Cognitive Construction Grammar Oriented English Verb Pattern Recognition

There are large amount of exceptional linguistic phenomena in natural language,such as the proverbs,slangs,idioms etc.These types of phenomena are quit difficult to be parsed by traditional syntax parsers or by framework semantics.Motivated by explaining these linguistic phenomena,the cognitive construction grammar has been proposed and then widely accepted as one of the most important and promising linguistic theory.By cognitive construction grammar,a construction is defined as the basic semantic unit of a language.Different from traditional framework semantics and productive syntax,the scale of basic semantic unit is unlimited.So it is impossible to manually construct the full set of constructions for a natural language,which brings a big challenge: how to discover and mine constructions of a language.Obviously,this challenge can not be solved easily.Recently,to seek for the solution of discovering constructions,researchers focus on the mining of English verb patterns.This thesis is just following this research work and is trying to put forward the method to mine English verb patterns.The main work of this thesis include two parts: Researching on the discovering of typical English verb patterns via semantic clustering methods;Constructing the English verb patterns through the semantic tagging of construction arguments.Since the contexts of the same verb are various in different sentences,it is a really challenging task to cluster different contexts of the same verb into the same cluster.To address this issue,we have compared numbers of sentence semantic representation methods,different sentence distance computing methods and two clustering methods.Finally,we have proposed the verb pattern clustering method based on the word mover's distance(WMD)and R&L clustering method.Our experimental results show that the proposed method has reached the best performance among all compared methods.Based on the verb pattern clustering results,we further research on the tagging of arguments in each verb pattern.The purpose of this step is to construct the abstract patterns of the verbs.The support vector machine(SVM)and conditional random field(CRF)method are used as the semantic tagging methods.Comparing with our experimental results and the results published on Sem Eval 2015 shows that,the proposed method of cognitive construction grammar oriented English verb pattern recognition obviously increases the precision and the result of verb pattern clustering is reached a 21.6% of performance gain.