摘要： 利用电话录音的汉维平行语料库和开源的Moses系统构建一个基于短语的统计机器翻译系统。针对汉维平行语料库规模较小和维吾尔语形态变化比较丰富的特点，通过对词级的语料库进行切分得到词素级的语料库，并分别进行词一级的实验和词素级的实验。实验表明，词素级的实验能降低无法识别的词的概率，提高翻译的质量。

关键词: 汉维, 维汉, 词素, 预处理, 后处理

Abstract: This paper gives a description of implementing a phrase-based machine translation system for Chinese-Uyghur, by the Moses toolkit, using a parallel corpus which is based on telephone recording. For the small scale parallel corpus and highly-inflected characteristics for Uyghur, it splits the Uyghur words into morphemes, and it gets another parallel corpus on morpheme-level. Experiments are carried out on word-level and morpheme-level separately, and show it can reduce the probability of Out-Of-Vocabulary(OOV) and improve the translation quality.

Key words: Chinese-Uyghur, Uyghur-Chinese, morpheme-level, preprocessing, postprocessing

中图分类号: 

• TP391.1