基于MDL的汉语语义选择限制自动获取  

Acquiring Chinese Selectional Preferences Using the MDL Principle

作　　者：贾玉祥[1] 李育光 昝红英[1] JIA Yuxiang;LI Yuguang;ZAN Hongying(School of Information Engineering,Zhengzhou University,Zhengzhou 450001,China)

机构地区：[1]郑州大学信息工程学院,河南郑州450001

出　　处：《郑州大学学报（理学版）》2018年第1期66-71,共6页Journal of Zhengzhou University:Natural Science Edition

基　　金：国家自然科学基金项目(61402419);国家社会科学基金项目(14BYY096);国家高技术研究发展计划863课题(2012AA011101);国家重点基础研究发展计划973课题(2014CB340504)

摘　　要：语义选择限制是一种重要的词汇语义知识,有助于自然语言的句法语义分析,也有助于解决自然语言处理中的数据稀疏问题.提出了基于语义分类体系和最小描述长度原则的汉语语义选择限制知识自动获取方法,对现有的名词语义分类体系进行改造,实现了一个知识获取的高效算法.基于大规模语料进行知识获取,对获取的优选语义类进行了分析,并进行了伪消歧实验,与基于KL距离的方法进行了对比,体现了所用方法的有效性.Selectional preference was an important lexical knowledge which could be applied to syntactic and semantic analysis of natural languages and solving data sparseness problems.A method for Chinese selectional preferences acquisition using a semantic taxonomy and the MDL principle was proposed.Firstly,modifications were made on the existing noun taxonomy and a high efficient algorithm was implemented.Then,knowledge acquisition was carried out on a large scale corpus.The effectiveness of the proposed method was shown by the analysis of the acquired preferred semantic classes and the comparison with the KL divergence-based method on the pseudo-disambiguation experiments.

关 键 词：语义选择限制 词汇知识获取 最小描述长度 语义分类体系 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]