Research On Identification Method Of Chinese Phrase Based On Semantic Analysis

Chinese base phrase identification and analysis are one of the important tasks of natural language shallow parsing. The analysis results of base phrase make the sentence hierarchy clearly and lay the foundation for syntactic analysis further. Chinese noun phrases and prepositional phrases account for a large part of Chinese base phrases type. The recognition rate increasing about base noun phrases and prepositional phrases has an important significance on machine translation and information retrieval. This thesis focuses on the task of shallow parsing and employs various methods of machine learning to identify noun phrases and prepositional phrases in texts. We also construct different identifying system and acquire satisfying result. The research of this paper as follows:Noun phrase recognition: research on the basis of noun phrases recognition employing the maximum entropy and transformation rules, this paper presents the recognition of noun phrase based on the maximum entropy and transformation rules combined. We construct the basic framework for the maximum entropy while using it to construct the noun phrase recognizing model. According to the structural characteristics and context features, we extract features firstly, then select some from them, lastly estimate parameter's value for every feature. So we have established a maximum entropy model recognizing noun phrase. When we recognize noun phrase based on transformation rules, we learn rules making full use of the context, so get the ordered sequence rules identified the noun phrase. Through analyzing characteristic of maximum entropy and transformation rules, we present the method that recognizes noun phrase based on combining both them. Experimental results improve the speed and accuracy of recognition system.Prepositional phrase recognition: Based on research noun phrase recognition employing maximum entropy, we proposed a maximum entropy-based method for automatic identification of prepositional phrase, which focus on multiple error identification problems on the right boundary of prepositional phrase. It combines the dependency grammar knowledge of prepositional phrase boundary. In the process of recognition, firstly, we apply maximum entropy to identify prepositional phrase, then fine-tune the results with dependency grammar knowledge generated by dependency Treebank. It improves the recognition rate of prepositional phrase.