EM算法在统计自然语言处理中的应用  

Application of EM algorithm in statistical natural language processing

作　　者：王继曾[1] 刘宽[1] 任浩征[1] 罗恒[1] 

机构地区：[1]兰州理工大学计算机与通信学院,甘肃兰州730050

出　　处：《计算机工程与设计》2006年第19期3715-3717,共3页Computer Engineering and Design

摘　　要：在统计自然语言处理中会经常遇到一类参数估值问题,就是当观察数据为不完全数据时如何求解参数的最大似然估计,EM算法就是解决这类问题的经典算法。给出了EM算法的基本框架,结合HMM和PCFG模型给出如何应用EM算法求解参数的极大似然估计,讨论了EM算法的优点和不足之处。In statistical natural language processing, one class problem is often encountered that how to estimate the parameter＇s maximumlikelihood estimation when observed data set is incomplete. EM algorithm is the classical method to solve this problem. The basic framework of the EM algorithm is described, and then how to apply the EM algorithm is demonstrated to solve the problem of maximumlikelihood parameters estimation combine with the models ofHMM and PCFG. Finally, the advantages and disadvantages of EM algorithm are discussed.

关 键 词：自然语言 EM算法 参数估计 似然函数 隐马尔科夫模型 概率上下文无关文法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]