统计机器翻译综述刘群.pdf

17 4 JOURNAL OF CHINESE INFORMATION PROCESSING Vol117 No14 : 1003- 0077- (2003) 04- 0001- 12 ¹ (, 100871; , 100080) :本文综述了基于信源信道思想和基于最大熵思想的统计机器翻译方法并介绍了统计机器翻译的 评测方法基于信源信道的方法将翻译 率表示为一个语言模型和 一个翻译模型而基于最大熵的方法则 是利用 一系列实数值特征函数的线性组合来求解最优的译文基于最大熵的统计机器翻译方法比基于信源 信道的方法更具有 一般性, 后者可以看做前者的一个特例 :人工智能; 机器翻译;综述; 统计机器翻译;信源信道模型; 最大熵方法 :TP391.2 :A Survey on Statistical Mac ine Translation LIU Qun (Institute of Co putational Linguistics, Peking University, Beijing 100871, China; Institute of Co puting T echnology, ChineseAcade y of Sciences, Beijing 100080, China) Abstract:Thepaper gives a survey on three approaches of statistical achine translation and the evaluation ethods used in SMT. Thebasic idea of parallel gra ar based approach is to build parallel gra ars for source and target languages, which confor the sa e probabilistic distribution. In the source-channel approach, thetranslation proba- bility is expressed as a language odel and a translation odel. In the axi u entropy approach, the opti al translation is searched accordingto a linear co bination of a series of real-valued featurefunctions. T he source-chan- nel approach can beregard as a special case of axi u entropy approach. Key words: artificial intelligence; achine translation; overview; statistical achine translation; source channel odel; axi u entropy ethod , (data-driven) 1949 ,Weaver 56:/ , , , ,,0 , , , Cho sky , 90 , IBM Brown ¹ :2003- 02- 10 :(973) (G1998030507- 4;G1998030510) :( 1966) ), ,, ,. 1 ,, ,, ,, , , : , , , , , Als