Deterministic Chinese Dependency Analysis Based On Support Vector Machine

Syntax analysis is the crucial section of natural language processing and machine translation. The mission of syntax analysis automatically outputs sentence structure and sentence relations, which turns the linear sequence sentence to the structural syntax tree. According on the different grammar system, analysis results represent different format. The paper research dependency syntax.Chinese dependency relation resulting from the dependency analysis can represent the further syntax relations between the words in the sentence based on dependency constrains. Word is the smallest element of sentence, the dependency based on words analysis can represent deep syntax relation so that the paper research Chinese dependency relation between each word. Based on dependency axiom, defined the thirty-eight dependency types standard of Chinese words for corpus by nature language processing laboratory, Dalian university of technology.This paper proposes deterministic Chinese dependency analysis method which considering long-distance dependency. Because Nivre algorithm have been used for English dependency analysis, and the syntax structure also resemble between Chinese and English. So choose the Nivre algorithm for the deterministic algorithm. Deterministic Chinese dependency analysis is to parse a sentence only deciding whether the current word modifies words immediately. beside it. However in some Chinese sentences, the children of the focused word may be far away from it. It is difficult to parse this condition with conventional deterministic dependency parser. The proposed method parses a sentence deterministically with consideration of long-distance dependency. Support Vector Machines is applied to identify Chinese dependency.Experiments using the Harbin University of Technology Corpus show that the method outperforms previous system by 5.32% accuracy. The dependency accuracy achieves 78.30%. The close test dependency accuracy achieves 97.64%. The results prove that the method proposed in this paper fit for Chinese characteristic and achieve a better parsing accuracy.