基于《知网》义原搭配的中文文本语义级自动查错研究  

Study of semantic automatic error-detecting for Chinese text based on sememe matching of Hownet

作　　者：郭充[1] 张仰森[1] 

机构地区：[1]北京信息科技大学计算机学院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17期3924-3928,共5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：语义级的自动查错一直是汉语文本自动校对技术的难点。针对汉语文本中的语义搭配错误,提出了一种基于《知网》义原搭配的有效的自动查错方法,主要包括语义知识库的构建和自动查错算法。语义知识库的构建分为两步:①根据规则从训练语料中统计出大量的动词与名词之间的二元搭配组合,并利用互信息等因素筛选记录;②利用《知网》对词语义项的义原描述(对于动词抽取其义项中的主义原,名词抽取其义项中的主义原和常见的4个辅助义原),将动词与名词的二元搭配组合转变为义原之间相互制约的多元组合。在语义知识库的基础上设计相应的自动查错算法。实验结果显示,该方法的召回率为35%,精确率为82.3%,具有较好的应用前景。The study of automatic proofreading of Chinese text is always very difficult at the semantic level.The semantic matching error of Chinese text is discussed and an effective automatic error-detecting method is proposed based on the sememes matching of Hownet.This method includes two parts： the constructing of the semantic repository and the algorithm of automatic error-detecting.The procedure of constructing the repository is： ① Statistics huge amounts of the binary-combination of verbs and nouns from training corpus by a group of rules,than filter the records through the factors： mutual-information etc.② Use the sememe descriptions of words’ meanings defined in Hownet（for verbs choose the main sememe and for nouns choose the main sememe and the other four common sememes） to convert the binary-combination of verbs and nouns into the restricted multi-combination of sememes.The automatic errordetecting algorithm could be devised appropriately on the basis of the semantic repository.The experimental result shows that the recall rate is 35% and accuracy rate is 82.3%,which reflects the good application prospect.

关 键 词：语义搭配错误 互信息 知网 义原 语义知识库 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]