Presentation

Chao Han¹, Arild Hestvik¹, Ryan Rhodes²; ¹University of Delaware, ²Rutgers University

The varying-standards paradigm (Phillips et al., 2000) varies standards within a phoneme category and is believed to enforce a phoneme representation as a memory trace for deviance detection. Studies adopting the paradigm have assumed that varying the standards precludes a phonetic MMN (Eulitz & Lahiri, 2004; Hestvik & Durvasula, 2016). Specifically, if the phoneme-based memory trace is purely abstract, there should be no within-category MMN. However, if the memory trace encodes phonetic information, a within-category MMN is expected. In two previous experiments (Han et al, submitted), we did observe an MMN to a within-category contrast, suggesting that the memory trace activated by the varying standards must encode phonetic information. This does not necessarily invalidate Phillips et al, as they could be right that varying the standards recruits a phoneme, but that representation may contain fine-grained acoustic information. However, an alternative interpretation is that varying the standards does not recruit a phoneme at all, but rather results in a memory trace that is a statistical summary of the presented stimuli, as demonstrated by Garrido et al (2013, 2016) with pure tones. To distinguish between the two interpretations, we replicate Garrido’s with VOT-varying speech sounds. Participants will hear 840 standard [tæ]s of VOT drawn from a normal distribution with a mean of 128ms and a standard deviation of 10ms. The deviants are 105 tokens of [tæ] with 64ms VOT and were compared to 105 tokens of [tæ]s with 128ms VOT. The deviant VOT is an outlier to the stimuli presented in the experiment but is approximately equivalent to the mean VOT of /t/ (~60ms, estimated by the empirical VOT in Chodroff & Wilson, 2018) which could be stored with the phoneme. If this deviant VOT results in an outlier MMN effect when compared to standards with a mean of 128ms, then it can only be a result of the memory trace being a statistical summary of the standards. On the other hand, if the standards invoke a latent phoneme with a mean VOT of ~60ms, then the deviant would be identical to the memory trace and no MMN is expected. Combined with our previous experiments, this would provide decisive evidence that varying standards evoke a memory trace containing prototypical phonetic information. To interpret the possible absence of MMN, we are adding a non-speech control condition, where participants will hear a spectral-rotated version of the same stimuli as presented in the speech condition. The spectral-rotated stimuli preserve the acoustic difference between the standards and deviants as the standard-deviant VOT difference in the speech condition but remove the linguistic information. The presence of MMN in the control condition can thus inform us that the possible absence of MMN in the speech condition is not due to the lack of VOT difference between standards and deviants. Assuming the same effect size (d = -0.83) as one of our previous experiments, a power analysis suggests 21 subjects to get 90% power for a standard-deviant a standard-deviant comparison in a two-tailed t-test (\alpha = .05)

Topic Areas: Speech Perception, Phonology and Phonological Working Memory