Talk Title: When do listeners mis-perceive co-articulatory variation? Experimental evidence for the role of temporal dynamics.
(Joint work with Georgia Zellou, University of California, Davis)
Co-articulation makes vowels in context acoustically different from context-free vowels. While listeners often compensate by ascribing these acoustic effects to their articulatory source, they sometimes fail to do so. In three perceptual experiments, we sought to pinpoint the conditions under which such mis-attribution occurs. We focused on American English high front vowels, which display systematically different co-articulation dynamics from coda consonants: F2-lowering formant transitions occupy a very small, local portion of the duration of tense [i], but an extensive portion of the duration of lax [ɪ] (Pycha, 2016). In Experiment 1, listeners heard natural productions of CVC nonwords, such as [gib] and [gɪb], and estimated speaker height. For tokens with [ɪ], there was a reliable gradient effect: height judgments correlated with the temporal extent of co-articulation. Experiment 2a was similar, except that stimulus tokens varied across a wider range of temporal extent of co-articulation than in Experiment 1.
Again, for tokens with [ɪ], there was a reliable gradient effect. In Experiment 2b, listeners heard the same tokens in an AXB discrimination task, where the flanking vowels were synthetic, steady-state [ɪ] and [ʌ]. Results show that, as temporal extent of co-articulation on [ɪ] increases, so does the likelihood that listeners mis-perceive the vowel as [ʌ]. These results, which we interpret in light of Ohala’s (1993) proposals for a theory of sound change, show that temporal dynamics do indeed play a strong role in modulating perceptual compensation, but also that perceptual compensation failures can operate gradually, rather than abruptly.