基于规则的彝汉双语调序模型研究

A Rule-based Reordering Model of Yi-Chinese Translation

贵州古彝文与汉语在句法结构上属于非单调对称语言,在机器翻译过程中,这种差异将严重影响译文质量。本文通过研究贵州古彝文与汉语在句法结构上的差异,提出一种基于规则的彝汉双语调序模型,用于调整彝文语序,以提高彝汉翻译译文质量。该模型中针对贵州古彝文句法特点,总结了一系列彝语句法规则,通过对比彝语和汉语句法结构上的差异,利用移动、替换等操作调整彝语语序,使其在结构上更符合汉语语序。最后通过实验对所提调序规则进行验证,实验结果表明,通过调序模型得到的符合目标语言句法的源语言调序结果正确率为0.635;利用词对齐模板,对源语言进行直译,译文的BLUE评分值由调序前的0.154提高到调序后0.248。

There is a non-monotonic symmetric relationship between Guizhou ancient Yi language and Chinese. In order to improve the translation quality, a rule-based reordering model was proposed to adjust the source-side language order. In this model, by studying the differences of them, a series summarized, and utilized to decide the source-side language＇s operators, of syntactic rules of Yi language were such as exchange, replacement, by comparing the structure different information between Yi and Chinese. At the last, some experiments were de- signed to verify the correctness of these rules. The result shows the correctness, that target language＇s syntactic phase tree produced by these rules, is 0.635.The BLEU score is improved from 0.154 before reordering to 0.248 after.