Empirical Analysis of Three Dimensions of Spoken Discourse: Segmentation, Coherence, and Linguistic Devices

A discourse consists not simply of a linear sequence of utterances,1 but of meaningful semantic or pragmatic relations among utterances. Each utterance of a discourse either bears a relation to a preceding utterance or constitutes the onset of a new unit of meaning or action that subsequent utterances may add to. The need to model the relation between such units and linguistic features of utterances is almost universally acknowledged in the literature on discourse. For example, previous work argues for an interdependence between particular cue words and phrases such as anyway, and their location relative to an utterance or text, (e.g., Hirschberg and Litman 1993, Grosz and Sidner 1986, Reichman 1985, Cohen 1984); the distribution and duration of pauses relative to multi-utterance units (e.g., Grosz and Hirschberg 1992, Hirschberg and Grosz 1992, Chafe 1980, Butterworth 1980); and the interdependence between the form of discourse anaphoric noun phrases and the relation of the current utterance to a hierarchical model of utterance actions, or to a model of focus of attention (e.g., Grosz 1977, Grosz and Sidner 1986, Reichman 1985, Sidner 1979, Passonneau 1985). However, there are a variety of distinct proposals regarding how to model the interdependence among the three dimensions of: 1. sequences of semantically and pragmatically related utterances, 2. the units or relations they reflect, and 3. lexico-grammatical or prosodic features. We refer to these dimensions respectively as segmentation, coherence, and linguistic devices.