You are viewing the SNL 2017 Archive Website. For the latest information, see the Current Website.

 
Poster E24, Friday, November 10, 10:00 – 11:15 am, Harborview and Loch Raven Ballrooms

Decoding the cortical sensitivity of spoken acoustic variability in persons with aphasia

Caroline Niziolek1, Sara Beach2, Swathi Kiran1;1Boston University, 2Harvard Medical School

Although aphasia is primarily considered a disorder of language, higher-level linguistic deficits are often accompanied by lower-level deficits in auditory perception, feedback processing, and the stable production of consonants and vowels. This project investigates the functional source of low-level speech production deficits in aphasia by assessing the extent to which persons with aphasia (PWA) are sensitive to deviations in their own speech feedback. We consider two measures of feedback sensitivity: detection, the cortical sensitivity to acoustic deviations in one’s own speech, and correction, the extent to which these deviations are behaviorally corrected online. Ten PWA and ten age-matched controls took part in a magnetoencephalography (MEG) study. PWA were 51-59 years of age (mean=54.9; SD=3.0), had chronic aphasia (months post-stroke: mean=83.9; SD=45.7), and were assessed using the Western Aphasia Battery (WAB-R) and Psycholinguistic Assessments of Language Processing in Aphasia (PALPA). Participants produced 200 repetitions of three monosyllabic words (“eat”, “Ed”, “add”) while neural activity was recorded from 306 channels (Elekta Neuromag Triux). This “speak” condition was interleaved with a “listen” condition in which recorded audio from the speak trials was played back to participants through earphones. To assess detection ability, we compared the auditory M100 response between the listen and speak conditions, using the difference between the two, or speaking-induced suppression, as an index of how well the feedback matched the intended sound. Feedback sensitivity was defined as a reduction in suppression for deviations from typical vowel acoustics, and was measured separately for left and right auditory cortex. To further probe feedback sensitivity, we trained a machine learning classifier (Meyers, 2013) to distinguish both coarse (word identity) and fine-grained (acoustic prototypicality) phonemic information in the whole-brain MEG signal and tested it on held-out portions of the data. To assess correction ability, formants for each vowel were calculated in two time windows, the onset and midpoint of the syllable. Vowel centering was defined as formant movement toward the median over the time course of the syllable, lessening acoustic deviation. Evidence for detection of self-produced acoustic deviations was found in both controls and PWA. In controls, speaking-induced suppression was consistently modulated by acoustic deviation only in the left hemisphere; that is, only left auditory activity was relatively greater during the production of more deviant utterances. In contrast, PWA showed only weak activation of left auditory regions, but speaking-induced suppression in the right hemisphere was modulated by acoustic deviation. Further, both word identity and acoustic prototypicality can be read out from the neural data of PWA and controls passively listening to their own speech. Behavioral correction was also largely intact: PWA had greater acoustic variability than controls at vowel onset, but both groups exhibited vowel centering, significantly decreasing variability over the course of the syllable. The hemispheric shift of modulatory responses in PWA is suggestive of plasticity in the neural mechanisms that underlie this sensitivity, and may also enable the neural decoding accuracy and intact behavioral correction seen in PWA. These analyses inform theories of error detection and correction in aphasia.

Topic Area: Speech Motor Control and Sensorimotor Integration

Back to Poster Schedule