ACL 2021 论文列表

Recently, mT5 - a massively multilingual version of T5 - leveraged a unified text-to-text format to attain state-of-the-art results on a wide variety of multilingual NLP tasks. In this paper, we investigate the impact of incorporating parallel data into mT5 pre-training. We find that multi-tasking language modeling with objectives such as machine translation during pre-training is a straightforward way to improve performance on downstream multilingual and cross-lingual tasks. However, the gains start to diminish as the model capacity increases, suggesting that parallel data might not be as essential for larger models. At the same time, even at larger model sizes, we find that pre-training with parallel data still provides benefits in the limited labelled data regime

We introduce mTVR, a large-scale multilingual video moment retrieval dataset, containing 218K English and Chinese queries from 21.8K TV show video clips. The dataset is collected by extending the popular TVR dataset (in English) with paired Chinese queries and subtitles. Compared to existing moment retrieval datasets, mTVR is multilingual, larger, and comes with diverse annotations. We further propose mXML, a multilingual moment retrieval model that learns and operates on data from both languages, via encoder parameter sharing and language neighborhood constraints. We demonstrate the effectiveness of mXML on the newly collected mTVR dataset, where mXML outperforms strong monolingual baselines while using fewer parameters. In addition, we also provide detailed dataset analyses and model ablations. Data and code are publicly available at https://github.com/jayleicn/mTVRetrieval

BERT has been shown to be extremely effective on a wide variety of natural language processing tasks, including sentiment analysis and emotion detection. However, the proposed pretraining objectives of BERT do not induce any sentiment or emotion-specific biases into the model. In this paper, we present Emotion Masked Language Modelling, a variation of Masked Language Modelling aimed at improving the BERT language representation model for emotion detection and sentiment analysis tasks. Using the same pre-training corpora as the original model, Wikipedia and BookCorpus, our BERT variation manages to improve the downstream performance on 4 tasks from emotion detection and sentiment analysis by an average of 1.2% F-1. Moreover, our approach shows an increased performance in our task-specific robustness tests.

Parsing spoken dialogue poses unique difficulties, including disfluencies and unmarked boundaries between sentence-like units. Previous work has shown that prosody can help with parsing disfluent speech (Tran et al. 2018), but has assumed that the input to the parser is already segmented into sentence-like units (SUs), which isn’t true in existing speech applications. We investigate how prosody affects a parser that receives an entire dialogue turn as input (a turn-based model), instead of gold standard pre-segmented SUs (an SU-based model). In experiments on the English Switchboard corpus, we find that when using transcripts alone, the turn-based model has trouble segmenting SUs, leading to worse parse performance than the SU-based model. However, prosody can effectively replace gold standard SU boundaries: with prosody, the turn-based model performs as well as the SU-based model (91.38 vs. 91.06 F1 score, respectively), despite performing two tasks (SU segmentation and parsing) rather than one (parsing alone). Analysis shows that pitch and intensity features are the most important for this corpus, since they allow the model to correctly distinguish an SU boundary from a speech disfluency – a distinction that the model otherwise struggles to make.

Context-aware neural machine translation (NMT) remains challenging due to the lack of large-scale document-level parallel corpora. To break the corpus bottleneck, in this paper we aim to improve context-aware NMT by taking the advantage of the availability of both large-scale sentence-level parallel dataset and source-side monolingual documents. To this end, we propose two pre-training tasks. One learns to translate a sentence from source language to target language on the sentence-level parallel dataset while the other learns to translate a document from deliberately noised to original on the monolingual documents. Importantly, the two pre-training tasks are jointly and simultaneously learned via the same model, thereafter fine-tuned on scale-limited parallel documents from both sentence-level and document-level perspectives. Experimental results on four translation tasks show that our approach significantly improves translation performance. One nice property of our approach is that the fine-tuned model can be used to translate both sentences and documents.

Neural models for automated fact verification have achieved promising results thanks to the availability of large, human-annotated datasets. However, for each new domain that requires fact verification, creating a dataset by manually writing claims and linking them to their supporting evidence is expensive. We develop QACG, a framework for training a robust fact verification model by using automatically generated claims that can be supported, refuted, or unverifiable from evidence from Wikipedia. QACG generates question-answer pairs from the evidence and then converts them into different types of claims. Experiments on the FEVER dataset show that our QACG framework significantly reduces the demand for human-annotated training data. In a zero-shot scenario, QACG improves a RoBERTa model’s F1 from 50% to 77%, equivalent in performance to 2K+ manually-curated examples. Our QACG code is publicly available.

Event extraction has long been a challenging task, addressed mostly with supervised methods that require expensive annotation and are not extensible to new event ontologies. In this work, we explore the possibility of zero-shot event extraction by formulating it as a set of Textual Entailment (TE) and/or Question Answering (QA) queries (e.g. “A city was attacked” entails “There is an attack”), exploiting pretrained TE/QA models for direct transfer. On ACE-2005 and ERE, our system achieves acceptable results, yet there is still a large gap from supervised approaches, showing that current QA and TE technologies fail in transferring to a different domain. To investigate the reasons behind the gap, we analyze the remaining key challenges, their respective impact, and possible improvement directions.

Due to the scarcity of annotated data, Abstract Meaning Representation (AMR) research is relatively limited and challenging for languages other than English. Upon the availability of English AMR dataset and English-to- X parallel datasets, in this paper we propose a novel cross-lingual pre-training approach via multi-task learning (MTL) for both zeroshot AMR parsing and AMR-to-text generation. Specifically, we consider three types of relevant tasks, including AMR parsing, AMR-to-text generation, and machine translation. We hope that knowledge gained while learning for English AMR parsing and text generation can be transferred to the counterparts of other languages. With properly pretrained models, we explore four different finetuning methods, i.e., vanilla fine-tuning with a single task, one-for-all MTL fine-tuning, targeted MTL fine-tuning, and teacher-studentbased MTL fine-tuning. Experimental results on AMR parsing and text generation of multiple non-English languages demonstrate that our approach significantly outperforms a strong baseline of pre-training approach, and greatly advances the state of the art. In detail, on LDC2020T07 we have achieved 70.45%, 71.76%, and 70.80% in Smatch F1 for AMR parsing of German, Spanish, and Italian, respectively, while for AMR-to-text generation of the languages, we have obtained 25.69, 31.36, and 28.42 in BLEU respectively. We make our code available on github https://github.com/xdqkid/XLPT-AMR.

In this work, we introduce : the largest publicly available multilingual dataset for factual verification of naturally existing real-world claims. The dataset contains short statements in 25 languages and is labeled for veracity by expert fact-checkers. The dataset includes a multilingual evaluation benchmark that measures both out-of-domain generalization, and zero-shot capabilities of the multilingual models. Using state-of-the-art multilingual transformer-based models, we develop several automated fact-checking models that, along with textual claims, make use of additional metadata and evidence from news stories retrieved using a search engine. Empirically, our best model attains an F-score of around 40%, suggesting that our dataset is a challenging benchmark for the evaluation of multilingual fact-checking models.

Medical report generation is one of the most challenging tasks in medical image analysis. Although existing approaches have achieved promising results, they either require a predefined template database in order to retrieve sentences or ignore the hierarchical nature of medical report generation. To address these issues, we propose MedWriter that incorporates a novel hierarchical retrieval mechanism to automatically extract both report and sentence-level templates for clinically accurate report generation. MedWriter first employs the Visual-Language Retrieval (VLR) module to retrieve the most relevant reports for the given images. To guarantee the logical coherence between generated sentences, the Language-Language Retrieval (LLR) module is introduced to retrieve relevant sentences based on the previous generated description. At last, a language decoder fuses image features and features from retrieved reports and sentences to generate meaningful medical reports. We verified the effectiveness of our model by automatic evaluation and human evaluation on two datasets, i.e., Open-I and MIMIC-CXR.

Lately proposed Word Sense Disambiguation (WSD) systems have approached the estimated upper bound of the task on standard evaluation benchmarks. However, these systems typically implement the disambiguation of words in a document almost independently, underutilizing sense and word dependency in context. In this paper, we convert the nearly isolated decisions into interrelated ones by exposing senses in context when learning sense embeddings in a similarity-based Sense Aware Context Exploitation (SACE) architecture. Meanwhile, we enhance the context embedding learning with selected sentences from the same document, rather than utilizing only the sentence where each ambiguous word appears. Experiments on both English and multilingual WSD datasets have shown the effectiveness of our approach, surpassing previous state-of-the-art by large margins (3.7% and 1.2% respectively), especially on few-shot (14.3%) and zero-shot (35.9%) scenarios.

Recent works made significant advances on summarization tasks, facilitated by summarization datasets. Several existing datasets have the form of coherent-paragraph summaries. However, these datasets were curated from academic documents that were written for experts, thus making the essential step of assessing the summarization output through human-evaluation very demanding. To overcome these limitations, we present a dataset based on article summaries appearing on the WikiHow website, composed of how-to articles and coherent-paragraph summaries written in plain language. We compare our dataset attributes to existing ones, including readability and world-knowledge, showing our dataset makes human evaluation significantly easier and thus, more effective. A human evaluation conducted on PubMed and the proposed dataset reinforces our findings.

Many Question-Answering (QA) datasets contain unanswerable questions, but their treatment in QA systems remains primitive. Our analysis of the Natural Questions (Kwiatkowski et al. 2019) dataset reveals that a substantial portion of unanswerable questions (~21%) can be explained based on the presence of unverifiable presuppositions. Through a user preference study, we demonstrate that the oracle behavior of our proposed system—which provides responses based on presupposition failure—is preferred over the oracle behavior of existing QA systems. Then, we present a novel framework for implementing such a system in three steps: presupposition generation, presupposition verification, and explanation generation, reporting progress on each. Finally, we show that a simple modification of adding presuppositions and their verifiability to the input of a competitive end-to-end QA system yields modest gains in QA performance and unanswerability detection, demonstrating the promise of our approach.

Subword segmentation algorithms have been a de facto choice when building neural machine translation systems. However, most of them need to learn a segmentation model based on some heuristics, which may produce sub-optimal segmentation. This can be problematic in some scenarios when the target language has rich morphological changes or there is not enough data for learning compact composition rules. Translating at fully character level has the potential to alleviate the issue, but empirical performances of character-based models has not been fully explored. In this paper, we present an in-depth comparison between character-based and subword-based NMT systems under three settings: translating to typologically diverse languages, training with low resource, and adapting to unseen domains. Experiment results show strong competitiveness of character-based models. Further analyses show that compared to subword-based models, character-based models are better at handling morphological phenomena, generating rare and unknown words, and more suitable for transferring to unseen domains.

NLP is currently dominated by language models like RoBERTa which are pretrained on billions of words. But what exact knowledge or skills do Transformer LMs learn from large-scale pretraining that they cannot learn from less data? To explore this question, we adopt five styles of evaluation: classifier probing, information-theoretic probing, unsupervised relative acceptability judgments, unsupervised language model knowledge probing, and fine-tuning on NLU tasks. We then draw learning curves that track the growth of these different measures of model ability with respect to pretraining data volume using the MiniBERTas, a group of RoBERTa models pretrained on 1M, 10M, 100M and 1B words. We find that these LMs require only about 10M to 100M words to learn to reliably encode most syntactic and semantic features we test. They need a much larger quantity of data in order to acquire enough commonsense knowledge and other skills required to master typical downstream NLU tasks. The results suggest that, while the ability to encode linguistic features is almost certainly necessary for language understanding, it is likely that other, unidentified, forms of knowledge are the major drivers of recent improvements in language understanding among large pretrained models.

Whereas much of the success of the current generation of neural language models has been driven by increasingly large training corpora, relatively little research has been dedicated to analyzing these massive sources of textual data. In this exploratory analysis, we delve deeper into the Common Crawl, a colossal web corpus that is extensively used for training language models. We find that it contains a significant amount of undesirable content, including hate speech and sexually explicit content, even after filtering procedures. We discuss the potential impacts of this content on language models and conclude with future research directions and a more mindful approach to corpus collection and analysis.

When evaluating an article and the claims it makes, a critical reader must be able to assess where the information presented comes from, and whether the various claims are mutually consistent and support the conclusion. This motivates the study of claim provenance, which seeks to trace and explain the origins of claims. In this paper, we introduce new techniques to model and reason about the provenance of multiple interacting claims, including how to capture fine-grained information about the context. Our solution hinges on first identifying the sentences that potentially contain important external information. We then develop a query generator with our novel rank-aware cross attention mechanism, which aims at generating metadata for the source article, based on the context and the signals collected from a search engine. This establishes relevant search queries, and it allows us to obtain source article candidates for each identified sentence and propose an ILP based algorithm to infer the best sources. We experiment with a newly created evaluation dataset, Politi-Prov, based on fact-checking articles from www.politifact.com; our experimental results show that our solution leads to a significant improvement over baselines.

According to the self-determination theory, the levels of satisfaction of three basic needs (competence, autonomy and relatedness) have implications on people’s everyday life and career. We benchmark the novel task of automatically detecting those needs on short posts in English, by modelling it as a ternary classification task, and as three binary classification tasks. A detailed manual analysis shows that the latter has advantages in the real-world scenario, and that our best models achieve similar performances as a trained human annotator.

Crowdsourcing is widely used to create data for common natural language understanding tasks. Despite the importance of these datasets for measuring and refining model understanding of language, there has been little focus on the crowdsourcing methods used for collecting the datasets. In this paper, we compare the efficacy of interventions that have been proposed in prior work as ways of improving data quality. We use multiple-choice question answering as a testbed and run a randomized trial by assigning crowdworkers to write questions under one of four different data collection protocols. We find that asking workers to write explanations for their examples is an ineffective stand-alone strategy for boosting NLU example difficulty. However, we find that training crowdworkers, and then using an iterative process of collecting data, sending feedback, and qualifying workers based on expert judgments is an effective means of collecting challenging data. But using crowdsourced, instead of expert judgments, to qualify workers and send feedback does not prove to be effective. We observe that the data from the iterative protocol with expert assessments is more challenging by several measures. Notably, the human–model gap on the unanimous agreement portion of this data is, on average, twice as large as the gap for the baseline protocol data.