Stroke Lesions as a Rosetta Stone for Language Model Interpretability

TL;DR

This paper introduces BLUM, a framework that uses lesion-symptom mapping from stroke patients to externally validate and interpret the internal components of large language models, bridging neuroscience and AI.

Contribution

The study presents a novel approach that leverages human brain lesion data to evaluate and interpret LLM perturbations, providing external validation for model interpretability.

Findings

LLM error profiles align with human lesion patterns in 67-68% of cases.

Semantic errors map onto ventral-stream lesions; phonemic errors onto dorsal-stream.

External validation using stroke data enhances understanding of LLM internal mechanisms.

Abstract

Large language models (LLMs) have achieved remarkable capabilities, yet methods to verify which model components are truly necessary for language function remain limited. Current interpretability approaches rely on internal metrics and lack external validation. Here we present the Brain-LLM Unified Model (BLUM), a framework that leverages lesion-symptom mapping, the gold standard for establishing causal brain-behavior relationships for over a century, as an external reference structure for evaluating LLM perturbation effects. Using data from individuals with chronic post-stroke aphasia (N = 410), we trained symptom-to-lesion models that predict brain damage location from behavioral error profiles, applied systematic perturbations to transformer layers, administered identical clinical assessments to perturbed LLMs and human patients, and projected LLM error profiles into human lesion space. LLM error profiles were sufficiently similar to human error profiles that predicted lesions corresponded to actual lesions in error-matched humans above chance in 67% of picture naming conditions (p < 10^{-23}) and 68.3% of sentence completion conditions (p < 10^{-61}), with semantic-dominant errors mapping onto ventral-stream lesion patterns and phonemic-dominant errors onto dorsal-stream patterns. These findings open a new methodological avenue for LLM interpretability in which clinical neuroscience provides external validation, establishing human lesion-symptom mapping as a reference framework for evaluating artificial language systems and motivating direct investigation of whether behavioral alignment reflects shared computational principles.