基于句子级BLEU指标挑选数据的半监督神经机器翻译.docxVIP

基于句子级BLEU指标挑选数据的半监 督神经机器翻译 叶绍林郭武 验室 摘要： 在单语语料的使用上，统计机器翻译可通过利用语言模型提高性能，而神经机 器翻译很难通过这种方法有效利用单语语料?针对此问题，文中提出基于句子级 双语评估替补(BLEU)指标挑选数据的半监督神经网络翻译模型?分别利用统计 机器翻译和神经机器翻译模型对无标注数据生成候选翻译，然后通过句子级 BLEU指标挑选单语候选翻译，加入到有标注的数据集中进行半监督联合训练. 实验表明，文中方法能高效利用无标注的单语语料，在N1ST汉英翻译任务上， 相比仅使用精标的有标注数据单系统，文中方法BLEU值有所提升. 关键词： 半监督;句子级双语评估替补(BLEU);神经机器翻译; 作者简介：叶绍林，男，1993年生，硕士研究生，主要研究方向为机器翻 译.E-mai 1: slye@ma订.ustc. edu. cn. 作者简介：郭武(通讯作者)，男，1973年生，博士，副教授，主要研究方向 为语咅信号处理、自然语言处理.Ena订:guowu@ustc. edu. cri. 收稿日期：2017-05-12 基金：国家重点研发计划专项项目(NO.2016YFB1001303)资助 Semi-supervised Neural Machine Translation Based on Sentence-Level BLEU Metric Data Selection YE Shaolin GUO Wu NationalEngineering Laboratory for Speech and Language Information Processing, University of Science and Technology of China; Abstract： The performanee of statistical machine translation is improved by language model. However, the monolingual corpus is not equal to be effectively used by ncureil machine translation. To solve this problem, a semi-supervised neural machine translation model based on sentence-level bilingual evaluation understudy ( BLEU) metric data selection is proposed. The Candidate translations for non-labeled data are firstly generated by statistical machine translation and neural machinc translation models, respectively. Then the candidatc translations are selected through sentence-level BLEU, and the selected candidate translations are added to the labeled dataset to conduct semi-supervised joint training. The experimental resuIts demonstrate the effectiveness of the proposed algorithm in the usage of non-labeled data. Tn the NTST Chin csc-E nglish trans lati on t asks, the proposed met hod obta ins an obvious improvement over the baseline system only with the fine labeled data. Keyword： Semi-supervise; Sentence-Level B订ingual Evaluation Understudy (BLEU); Neural Machine Tremslation; Received： 2017-05-12 引用格式叶绍林，郭武?基于句子级BLEU指标挑选数据的半监督神经机器翻译. 模式识别与人工智能，2017, 30 (12) :937-942. 随着端到端(End-to-End)的神经机器翻译(Neural Machine Transla