Bob McMurray,1 Michael K. Tanenhaus,1 Richard N. Aslin1 & Michael J. Spivey2
1 University of Rochester, 2 Cornell University
Classic studies on phoneme recognition for syllable initial voiced and voiceless stop consonants demonstrated remarkably "categorical" perception with (a) step-like identification functions, and (b) nearly perfect discrimination of small acoustic differences when the tokens straddled the category boundary and near chance discrimination when the same magnitude difference was within a category. These results were consistent with a view of speech recognition in which fine-grained acoustic-phonetic differences within phonemic categories have minimal effects on lexical access. However, recent work in articulatory phonetics suggests that position in a prosodic domain has consistent and reliable effects on the production of word-initial consonants. These prosodic differences result in small but reliable variations in word-initial VOT [1, 2]. Thus, a complementary word recognition system should also be sensitive to fine-grained differences in VOT.
We used an eye-tracking paradigm in which subjects heard either a single token from a 9-step synthetic /b/-/p/ VOT continuum (0 to 40 ms) embedded in a word, or an l- or sh- initial filler word and responded by clicking on one of four pictures on each trial (e.g., a picture of a peach, beach, ship or leaf). We analyzed the proportion of fixations to the 'target' (e.g., the picture of a beach when a 0 ms VOT token was presented), the most similar alternative or 'competitor' (e.g., the peach), and the two unrelated alternatives. We examined prototypical VOTs (90 % or more of responses to one category): 0, 5 and 10 ms for /b/; 25, 30, 35 and 40 ms for /p/. We excluded any trial in which the "incorrect" word for that category was selected. As VOT approached the category boundary, the duration (and proportion) of fixations to the competitor showed gradient increases, resulting in a significant effect of within-category VOT (/b/: F(3,48)=6.3, p=.001; /p/: F(4,64)=8.0, p<.0001) and highly reliable linear trends. For example, the competitor word (e.g., /peach/ when /beach/ is the stimulus) became more active and remained active longer as VOT increased from 0 to 5 to 10 ms. Additional experiments demonstrated that limited sensitivity to within-category differences in VOT results from the combined effects of using a meta-linguistic task with non-word syllables and only two alternatives.
These results suggest that sensitivity to fine-grained acoustic differences, as preserved in patterns of lexical activation, could be an important mechanism for taking into account prosodically-conditioned acoustic differences, integrating acoustic/phonetic features across time, and resolving temporal ambiguities.
References
[1] Fougeron, C., & Keating, P. (1997). Articulatory strengthening at edges of prosodic domains. Journal of the Acoustical Society of America, 101, 3728-3740.
[2] Crosswhite, K., Masharov, M., McDonough, J.M., & Tanenhaus, M. K. (2001). Acoustic/phonetic differences in onset embedded words (doll/dolphin) within and across prosodic domains: An instrumental and perceptual study.