利用浅层句法分析提取特征的词义消歧  

Chinese WSD based on features obtaining with shallow parsing

作　　者：孙超[1] 张仰森[1] 

机构地区：[1]北京信息科技大学智能信息处理研究所,北京100192

出　　处：《计算机工程与设计》2010年第21期4704-4707,共4页Computer Engineering and Design

基　　金：国家自然科学基金项目(60873013);北京市自然科学基金B类重点项目(KZ200811232019);北京市属市管高校人才强教计划基金项目(PXM2008_014215_055942)

摘　　要：针对如何从文本中提取高质量消歧特征的问题,提出了基于浅层句法分析的消歧特征提取算法,建立了以语块分析识别为核心的特征提取模型。该模型通过对实词类型语块识别、分析中心词语词性和虚词类型语块分析,得到多义词的消歧特征。以北京大学计算语言研究所的现代汉语基本标注语料库为基础,选取了44个多义词,通过使用最大熵消歧模型进行训练和预测实验,准确率达到了78.71%。To solve the problem of obtaining high-quality features in text, the features obtaining algorithm based on shallow parsing is proposed, and the features obtaining model with chunk analysis and identify as core is constructed. By identifying substantive chunks, analyzing center word and part of speech, and identifying empty word chunks, the disambiguation features ofpolysemy are obtained. On the basis of Word-Sense Tagging Corpus of Institute of Computational Linguistics, Peking University, there are 44 polysemies selected. On the experiment of trainning and predicting using maximum entropy disambiguation model, the accuracy rate of disambiguation arrives at 78.71%.

关 键 词：词义消歧 浅层句法分析 语块 特征提取 最大熵模型 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]