About variable thresholds accounting for the perception of prominent syllables in spontaneous speech

Denise Deshaies & Claude Paradis

Fri. 2:00-4:05 A

It has been shown quite convincingly that variation at the phonetic, and more specifically, at the segmental level can be constrained by taking into consideration linguistic, geographic, social, and stylistic or contextual factors. However, in the sociolinguistic paradigm, there has been very few studies carried out on prosodic variables. Several factors, such as the difficulty in defining variables, the complexity of the instrumental analysis, the lack of good phonological models until recently, may explain this state of affairs.

For some times, based on the insights of Mertens (1987), Hirst & Di Cristo (to be published), we have been working on a model for representing and transcribing the intonation patterns of spontaneous discourse in Quebec French. Since all models of intonation rely on the relative position of stressed syllables, we have first tried to identify these as objectively as possible on the one hand by asking large groups of subjects to mark them and, on the other hand, by considering the traditional acoustic parameters (F0, duration, intensity) that are said to be related to the perception of prominent syllables.

In this paper, we shall deal with the variability found in the acoustic data of the acknowledged prominent syllables. More specifically, we shall show that using fixed thresholds, as those used by Mertens (1987) for F0 and Pasdeloup (1990) for duration, in order to interpret acoustic data and to distinguish the stressed from the unstressed syllables, is bound to fail. We shall thus argue that such thresholds have to be replaced, for the two corpora considered, by variable ones tuned for each speaker and for each utterance.

This study is based on the results of perceptual tests that were carried out with naive and non-naive French speakers from Quebec and France for establishing the metric structure of 40 utterances extracted from "sociolinguistic interviews" done in the Quebec Province and in France; it is also based on precise acoustical analysis carried out with CSL on the same utterances used for the perception tests.

Preliminary results indicate that the agreement between the perception of prominence and an above average duration is less frequent in the Quebec French utterances than in the France French ones and that the perception of prominence is not only linked to the increase in fundamental frequency on one syllable but also to the nature of the following reset, in other words to the degree of the lowering slope of the pitch contour following the prominent syllable.

We also found, at least for non laboratory speech, that if no robust correlation has ever been uncovered between the perception of prominence at the syllable level and one or several of the acoustic parameters thought to be relevant, it is due to the fact that the acoustic data cannot be considered in absolute terms. A F0 value or a duration measurement in a specific utterance is to be weighed in relation to the others in the same utterance; for example, the same F0 difference between a peak and a valley of 3 semi-tones might lead to the perception of a prominent syllable in one context, while it won't in another.

References

Hirst, D. & A. Di Cristo. (to be published). A survey of intonation systems, in D. Hirst & A. Di Cristo (eds.), Intonation systems. A survey of twenty languages. Cambridge: Cambridge University Press.

Mertens, P. 1987. L'intonation du français. De la description linguistique à la reconnaissance automatique. Université catholique de Louvain PhD thesis.

Pasdeloup, V. (1990) Modèle de règles rythmiques du français appliqué à la synthèse de la parole. Université de Provence (Aix-en-Provence) PhD thesis.