PolyFrame at MWE-2026 AdMIRe 2: When Words Are Not Enough: Multimodal Idiom Disambiguation

Nina Hosseini-Kivanani · 2026

Read PolyFrame at MWE-2026 AdMIRe 2: When Words Are Not Enough: Multimodal Idiom Disambiguation by Nina Hosseini-Kivanani (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-2602-18652.txt.

Description

Nina Hosseini-Kivanani tackles idioms that trip models. Her 2026 paper presents PolyFrame for the MWE-2026 AdMIRe2 task on idiom disambiguation. Idioms carry meanings not following from their words, a problem growing across languages. PolyFrame runs one pipeline for image-plus-text and text-only caption ranking.

All variants keep frozen CLIP-style encoders plus the multilingual BGE M3 encoder, training light modules only: logistic regression and sentence-type prediction, idiom synonym substitution, distractor-aware scoring, Borda fusion. From a CLIP baseline, idiom-aware paraphrasing and sentence-type classification lift English top-rank accuracy to sixty percent, with the same score in zero-shot Portuguese transfer. On the multilingual blind test across fifteen languages, systems average roughly one third top-rank. Ablations point to idiom-aware rewriting as the main driver, with sentence-type prediction and fusion adding robustness. The findings show effective idiom disambiguation is possible without fine-tuning encoders.

Catalog record

CREATORNina Hosseini-Kivanani
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE23 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-18652

Browse the full library · About the Tower of Babel