The Automatic Identification Of The Semantic Core Words For Frame Elements

Frame Semantic Dependency Graph based on the Chinese FrameNet is an effective way to formalize the Chinese sentence. Frame Kernel Dependency Graph extract kernel elements from Frame Semantic Dependency Graph in order to achieve deep semantic understanding of a Chinese sentence. This paper explores extracting the semantic core words of Frame Elements to obtain the Frame Kernel Dependency Graph.The main work of this paper is as follows:(1) Systematically discuss the identification of semantic core words of Frame Elements and propose to use machine learning methods based on statistical theory, Such as Conditional Random Fields model and Maximum entropy model and Support Vector Machine model to extract the semantic core words of Frame Elements.(2) For three different models, feature set selection is analyzed. On the basis of the base features, different feature templates setting were compared. Lastly, the optimum template and model is selected to the better method for identification of semantic core words of Frame Elements.(3) Improvement of experiments. To further improve the efficiency of recognition, we explore to improve method of feature template on CRF model. When choosing the context information, the information content words long-distance notional is selected. This approach makes the identification certain improvement in efficiency.Experimental results showed that CRF model has better performance to the identification of semantic core words of Frame Elements. The average precision of experimental result achieved 97.34% and 94.03% for frame elements of simple and complex phrase type respectively.