Research And Application On Semantic Analysis Of Chinese Terms

Chinese semantic analysis is one of key technologies of natural language processing, and contributes to well comprehend Chinese. The improvement of Chinese semantic analysis will play an important role in information retrieval and machine translation. As the development of technology, terms appear continuously, so the semantic analysis of terms is of great significance.Based on the research of the term, the paper introduces a term semantic analysis system. The system includes two parts: dependency analysis and semantic analysis. The two parts are based on the method of machine learning. Based on the semantic analysis, the Chinese term translation system is realized, and the specific contents are as follows:Firstly, we analyze a lot of terms, and find out most of terms are noun phrase and low recurrence rate of terms, so we choose the proper features for terms in the dependency analysis stage based on the two features and we train the support vector machine (SVM) dependency model to identify the dependency relationships within terms.The proper features include basic feature, mutual information and the first sememe of words in hownet,Secondely, this paper proposes an approach for Chinese term semantic analysis. Firstly, we define 14 semantic relationships and then train CRF model to identify the semantic relationships between two words. But the range of semantic relationship within terms is so finite that the semantic model cannot comprehensively identify the confusion categories. So we train SVM model to solve this problem. After the CRF model outputs 2-best semantic results, we use SVM model to identify the final result in 2-best results.At last, the result of the semantic analys is applied to term translation. In the first stage, we extract the constituent pharses based on the result of the dependency analysis and extract the non-constituent phrases by the method of GROW-DIAL-FINAL, and then abstract the ranking template. On the basis of the ranked source language, we use Moses to decode the ranked terms. Experimental results show that the method is effective, and the accuracy of semantic analysis reaches 77.13% and 69.05% respectively in parent-category and sub-category. the result of the semantic analys is applied to term translation and the translation result are better than before.