Recent work shows that syllable-onset /l/s are longer and darker in syllables with voiced codas compared with voiceless codas. In some circumstances, listeners are sensitive to these coda-dependent acoustic differences in onset /l/s as a cue to coda voicing in word recognition. These findings are at variance with segmental models of word recognition such as TRACE and they provide support for an alternative approach in which partitioning the speech signal into segments is not essential to lexical access. The present experiment further explores the effects of the acoustic characteristics of an onset /l/ on the perception of coda voicing.
The word 'led' was synthesized from a natural token. From this stimulus, five more were made such that the duration of /l/ varied in 20-ms steps over the range 70-170 ms. At each of the six durations, a total of 4 stimuli were made by using different combinations of F2 frequency and f0 during the /l/. F2 was constant during the /l/, at either 1850 Hz or 1740 Hz, jumping to 1860 Hz at vowel onset. F0 began at either 180 Hz or 168 Hz, and fell to 162 Hz at vowel onset. Thus there were in total 24 stimuli, identical except for the properties of the initial /l/. Each stimulus was truncated 80 ms after vowel onset, and 300 ms of white noise appended to it, so that the end of the stimulus seemed to be obliterated by noise. The randomised stimuli were played ten times each to 12 listeners in a two-alternative word-identification task (led / let).
Both duration and f0 had the predicted effect on responses: shorter /l/s and higher f0 produced more let responses. Duration had the greatest influence, with a mean of 73% let responses when /l/ was shortest, falling to 36% when /l/ was longest (F(5,55) = 16.21, p < .0001). F0 alone had a smaller effect: 59% vs. 51% let to stimuli with high vs. low f0 respectively (F(1,11) = 12.12, p = .005). There were no significant interactions. Contrary to expectations derived from production data, F2 frequency did not significantly affect the listeners' responses. The potential effect of F2 on the perceived coda voicing in the absence of durational and f0 differences remains to be determined in ongoing experiments. Overall, however, our results confirm that listeners use acoustic properties of onset /l/ to predict the voicing of the tautosyllabic coda.
Hawkins, S., & Nguyen, N. (to appear). Effects on word recognition of syllable-onset cues to syllable-coda voicing, Papers in Laboratory Phonology VI (Cambridge University Press, Cambridge, UK).
Nguyen, N., & Hawkins, S. (1998). Syllable-onset acoustic properties associated with syllable-coda voicing, Proceedings of the 5th International Conference on Spoken Language Processing, Sydney, 30 Nov. - 4 Dec. 1998.
Nguyen, N., & Hawkins, S. (1999). Implications for word recognition of phonetic dependencies between syllable onsets and codas, Proceedings of the 14th International Congress of Phonetic Sciences, San Francisco, 1-7 August 1999.