基于TCT语料库的汉语词类原型及隶属度计算研究

Research on Chinese Lexical Prototype and Membership Calculation Based on TCT Corpus

摘要: 汉语词类划分在汉语语法学界和计算语言学界都是一个棘手的话题。本文基于袁毓林“汉语词类是模糊集合”的假设，基于语料库研究方法和特征工程的思想，利用TCT语料库中的句法结构标注，构建词的“句法位置–词频”向量。最后基于簇识别的思想，找出汉语词类的原型词，并通过原型词的“句法位置–词频”向量，计算每个词在该词类标准下的隶属度。该模型在九个词类大类研究中的平均F1值为81.079%，显示出了较好的效果。

Abstract: The division of Chinese word classes is a thorny topic in the field of Chinese grammar and computational linguistics. Based on Yuan Yulin’s assumption that “Chinese word classes are fuzzy sets”, this paper uses syntactic structure annotation in TCT corpus to construct the vector of “syntactic position-word frequency” of words. Finally, based on the idea of cluster recognition, we find out the prototype words of Chinese word class, and calculate the membership degree of each word under the standard of the word class through the vector of “syntactic position-word frequency” of the prototype words. The average F1 value of this model is 81.079% in the nine word classes studies, showing a good effect.