Research On Some Key Issues Of Metaphor Computation

Language understanding is one of the most important characteristics for human beings. As a pervasive phenomenon in natural language, metaphor is not only an essential thinking approach, but also an ingredient in human conceptual system. Many of our ways of thinking and experiences are virtually represented metaphorically. With the development of the cognitive research on metaphor, it is urgent to formulate a computational model for metaphor understanding based on the cognitive mechanism of metaphor, especially for promoting natural language understanding (NLU).This dissertation first reviews the computational models for metaphor, and then addresses three key issues in metaphor computation to be resolved, which are formalization, recognition and classification of metaphor language, knowledge representation, metaphor mapping and the agent's epistemic state involved in metaphor understanding.According to its linguistic definition, this dissertation classifies metaphorical language into two categories, i.e., denotational metaphor and connotational metaphor. Based on the dependency grammar, we describe the sentence as a dependency tree. Various dependency models of metaphorical sentences are brought forward following the analysis and classification of the dependency trees of metaphorical sentences in a large corpus. An efficient tree model matching algorithm is also proposed, which can determine the occurrence of the metaphor in the dependency tree. For denotational metaphors, a measure formula called Denotational Adaptability (DA) is given based on the hypernymy relation: if the DA value between two words is lower than the threshold, we consider the existence of a denotational metaphor. On the other hand, for collocational metaphors, the measure formula called Collocational Adaptability (CA) is given based on case and semantic knowledge base: if the CA value between two words is lower than the threshold, we consider the existence of a collocational metaphor. After the metaphor is recognized, the metaphor categories and metaphor role are tagged according to the metaphor dependency model.As for metaphor understanding, many works have been done in pragmatics and cognitive linguistics, especially the discussion on metaphor understanding process in pragmatics and metaphor mapping representation in cognitive linguistics. In line with the conceptual blending framework, we represent conceptual knowledge as ontology, and denote mental spaces as ontologies. Therefore, metaphor mapping is converted to conceptual mapping between mental spaces. The generic space is regarded as a common ontology extended from the Sememe system of HowNet. Meanwhile, the input spaces are represented as online knowledge structures which are also represented as ontology. The input space ontology is acquired through the metaphor roles in the dependency tree, in which the source role (Met_S) and target role (Met_T) refer to concepts based on the lexical ontology and the thematic roles derived from semantic role parsing of the sentence. The blending runs the mapping algorithm on the input space ontologies, using the triangulation rule and squaring rule, and employs the Integrated Degree to measure the mapping results.Metaphor understanding involves the agent's subjective factors, including his world knowledge of the metaphorical topic, belief and intention, etc. However, these factors are seldom touched upon by most existing research which merely focuses on the linguistic characteristics of metaphor and the representation of the world knowledge, over-emphasizing the role of objective knowledge. Therefore, in this dissertation, the agent's subjective factors are introduced. An epistemic modal operator of epistemic logic is also used in the logical characterization of metaphor understanding. The obtained logic system is applied to the comprehension of Chinese metaphors.The final part of the dissertation implements a metaphor understanding system -MetaphorSuite. This system integrates dependency parsing and its visual revising tool, algorithm of metaphor dependency model matching and metaphor role tagging, metaphor mapping and interpretation, etc. MetaphorSuite system provides a complete platform for further development of metaphor computation, especially for the construction of basic resources, such as metaphor corpus and metaphor knowledge base.