Research On Deterministic Dependency Parsing Method Based On Corpus Statistics In Chinese Compound Sentences

Chinese syntactic parsing is one of the interesting and challenging topics In the field of Chinese information processing.The task for it is:for a given Chinese sentence,automatically infer its syntactic structure according to the specific syntax,and ensure that the logic and semantic relation between every grammatical units in sentence.It is more and more widely used in all kinds of NLP system, for example: machine translation, question answering system, text mining,information retrieval etc.Syntactic parsing technology is divided into the two:based on rules and based on statistics.Statistical methods based on large-scale corpus and mathematical model does not rely on subjective prior knowledge,and algorithm and knowledge are separated.This paper mainly uses statistical dependency syntactic parsing based on policy decision.In addition,syntactic parsing of Chinese sentence not only need to derive the dependency relation between every syntactic units,also need to identify the hierarchical relationship between the clauses,to correctly grasp the sentence semantics,and be better applied to the language processing.Usually,there are more words in Chinese compound sentences.parsing for Chinese compound sentences is easy to produce error accumulation in long range right dependency.This paper,to solve it,firstly put forward a kind of "divide and conquer strategy" for analysis of syntactic parsing in Chinese sentence:get the clause recognition for a complex sentence and automatically identify the logical and semantic relationship between the clauses;then extract the clauses,and translate complex process of syntactic parsing for a complex sentence into the analysis of each of its clauses,finally integrate the analysis results of each sub process on the basis of the hierarchy and get the complex sentence dependency relation graph. Secondly,this paper put forward an analysis model for sentence hierarchy with the combination of semantic and rules:according to the analysis of the hierarchy relationship of marked complex sentence,in order to solve the accuracy decline of automatic recognition caused by the inadequate relationship mark,this paper analyzed formal semantic knowledge that effect sentence association,on this basis constructed Small sentence Association Recognition algorithm and apply it to the corresponding complex sentence hierarchy decision rule,to assist the analysis of the hierarchical relationship;proposed a model of marked complex sentence hierarchy analysis, combined with semantics and rules. Thirdly,in the basis of the above clause partition and according to "divide and conquer strategy" further improved the Nivre algorithm.Experimental results show that:deterministic dependency parsing method in this paper,than original deterministic parsing algorithm,improved the accuracy of parsing to a certain extent.