MWE-2026 Shared Task: AdMIRe 2 Advancing Multimodal Idiomaticity Representation

Idiomatic expressions present a unique challenge in NLP, as their meanings are often not directly inferable from their constituent words. Despite recent advancements in large language models, idiomaticity remains a significant obstacle to robust semantic representation. We present datasets and task results for MWE-2026 Shared Task 2: Advancing Multimodal Idiomaticity Representation 2 (AdMIRe 2), which challenges the community to assess and improve models’ ability to interpret idiomatic expressions in multimodal contexts across multiple languages. Participants competed in an image ranking task in which, for each item, systems receive a context sentence containing a potentially idiomatic expression (PIE) and five candidate images. Participating systems are required to predict the sentence type (i.e., idiomatic vs. literal) for the given context and rank the images by how well they depict the intended meaning in that context. Among the participating systems the most effective methods include pipelines utilizing closed-source commercial models such as Gemini 2.5 and GPT-5, and employing chain-of-thought reasoning strategies. Methods to mitigate language models’ bias towards literal interpretations and ensembles to smooth out variance were common.