Whether, Not Which: Mechanistic Interpretability Reveals Dissociable Affect Reception and Emotion Categorization in LLMs

Michael Keeman · 2026

Read Whether, Not Which: Mechanistic Interpretability Reveals Dissociable Affect Reception and Emotion Categorization in LLMs by Michael Keeman (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-22295.txt.

Description

Keeman tests whether emotion circuits in language models read real feeling or just emotion words. Past studies used prompts with clear emotion keywords, leaving that doubt open. Keeman uses clinical vignettes that show emotion through situations and behavior, with keywords removed. He tests six models with probing, activation patching, knockouts, and geometry checks. One mechanism detects that content is emotional, with near-perfect accuracy in all six models. The other maps feelings to labels and leans partly on keywords. Patching shows both kinds share one inner space.

Catalog record

CREATORMichael Keeman
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE74517
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.22295

Browse the full library · About the Tower of Babel