Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Separating Acoustics from Phonology with Rotated Speech
Poster Session B, Friday, September 12, 4:30 - 6:00 pm, Field House
Jacob Burger1, Arild Hestvik1, William Idsardi2, Chao Han3, Azia Knox1, Tarleton Hill1; 1University of Delaware, 2University of Maryland, 3University of Toronto
Underspecification theory claims that specific contrasting phonological features are absent from the lexical representation of phonemes (Lahiri & Reetz, 2002; 2010), predicting an interaction between Mismatch Negativity (MMN) amplitude and phoneme when minimally contrasting pairs are presented to listeners in a varying standards paradigm (Eulitz and Lahiri, 2004). Previous research (Hestvik & Durvasula, 2016; Schluter et al., 2017) observed larger MMN to deviant [d] compared to standard /t/ than deviant [t] compared to standard /d/, which follows if /d/ is underspecified for [-spread (glottis)], because the phonetic deviant [t] with [+spread] is not contradicting the underspecified memory trace for /d/. Alternatively, the asymmetry could be due to the acoustic VOT differences between [d] and [t], not phonemic interpretation. We tested this by measuring MMN to “rotated speech”. Rotation perturbs the formant structure of the vowel, rendering the signal unintelligible as speech while preserving VOT (Blesser, 1972). If the MMN asymmetry is simply driven by short vs long VOT, the same asymmetry should then be observed with rotated speech. Methods: 46 native English speakers aged 18-35 participated. The rotated stimuli were generated from the original speech syllables of [ta] and [da] by multiplying the speech waveform by a sine wave with a carrier frequency of 7 kHz. Participants heard the stimuli in a varying standards paradigm, with one group hearing [d] as deviant and [t] as standard, and the other group hearing [t] as deviant and [d] as standard. Rotated speech was presented in the first block, and normal speech in the second, resulting in deviant type (deviant [d] vs. deviant [t]) as between-subject, and stimulus type (standard vs deviant) x speech type (rotated vs normal speech) as within-subject. Results: Two difference waves for speech and rotated speech were decomposed into 14 temporo-spatial PCA components (Dien, 2010). One component (TF5SF2) was identified as the MMN, peaking at 204ms at FCz. Its factor scores were analyzed as dependent measures in a 2 x 2 repeated measures ANOVA, with group as between-subjects factor and speech type as within-subject factor. No main effect of speech vs rotated speech or group was observed in the difference wave, but the interaction between speech type and deviant group was significant (F(1,44)=33.46, p< 0.00001), stemming from a crossing interaction: for normal speech, the difference score for deviant [d] was significantly below zero, deviant [t] was not. For rotated speech, the opposite pattern was observed, with significantly larger MMN for the rotation of deviant [t] than deviant [d]. Discussion: The speech stimuli replicated previous findings with larger MMN for deviant [d] than deviant [t], while the rotated speech did not exhibit this pattern, leading to the inference that the speech MMN asymmetry reflects phonemic underspecification rather than differences in acoustics. Furthermore, rotated speech resulted in the opposite pattern. We speculate this could be acoustically driven, stemming from the “double-on” vs “single on” neuronal response to short vs. long VOT observed in monkey brains (Steinschneider, 2013), which is presumably also true for humans when purely acoustic VOT differences are processed.
Topic Areas: Phonology,