Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Neural bases of speech categorization sensitive to articulatory perturbation
Poster Session C, Friday, October 25, 4:30 - 6:00 pm, Great Hall 3 and 4
Olivia Bizimungu1, Lucie Ménard2, Sylvain Baillet1; 1McGill University, 2Université du Québec à Montreal
Despite large acoustic variability between spoken utterances, our brains rapidly transform sound waves into meaningful phonetic categories. This process of categorical perception facilitates our understanding of speech in challenging listening environments, although it is itself sensitive to context contingencies. Numerous studies have demonstrated that perturbing speech articulators, such as the lips, tongue, and jaw, during active listening modifies the auditory-to-perceptual mapping of sounds produced using the affected articulator (Ito et al., 2009, Möttönen & Watkins, 2009). This phenomenon suggests that sensorimotor articulatory representations may play a role in the perceptual processes underlying speech categorization, even in the absence of overt articulation. While the involvement of sensorimotor systems in speech and other types of auditory perception has been documented, the underlying neurophysiological mechanisms remain to be elucidated. We recorded magnetoencephalographic (MEG) brain responses from 17 healthy, young adults (mean age = 22.29 years, SD = 3.70; 13 women) as they classified sounds from an acoustic continuum synthesized between the vowels /u/ (small lip area) and /œ/ (large lip area). Participants performed the task both at baseline and while holding a 2.5cm tube between the lips, intended to expand their lip aperture (mimicking the production of /œ/). We trained MEG signal decoders using a support vector machine (SVM) design to predict the phonetic category that participants assigned to each speech sound on a single-trial basis from the corresponding MEG recordings. The SVM classifier successfully predicted behavioral responses with an accuracy of up to 65%. The SVM decoders were trained exclusively on trials corresponding to stimuli with a clear phonetic identity, specifically the continuum endpoints. When applied to intermediate stimuli along the phonetic continuum, we found that SVM predictions closely mirrored participants’ perceptual reports. The decoding probability of /œ/ increased across the phonetic continuum, correlating significantly with psychometric response curves (r = .61, p < .001). We then investigated whether the decoders would generalize in the presence of the articulatory perturbator. Outcomes of trials performed with the articulatory perturbation were better predicted by models trained on data collected under the same perturbation condition than by models trained on baseline (non-perturbation) data (p = .048), and the reverse was true for trials performed without perturbation (p = .043). This experiment demonstrates that the brain processes underlying categorical speech perception are sensitive to sensorimotor perturbations, even in the absence of overt speech production. We interpreted the decoders’ weights applied to the MEG sensor data (N=272) and found that the brain regions with greater importance in phonetic decoding included the primary auditory cortex and superior temporal sulcus, and to a lesser degree, the inferior frontal gyrus, pre- and post-central gyri, and the temporo-parietal junction. Taken together, our data contribute to clarifying categorical speech perception processes, demonstrating that they emerge rapidly across multiple brain regions involved in language processing. Future analyses will determine whether internal, articulatory representations, primarily involving sensorimotor brain regions, modulate perceptual processes and contribute to altered phonetic perception due to articulatory perturbation.
Topic Areas: Computational Approaches, Multisensory or Sensorimotor Integration