Multimodal UNcommonsense: From Odd to Ordinary and Ordinary to Odd

Yejin Son, Saejin Kim, Dongjun Min, Youngjae Yu · 2026

Read Multimodal UNcommonsense: From Odd to Ordinary and Ordinary to Odd by Yejin Son, Saejin Kim, Dongjun Min, Youngjae Yu (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-2602-01561.txt.

Description

Yejin Son and co-authors introduce Multimodal UNcommonsense, a benchmark for cases defying typical norms. It pairs visual scenes with odd or strange results in natural language. Models must explain odd images with plain logic or uncover unexpected interpretations in normal scenes. Their search-based in-context learning system transfers reasoning from larger to smaller models without extra training. A multimodal ensemble retriever finds relevant cases even when image-text pairs are planned to clash. Standard in-context learning methods improve by an average of 8.3% in rare, odd settings.

Catalog record

CREATORYejin Son, Saejin Kim, Dongjun Min, Youngjae Yu
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE59508
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602.01561

Browse the full library · About the Tower of Babel