ACL 2022 论文列表

Zero-shot stance detection (ZSSD) aims to detect the stance for an unseen target during the inference stage. In this paper, we propose a joint contrastive learning (JointCL) framework, which consists of stance contrastive learning and target-aware prototypical graph contrastive learning. Specifically, a stance contrastive learning strategy is employed to better generalize stance features for unseen targets. Further, we build a prototypical graph for each instance to learn the target-based representation, in which the prototypes are deployed as a bridge to share the graph structures between the known targets and the unseen ones. Then a novel target-aware prototypical graph contrastive learning strategy is devised to generalize the reasoning ability of target-based stance representations to the unseen targets. Extensive experiments on three benchmark datasets show that the proposed approach achieves state-of-the-art performance in the ZSSD task.

Existing question answering (QA) techniques are created mainly to answer questions asked by humans. But in educational applications, teachers often need to decide what questions they should ask, in order to help students to improve their narrative understanding capabilities. We design an automated question-answer generation (QAG) system for this education scenario: given a story book at the kindergarten to eighth-grade level as input, our system can automatically generate QA pairs that are capable of testing a variety of dimensions of a student’s comprehension skills. Our proposed QAG model architecture is demonstrated using a new expert-annotated FairytaleQA dataset, which has 278 child-friendly storybooks with 10,580 QA pairs. Automatic and human evaluations show that our model outperforms state-of-the-art QAG baseline systems. On top of our QAG system, we also start to build an interactive story-telling application for the future real-world deployment in this educational scenario.

The recent success of distributed word representations has led to an increased interest in analyzing the properties of their spatial distribution. Several studies have suggested that contextualized word embedding models do not isotropically project tokens into vector space. However, current methods designed to measure isotropy, such as average random cosine similarity and the partition score, have not been thoroughly analyzed and are not appropriate for measuring isotropy. We propose IsoScore: a novel tool that quantifies the degree to which a point cloud uniformly utilizes the ambient vector space. Using rigorously designed tests, we demonstrate that IsoScore is the only tool available in the literature that accurately measures how uniformly distributed variance is across dimensions in vector space. Additionally, we use IsoScore to challenge a number of recent conclusions in the NLP literature that have been derived using brittle metrics of isotropy. We caution future studies from using existing tools to measure isotropy in contextualized embedding space as resulting conclusions will be misleading or altogether inaccurate.

Modern neural language models can produce remarkably fluent and grammatical text. So much, in fact, that recent work by Clark et al. (2021) has reported that conventional crowdsourcing can no longer reliably distinguish between machine-authored (GPT-3) and human-authored writing. As errors in machine generations become ever subtler and harder to spot, it poses a new challenge to the research community for robust machine text evaluation. We propose a new framework called Scarecrow for scrutinizing machine text via crowd annotation. To support the broad range of real machine errors that can be identified by laypeople, the ten error categories of Scarecrow—such as redundancy, commonsense errors, and incoherence—are identified through several rounds of crowd annotation experiments without a predefined ontology. We then use Scarecrow to collect over 41k error spans in human-written and machine-generated paragraphs of English language news text. We isolate factors for detailed analysis, including parameter count, training data, and various decoding-time configurations. Our approach successfully quantifies measurable gaps between human authored text and generations from models of several sizes, including fourteen configurations of GPT-3. In addition, our analysis unveils new insights, with detailed rationales provided by laypeople, e.g., that the commonsense capabilities have been improving with larger models while math capabilities have not, and that the choices of simple decoding hyperparameters can make remarkable differences on the perceived quality of machine text. We release our training material, annotation toolkit and dataset at https://yao-dou.github.io/scarecrow/.

The performance of deep learning models in NLP and other fields of machine learning has led to a rise in their popularity, and so the need for explanations of these models becomes paramount. Attention has been seen as a solution to increase performance, while providing some explanations. However, a debate has started to cast doubt on the explanatory power of attention in neural networks. Although the debate has created a vast literature thanks to contributions from various areas, the lack of communication is becoming more and more tangible. In this paper, we provide a clear overview of the insights on the debate by critically confronting works from these different areas. This holistic vision can be of great interest for future works in all the communities concerned by this debate. We sum up the main challenges spotted in these areas, and we conclude by discussing the most promising future avenues on attention as an explanation.

Creating chatbots to behave like real people is important in terms of believability. Errors in general chatbots and chatbots that follow a rough persona have been studied, but those in chatbots that behave like real people have not been thoroughly investigated. We collected a large amount of user interactions of a generation-based chatbot trained from large-scale dialogue data of a specific character, i.e., target person, and analyzed errors related to that person. We found that person-specific errors can be divided into two types: errors in attributes and those in relations, each of which can be divided into two levels: self and other. The correspondence with an existing taxonomy of errors was also investigated, and person-specific errors that should be addressed in the future were clarified.

In order to equip NLP systems with ‘selective prediction’ capability, several task-specific approaches have been proposed. However, which approaches work best across tasks or even if they consistently outperform the simplest baseline MaxProb remains to be explored. To this end, we systematically study selective prediction in a large-scale setup of 17 datasets across several NLP tasks. Through comprehensive experiments under in-domain (IID), out-of-domain (OOD), and adversarial (ADV) settings, we show that despite leveraging additional resources (held-out data/computation), none of the existing approaches consistently and considerably outperforms MaxProb in all three settings. Furthermore, their performance does not translate well across tasks. For instance, Monte-Carlo Dropout outperforms all other approaches on Duplicate Detection datasets but does not fare well on NLI datasets, especially in the OOD setting. Thus, we recommend that future selective prediction approaches should be evaluated across tasks and settings for reliable estimation of their capabilities.

Thanks to the strong representation power of neural encoders, neural chart-based parsers have achieved highly competitive performance by using local features. Recently, it has been shown that non-local features in CRF structures lead to improvements. In this paper, we investigate injecting non-local features into the training process of a local span-based parser, by predicting constituent n-gram non-local patterns and ensuring consistency between non-local patterns and local constituents. Results show that our simple method gives better results than the self-attentive parser on both PTB and CTB. Besides, our method achieves state-of-the-art BERT-based performance on PTB (95.92 F1) and strong performance on CTB (92.31 F1). Our parser also outperforms the self-attentive parser in multi-lingual and zero-shot cross-domain settings.

Transformer-based models are the modern work horses for neural machine translation (NMT), reaching state of the art across several benchmarks. Despite their impressive accuracy, we observe a systemic and rudimentary class of errors made by current state-of-the-art NMT models with regards to translating from a language that doesn’t mark gender on nouns into others that do. We find that even when the surrounding context provides unambiguous evidence of the appropriate grammatical gender marking, no tested model was able to accurately gender occupation nouns systematically. We release an evaluation scheme and dataset for measuring the ability of NMT models to translate gender morphology correctly in unambiguous contexts across syntactically diverse sentences. Our dataset translates from an English source into 20 languages from several different language families. With the availability of this dataset, our hope is that the NMT community can iterate on solutions for this class of especially egregious errors.

Current practices in metric evaluation focus on one single dataset, e.g., Newstest dataset in each year’s WMT Metrics Shared Task. However, in this paper, we qualitatively and quantitatively show that the performances of metrics are sensitive to data. The ranking of metrics varies when the evaluation is conducted on different datasets. Then this paper further investigates two potential hypotheses, i.e., insignificant data points and the deviation of i.i.d assumption, which may take responsibility for the issue of data variance. In conclusion, our findings suggest that when evaluating automatic translation metrics, researchers should take data variance into account and be cautious to report the results on unreliable datasets, because it may leads to inconsistent results with most of the other datasets.

Prompting methods recently achieve impressive success in few-shot learning. These methods modify input samples with prompt sentence pieces, and decode label tokens to map samples to corresponding labels. However, such a paradigm is very inefficient for the task of slot tagging. Since slot tagging samples are multiple consecutive words in a sentence, the prompting methods have to enumerate all n-grams token spans to find all the possible slots, which greatly slows down the prediction. To tackle this, we introduce an inverse paradigm for prompting. Different from the classic prompts mapping tokens to labels, we reversely predict slot values given slot types. Such inverse prompting only requires a one-turn prediction for each slot type and greatly speeds up the prediction. Besides, we propose a novel Iterative Prediction Strategy, from which the model learns to refine predictions by considering the relations between different slot types. We find, somewhat surprisingly, the proposed method not only predicts faster but also significantly improves the effect (improve over 6.1 F1-scores on 10-shot setting) and achieves new state-of-the-art performance.

Modern Natural Language Processing (NLP) models are known to be sensitive to input perturbations and their performance can decrease when applied to real-world, noisy data. However, it is still unclear why models are less robust to some perturbations than others. In this work, we test the hypothesis that the extent to which a model is affected by an unseen textual perturbation (robustness) can be explained by the learnability of the perturbation (defined as how well the model learns to identify the perturbation with a small amount of evidence). We further give a causal justification for the learnability metric. We conduct extensive experiments with four prominent NLP models — TextRNN, BERT, RoBERTa and XLNet — over eight types of textual perturbations on three datasets. We show that a model which is better at identifying a perturbation (higher learnability) becomes worse at ignoring such a perturbation at test time (lower robustness), providing empirical support for our hypothesis.

Character-level information is included in many NLP models, but evaluating the information encoded in character representations is an open issue. We leverage perceptual representations in the form of shape, sound, and color embeddings and perform a representational similarity analysis to evaluate their correlation with textual representations in five languages. This cross-lingual analysis shows that textual character representations correlate strongly with sound representations for languages using an alphabetic script, while shape correlates with featural scripts. We further develop a set of probing classifiers to intrinsically evaluate what phonological information is encoded in character embeddings. Our results suggest that information on features such as voicing are embedded in both LSTM and transformer-based representations.

Research Replication Prediction (RRP) is the task of predicting whether a published research result can be replicated or not. Building an interpretable neural text classifier for RRP promotes the understanding of why a research paper is predicted as replicable or non-replicable and therefore makes its real-world application more reliable and trustworthy. However, the prior works on model interpretation mainly focused on improving the model interpretability at the word/phrase level, which are insufficient especially for long research papers in RRP. Furthermore, the existing methods cannot utilize a large size of unlabeled dataset to further improve the model interpretability. To address these limitations, we aim to build an interpretable neural model which can provide sentence-level explanations and apply weakly supervised approach to further leverage the large corpus of unlabeled datasets to boost the interpretability in addition to improving prediction performance as existing works have done. In this work, we propose the Variational Contextual Consistency Sentence Masking (VCCSM) method to automatically extract key sentences based on the context in the classifier, using both labeled and unlabeled datasets. Results of our experiments on RRP along with European Convention of Human Rights (ECHR) datasets demonstrate that VCCSM is able to improve the model interpretability for the long document classification tasks using the area over the perturbation curve and post-hoc accuracy as evaluation metrics.

Grammatical Error Correction (GEC) should not focus only on high accuracy of corrections but also on interpretability for language learning. However, existing neural-based GEC models mainly aim at improving accuracy, and their interpretability has not been explored.A promising approach for improving interpretability is an example-based method, which uses similar retrieved examples to generate corrections. In addition, examples are beneficial in language learning, helping learners understand the basis of grammatically incorrect/correct texts and improve their confidence in writing. Therefore, we hypothesize that incorporating an example-based method into GEC can improve interpretability as well as support language learners. In this study, we introduce an Example-Based GEC (EB-GEC) that presents examples to language learners as a basis for a correction result. The examples consist of pairs of correct and incorrect sentences similar to a given input and its predicted correction. Experiments demonstrate that the examples presented by EB-GEC help language learners decide to accept or refuse suggestions from the GEC output. Furthermore, the experiments also show that retrieved examples improve the accuracy of corrections.

The largest store of continually updating knowledge on our planet can be accessed via internet search. In this work we study giving access to this information to conversational agents. Large language models, even though they store an impressive amount of knowledge within their weights, are known to hallucinate facts when generating dialogue (Shuster et al., 2021); moreover, those facts are frozen in time at the point of model training. In contrast, we propose an approach that learns to generate an internet search query based on the context, and then conditions on the search results to finally generate a response, a method that can employ up-to-the-minute relevant information. We train and evaluate such models on a newly collected dataset of human-human conversations whereby one of the speakers is given access to internet search during knowledgedriven discussions in order to ground their responses. We find that search-query based access of the internet in conversation provides superior performance compared to existing approaches that either use no augmentation or FAISS-based retrieval (Lewis et al., 2020b).

The composition of richly-inflected words in morphologically complex languages can be a challenge for language learners developing literacy. Accordingly, Lane and Bird (2020) proposed a finite state approach which maps prefixes in a language to a set of possible completions up to the next morpheme boundary, for the incremental building of complex words. In this work, we develop an approach to morph-based auto-completion based on a finite state morphological analyzer of Plains Cree (nêhiyawêwin), showing the portability of the concept to a much larger, more complete morphological transducer. Additionally, we propose and compare various novel ranking strategies on the morph auto-complete output. The best weighting scheme ranks the target completion in the top 10 results in 64.9% of queries, and in the top 50 in 73.9% of queries.

Lexically constrained neural machine translation (NMT), which controls the generation of NMT models with pre-specified constraints, is important in many practical scenarios. Due to the representation gap between discrete constraints and continuous vectors in NMT models, most existing works choose to construct synthetic data or modify the decoding algorithm to impose lexical constraints, treating the NMT model as a black box. In this work, we propose to open this black box by directly integrating the constraints into NMT models. Specifically, we vectorize source and target constraints into continuous keys and values, which can be utilized by the attention modules of NMT models. The proposed integration method is based on the assumption that the correspondence between keys and values in attention modules is naturally suitable for modeling constraint pairs. Experimental results show that our method consistently outperforms several representative baselines on four language pairs, demonstrating the superiority of integrating vectorized lexical constraints.

Self-attention heads are characteristic of Transformer models and have been well studied for interpretability and pruning. In this work, we demonstrate an altogether different utility of attention heads, namely for adversarial detection. Specifically, we propose a method to construct input-specific attention subnetworks (IAS) from which we extract three features to discriminate between authentic and adversarial inputs. The resultant detector significantly improves (by over 7.5%) the state-of-the-art adversarial detection accuracy for the BERT encoder on 10 NLU datasets with 11 different adversarial attack types. We also demonstrate that our method (a) is more accurate for larger models which are likely to have more spurious correlations and thus vulnerable to adversarial attack, and (b) performs well even with modest training sets of adversarial examples.