Knowledge representation and metaphor (Q1188963)

This book considers metaphors and metaphor comprehension as a systematic process in natural language understanding, i.e. the non-deviational, non- exceptional nature of metaphors is stressed. A so-called Dynamic Type Hierarchy (DTH) theory is proposed in which non-literal, figurative aspects as well as literal usage of speech can be explained within a coherent formal framework. In particular, explicit descriptions are given how metaphors can be used to refine and reorganize the ontological level of conceptual knowledge bases. Whether an utterance is interpreted as literal or figurative language then depends on what locations in a semantic type hierarchy concepts are assigned to and the context behind a particular use of languages that triggers the dynamical readjustment of a standard concept hierarchy to a metaphoric framework. The book starts by exploring the nature of metaphor, its traditional conception and elaborates on its major characteristics. Next, the dominant theories of metaphor are considered and each is weighted against recent philosophical studies on the subject. This section also contains a review of the current psycholinguistic evidence on metaphor processing. A brief overview of major issues in knowledge representation for artificial intelligence emphasizes the role of logic, its advantages, disadvantages and apparent difficulties to cope with ordinary language phenomena. After this critical examination of using logic as a knowledge representation formalism, several other representation schemes are examined concentrating primarily on the one used in the remainder of the book, namely Sowa's conceptual graphs. This semantic network-based model combines major constructs underlying various AI representation models, such as type hierarchies, (multiple) inheritance, schemata, prototypes and case relations with a system of formal deduction. Based on this framework a new theory of metaphor is proposed. Type hierarchies and the associated schemata that represent backbround knowledge in a conceptual system are seen as modeling the ontology of the speaker/hearer's internal world. Metaphor is seen as a method generating new perspectives and new concepts in terms of the internal model represented by these hierarchies. Such hierarchies are conceived as dynamic rather than static, and mechanisms for their adaption and rearrangement are introduced. This theoretical account is then compared to and contrasted with other computational approaches to metaphor understanding. A special chapter delves into some philosophical aspects of concepts, their relationship to sets and classes (extension vs. intension, class inclusion, entailment, inheritance, etc.) and looks at the semantics of semantic networks and associated conceptual hierarchies. Furthermore, two rivaling views of language, that of ideal language philosophy (stressing rational and objective features of language) and that of ordinary language philosophy (concentrating on the social and cultural aspects of meaning and language use) are contrasted with respect to their possible contributions to the natural language processing task. Problems and criticism of the ideal language approach as raised by proponents of the ordinary language school are shown to be accomodated by DTH theory. This is illustrated by the explication of particular natural language phenomena, such as open texture, context, family resemblances, and concept formation. Finally, an outline of computational procedures to implement the proposed theory of metaphor comprehension is given, considering the Cgen system and its further extension by dynamic type hierarchy computation mechanisms.