared Tasks and we examine correlation with results from human evaluations of semantic adequacy. We show that while our metric correlates with human evaluation scores, this correlation varies with the specifics of the human evaluation setup. This suggests that in order to measure the entity-based adequacy of generated texts, an automatic metric such as the one proposed here might be more reliable, as less subjective and more focused on correct verbalisation of the input, than human evaluation measures.