Elamur Research⌕
☰
home / papers / 2026 / Team JAT at SemEval-2026 Task 9: Enhancing Polarization Detection with Cross-Lingual Transfer and Feature Fusion

Team JAT at SemEval-2026 Task 9: Enhancing Polarization Detection with Cross-Lingual Transfer and Feature Fusion

Paper · 2026 · Aleksandra Matkowska, Taya Lin and Yu-Chun Chao

Summary

The paper describes Team JAT's system for SemEval-2026 Task 9 (POLAR), Subtask 1. The task is to decide whether a short online post is polarized, which here means it frames an issue as a hostile, zero-sum conflict between groups. The authors cover 13 Indo-European languages from the organizers' datasets. Several of these are strongly class-imbalanced, with Hindi and Odia the most skewed.

They tried two approaches:

  • Feature fusion (English only): a RoBERTa classifier combined with handcrafted sentiment and intensity features, plus EDA augmentation and class weighting.
  • LoRA fine-tuning: XLM-R large trained on various language groupings, with weighted loss and per-language decision thresholds. HerBERT and RoBERTa served as monolingual reference models.

The handcrafted features added nothing, and augmentation hurt results. The submitted model, trained jointly on all 13 languages, reached a mean macro-F1 of 0.763. This is close to the official baseline (0.769) and far above random (0.491).

The study shows that pooling data from closely related languages can help a multilingual encoder. However, dedicated monolingual models still performed best for high-resource languages.

Key points

  • Fusion features: a 5-dimensional vector made of the VADER compound score, the share of uppercase characters, the number of exclamation marks, and NRC Emotion Lexicon counts for anger and fear.
    • Adding EDA lowered dev macro-F1 from 0.809 to 0.716.
    • On the English test set, plain fine-tuned RoBERTa scored 0.791 and RoBERTa with features scored 0.788.
  • Submitted 13-language XLM-R LoRA model:
    • It outperformed the official baseline in 8 of 13 languages, with the largest gains in Polish (+0.052), German (+0.050) and Spanish (+0.050).
    • It fell well short in Italian (−0.138), Punjabi (−0.083) and Farsi (−0.054).
    • Its best ranking was 2nd place for Odia.
  • Effect of related-language data:
    • Polish on XLM-R improved from 0.768 to 0.806 when Russian data was added, compared with 0.775 when English was added instead.
    • Adding Chinese slightly lowered English (0.720 → 0.713).
  • Monolingual models: the best Polish score (0.827) came from HerBERT with LoRA. Russian performed best when trained alone (0.790).
  • Generalization: the final model fell from 0.816 on the dev set to 0.763 on the test set, which suggests limited generalization.

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 TübingenTübingen · Germany75 works · 41,081 citations

 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 Aleksandra Matkowska, Taya Lin and Yu-Chun Chao. 2026. Team JAT at SemEval-2026 Task 9: Enhancing Polarization Detection with Cross-Lingual Transfer and Feature Fusion. 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-1.302

@inproceedings{matkowska2026team,
  title = {{Team JAT at SemEval-2026 Task 9: Enhancing Polarization Detection with Cross-Lingual Transfer and Feature Fusion}},
  author = {Aleksandra Matkowska, Taya Lin and Yu-Chun Chao},
  booktitle = {Proceedings of the 20th International Workshop on Semantic Evaluation (2026)},
  year = {2026},
  doi = {10.18653/v1/2026.semeval-1.302},
  url = {https://doi.org/10.18653/v1/2026.semeval-1.302}
}

References

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

  1. Hunt Allcott, Luca Braghieri, Sarah Eichmeyer, and Matthew Gentzkow. 2020. The welfare effects American Economic Review, of social media. 110(3):629–676. find
  2. Jason Ansel, Edward Yang, Horace He, Natalia Gimelshein, Animesh Jain, Michael Voznesensky, Bin Bao, Peter Bell, David Berard, Evgeni Burovski, Geeta Chauhan, Anjali Chourdia, Will Constable, Alban Desmaison, Zachary DeVito, Elias Ellison, Will Feng, Jiong Gong, Michael Gschwind, Brian Hirsh, Sherlo… find
  3. A. Bankula. 2025. Cross-linguistic transfer in multilingual NLP: The role of language families and morphology. CoRR, abs/2505.13908. find
  4. Cambridge University Press. n.d. Polarization. Cambridge Dictionary. Accessed: 2025-12-21. find
  5. Matteo Cinelli, Gianmarco De Francisci Morales, Alessandro Galeazzi, Walter Quattrociocchi, and Michele Starnini. 2021. The echo chamber effect on social media. Proceedings of the National Academy of Sciences, 118(9):e2023301118. find
  6. Alexis Conneau, Kartikay Khandelwal, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer, and Veselin Stoyanov. 2020. Unsupervised cross-lingual representation learning at scale. In Proceedings of the 58th Annual Meeting of the Association fo… find
  7. Sophie Henning, William Beluch, Alexander Fraser, and Annemarie Friedrich. 2023. A survey of methods for addressing class imbalance in deep-learning based natural language processing. In Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics, page… find
  8. Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, and Weizhu Chen. 2021. LoRA: Low-rank adaptation of large language models. CoRR, abs/2106.09685.
  9. C. Hutto and Eric Gilbert. 2014. Vader: A parsimonious rule-based model for sentiment analysis of social media text. Proceedings of the International AAAI Conference on Web and Social Media, 8(1):216–225. find
  10. Saif M Mohammad and Peter D Turney. 2013. Nrc emotion lexicon. National Research Council, Canada, 2:234. find
  11. Robert Mroczkowski, Piotr Rybak, Alina Wróblewska, and Ireneusz Gawlik. 2021. HerBERT: Efficiently pretrained transformer-based language model for Polish. In Proceedings of the 8th Workshop on Balto-Slavic Natural Language Processing, pages 1–10, Kiyv, Ukraine. Association for Computational Linguist… find
  12. Ajay Nagar, Anmol Bhasin, and Gaurav Mathur. 2019. Text classification using gated fusion of n-gram features and semantic features. Computación y Sistemas, 23(3):1015–1020. find
  13. Usman Naseem, Robert Geislinger, Juan Ren, Sarah Kohail, Rudy Garrido Veliz, P Sam Sahil, Yiran Zhang, Marco Antonio Stranisci, Idris Abdulmumin, "Ozge Alacam, Cengiz Acar"urk, Aisha Jabr, Saba Anwar, Abinew Ali Ayele, Elena Tutubalina, Aung Kyaw Htet, Xintong Wang, Surendrabikram Thapa, Tanmoy Chak… — SemEval-2026 Task 9: Detecting Multilingual, Multicultural and Multievent Online Polarization
  14. 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
  15. Steve Nwaiwu. 2025. Parameter-efficient fine-tuning for low-resource text classification: a comparative study of LoRA, IA3, and ReFT. Frontiers in Big Data, 8(1677331). find
  16. Fabian Pedregosa, Gaël Varoquaux, Alexandre Gramfort, Vincent Michel, Bertrand Thirion, Olivier Grisel, Mathieu Blondel, Peter Prettenhofer, Ron Weiss, Vincent Dubourg, Jake Vanderplas, Alexandre Passos, David Cournapeau, Matthieu Brucher, Matthieu Perrot, and Édouard Duchesnay. 2011. Scikit-learn:… find
  17. POLAR SemEval Organizers. 2026. POLAR @ SemEval-2026: Official leaderboards. https://github.com/Polar-SemEval/Leaderboards. Accessed: 2026-05-01. ↗
  18. Almog Simchon, William J Brady, and Jay J Van Bavel. 2022. Troll and divide: the language of online polarization. PNAS Nexus, 1(1):pgac019. find
  19. Chi Sun, Xipeng Qiu, Yige Xu, and Xuanjing Huang. 2019. How to fine-tune BERT for text classification? CoRR, abs/1905.05583.
  20. Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pier-ric Cistac, Tim Rault, Rémi Louf, Morgan Funtowicz, Joe Davison, Sam Shleifer, Patrick von Platen, Clara Ma, Yacine Jernite, Julien Plu, Canwen Xu, Teven Le Scao, Sylvain Gugger, Mariama Drame, Quentin Lho… — Transformers: State-of-the-Art Natural Language Processing

Updated

↑ back to top

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