Perturbation-based Regional Interpretability through Subtraction Mapping (PRISM): naming-error dissociations in language models and post-stroke aphasia

2026-08-12

AI Summary

Current large language models lack spatially resolved, falsifiable tools to test whether their internal components specialize in distinct cognitive functions. This work adapts the subtraction analysis framework from neuroimaging to Transformer models, integrating hierarchical perturbations with error categorization from the Philadelphia Naming Test to propose PRISM—a method combining threshold-free clustering and lesion–symptom difference mapping. PRISM establishes, for the first time, a structurally aligned subtraction mapping between model layers and human brain regions. The framework robustly replicates a double dissociation pattern for phonological processing in both deep model layers and the human frontal perisylvian cortex, with convergent trends also observed along the semantic dimension. This provides a unified, falsifiable paradigm for evaluating functional specialization in both artificial models and the human brain.

Abstract

Mechanistic interpretability of large language models lacks spatially resolved, falsifiable tools for testing whether internal components are specialized for distinct cognitive operations. We adapt subtraction analysis, the standard framework of human neuroimaging, from biological brains to perturbed transformers, and apply the same logic to both substrates in parallel. Building on the Brain-LLM Unified Model (BLUM), which showed that layer-perturbed LLaVA-1.6-Vicuna-13B error profiles match the lesion patterns of aphasic patients, we develop PRISM (Perturbation-based Regional Interpretability through Subtraction Mapping). PRISM maps the seven clinical Philadelphia Naming Test categories, subtracts error classes pairwise, and treats each perturbation seed as a subject in a group analysis with threshold-free cluster enhancement along the layer axis. We run a structurally matched analysis on 213 chronic post-stroke aphasia patients using correlation-difference lesion-symptom mapping, and replicate both sides on held-out splits. The designs match in subject dimension (seeds, patients), spatial dimension (layers, atlas-parcellated cortex) and thresholding, but the contrast operator differs: a within-subject error-proportion difference for the LLM, a between-subject correlation difference for the cortex. Both substrates recover a robust phonemic-favoring dissociation, a deep layer cluster and a frontal-perisylvian cortical cluster, both replicating; the semantic-favoring direction is a consistently signed but non-significant trend on both. PRISM thus gives a falsifiable, spatially resolved test of functional-specialization claims in transformer language models. A confirmatory ROI-level intervention (PRISM Stage 3) licensing the strongest causal-mechanism claim is left to subsequent work.

Problem

• mechanistic interpretability
• functional specialization
• large language models
• spatially resolved analysis
• falsifiable testing

Innovation

• PRISM
• mechanistic interpretability
• subtraction mapping
• language models
• aphasia