Research On Chinese Parsing Based On Semantic Analysis And Its Implementation

In this thesis, the semantic level of natural language processing is researched and a hierarchy to represent Chinese semantic knowledge is proposed and then is used in a Chinese parsing system based on semantic analysis.Semantic analysis is the basis of natural language understanding. In this thesis, different semantic theories and knowledge representation methods are discussed; and after the knowledge expression method of knowledge graphs is improved, a Chinese semantic knowledge hierarchy based on knowledge graphs is proposed and is constructed. The hierarchy is composed of semantic unites and semantic-kind rules. The semantic unite includes word graphs, phrase graphs and sentence graphs. Word graphs are generated according to HowNet; phrase graphs are combined by word graphs when phrases are parsed; sentence graphs are combined by phrase graphs according to sentence partten rules. The semantic-kind rules include semantic rules and sentence pattern rules.After the Chinese semantic knowledge hierarchy is built, semantic knowledge is applied to basic semantic computation. The method to compute word similarity and the method to compute word relevancy are presented and are applied to the work to select the best semantic rule; also the method to compute knowledge graph value is proposed and is applied to the work to select the best knowledge graph; and then the method to compute phrase similarity and sentence similarity is.proposed. When computing, both syntax structure and semantic relation are given attention to and the different weights are distributed according to the different syntax structures and to different the semantic relations which are expressed in knowledge graphs, so phrase similarity computation is made reasonable, and so is sentence similarity computation.Then according to the idea of chunk parsing, the Chinese parsing system based on semantic analysis is designed and is implemented. First, the system identifies the basic predicate verb phrases and identifiers the extended predicate verb phrases; then identifiers and analyses the preposition phrases and the phrases of the other kinds andso on; in the end, analyses the sentence according to sentence pattern rules. When parsing phrases, both syntactic knowledge which is expressed by frames and semantic knowledge which is expressed by knowledge graphs are combined into the extended context free grammar. Each syntactic production has two precondition judge functions. When the analyzer needs to use a production to do reducing, one precondition judge function corresponding to the production is firstly activated and in it the corresponding syntactic parsing is performed. If the precondition judge function is successful, the other precondition judge function corresponding to the production is secondly activated and in it the corresponding semantic analysis is performed. The reduction can be done only if the syntactic parsing and the semantic analysis are reasonable, otherwise the current analysis tree should be terminated. When reducing, the frames of syntactic structures and the knowledge graphs of semantic information are generated together. So, both syntactic parsing and semantic analysis lead to the actions of the analyzer of the system.In the end, our work is summarized and future research direction is pointed out.In summary, our work has gone deep into the semantic level of Chinese analysis from the syntactic level. We have built the integrated and systematic Chinese semantic knowledge expressions, and use semantic knowledge in Chinese paring. The parsing result from the testing corpus is satisfying. What we have done provides a good theoretical and practical foundation for future research.