研究揭示大语言模型中隐喻表征的几何结构并可调控输出隐喻性

The Geometry of Metaphor: Probing and Steering Metaphor Representations in Large Language Models

AI 导读

研究者用机制可解释性技术在覆盖不同架构与规模的大语言模型中识别出因果子空间，可在不同词性的分类任务上操控输出。该技术同样适用于生成任务，表明可用于控制 LLM 输出的隐喻程度。成果将发表于 BlackboxNLP 2026 研讨会。

正文 · 原文

Abstract:Metaphor is a crucial aspect of human cognition. As such, we would like to understand how it is represented in large language models (LLMs) and how it might be controlled. We use mechanistic interpretability techniques to identify causal subspaces in a range of LLMs covering different architectures and sizes. We show that we are able to steer output on classification tasks across different parts of speech. We are also able to steer output on generation tasks, indicating that these techniques can be used to control the metaphoricity of LLM output.

Comments:	BlackboxNLP 2026: The 9th Workshop on Analyzing and Interpreting Neural Networks for NLP
Subjects:	Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
Cite as:	arXiv:2610.31520 [cs.CL]
	(or arXiv:2610.31520v1 [cs.CL] for this version)
	https://doi.org/10.48550/arXiv.2610.31520 arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Mette Friis Andersen [view email]
[v1] Thu, 22 Oct 2026 16:33:01 UTC (4,305 KB)

来源：arXiv cs.CL · arxiv.org