Stroke Lesions as a Rosetta Stone for Language Model Interpretability

2026-02-03

AI Summary

This work proposes the Brain–Language Model Unified Framework (BLUM), which introduces an external validation benchmark for large language models by leveraging lesion–symptom mappings from post-stroke aphasia patients. By systematically perturbing Transformer layers and comparing error patterns between models and humans on picture naming and sentence completion tasks, the study establishes a novel paradigm linking behavioral alignment with computational mechanisms. Using data from 410 chronic stroke patients, the framework integrates lesion prediction, clinical assessment, and error-profile projection to demonstrate that model-predicted lesion locations align with actual lesions at rates of 67% (p < 10⁻²³) and 68.3% (p < 10⁻⁶¹). Furthermore, semantic and phonological errors map onto ventral and dorsal language pathway damage, respectively, offering neurobiologically grounded validation of model components’ functional necessity.

Technology Category

• Natural Language Processing: (Large) Language Models
• Machine Learning: Large Multimodal Models (LMMs)
• Computer Vision: Language and Vision

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Large language models for search
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

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.

Problem

• language model interpretability
• external validation
• lesion-symptom mapping
• large language models
• causal brain-behavior relationships

Innovation

• lesion-symptom mapping
• large language models
• model interpretability
• clinical neuroscience
• brain-LLM alignment