The Study, Based On Chinese Syntactic Subcategorization Analysis Approach

Parsing is a fundamental problem in natural language processing. The main task of parsing is to determine the input sequence of words whether can constitute a sentence to comply with grammar rule, that is, to construct the hierarchical constituent structure of a sentence. Through using grammar rule and other knowledge of natural language processing, it changes the input sentence order between words from the linear order into a non-linear data structure. It has many applications, such as machine translation, information retrieval, automatic summarization, it has a important role . About based statistical syntactic parsing methods, there are two important issues. One is to establish statistical model of Ambiguity Resolution parsing algorithm, the other is the design of syntactic parsing postprocess, they determine the syntactic analysis system accuracy and efficiency. The paper is mainly engaged in the work of the syntactic parsing aspects of postprocessing, which based on verb subcategorization for Chinese syntactic analysis. The main research work is as follows:1. We study existing common statistical parsing model and syntactic analysis algorithm as well as its comprehensive analysis and comparison, and also explain the verb subcategorization syntactic model theory, formal description, syntactic trees and the corresponding relationship of verb subcategorization.2. With the statistic parsing model outputing the n-best parsing trees, by using the transfer-based error-driven learning approach to improve verb subcategorization extraction, we re-sort to find the optimal parsing tree and conduct syntactic analysis postprocessing experiment of rule-based verb subcategorization.3. In order to find the improving optimal solution of the syntactic analysis, we use the transfer-based error-driven learning approach to make better statistical verb subcategorization extraction. Moreover, we do the experiment using syntactic analysis postprocessing method based on the statistical verb subcategorization.We perform experiments on the CIPS ParsEval-2009 dataset from Tsinghua Chinese Treebank (TCT). Experimental results show that our final result is an F1-score of 88.759 %, close the previously best reported systems reported F1-score of 88.77% in the CIPS ParsEval-2009. This convincingly demonstrates the effectiveness and accuracy of our proposed Verb Subcategorization-Based parsing postprocess for Chinese language.