Chinese Computational Linguistics: 20th China National Conference, CCL 2021, Hohhot, China, August 13–15, 2021, Proceedings (Lecture Notes in Artificial Intelligence)

This book constitutes the proceedings of the 20th China National Conference on Computational Linguistics, CCL 2021, held in Hohhot, China, in August 2021. The 31 full presented in this volume were carefully reviewed and selected from 90 submissions. The conference papers covers the following topics such as Machine Translation and Multilingual Information Processing, Minority Language Information Processing, Social Computing and Sentiment Analysis, Text Generation and Summarization, Information Retrieval, Dialogue and Question Answering, Linguistics and Cognitive Science, Language Resource and Evaluation, Knowledge Graph and Information Extraction, and NLP Applications. Preface Organization Contents Machine Translation and Multilingual Information Processing Reducing Length Bias in Scoring Neural Machine Translation via a Causal Inference Method 1 Introduction 2 Related Work 3 Approach 3.1 Correcting Length Bias via Half-Sibling Regression 3.2 Re-scoring Translation Candidates 3.3 Discussion 4 Experiments 4.1 Datasets and Evaluation Metric 4.2 Length Normalization Baselines 4.3 Model Setups 4.4 Main Results 4.5 Performance on Wider Beam Size 5 Conclusion and Future Work References Low-Resource Machine Translation Based on Asynchronous Dynamic Programming 1 Introduction 2 Background 2.1 Neural Machine Translation 2.2 RL Based NMT 3 Approach 3.1 Priority Acquisition of Experience 3.2 Asynchronous Strategy 3.3 Training 4 Experiment and Analysis 4.1 Datasets and Preprocessing 4.2 Setting 4.3 Main Results and Analysis 5 Conclusion References Minority Language Information Processing Uyghur Metaphor Detection via Considering Emotional Consistency 1 Introduction 2 Related Work 2.1 Metaphor Detection 2.2 Uyghur Metaphor 3 Our Proposed Model 3.1 Basic Idea 3.2 Model Structure 4 Experiment 4.1 Dataset 4.2 Experimental Details 4.3 Result Analysis 5 Conclusion References Incorporating Translation Quality Estimation into Chinese-Korean Neural Machine Translation 1 Introduction 2 Related Work 3 Methodology 3.1 Model Overview 3.2 Generate Rewards Through Sentence-Level Quality Estimation 3.3 Reward Computation 3.4 The Training of Reinforcement Learning 4 Experiments 4.1 Datasets 4.2 Preprocessing 4.3 Setting 4.4 Main Results and Analysis 4.5 Performance Verification About QE 4.6 Example of Translation Results 5 Conclusion References Social Computing and Sentiment Analysis Emotion Classification of COVID-19 Chinese Microblogs Based on the Emotion Category Description 1 Introduction 2 Related Work 3 Methods 3.1 Definition and Strategy of Emotion Category Description 3.2 Emotion Classification Fine-Tuning Based on a Question Answering (QA) Method 4 Experiments 4.1 Experimental Dataset 4.2 Baseline Models 4.3 Implementation Details 4.4 Experimental Results 5 Conclusion References Multi-level Emotion Cause Analysis by Multi-head Attention Based Multi-task Learning 1 Introduction 2 Related Work 3 Methodology 3.1 Task Definition 3.2 Model Description 3.3 Training and Parameter Learning 4 Experiments and Results 4.1 Dataset and Settings 4.2 Comparison of Different Methods 4.3 Ablation Study 4.4 Case Study 4.5 Error Analysis 5 Conclusions References Text Generation and Summarization Using Query Expansion in Manifold Ranking for Query-Oriented Multi-document Summarization 1 Introduction 2 Related Work 3 Method 3.1 Manifold Ranking 3.2 Query Expansion 3.3 Sentence Similarity Calculation 4 Experimental Setup 4.1 Datasets and Evaluation 4.2 Parameter Settings 5 Results 5.1 Comparison with Query Expansion Methods 5.2 Comparison with Related Methods 5.3 Influence of Sentence Similarity Parameter Tuning 5.4 Query Relevance Performance 6 Conclusion References Jointly Learning Salience and Redundancy by Adaptive Sentence Reranking for Extractive Summarization 1 Introduction 2 Related Work 2.1 Salience Learning 2.2 Redundancy Learning 3 Method 3.1 Overall Architecture 3.2 Extraction Summarization as Ranking 4 Experiments 4.1 Datasets 4.2 Implementation Details 4.3 Evaluation Metric 4.4 Experimental Results 5 Qualitative Analysis 5.1 Ablation Study 5.2 Effect of Summary Length 5.3 Analysis of N-Gram Frequency 5.4 Human Evaluation 5.5 Case Study 6 Conclusions References Incorporating Commonsense Knowledge into Abstractive Dialogue Summarization via Heterogeneous Graph Networks 1 Introduction 2 Heterogeneous Dialogue Graph Construction 2.1 Graph Notation 2.2 Utterance-Knowledge Bipartite Graph Construction 2.3 Speaker-Utterance Bipartite Graph Construction 2.4 Heterogeneous Dialogue Graph Construction 3 Dialogue Heterogeneous Graph Network 3.1 Node Encoder 3.2 Graph Encoder 3.3 Pointer Decoder 3.4 Training 4 Experiments 4.1 Automatic Evaluation 4.2 Human Evaluation 4.3 Ablation Study 4.4 Zero-Shot Setting 4.5 Visualization 4.6 Case Study 5 Related Work 6 Conclusion References Information Retrieval, Dialogue and Question Answering Enhancing Question Generation with Commonsense Knowledge 1 Introduction 2 Related Work 3 Knowledge Extraction 4 Model Description 4.1 Multi-task Learning Framework 4.2 Iterative Training Framework 5 Experimental Settings 5.1 Dataset and Metrics 5.2 Baseline Models 5.3 Implementation Details 6 Results 6.1 Main Results 6.2 Ablation Study 6.3 Analysis of Auxiliary Tasks 6.4 Human Evaluation 6.5 Case Study 7 Conclusion References Topic Knowledge Acquisition and Utilization for Machine Reading Comprehension in Social Media Domain 1 Introduction 2 Related Work 3 Method 3.1 Knowledge Acquisition 3.2 Topic Knowledge Reader 4 Experiment 4.1 TweetQA Dataset 4.2 Implement Detail 4.3 Baselines 4.4 Main Results 4.5 Different Number of Concepts 4.6 Ablation Study 4.7 Extractive vs. Generative 4.8 Weakly Supervised Training 5 Conclusion References Category-Based Strategy-Driven Question Generator for Visual Dialogue 1 Introduction 2 Related Work 3 Proposed Model 3.1 The Design and Implement of Category Info to Questions 3.2 Multi-modality Encoder 3.3 Category Predictor 3.4 Generation Decoder 4 Experiments 4.1 Implementation Details on Guesser and Oracle 4.2 Contrast Experiments 4.3 Ablation Experiments 4.4 Good Case Study 5 Conclusions References From Learning-to-Match to Learning-to-Discriminate: Global Prototype Learning for Few-shot Relation Classification 1 Introduction 2 Background 3 Global Transformed Prototypical Networks 3.1 Instance Encoder 3.2 Relation Marker 3.3 Multi-view Global Transformation 3.4 Prototype-Based Classification 3.5 Model Learning 4 Experiments 4.1 Experimental Settings 4.2 Overall Results 4.3 Detailed Analysis 5 Related Work 6 Conclusions References Multi-strategy Knowledge Distillation Based Teacher-Student Framework for Machine Reading Comprehension 1 Introduction 2 Related Work 2.1 Machine Reading Comprehension 2.2 Knowledge Distillation 3 Methods 3.1 MRC Model 3.2 Knowledge Distillation Strategies 3.3 Overall Training Procedure 4 Experiments 4.1 Datasets 4.2 Baseline 4.3 Experimental Setup and Evaluation Metrics 4.4 Results 5 Conclusion References LRRA:A Transparent Neural-Symbolic Reasoning Framework for Real-World Visual Question Answering 1 Introduction 2 Related Work 3 Approach 4 Experiments 4.1 Dataset 4.2 Implementation Details 4.3 Results 4.4 Example Analysis 5 Conclusion References Linguistics and Cognitive Science Meaningfulness and Unit of Zipf's Law: Evidence from Danmu Comments 1 Introduction 1.1 Zipf's Law 1.2 Remaining Questions in Zipf's Law 1.3 Goal of the Current Study 1.4 Danmu Comments 2 Previous Work 2.1 Meaningfulness of Zipf's Law 2.2 Units in Zipf's Law 3 Research Method 3.1 Data 3.2 Hypotheses 3.3 Predictions for Danmu Comments 4 Results and Discussion 4.1 Danmu 4.2 Clauses 4.3 Words 4.4 N-grams 5 Conclusion References Language Resource and Evaluation Unifying Discourse Resources with Dependency Framework 1 Introduction 2 Background 3 Unifying Discourse Corpora 3.1 Conversion of HIT-CDTB 3.2 Conversion of SU-CDTB 3.3 Corpus Statistics 4 Dependency Discourse Parsing 5 Conclusions References A Chinese Machine Reading Comprehension Dataset Automatic Generated Based on Knowledge Graph 1 Introduction 2 Related Work 2.1 English Machine Reading Comprehension Datasets 2.2 Chinese Machine Reading Comprehension Datasets 2.3 Machine Reading Comprehension Models 3 CMedRC: A Chinese Medical MRC Dataset 3.1 Question and Answer Generation 3.2 Synonymous Sentences Generation 3.3 Answer Matching and Document Linkage 3.4 Dataset Cleaning 3.5 Data Statistics 4 Experiment 4.1 Evaluation Metric 4.2 Baselines 4.3 Evaluation 5 Conclusion and Further Research 5.1 Conclusion 5.2 Further Research References Morphological Analysis Corpus Construction of Uyghur 1 Introduction 2 Related Work 2.1 Agglutinative Language and Morphological Analysis 2.2 Morphological Analysis Corpus 3 Construction of Uyghur Morphological Analysis Corpus 3.1 Data Preparation and Preprocessing 3.2 The Annotation Scheme 3.3 Corpus Construction Process 4 Corpus Information Statistics and Evaluation 5 Summary References Knowledge Graph and Information Extraction Improving Entity Linking by Encoding Type Information into Entity Embeddings 1 Introduction 2 Background and Related Works 2.1 Local and Global Entity Linking Models 2.2 Related Works 3 Our Method 3.1 Entity Labeling 3.2 Word and Type Vectors Pre-training 3.3 Entity Embeddings Training 3.4 Entity Embeddings Fusing 4 Experiments 4.1 Datasets and Evaluation Metric 4.2 Experimental Settings 4.3 Results 4.4 Analysis 5 Conclusion References A Unified Representation Learning Strategy for Open Relation Extraction with Ranked List Loss 1 Introduction 2 Methodology 2.1 Neural Encoders 2.2 Ranked List Loss 2.3 Virtual Adversarial Training 3 Experiment 3.1 Datasets 3.2 Settings 3.3 Main Results 3.4 Visualization Analysis 3.5 Other Empirical Studies 4 Conclusion References NS-Hunter: BERT-Cloze Based Semantic Denoising for Distantly Supervised Relation Classification 1 Introduction 2 Related Work 3 Model 3.1 Source Entity and Target Entity 3.2 Discrimination on Dependency Features 3.3 Denoising and Classification 4 Experiments 4.1 Dataset and Evaluation 4.2 Implementation Details 4.3 Baselines 4.4 Main Results 4.5 Denoising 4.6 Effects of Our MASK-lhs Feature 4.7 Apply NS-Hunter to CNN 4.8 Effect of Entity on Denoising 5 Conclusion References A Trigger-Aware Multi-task Learning for Chinese Event Entity Recognition 1 Introduction 2 The Proposed Algorithm 2.1 Event Trigger Dictionary Construction 2.2 Trigger Detection Network 2.3 Event-Featured Transformer-CRF 2.4 Loss Function 3 Experiments 3.1 Experimental Setup 3.2 Results and Discussion 4 Related Work 5 Conclusion References Improving Low-Resource Named Entity Recognition via Label-Aware Data Augmentation and Curriculum Denoising 1 Introduction 2 Related Work 2.1 Low-Resource NER 2.2 Curriculum Learning 3 Proposed Method 3.1 Overview 3.2 Data Augmentation via Pre-trained BERT 3.3 Denoising via Curriculum Learning 4 Experimental Setups 4.1 Datasets 4.2 Implementations 4.3 Experimental Results 4.4 Discussion 5 Conclusion References Global Entity Alignment with Gated Latent Space Neighborhood Aggregation 1 Introduction 2 Related Work 2.1 KG Embedding 2.2 Embedding-Based Entity Alignment 3 Problem Formulation 4 The Proposed Model 4.1 Basic Embedding Module 4.2 Topological Neighborhood and Latent Space Neighborhood 4.3 Neighborhood Aggregation 4.4 Gated Topological and Latent Space Neighborhood Aggregation 4.5 Global Entity Alignment Strategy 5 Experiments 5.1 Datasets and Experiment Settings 5.2 Entity Alignment Results 5.3 Effectiveness of Latent Space Neighborhood 5.4 Impact of Distance When Generating Latent Space Neighborhood 5.5 Effectiveness of Global Entity Alignment Strategy 6 Conclusion References NLP Applications Few-Shot Charge Prediction with Multi-grained Features and Mutual Information 1 Introduction 2 Related Work 2.1 Charge Predication 2.2 Few-Shot Text Classification 3 Methodology 3.1 Task Formalization 3.2 Coarse-Grained Features 3.3 Fine-Grained Features 3.4 Capsule Layer 3.5 Attribute Features 3.6 Aggregation and Prediction 3.7 Optimization 4 Experiments 4.1 Datasets 4.2 Baselines 4.3 Evaluation Metrics 4.4 Experiment Settings 4.5 Experimental Results 4.6 Ablation Study 4.7 Efficiency Analysis 5 Conclusion References Sketchy Scene Captioning: Learning Multi-level Semantic Information from Sparse Visual Scene Cues 1 Introduction 2 Related Works 3 A New Dataset for Sketchy Scene Captioning 3.1 SketchyScene Dataset without Descriptions 3.2 Description Collection for SketchyScene Dataset 3.3 Dataset Analysis 4 Multi-level Sketchy Scene Captioning Through Sequence Learning 4.1 Sketchy Scene Encoder for Deep Visual Features 4.2 Sketchy Scene Decoder with Spatial Visual Attention 5 Experiment and Analysis 5.1 Dataset and Preprocessing 5.2 Model Learning and Inference 5.3 Quantitative Evaluation 5.4 Qualitative Evaluation 6 Conclusion and Future Work References BDCN: Semantic Embedding Self-explanatory Breast Diagnostic Capsules Network 1 Introduction 2 Model 2.1 Check the Report Semantic Segmentation Layer 2.2 Sem-Bert Layer 2.3 Muti-Cap Layer 2.4 Model Interpretability 3 Experiment 3.1 Dataset 3.2 Parameter Settings 3.3 Different Stage Selection 4 Conclusion References GCN with External Knowledge for Clinical Event Detection 1 Introduction 2 Related Work 2.1 Clinical Event Detection 2.2 Stanford CoreNLP 2.3 BioBERT 3 Model 3.1 Character-Word Embedding 3.2 Semantic GCN 3.3 CRF Decoding 4 Experiment 4.1 Dataset 4.2 Evaluation Metrics 4.3 Settings 4.4 Evaluation on CED 4.5 Ablation Study 4.6 Type Analysis 5 Conclusion References A Prompt-Independent and Interpretable Automated Essay Scoring Method for Chinese Second Language Writing 1 Introduction 2 Related Work 2.1 Prompt-Specific vs. Prompt-Independent 2.2 Interpretability and Feedback 2.3 Automated Essay Scoring of Chinese Essays 3 The Proposed Method 3.1 The Interpretable Representations of Essay Features 3.2 The Ordinal Logistic Regression Model 4 Experiments 4.1 Dataset and Preprocessing 4.2 Feature Selection 4.3 Models, Parameters and Evaluation Metrics 4.4 Results 5 Discussion 5.1 Analysis on Confusion Matrix 5.2 Revisiting Linear Regression: Interpretability and Potential of Providing Feedback 6 Conclusion and Future Work A The Linguistic Complexity and Writing Error Features B The Example Essays References A Robustly Optimized BERT Pre-training Approach with Post-training 1 Introduction 2 The Proposed Model: PPBERT 2.1 Pre-training 2.2 Post-training 2.3 Fine-Tuning 3 Experiments 3.1 Tasks 3.2 Datasets 3.3 Experimental Results 4 Ablation Study and Analyses 4.1 Cooperation with Other Pre-trained LMs 5 Conclusion References Author Index