COLING 2022 Papers

Emotion cause analysis (ECA) aims to extract emotion clauses and find the corresponding cause of the emotion. Existing methods adopt fine-tuning paradigm to solve certain types of ECA tasks. These task-specific methods have a deficiency of universality. And the relations among multiple objectives in one task are not explicitly modeled. Moreover, the relative position information introduced in most existing methods may make the model suffer from dataset bias. To address the first two problems, this paper proposes a universal prompt tuning method to solve different ECA tasks in the unified framework. As for the third problem, this paper designs a directional constraint module and a sequential learning module to ease the bias. Considering the commonalities among different tasks, this paper proposes a cross-task training method to further explore the capability of the model. The experimental results show that our method achieves competitive performance on the ECA datasets.

Fine-grained entity typing (FET) aims to deduce specific semantic types of the entity mentions in the text. Modern methods for FET mainly focus on learning what a certain type looks like. And few works directly model the type differences, that is, let models know the extent that which one type is different from others. To alleviate this problem, we propose a type-enriched hierarchical contrastive strategy for FET. Our method can directly model the differences between hierarchical types and improve the ability to distinguish multi-grained similar types. On the one hand, we embed type into entity contexts to make type information directly perceptible. On the other hand, we design a constrained contrastive strategy on the hierarchical structure to directly model the type differences, which can simultaneously perceive the distinguishability between types at different granularity. Experimental results on three benchmarks, BBN, OntoNotes, and FIGER show that our method achieves significant performance on FET by effectively modeling type differences.

Multi-hop question generation (QG) is the process of generating answer related questions, which requires aggregating multiple pieces of information and reasoning from different parts of the texts. This is opposed to single-hop QG which generates questions from sentences containing an answer in a given paragraph. Single-hop QG requires no reasoning or complexity, while multi-hop QG often requires logical reasoning to derive an answer related question, making it a dual task. Not enough research has been made on the multi-hop QG due to its complexity. Also, a question should be created using the question type and words related to the correct answer as a prompt so that multi-hop questions can get more information. In this view, we propose a new type-dependent prompt cycleQAG (cyclic question-answer-generation), with a cycle consistency loss in which QG and Question Answering (QA) are learnt in a cyclic manner. The novelty is that the cycle consistency loss uses the negative cross entropy to generate syntactically diverse questions that enable selecting different word representations. Empirical evaluation on the multi-hop dataset with automatic and human evaluation metrics outperforms the baseline model by about 10.38% based on ROUGE score.

Chinese Named Entity Recognition (NER) has continued to attract research attention. However, most existing studies only explore the internal features of the Chinese language but neglect other lingual modal features. Actually, as another modal knowledge of the Chinese language, English contains rich prompts about entities that can potentially be applied to improve the performance of Chinese NER. Therefore, in this study, we explore the bilingual enhancement for Chinese NER and propose a unified bilingual interaction module called the Adapted Cross-Transformers with Global Sparse Attention (ACT-S) to capture the interaction of bilingual information. We utilize a model built upon several different ACT-Ss to integrate the rich English information into the Chinese representation. Moreover, our model can learn the interaction of information between bilinguals (inter-features) and the dependency information within Chinese (intra-features). Compared with existing Chinese NER methods, our proposed model can better handle entities with complex structures. The English text that enhances the model is automatically generated by machine translation, avoiding high labour costs. Experimental results on four well-known benchmark datasets demonstrate the effectiveness and robustness of our proposed model.

Social media platforms host discussions about a wide variety of topics that arise everyday. Making sense of all the content and organising it into categories is an arduous task. A common way to deal with this issue is relying on topic modeling, but topics discovered using this technique are difficult to interpret and can differ from corpus to corpus. In this paper, we present a new task based on tweet topic classification and release two associated datasets. Given a wide range of topics covering the most important discussion points in social media, we provide training and testing data from recent time periods that can be used to evaluate tweet classification models. Moreover, we perform a quantitative evaluation and analysis of current general- and domain-specific language models on the task, which provide more insights on the challenges and nature of the task.

ICD coding is designed to assign the disease codes to electronic health records (EHRs) upon discharge, which is crucial for billing and clinical statistics. In an attempt to improve the effectiveness and efficiency of manual coding, many methods have been proposed to automatically predict ICD codes from clinical notes. However, most previous works ignore the decisive information contained in structured medical data in EHRs, which is hard to be captured from the noisy clinical notes. In this paper, we propose a Tree-enhanced Multimodal Attention Network (TreeMAN) to fuse tabular features and textual features into multimodal representations by enhancing the text representations with tree-based features via the attention mechanism. Tree-based features are constructed according to decision trees learned from structured multimodal medical data, which capture the decisive information about ICD coding. We can apply the same multi-label classifier from previous text models to the multimodal representations to predict ICD codes. Experiments on two MIMIC datasets show that our method outperforms prior state-of-the-art ICD coding approaches. The code is available at https://github.com/liu-zichen/TreeMAN.

Even though many recent semantic parsers are based on deep learning methods, we should not forget that rule-based alternatives might offer advantages over neural approaches with respect to transparency, portability, and explainability. Taking advantage of existing off-the-shelf Universal Dependency parsers, we present a method that maps a syntactic dependency tree to a formal meaning representation based on Discourse Representation Theory. Rather than using lambda calculus to manage variable bindings, our approach is novel in that it consists of using a series of graph transformations. The resulting UD semantic parser shows good performance for English, German, Italian and Dutch, with F-scores over 75%, outperforming a neural semantic parser for the lower-resourced languages. Unlike neural semantic parsers, our UD semantic parser does not hallucinate output, is relatively easy to port to other languages, and is completely transparent.

Semantic parsing considers the task of mapping a natural language sentence into a target formal representation, where various sophisticated sequence-to-sequence (seq2seq) models have been applied with promising results. Generally, these target representations follow a syntax formalism that limits permitted forms. However, it is neither easy nor flexible to explicitly integrate this syntax formalism into a neural seq2seq model. In this paper, we present a structure-aware self-attention language model to capture structural information of target representations and propose a knowledge distillation based approach to incorporating the target language model into a seq2seq model, where grammar rules or sketches are not required in the training process. An ablation study shows that the proposed language model can notably improve the performance of the baseline model. The experiments show that our method achieves new state-of-the-art performance among neural approaches on four semantic parsing (ATIS, GEO) and Python code generation (Django, CoNaLa) tasks.

Relevant to all application domains where it is important to get at the reasons underlying sentiments and decisions, argument mining seeks to obtain structured arguments from unstructured text and has been addressed by approaches typically involving some feature and/or neural architecture engineering. By adopting a transfer learning methodology, and by means of a systematic study with a wide range of knowledge sources promisingly suitable to leverage argument mining, the aim of this paper is to empirically assess the potential of transferring such knowledge learned with confluent tasks. By adopting a lean approach that dispenses with heavier feature and model engineering, this study permitted both to gain novel empirically based insights into the argument mining task and to establish new state of the art levels of performance for its three main sub-tasks, viz. identification of argument components, classification of the components, and determination of the relation among them.

Due to the increasing use of service chatbots in E-commerce platforms in recent years, customer satisfaction prediction (CSP) is gaining more and more attention. CSP is dedicated to evaluating subjective customer satisfaction in conversational service and thus helps improve customer service experience. However, previous methods focus on modeling customer-chatbot interaction across different turns, which are hard to represent the important dynamic satisfaction states throughout the customer journey. In this work, we investigate the problem of satisfaction states tracking and its effects on CSP in E-commerce service chatbots. To this end, we propose a dialogue-level classification model named DialogueCSP to track satisfaction states for CSP. In particular, we explore a novel two-step interaction module to represent the dynamic satisfaction states at each turn. In order to capture dialogue-level satisfaction states for CSP, we further introduce dialogue-aware attentions to integrate historical informative cues into the interaction module. To evaluate the proposed approach, we also build a Chinese E-commerce dataset for CSP. Experiment results demonstrate that our model significantly outperforms multiple baselines, illustrating the benefits of satisfaction states tracking on CSP.

In this paper, we explore the relation between gestures and language. Using a multimodal dataset, consisting of Ted talks where the language is aligned with the gestures made by the speakers, we adapt a semi-supervised multimodal model to learn gesture embeddings. We show that gestures are predictive of the native language of the speaker, and that gesture embeddings further improve language prediction result. In addition, gesture embeddings might contain some linguistic information, as we show by probing embeddings for psycholinguistic categories. Finally, we analyze the words that lead to the most expressive gestures and find that function words drive the expressiveness of gestures.

The current advancement in abstractive document summarization depends to a large extent on a considerable amount of human-annotated datasets. However, the creation of large-scale datasets is often not feasible in closed domains, such as medical and healthcare domains, where human annotation requires domain expertise. This paper presents a novel data selection strategy to generate diverse and semantic questions in a low-resource setting with the aim to summarize healthcare questions. Our method exploits the concept of guided semantic-overlap and diversity-based objective functions to optimally select the informative and diverse set of synthetic samples for data augmentation. Our extensive experiments on benchmark healthcare question summarization datasets demonstrate the effectiveness of our proposed data selection strategy by achieving new state-of-the-art results. Our human evaluation shows that our method generates diverse, fluent, and informative summarized questions.

Previous works have demonstrated the effectiveness of utilising pre-trained sentence encoders based on their sentence representations for meaning comparison tasks. Though such representations are shown to capture hidden syntax structures, the direct similarity comparison between them exhibits weak sensitivity to word order and structural differences in given sentences. A single similarity score further makes the comparison process hard to interpret. Therefore, we here propose to combine sentence encoders with an alignment component by representing each sentence as a list of predicate-argument spans (where their span representations are derived from sentence encoders), and decomposing the sentence-level meaning comparison into the alignment between their spans for paraphrase identification tasks. Empirical results show that the alignment component brings in both improved performance and interpretability for various sentence encoders. After closer investigation, the proposed approach indicates increased sensitivity to structural difference and enhanced ability to distinguish non-paraphrases with high lexical overlap.

Research on neural IR has so far been focused primarily on standard supervised learning settings, where it outperforms traditional term matching baselines. Many practical use cases of such models, however, may involve previously unseen target domains. In this paper, we propose to improve the out-of-domain generalization of Dense Passage Retrieval (DPR) - a popular choice for neural IR - through synthetic data augmentation only in the source domain. We empirically show that pre-finetuning DPR with additional synthetic data in its source domain (Wikipedia), which we generate using a fine-tuned sequence-to-sequence generator, can be a low-cost yet effective first step towards its generalization. Across five different test sets, our augmented model shows more robust performance than DPR in both in-domain and zero-shot out-of-domain evaluation.

Most existing methods on robust neural machine translation (NMT) construct adversarial examples by injecting noise into authentic examples and indiscriminately exploit two types of examples. They require the model to translate both the authentic source sentence and its adversarial counterpart into the identical target sentence within the same training stage, which may be a suboptimal choice to achieve robust NMT. In this paper, we first conduct a preliminary study to confirm this claim and further propose an Iterative Scheduled Data-switch Training Framework to mitigate this problem. Specifically, we introduce two training stages, iteratively switching between authentic and adversarial examples. Compared with previous studies, our model focuses more on just one type of examples at each single stage, which can better exploit authentic and adversarial examples, and thus obtaining a better robust NMT model. Moreover, we introduce an improved curriculum learning method with a sampling strategy to better schedule the process of noise injection. Experimental results show that our model significantly surpasses several competitive baselines on four translation benchmarks. Our source code is available at https://github.com/DeepLearnXMU/RobustNMT-ISDST.

Multi-class unknown intent detection has made remarkable progress recently. However, it has a strong assumption that each utterance has only one intent, which does not conform to reality because utterances often have multiple intents. In this paper, we propose a more desirable task, multi-label unknown intent detection, to detect whether the utterance contains the unknown intent, in which each utterance may contain multiple intents. In this task, the unique utterances simultaneously containing known and unknown intents make existing multi-class methods easy to fail. To address this issue, we propose an intuitive and effective method to recognize whether All Intents contained in the utterance are Known (AIK). Our high-level idea is to predict the utterance’s intent number, then check whether the utterance contains the same number of known intents. If the number of known intents is less than the number of intents, it implies that the utterance also contains unknown intents. We benchmark AIK over existing methods, and empirical results suggest that our method obtains state-of-the-art performances. For example, on the MultiWOZ 2.3 dataset, AIK significantly reduces the FPR95 by 12.25% compared to the best baseline.

Cross-lingual word embeddings (CLWE) have been proven useful in many cross-lingual tasks. However, most existing approaches to learn CLWE including the ones with contextual embeddings are sense agnostic. In this work, we propose a novel framework to align contextual embeddings at the sense level by leveraging cross-lingual signal from bilingual dictionaries only. We operationalize our framework by first proposing a novel sense-aware cross entropy loss to model word senses explicitly. The monolingual ELMo and BERT models pretrained with our sense-aware cross entropy loss demonstrate significant performance improvement for word sense disambiguation tasks. We then propose a sense alignment objective on top of the sense-aware cross entropy loss for cross-lingual model pretraining, and pretrain cross-lingual models for several language pairs (English to German/Spanish/Japanese/Chinese). Compared with the best baseline results, our cross-lingual models achieve 0.52%, 2.09% and 1.29% average performance improvements on zero-shot cross-lingual NER, sentiment classification and XNLI tasks, respectively.

The task of shallow discourse parsing in the Penn Discourse Treebank (PDTB) framework has traditionally been restricted to identifying those relations that are signaled by a discourse connective (“explicit”) and those that have no signal at all (“implicit”). The third type, the more flexible group of “AltLex” realizations has been neglected because of its small amount of occurrences in the PDTB2 corpus. Their number has grown significantly in the recent PDTB3, and in this paper, we present the first approaches for recognizing these “alternative lexicalizations”. We compare the performance of a pattern-based approach and a sequence labeling model, add an experiment on the pre-classification of candidate sentences, and provide an initial qualitative analysis of the error cases made by both models.

Sentiment analysis in social media is challenging since posts are short of context. As a popular way to express emotion on social media, stickers related to these posts can supplement missing sentiments and help identify sentiments precisely. However, research about stickers has not been investigated further. To this end, we present a Chinese sticker-based multimodal dataset for the sentiment analysis task (CSMSA). Compared with previous real-world photo-based multimodal datasets, the CSMSA dataset focuses on stickers, conveying more vivid and moving emotions. The sticker-based multimodal sentiment analysis task is challenging in three aspects: inherent multimodality of stickers, significant inter-series variations between stickers, and complex multimodal sentiment fusion. We propose SAMSAM to address the above three challenges. Our model introduces a flexible masked self-attention mechanism to allow the dynamic interaction between post texts and stickers. The experimental results indicate that our model performs best compared with other models. More researches need to be devoted to this field. The dataset is publicly available at https://github.com/Logos23333/CSMSA.