Discourse Relation Discovery Based On Chinese FrameNet

Researching the sentential relations in discourses has always been a very important task in the field of Natural Language Processing(NLP).It is of great significance to understand the overall discourse semantics and the logical structure between sentences.With the development of NLP,the understanding of semantics has gradually shifted from lexical or sentence-level semantics to discourse-level semantics.Therefore,automatically identifying the sentential relations in discourses via computer has profound implications.This paper,on the basis of China National Science Foundation titled " Chinese discourse frame semantic network automatically extract and its semantic reasoning ",employs the knowledge base,Chinese FrameNet(CFN),builds two models to identify implicit sentential relations respectively based on neural network and frame relations and then,by fusing the two models to enhance the performance,develops an automatic frame-semantic extraction system for Chinese texts.The main work of this article is as follows:(1)by using the CFN frame elements and lexical information,sentence semantics are reprensented and then utilized as the input layer of the neural network model.Analyzing the sentential relations in discourses based on the self-building sentential relation system of Shanxi University and regulating the parameters of each layer in the neural network model,the best experimental results have been achieved.(2)After disambiguating the frames evoked by the sentences in discourses,sentential semantics are represented by a frame,and then a graph-related algorithm is used to construct the discourse-level frame-to-frame relations graph.Afterwards,the algorithms for mining association rules and frame-to-frame relations are used to identify the sentential relations in discourses.(3)Designing and implementing a preliminary system to automatically extract frame semantics for Chinese texts,and then using UI-related techniques to achieve the visualization of the frame-to-frame relations and the sentential relations in discourses.The system verifies the effectiveness of the methods proposed in this paper.At the same time,the system also reserves some interfaces for future maintenance and development.