《厦门大学学报（自然科学版）》

目前,基于深度学习的神经机器翻译已经成为机器翻译领域的主流方法.神经机器翻译模型相较于统计机器翻译模型具有更庞大的参数规模,因此其翻译质量取决于训练数据是否充足.由于与维吾尔语相关的平行语料资源严重匮乏,神经机器翻译模型在维汉翻译任务上表现不佳,为此提出了一种利用伪语料对神经机器翻译模型进行增量训练的方法,可有效提升神经机器翻译在维汉翻译任务上的质量.

At present,the neural machine translation based on deep learning has become the mainstream method in the field of machine translation.The neural machine translation model requires a larger parameter size than the statistical machine translation model does. Therefore, its translation quality depends on the sufficiency of the training data.Due to the serious lack of parallel corpus resources related to Uyghur,the neural machine translation model performs poorly on Uyghur-to-Chinese translation tasks.This paper proposes a method of incremental training of neural machine translation models using pseudo-corpus,which effectively improves the quality of neural machine translation in Uyghur-to-Chinese translation tasks.

引言
1 系统架构
2 数据
3 实验
4 结论

图1 Transformer模型架构(修改自文献[4])<br/>Fig.1 Architecture of the Transformer model(modified from reference[4])

图1 Transformer模型架构(修改自文献[4])
Fig.1 Architecture of the Transformer model(modified from reference[4])

图2 多头自注意力模块(修改自文献[4])<br/>Fig.2 Multi-head self-attention module(modified from reference[4])

图2 多头自注意力模块(修改自文献[4])
Fig.2 Multi-head self-attention module(modified from reference[4])

图3 利用反向翻译方法构建伪语料<br/>Fig.3 Constructing synthetic corpus using back-translation method

图3 利用反向翻译方法构建伪语料
Fig.3 Constructing synthetic corpus using back-translation method

图4 利用伪平行语料的增量训练方法<br/>Fig.4 Incremental training method using pseudo-parallel corpus

图4 利用伪平行语料的增量训练方法
Fig.4 Incremental training method using pseudo-parallel corpus

表1 基于不同评价指标的测试结果<br/>Tab.1 Testing results based on different evaluation metrics

表1 基于不同评价指标的测试结果
Tab.1 Testing results based on different evaluation metrics

[1] CHO K,VAN MERRIENBOER B,GULCEHRE C,et al.Learning phrase representations using RNN encoder-decoder for statistical machine translation[C]∥Proceedings of EMNLP 2014.Baltimore:Association for Computational Linguistics,2014:1724-1734.
[2] BAHDANAU D,CHO K,BENGIO Y.Neural machine translation by jointly learning to align and translate[C]∥Proceedings of ICLR 2015.San Diego:International Conference on Learning Representations,2015:1409.0473.
[3] GEHRING J,AULI M,GRANGIER D,et al.Convolutional sequence to sequence learning[C]∥Proceedings of ICML 2017.Sydney:International Conference on Machine Learning,2017:1705.03122.
[4] VASWANI A,SHAZEER N,PARMAR N,et al.Attention is all you need[C]∥Proceedings of NIPS 2017.Long Beach:Conference on Neural Information Processing Systems,2017:1706.03762.
[5] SENNRICH R,HADDOW B,BIRCH A.Improving neural machine translation models with monolingual data[C]∥Proceedings of ACL 2016.Berlin:Association for Computational Linguistics,2016:86-96.
[6] LUONG M,PHAM H,MANNING C D.Effective approaches to attention-based neural machine translation[C]∥Proceedings of EMNLP 2015.Lisbon:Association for Computational Linguistics,2015:1412-1421.
[7] WU Y,SCHUSTER M,CHEN Z,et al.Google's neural machine translation system:bridging the gap between human and machine translation[EB/OL].[2018-11-27].https:∥arxiv.org/pdf/1609.08144.
[8] SUTSKEVER I,VINYALS V,LE Q V.2014.Sequence to sequence learning with neural networks[EB/OL].[2018-11-27].https:∥arxiv.org/pdf/1409.3215.
[9] SENNRICH R,HADDOW B,BIRCH A.Neural machine translation of rare words with subword units[C]∥Proceedings of ACL 2016.Berlin:Association for Computational Linguistics,2016:1715-1725.
[10] JOHNSON M,SCHUSTER M,LE Q V,et al.2016 Google's multilingual neural machine translation system:enabling zero-shot translation[EB/OL].[2018-11-27].https:∥arxiv.org/pdf/1611.04558.
[11] CHIANG D,DENEEFE S,CHAN Y S,et al.Decomposability of translation metrics for improved evaluation and efficient algorithms[EB/OL].[2018-11-27].https:∥www3.nd.edu/～dchiang/papers/bleu.pdf.
[12] DODDINGTON G.Automatic evaluation of machine translation quality using N-gram co-occurrence statistics[EB/OL].[2018-11-27].http:∥www.mt-archive.info/HLT-2002-Doddington.pdf.
[13] MELAMED I D,GREEN R,TURIAN J P.Precision and recall of machine translation[C]∥Proceedings of HLT-NAACL 2003.Edmonton:Association for Computational Linguistics,2003:61-63.
[14] KLAKOW D,PETERS J.Testing the correlation of word error rate and perplexity[J].Speech Communication,2002,38(1/2):19-28.
[15] LEUSCH G,UEFFING N,NEY H.A novel string-to-string distance measure with applications to machine translation evaluation[C]∥Proceedings of MT Summit Ⅸ 2003.New Orleans:[s.n.],2003:240-247.
[16] 刘群,刘洋.一种机器翻译自动评测方法及其系统:中国,ZL200410000628.8[P].2009-10-28.
[17] BANERJEE S,LAVIE A.METEOR:an automatic metric for MT evaluation with improved correlation with human judgments[C]∥Proceedings of ACL 2005.Ann Arbor:Association for Computational Linguistics,2005:65-72.
[18] SNOVER M,DORR B,SCHWARTZ R,et al.A study of translation edit rate with targeted human annotation[C]∥Proceedings of AMTA 2006.Cambridge:Association for Machine Translation in the Americas,2006:223-231.

备注

引言

1 系统架构

2 数据

3 实验

4 结论

学报简介

备注

引言

1 系统架构

2 数 据

3 实 验

4 结 论

学报简介

2 数据

3 实验

4 结论