Segmental anchoring of pitch movements: autosegmental phonology or speech production?

Arvaniti, Ladd & Mennen (1998) reported a phenomenon of “segmental anchoring”: the beginning and end of a pitch movement are anchored to specific locations in segmental structure, while the slope and duration of the pitch movement vary according to the segmental material with which it is associated. This finding has more recently been replicated and extended in several languages. In a superficially attractive autosegmental interpretation of the phenomenon, the pitch movement is analysed as a sequence of tones which independently undergo “secondary association” to specific points in structure. However, recent empirical findings make such a phonological account less plausible, and suggest that some aspects of segmental anchoring need to be explained in terms of quantitative language-specific phonetic detail. Analogies to diphthong dynamics and to voice onset time, meanwhile, provide an impetus for detailed studies of the production and perception of pitch movements that may advance our understanding of segmental anchoring in general. 1 Segmental anchoring: the basic finding In some intuitively clear way, F0 features such as tone and accent belong with specific elements of the segmental string: Chinese tones go with syllables (or possibly syllable rhymes), English pitch accents go with stressed syllables, Japanese word accents go with a specific mora, etc. This loose belonging together is known in the autosegmental phonological literature as “association”. However, it has been clear for some time that the precise temporal coordination or alignment of F0 events with segmental events is quite complex, and does not follow straightforwardly from the mere fact of association. For example, the peak of a pitch accent may be aligned outside (usually after) the stressed syllable with which it is intuitively associated. Various studies in the early 1990s (notably Silverman & Pierrehumbert, 1990, on English and Prieto, van Santen & Hirschberg, 1995, on Spanish) attempted to identify the interacting factors that contribute to the extent of this phenomenon of “peak delay”. Their principal finding was that an upcoming word boundary and/or an immediately following pitch accent (“stress clash”) can cause the peak to occur earlier; in effect, the peak can be “delayed” when the distance to the next prosodically relevant event is increased. About the same time, Caspers & van Heuven (1993) investigated the effect of “time pressure” on pitch movements in Dutch. They studied the alignment of rising and falling pitch movements with the associated accented syllable as they manipulated speech rate, phonological vowel length, and distance to the following pitch movement. They found a complex set of effects, apparently different for rises and falls, but did not arrive at any general principles governing the alignment of pitch movements with the associated segmental material. However, one of their key findings is relevant for this paper, namely that irrespective of the “time pressure” manipulations, the beginning of an accentual pitch rise in Dutch is aligned with the beginning of the accented syllable. The same was subsequently reported for Spanish by Prieto et al. (1995).