摘要
为了使中英文翻译更加智能以及更加合理,采用句子级的释义对译文进行改写,将其视为同一语言之间的翻译任务;在没有大规模平行释义语料库的情况下,利用机器翻译结果和源语言的参考翻译来近似平行释义语料库;然后,利用该模型训练一个从机器翻译结果到参考翻译的重复系统,生成语义一致的句子级重复结果;在此基础上,将重述结果引入系统整合的翻译假设中;最后,在翻译和释义的基础上,进行了面向移动应用的设计和开发,实现了中英文机器翻译;通过实验发现,该方法相对于经典的基线系统提高了1.02-1.71BLEU分数。
In order to make Chinese and English translation more intelligent and more reasonable,this article uses sentence-level interpretation to rewrite the translation as a translation task between the same language.In the absence of a large-scale parallel paraphrasing corpus,the machine translation results and the reference translation of the source language are used to approximate the parallel paraphrasing corpus.Then,the model is used to train a repetitive system from machine translation results to reference translations to generate semantically consistent sentence-level repetition results.Based on this,the results of restatement are introduced into the translation assumption of system integration.Finally,on the basis of translation and interpretation,the mobile application-oriented design and development were carried out to realize Chinese and English machine translation.It was found through experiments that the proposed method improved 1.02-1.71 BLEU scores compared to the classic baseline system.
作者
成洁
Cheng Jie(Shaanxi Institute of International Trade and Commerce,Xi an 712046,China)
出处
《计算机测量与控制》
2020年第10期186-190,共5页
Computer Measurement &Control
基金
2019陕西省教育厅专项科研计划项目(19JK0090)
2019陕西省社会科学基金项目(2019M032)。
关键词
机器学习
机器翻译
RNN
神经网络
释义
语料库
machine learning
machine translation
RNN
neural network
paraphrase
corpus