Speaker
Description
Psychiatric disorders including ADHD, depression, and schizophrenia are associated with characteristic alterations in semantic language processing. Recent work demonstrates that fMRI activation patterns in healthy individuals during naturalistic language processing share similarity with large language model (LLM) internal representations. Whether this alignment differs in clinical populations remains unexplored.
PSYCH-GPT is a prospective, observational, parallel-group fMRI study (N = 120; n = 30 per group: ADHD, depression, schizophrenia, healthy controls). Participants listen passively to a naturalistic audiobook passage during 3T fMRI acquisition. BOLD signal time series are compared to word-level contextual embeddings from locally hosted LLMs via Representational Similarity Analysis (RSA). Secondary analyses examine correlations with symptom severity and compare LLM architectures.
We hypothesise that brain-LLM alignment is reduced in clinical groups, with disorder-specific divergence patterns reflecting known alterations in semantic coherence.