Machine translation : 17th China Conference, CCMT 2021, Xining, China, October 8-10, 2021 : revised selected papers

This book constitutes the refereed proceedings of the 17th China Conference on Machine Translation, CCMT 2020, held in Xining, China, in October 2021. The 10 papers presented in this volume were carefully reviewed and selected from 25 submissions and focus on all aspects of machine translation, including preprocessing, neural machine translation models, hybrid model, evaluation method, and post-editing. Preface 6 Organization 7 Contents 12 A Document-Level Machine Translation Quality Estimation Model Based on Centering Theory 13 1 Introduction 13 2 Related Work 14 3 Centering Theory and Extraction of the Preferred Centers 15 3.1 Centering Theory and Preferred Centers 15 3.2 The Preferred Centers Extraction Model 15 3.3 The Semi-supervised Preferred Center Extraction Method 17 4 The Quality Estimation Model 17 4.1 The Inner-Extractor 18 4.2 The Outer-Extractor 18 4.3 The Evaluator 20 5 Experiments 20 5.1 Metrics 20 5.2 Dataset Description 21 5.3 Preferred Centers Extraction 22 5.4 QE Results 22 5.5 Case Study 23 6 Conclusion 23 A Appendix 24 B Appendix 25 C Appendix 25 References 26 SAU’S Submission for CCMT 2021 Quality Estimation Task 28 1 Introduction 28 2 Methods 29 2.1 Basic Model 29 2.2 Proposed Method 30 2.3 Ensemble 31 3 Experiment 31 3.1 Dataset 31 3.2 Settings 32 3.3 Results of the Single Model 32 3.4 Results of the Ensemble Methods 33 4 Conclusion 34 References 35 BJTU-Toshiba's Submission to CCMT 2021 QE and APE Task 37 1 Introduction 37 2 Chinese-English Sentence-Level Quality Estimation 38 2.1 Model Description 38 2.2 Multi-phase Pre-finetuning 40 2.3 Partial-Input Estimation 42 2.4 Model Ensemble 43 3 Chinese-English Automatic Post-Editing 43 3.1 BERT-initialized Transformer 43 3.2 Domain Selection 45 3.3 Data Augmentation Techniques 45 4 Conclusion 48 References 48 Low-Resource Neural Machine Translation Based on Improved Reptile Meta-learning Method 51 1 Introduction 51 2 Background 52 3 Our Approach 54 3.1 Unified Word Embedding Representation 54 3.2 NMT Method Based on Improved Reptile Meta-learning 55 4 Experiments 57 4.1 Datasets 57 4.2 Setting and Baseline 58 4.3 Result and Analysis 58 4.4 Ablation Experiments 60 4.5 Case Study 60 5 Conclusion 61 References 62 Semantic Perception-Oriented Low-Resource Neural Machine Translation 63 1 Introduction 63 2 Background 65 3 Method 66 3.1 Semantic Perception-Assisted Pre-training Model 66 3.2 Hierarchical Knowledge Distillation Training Process 68 4 Experiments 69 4.1 Datasets and Configuration 69 4.2 Results and Analysis 70 4.3 Ablation Experiment 71 4.4 Case Study 72 5 Conclusion 73 References 73 Semantic-Aware Deep Neural Attention Network for Machine Translation Detection 75 1 Introduction 75 2 Related Work 76 3 Model Overview 77 3.1 Semantic-Aware Influencing Attention Network (SIAN) in Monolingual Scenario 78 3.2 Semantic Consistency-Aware Interactive Attention Network (SCIA) in Bilingual Scenario 80 4 Experiments 82 4.1 Data Preparation 82 4.2 Model Parameters Settings 82 4.3 Evaluation Metric 83 4.4 Model Comparison and Analysis in Monolingual Scenario 83 4.5 Model Comparison and Analysis in the Bilingual Scenario 84 4.6 Case Study 85 4.7 Evaluation on Neural Machine Translation Systems 85 5 Conclusion 86 References 87 Routing Based Context Selection for Document-Level Neural Machine Translation 89 1 Introduction 89 2 Related Work 90 3 Background 91 3.1 Document-Level NMT 91 3.2 Transformer 92 3.3 Conditional Language-Specific Routing (CLSR) 92 4 Method 93 4.1 Inputs of Our Model 93 4.2 Context Attention 94 4.3 Auto-selection 95 5 Experiments 96 5.1 Datasets 96 5.2 Training Detail 97 5.3 Main Results 97 5.4 Ablation Study 98 5.5 Analysis 99 6 Conclusion and Future Work 100 References 101 Generating Diverse Back-Translations via Constraint Random Decoding 104 1 Introduction 104 2 Related Work 105 3 Proposed Methods 106 3.1 Fluency Boost Learning 106 3.2 Evolution Decoding Algorithm 107 3.3 Joint Training 109 4 Experimental Setting 109 4.1 Metrics 109 4.2 Dataset 110 4.3 Experiment Settings 110 5 Results and Analysis 111 5.1 Main Results 111 5.2 Quantitative Analysis 112 5.3 Qualitative Analysis 113 6 Conclusion 114 References 114 ISTIC's Neural Machine Translation System for CCMT' 2021 117 1 Introduction 117 2 System Architecture 118 2.1 Baseline System 118 2.2 Our System 120 3 Methods in Different Tasks 121 3.1 M2C Task, U2C Task, and T2C Task 122 3.2 R2C Low Resource Task 122 4 Experiments 123 4.1 System Settings 123 4.2 Data Pre-processing 123 4.3 Experimental Results 125 5 Conclusions 127 References 127 BJTU's Submission to CCMT 2021 Translation Evaluation Task 129 1 Introduction 129 2 Data 130 2.1 Chinese-English 130 2.2 UyghurChinese 130 2.3 TibetanChinese 131 3 Model 131 4 Method 131 4.1 Data Augmentation 131 4.2 Model Average 132 4.3 Finetune 132 4.4 Model Ensemble 133 4.5 Reranking 133 5 Experiment 133 5.1 ChineseEnglish 133 5.2 EnglishChinese 134 5.3 UyghurChinese 134 5.4 TibetanChinese 135 6 Conclusion 135 References 136 Author Index 137