Computational Model Of Chinese Metaphor Recognition And Interpretation

Metaphor is not only a linguistic phenomenon but also a kind of cognitive phenomenon, spreading over languages and thoughts. It is regarded as the central problem of languages and thoughts. Researching on computational model of metaphor recognition and interpretation supplies an in-depth and dynamic perspective for natural language understanding.The dissertation addresses the research on Chinese metaphor recognition and interpretatioin at sentence level. It aims to meet the requirement on concise information with small granularity and formalism. The linguistic features and regulations of metaphorical sentences are analyzed. The metaphor-receiver-oriented computational system is established. A computable classification system of Chinese metaphor is then set up. A classified and schematized Chinese metaphor recognition system which can process large-scale metaphorical expressions is developed. A target-driven computational approach for metaphor interpretation is proposed, which can represent the literal meaning, acquire the similarities and interpret the true meaning of recognized metaphors. According to our theories and technieques, the first Chinese computational system CHMeta is implemented. It indicates that the computaitional theories and techniques applied in the system are valid and effective.The computational system for metaphor processing is established from the perspective of people who receive and understand metaphors. We propose that people are able to identify a metaphor without inferring its meaning or even without understanding its true meaning. Thus metaphor recognition is isolated from metaphor inference work and is carried out in its own right. Then the computational process of metaphor could follow the three-stage process: Recognition-Analysis-Interpretaion.On metaphor recognition, Metaphor Role Dependency Schema Matching Theory is developed which transforms complex metaphor expressions into combinations of semantic dependency schemas of metaphor roles and convert metaphor recognition into the range of pattern recognition. Chinese metaphorical units are classified into 32 class schemas, each of which has a special semantic dependency structure of metaphor roles and corresponding rules for measurement calculation and tagging. The computational classification system covers 95% of Chinese metaphorical phenomena. The 32 categories used to check potential metaphors are normalized and stored as class schemas. As for an input sentence it is first parsed and formalized as a metaphor role dependency schema and then the classifier will judge whether the input dependency schema contains one or more of the 32 class schemas by schema matching algorithm. Referencial Unconventionality and Collocational Unconventionality are defiend as parameters for metaphor judgement. If there are class schemas matched and the parameters satisfy the threshold, then the potential metaphorical units in the input sentence are recognized and tagged. Experiment results support the validity and efficiency of our metaphor recognition strategy.On true meaning acquirement and interpretation, a representation method and metaphor role framework is developed to represent literal structure and meaning of recognized metaphors. A target-driven mechanism combied with target concept attributes, source concept features (attribute values) and mutual-information based mapping function is developed for similarity acquirement and true meaning creation. The interpretation rules are also developed to output the true meaning of metaphors in natural language.Computational system of metaphor recognition and interpretation is the target of the research. The theories and approaches are combied to implement the system which we call CHMeta. CHMeta is finally implemented as the first integrated computational system with large-scale processing ability of Chinese metaphor computation. The work is encouraging and has made a crutial step of Chinese metaphor computation and it also lays a foundation for further development of metaphor computation.