EMNLP 2022 论文列表

Topic taxonomies display hierarchical topic structures of a text corpus and provide topical knowledge to enhance various NLP applications. To dynamically incorporate new topic information, several recent studies have tried to expand (or complete) a topic taxonomy by inserting emerging topics identified in a set of new documents. However, existing methods focus only on frequent terms in documents and the local topic-subtopic relations in a taxonomy, which leads to limited topic term coverage and fails to model the global taxonomy structure. In this work, we propose a novel framework for topic taxonomy expansion, named TopicExpan, which directly generates topic-related terms belonging to new topics. Specifically, TopicExpan leverages the hierarchical relation structure surrounding a new topic and the textual content of an input document for topic term generation. This approach encourages newly-inserted topics to further cover important but less frequent terms as well as to keep their relation consistency within the taxonomy. Experimental results on two real-world text corpora show that TopicExpan significantly outperforms other baseline methods in terms of the quality of output taxonomies.

Topic models are powerful tools to get an overview of large collections of text data, a situation that is prevalent in industry applications. A rising trend within topic modeling is to directly cluster dimension-reduced embeddings created with pretrained language models. It is difficult to evaluate these models because there is no ground truth and automatic measurements may not mimic human judgment. To address this problem, we created a tool called STELLAR for interactive topic browsing which we used for human evaluation of topics created from a real-world dataset used in industry. Embeddings created with BERT were used together with UMAP and HDBSCAN to model the topics. The human evaluation found that our topic model creates coherent topics. The following discussion revolves around the requirements of industry and what research is needed for production-ready systems.

Recent unsupervised topic modelling ap-proaches that use clustering techniques onword, token or document embeddings can ex-tract coherent topics. A common limitationof such approaches is that they reveal noth-ing about inter-topic relationships which areessential in many real-world application do-mains. We present an unsupervised topic mod-elling method which harnesses TopologicalData Analysis (TDA) to extract a topologicalskeleton of the manifold upon which contextu-alised word embeddings lie. We demonstratethat our approach, which performs on par witha recent baseline, is able to construct a networkof coherent topics together with meaningfulrelationships between them.

The predictions of question answering (QA) systems are typically evaluated against manually annotated finite sets of one or more answers. This leads to a coverage limitation that results in underestimating the true performance of systems, and is typically addressed by extending over exact match (EM) with predefined rules or with the token-level F1 measure.In this paper, we present the first systematic conceptual and data-driven analysis to examine the shortcomings of token-level equivalence measures.To this end, we define the asymmetric notion of answer equivalence (AE), accepting answers that are equivalent to or improve over the reference, and publish over 23k human judgements for candidates produced by multiple QA systems on SQuAD.Through a careful analysis of this data, we reveal and quantify several concrete limitations of the F1 measure, such as a false impression of graduality, or missing dependence on the question.Since collecting AE annotations for each evaluated model is expensive, we learn a BERT matching (BEM) measure to approximate this task. Being a simpler task than QA, we find BEM to provide significantly better AE approximations than F1, and to more accurately reflect the performance of systems.Finally, we demonstrate the practical utility of AE and BEM on the concrete application of minimal accurate prediction sets, reducing the number of required answers by up to X2.6.

End-to-end spoken language understanding (SLU) systems are gaining popularity over cascaded approaches due to their simplicity and ability to avoid error propagation. However, these systems model sequence labeling as a sequence prediction task causing a divergence from its well-established token-level tagging formulation. We build compositional end-to-end SLU systems that explicitly separate the added complexity of recognizing spoken mentions in SLU from the NLU task of sequence labeling. By relying on intermediate decoders trained for ASR, our end-to-end systems transform the input modality from speech to token-level representations that can be used in the traditional sequence labeling framework. This composition of ASR and NLU formulations in our end-to-end SLU system offers direct compatibility with pre-trained ASR and NLU systems, allows performance monitoring of individual components and enables the use of globally normalized losses like CRF, making them attractive in practical scenarios. Our models outperform both cascaded and direct end-to-end models on a labeling task of named entity recognition across SLU benchmarks.

Hate speech detection is complex; it relies on commonsense reasoning, knowledge of stereotypes, and an understanding of social nuance that differs from one culture to the next. It is also difficult to collect a large-scale hate speech annotated dataset. In this work, we frame this problem as a few-shot learning task, and show significant gains with decomposing the task into its “constituent” parts. In addition, we see that infusing knowledge from reasoning datasets (e.g. ATOMIC2020) improves the performance even further. Moreover, we observe that the trained models generalize to out-of-distribution datasets, showing the superiority of task decomposition and knowledge infusion compared to previously used methods. Concretely, our method outperforms the baseline by 17.83% absolute gain in the 16-shot case.

Event extraction (EE) is crucial to downstream tasks such as new aggregation and event knowledge graph construction. Most existing EE datasets manually define fixed event types and design specific schema for each of them, failing to cover diverse events emerging from the online text. Moreover, news titles, an important source of event mentions, have not gained enough attention in current EE research. In this paper, we present Title2Event, a large-scale sentence-level dataset benchmarking Open Event Extraction without restricting event types. Title2Event contains more than 42,000 news titles in 34 topics collected from Chinese web pages. To the best of our knowledge, it is currently the largest manually annotated Chinese dataset for open event extraction. We further conduct experiments on Title2Event with different models and show that the characteristics of titles make it challenging for event extraction, addressing the significance of advanced study on this problem. The dataset and baseline codes are available at https://open-event-hub.github.io/title2event.

News recommendation is a widely adopted technique to provide personalized news feeds for the user. Recently, pre-trained language models (PLMs) have demonstrated the great capability of natural language understanding and benefited news recommendation via improving news modeling. However, most existing works simply finetune the PLM with the news recommendation task, which may suffer from the known domain shift problem between the pre-training corpus and downstream news texts. Moreover, PLMs usually contain a large volume of parameters and have high computational overhead, which imposes a great burden on low-latency online services. In this paper, we propose Tiny-NewsRec, which can improve both the effectiveness and the efficiency of PLM-based news recommendation. We first design a self-supervised domain-specific post-training method to better adapt the general PLM to the news domain with a contrastive matching task between news titles and news bodies. We further propose a two-stage knowledge distillation method to improve the efficiency of the large PLM-based news recommendation model while maintaining its performance. Multiple teacher models originated from different time steps of our post-training procedure are used to transfer comprehensive knowledge to the student model in both its post-training stage and finetuning stage. Extensive experiments on two real-world datasets validate the effectiveness and efficiency of our method.

Adapter-tuning is a paradigm that transfers a pretrained language model to downstream tasks by adding and tuning a small number of new parameters. Previously proposed adapter architectures are all feed-forward neural networks. In this paper, we investigate the effectiveness of using tiny-attention—i.e., attention with extremely small per-head dimensionality—as adapters. Our tiny-attention adapter learns to modify the hidden states at each position directly conditioned on the hidden states at all the other positions, which is missed by the previously proposed adapters. Moreover, we view its multiple attention heads as a mixture of experts and propose to average their weights during deployment, which further reduces its inference computation cost. On the GLUE benchmark, our tiny-attention adapter outperforms the other parameter-efficient transfer learning methods as well as full fine-tuning while only updating 0.05% of the parameters. On the FewGLUE benchmark, its performance is comparable to that of GPT-3 and PET.

In this paper, we study the effects of incorporating timestamps, such as document creation dates, into generation systems. Two types of time-aware prompts are investigated: (1) textual prompts that encode document timestamps in natural language sentences; and (2) linear prompts that convert timestamps into continuous vectors. To explore extrapolation to future data points, we further introduce a new data-to-text generation dataset, TempWikiBio, containing more than 4 millions of chronologically ordered revisions of biographical articles from English Wikipedia, each paired with structured personal profiles.Through data-to-text generation on TempWikiBio, text-to-text generation on the content transfer dataset, and summarization on XSum,we show that linear prompts on encoder and textual prompts improve the generation quality on all datasets.Despite having less performance drop when testing on data drawn from a later time, linear prompts focus more on non-temporal information and are less sensitive to the given timestamps, according to human evaluations and sensitivity analyses.Meanwhile, textual prompts establish the association between the given timestamps and the output dates, yielding more factual temporal information in the output.

Patent Landscaping, one of the central tasks of intellectual property management, includes selecting and grouping patents according to user-defined technical or application-oriented criteria. While recent transformer-based models have been shown to be effective for classifying patents into taxonomies such as CPC or IPC, there is yet little research on how to support real-world Patent Landscape Studies (PLSs) using natural language processing methods. With this paper, we release three labeled datasets for PLS-oriented classification tasks covering two diverse domains. We provide a qualitative analysis and report detailed corpus statistics.Most research on neural models for patents has been restricted to leveraging titles and abstracts. We compare strong neural and non-neural baselines, proposing a novel model that takes into account textual information from the patents’ full texts as well as embeddings created based on the patents’ CPC labels. We find that for PLS-oriented classification tasks, going beyond title and abstract is crucial, CPC labels are an effective source of information, and combining all features yields the best results.

Word alignment is to find translationally equivalent words between source and target sentences. Previous work has demonstrated that self-training can achieve competitive word alignment results. In this paper, we propose to use word alignments generated by a third-party word aligner to supervise the neural word alignment training. Specifically, source word and target word of each word pair aligned by the third-party aligner are trained to be close neighbors to each other in the contextualized embedding space when fine-tuning a pre-trained cross-lingual language model. Experiments on the benchmarks of various language pairs show that our approach can surprisingly do self-correction over the third-party supervision by finding more accurate word alignments and deleting wrong word alignments, leading to better performance than various third-party word aligners, including the currently best one. When we integrate all supervisions from various third-party aligners, we achieve state-of-the-art word alignment performances, with averagely more than two points lower alignment error rates than the best third-party aligner.We released our code at https://github.com/sdongchuanqi/Third-Party-Supervised-Aligner.

Large pre-trained language models (PLMs) such as GPT-3 have shown strong in-context learning capabilities, which are highly appealing for domains such as biomedicine that feature high and diverse demands of language technologies but also high data annotation costs. In this paper, we present the first systematic and comprehensive study to compare the few-shot performance of GPT-3 in-context learning with fine-tuning smaller (i.e., BERT-sized) PLMs on two representative biomedical information extraction (IE) tasks: named entity recognition and relation extraction. We follow the true few-shot setting to avoid overestimating models’ few-shot performance by model selection over a large validation set. We also optimize GPT-3’s performance with known techniques such as contextual calibration and dynamic in-context example retrieval. However, our results show that GPT-3 still significantly underperforms compared to simply fine-tuning a smaller PLM. In addition, GPT-3 in-context learning also yields smaller gains in accuracy when more training data becomes available. More in-depth analyses further reveal issues of in-context learning that may be detrimental to IE tasks in general. Given the high cost of experimenting with GPT-3, we hope our study provides helpful guidance for biomedical researchers and practitioners towards more practical solutions such as fine-tuning small PLMs before better in-context learning is available for biomedical IE.

Commonsense knowledge has been widely considered for building intelligent open-domain dialogue agents, aiming to generate meaningful and diverse responses. Previous works in this field usually lack the ability to effectively obtain and utilize auxiliary commonsense from the external visual world. In this paper, we argue that exploiting logical information in images related to context can be effective to enrich and steer the generation process. In view of this, we propose VICTOR, a context-relevant VIsual Commonsense enhanced dialogue generaTOR for generating coherent and informative responses. To obtain the associated visual commonsense, we devise a novel approach that expands topic words on the knowledge graph and maps them into daily scenarios. During the generation, the model adopts multimodal fusion mechanism to integrate visual and textual information, and adaptively combine their decoding distributions for better response generation. The experimental results on two public datasets show that our proposed method outperforms the latest competitive methods in terms of coherence and diversity.

Knowledge-grounded dialogue (KGC) shows excellent potential to deliver an engaging and informative response. However, existing approaches emphasize selecting one golden knowledge given a particular dialogue context, overlooking the one-to-many phenomenon in dialogue. As a result, existing paradigm limits the diversity of knowledge selection and generation. To this end, we establish a multi-reference KGC dataset and propose a series of metrics to systematically assess the one-to-many efficacy of existing KGC models. Furthermore, to extend the hypothesis space of knowledge selection to enhance the mapping relationship between multiple knowledge and multiple responses, we devise a span-based variational model and optimize the model in a wake-sleep style with an ameliorated evidence lower bound objective to learn the one-to-many generalization. Both automatic and human evaluations demonstrate the efficacy of our approach.

Human variation in labeling is often considered noise. Annotation projects for machine learning (ML) aim at minimizing human label variation, with the assumption to maximize data quality and in turn optimize and maximize machine learning metrics. However, thisconventional practice assumes that there exists a *ground truth*, and neglects that there exists genuine human variation in labeling due to disagreement, subjectivity in annotation or multiple plausible answers.In this position paper, we argue that this big open problem of human label variation persists and critically needs more attention to move our field forward. This is because human label variation impacts all stages of the ML pipeline: *data, modeling and evaluation*. However, few works consider all of these dimensions jointly; and existing research is fragmented. We reconcile different previously proposed notions of human label variation, provide a repository of publicly-available datasets with un-aggregated labels, depict approaches proposed so far, identify gaps and suggest ways forward. As datasets are becoming increasingly available, we hope that this synthesized view on the “problem” will lead to an open discussion on possible strategies to devise fundamentally new directions.

Language use changes over time, and this impacts the effectiveness of NLP systems. This phenomenon is even more prevalent in social media data during crisis events where meaning and frequency of word usage may change over the course of days. Contextual language models fail to adapt temporally, emphasizing the need for temporal adaptation in models which need to be deployed over an extended period of time. While existing approaches consider data spanning large periods of time (from years to decades), shorter time spans are critical for crisis data. We quantify temporal degradation for this scenario and propose methods to cope with performance loss by leveraging techniques from domain adaptation. To the best of our knowledge, this is the first effort to explore effects of rapid language change driven by adversarial adaptations, particularly during natural and human-induced disasters. Through extensive experimentation on diverse crisis datasets, we analyze under what conditions our approaches outperform strong baselines while highlighting the current limitations of temporal adaptation methods in scenarios where access to unlabeled data is scarce.

Construction Grammar (CxG) is a paradigm from cognitive linguistics emphasising the connection between syntax and semantics. Rather than rules that operate on lexical items, it posits constructions as the central building blocks of language, i.e., linguistic units of different granularity that combine syntax and semantics. As a first step towards assessing the compatibility of CxG with the syntactic and semantic knowledge demonstrated by state-of-the-art pretrained language models (PLMs), we present an investigation of their capability to classify and understand one of the most commonly studied constructions, the English comparative correlative (CC). We conduct experiments examining the classification accuracy of a syntactic probe on the one hand and the models’ behaviour in a semantic application task on the other, with BERT, RoBERTa, and DeBERTa as the example PLMs. Our results show that all three investigated PLMs are able to recognise the structure of the CC but fail to use its meaning. While human-like performance of PLMs on many NLP tasks has been alleged, this indicates that PLMs still suffer from substantial shortcomings in central domains of linguistic knowledge.

Numerous works use word embedding-based metrics to quantify societal biases and stereotypes in texts. Recent studies have found that word embeddings can capture semantic similarity but may be affected by word frequency. In this work we study the effect of frequency when measuring female vs. male gender bias with word embedding-based bias quantification methods. We find that Skip-gram with negative sampling and GloVe tend to detect male bias in high frequency words, while GloVe tends to return female bias in low frequency words. We show these behaviors still exist when words are randomly shuffled. This proves that the frequency-based effect observed in unshuffled corpora stems from properties of the metric rather than from word associations. The effect is spurious and problematic since bias metrics should depend exclusively on word co-occurrences and not individual word frequencies. Finally, we compare these results with the ones obtained with an alternative metric based on Pointwise Mutual Information. We find that this metric does not show a clear dependence on frequency, even though it is slightly skewed towards male bias across all frequencies.