Recovering Lesion Parameters from Aphasic Picture Naming Error Profiles in Large Language Models

2026-08-05

AI Summary

This study addresses the inverse problem of inferring lesion parameters in large language models from aphasic picture-naming error patterns to elucidate internal computational mechanisms of Transformers. By systematically inducing targeted lesions—varying in layer, intensity, and noise—within LLaVA-Vicuna 13B, the authors generate clinically aligned naming error profiles. A multitask neural network is then trained to invert this mapping, recovering lesion parameters from observed behavioral errors. This work presents the first successful inversion from behavioral error profiles to precise lesion parameters in a large model, supported by counterfactual validation and cross-distribution generalization tests within an interpretable framework. Experimental results demonstrate accurate recovery of lesion intensity and noise, localizable layer indices within neighboring ranges, 81.4% counterfactual behavioral replication accuracy, and strong syndrome discrimination performance on a clinical dataset of 278 stroke patients.

Abstract

Interpretability methods for large language models (LLMs) describe internal state but do not directly test whether that state is causally sufficient to produce the observed behavior. In earlier work, we lesioned LLMs to produce error profiles in picture naming, a central task for assessing aphasia, and found that specific lesions produced errors resembling those of individual stroke survivors. Here we ask the inverse question: given an error profile, can the lesion parameters that produced it be recovered, and what does this inverse problem reveal about transformer computation? Lesions in LLaVA-Vicuna 13B were parameterized by layer index, modification percentage, and noise sigma across 4,840 configurations, and error profiles were characterized by a seven-category clinical taxonomy (correct, semantic, unrelated, formal, mixed, neologism, no-response). We trained a multi-task neural network to map error profiles back to perturbation parameters. The problem admitted a partial solution: across 10 independently trained inverse models, modification percentage and noise sigma were recoverable, whereas layer index was recoverable only within a neighborhood. In counterfactual validation, a fresh model instance perturbed with the recovered parameters reproduced the target behavior in 81.4% of cases. This dissociation between low layer recovery and high counterfactual fidelity is consistent with functional redundancy across transformer layers, a property not captured by standard interpretability methods. As an out-of-distribution test, we applied the trained model to picture-naming error profiles from 278 stroke survivors; recovered parameters were syndrome-discriminative, most strongly for perturbation intensity, indicating generalization beyond the training distribution. Counterfactual validation provides a general framework for LLM interpretability claims beyond inverse mapping.

Problem

• lesion parameter recovery
• aphasic error profiles
• picture naming
• large language models
• inverse problem

Innovation

• inverse modeling
• lesion recovery
• counterfactual validation
• functional redundancy
• aphasia error profiles