改进的模板驱动的神经机器翻译  

Improved template-driven neural machine translation

作　　者：詹聪 李盛乐[1] ZHAN Cong;LI Sheng-le(Key laboratory of Earthquake Geodesy,Institute of Seismology,Wuhan 430071,China)

机构地区：[1]中国地震局地震研究所地震大地测量重点实验室,湖北武汉430071

出　　处：《微电子学与计算机》2019年第5期29-32,37,共5页Microelectronics & Computer

基　　金：中国地震局地震研究所基本科研业务费专项资助项目;中国地震局地壳应力研究所基本科研业务费专项资助项目(IS201656260)

摘　　要：为解决神经机器翻译对语料的数量、规模要求较高,在语料不充足的情形下对结构复杂的长句翻译效果不佳的问题,本文对模板驱动的神经机器翻译进行了改进,在句子级模板的基础上增加了亚句子级的子块模板,提出了一种将句子进行压缩后与句子级模板匹配,以压缩后的句子单词为中心词提取名词短语和动词短语与子块模板进行匹配的方法.实验结果表明,本文提出的方法能有效改善机器翻译质量.At present,neural networks are widely used in machine translation and have achieved good translation results.However,neural machine translation is more sensitive to sentence length and has poor translation effect on long and complicated sentences.In order to solve this problem,based on template-based neural machine translation,this paper adds a sub-sentence level template named sub-block template to the sentence level template and proposed a method using compressed sentences to match the sentence level template,and extract the noun phrase and verb phrase with the compressed sentence word as the center word and match the sub-block template.The experimental results show that the proposed method can effectively improve the quality of machine translation.

关 键 词：神经机器翻译 子块模板 短语提取 

分 类 号：H085[语言文字—语言学] TP391.1[自动化与计算机技术—计算机应用技术]