中软译星翻译质量改进策略

Transtar Approaches for Translation Improvement

本文总结了译星翻译系统所采用的技术方法。以及译星系统针对一般篇章翻译所采取的一些策略。在技术方法中，介绍了译星系统的语言模型、结构设计、源文分析与译文生成思想。对于翻译策略，提出了基于词组的语言单元分析方法，以及语义结构优先语法结构的处理原则，从而加强了词组和语义在机器翻译中的地位与作用。同时还介绍了如何加强语义分析的具体实施方法。以这些翻译策略为改进原则，译星翻译系统的翻译质量得到了提高。其中，译星汉英系统的篇章翻译忠实度达到了54％，译星英汉系统篇章翻译达到了51％。

This paper summarizes the technical model in Transtar system. It also introduces some improving approaches of Transtar in its general-article translation. In terms of technical model, it introduces Transtar linguistic model, structural design, basic idea on source analysis and target generation. In terms of translation approach, it puts forward the idea of phrase as sentence unit and semantic function as analysis preference. Therefore, it puts more stresses on phrase and semantics during machine translation. It also gives a brief introduction on how to conduct semantic function analysis. With these new approaches, Transtar has improved its translation performance. Transtar ChineseEnglish translation accuracy reaches 54%. And Transtar English-Chinese translation reaches 51%.