Elamur Research⌕
☰
home / papers / 2026 / Taien at SemEval-2026 Task 9: Multilingual Polarization Detection Using Transformer-based Models

Taien at SemEval-2026 Task 9: Multilingual Polarization Detection Using Transformer-based Models

Paper · 2026 · Saida Taien and Palash Hossen

Summary

The authors tackle Subtask 1 of SemEval-2026 Task 9. The task is to decide whether a short text carries ideologically polarized content, a binary judgment that differs from ordinary sentiment analysis. They use the POLAR dataset, which contains social-media and news texts on elections, conflicts, gender rights and migration in 22 languages. The languages range from well-resourced ones like English and Chinese to low-resource ones such as Hausa, Odia and Burmese.

They fine-tune XLM-RoBERTa-base and mDeBERTa-v3-base with one shared pipeline for all languages: light Unicode and whitespace cleanup, 128-token inputs, AdamW and 3 epochs. The final prediction averages the two models' class probabilities (soft voting). The system uses no external corpora and no explicit handling of class imbalance.

On the official test sets, the ensemble reaches an average weighted F1 of 0.788 and an accuracy of 0.822. It ranks first for Burmese, but Italian and Odia remain weak. The work matters as a simple, reproducible multilingual baseline, and it shows that implicit, context-dependent polarization is still hard to detect.

Key points

  • Data size: Training sets range from about 1,700 texts (Punjabi) to about 7,000 (Swahili). Test sets contain 809 to 3,147 texts per language.
  • Model comparison: Average weighted F1 is 0.781 for XLM-RoBERTa-base, 0.774 for mDeBERTa-v3-base and 0.788 for the ensemble. The paper reports these as validation scores, yet the same 0.788 is also given as the test average.
  • Leaderboard positions:
    • Burmese: 1st place (F1 0.8913).
    • Persian: 2nd place.
    • Swahili: 3rd place.
    • Hausa: 5th place.
    • Weakest languages: Italian (F1 0.5798) and Odia (F1 0.7309).
  • Ablations:
    • Soft voting adds roughly +0.9 Macro-F1 over majority voting.
    • Selecting the best validation checkpoint beats keeping last-epoch weights.
  • Error analysis: Errors are mostly false negatives, caused by non-literal phrasing, implicit framing and missing discourse context.
  • Code: Available on GitHub at saida-taien/Multilingual-Polarization-22lang.

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

 Institutions

🎓University of ChittagongChittagong · Bangladesh4 works · 5 citations

Also on the byline: BGC Trust University Bangladesh.

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 works

🧭 Topics 1Research subjects and languages the work is about.Multilingual & cross-lingual NLP · 828 works

Cite

Cite as Saida Taien and Palash Hossen. 2026. Taien at SemEval-2026 Task 9: Multilingual Polarization Detection Using Transformer-based Models. Proceedings of the 20th International Workshop on Semantic Evaluation (2026).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.semeval-1.260

@inproceedings{taien2026taien,
  title = {{Taien at SemEval-2026 Task 9: Multilingual Polarization Detection Using Transformer-based Models}},
  author = {Saida Taien and Palash Hossen},
  booktitle = {Proceedings of the 20th International Workshop on Semantic Evaluation (2026)},
  year = {2026},
  doi = {10.18653/v1/2026.semeval-1.260},
  url = {https://doi.org/10.18653/v1/2026.semeval-1.260}
}

References

21 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. gatekeepers, and the price of bipartisanship. In Proceedings of the 2018 World Wide Web Conference, WWW ’18, page 913–922, Republic and Canton of Geneva, CHE. International World Wide Web Conferences Steering Committee. find
  2. Abdullah Ali and Birhanu Eshete. 2020. Best-effort adversarial approximation of black-box malware classifiers. Preprint, arXiv:2006.15725. — [arXiv:2006.15725](https://arxiv.org/abs/2006.15725) find
  3. Adem Chanie Ali, Seid Muhie Yimam, Abinew Ali Ayele, Chris Biemann, and Martin Semmann. 2025. Silenced voices: social media polarization and women’s marginalization in peacebuilding during the northern ethiopia war. i-com, 24(2):407–432. find
  4. G. Banim. 2025. Very large online platforms—how big is your polarization footprint? towards a metric to give eu citizens transparency around an online systemic risk driving conflict in our societies. Build Up. find
  5. Alexis Conneau, Kartikay Khandelwal, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer, and Veselin Stoyanov. 2020a. Unsupervised cross-lingual representation learning at scale. In Proceedings of the 58th Annual Meeting of the Association f… find
  6. Alexis Conneau, Kartikay Khandelwal, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer, and Veselin Stoyanov. 2020b. Unsupervised cross-lingual representation learning at scale. Preprint, arXiv:1911.02116. find
  7. Alexis Conneau, Kartikay Khandelwal, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer, and Veselin Stoyanov. 2020c. Unsupervised cross-lingual representation learning at scale. In Proceedings of the 58th Annual Meeting of the Association f… find
  8. Michael Conover, Jacob Ratkiewicz, Matthew Francisco, Bruno Goncalves, Filippo Menczer, and Alessandro Flammini. 2021. Political polarization on twitter. Proceedings of the International AAAI Conference on Web and Social Media, 5(1):89–96. find
  9. Yi Dai, Hao Lang, Yinhe Zheng, Fei Huang, and Yongbin Li. 2023. Long-tailed question answering in an open world. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 6362–6382, Toronto, Canada. Association for Computational Linguis… find
  10. Rajesh Kumar Das, Mirajul Islam, Md Mahmudul Hasan, Sultana Razia, Mocksidul Hassan, and Sharun Akter Khushbu. 2023. Sentiment analysis in multilingual context: Comparative analysis of machine learning and hybrid deep learning models. Heliyon, 9(9):e20281. find
  11. Kiran Garimella, Gianmarco De Francisci Morales, Aristides Gionis, and Michael Mathioudakis. 2018. Political discourse on social media: Echo chambers, Pengcheng He, Jianfeng Gao, and Weizhu Chen. 2021. Debertav3: Improving deberta using electra-style pre-training with gradient-disentangled embedding… find
  12. Dou Hu, Lingwei Wei, Yaxin Liu, Wei Zhou, and Songlin Hu. 2023. Ucas-iie-nlp at semeval-2023 task 12: Enhancing generalization of multilingual bert for low-resource sentiment analysis. Preprint, arXiv:2306.01093. find
  13. Ankit Kumar, Shivam Mishra, and Anil Singh. 2024. Multilingual hate speech and offensive language detection using transformer-based models. Scientific Reports, 14(1):60210. find
  14. Saif Mohammad, Svetlana Kiritchenko, Parinaz Sobhani, Xiaodan Zhu, and Colin Cherry. 2016. SemEval-2016 task 6: Detecting stance in tweets. In Proceedings of the 10th International Workshop on Semantic Evaluation (SemEval-2016), pages 31– 41, San Diego, California. Association for Computational Ling…
  15. Usman Naseem, Robert Geislinger, Juan Ren, Sarah Kohail, Rudy Garrido Veliz, P Sam Sahil, Yiran Zhang, Marco Antonio Stranisci, Idris Abdulmumin, Özge Alacam, Cengiz Acarürk, Aisha Jabr, Saba Anwar, Abinew Ali Ayele, Elena Tutubalina, Aung Kyaw Htet, Xintong Wang, Surendrabikram Thapa, Tanmoy Chakra… — SemEval-2026 Task 9: Detecting Multilingual, Multicultural and Multievent Online Polarization
  16. Usman Naseem, Robert Geislinger, Juan Ren, Sarah Kohail, Rudy Garrido Veliz, P Sam Sahil, Yiran Zhang, Marco Antonio Stranisci, Idris Abdulmumin, Özge Alacam, Cengiz Acartürk, Aisha Jabr, Saba Anwar, Abinew Ali Ayele, Simona Frenda, Alessandra Teresa Cignarella, Elena Tutubalina, Oleg Rogov, Aung Ky… — POLAR: A Benchmark for Multilingual, Multicultural, and Multi-Event Online Polarization
  17. Matúš Pikuliak, Marián Šimko, and Mária Bieliková. 2021. Cross-lingual learning for text processing: A survey. Expert Systems with Applications, 165:113765. find
  18. Pradeep Kumar Roy. 2024. Deep ensemble network for sentiment analysis in bi-lingual low-resource languages. ACM Trans. Asian Low-Resour. Lang. Inf. Process., 23(1). find
  19. Dominik Schlechtweg, Barbara McGillivray, Simon Hengchen, Haim Dubossarsky, and Nina Tahmasebi. 2020. SemEval-2020 task 1: Unsupervised lexical semantic change detection. In Proceedings of the Fourteenth Workshop on Semantic Evaluation, pages 1–23, Barcelona (online). International Committee for Com…
  20. Weiqi Wang, Baixuan Xu, Tianqing Fang, Lirong Zhang, and Yangqiu Song. 2023. KnowComp at SemEval-2023 task 7: Fine-tuning pre-trained language models for clinical trial entailment identification. In Proceedings of the 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 1–9, Toro… find
  21. Yuxuan Wang, Wanxiang Che, Ivan Titov, Shay B. Cohen, Zhilin Lei, and Ting Liu. 2021. A closer look into the robustness of neural dependency parsers using better adversarial examples. In Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021, pages 2344–2354, Online. Association…

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/taien-2026-taien-semeval-2026-task-9/