Chapter 11

Derivation of the prosodic structure

This chapter presents the rules implemented for the automatic generation of the internal prosodic structure of an intonational period based on prominence annotations, from the metrical foot to the intonation package. A special feature of the model is that it accounts for performance features such as filled pauses, interruptions and non-prominent disfluent segments, to differentiate two types of constituents, that we called standard and extended units, at each level of the prosodic hierarchy. Each case is illustrated by examples extracted from the Rhapsodie database. The diversity of the output is discussed in the light of the aim of the project: An exhaustive study of the correlations between situational variables and prosodic constructions implies a detailed characterization of the prosodic objects to be observed. Finally, the issue of the representativeness of each type of unit, standard vs. non standard, is discussed in the context of prosodic performance modeling.