利用语义信息的句法分析统计模型

Statistical Syntactic Parsing Model Utilizing Semantic Information

句法结构是句法形式和语义内容的结合.中文配价结构能够准确地描绘中文句子的语义联系及语法结构,本文利用配价结构有关语法理论探索了融合配价信息的语义角色标记,进而构建了语义角色标记与语法结构分析并行学习方法:在语法分析的进程中,实施语义分析及标记;并把语义融入语法规则的概率计算.本文在语义信息标记基础上提出了基于语义类的句子语法结构分析模型,明显地提高了句法分析系统性能.句子语法结构分析试验数据说明,利用配价信息并基于词类的语义角色标记与句子语法结构分析联合学习方法,其召回率、精确率相应为88. 26%、88. 73%,综合指标相比头驱动句子语法结构分析方法提高了8. 39%.

Syntactic structures are unities of syntactic forms and semantic contents. Chinese valence structures can well characterize syntactic structures and semantic constitution relationships of Chinese sentences,this paper explored the semantic role labeling fusing valence information based on the studying of related theories of valence grammar,and established a joint syntactic and semantic parsing model: the labeling and parsing of semantic information are carried on during the process of syntactic parsing;and simultaneously the labeled semantic information is integrated in the probability calculations of rules. Based on the semantic information annotation,this paper proposes a syntactic parsing model based on semantic classes,the system performances of syntactic parsing were obviously enhanced. Experiments are conducted for the joint syntactic and semantic parser fusing valence information,it achieves 88. 73% precision and 88. 26% recall,F measure is improved 8. 39% comparing with head-driven parsing method.