Research On The Models Of Computational Pragmatics Based On Data Mining

This dissertation establishes corresponding computational models for the existing typical theories in pragmatics,such as deixis,speech act theory and linguistic adaptation theory.By comparing with adjacent disciplines such as formal pragmatics,the research scope of computational pragmatics between computer science and technology and pragmatics is clarified.The Bayesian belief network model of indirect speech act theory,the expectation maximization context estimation model of dynamic adaptation process and the association rule mining model of anaphora phenomenon are established in this thesis.The Bayesian belief network model of indirect speech act theory studies the relationship between context and discourse from a static perspective.Since the indirect speech act and the Bayesian formula are abduction reasoning,a naive Bayesian model of indirect speech act can be constructed.To break the restriction of the class independent conditions in the Bayesian formula,the Bayesian belief network model is built.This model is mainly composed of two parts: the internal cognitive process and the external social and cultural influence.In order to improve the accuracy of the model,a belief network of cognitive processes is constructed based on the idealized cognitive model in cognitive linguistics.The example shows that the model is consistent with pragmatic theory and objective reality.The expectation maximization context estimation model of dynamic adaptation process explains the relationship between context and discourse from a dynamic perspective.If the context and chosen discourses are treated as distributions and samples respectively,they will satisfy the Gaussian mixture model according to the central limit theorem.In this thesis,the expectation maximization algorithm is used to describe the language selection and its adaptation process.The temporal characteristics of the model are reflected by hidden Markov chain,and context can be described by infinity and hierarchical abstraction.The above two models were validated by a corpus analysis of The Berkeley Coffee Shop.The association rule mining model of anaphora phenomenon finds the association rules existing among the properties of the anaphora through experiments.Based on these rules,the differences in the use of anaphora in news and speech are compared.Those experiments prove the operability of computational pragmatics,and the conclusions are consistent with the principles in pragmatics and stylistics.A preliminary exploration of computational pragmatics is conducted in this dissertation from a new perspective,trying to combine computer science with pragmatics which contains strong subjectivity and many influencing factors.It provides a reference for the establishment of a computational model of disciplines with strong human characteristics.