Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Hierarchical speech encoding in the developing brain: spatio-temporal insights from intracranial EEG
Poster Session C, Thursday, October 1, 10:45 am - 12:45 pm, Wangari Maathai
Jill Kries1, Brian Ervin2, Jason Buroker2, Gregory Hickok3, Ravindra Arya2, Laura Gwilliams1; 1Stanford University, Stanford, CA, USA, 2Cincinnati Children's Hospital Medical Center, Cincinnati, OH, US, 3University of California, Irvine, CA, USA
Introduction: Typically developing (TD) children acquire language at a remarkable pace, progressing from babbling to full sentences within two years. Yet acquisition continues well into young adulthood, relying on the coordinated development of a distributed brain network linking speech sounds to meaning. How this network matures remains poorly understood. In children with epilepsy, this developmental trajectory can be disrupted; compared to TD peers, they show reliably lower verbal comprehension, and prolonged seizure activity is associated with progressive language decline. Yet which specific subprocesses are affected remains unclear, in part because we lack a full characterization of how the developing brain encodes speech across acoustic, phonetic, and lexical levels. We expect acoustic and phonetic encoding to be established early, with lexical encoding maturing later in development, and hypothesize that verbal comprehension selectively modulates lexical encoding while seizure duration more broadly disrupts encoding across the feature hierarchy. Methods: We ask: (1) What are the spatio-temporal neural correlates of speech encoding in the developing brain across acoustic, phonetic, and lexical levels? (2) How do age, verbal comprehension, and years lived with epilepsy modulate the spatio-temporal dynamics of neural speech encoding? We recorded intracranial EEG from 101 children with epilepsy (ages 3-23; 13k electrodes) as they listened to 3 minutes of natural speech. We annotated the stimuli with features spanning three hierarchical levels, characterizing acoustic (peak rate), phonetic (voicing, manner and place of articulation, roundness), and lexical (phoneme surprisal and entropy) properties, alongside baseline onset features. Neural responses were modeled using the multivariate temporal response function (mTRF) framework, which predicts electrode-level activity from continuous speech features. Results: Speech encoding was robust across all participants and localized to a bilateral temporal-frontal network, with electrode placement accounting for 12% of variance in encoding performance. Phonetic and lexical features showed stronger unique encoding than the acoustic feature across this network, whereas the acoustic feature uniquely outperformed lexical features only in right STG. Phonetic features yielded 95% fewer significant electrodes than acoustic or lexical features, suggesting largely shared encoding with other features. TRF peak latencies revealed a temporal hierarchy, with lexical features peaking later (205 ms) than acoustic (187 ms) or phonetic (185 ms) features. Finally, speech encoding strength scaled positively with age for acoustic and lexical but not phonetic features; verbal comprehension and years lived with epilepsy did not reach significance. Conclusion: Together, these findings reveal a bilateral temporal-frontal network underlying hierarchical speech encoding in the developing brain, with acoustic, phonetic, and lexical features engaging partially dissociable regions at distinct latencies. Encoding strength increased with age, paralleling the well-documented growth in behavioral language proficiency across development. The absence of epilepsy-related effects may reflect limited statistical power or insensitivity of the current feature set to the most affected subprocesses. Future research should examine semantic and syntactic features to determine whether these findings extend to higher-order language processing.
Topic Areas: Language Development/Acquisition, Disorders: Developmental