Poster E8, Saturday, August 18, 3:00 – 4:45 pm, Room 2000AB

Word Learning Influences Phonotactic Repair: A Granger Analysis of MR-constrained MEG/EEG data

Adriana Schoenhaut1, Mulliner Cody3, Ahlfors Seppo1,2, Gow David1,2,3;1Massachusetts General Hospital, 2Athinoula A. Martinos Center for Biomedical Imaging, 3Salem State University

Introduction. All languages show constraints on how speech sounds are combined to form syllables and words (phonotactic constraints). Evidence for the systematic modification of loan words (e.g. English drive -> Japanese [doraibu]) and perceptual repair of banned onset structures (e.g. sr – perceived as [shr-]) demonstrates that constraints actively influence speech perception. These effects have been alternately attributed to mediation by abstract phonological rules/constraints and to top-down lexical influences on speech perception. To distinguish between these accounts, we examine the behavioral and neural dynamic effects of a specifically lexical manipulation (word learning) on perceptual repair of banned onset structures by English speakers. Methods: We conducted two experiments. In both, English-speaking participants learned a set of novel words with visual associates (eg sradex). In a behavioral pilot study, subjects performed a discrimination task with pairs syllables from a five step /sr-V/ to a five-step /shr-V/ continuum before and after sleep-consolidated learning of a set of novel words including a subset that contained forms with illicit (/sr-/) onsets. In a subsequent task, a different set of subjects learned a new set of words, again including a subset of words with illicit forms. After sleep consolidation these subjects performed a 2AFC discrimination task involving the same 5-step srV-shrV continuum while simultaneous MEG and EEG data were collected. These neural data were reconstructed in sourcespace and analyzed using Kalman-filter based form of Granger causality analysis. Results: In the behavioral task, subjects showed superior discrimination performance after word learning. A similar behavioral effect was found in the scanning task. Participants were showed lower rates of perceptual repair and were more likely to accept illicit forms (/sr-). Analysis of activation patterns showed involvement of word form areas (supramarginal gyrus, posterior middle temporal gyrus) but not known rule areas (e.g. LIFG, insula, basal ganglia). Granger analysis showed that word learning influenced patterns of influence by word form areas on brain regions involved in acoustic phonetic processing (posterior superior temporal gyrus). Conclusion: Behavioral evidence for an effect of word learning on repair of banned phonotactic patterns is consistent with the hypothesis that phonotactic constraints reflect top-down lexical influences on speech perception. Moreover, the results of Granger analysis are consistent with previous results showing a central rule of top-down influence by word form areas on acoustic-phonetic regions in the Ganong effect and a variety of phonotactic phenomena in speech perception.

Topic Area: Perception: Speech Perception and Audiovisual Integration