COLING 2022 论文列表

This paper describes a resource of Wikipedias in 31 languages categorized into Extended Named Entity (ENE), which has 219 fine-grained NE categories. We first categorized 920 K Japanese Wikipedia pages according to the ENE scheme using machine learning, followed by manual validation. We then organized a shared task of Wikipedia categorization into 30 languages. The training data were provided by Japanese categorization and the language links, and the task was to categorize the Wikipedia pages into 30 languages, with no language links from Japanese Wikipedia (20M pages in total). Thirteen groups with 24 systems participated in the 2020 and 2021 tasks, sharing their outputs for resource-building. The Japanese categorization accuracy was 98.5%, and the best performance among the 30 languages ranges from 80 to 93 in F-measure. Using ensemble learning, we created outputs with an average F-measure of 86.8, which is 1.7 better than the best single systems. The total size of the resource is 32.5M pages, including the training data. We call this resource creation scheme “Resource by Collaborative Contribution (RbCC)”. We also constructed structuring tasks (attribute extraction and link prediction) using RbCC under our ongoing project, “SHINRA.”

There is a lack of reproducibility in results from experiments that apply the Appraisal taxonomy. Appraisal is widely used by linguists to study how people judge things or people. Automating Appraisal could be beneficial for use cases such as moderating online comments. Past work in Appraisal annotation has been descriptive in nature and, the lack of publicly available data sets hinders the progress of automation. In this work, we are interested in two things; first, measuring the performance of automated approaches to Appraisal classification in the publicly available Australasian Language Technology Association (ALTA) Shared Task Challenge data set. Second, we are interested in reproducing the annotation of the ALTA data set. Four additional annotators, each with a different linguistics background, were employed to re-annotate the data set. Our results show a poor level of agreement at more detailed Appraisal categories (Fleiss Kappa = 0.059) and a fair level of agreement (Kappa = 0.372) at coarse-level categories. We find similar results when using automated approaches that are available publicly. Our empirical evidence suggests that at present, automating classification is practical only when considering coarse-level categories of the taxonomy.

Open-source platforms such as GitHub and Stack Overflow both play significant roles in current software ecosystems. It is crucial but time-consuming for developers to raise programming questions in coding forums such as Stack Overflow and be navigated to actual solutions on GitHub repositories. In this paper, we dedicate to accelerating this activity. We find that traditional information retrieval-based methods fail to handle the long and complex questions in coding forums, and thus cannot find suitable coding repositories. To effectively and efficiently bridge the semantic gap between repositories and real-world coding questions, we introduce a specialized dataset named Repo4QA, which includes over 12,000 question-repository pairs constructed from Stack Overflow and GitHub. Furthermore, we propose QuRep, a CodeBERT-based model that jointly learns the representation of both questions and repositories. Experimental results demonstrate that our model simultaneously captures the semantic features in both questions and repositories through supervised contrastive loss and hard negative sampling. We report that our approach outperforms existing state-of-art methods by 3%-8% on MRR and 5%-8% on P@1.

Applying Reinforcement learning (RL) following maximum likelihood estimation (MLE) pre-training is a versatile method for enhancing neural machine translation (NMT) performance. However, recent work has argued that the gains produced by RL for NMT are mostly due to promoting tokens that have already received a fairly high probability in pre-training. We hypothesize that the large action space is a main obstacle to RL’s effectiveness in MT, and conduct two sets of experiments that lend support to our hypothesis. First, we find that reducing the size of the vocabulary improves RL’s effectiveness. Second, we find that effectively reducing the dimension of the action space without changing the vocabulary also yields notable improvement as evaluated by BLEU, semantic similarity, and human evaluation. Indeed, by initializing the network’s final fully connected layer (that maps the network’s internal dimension to the vocabulary dimension), with a layer that generalizes over similar actions, we obtain a substantial improvement in RL performance: 1.5 BLEU points on average.

Deep neural models have become the mainstream in answer selection, yielding state-of-the-art performance. However, these models tend to rely on spurious correlations between prediction labels and input features, which in general suffer from robustness and generalization. In this paper, we propose a novel Spurious Correlation reduction method to improve the robustness of the neural ANswer selection models (SCAN) from the sample and feature perspectives by removing the feature dependencies and language biases in answer selection. First, from the sample perspective, we propose a feature decorrelation module by learning a weight for each instance at the training phase to remove the feature dependencies and reduce the spurious correlations without prior knowledge of such correlations. Second, from the feature perspective, we propose a feature debiasing module with contrastive learning to alleviate the negative language biases (spurious correlations) and further improve the robustness of the AS models. Experimental results on three benchmark datasets show that SCAN achieves substantial improvements over strong baselines. For reproducibility, we will release our code and data upon the publication of this paper.

Speech information in a pretrained wav2vec2.0 model is usually leveraged through its encoder, which has at least 95M parameters, being not so suitable for small footprint Keyword Spotting. In this work, we show an efficient way of profiting from wav2vec2.0’s linguistic knowledge, by recycling the phonetic information encoded in its latent codebook, which has been typically thrown away after pretraining. We do so by transferring the codebook as weights for the latent bottleneck of a Keyword Spotting Perceiver, thus initializing such model with phonetic embeddings already. The Perceiver design relies on cross-attention between these embeddings and input data to generate better representations. Our method delivers accuracy gains compared to random initialization, at no latency costs. Plus, we show that the phonetic embeddings can easily be downsampled with k-means clustering, speeding up inference in 3.5 times at only slight accuracy penalties.

Grounding dialogue agents with knowledge documents has sparked increased attention in both academia and industry. Recently, a growing body of work is trying to build retrieval-based knowledge-grounded dialogue systems. While promising, these approaches require collecting pairs of dialogue context and the corresponding ground-truth knowledge sentences that contain the information regarding the dialogue context. Unfortunately, hand-labeling data to that end is time-consuming, and many datasets and applications lack such knowledge annotations. In this paper, we propose a reciprocal learning approach to jointly optimize a knowledge retriever and a response ranker for knowledge-grounded response retrieval without ground-truth knowledge labels. Specifically, the knowledge retriever uses the feedback from the response ranker as pseudo supervised signals of knowledge retrieval for updating its parameters, while the response ranker also receives the top-ranked knowledge sentences from knowledge retriever for optimization. Evaluation results on two public benchmarks show that our model can significantly outperform previous state-of-the-art methods.

Text-to-SQL has attracted attention from both the natural language processing and database communities because of its ability to convert the semantics in natural language into SQL queries and its practical application in building natural language interfaces to database systems. The major challenges in text-to-SQL lie in encoding the meaning of natural utterances, decoding to SQL queries, and translating the semantics between these two forms. These challenges have been addressed to different extents by the recent advances. However, there is still a lack of comprehensive surveys for this task. To this end, we review recent progress on text-to-SQL for datasets, methods, and evaluation and provide this systematic survey, addressing the aforementioned challenges and discussing potential future directions. We hope this survey can serve as quick access to existing work and motivate future research.

Intelligent medical services have attracted great research interests for providing automated medical consultation. However, the lack of corpora becomes a main obstacle to related research, particularly data from real scenarios. In this paper, we construct RealMedDial, a Chinese medical dialogue dataset based on real medical consultation. RealMedDial contains 2,637 medical dialogues and 24,255 utterances obtained from Chinese short-video clips of real medical consultations. We collected and annotated a wide range of meta-data with respect to medical dialogue including doctor profiles, hospital departments, diseases and symptoms for fine-grained analysis on language usage pattern and clinical diagnosis. We evaluate the performance of medical response generation, department routing and doctor recommendation on RealMedDial. Results show that RealMedDial are applicable to a wide range of NLP tasks with respect to medical dialogue.

The introduction of multimodal information and pretraining technique significantly improves entity recognition from visually-rich documents. However, most of the existing methods pay unnecessary attention to irrelevant regions of the current document while ignoring the potentially valuable information in related documents. To deal with this problem, this work proposes a cross-document semantic enhancement method, which consists of two modules: 1) To prevent distractions from irrelevant regions in the current document, we design a learnable attention mask mechanism, which is used to adaptively filter redundant information in the current document. 2) To further enrich the entity-related context, we propose a cross-document information awareness technique, which enables the model to collect more evidence across documents to assist in prediction. The experimental results on two documents understanding benchmarks covering eight languages demonstrate that our method outperforms the SOTA methods.

We present a corrected version of a subset of the FactBank data set. Previously published results on FactBank are no longer valid. We perform experiments on FactBank using multiple training paradigms, data smoothing techniques, and polarity classifiers. We argue that f-measure is an important alternative evaluation metric for factuality. We provide new state-of-the-art results for four corpora including FactBank. We perform an error analysis on Factbank combined with two similar corpora.

We examine the inducement of rare but severe errors in English-Chinese and Chinese-English in-domain neural machine translation by minimal deletion of source text with character-based models. By deleting a single character, we can induce severe translation errors. We categorize these errors and compare the results of deleting single characters and single words. We also examine the effect of training data size on the number and types of pathological cases induced by these minimal perturbations, finding significant variation. We find that deleting a word hurts overall translation score more than deleting a character, but certain errors are more likely to occur when deleting characters, with language direction also influencing the effect.

Temporal relation extraction aims to extract temporal relations between event pairs, which is crucial for natural language understanding. Few efforts have been devoted to capturing the global features. In this paper, we propose RSGT: Relational Structure Guided Temporal Relation Extraction to extract the relational structure features that can fit for both inter-sentence and intra-sentence relations. Specifically, we construct a syntactic-and-semantic-based graph to extract relational structures. Then we present a graph neural network based model to learn the representation of this graph. After that, an auxiliary temporal neighbor prediction task is used to fine-tune the encoder to get more comprehensive node representations. Finally, we apply a conflict detection and correction algorithm to adjust the wrongly predicted labels. Experiments on two well-known datasets, MATRES and TB-Dense, demonstrate the superiority of our method (2.3% F1 improvement on MATRES, 3.5% F1 improvement on TB-Dense).

Question generation is the task of automatically generating questions based on given context and answers, and there are problems that the types of questions and answers do not match. In minority languages such as Tibetan, since the grammar rules are complex and the training data is small, the related research on question generation is still in its infancy. To solve the above problems, this paper constructs a question type classifier and a question generator. We perform fine-grained division of question types and integrate grammatical knowledge into question type classifiers to improve the accuracy of question types. Then, the types predicted by the question type classifier are fed into the question generator. Our model improves the accuracy of interrogative words in generated questions, and the BLEU-4 on SQuAD reaches 17.52, the BLEU-4 on HotpotQA reaches 19.31, the BLEU-4 on TibetanQA reaches 25.58.

Natural language generation (NLG) models can propagate social bias towards particular demography. Though several studies investigated bias from data and model, NLG task distinctively uses stochastic decoder that can positively or negatively impact the bias-sensitive tokens initially predicted by the model. To address this gap in research, we present an extensive analysis of bias from decoding techniques for open-domain language generation considering the entire decoding space. We analyze to what extent bias metrics like toxicity and sentiment are impacted by the individual components of decoder algorithms. To this extent, we also analyze the trade-off between bias scores and human-annotated generation quality throughout the decoder space. Together, these methods reveal the imperative of testing inference time bias and provide evidence on the usefulness of inspecting the entire decoding spectrum.

With the recent advance in neural machine translation demonstrating its importance, research on quality estimation (QE) has been steadily progressing. QE aims to automatically predict the quality of machine translation (MT) output without reference sentences. Despite its high utility in the real world, there remain several limitations concerning manual QE data creation: inevitably incurred non-trivial costs due to the need for translation experts, and issues with data scaling and language expansion. To tackle these limitations, we present QUAK, a Korean-English synthetic QE dataset generated in a fully automatic manner. This consists of three sub-QUAK datasets QUAK-M, QUAK-P, and QUAK-H, produced through three strategies that are relatively free from language constraints. Since each strategy requires no human effort, which facilitates scalability, we scale our data up to 1.58M for QUAK-P, H and 6.58M for QUAK-M. As an experiment, we quantitatively analyze word-level QE results in various ways while performing statistical analysis. Moreover, we show that datasets scaled in an efficient way also contribute to performance improvements by observing meaningful performance gains in QUAK-M, P when adding data up to 1.58M.

While question generation (QG) has received significant focus in conversation modeling and text generation research, the problems of comparing questions and evaluation of QG models have remained inadequately addressed. Indeed, QG models continue to be evaluated using traditional measures such as BLEU, METEOR, and ROUGE scores which were designed for other text generation problems. We propose QSTS, a novel Question-Sensitive Text Similarity measure for comparing two questions by characterizing their target intent based on question class, named-entity, and semantic similarity information from the two questions. We show that QSTS addresses several shortcomings of existing measures that depend on n-gram overlap scores and obtains superior results compared to traditional measures on publicly-available QG datasets. We also collect a novel dataset SimQG, for enabling question similarity research in QG contexts. SimQG contains questions generated by state-of-the-art QG models along with human judgements on their relevance with respect to the passage context they were generated for as well as when compared to the given reference question. Using SimQG, we showcase the key aspect of QSTS that differentiates it from all existing measures. QSTS is not only able to characterize similarity between two questions, but is also able to score questions with respect to passage contexts. Thus QSTS is, to our knowledge, the first metric that enables the measurement of QG performance in a reference-free manner.

Controllable story generation is a challenging task in the field of NLP, which has attracted increasing research interest in recent years. However, most existing works generate a whole story conditioned on the appointed keywords or emotions, ignoring the psychological changes of the protagonist. Inspired by psychology theories, we introduce global psychological state chains, which include the needs and emotions of the protagonists, to help a story generation system create more controllable and well-planned stories. In this paper, we propose a Psychology-guided Controllable Story Generation System (PICS) to generate stories that adhere to the given leading context and desired psychological state chains for the protagonist. Specifically, psychological state trackers are employed to memorize the protagonist’s local psychological states to capture their inner temporal relationships. In addition, psychological state planners are adopted to gain the protagonist’s global psychological states for story planning. Eventually, a psychology controller is designed to integrate the local and global psychological states into the story context representation for composing psychology-guided stories. Automatic and manual evaluations demonstrate that PICS outperforms baselines, and each part of PICS shows effectiveness for writing stories with more consistent psychological changes.

Pre-trained Language Models (PLMs) have been applied in NLP tasks and achieve promising results. Nevertheless, the fine-tuning procedure needs labeled data of the target domain, making it difficult to learn in low-resource and non-trivial labeled scenarios. To address these challenges, we propose Prompt-based Text Entailment (PTE) for low-resource named entity recognition, which better leverages knowledge in the PLMs. We first reformulate named entity recognition as the text entailment task. The original sentence with entity type-specific prompts is fed into PLMs to get entailment scores for each candidate. The entity type with the top score is then selected as final label. Then, we inject tagging labels into prompts and treat words as basic units instead of n-gram spans to reduce time complexity in generating candidates by n-grams enumeration. Experimental results demonstrate that the proposed method PTE achieves competitive performance on the CoNLL03 dataset, and better than fine-tuned counterparts on the MIT Movie and Few-NERD dataset in low-resource settings.