层次化蒙古语统计语言模型  

A Layered Statistical Mongolian Language Model

作　　者：侯宏旭[1] 张国强[1] 刘志文[1] 

机构地区：[1]内蒙古大学计算机学院,呼和浩特010021

出　　处：《内蒙古大学学报（自然科学版）》2009年第3期336-340,共5页Journal of Inner Mongolia University：Natural Science Edition

基　　金：内蒙古自然科学基金资助项目(200607010805)

摘　　要：蒙古文具有典型的构词、构形词缀的特点,一个蒙古语单词往往可以切分成词干和词缀等若干个部分.根据蒙古语的特点,提出了一种层次化的蒙古语语言模型,将蒙古语语言模型分为词干和词干、词干和词缀、词缀和词缀三个层次.在这三个层次上分别计算出独立的语言模型.把这种层次化的蒙古语语言模型方案应用到了一个统计汉蒙机器翻译系统中,实验证明所构造的层次化的蒙古语语言模型能够有效地提高汉蒙机器翻译的效果.The Mongolian language has the character of word-building affix formation. A Mongolian word usually can be divided into several parts such as etymas and affixes. According to this character of Mongolian, a new language model called layered Mongolian language model is presented. Language model is divided into three administrative levels,which are etyma and etyma, etyma and affix,affix and affix. There are independent language models in each level. The method is realized,and has been applied into a statistical Chinese-Mongolian machine translation system. The experiments proved that the layered Mongolian language model could improve the translation result.

关 键 词：蒙古语 语言模型 机器翻译 词干 词缀 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]