Modeling the Articulatory Dynamics of Two Levels of Stress Contrast

Abstract

One of the most intractable problems in the description of utterance prosody is that of understanding speech timing control. Gösta Bruce’s seminal monograph on Swedish word accents in sentence context (Bruce 1977) highlighted an aspect of this problem that is crucially important for the perception of intonational contrasts in many languages — that of modeling what speakers do when they utter a given tune for a sentence and coordinate it with the string of words that constitute its text. By decomposing the tune into independent contributions from word accent, focal phrase accent, and final boundary events, Bruce was able to show that a robust correlate of the contrast between Accent I and Accent II involves a difference in timing, relative to rhythmically critical events in the text, of tonal events that are common to both word accent types. While much work remained to be done (cf. Bruce 1990 and the literature reviewed there), Bruce’s careful attention to prosodic structure, both at the level of lexical contrasts and at the level of focal prominence contrasts for the phrase or utterance as a whole, allowed him to build a phonological model of Swedish intonation contours that could generate precise, testable quantitative predictions. This work was a foundation stone for Pierrehumbert’s later phonological model for English intonation (Pierrehumbert 1980; Liberman and Pierrehumbert 1984; Beckman and Pierrehumbert 1986), a model which produced a synthesis system capable of generating by rule the full range of grammatical intonations. for English (Anderson et al. 1984). Summarizing such developments in intonation synthesis over the last fifteen years, we can say that Bruce’s work on the Swedish model has inspired, directly or indirectly, implementable phonological models of intonation systems for many languages (see reviews in Ladd 1992, 1996; Pierrehumbert, This volume). The English model, in particular, underscores the importance of accurate tonal timing, since there are pairs of accent types that are like the Swedish word accents in contrasting just in tune-text alignment (Pierrehumbert and Steele 1989), and at the same time, there are differences in timing for the same accent type when it is coordinated with rhythmic events at different structural positions in the text (Silverman and Pierrehumbert 1990).