Unmediated data-oriented generation

Abstract

This thesis describes the development of a system of Data-Oriented Generation (DOG) wherein noun-phrases are produced as descriptions of simple visual stimuli. This is work towards a broader goal of developing a psycholinguistically realistic Data-Oriented theory of Sentence Generation. Technologically, this is timely because, after sixteen years of research into Data-Oriented Parsing (DOP; the formalism was first proposed by Scha (1990), first implemented by Bod (1992) and has been further developed , for example, by Bod, (1998, 2003, 2006b), Bod, Bonnema and Scha (1996), Bod and Kaplan (1998) Goodman (2003), Hoogweg (2000), no-one has yet produced a system for Data-Oriented Generation. Rather than use a logic-like formalism to encode meaning, the model of generation proposed operates by directly coupling linguistic exemplars with exemplars in other modalities – vision, in the present case, though, it is hoped that the model could be extended into other meaning-providing modalities.