Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Effects of acoustic interruptions on gradient speech representations and predictive language processing
Poster Session E, Friday, October 2, 11:00 am - 1:00 pm, Wangari Maathai
This poster is part of the Sandbox Series.
Chiara Repetti-Ludlow3, Jenah Black1, Joseph Toscano2, Barbara Shinn-Cunningham1; 1Carnegie Mellon University, 2Villanova University, 3Georgia Institute of Technology
Speech unfolds rapidly, and listeners rely on gradient acoustic cues (such as voice onset time [VOT]) to predict upcoming sounds. Under ideal listening conditions, N1 amplitude tracks VOT continuously, with shorter VOTs eliciting more negative N1s (Toscano et al., 2010). Further, clear lexical context can shift the N1 of an ambiguous onset toward a predicted phoneme (Getz & Toscano, 2010). Yet speech rarely unfolds under these conditions: predictive cues are often gradient and ambiguous, and the signal itself may be disrupted. Indeed, recent findings suggest that acoustic interruptions alter listeners' use of fine-grained detail, impacting syllable recall (Liang et al., 2022) and resulting in more categorical behavioral responses to voicing perception (Black et al., in prep). In the present study, we ask (1) do predictive gradient acoustic cues drive continuous differences in the encoding of subsequent ambiguous sounds; and (2) does an acoustic interruption disrupt the use of those gradient representations? Following a forced-choice task to establish each participant’s perceptual boundaries, we carried out an EEG task that used two compound words (beachball and peachpit). We varied the onset /b~p/ in 5 VOT steps and the word-medial /b-p/ in 3 VOT steps, resulting in some clear stimuli (e.g. beachball), some disambiguated stimuli (e.g. ?eachball), and some completely ambiguous stimuli (e.g. ?each?all). Half of the trials carried a brief acoustic interruption (similar to a door slam) between the first and second syllable. Precise stimulus timing was jittered to allow the interruption-evoked response to be deconvolved. Each participant completed 900 trials across 10 blocks of 90. The task was primarily passive; however, a lexical-decision probe occurred after six random trials per block to ensure participant attention. The experiment was conducted using a 32-channel EEG system sampled at 2048 Hz in a sound- and electronically-shielded booth. Recordings were referenced to the average of the mastoid electrodes. We predict that auditory N1 at the first syllable will scale linearly with the onset VOT, replicating prior work. Critically, the N1 at the second syllable should index how onset gradiency cascades forward: an unambiguous onset (e.g., beach) should bias the N1 toward the real-word completion (ball), with bias scaling with onset category membership. An ambiguous onset should yield no such predictive bias; instead, the N1 at the second syllable will vary linearly with word-medial VOT. Pilot data (N=3) demonstrates trends in line with these predictions. With a larger sample size, we further predict that an interruption after the first-syllable onset will disrupt its cascading gradient representation, driving a weakly-predictive context to behave similarly to the strongly-predictive context and subsequently biasing the N1 at the ambiguous second-syllable toward real-word expectations in a categorical manner. This study will thus shed light on the ongoing effect of interruptions on speech processing and the strategies listeners use to accommodate imperfect speech conditions.
Topic Areas: Speech Perception,