A Study On Chinese Discourse Coherence Based On Frame Semantics

Chinese discourse coherence theory and technology research is an important research task in the field of discourse analysis.From the perspective of linguistics,discourse coherence refers to that discourse is organized by various concepts expressed in the discourse.From the perspective of computer operability,we need a system description to express the discourse coherence,and carry on the related research in this foundation.On the problem of the representation of discourse coherence,this paper constructed a coherence description scheme for Chinese discourse which was based on Chinese FrameNet(CFN).At the same time,we built a corpus for expanding discourse coherence technology research.According to the description scheme,Chinese discourse coherence technology research mainly includes three subtasks:discourse segmentation,discourse structure construction and discourse relation recognition.On the self-built corpus,automatic analysis of the three subtasks has carried on the sentence.The main research contents and research results are as follows:(1)For the task of representation of discourse coherence,this paper constructed a coherence description scheme for Chinese discourse which was based on Chinese FrameNet(CFN).The discourse coherence was transformed into a computable frame semantic problem,so as to provide appropriate presentation mechanism and calculation basis for discourse coherence.(2)For the problem of insufficient for Chinese coherence corpus,this paper built a corpus which contains 496 discourses,and carried on artificial consistency check.The construction of corpus not only solved the problem of insufficient corpus,also provided resources for further Chinese discourse coherence analysis research.(3)For the task of discourse segmentation,this paper combined with characteristics of Chinese punctuation and frame semantic to develop a series of rules,implement the segmentation of discourse.Experimental results showed that frame semantics can effectively segment discourse unit.(4)For the task of discourse structure construction,maximum entropy classifier was applied to recognize the relation existence between discourse units with dependency parser feature,syntactic parser feature,target feature and frame semantic feature respectively.Then,we used probability of the relation existence between discourse units to construct the discourse structure by greedy bottom-up algorithm.Experimental results showed that frame semantic feature can improve the performance of discourse structure construction.(5)For the task of discourse relation recognition,maximum entropy classifier was applied to recognize the class of discourse relation with lexical feature,dependency parser feature,syntactic parser feature,target feature and frame semantic feature respectively.Experimental results showed that frame semantic feature can improve the performance of discourse relation recognition.Studies on Chinese discourse coherence,this paper proposed description scheme for Chinese discourse based on frame semantic,and carried on experiment on self-build corpus.The experimental results proved that frame semantics has excellent function in discourse coherence.It can not only represent the discourse coherence from the formal,but also can effectively improve the accuracy of discourse coherence on three subtasks.The research on discourse coherence offered a new discourse coherence description system and technology for discourse analysis field,and provided a strong support for other natural language processing research.