Recognition Of Referring Expressions

Computational models of referring such as those of Kronfeld (1990) and Heeman & Hirst (1995), being based on the view of language as goal-directed behaviour, assume that the act of referring includes making the hearer recognize the speaker's communicative goal. My present work centres on the question of how we recognize an expression as a referring expression, i.e. how do we know that the use of a noun phrase is intended to indicate a particular object? Working in a Kronfeldian framework, I assume that referring consists of a "literal goal" (making the hearer recognize the np as a referring expression) and a "dis-course purpose" (making the hearer recognize and apply the right "identification constraints", such as the requirement that the referent should be identified perceptually , or should be identified with an entity introduced at a previous stage in the discourse). I propose that in some cases the recognition of the np as a referring expression depends on the recognition of the identification constraints. I work primarily with referring in the literary mode, looking for computation-ally manageable triggers such as lexical anchorings and information derivable from knowledge of genre, but I am also concerned with the questions of whether a unified approach to referring is possible and the extent to which text-based models can be generalized to a multimodal context.