Boundaries and groupings - the structuring of speech in different communicative situations: a description of the GROG project

The goal of the project is to model the prosodic structuring of speech in terms of boundaries and groupings. The modeling will include different communicative situations and be based on existing as well as new speech corpora. Production and perception studies will be used in parallel with automatic methods developed for analysis, modeling and prediction of prosody. The model will be perceptually evaluated using synthetic speech. Introduction The objective of the GROG project is to model the structuring of speech in terms of prosodic boundaries and groupings. This structuring is fundamental for spoken communication in that it reflects the speaker’s internal organization of the information, and facilitates the listener’s processing of the message (e.g. Gee & Grosjean, 1983). The later aspect is important for various speech technological applications. Moreover, the structuring of speech is known to vary in different communicative situations. Terken (2001) suggests that communicative situation can be characterized in terms of at least three dimensions: the content, the speaker, and the communicative context. Several kinds of information need to be taken into account to model the structuring of speech. First, we need a classification of prosodic boundaries. Although most researchers agree that several boundary strengths must be assumed, there is no general agreement on issues such as the number and types of boundaries that need to be distinguished. Moreover, the knowledge of how boundaries and groupings are signaled and perceived, and the relationship between those is far from complete. This is reflected in the multitude of prosodic transcription systems available; ToBI 1 GROG: Gräns och gruppering – Strukturering av talet i olika kommunikativa situationer, supported by The Swedish Research Council (VR) 2002–2004. (Beckman & Ayers Elam, 1997) is perhaps the best known, but there are several alternative systems. Second, we need an acoustic modeling of prosodic boundaries tuned to different communicative situations. It is widely known that prosodic boundaries and groupings are signaled mainly by features such as silent or filled pauses, breathing, F0 movements, preboundary lengthening, speaking rate, intensity variation, voice quality and reduction phenomena. We also know that there is a complex interplay between e.g. tonal and temporal features in this signaling (Bruce et al., 1993a). Third, several researchers have shown that morphosyntactic structure and prosody are related to each other, although this relationship is not clearly understood (e.g. Strangert et al., 1993; Ostendorf & Veilleux, 1994). The same holds for the relationship between discourse structure and prosody (e.g. Swerts & Geluykens, 1994). An area of specific interest (see Production and perception studies below) concerns the relations between discourse structure and prosody, where an interplay can be expected between prominence (focus) and boundary phenomena. Thus, we need to explore what kinds of linguistic features and what detail of linguistic analysis that are needed for making correct predictions about prosody. We also need to study the relation between morpho-syntax, discourse and prosody. Speech, Music and Hearing 2 Previous research on Swedish The areas outlined above have all been addressed in previous research on Swedish. Bruce (1995) includes a classification system for prosodic boundaries based on auditive analysis. Studies concerning prosodic characteristics in the vicinity of boundaries include Fant & Kruckenberg (1989), Bruce et al. (1993a). An intonation model suggested by Fant & Kruckenberg (2002) takes duration and pausing into account, as well as clause structure and part-of-speech (PoS). The relationship between syntactic structure and prosody has been investigated by Strangert (1990) and Strangert et al. (1993), Gustafson-Capkova & Megyesi (2002) and the relationship between prosodic structure and prosody by Horne et al. (1995). Horne et al. (2001) concerns prosodic correlates related to topic structure of spoken discourse. Finally, speech style variation related to boundaries and grouping has been investigated by Strangert (1993) and Bruce (1995).