《厦门大学学报（自然科学版）》

针对维汉机器翻译中存在的维吾尔语(下文简称维语)形态复杂性和数据稀疏性问题,提出了一种层次化融合多个维语语法特征的神经网络机器翻译模型.该模型采用4种特征(词干、词性、词缀、词缀形态)作为源端语言附加信息,用于辅助单一词汇形式表示的维语语句; 同时引入层次化多特征融合的神经网络结构,用于分层处理维语的词干级和词缀级特征,以增强机器翻译系统对维语的句法结构和语义知识的学习能力,从而提高维汉机器翻译质量.在维汉公开数据集上的实验结果表明,所提出的层次化多特征融合模型可以有效提高维汉机器翻译系统性能,其双语互译评估(BLEU)值和字符匹配度(ChrF3)值均有明显提升.

Focusing on the issue of the complex morphology and data sparseness of Uyghur in Uyghur-Chinese machine translation,we proposes a neural hierarchical combination model for multiple Uyghur linguistic features.This model employs four features(lemma,part-of-speech tag,affix and affix morphology)as additional information to enrich the Uyghur sentences with single word surface form.Moreover,in the model we introduces a hierarchical multi-features combined neural network to hierarchically process the lemma-level and affix-level Uyghur features to enhance the ability of machine translation system and learn the Uyghur syntactic structure and semantic knowledge accordingly.Experimental results on Uyghur-Chinese public dataset show that the hierarchical multi-features combination model can effectively improve the performance of Uyghur-Chinese machine translation system on BLEU and ChrF3 scores.

引言
1 本文NMT模型
2 维语的语言学特征
3 实验
4 结论

图1 基于注意力机制的编码器-解码器模型<br/>Fig.1 Attentional-based encoder-decoder model

图1 基于注意力机制的编码器-解码器模型
Fig.1 Attentional-based encoder-decoder model

图2 多特征融合模型<br/>Fig.2 Multi-features combination model

图2 多特征融合模型
Fig.2 Multi-features combination model

图3 层次化的多特征融合模型<br/>Fig.3 Hierarchical multi-features combination model

图3 层次化的多特征融合模型
Fig.3 Hierarchical multi-features combination model

图4 维语形态结构<br/>Fig.4 Morphology structure of Uyghur language

图4 维语形态结构
Fig.4 Morphology structure of Uyghur language

表1 维语词汇的词缀个数统计<br/>Tab.1 Statistics of Uyghur affixes

表1 维语词汇的词缀个数统计
Tab.1 Statistics of Uyghur affixes

表2 维吾尔语语法特征序列<br/>Tab.2 linguistic feature sequences of Uyghur sentence

表2 维吾尔语语法特征序列
Tab.2 linguistic feature sequences of Uyghur sentence

表3 维汉机器翻译数据集统计<br/>Tab.3 Corpus statistics of Uyghur-Chinese machine translation

表3 维汉机器翻译数据集统计
Tab.3 Corpus statistics of Uyghur-Chinese machine translation

表4 各个系统的实验结果对比<br/>Tab.4 Experimental results comparison of different systems

表4 各个系统的实验结果对比
Tab.4 Experimental results comparison of different systems

表5 融合不同语法特征的实验结果对比<br/>Tab.5 Experimental results comparison of different linguistic features combination

表5 融合不同语法特征的实验结果对比
Tab.5 Experimental results comparison of different linguistic features combination

表6 维汉机器翻译实例<br/>Tab.6 Examples of Uyghur-Chinese translation

表6 维汉机器翻译实例
Tab.6 Examples of Uyghur-Chinese translation

[1] SUTSKEVER I,VINYALS O,LE Q V.Sequence to sequence learning with neural networks[EB/OL].[2019-08-01].https:∥arxiv.org/pdf/1409.3215.pdf.
[2] CHO K,VAN MERRIËNBOER B,GULCEHRE C,et al.Learning phrase representations using RNN encoder-decoder for statistical machine translation[EB/OL].[2019-08-01].https:∥arxiv.org/pdf/1406.1078.pdf.
[3] WANG M,LU Z,LI H,et al.Memory-enhanced decoder for neural machine translation[EB/OL].[2019-08-01].https:∥arxiv.org/pdf/1606.02003.pdf.
[4] VASWANI A,SHAZEER N,PARMAR N,et al.Attention is all you need[EB/OL].[2019-08-01].https:∥arxiv.org/pdf/1706.03762.pdf.
[5] WU Y,SCHUSTER M,CHEN Z,et al.Google's neural machine translation system:bridging the gap between human and machine translation[EB/OL].[2019-08-01].https:∥arxiv.org/pdf/1609.08144.pdf.
[6] JOHNSON M,SCHUSTER M,LE Q V,et al.Google's multilingual neural machine translation system:enabling zero-shot translation[J].Transactions of the Association for Computational Linguistics,2017,5:339-351.
[7] 米莉万.基于维吾尔语词干词缀粒度的汉维机器翻译[J].中文信息学报,2015,29(3):201-206.
[8] HALIDANMU A,YONG C,YANG L I U,et al.Uyghur morphological segmentation with bidirectional GRU neural networks[J].Journal of Tsinghua University(Science and Technology),2017,57(1):1-6.
[9] 熊德意,李军辉,王星,等.基于约束的神经机器翻译[J].中国科学:信息科学,2018(5):574-588.
[10] ALEXANDRESCU A,KIRCHHOFF K.Factored neural language models[C]∥Proceedings of the Human Language Technology Conference of the North American Chapter of the Association of Computational Linguistics.New York:ACL,2006:1-4.
[11] KOEHN P,HOANG H.Factored translation models[C]∥Proceedings of the 2007 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning(EMNLP-CoNLL).Prague:ACL,2007:868--876.
[12] CHEN D,MANNING C.A fast and accurate dependency parser using neural networks[C]∥Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing(EMNLP).Doha:ACL,2014:740-750.
[13] SENNRICH R,HADDOW B.Linguistic input features improve neural machine translation[EB/OL].[2019-08-01].https:∥arxiv.org/pdf/1606.02892.pdf.
[14] AQLAN F,FAN X,ALQWBANI A,et al.Improved Arabic-Chinese machine translation with linguistic input features[J].Future Internet,2019,11(1):22.
[15] 哈里旦木·阿布都克里木,刘洋,孙茂松.神经机器翻译系统在维吾尔语-汉语翻译中的性能对比[J].清华大学学报(自然科学版),2017(8):96-101.
[16] LUONG M T,SUTSKEVER I,LE Q V,et al.Addressing the rare word problem in neural machine translation[EB/OL].[2019-08-01].https:∥arxiv.org/pdf/1410.8206.pdf.
[17] BAHDANAU D,CHO K,BENGIO Y.Neural machine translation by jointly learning to align and translate[EB/OL].[2019-08-01].https:∥arxiv.org/pdf/1409.0473.pdf.
[18] GRAVES A,SCHMIDHUBER J.Framewise phoneme classification with bidirectional LSTM and other neural network architectures[J].Neural Networks,2005,18(5/6):602-610.
[19] LUONG M T,PHAM H,MANNING C D.Effective approaches to attention-based neural machine translation[EB/OL].[2019-08-01].https:∥arxiv.org/pdf/1508.04025.pdf.
[20] ZOPH B,KNIGHT K.Multi-source neural translation[EB/OL].[2019-08-01].https:∥arxiv.org/pdf/1601.00710.pdf.
[21] TURSUN E,GANGULY D,OSMAN T,et al.A semisupervised tag-transition-based markovian model for Uyghur morphology analysis[J].ACM Transactions on Asian and Low-Resource Language Information Processing(TALLIP),2016,16(2):8.
[22] PAPINENI K,ROUKOS S,WARD T,et al.BLEU:a method for automatic evaluation of machine translation[C]∥Proceedings of the 40th Annual Meeting on Association for Computational Linguistics.Stroudsburg:ACL,2002:311-318.
[23] POPOVIC' M.ChrF:character n-gram)的F3(ChrF3)值[23]作为机器翻译的评价指标.
[24] KINGMA D P,BA J.Adam:a method for stochastic optimization[EB/OL].[2019-08-01].https:∥arxiv.org/pdf/1412.6980.pdf.
[25] SENNRICH R,HADDOW B,BIRCH A.Neural machine translation of rare words with subword units[EB/OL].[2019-08-01].https:∥arxiv.org/pdf/1508.07909.pdf.

备注

引言

1 本文NMT模型

2 维语的语言学特征

3 实验

4 结论

学报简介

备注

引言

1 本文NMT模型

2 维语的语言学特征

3 实 验

4 结 论

学报简介

3 实验

4 结论