Discourse Processing based on Discourse Management Theory : Analysis and Generation of Noun Phrases

Abstract

In the current researches in natural language processing and formal semantics, several studies have been made for modeling discourse processing. One of the powerful approaches is to build a model based on the relation theory of meaning, in which utterances are interpreted as mappings between situations or beliefs. In this paper, we construct the model of discourse processing based on the Discourse Management Theory (DMT), which is a kind of semantic theory inspired by the idea of the relation theory of meaning. As is discussed in detail in DMT, some natural language expressions, for instance, Japanese noun phrase expressions such as 'Tanaka', 'Tanaka-toiu-hito (a man named Tanaka)', 'ano-hito (that man)', and 'sono-hito (also that man)', are used selectively depending on the contents of the speaker's and hearer's knowledge, that is, the context when those expressions are uttered. A principle for choosing an appropriate expression to a context is formalized by considering constraints which those expressions have. These constraints are processed uniformly according to the same schema of DMT. Context processing, both context analysis and generation, is modelled as the mapping between contexts using constraints ; that is, the schema of DMT. Using each constraint one context is mapped to another, and the whole context processing is constituted by the composition of these mappings. In this model, no specific procedures, such as analysis and generation of anaphoric expressions, are used. All contextual processes, for instance, identifying the antecedent of an anaphora in context analysis and deciding which expression to use for a description in context generation, are done by the mapping using concerned constraints as a part of the course of the whole context processing according to the DMT schema. We will illustrate these processes with some example dialogues.