Chinese Computational Linguistics: 21st China National Conference, CCL 2022, Nanchang, China, October 14–16, 2022, Proceedings (Lecture Notes in Artificial Intelligence)

This book constitutes the proceedings of the 21st China National Conference on Computational Linguistics, CCL 2022, held in Nanchang, China, in October 2022. The 22 full English-language papers in this volume were carefully reviewed and selected from 293 Chinese and English submissions. The conference papers are categorized into the following topical sub-headings: Linguistics and Cognitive Science; Fundamental Theory and Methods of Computational Linguistics; Information Retrieval, Dialogue and Question Answering; Text Generation and Summarization; Knowledge Graph and Information Extraction; Machine Translation and Multilingual Information Processing; Minority Language Information Processing; Language Resource and Evaluation; NLP Applications. Preface Organization Contents Linguistics and Cognitive Science Discourse Markers as the Classificatory Factors of Speech Acts 1 Discourse Markers and the (Dis)agreement Continuum 2 Methods and Materials 2.1 The Hierarchical Cluster Analysis (HCA) 2.2 The Switchboard Dialog Act Corpus (SwDA) 3 Results and Discussions 4 Conclusions and Implications References Fundamental Theory and Methods of Computational Linguistics DIFM: An Effective Deep Interaction and Fusion Model for Sentence Matching 1 Introduction 2 BIDAF Model Based on Bi-directional Attention Flow 3 Method 3.1 Embedding Layer 3.2 Contextual Encoder Layer 3.3 Interaction Layer 3.4 Fusion Layer 3.5 Output Layer 4 Experimental Results and Analysis 4.1 Experimental Details 4.2 Experimental Results and Analysis 4.3 Ablation Experiments 5 Conclusion References ConIsI: A Contrastive Framework with Inter-sentence Interaction for Self-supervised Sentence Representation 1 Introduction 2 Related Work 3 Methodology 3.1 Model 3.2 Sentence Construction Techniques 4 Experiments 4.1 Data 4.2 Evaluation Setup 4.3 Training Details 4.4 Baselines 4.5 Main Results 4.6 Ablation Study 4.7 Analysis 5 Conclusion References Information Retrieval, Dialogue and Question Answering Data Synthesis and Iterative Refinement for Neural Semantic Parsing without Annotated Logical Forms 1 Introduction 2 Background 2.1 Base Semantic Parsing Model 2.2 SCFG for Data Synthesization 3 Approach 3.1 Data Synthesis 3.2 Iterative Data-Model Refining 4 Experiments 4.1 Experimental Settings 4.2 Experimental Results 5 Related Work 6 Conclusions References EventBERT: Incorporating Event-Based Semantics for Natural Language Understanding 1 Introduction 2 Related Work 2.1 Semantics in Language Representation 2.2 Graph Modeling for Language Understanding 3 Model 3.1 Context Encoder 3.2 Event-Based Encoder 4 Experiments 4.1 Setup 4.2 Results 5 Analysis 5.1 Ablation Study 5.2 Methods of Aggregation 5.3 Effectiveness of Semantic Structures 5.4 Interpretability: Case Study 5.5 Error Analysis 6 Conclusion References An Exploration of Prompt-Based Zero-Shot Relation Extraction Method 1 Introduction 2 Related Work 2.1 Knowledge in Pretrained Language Model 2.2 Prompt-Based Optimization 3 Method 3.1 Problem Definition 3.2 Task Reformulation 3.3 Model with Prompt Tuning 3.4 Training and Inference 4 Experimental Setup 4.1 Datasets 4.2 Compared Methods 4.3 Implementation Details 5 Results and Discussion 5.1 Main Results 5.2 Cross Domain Analysis 5.3 Influence of Pre-defined Relation Number 5.4 Analysis on Different Prompt Forms 6 Conclusions References Abstains from Prediction: Towards Robust Relation Extraction in Real World 1 Introduction 2 Related Work 2.1 Relation Extraction 2.2 Classification with Rejection Option 3 Approach 3.1 Method Overview 3.2 Relation Representation Encoder 3.3 Confidence-Calibrated Classifier 3.4 Class-Preserving Transformations 3.5 OpenRE Module 4 Experimental Setup 4.1 Datasets 4.2 Baselines and Evaluation Metrics 4.3 Implementation Details 5 Results and Analysis 5.1 Main Results 5.2 Ablation Study 5.3 Relation Representation Visualization 5.4 A Case Study on OpenRE 6 Conclusions References Using Extracted Emotion Cause to Improve Content-Relevance for Empathetic Conversation Generation 1 Introduction 2 Related Work 3 Task Formulation 4 Approach 4.1 Emotion Cause Extractor 4.2 Empathetic Conversation Generator 4.3 Training Strategy 5 Experiments 5.1 Datasets 5.2 Comparison Methods 5.3 Evaluation Metrics 5.4 Parameter Settings 5.5 Experimental Results and Analysis 5.6 Case Study 6 Conclusion References Text Generation and Summarization To Adapt or to Fine-Tune: A Case Study on Abstractive Summarization 1 Introduction 2 Related Work 3 Methodology 3.1 Method Overview 3.2 Adapter Variants 3.3 Evaluation 4 Language Experiments 4.1 Experimental Setup 4.2 Results 4.3 Convergence 5 Domain Adaptation Experiments 5.1 Experimental Setup 5.2 Results 5.3 Qualitative Analysis 6 Task Transfer Experiments 6.1 Experimental Setup 6.2 Results 6.3 Model Robustness 6.4 Effect of Data Availability on Performance 7 Conclusions and Future Work A Model Configurations References Knowledge Graph and Information Extraction MRC-Based Medical NER with Multi-task Learning and Multi-strategies 1 Introduction 2 Related Work 3 The MRC-MTL-MS Model 3.1 Multi-task Learning (MTL) 3.2 Multi-strategies (MS) 4 Datasets 5 Experiments 5.1 Query Generation 5.2 Experimental Settings 5.3 Comparison with Previous Models 5.4 Ablation Experiments 5.5 Experiments on Different Types of NEs 5.6 Case Study 6 Conclusion References A Multi-Gate Encoder for Joint Entity and Relation Extraction 1 Introduction 2 Related Work 3 Method 3.1 Problem Definition 3.2 Multi-gate Encoder 3.3 Table-Filling Modules 3.4 Loss Function 3.5 Differences from PFN 4 Experiments 4.1 Dataset 4.2 Evaluation 4.3 Implementation Details 4.4 Baselines 4.5 Main Results 4.6 Inference Speed 5 Analysis 5.1 Effect of Task Gates 5.2 Effect of Interaction Gates 5.3 Bidirectional Interaction vs Unidirectional Interaction 6 Conclusion References Improving Event Temporal Relation Classification via Auxiliary Label-Aware Contrastive Learning 1 Introduction 2 Related Work 2.1 Event Temporal Relation Classification 2.2 Contrastvie Learning 3 Our Baseline Model 4 Self-supervised Contrastive Learning 5 TempACL Approach 5.1 Training Encoder K 5.2 Joint Training with Patient Label-Aware Contrastive Loss 6 Experiments and Results 6.1 Dataset 6.2 Experimental Setup 6.3 Main Results 6.4 Ablation Study and Qualitative Analysis 7 Conclusion References Machine Translation and Multilingual Information Processing Towards Making the Most of Pre-trained Translation Model for Quality Estimation 1 Introduction 2 Background 2.1 Task Description 2.2 Previous Work 3 Approach 3.1 QE Architecture 3.2 Conditional Masked Language Modeling 3.3 Denoising Restoration 4 Experiments 4.1 Settings 4.2 Main Results 5 Analysis 5.1 The Impact of Mask Ratio and [MASK] Symbol 5.2 The Impact of Knowledge Distillation 5.3 Different Loss Calculation Methods 6 Conclusion References Supervised Contrastive Learning for Cross-Lingual Transfer Learning 1 Introduction 2 Background 2.1 Contrastive Learning 2.2 Cross-Lingual Transfer 3 Approach 3.1 Cross-Lingual Data Augmentation 3.2 Cross-Lingual Alignment: What Do We Want? 3.3 Better Alignment with SCL 4 Experiments 4.1 Data Preparation 4.2 Setup 4.3 Main Results 5 Analysis and Discussion 5.1 Different Augmentations 5.2 Similarity Measure 5.3 Contrast Temperature 5.4 SCL for Cross-Lingual Retrieval 6 Conclusion References Minority Language Information Processing Interactive Mongolian Question Answer Matching Model Based on Attention Mechanism in the Law Domain 1 Introduction 2 Related Work 3 Model Architecture 3.1 Input Layer 3.2 Context Encoding Layer 3.3 Interaction Layer 3.4 Matching Layer 3.5 Aggregation Layer 3.6 Prediction Layer 3.7 Model Training 4 Experiments 4.1 Data Set and Evaluation Metrics 4.2 Model Configuration 4.3 Baselines 4.4 Results 4.5 Ablation Study 5 Conclusion References Language Resource and Evaluation TCM-SD: A Benchmark for Probing Syndrome Differentiation via Natural Language Processing 1 Introduction 2 Preliminaries 2.1 Characteristics of Traditional Chinese Medicine (TCM) Diagnosis 2.2 Differences Between ICD Coding and Syndrome Differentiation 3 Related Works 3.1 Medical Datasets 3.2 Natural Language Processing (NLP) in Syndrome Differentiation 3.3 Domain Specific Pre-trained Language Model 4 Benchmark and Methods 4.1 Syndrome Normalization 4.2 Dataset Statistics 4.3 External Knowledge 4.4 ZY-BERT 5 Experiments 5.1 Baseline 5.2 Main Results 6 Discussion 6.1 Effect of Domain-Specific Pre-training 6.2 Effect of Knowledge 6.3 Ablation Study 6.4 Error Analysis 7 Conclusions References COMPILING: A Benchmark Dataset for Chinese Complexity Controllable Definition Generation 1 Introduction 2 Related Work 2.1 Definition Generation 2.2 Controllable Generation 2.3 Prompt Learning 3 Problem Formulation 4 Dataset Construction 4.1 Data Structured Annotation 4.2 Example Sentences Expansion 4.3 Post Processing 5 Dataset Analysis 6 Experiments 6.1 Baselines 6.2 Settings 6.3 Evaluation Metrics 6.4 Results and Analysis 7 Conclusion References NLP Applications Can We Really Trust Explanations? Evaluating the Stability of Feature Attribution Explanation Methods via Adversarial Attack 1 Introduction 2 Related Work 2.1 Feature Attribution Explanation Method 2.2 Evaluation of Explanation Methods 3 Formulation 3.1 Textual Adversarial Attack 3.2 Explanation Adversarial Attack 4 Attack Method 4.1 Measuring the Explanation Difference 4.2 Attack Strategies 5 Experiments 5.1 Datasets and Models 5.2 Explanation Methods 5.3 Experimental Settings and Results 6 Discussion 6.1 Correlation Analysis Between the Two Attack Steps and the Evaluation Results 6.2 Simply Improving Stability of Explanation Method 7 Conclusion 8 Limitations References Dynamic Negative Example Construction for Grammatical Error Correction Using Contrastive Learning 1 Introduction 2 Related Work 3 Our Method 3.1 Overall Architecture 3.2 Negative Example Construction Strategy 3.3 Dynamic Construction 3.4 Model Training 4 Experiments 4.1 Datasets 4.2 Experiment Settings 4.3 Compared Models 4.4 Overall Results and Analysis 5 Discussion and Analysis 5.1 Case Study 5.2 Effect of Dynamic Construction 5.3 Effect of the Noising Probability 6 Conclusion References SPACL: Shared-Private Architecture Based on Contrastive Learning for Multi-domain Text Classification 1 Introduction 2 Related Work 2.1 Multi-domain Text Classification 2.2 Contrastive Learning 3 Methodology 3.1 Model Architecture 3.2 Domain-Specific Representation Learning 3.3 Conditional Adversarial Network 3.4 Contrastive Learning 3.5 Objective Function 4 Experiment 4.1 Dataset 4.2 Baselines 4.3 Experimental Setting 4.4 Results 4.5 Ablation Study 5 Conclusion References Low-Resource Named Entity Recognition Based on Multi-hop Dependency Trigger 1 Introduction 2 Model 2.1 DepTrigger 2.2 Trigger Match Network 2.3 Entity Recognition Network 2.4 Inference 3 Experiments 3.1 Experiments Setup 3.2 Results 4 Conclusion and Future Work References Fundamental Analysis Based Neural Network for Stock Movement Prediction 1 Introduction 2 Related Work 2.1 Technical Analysis Based Approach 2.2 Fundamental Analysis Based Approach 3 Our Method 3.1 Task Definition 3.2 Overall Architecture 3.3 Text Encoding 3.4 Price Features 3.5 Temporal Fusion by Coattention Neural Network 3.6 Global Fusion by Sequential Encoding 3.7 Model Training 4 Experiments 4.1 Dataset 4.2 Experiment Settings 4.3 Compared Models 4.4 Experimental Results 4.5 Ablation Study 4.6 Case Study 5 Conclusion References Author Index