arXiv cs.CL· Muxuan He, Wang Bill Zhu, Yue Chen, Elsi Kaiser, Weizhi Du, Yao Du, Samuel Nastase, Yanjun Gao·· 13 小时前AI 评分31
把失语症当作透镜：研究 LLM 语言功能的结构与纠缠
Language Impairment as a Lens onto Large Language Models
AI 导读
研究者提出受失语症启发的机制可解释性框架，用分类器引导的损伤与连续激活引导两种方法在 LLM 中系统性诱发类失语症故障。结果显示语言功能可被定位到离散结构回路和连续表征向量，但不同失语症画像间存在显著回路重叠与向量相似性，功能高度纠缠；激活引导可稳健泛化到下游任务，结构损伤则受非对称约束。
正文 · 原文
Abstract:Although large language models (LLMs) encode language in highly distributed spaces, they remain vulnerable to focal disruptions---a phenomenon strikingly similar to how localized brain damage causes clinical aphasia in humans. To investigate this theoretical tension, we ask whether LLMs possess a functional structure analogous to the human language system, by introducing a novel mechanistic interpretability framework inspired by aphasia. Through two complementary approaches, classifier-guided lesioning and continuous activation steering, we systematically induce aphasia-like breakdowns in LLMs. We find that language functions can be successfully isolated to discrete structural circuits and continuous representational vectors, yet these functions remain densely entangled with significant circuit overlap and vector similarities across distinct aphasic profiles. Furthermore, while steering generalizes robustly across downstream tasks, structural lesioning remains asymmetrically constrained. By bridging neuropsychology and mechanistic interpretability, we demonstrate that LLMs, functionally analogous to biological brains, rely on a densely entangled architecture that can be systematically mapped through clinically-informed interventions.
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2610.28992 [cs.CL] |
| (or arXiv:2610.28992v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.28992 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Wang Bill Zhu [view email]
[v1]
Tue, 20 Oct 2026 23:42:36 UTC (1,917 KB)
来源：arXiv cs.CL · arxiv.org