基于树库的汉语依存句法分析  

Chinese Dependency Parsing Based on Treebank

作　　者：刘海涛[1] 赵怿怡[1] 

机构地区：[1]中国传媒大学应用语言学研究所,北京100024

出　　处：《模式识别与人工智能》2009年第1期17-21,共5页Pattern Recognition and Artificial Intelligence

摘　　要：旨在探索利用语言学手段来提高句法分析精度的可能性.采用MaltParser和自建的汉语依存树库进行相关汉语依存句法分析实验.通过对句法分析结果的分析,找出影响句法分析精度的主要因素,并据此对树库中处理某些语言结构的方式进行修改.然后再对得到的句法分析数据进行进一步分析,以确定所用方法的有效性.结果表明,无标记依存句法分析精度提高了5.5%,有标记依存句法分析精度提高了7.5%.Aiming at exploring the possibility of increasing the parsing accuracy by linguistic means, an experiment of Chinese dependency parsing is conducted by using MaltParser and a self-buih treebank. Through the detailed analysis for the parsing results, the possible suggestion about improving the performance of the parser is provided and it is used as the guidance to modify the annotation scheme of the treebank. Experimental results show that the accuracy of unlabeled dependency attachment score increases 5.5%, and the accuracy of labeled score raises 7.5%.

关 键 词：依存句法分析 树库 中文信息处理 标注体系 语料库语言学 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]