EMNLP 2023 论文列表

Metaphor is a pervasive aspect of human communication, and its presence in multimodal forms has become more prominent with the progress of mass media. However, there is limited research on multimodal metaphor resources beyond the English language. Furthermore, the existing work in natural language processing does not address the exploration of categorizing the source and target domains in metaphors. This omission is significant considering the extensive research conducted in the fields of cognitive linguistics, which emphasizes that a profound understanding of metaphor relies on recognizing the differences and similarities between domain categories. We, therefore, introduce MultiCMET, a multimodal Chinese metaphor dataset, consisting of 13,820 text-image pairs of advertisements with manual annotations of the occurrence of metaphors, domain categories, and sentiments metaphors convey. We also constructed a domain lexicon that encompasses categorizations of metaphorical source domains and target domains and propose a Cascading Domain Knowledge Integration (CDKI) benchmark to detect metaphors by introducing domain-specific lexical features. Experimental results demonstrate the effectiveness of CDKI. The dataset and code are publicly available.

Inexhaustible web content carries abundant perceptible information beyond text. Unfortunately, most prior efforts in pre-trained Language Models (LMs) ignore such cyber-richness, while few of them only employ plain HTMLs, and crucial information in the rendered web, such as visual, layout, and style, are excluded. Intuitively, those perceptible web information can provide essential intelligence to facilitate content understanding tasks. This study presents an innovative Gestalt Enhanced Markup (GEM) Language Model inspired by Gestalt psychological theory for hosting heterogeneous visual information from the render tree into the language model without requiring additional visual input. Comprehensive experiments on multiple downstream tasks, i.e., web question answering and web information extraction, validate GEM superiority.

Several recent papers have published good solutions for language identification (LID) for about 300 high-resource and medium-resource languages. However, there is no LID available that (i) covers a wide range of low-resource languages, (ii) is rigorously evaluated and reliable and (iii) efficient and easy to use. Here, we publish GlotLID-M, an LID model that satisfies the desiderata of wide coverage, reliability and efficiency. It identifies 1665 languages, a large increase in coverage compared to prior work. In our experiments, GlotLID-M outperforms four baselines (CLD3, FT176, OpenLID and NLLB) when balancing F1 and false positive rate (FPR). We analyze the unique challenges that low-resource LID poses: incorrect corpus metadata, leakage from high-resource languages, difficulty separating closely related languages, handling of macrolanguage vs varieties and in general noisy data. We hope that integrating GlotLID-M into dataset creation pipelines will improve quality and enhance accessibility of NLP technology for low-resource languages and cultures. GlotLID-M model, code, and list of data sources are available: https://github.com/cisnlp/GlotLID.

In-context learning is a new learning paradigm where a language model observes a few examples and directly outputs the test input's prediction. Previous works have shown that it is sensitive to the provided examples and randomly sampled examples probably cause inferior performance. In this paper, we propose finding ``support examples'' for in-context learning: Given a training dataset, it aims to select one permutation of a few examples, which can well characterize the task for in-context learning and thus lead to superior performance. Although for traditional gradient-based training, there are extensive methods to find a coreset from the entire dataset, they struggle to find important in-context examples, because in-context learning occurs in the language model's forward process without gradients or parameter updates and thus has a significant gap with traditional training. Additionally, the strong dependence among in-context examples makes it an NP-hard combinatorial optimization problem and enumerating all permutations is infeasible. Hence we propose **LENS**, a fi**L**ter-th**EN**-**S**earch method to tackle this challenge in two stages: irst we filter the dataset to obtain individually informative in-context examples. Specifically, we propose a novel metric, InfoScore, to evaluate the example's in-context informativeness based on the language model's feedback, and further propose a progressive filtering process to filter out uninformative examples. Then we propose diversity-guided example search which iteratively refines and evaluates the selected example permutations, to find examples that fully depict the task. The experimental results show that LENS significantly outperforms a wide range of baselines and further analyses show that each component contribute critically to the improvements and shed light on the principles of supporting examples and in-context learning.

While many prior studies have applied computational approaches, such as machine learning, to detect and moderate hate speech, only scant attention has been paid to the task of identifying the underlying cause of hate speech. In this study, we introduce the concept of hate instigating speech, which refers to a specific type of textual posts on online platforms that stimulate or provoke others to engage in hate speech. The identification of hate instigating speech carries substantial practical implications for effective hate speech moderation. Rather than targeting individual instances of hate speech, by focusing on their roots, i.e., hate instigating speech, it becomes possible to significantly reduce the volume of content that requires review for moderation. Additionally, targeting hate instigating speech enables early prevention of the spread and propagation of hate speech, further enhancing the effectiveness of moderation efforts. However, several challenges hinder researchers from addressing the identification of hate instigating speech. First, there is a lack of comprehensive datasets specifically annotated for hate instigation, making it difficult to train and evaluate computational models effectively. Second, the subtle and nuanced nature of hate instigating speech (e.g., seemingly non-offensive texts serve as catalysts for triggering hate speech) makes it difficult to apply off-the-shelf machine learning models to the problem. To address these challenges, in this study, we have developed and released a multilingual dataset specifically designed for the task of identifying hate instigating speech. Specifically, it encompasses both English and Korean, allowing for a comprehensive examination of hate instigating speech across different linguistic contexts. We have applied existing machine learning models to our dataset and the results demonstrate that the extant models alone are insufficient for effectively detecting hate instigating speech. This finding highlights the need for further attention from the academic community to address this specific challenge. We expect our study and dataset to inspire researchers to explore innovative methods that can enhance the accuracy of hate instigating speech detection, ultimately contributing to more effective moderation and prevention of hate speech propagation online.

Open Information Extraction (OpenIE) is a traditional NLP task that extracts structured information from unstructured text to be used for other downstream applications. Traditionally, OpenIE focuses on extracting the surface forms of relations as they appear in the raw text, which we term extractive OpenIE. One of the main drawbacks of this approach is that implicit semantic relations (inferred relations) can not be extracted, compromising the performance of downstream applications. In this paper, we broaden the scope of OpenIE relations from merely the surface form of relations to include inferred relations, which we term abstractive OpenIE. This new task calls for the development of a new abstractive OpenIE training dataset and a baseline neural model that can extract those inferred relations. We also demonstrate the necessity for a new semantics-based metric for evaluating abstractive OpenIE extractions. Via a case study on Complex QA, we demonstrate the effectiveness of abstractive OpenIE.

The tremendous growth of social media users interacting in online conversations has led to significant growth in hate speech affecting people from various demographics. Most of the prior works focus on detecting explicit hate speech, which is overt and leverages hateful phrases, with very little work focusing on detecting hate speech that is implicit or denotes hatred through indirect or coded language. In this paper, we present CoSyn, a context synergized neural network that explicitly incorporates user- and conversational-context for detecting implicit hate speech in online conversations. CoSyn introduces novel ways to encode these external contexts and employs a novel context interaction mechanism that clearly captures the interplay between them, making independent assessments of the amounts of information to be retrieved from these noisy contexts. Additionally, it carries out all these operations in the hyperbolic space to account for the scale-free dynamics of social media. We demonstrate the effectiveness of CoSyn on 6 hate speech datasets and show that CoSyn outperforms all our baselines in detecting implicit hate speech with absolute improvements in the range of 1.24% - 57.8%. We make our code available.

Evaluating the performance of Grammatical Error Correction (GEC) systems is a challenging task due to its subjectivity. Designing an evaluation metric that is as objective as possible is crucial to the development of GEC task. However, mainstream evaluation metrics, i.e., reference-based metrics, introduce bias into the multi-reference evaluation by extracting edits without considering the presence of multiple references. To overcome this issue, we propose Chunk-LE Multi-reference Evaluation (CLEME), designed to evaluate GEC systems in the multi-reference evaluation setting. CLEME builds chunk sequences with consistent boundaries for the source, the hypothesis and references, thus eliminating the bias caused by inconsistent edit boundaries. Furthermore, we observe the consistent boundary could also act as the boundary of grammatical errors, based on which the F$_{0.5}$ score is then computed following the correction independence assumption. We conduct experiments on six English reference sets based on the CoNLL-2014 shared task. Extensive experiments and detailed analyses demonstrate the correctness of our discovery and the effectiveness of CLEME. Further analysis reveals that CLEME is robust to evaluate GEC systems across reference sets with varying numbers of references and annotation styles. All the source codes of CLEME are released at https://github.com/THUKElab/CLEME.

AI and NLP publication venues have increasingly encouraged researchers to reflect on possible ethical considerations, adverse impacts, and other responsible AI issues their work might engender. However, for specific NLP tasks our understanding of how prevalent such issues are, or when and why these issues are likely to arise, remains limited. Focusing on text summarization—a common NLP task largely overlooked by the responsible AI community—we examine research and reporting practices in the current literature. We conduct a multi-round qualitative analysis of 333 summarization papers from the ACL Anthology published between 2020–2022. We focus on how, which, and when responsible AI issues are covered, which relevant stakeholders are considered, and mismatches between stated and realized research goals. We also discuss current evaluation practices and consider how authors discuss the limitations of both prior work and their own work. Overall, we find that relatively few papers engage with possible stakeholders or contexts of use, which limits their consideration of potential downstream adverse impacts or other responsible AI issues. Based on our findings, we make recommendations on concrete practices and research directions.

In speech translation, leveraging multimodal data to improve model performance and address limitations of individual modalities has shown significant effectiveness. In this paper, we harness the complementary strengths of speech and text to improve speech translation. However, speech and text are disparate modalities, we observe three aspects of modality gap that impede their integration in a speech translation model. To tackle these gaps, we propose **Fuse**-**S**peech-**T**ext (**FuseST**), a cross-modal model which supports three distinct input modalities for translation: speech, text and fused speech-text. We leverage multiple techniques for cross-modal alignment and conduct a comprehensive analysis to assess its impact on speech translation, machine translation and fused speech-text translation. We evaluate FuseST on MuST-C, GigaST and newstest benchmark. Experiments show that the proposed FuseST achieves an average 34.0 BLEU on MuST-C En$\rightarrow$De/Es/Fr (vs SOTA +1.1 BLEU). Further experiments demonstrate that FuseST does not degrade on MT task, as observed in previous works. Instead, it yields an average improvement of 3.2 BLEU over the pre-trained MT model. Code is available at https://github.com/WenbiaoYin/FuseST.

To create a captivating story, a writer often plans a sequence of logically coherent events and ingeniously manipulates the narrative order to generate flashback in place. However, existing storytelling systems suffer from both insufficient understanding of event correlations and inadequate awareness of event temporal order (e.g., go to hospital <after> get ill), making it challenging to generate high-quality events that balance the logic and narrative order of story. In this paper, we propose a narrative order aware framework BPOT (Bidirectional Pretraining Model with Optimal Transport Reward) for story generation, which presents a bidirectional pretrained model to encode event correlations and pairwise event order. We also design a reinforcement learning algorithm with novel optimal transport reward to further improve the quality of generated events in the fine-tuning stage. Specifically, a narrative order aware event sequence model is pretrained with the joint learning objectives of event blank infilling and pairwise order prediction. Then, reinforcement learning with novel optimal transport reward is designed to further improve the generated event quality in the fine-tuning stage. The novel optimal transport reward captures the mappings between the generated events and the sentences in the story, effectively measuring the quality of generated events. Both automatic and manual evaluation results demonstrate the superiority of our framework in generating logically coherent stories with flashbacks.

Claim verification plays a crucial role in combating misinformation. While existing works on claim verification have shown promising results, a crucial piece of the puzzle that remains unsolved is to understand how to verify claims without relying on human-annotated data, which is expensive to create at a large scale. Additionally, it is important for models to provide comprehensive explanations that can justify their decisions and assist human fact-checkers. This paper presents First-Order-Logic-Guided Knowledge-Grounded (FOLK) Reasoning that can verify complex claims and generate explanations without the need for annotated evidence using Large Language Models (LLMs). FOLK leverages the in-context learning ability of LLMs to translate the claim into a First-Order-Logic (FOL) clause consisting of predicates, each corresponding to a sub-claim that needs to be verified. Then, FOLK performs FOL-Guided reasoning over a set of knowledge-grounded question-and-answer pairs to make veracity predictions and generate explanations to justify its decision-making process. This process makes our model highly explanatory, providing clear explanations of its reasoning process in human-readable form. Our experiment results indicate that FOLK outperforms strong baselines on three datasets encompassing various claim verification challenges. Our code and data are available.

Unlike the Open Domain Question Answering (ODQA) setting, the conversational (ODConvQA) domain has received limited attention when it comes to reevaluating baselines for both efficiency and effectiveness. In this paper, we study the State-of-the-Art (SotA) Dense Passage Retrieval (DPR) retriever and Fusion-in-Decoder (FiD) reader pipeline, and show that it significantly underperforms when applied to ODConvQA tasks due to various limitations. We then propose and evaluate strong yet simple and efficient baselines, by introducing a fast reranking component between the retriever and the reader, and by performing targeted finetuning steps. Experiments on two ODConvQA tasks, namely TopiOCQA and OR-QuAC, show that our method improves the SotA results, while reducing reader's latency by 60%. Finally, we provide new and valuable insights into the development of challenging baselines that serve as a reference for future, more intricate approaches, including those that leverage Large Language Models (LLMs).

Feature attribution scores are used for explaining the prediction of a text classifier to users by highlighting a k number of tokens. In this work, we propose a way to determine the number of optimal k tokens that should be displayed from sequential properties of the attribution scores. Our approach is dynamic across sentences, method-agnostic, and deals with sentence length bias. We compare agreement between multiple methods and humans on an NLI task, using fixed k and dynamic k. We find that perturbation-based methods and Vanilla Gradient exhibit highest agreement on most method--method and method--human agreement metrics with a static k. Their advantage over other methods disappears with dynamic ks which mainly improve Integrated Gradient and GradientXInput. To our knowledge, this is the first evidence that sequential properties of attribution scores are informative for consolidating attribution signals for human interpretation.

Online sentiment analysis has emerged as a crucial component in numerous data-driven applications, including social media monitoring, customer feedback analysis, and online reputation management. Despite their importance, current methodologies falter in effectively managing the continuously evolving nature of data streams, largely due to their reliance on substantial, pre-existing labelled datasets. This paper presents $\textbf{sentistream}$, a novel co-training framework specifically designed for efficient sentiment analysis within dynamic data streams. Comprising unsupervised, semi-supervised, and stream merge modules, $\textbf{ sentistream}$ guarantees constant adaptability to evolving data landscapes. This research delves into the continuous adaptation of language models for online sentiment analysis, focusing on real-world applications. Experimental evaluations using data streams derived from three benchmark sentiment analysis datasets confirm that our proposed methodology surpasses existing approaches in terms of both accuracy and computational efficiency.

Few-shot relation extraction (FSRE) aims to train a model that can deal with new relations using only a few labeled examples. Most existing studies employ Prototypical Networks for FSRE, which usually overfits the relation classes in the training set and cannot generalize well to unseen relations. By investigating the class separation of an FSRE model, we find that model upper layers are prone to learn relation-specific knowledge. Therefore, in this paper, we propose a HyperNetwork-based Decoupling approach to improve the generalization of FSRE models. Specifically, our model consists of an encoder, a network generator (for producing relation classifiers) and the produced-then-finetuned classifiers for every N-way-K-shot episode. Meanwhile, we design a two-step training framework along with a class-agnostic aligner, in which the generated classifiers focus on acquiring relation-specific knowledge and the encoder is encouraged to learn more general relation knowledge. In this way, the roles of upper and lower layers in an FSRE model are explicitly decoupled, thus enhancing its generalizing capability during testing. Experiments on two public datasets demonstrate the effectiveness of our method.

Modelling how concepts are related is a central topic in Lexical Semantics. A common strategy is to rely on knowledge graphs (KGs) such as ConceptNet, and to model the relation between two concepts as a set of paths. However, KGs are limited to a fixed set of relation types, and they are incomplete and often noisy. Another strategy is to distill relation embeddings from a fine-tuned language model. However, this is less suitable for words that are only indirectly related and it does not readily allow us to incorporate structured domain knowledge. In this paper, we aim to combine the best of both worlds. We model relations as paths but associate their edges with relation embeddings. The paths are obtained by first identifying suitable intermediate words and then selecting those words for which informative relation embeddings can be obtained. We empirically show that our proposed representations are useful for solving hard analogy questions.

Current language models are mainly trained on snap-shots of data gathered at a particular time, which decreases their capability to generalize over time and model language change. To model the \textit{time} variable, existing works have explored temporal language models (e.g., TempoBERT) by directly incorporating the timestamp into the training process. While effective to some extent, these methods are limited by the superficial temporal information brought by timestamps, which fails to learn the inherent changes of linguistic components. In this paper, we empirically confirm that the performance of pre-trained language models (PLMs) is closely affiliated with syntactically changed tokens. Based on this observation, we propose a simple yet effective method named \textit{\textbf{S}yntax-\textbf{G}uided \textbf{T}emporal \textbf{L}anguage \textbf{M}odel}~(SG-TLM), which could learn the inherent language changes by capturing an intrinsic relationship between the \textit{time} prefix and the tokens with salient syntactic change. Experiments on two datasets and three tasks demonstrate that our model outperforms existing temporal language models in both memorization and generalization capabilities. Extensive results further confirm the effectiveness of our approach across different model frameworks, including both encoder-only and decoder-only models (e.g., LLaMA). Our code is available at \url{https://github.com/zhaochen0110/TempoLM}.

Retrieval augmentation enhances generative language models by retrieving informative exemplars relevant for output prediction. However, in realistic graph parsing problems where the output space is large and complex, classic retrieval methods based on input-sentence similarity can fail to identify the most informative exemplars that target graph elements the model is most struggling about, leading to suboptimal retrieval and compromised prediction under limited retrieval budget. In this work, we improve retrieval-augmented parsing for complex graph problems by exploiting two unique sources of information (1) structural similarity and (2) model uncertainty. We propose $\textit{\textbf{S}tructure-aware and \textbf{U}ncertainty-\textbf{G}uided \textbf{A}daptive \textbf{R}etrieval} \textbf{(SUGAR)} $ that first quantify the model uncertainty in graph prediction and identify its most uncertain subgraphs, and then retrieve exemplars based on their structural similarity with the identified uncertain subgraphs. On a suite of real-world parsing benchmarks with non-trivial graph structure (SMCalflow and E-commerce), SUGAR exhibits a strong advantage over its classic counterparts that do not leverage structure or model uncertainty.