High accuracy Chinese predicate recognition method combining lexical and syntactic feature

A method which merges lexical with syntactic features and combines C4.5 algorithm and rules was proposed for the systematic study of Chinese predicates. The method extracts lexical and syntactic features respectively. According to the lexical features, the suspicious predicate and its number are obtained. The lexical features are filtered by manual rules to identify the predicate which conforms to the rules. Basing on the lexical and syntactic features, the ones which do not conform to the rules are identified using C4.5. On the basis of Beijing forest studio-chinese tagged corpus(BFS-CTC) whose total number of predicates is more than 20 000, features and parameter choice experiment, syntactic features verification experiment, the amount of training data choice experiment and the method precision experiment were carried out to study the relations between predicate recognition results and the factors including lexical and syntactic features, function of the syntactic features and the amount of training data. The results show that the syntactic features effectively improves the effect of predicate recognition, as the amount of training data increasing the precision is convergence and the high precision reaches 99%.