Discourse Measurement Method Based On Quantitative Analysis Of Clause Relations

In the past ten years,with the development of computer science and information technology,computational linguistics goes through a rapid development in the fields of machine translation,voice recognition and human-computer interaction.However,limitations of these artificial intelligence products still exist.Take machine translation as an example,it seems incompetent when dealing with a long discourse,only providing a lousy translation.The fundamental reason is that the cognitive ability of a machine is based on binary information and simple algorithms,while people's understanding of discourses is based on the understanding of the relationship between clauses.On the other hand,different discourse genres are distinct in distribution characteristics of clause relations according to Ding Jianxin and Chen Anling,which provides a theoretical basis for discourse genre recognition based on clause relations.On the basis of extensive literature research,this dissertation selects the clause relations of Halliday's systemic functional linguistics as the breakthrough point.Taking both the advantages of Halliday's clause complex theory and the problems of his classification framework of the clause relations into account,this dissertation adopts Cheng Xiaotang's improved framework of clause relations,and puts forward two important concepts,Characteristic Matrix of Clause Relations(CMCR for short)and Correlation of Discourses(COD for short),as well as a discourse measurement method based on quantitative analysis of clause relations.In Halliday's systemic functional linguistics,clause complex is the largest unit,and clause relations contain a wealth of discourse information.Therefore,we can analyze the textual features in a discourse comprehensively by means of detecting its CMCR.COD is a quantitative description of correlation between different discourse genres from the perspective of statistics.Extended from COD,discourse differences can reflect the clause relation distribution differences between different genres from another side,and illustrate specific differences intuitively.With the introduction of these concepts and methods,we can implement the large-scale training for the machine on discourse genres,thus to realize the automatic genre recognition and classification of massive text.This dissertation mainly combines quantitative analysis with other research methods,such as statistical analysis,case analysis,deduction,induction,and literature retrieval,etc.In the measurement procedures,different discourses are chosen as the sample data and the corresponding CMCRs are chosen as the model parameters.After analyzing clause relations in the sample discourses,we can get the CMCRs we desire.Subsequently,the CMCRs are processed for error correction firstly and then calculated together for correlation test.At last,COD,a weighted average of semantic,rhetorical and projection correlation coefficients,is figured out to characterize the similarity between these two discourses.The whole procedures establish a systematic discourse measurement method based on quantitative analysis of clause relations.Through the actual verification of several discourses,the result is in good accordance with the expectation,and fully illustrates the rationality,correctness and feasibility of the theory.This discourse measurement method based on quantitative analysis of clause relations proposed in this dissertation can not only infer meaningful linguistic information about the discourse itself at a micro level,but also give a quantitative description of similarity between different discourses at a macro level.This method has great maneuverability and application value in automatic discourse analysis,which opens up a new research approach for scientific,objective and systematic discourse analysis.