Identification Of Dependency Relations Between Clauses Within Clause Complexes Based On The MOOD System

Having clause as the fundamental grammatical unit,Systemic Functional Linguistics regards the three Metafunctions,which is made up of systems of Theme,MOOD and Transitivity,as the starting points to analyze various configurations selected by different parts of the clause to achieve syntactic and semantic functions.The three Metafunctions in Systemic Functional Linguistics were proved to be powerful in the analysis of various types of discourses.However,it should be noticed that they were often used to explore constituents on or below the clause level.When investigating the level above the clause,such as clause complex,two other systems were put forward,which are the taxis and logico-semantic relations.Theoretically speaking,these two systems seem to be rather comprehensive.However,when used to analyze texts,the logico-semantic relation was in fact proved to be quite complex and vague.Therefore,this study intended to re-classify the relations between clauses in clause complexes based on the Interpersonal Metafunction in Systemic Functional Linguistics.Clause automatic identification has always been paid much attention in the field of Natural Language Processing.However,most researches focused more on the improvement of the identification performance from the perspective of the algorithm.And rarely have they attempted to explore whether different grammatical system might have an impact in this regard.Therefore,this study intended to investigate the Conditional Random Fields(CRFs)'s automatic clause relation identification performance,under the guidance of the clause relations re-classified based on the Interpersonal Metafunction.By the light of Halliday's definition of clause constituents from the perspective of the MOOD system,which is the principal grammatical system of Interpersonal Metafunction,this study re-classified the relations between clauses within clause complexes and tagged a corpus of business domain to investigate the reclassified result.The tagging result showed that the new classification was scientific and reasonable.Then based on that,the automatic identification performance of the CRFs model was tested,and a positive outcome was achieved,with an F-measure of 85.04%,Precision of 85.55% and Recall 84.53%.To be specific,the CRFs model performed especially well at the automatic identification of the complement clause and adjunct clause.The automatic identification of the subject clause as well as the appositive result were not so satisfactory,but the main reasons could be attributed to the small number of these two kinds of clause relations in this corpus.In addition,through the analysis of the complement clause and adjunct clauses within this corpus,this study stepped further to investigate the factors that might have an impact on the automatic identification results.The major contribution of this study is to provide insight to the exploration of different types of clause relations within clause complexes from the perspective of MOOD system,which might enrich the functional analysis of discourses.Besides,some factors are also investigated through the analysis of the complement clauses and the adjunct clauses,in the hope of inspiring those who are interested in the field of business texts automatic identification as well.