机器翻译词语漏译的语料库语境策略研究  

A study on the corpus context strategies for the missing words in Machine Translation

作　　者：赵会军 林国滨 ZHAO Huijun;LIN Guobin(School of International Studies,Yango University,Fuzhou 350015,China)

机构地区：[1]阳光学院外国语与海外教育学院,福建省350015

出　　处：《外语教学与研究》2022年第2期277-287,F0003,共12页Foreign Language Teaching and Research

基　　金：教育部人文社会科学研究规划基金项目“机器翻译漏译三维语境判断机理研究”(19YJAZH110);福建省社会科学规划项目“茶产业汉英双语语料库智能构建研究”(FJ2021B106)的阶段性成果。

摘　　要：机器翻译漏译错误有语用、语法层面的,也有词语层面的,后者占比较大。本文从机器翻译译后和机器翻译应用两个角度总结漏译的语言学应对策略。从词、短语和句子三个层级入手,采用本地语料库和机器翻译数据的语境交叉确认策略确定漏译的词语,采用词向量语境关联搭配策略降低机器翻译漏译数量。在两项实验中,与词语漏译评测强相关的BLEU值以及人工评测的结果都显示,在语言学干预语料库因素和词向量语境因素后,漏译比例大幅减少,相关词语错译和语序错误也同步减少。Word missing errors in Machine Translation(MT) can be found at the level of pragmatics,grammar,and words,with the last accounting for the majority.The linguistic strategies of identifying and reducing the missing translation are discussed in this paper from the perspectives of MT and post-translation.Starting from the three levels of words,phrases and sentences,the strategy of context cross-confirmation of local corpus and MT data is used to identify the missing words,and the strategy of context correlation matching of word embedding is used to reduce the number of missing words in MT.In the two experiments,both the bilingual evaluation understudy(BLEU) values that strongly correlate with the evaluation of word missing and the human evaluation show that after linguistic intervention in corpus and word embedding factors,the proportion of missing translation is greatly reduced,and the related word mistranslation and word order errors are also synchronously reduced.

关 键 词：机器翻译 漏译 语料库 语境 

分 类 号：H059[语言文字—语言学]