关键词: 维吾尔语, 多音词, 最大熵模型

Abstract: The correct pronunciation of polyphonic word is one of the important factors that affect the Uyghur speech synthesis intelligibility.A word consists of stem and affix in Uyghur language，although there is a few polyphone stems，but a large number of polyphonic words are constituted by jointing of affix and polyphonic stem.This paper selects 16 polyphonic stems which are frequently used and often read wrong in Uyghur language to study，presents a different rule based method and adopts the maximum entropy model for disambiguation of polyphonic words which does not meet the rules on the basis of the different features of polyphones.Simultaneously，log-likelihood ratio is used to extract keywords and greedy algorithm is used to select best feature set.The performance test of the algorithm shows that the average precision of polyphonic word disambiguation is up to 87.7%.

Key words: Uyghur language, polyphonic word, maximum entropy model