Elamur Research⌕
☰
home / papers / 2026 / UNED at SemEval-2026 Task 9: Sentiment-Aware Transformer Models with Back-Translation Augmentation for Online polarisation Detection

UNED at SemEval-2026 Task 9: Sentiment-Aware Transformer Models with Back-Translation Augmentation for Online polarisation Detection

Paper · 2026 · Victor Garcia Sanabria, Alvaro Rodrigo and Roberto Centeno

Summary

The UNED team tackles Subtask 1 of SemEval-2026 Task 9, which is binary detection of polarised language in short Spanish social media posts. Their hypothesis is that polarised discourse relies heavily on evaluative and emotionally loaded wording, so encoders already tuned for affect should help. Using the same fine-tuning setup for every model, they compare three checkpoints: plain XLM-RoBERTa-base, a RoBERTa-base model tuned on GoEmotions, and the Twitter-specific sentiment model XLM-T (cardiffnlp). The training set is small (3,305 posts), so they enlarge it by back-translating each post Spanish→English→Spanish with Helsinki-NLP OPUS-MT models and pairing every original with one paraphrase. XLM-T gave the best validation macro-F1. Back-translation mainly made results more stable, while expanding the data with Italian translations reduced performance. The submitted system scored 0.743 macro-F1 and placed 29th of 37 teams, slightly above the baseline. The authors take this as evidence that affective pretraining carries useful signal for polarisation but does not capture its group-oriented, relational structure.

Key points

  • Data: 3,305 Spanish training posts, nearly balanced (50.2% polarised), averaging about 11 tokens each. The authors used an 80/20 stratified split for local validation and trained each configuration only once.
  • Validation macro-F1: XLM-R (no fine-tuning) scored 0.49, RoBERTa-GoEmotions 0.64, and XLM-T 0.69. XLM-T with back-translation reached a best score of 0.69, ranging from 0.65 to 0.69 across runs. The fine-tuned base XLM-R overfit badly, reaching a training F1 of 1.0.
  • Augmentation: Back-translation used nlpaug's BackTranslationAug with Helsinki-NLP/opus-mt-es-en and opus-mt-en-es, mixed 1:1 with the original data. Cross-lingual expansion from Italian dropped validation F1 to about 0.50.
  • Training setup: Hugging Face Trainer with AdamW, 2 epochs, learning rate 1e-5, batch size 16, weight decay 0.01, and maximum length 256, run on a single Tesla T4 GPU in Colab.
  • Official result: 0.743 macro-F1, ranked 29/37, compared with a baseline of 0.7266.

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

🎓Universidad Nacional de Educación a DistanciaMadrid · Spain6 works · 269 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksSocial Sciences · 1,954 worksSociology and Political Science · 626 works

🧭 Topics 1Research subjects and languages the work is about.Sentiment & emotion · 1,035 works

Cite

Cite as Victor Garcia Sanabria, Alvaro Rodrigo and Roberto Centeno. 2026. UNED at SemEval-2026 Task 9: Sentiment-Aware Transformer Models with Back-Translation Augmentation for Online polarisation Detection. 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-1.176

@inproceedings{sanabria2026uned,
  title = {{UNED at SemEval-2026 Task 9: Sentiment-Aware Transformer Models with Back-Translation Augmentation for Online polarisation Detection}},
  author = {Victor Garcia Sanabria, Alvaro Rodrigo and Roberto Centeno},
  booktitle = {Proceedings of the 20th International Workshop on Semantic Evaluation (2026)},
  year = {2026},
  doi = {10.18653/v1/2026.semeval-1.176},
  url = {https://doi.org/10.18653/v1/2026.semeval-1.176}
}

References

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

  1. Francesco Barbieri, Luis Espinosa Anke, and Jose Camacho-Collados. 2022. Xlm-t: Multilingual language models in twitter for sentiment analysis and beyond. In Proceedings of the Thirteenth Language Resources and Evaluation Conference (LREC 2022), pages 258–266.
  2. Alexis Conneau, Kartik 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 for…
  3. Dorottya Demszky, Dana Movshovitz-Attias, Jeongwoo Ko, Alan Cowen, Gaurav Nemade, and Sujith Ravi. 2020. Goemotions: A dataset of fine-grained emotions. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics (ACL 2020), pages 4040–4054. find
  4. Suchin Gururangan, Ana Marasović, Swabha Swayamdipta, Kyle Lo, Iz Beltagy, Doug Downey, and Noah A. Smith. 2020. Don’t stop pretraining: Adapt language models to domains and tasks. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 8342–8360, Online. A… find
  5. Spencer Kiesel and Sharif Amlani. 2025. Affective polarization in a word: Open-ended and self-coded evaluations of partisan affect. PLOS ONE, 20(1):e0310772. find
  6. 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
  7. 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
  8. Rico Sennrich, Barry Haddow, and Alexandra Birch. 2016. Improving neural machine translation models with monolingual data. In Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (ACL 2016), pages 86–96. find
  9. Jörg Tiedemann and Santhosh Thottingal. 2020. Opus-mt — building open translation services for the world. In Proceedings of the 22nd Annual Conference of the European Association for Machine Translation (EAMT 2020), pages 479–480. find

Updated

↑ back to top

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