神经机器翻译的联合训练(英文版)

传统的神经机器翻译都是用一个有向的模型来对一种语言到另外一种语言的翻译进行建模。相比于传统的学习方法，本书的联合训练可以使得独立的有向模型共享双方的信息，得到的更好的交互，从而使得训练出来的模型表现更好。本书可供高校和科研院所计算机科学与技术专业的科研人员、学生以及机器智能等相关行业的工程技术人员阅读和参考。

神经机器翻译的出现对于机器翻译的发展有着根本性的推动，关于神经机器翻译的研究也成为近几年的学术热点之一。传统的神经机器翻译是用一个有向的模型来对一种语言到另外一种语言的翻译进行建模。相比于传统的学习方法，本书中的联合训练可以使得独立的有向模型共享双方的信息，得到更好的交互，从而使得训练出来的模型表现更好。本书可供高校和科研院所计算机科学与技术相关专业的科研人员、学生以及机器智能等相关行业的工程技术人员阅读和参考。

2012年毕业于北京交通大学，2017年取得清华大学大学工学博士学位，2017年加入腾讯担任高级研究员，主要研究领域为机器翻译，在国际重要会议诸如ACL、IJCAI、AAAI等发表论文10多篇。

1 Neural Machine Translation
1.1 Introduction
1.2 Neural Machine Translation
References
2 Agreement-Based Joint Training for Bidirectional Attention-Based Neural Machine Translation
2.1 Introduction
2.2 Agreement-Based Joint Training
2.3 Experiments
2.3.1 Setup
2.3.2 Comparison of Loss Functions
2.3.3 Results on Chinese-English Translation
2.3.4 Results on Chinese-English Alignment
2.3.5 Analysis of Alignment Matrices
2.3.6 Results on English-to-French Translation
2.4 Summary
References
3 Semi-supervised Learning for Neural Machine Translation
3.1 Introduction
3.2 Semi-supervised Learning for Neural Machine Translation
3.2.1 Supervised Learning
3.2.2 Autoencoders on Monolingual Corpora
3.2.3 Semi-supervised Learning
3.2.4 Training
3.3 Experiments
3.3.1 Setup
3.3.2 Effect of Sample Size k
3.3.3 Effect of OOV Ratio
3.3.4 Comparison with SMT
3.3.5 Comparison with Previous Work
3.4 Summary
References
4 Joint Training for Pivot-Based Neural Machine Translation
4.1 Introduction
4.2 Pivot-Based NMT
4.3 joint Training for Pivot-Based NMT
4.3.1 Training Objective
4.3.2 Connection Terms
4.3.3 Training
4.4 Experiments
4.4.1 Setup
4.4.2 Results on the Europarl Corpus
4.4.3 Results on the WMT Corpus
4.4.4 Effect of Bridging Corpora
4.5 Summary
References
5 Joint Modeling for Bidirectional Neural Machine Translation with Contrastive Learning
5.1 Introduction
5.2 Unidirectional Neural Machine Translation
5.3 Bidirectional Neural Machine Translation
5.4 Decoding Strategies
5.5 Experiments
5.5.1 Setup
5.5.2 Effect of Translation Strategies
5.5.3 Comparison with SMT and Standard NMT
5.5.4 BLEU Scores Over Sentence Length
5.5.5 Comparison of Learning Curves
5.5.6 Analysis of Expected Embeddings
5.5.7 Results on English-German Translation
5.6 Summary
References
6 Related Work
6.1 Atentional Mechanisms in Neural Machine Translation
6.2 Capturing Bidirectional Dependencies
6.2.1 Capturing Bidirectional Dependencies
6.2.2 Agreement-Based Learning
6.3 Incorporating Additional Data Resources
6.3.1 Exploiting Monolingual Corpora for Machine Translation
6.3.2 Autoencoders in Unsupervised and Semi-supervised Learning
6.3.3 Machine Translation with Pivot Languages
6.4 Contrastive Learning
References
7 Conclusion
7.1 Conclusion
7.2 Future Directions
7.2.1 Joint Modeling
7.2.2 Joint Training
7.2.3 More Tasks
References