Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Feedback-Related Negativity Predicts Flexible Generalization in Non-Native Speech Category Learning
Poster Session E, Sunday, September 14, 11:00 am - 12:30 pm, Field House
Liron Shlesinger1, Teagan E. Esther1, Fernando Llanos1; 1University of Texas at Austin
Acquiring non-native speech categories can be particularly challenging for adults. To succeed, they must identify the acoustic features that are linguistically more relevant. This can be particularly difficult when non-native speech is produced by multiple speakers across varying word contexts., as these additional sources of variability can hinder the perception of linguistically relevant features. Previous fMRI work [1] has shown that trial-by-trial feedback can significantly improve the acquisition of non-native speech categories, rewarding correct behavior through engagement of cortico-striatal pathways. In the present study, we used the Feedback-Related Negativity (FRN)—an EEG marker of reward prediction error in response to feedback—to model the neural mismatch between expected and actual outcomes and its impact on non-native speech category learning. Greater FRN amplitudes are thought to reflect larger discrepancies between expected and actual errors, offering a neural index of expectation-outcome dynamics during learning. Native English speakers (N = 36) were trained to categorize 36 exemplars of three French nasal vowels that are not contrastive in English. Stimuli were produced by two native speakers across six word contexts, and presented in random order across five training blocks, with each token repeated once per block. On each trial, participants received immediate visual feedback indicating whether their categorization was correct or incorrect. Following the training phase, participants completed a brief generalization block in which they categorized 36 novel tokens produced by unfamiliar speakers, without receiving feedback. Individual accuracy during the training and generalization phases was calculated as the percentage of correct categorization responses. FRN magnitude was measured following established procedures [2], by computing the difference wave between ERP responses to incorrect and correct feedback. The mean amplitude of the difference wave was then averaged within a 325-425 ms time window across midline electrodes (Fz, Cz, and Pz). Participants successfully learned the vowel categories above chance level (mean training accuracy = 52.9% correct; logistic mixed-effects model: z = 3.72, p < 0.001), indicating that the training paradigm was effective. During the training phase, we found no significant difference in FRN magnitude between learners performing above and below the 50th percentile in categorization accuracy. However, in the generalization phase, learners in the top half conveyed significantly less negative FRN magnitudes in response to unexpected errors compared to those in the bottom half. This indicates that learners with more realistic expectations, operationalized as smaller neural discrepancies between expected and actual outcomes, are more likely to form flexible and generalizable speech representations. In contrast, consistent with the Dunning-Kruger effect [3], learners who overestimate their performance develop poorer generalization skills. References: [1] Yi, H. G., … & Chandrasekaran, B. (2016). The role of corticostriatal systems in speech category learning. Cerebral Cortex, 26(4), 1409-1420. [2] Opitz, B., Ferdinand, N. K., & Mecklinger, A. (2011). Timing matters: The impact of immediate and delayed feedback on artificial language learning. Frontiers in Human Neuroscience, volume 5-2011. [3] Dunning, D. (2011). The Dunning–Kruger effect: On being ignorant of one's own ignorance. In Advances in experimental social psychology (Vol. 44, pp. 247-296). Academic Press.
Topic Areas: Speech Perception,