Coarticulation, the overlapping articulation of adjacent or nearby sounds, is a major area of investigation within current theories of speech production and perception, and laboratory approaches to phonology. The theoretical importance of coarticulation rests in the challenges that coarticulatory variation presents for phonetic and phonological models, combined with the rich information gained from its detailed investigation.
Supported in part by NSF grant BCS-0118684, we are nearing completion of a series of acoustic, perceptual, and phonological investigations of the interplay between segmental timing and coarticulation. On the acoustics/production side, our aim has been to determine whether, under certain contextual conditions, the duration of the segment that serves as the source of coarticulation is inversely related to the extent of its coarticulatory influences. On the perception side, the question has been whether, again under selected contextual conditions, the spatiotemporal extent of coarticulatory information perceptually trades with source segmental duration, yielding a constant phonetic percept. A phonological motivation for this work is that, over time, coarticulatory effects may become phonologized, sometimes with loss of the coarticulatory source. The working hypothesis is that, for coarticulatory effects whose conditioning environment is lost, the process of phonologization involves a stage of coarticulatory variation in which shortened duration of the source segment is offset by more extensive coarticulatory influences. A further hypothesis is that such interactions are facilitated by perceptual equivalence between coarticulatory information on a target sound and source segmental duration. To date, the targeted coarticulatory patterns have primarily been vowel-nasal (VN) sequences - that is, sequences that historically give rise to distinctive nasal vowels, with loss of the nasal consonant constriction.
The production data provide strong, consistent evidence of an inverse relation between the duration of a nasal consonant and the temporal extent of coarticulatory nasalization on a flanking vowel. For example, our acoustic measures of VNC productions by five English speakers show that nasal consonants are shorter before voiceless than before voiced consonants, and vowels are correspondingly more nasalized when followed by NCvoiceless than by NCvoiced. As another example, in Thai, which has contrastive vowel length, nasal codas are relatively short after long vowels, and research in our lab has shown that these long vowels are produced with relatively extensive nasalization (i.e., vowel nasalization extends through more of the vowel in V:N than in VN: sequences; see Onsuwan's 2005 U-M doctoral dissertation).
Our perceptual findings suggest that, in arriving at phonological representations that encompass the wide range of phonetic variants found in production, listeners formulate equivalence categories in which the two 'sites' of a lowered velum in a VN sequence,
and N, are perceptually equivalent. Specifically, when presented with stimuli differing in the temporal extent of vowel nasalization and the duration of the nasal consonant, listeners are accurate at discriminating pairs whose members differ substantially in total nasalization across the
sequence (i.e., pairs of the type
SNS-
LNL, where S = slight V nasalization or shorter N duration and L = longer V nasalization or longer N duration); listeners are poorer discriminators of pairings in which the total nasalization in the VN sequence is held constant (i.e.,
LNS-
SNL). This pattern of perceptual equivalence is robust, holding for listeners whose languages differ substantially in the temporal extent of vowel nasalization (e.g., English, Ikalanga, and Italian).
We speculate that this production-perception scenario is a factor in the historical change VN >
. Generally speaking, when coarticulation is phonologized in the transmission from a speaker to a listener or possibly, over time, within a listener, a predictable property of the signal becomes a distinctive one. We propose that, for coarticulatory nasalization, phonologization is facilitated by co-variation in production and equivalence in perception between
and N. Although the speaker may intend /VN/, under equivalence listeners attend less to the segmental source of nasalization and abstract away 'nasal' rather than /VN/ - or /
/. Support for this speculation is provided by the ways in which the phonetic patterns mirror the historical situation in many languages of the world: the contexts that give rise to the concomitant processes of nasal coda shortening and heavier vowel nasalization are precisely those contexts in which phonological nasal vowels have been found to be especially likely to develop historically.