Japanese Dependancy Analysis Based On NN-LSVM

Japanese dependency analysis is to determine an optimal combination of dependencies based on dependency constrains. It is an important part of Japanese syntax parsing and plays an important role in many fields such as Japanese machine interpretation and sentence analysis. In consequence, research on the dependency analysis makes great theoretical and practical sense in the Japanese natural language processing. This paper puts focus on dependency analysis method based on NN-LSVM with the support of marked corpus .Japanese is a compact-structured language, in which dependency lies between two adjacent segments. This paper adopts segment stepwise analysis algorithm based on chunk to control the analysis. In the analysis procedure, only whether the direct adjacent segments has a dependency is checked instead of all segments pairs, which reduce s the number of candidate dependencies and the computation of the resolution. Consequently, the parsing accuracy and speed is improved.In this paper, a classifier based on SVM(Support Vector Machine) model is generated using the marked corpus, which in turn determines whether there is a dependency between two segments. To improve the parsing accuracy and speed, a method for Japanese dependency structure analysis that using NN-LSVM to prune large training set is proposed. This method first uses LSVM and NN to prune those samples that unused or not good to improve the classifier's performance, then obtain the parser.Experiments using the Kyoto University Corpus show that the segment parsing, accuracy can reach 89.23% and the sentence parsing accuracy is up to 47.82%; the parsing speed is 0.8 second per sentence. The results prove that the method proposed in this paper can resolve the dependency of Japanese and achieve a better parsing accuracy and speed..