ACL 2023 论文列表

Intent detection, which estimates diverse intents behind user utterances, is an essential component of task-oriented dialogue systems. Previous intent detection models are usually trained offline, which can only handle predefined intent classes. In the real world, new intents may keep challenging deployed models. For example, with the prevalence of the COVID-19 pandemic, users may pose various issues related to the pandemic to conversational systems, which brings many new intents. A general intent detection model should be intelligent enough to continually learn new data and recognize new arriving intent classes. Therefore, this work explores Class Lifelong Learning for Intent Detection (CLL-ID), where the model continually learns new intent classes from new data while avoiding catastrophic performance degradation on old data. To this end, we propose a novel lifelong learning method, called Structure Consolidation Networks (SCN), which consists of structure-based retrospection and contrastive knowledge distillation to handle the problems of expression diversity and class imbalance in the CLL-ID task. In addition to formulating the new task, we construct 3 benchmarks based on 8 intent detection datasets. Experimental results demonstrate the effectiveness of SCN, which significantly outperforms previous lifelong learning methods on the three benchmarks.

Context is everything, even in commonsense moral reasoning. Changing contexts can flip the moral judgment of an action; Lying to a friend is wrong in general, but may be morally acceptable if it is intended to protect their life. We present ClarifyDelphi, an interactive system that learns to ask clarification questions (e.g., why did you lie to your friend?) in order to elicit additional salient contexts of a social or moral situation. We posit that questions whose potential answers lead to diverging moral judgments are the most informative. Thus, we propose a reinforcement learning framework with a defeasibility reward that aims to maximize the divergence between moral judgments of hypothetical answers to a question. Human evaluation demonstrates that our system generates more relevant, informative and defeasible questions compared to competitive baselines. Our work is ultimately inspired by studies in cognitive science that have investigated the flexibility in moral cognition (i.e., the diverse contexts in which moral rules can be bent), and we hope that research in this direction can assist both cognitive and computational investigations of moral judgments.

With the growing importance of detecting misinformation, many studies have focused on verifying factual claims by retrieving evidence. However, canonical fact verification tasks do not apply to catching subtle differences in factually consistent claims, which might still bias the readers, especially on contentious political or economic issues. Our underlying assumption is that among the trusted sources, one’s argument is not necessarily more true than the other, requiring comparison rather than verification. In this study, we propose ClaimDIff, a novel dataset that primarily focuses on comparing the nuance between claim pairs. In ClaimDiff, we provide human-labeled 2,941 claim pairs from 268 news articles. We observe that while humans are capable of detecting the nuances between claims, strong baselines struggle to detect them, showing over a 19% absolute gap with the humans. We hope this initial study could help readers to gain an unbiased grasp of contentious issues through machine-aided comparison.

We present Claim-Dissector: a novel latent variable model for fact-checking and analysis, which given a claim and a set of retrieved evidence jointly learns to identify: (i) the relevant evidences to the given claim (ii) the veracity of the claim. We propose to disentangle the per-evidence relevance probability and its contribution to the final veracity probability in an interpretable way — the final veracity probability is proportional to a linear ensemble of per-evidence relevance probabilities. In this way, the individual contributions of evidences towards the final predicted probability can be identified. In per-evidence relevance probability, our model can further distinguish whether each relevant evidence is supporting (S) or refuting (R) the claim. This allows to quantify how much the S/R probability contributes to final verdict or to detect disagreeing evidence. Despite its interpretable nature, our system achieves results competetive with state-of-the-art on the FEVER dataset, as compared to typical two-stage system pipelines, while using significantly fewer parameters. Furthermore, our analysis shows that our model can learn fine-grained relevance cues while using coarse-grained supervision and we demonstrate it in 2 ways. (i) We show that our model can achieve competitive sentence recall while using only paragraph-level relevance supervision. (ii) Traversing towards the finest granularity of relevance, we show that our model is capable of identifying relevance at the token level. To do this, we present a new benchmark TLR-FEVER focusing on token-level interpretability — humans annotate tokens in relevant evidences they considered essential when making their judgment. Then we measure how similar are these annotations to the tokens our model is focusing on.

In this paper, we introduce the benchmark datasets named CLUB (Chemical Language Understanding Benchmark) to facilitate NLP research in the chemical industry. We have 4 datasets consisted of text and token classification tasks. As far as we have recognized, it is one of the first examples of chemical language understanding benchmark datasets consisted of tasks for both patent and literature articles provided by industrial organization. All the datasets are internally made by chemists from scratch. Finally, we evaluate the datasets on the various language models based on BERT and RoBERTa, and demonstrate the model performs better when the domain of the pretrained models are closer to chemistry domain. We provide baselines for our benchmark as 0.8054 in average, and we hope this benchmark is used by many researchers in both industry and academia.

We present a new fact-checking benchmark, Check-COVID, that requires systems to verify claims about COVID-19 from news using evidence from scientific articles. This approach to fact-checking is particularly challenging as it requires checking internet text written in everyday language against evidence from journal articles written in formal academic language. Check-COVID contains 1, 504 expert-annotated news claims about the coronavirus paired with sentence-level evidence from scientific journal articles and veracity labels. It includes both extracted (journalist-written) and composed (annotator-written) claims. Experiments using both a fact-checking specific system and GPT-3.5, which respectively achieve F1 scores of 76.99 and 69.90 on this task, reveal the difficulty of automatically fact-checking both claim types and the importance of in-domain data for good performance. Our data and models are released publicly at https://github.com/posuer/Check-COVID.

Recent research on dialog state tracking (DST) focuses on methods that allow few- and zero-shot transfer to new domains or schemas. However, performance gains heavily depend on aggressive data augmentation and fine-tuning of ever larger language model based architectures. In contrast, general purpose language models, trained on large amounts of diverse data, hold the promise of solving any kind of task without task-specific training. We present preliminary experimental results on the ChatGPT research preview, showing that ChatGPT achieves state-of-the-art performance in zero-shot DST. Despite our findings, we argue that properties inherent to general purpose models limit their ability to replace specialized systems. We further theorize that the in-context learning capabilities of such models will likely become powerful tools to support the development of dedicated dialog state trackers and enable dynamic methods.

Building robust deep neural networks (DNNs) against adversarial attacks is an important but challenging task. Previous defense approaches mainly focus on developing new model structures or training algorithms, but they do little to tap the potential of training instances, especially instances with robust patterns carring innate robustness. In this paper, we show that robust and non-robust instances in the training dataset, though are both important for test performance, have contrary impacts on robustness, which makes it possible to build a highly robust model by leveraging the training dataset in a more effective way. We propose a new method that can distinguish between robust instances from non-robust ones according to the model’s sensitivity to perturbations on individual instances during training. Surprisingly, we find that the model under standard training easily overfits the robust instances by relying on their simple patterns before the model completely learns their robust features. Finally, we propose a new mitigation algorithm to further release the potential of robust instances. Experimental results show that proper use of robust instances in the original dataset is a new line to achieve highly robust models.

NLP models often degrade in performance when real world data distributions differ markedly from training data. However, existing dataset drift metrics in NLP have generally not considered specific dimensions of linguistic drift that affect model performance, and they have not been validated in their ability to predict model performance at the individual example level, where such metrics are often used in practice. In this paper, we propose three dimensions of linguistic dataset drift: vocabulary, structural, and semantic drift. These dimensions correspond to content word frequency divergences, syntactic divergences, and meaning changes not captured by word frequencies (e.g. lexical semantic change). We propose interpretable metrics for all three drift dimensions, and we modify past performance prediction methods to predict model performance at both the example and dataset level for English sentiment classification and natural language inference. We find that our drift metrics are more effective than previous metrics at predicting out-of-domain model accuracies (mean 16.8% root mean square error decrease), particularly when compared to popular fine-tuned embedding distances (mean 47.7% error decrease). Fine-tuned embedding distances are much more effective at ranking individual examples by expected performance, but decomposing into vocabulary, structural, and semantic drift produces the best example rankings of all considered model-agnostic drift metrics (mean 6.7% ROC AUC increase).

Widespread disparities in clinical outcomes exist between different demographic groups in the United States. A new line of work in medical sociology has demonstrated physicians often use stigmatizing language in electronic medical records within certain groups, such as black patients, which may exacerbate disparities. In this study, we characterize these instances at scale using a series of domain-informed NLP techniques. We highlight important differences between this task and analogous bias-related tasks studied within the NLP community (e.g., classifying microaggressions). Our study establishes a foundation for NLP researchers to contribute timely insights to a problem domain brought to the forefront by recent legislation regarding clinical documentation transparency. We release data, code, and models.

Current image generation models struggle to reliably produce well-formed visual text. In this paper, we investigate a key contributing factor: popular text-to-image models lack character-level input features, making it much harder to predict a word’s visual makeup as a series of glyphs. To quantify this effect, we conduct a series of experiments comparing character-aware vs. character-blind text encoders. In the text-only domain, we find that character-aware models provide large gains on a novel spelling task (WikiSpell). Applying our learnings to the visual domain, we train a suite of image generation models, and show that character-aware variants outperform their character-blind counterparts across a range of novel text rendering tasks (our DrawText benchmark). Our models set a much higher state-of-the-art on visual spelling, with 30+ point accuracy gains over competitors on rare words, despite training on far fewer examples.

Movie screenplays have a distinct narrative structure. It segments the story into scenes containing interleaving descriptions of actions, locations, and character dialogues.A typical screenplay spans several scenes and can include long-range dependencies between characters and events.A holistic document-level understanding of the screenplay requires several natural language processing capabilities, such as parsing, character identification, coreference resolution, action recognition, summarization, and attribute discovery. In this work, we develop scalable and robust methods to extract the structural information and character coreference clusters from full-length movie screenplays. We curate two datasets for screenplay parsing and character coreference — MovieParse and MovieCoref, respectively.We build a robust screenplay parser to handle inconsistencies in screenplay formatting and leverage the parsed output to link co-referring character mentions.Our coreference models can scale to long screenplay documents without drastically increasing their memory footprints.

BIG-Bench (Srivastava et al., 2022) is a diverse evaluation suite that focuses on tasks believed to be beyond the capabilities of current language models. Language models have already made good progress on this benchmark, with the best model in the BIG-Bench paper outperforming average reported human-rater results on 65% of the BIG-Bench tasks via few-shot prompting. But on what tasks do language models fall short of average human-rater performance, and are those tasks actually unsolvable by current language models? In this work, we focus on a suite of 23 challenging BIG-Bench tasks which we call BIG-Bench Hard (BBH). These are the tasks for which prior language model evaluations did not outperform the average human-rater. We find that applying chain-of-thought (CoT) prompting to BBH tasks enables PaLM to surpass the average human-rater performance on 10 of the 23 tasks, and Codex (code-davinci-002) to surpass the average human-rater performance on 17 of the 23 tasks. Since many tasks in BBH require multi-step reasoning, few-shot prompting without CoT, as done in the BIG-Bench evaluations (Srivastava et al., 2022), substantially underestimates the best performance and capabilities of language models, which is better captured via CoT prompting. As further analysis, we explore the interaction between CoT and model scale on BBH, finding that CoT enables emergent task performance on several BBH tasks with otherwise flat scaling curves.

The retrieval model is an indispensable component for real-world knowledge-intensive tasks, e.g., open-domain question answering (ODQA). As separate retrieval skills are annotated for different datasets, recent work focuses on customized methods, limiting the model transfer- ability and scalability. In this work, we propose a modular retriever where individual modules correspond to key skills that can be reused across datasets. Our approach supports flexible skill configurations based on the target domain to boost performance. To mitigate task interference, we design a novel modularization parameterization inspired by sparse Transformer. We demonstrate that our model can benefit from self-supervised pretraining on Wikipedia and fine-tuning using multiple ODQA datasets, both in a multi-task fashion. Our approach outperforms recent self-supervised retrievers in zero-shot evaluations and achieves state-of-the-art fine-tuned retrieval performance on NQ, HotpotQA and OTT-QA.

The knowledge-augmented deep learning paradigm refers to a paradigm in which domain knowledge is identified and integrated into deep models. Conventional methods typically employ task-specific approaches to gather external knowledge from various sources. In contrast, large language models are extensively pre-trained and can serve as a comprehensive source of external knowledge. In this paper, we propose CoT-KA, a Chain-of-Thought-based method that augments knowledge for deep learning. CoT-KA avoids the need for additional knowledge retrieval or knowledge reasoning models, as required in conventional augmentation methods. Our results demonstrate that CoT-KA outperforms both pure CoT-based methods and the non-augmented method across the majority of eleven publicly available benchmarks for various reasoning tasks.

Multilingual machine translation models can benefit from synergy between different language pairs, but also suffer from interference. While there is a growing number of sophisticated methods that aim to eliminate interference, our understanding of interference as a phenomenon is still limited. This work identifies the main factors that contribute to interference in multilingual machine translation. Through systematic experimentation, we find that interference (or synergy) are primarily determined by model size, data size, and the proportion of each language pair within the total dataset. We observe that substantial interference occurs mainly when the model is very small with respect to the available training data, and that using standard transformer configurations with less than one billion parameters largely alleviates interference and promotes synergy. Moreover, we show that tuning the sampling temperature to control the proportion of each language pair in the data is key to balancing the amount of interference between low and high resource language pairs effectively, and can lead to superior performance overall.

Concepts benefit natural language understanding but are far from complete in existing knowledge graphs (KGs). Recently, pre-trained language models (PLMs) have been widely used in text-based concept extraction (CE). However, PLMs tend to mine the co-occurrence associations from massive corpus as pre-trained knowledge rather than the real causal effect between tokens. As a result, the pre-trained knowledge confounds PLMs to extract biased concepts based on spurious co-occurrence correlations, inevitably resulting in low precision. In this paper, through the lens of a Structural Causal Model (SCM), we propose equipping the PLM-based extractor with a knowledge-guided prompt as an intervention to alleviate concept bias. The prompt adopts the topic of the given entity from the existing knowledge in KGs to mitigate the spurious co-occurrence correlations between entities and biased concepts. Our extensive experiments on representative multilingual KG datasets justify that our proposed prompt can effectively alleviate concept bias and improve the performance of PLM-based CE models.

Over the past few years, we’ve witnessed an enormous interest in stock price movement prediction using AI techniques. In recent literature, auxiliary data has been used to improve prediction accuracy, such as textual news. When predicting a particular stock, we assume that information from other stocks should also be utilized as auxiliary data to enhance performance. In this paper, we propose the Causality-guided Multi-memory Interaction Network (CMIN), a novel end-to-end deep neural network for stock movement prediction which, for the first time, models the multi-modality between financial text data and causality-enhanced stock correlations to achieve higher prediction accuracy. CMIN transforms the basic attention mechanism into Causal Attention by calculating transfer entropy between multivariate stocks in order to avoid attention on spurious correlations. Furthermore, we introduce a fusion mechanism to model the multi-directional interactions through which CMIN learns not only the self-influence but also the interactive influence in information flows representing the interrelationship between text and stock correlations. The effectiveness of the proposed approach is demonstrated by experiments on three real-world datasets collected from the U.S. and Chinese markets, where CMIN outperforms existing models to establish a new state-of-the-art prediction accuracy.

Despite their widespread adoption, neural conversation models have yet to exhibit natural chat capabilities with humans. In this research, we examine user utterances as causes and generated responses as effects, recognizing that changes in a cause should produce a different effect. To further explore this concept, we have compiled and expanded upon a new dataset called CausalDialogue through crowd-sourcing. This dataset includes multiple cause-effect pairs within a directed acyclic graph (DAG) structure. Our analysis reveals that traditional loss functions struggle to effectively incorporate the DAG structure, leading us to propose a causality-enhanced method called Exponential Maximum Average Treatment Effect (ExMATE) to enhance the impact of causality at the utterance level in training neural conversation models. To evaluate the needs of considering causality in dialogue generation, we built a comprehensive benchmark on CausalDialogue dataset using different models, inference, and training methods. Through experiments, we find that a causality-inspired loss like ExMATE can improve the diversity and agility of conventional loss function and there is still room for improvement to reach human-level quality on this new dataset.