Automatic Modelling of Intonation

In a recent paper (Hirst, Di Cristo & Espesser "Levels of representation and levels of analysis for the description of intonation systems" in Merle Horne (ed.) 2000 Intonation : Theory and Experiment. Kluwer Academic Press, Dordrecht). we argue that a satisfactory model of an intonation system will need to account for four distinct levels ordered from the most concrete to the most abstract.

Physical level
Phonetic level
Surface phonological level
Underlying phonological level

Representations at each level must be interpretable in terms of representations at adjacent levels. A "model" of an intonation system is conceived of as an attempt to provide mechanisms which allow us to derive representations at each level from representations at other levels.

Physical level

The level of acoustic data and/or physiological measurement.

Phonetic level

In the case of fundamental frequency we assume that at the phonetic level the fundamental frequency curve has been factored into two components :

a macroprosodic component: This consists of a smooth continuous curve reflecting the intonation contour with which the utterance is spoken. The simplest mathematical function respecting the constraints of continuity and smoothness is a quadratic spline function.
a microprosodic component: This consists of deviations from the smooth continuous macroprosodic curve reflecting the influence of the segmental content of the utterance.

An algorithm MOMEL has been developped (Hirst & Espesser 1993) which allows the automatic modelling of a raw fundamental frequency curve as a sequence of target points defining a quadratic spline function.

Surface phonological level

Following Trubetkoy, we take the phonology/phonetics boundary to be the borderline between continuously variable and discrete phnomena. We further assume a distinction between surface phonological form - which is interfaced with phonetics and underlying phonological form - the level which is interpreted as meaningful.

For intonation patterns we have proposed (Hirst & Di Cristo in press) an INternational Transcription System for INTonation (INTSINT) as a first approximation to a surface phonological representation of intonation patterns.

Underlying phonological level

Other research projects on prosody.