基于最大熵模型的汉语依存分析

采用最大熵模型实现中文依存语法的分析。用自底而上的方式构建语句的依存关系树，构建过程每一步在向左连接、向右连接以及不连接3种动作选取其一。用最大熵原理判断每个动作的概率，得到依存树中各边的概率，然后找出具有最大概率的依存关系树。实验结果表明，该模型具有较好的分析精度。目前，该模型已被应用于基于自然语言的信息检索项目中。

Chinese Dependency Parsing with Maximum Entropy Principle

This paper uses maximum entropy(ME) model to parse chinese sentence with dependency grammar.The dependency-tree is constructed with a bottom-up process,and one of the three actions(left-concatenation,right-concatenation,non-concatenation) is selected in every step of the constructing process.The maximum entropy principle is used to compute the probability of the actions.Thus the dependency-tree with maximum probability can be obtained.The model is experimentally proved satisfying in precision and has been applied in a Chinese natural language retrieval project.