Study On Dependency Parsing For Chinese Based On SVM

Syntactic analysis is one of the most important tasks in Natural Language Processing. This paper aims to automatically identify the syntactic structure of Chinese according to the given grammar and parses the relationship between the syntactic units.There has been a steadily increasing interest in syntactic parsing based on depen-dency parsing in recent years. By far the most effective strategy of dependency paring is to combine training algorithm based on the classifier with the model of determinest-ic parsing. So this paper constructs a dependency parser for Chinese based on SVM.The algorithm is based on Nivre'Arc-eager algorithm.However, the shortcomin-gs of Nivre's lead to the problem of Early-reduce when dealing with the right long-distance dependency, so we will improve it. The parser' main ideas is as follows:This paper does pre-processing before parsing sentences to reduce the error propagation caused by the greedy characteristic of deterministic parsing. First this paper utilizes SVM to construct a root finder to divide a sentence into two sub-sentences and then extracts the prepositional phrases from sub-sentence. The pre-processing can decrease the complexity of the sentence and improve the parsing accuracy consequently. Improved Nivre's algorithm is adopted to parse sub-sentence lastly. In Chinese, only prepositions and verbs have right-side dependents. For the problem of Early-reduce caused by prepositions, this paper resolves it by extracting the prepositional phrases.This paper imports the global features and defines a new action—Verb-Shift to settle the problem of Early-reduce caused by verbs.Finally this paper does a comprehensive evaluation for the performance of the parser. Experimental evaluation shows the strategy improves the accuracy obviously.