Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Naturalistic speech tracking reveals developmental reorganization of children’s word-learning system
Poster Session C, Thursday, October 1, 10:45 am - 12:45 pm, Wangari Maathai
Chen Hong1, Rujun Duan1, Hanlin Wu1, Gangyi Feng1; 1The Chinese University of Hong Kong
Children learn language from continuous natural speech, requiring the developing brain to extract multiple speech cues from rapidly changing acoustic input. Neural tracking provides a powerful way to characterize this process, but it remains unclear how multi-level speech tracking changes from childhood to adulthood and whether these neural dynamics explain individual differences in children’s language-learning ability. Here, we used EEG and temporal response function modeling to examine neural tracking of naturalistic Cantonese speech in 82 children aged 4.9-12.4 years (M = 8.9 ± 1.8 years) and 61 adults. Participants listened to six or eight Cantonese stories, each lasting 3.5 ± 0.7 min. We modeled responses to six speech dimensions spanning acoustic, phonetic, pitch, and lexical-tone processing. Children also completed assessments of nonverbal intelligence, working memory, LLAMA word learning, visual memory, and probabilistic learning. Fivefold cross-validated partial least squares regression tested whether temporal patterns of speech tracking predicted these scores. Both children and adults showed significant tracking over bilateral frontal and temporal regions. Children showed stronger responses over bilateral temporal regions, whereas adults showed stronger responses over prefrontal regions, with additional adult-dominant occipital responses for acoustic, consonant, and vowel features. Most features elicited both early responses at 30-150 ms and late responses at 150-500 ms. Children showed longer peak latencies for acoustic, consonant, vowel, and early pitch-change responses. For pitch height, children showed a later late component but no early latency difference. Tone-category responses were earlier in children than in adults in both time windows. Linear classification showed that most feature pairs were distinguishable within each group, except pitch height versus pitch change, and that six-feature spatiotemporal patterns distinguished children from adults with a high accuracy of 94.1%. Critically, multi-feature TRF patterns selectively predicted children’s LLAMA word-learning scores, Pearson’s r = 0.331, p = 0.03, but not other cognitive scores. Predictive windows emerged at approximately 160-450 ms for acoustics, pitch height, pitch change, and tone category, at 300-450 ms for consonants, and at -350 to -100 ms for tone category. These findings show that naturalistic speech tracking captures developmental reorganization of speech representations and selective individual differences in children’s word-learning ability.
Topic Areas: Language Development/Acquisition, Speech Perception