Emergent Analogical Reasoning in Transformers

Gouki Minegishi, Jingyuan Feng, Hiroki Furuta, Takeshi Kojima Yusuke Iwasawa Yutaka Matsuo, The University of Tokyo · 2026

Read Emergent Analogical Reasoning in Transformers by Gouki Minegishi, Jingyuan Feng, Hiroki Furuta, Takeshi Kojima Yusuke Iwasawa Yutaka Matsuo, The University of Tokyo (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-2602-01992.txt.

Description

Gouki Minegishi and colleagues frame analogical reasoning as inferring links across categories, via class functors. They build synthetic tasks testing emergence, finding it keyed to data traits, optimization choices and model scale. Mechanistic study shows reasoning splits into geometric alignment of relational links in vector space and functor use within the Transformer. These mechanisms move relational patterns from one category to another, making analogy, with the same trends in trained models. The work moves analogy from abstract notion to concrete fact in neural networks.

Catalog record

CREATORGouki Minegishi, Jingyuan Feng, Hiroki Furuta, Takeshi Kojima Yusuke Iwasawa Yutaka Matsuo, The University of Tokyo
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE60269
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602.01992

Browse the full library · About the Tower of Babel