日汉机器翻译误译分析及改善策略探讨——以日本首相的致辞文本为例

Analysis of Japanese-Chinese Machine Mistranslation and Discussion of Improvement Strategies—Using the Text of Japanese Prime Minister’s Speech as an Example

本文以日本首相的致辞文本为例,研究了百度机器翻译系统在致辞文本翻译上的表现,统计了机器翻译误译类型的分布,分析了机器误译的主要类型及原因,并从译者编辑的角度提出了改进策略。研究结果表明:百度翻译在致辞类文本上的整体表现较差,有超过半数的句子需要进行不同程度的译后编辑。在影响原文理解的机器翻译误译类型中,句段逻辑误译和词组搭配误译分别占32%,其次是专业术语误译,占12%,一般词义误译占9%,多译有12%,最后漏译有3%。针对这些机译问题,提出了译前简化长难句;在计算机翻译技术合成环境中工作;先修改后校对的策略,并通过实例进行验证,以期提高人机合作翻译工作的效率。

This paper takes the speech text of the Prime Minister of Japan as an example and studies the performance of Baidu’s machine translation system on the translation of speech texts, statistics on the distribution of machine translation mistranslation types, analyses the main types of machine mistranslation and their causes. It also proposes improvement strategies from the perspective of translator edits. The research results demonstrate that the overall performance of Baidu Translation on speech texts is weak with more than half of the sentences requiring different degrees of post-translation editing. Among the types of machine translation mistranslation that affect the comprehension of the original text, sentence paragraph logic mistranslation and phrase collocation mistranslation respectively account for 32%, followed by terminology mistranslation at 12%, general word sense mistranslation at 9%, multiple translations at 12%, and finally omission at 3%. In order to address these machine translation problems, the strategies of simplifying long and difficult sentences before translation;working in the synthetic environment of computer translation technology;and revising before proofreading are proposed and verified by examples, with a view to improving the efficiency of human-computer cooperative translation work.