Research Of Chinese Syntax Analysis Based On Machine Learning

Syntax analysis is one of the important issues of natural language processing. Inrecent years, the dependency syntax analysis gets more and more attention. Relativeto the research of the English dependency analysis, Chinese dependency parsingtechnology is still in its infancy. So the methods of Chinese dependency parsing areinvestigated based on the Chinese dependency parsing research results and the uniquecharacteristics of Chinese grammar in this thesis. The main goal of this paper is totake the strategy of dependency parsing which is to combine training algorithm basedon the classifier with the model of deterministic parsing to solve some problems ofChinese dependency parsing with prepositional phrases.This paper first describes some existed dependency parsing algorithms，amongwhich Arc-eager algorithm and RPP algorithm are studied in detail. In order tosolve the problem of Early-reduce caused by Arc-eager algorithm, the RPP algorithmproposed by Yao is further improved in this paper. Arc-eager algorithm can only dealwith the the dependencies between the adjacent words and the RPP algorithm alwayshave erros when identify the Prepositional phrase which contains verbs. Whenananlysis the dependencies of Chinese sentences, this paper identify the boundary ofthe prepositional phrase by drawing on from the prepositional phrase follow-up wordfeatures,and then improve the RPP algorithm. The improved parsing method can notonly solve the problem of the Early-reduce when analysis the prepositional phraseslong-distance dependencies，but also solve the boundary identification problem of theprepositional phrase which contains a verb.Finally, by using the Harbin University of Technology Corpus as theexperimental data and using the syntax parser as the experimental tool, this paper doesa comprehensive evaluation for the strategy proposed by this paper. Experiments show that this strategy is effective to improve the accuracy.