基于依存树的中文语义角色标注  

Chinese Semantic Role Labeling Based on Dependency Trees

作　　者：安强强[1] 张蕾[1] 

机构地区：[1]西北大学信息科学与技术学院,西安710127

出　　处：《计算机工程》2010年第4期161-163,共3页Computer Engineering

基　　金：陕西省教育厅专项科研基金资助项目(HD01302)

摘　　要：现有中文语义角色标注主要集中在基于短语结构句法树的标注。基于此,提出一种基于依存树的中文语义角色标注方法。将中文句子转化为标准的依存树,作为实验数据集,特征选取时结合知网,将语义信息引入特征集,以提高系统的召回率,并采用最大熵分类器进行实验,获得90.68%的F值。结果表明,在标准的句法树上,当基于依存关系的标注系统中加入新特征时,该中文语义角色标注取得了比基于句法成分标注更好的成绩。Current Chinese semantic role labeling mainly focuses on using phrase structure trees. This paper presents an approach of Chinese semantic role labeling method which is based on dependency trees. Chinese sentences are converted into gold dependency trees which are divided into training and testing set. By using maximum entropy classifier and adding the first sememe of word concept to the feature set, the system gets an F-score of 90.68%. Results show that dependency-based system adding new features performs better than constituent-based system on gold standard parses.

关 键 词：最大熵分类器 语义角色标注 依存树 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]