Newest papers

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

Shihui Li, Mette Andersen, Martha Lewis · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.31520 · cited by 0 works (OpenAlex) · open-access PDF

Portuguese

A metáfora é um aspecto crucial da cognição humana. Como tal, gostaríamos de entender como ela é representada em grandes modelos de linguagem (LLMs) e como ela pode ser controlada. Usamos técnicas de interpretação mecânica para identificar subespaços causais em uma variedade de LLMs que cobrem diferentes arquiteturas e tamanhos. Mostramos que somos capazes de direcionar a saída em tarefas de classificação em diferentes partes da fala. Também somos capazes de direcionar a saída em tarefas de geração, indicando que essas técnicas podem ser usadas para controlar a metafórica da saída do LLM.

Machine translation (Meta M2M100); the English original follows.

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.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Interpretability, Metaphor, Computer science, Control (management), Artificial intelligence, Range (aeronautics)

Is this paper real? This page was built from OpenAlex and arXiv records, not generated. Software can verify any reference in one call: GET https://store.agentexchange.work/gw/papers/verify?title=...&year=... (0.003 USD), fetch this record as JSON with /gw/papers/paper?id=2610.31520, or list what it cites with /gw/papers/references?id=.... Pay per call in USDC (x402) or with a 19 USD card key for 1,000 calls. Catalogue · llms.txt

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange