Learning Chinese Attribute Nouns Using Lexico-Syntactic Patterns

Lexical knowledge sources or lexical ontologies are very important to the knowledge grid and semantic computing. Attribute information are key elements for defining concepts in lexical sources. This paper explores the idea of creating corpus-based attribute classifiers using lexico-syntactic patterns. Two novel attribute classifiers are proposed, one is a likelihood ratio classifier using hand-coded lexico- syntactic patterns, the other is a maximum entropy (ME) classifier exploiting automatic extracted patterns. The performance of the method is compared to both the direct pattern matching approach and the human performance, which indicates that the proposed method for attribute learning is very effective.