现代汉语词义分类体系的建立和自动标注 = Building Word Sense Taxonomy and Automatic Annotation For Mandarin Chinese

Abstract

In this dissertation, we created word sense taxonomy for Chinese noun, verb and adjective, in terms of natural language processing. Then conduct automatic word sense class annotation, generating Chinese word sense class corpus. We have operatable definition for each class in the taxonomy, and make it suitable for corpus annotation, which are distinguishing characteristics of our word sense taxonomy. We study the issue of word sense taxonomy in the frame of distributional theory, semantic selectional restrictions theory and syntagmatic theory. Each class in the word sense taxonomy is defined with three types of features: syntactic performance, semantic role (for noun)/ argument structure (for verb), and semantic selectional restrictions. The result shows that the description of word sense class definition makes the taxonomy operateble in the process of sense class annotation. The methodology applied for building the taxonomy is one of the contributions we made through this dissertation. Automatic classification experiments are performed for multi-class words. The result of the experiments is quite encouraging: 84.1% words get the precision of over 90%; 96% words get the precision of over 85%. The disambiguation results show that the word sense taxonomy has enough distinction among sense classes, and verify that the word sense taxonomy is applicable in automatic annotation.