| Parsing is a key issue in the field of natural language processing. The goal of parsing is to derive the syntactic structure of a sentence automatically according to a certain grammar. In recent years dependency parsing gradually be taken seriously. However, dependency parsing has more research on simple sentence, but pays less attention to compound sentences. In the field of Chinese studies, compound sentences is an important entity of grammar, the semantic information it expresses is rich and complex, therefore it has more important research value in the field of dependency parsing.Based on the above reasons,the mainly purpose of this paper is to take dependency parsing on coordinative compound sentences in CCCS. The coordinative compound sentences in CCCS is mostly long sentences. In order to solve the difficult problem of dependency parsing from long sentences, combine with the characteristics of the coordinative compound sentences, Firstly, according to the relation words collocation and comma to divide coordinative compound sentences into two sentences, which can reduce the length of compound sentences in a degree. However, after the division,there still exists long sentence in clauses. By the definition of dependency grammar, there is no cross dependencies phenomenon on both sides of the root word in Chinese sentences. So A root finder was constructed by machine learning method,which can find the root word of the clause. then the root word can divide the long clause again. After two times division, which can not only reduce the length of compound sentences better, and reduce the complexity of sentence dependency parsing, but also improve the accuracy of dependency parsing on coordinative compound sentences.In the analysis of compound sentences’s dependency relation, we select the Arc-eager algorithm which based on deterministic dependency parsing. However, In Chinese, this method is prone to the Early-reduce problem of verbs and verbs as well as verbs and prepositions. As a result, the improved two segment dependency parsing method is used to solve the two Early-reduce problem.Before the start of the experiment, select 2000 sentences from CCCS to annotate their dependency relations, which act as the coordinative compound sentences tree bank in experimental use. Then the experiments and experimental results of the root finder and the dependency parser are carried out on different number of training sets and test sets. Compared with one analysis and two stage analysis, it proves the validity of the two stage analysis. |
