Elamur Research⌕
☰
home / papers / 2026 / LST at MWE-2026 AdMIRe 2: Advancing Multimodal Idiomaticity Representation

LST at MWE-2026 AdMIRe 2: Advancing Multimodal Idiomaticity Representation

Summary

The paper describes a text-only submission to the AdMIRe 2.0 shared task. In this task, a system receives a potentially idiomatic nominal compound, a context sentence, and five image captions. It must order the captions by how well each one matches the compound's meaning in context. AdMIRe 2.0 expands the earlier English–Portuguese benchmark to 15 languages. Labeled data exists only for the original two, and the test languages stayed hidden until evaluation. With so little supervision, the authors skip training entirely. They prompt three LLMs (GPT-4o, Qwen-Plus and DeepSeek) zero-shot, telling each model the language of the expression, and combine the three rankings with hard majority voting. This simple ensemble placed second on the official leaderboard. The authors also tested a pairwise procedure guided by usage type on Chinese. It gave no overall gain, though it helped on idiomatic cases. The work shows that off-the-shelf multilingual LLMs can handle idiom interpretation competitively, including in low-resource languages, without any task-specific training.

Key points

  • AdMIRe 1.0 Subtask A gave participants only 70 English and 32 Portuguese training items. AdMIRe 2.0 tests on 15 languages, among them several with scarce resources, for example Georgian, Igbo, Kazakh and Uzbek.
  • The prompt names the language explicitly ([LANG]), which the authors found can raise accuracy. The models return only a reordered list of caption indices.
  • GPT-4o and Qwen-Plus were chosen because, in pilot runs on AdMIRe 1.0, they scored close to the best team (CTYUN-AI). DeepSeek was added as a third model so that majority voting stays stable.
  • On the official AdMIRe 2.0 test set, Top 1 Acc ranged from 0.32 (Uzbek) to 0.53 (Portuguese-Brazil). NDCG@5 ranged from 0.71 (Igbo) to 0.81 (Portuguese-Brazil).
  • On the Chinese subset, the type-aware pairwise ranking left Top 1 Acc unchanged at 0.36 and lowered DCG by 0.01. Idiomatic DCG improved by 0.17, but literal scores fell.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

Fields: Linguistics & languages

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.mwe-1.27

@inproceedings{qiu2026lst,
  title = {{LST at MWE-2026 AdMIRe 2: Advancing Multimodal Idiomaticity Representation}},
  author = {Qiu, Le and Hsu, Yu-Yin and Chersoni, Emmanuele},
  booktitle = {Proceedings of the 22nd Workshop on Multiword Expressions (MWE 2026)},
  year = {2026},
  doi = {10.18653/v1/2026.mwe-1.27},
  url = {https://doi.org/10.18653/v1/2026.mwe-1.27}
}

References

21 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Doğukan Arslan, Rodrigo Wilkens, Wei He, Dilara Torunoğlu-Selamet, Thomas Pickard, Aline Villavicencio, Adriana S. Pagano, and Gülşen Eryiğit. 2026. MWE-2026 Shared Task 2: AdMIRe 2 - Advancing Multimodal Idiomaticity Representation. In Proceedings of the 22nd Workshop on Multiword Expressions (… find
  2. Yuming Fan, Dongming Yang, Zefeng Cai, and Binghuai Lin. 2025. CTYUN-AI at SemEval-2025 task 1: Learning to rank for idiomatic expressions. In Proceedings of the 19th International Workshop on Semantic Evaluation (SemEval-2025), pages 16– 19, Vienna, Austria. Association for Computational Linguistic… find
  3. Wei He, Marco Idiart, Carolina Scarton, and Aline Villavicencio. 2024. Enhancing idiomatic representation in multiple languages via an adaptive contrastive triplet loss. arXiv preprint arXiv:2406.15175. find
  4. Aixin Liu, Bei Feng, Bing Xue, Bingxuan Wang, Bochao Wu, Chengda Lu, Chenggang Zhao, Chengqi Deng, Chenyu Zhang, Chong Ruan, and 1 others. 2024. Deepseek-v3 technical report. arXiv preprint arXiv:2412.19437. find
  5. Siyu Long, Ran Wang, Kun Tao, Jiali Zeng, and Xinyu Dai. 2020. Synonym knowledge enhanced reader for chinese idiom reading comprehension. In Proceedings of the 28th International Conference on Computational Linguistics, pages 3684–3695. find
  6. Tomas Mikolov. 2013. Efficient estimation of word representations in vector space. arXiv preprint arXiv:1301.3781, 3781. find
  7. Jeff Mitchell and Mirella Lapata. 2010. Composition in distributional models of semantics. Cognitive science, 34(8):1388–1429. find
  8. Wiebke Petersen, Lara Eulenpesch, Ann Piho, Julio Julio, and Victoria Lohner. 2025. Transformer25 at SemEval-2025 task 1: A similarity-based approach. In Proceedings of the 19th International Workshop on Semantic Evaluation (SemEval-2025), pages 2311– 2317, Vienna, Austria. Association for Computati… find
  9. Thomas Pickard, Aline Villavicencio, Maggie Mi, Wei He, Dylan Phelps, and Marco Idiart. 2025. SemEval-2025 task 1: AdMIRe - advancing multimodal idiomaticity representation. In Proceedings of the 19th International Workshop on Semantic Evaluation (SemEval-2025), pages 2597–2609, Vienna, Austria. Ass… find our card
  10. Vikas Raunak, Arul Menezes, Matt Post, and Hany Hassan. 2023. Do GPTs produce less literal translations? In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 1041–1050, Toronto, Canada. Association for Computational Linguistics. find
  11. Runyang You, Xinyue Mei, and Mengyuan Zhou. 2025. PALI-NLP at SemEval 2025 task 1: Multimodal idiom recognition and alignment. In Proceedings of the 19th International Workshop on Semantic Evaluation (SemEval-2025), pages 1211–1216, Vienna, Austria. Association for Computational Linguistics. find
  12. Mo Yu and Mark Dredze. 2015. Learning composition models for phrase embeddings. Transactions of the Association for Computational Linguistics, 3:227– 242. find
  13. Ziheng Zeng and Suma Bhat. 2022. Getting bart to ride the idiomatic train: Learning to represent idiomatic expressions. Transactions of the Association for Computational Linguistics, 10:1120–1137. find
  14. Ying Sha, Mingmin Wu, Zhi Zeng, Xing Ge, Zhongqiang Huang, and Huan Wang. 2023. A prompt-based representation individual enhancement method for chinese idiom reading comprehension. In International Conference on Database Systems for Advanced Applications, pages 682–698. Springer. find
  15. Vered Shwartz and Ido Dagan. 2019. Still a pain in the neck: Evaluating text representations on lexical composition. Transactions of the Association for Computational Linguistics, 7:403–419. find our card
  16. Qwen Team. 2025. Qwen3 technical report. Preprint, arXiv:2505.09388. find our card
  17. Dilara Torunoğlu-Selamet, Dogukan Arslan, Rodrigo Wilkens, Wei He, Doruk Eryiğit, Thomas Pickard, Adriana S. Pagano, Aline Villavicencio, Gülşen Eryiğit, Ágnes Abuczki, Aida Cardoso, Alesia Lazarenka, Dina Almassova, Amalia Mendes, Anna Kanellopoulou, Antoni Brosa-Rodríguez, Baiba Saulite, Beata… find our card
  18. Xinyu Wang, Hongsheng Zhao, Tan Yang, and Hongbo Wang. 2020. Correcting the misuse: A method for the chinese idiom cloze test. In Proceedings of Deep Learning Inside Out (DeeLIO): The First Workshop on Knowledge Extraction and Integration for Deep Learning Architectures, pages 1–10. find
  19. Mingmin Wu, Yuxue Hu, Yongcheng Zhang, Zeng Zhi, Guixin Su, and Ying Sha. 2024. Mitigating idiom inconsistency: A multi-semantic contrastive learning method for chinese idiom reading comprehension. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 19243–19251. find
  20. Wenpeng Yin and Hinrich Schütze. 2014. An exploration of embeddings for generalized phrases. In Proceedings of the ACL 2014 Student Research Workshop, pages 41–47. find
  21. Wenpeng Yin and Hinrich Schütze. 2016. Discriminative phrase embedding for paraphrase identification. arXiv preprint arXiv:1604.00503. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/qiu-2026-lst-mwe-2026-admire-2/