Poster B10, Thursday, August 16, 3:05 – 4:50 pm, Room 2000AB

Early neural reconfiguration predicts future sound-to-word learning success

Gangyi Feng1, Bharath Chandrasekaran2, Patrick C.M. Wong1;1The Chinese University of Hong Kong, 2The University of Texas at Austin

Learning a foreign language in adulthood is challenging, especially when it comes to mapping novel, non-native sound categories onto meanings. The present study examined the neural dynamics underlying novel sound-to-meaning mapping acquisition across different learning stages in the adult brain and make an attempt to uncover the neural reconfiguration that is tightly related to the word learning success. Previous studies have examined either the neural end-stage of learning (e.g., pre- vs. post-neural differences) or have examined pre-training neural predictors of learning. Less is known about how neural reconfigurations at different stages of learning contribute to the individual learning success. Here, we trained 19 native English speakers (mean age = 25.9 y, SD = 4.6 y; no tonal language experience) to learn pseudo-words that entailed learning lexically meaningful pitch patterns (i.e., four Mandarin tones). Participants underwent nine behavioral training sessions, in which they were required to associate 24 pseudo-words with common object meanings, and three functional Magnetic Resonance Imaging (fMRI) scan sessions (i.e., Pre-, Mid, and Post-training sessions), during which they performed a loudness judgment task with a repetition priming paradigm (two repetition effects: tone repetition [tone change vs. no change] and talker repetition [talker change vs. no change] effects) to assess brain representation of sound. We calculated the univariate brain activations (BA) that related to tone/talker repetition effect and the interregional functional connectivity (FC). We further measured the neural pattern reconfiguration (pattern similarity) of these two metrics between scan sessions (e.g., Pre-vs-Mid reconfiguration). Machine-learning approach (i.e., support vector regression with leave-one-out cross-validation) was employed to build and validate prediction models with those reconfiguration metrics as predictors to predict individual learning outcomes. We found that both BA and FC patterns reconfigurations at the early stage of training (i.e., Pre-vs.-Mid) significantly predicted individual learning outcomes (Ps < 0.001). In contrast, we did not find significant outcome prediction at the later stage of training (i.e., Mid-vs.-Post). This finding was confirmed by using different FC network-construction and a permutation test with 10,000-iteration. Direct comparison between early and late models in predictive power revealed significant differences for both BA and FC reconfiguration (Ps < 0.01). Further prediction analysis on individual connections or voxels revealed that the early stage of neural reconfiguration showed different learning-related reconfiguration patterns compared to that at the later stage of training. Our findings represent a rare example of the neural dynamics of language learning across multi-day training and imaging sessions. Sound-to-meaning mapping at the early stage is critical for learners’ brain connectivity and local activation pattern to adapt and reconfigure according to the content of learning. The pattern of functional neural reconfiguration at the early stage of learning could be a powerful neural marker of individual learning success. Our study provides an avenue for understanding the neural mechanism of learning more generally, which may aid in designing personalized training protocols to optimize learning outcomes for all learners.

Topic Area: Perception: Speech Perception and Audiovisual Integration