Tag
ADAGE is a language-agnostic pipeline for constructing challenging, translation-free benchmarks for abstract analogical reasoning. It reveals a consistent cultural reasoning gap in models across Arabic, Amharic, and Japanese.
This paper formalizes analogical reasoning in Transformers using category theory, introduces synthetic tasks to study its emergence, and reveals that it arises from geometric alignment of relational structures and functor application, with signatures also found in pretrained LLMs. The work was accepted as a Spotlight at ICML 2026.