Decoding and Characterizing the Intracranial Representation of Semantic Information
- 1University of Alabama at Birmingham
Abstract
Brain-computer interfaces (BCIs) have achieved impressive performance by decoding motor and articulatory signals associated with speech production. However, considerably less is known about whether higher-level semantic representations can be decoded from human cortical activity. Demonstrating semantic decoding would advance both our understanding of language organization and the development of BCIs that rely on conceptual rather than purely articulatory information. We recorded intracranial neural activity from patients undergoing stereotactic electroencephalography (sEEG) for clinical epilepsy monitoring while they performed language tasks requiring semantic processing. High-gamma power was extracted from local field potentials and used to generate trial-level features for supervised machine-learning classification. Classification performance was evaluated using cross-validation. Semantic category information was decoded significantly above chance, with mean classification accuracy reaching 29.8% across 15 semantic categories (chance = 6.7%). These findings demonstrate that high-gamma activity contains information about conceptual category membership that can be extracted on individual trials. These results provide evidence that semantic information is accessible from intracranial population recordings and support the feasibility of semantic decoding as a complementary direction for future language BCIs. Beyond neuroprosthetic applications, this work contributes to understanding how conceptual knowledge is represented in the distributed human language network.
Related
- Multilingual Brain Decoding Reveals Shared Cortical Semantic Geometry
- Neural latent representation of implicit perceptual decision confidence signatures from brain-wide intracranial EEG
- Many Brains, One Geometry: A Shared Visual-Semantic Space for Cross-Dataset fMRI Decoding
- Semantic Alignment of EEG and Text: A Contrastive Learning Framework for Decoding Imagined Speech
- Frequency-specific oscillations track parallel processing of linguistic structural and lexical-semantic information from perception to decision
BibTeX
@misc{smith2026decoding,
title = {{Decoding and Characterizing the Intracranial Representation of Semantic Information}},
author = {C. Smith and S. Inchyna and B. Barrentine and M. J. Nelson},
howpublished = {bioRxiv},
year = {2026},
month = jul,
doi = {10.64898/2026.07.13.738249},
url = {https://www.biorxiv.org/content/10.64898/2026.07.13.738249v2},
note = {CC BY-NC},
}
RIS
TY - UNPB
TI - Decoding and Characterizing the Intracranial Representation of Semantic Information
AU - C. Smith
AU - S. Inchyna
AU - B. Barrentine
AU - M. J. Nelson
PY - 2026
DA - 2026/07/13
PB - bioRxiv
DO - 10.64898/2026.07.13.738249
UR - https://www.biorxiv.org/content/10.64898/2026.07.13.738249v2
AB - Brain-computer interfaces (BCIs) have achieved impressive performance by decoding motor and articulatory signals associated with speech production. However, considerably less is known about whether higher-level semantic representations can be decoded from human cortical activity. Demonstrating semantic decoding would advance both our understanding of language organization and the development of BCIs that rely on conceptual rather than purely articulatory information. We recorded intracranial neural activity from patients undergoing stereotactic electroencephalography (sEEG) for clinical epilepsy monitoring while they performed language tasks requiring semantic processing. High-gamma power was extracted from local field potentials and used to generate trial-level features for supervised machine-learning classification. Classification performance was evaluated using cross-validation. Semantic category information was decoded significantly above chance, with mean classification accuracy reaching 29.8% across 15 semantic categories (chance = 6.7%). These findings demonstrate that high-gamma activity contains information about conceptual category membership that can be extracted on individual trials. These results provide evidence that semantic information is accessible from intracranial population recordings and support the feasibility of semantic decoding as a complementary direction for future language BCIs. Beyond neuroprosthetic applications, this work contributes to understanding how conceptual knowledge is represented in the distributed human language network.
ER -