Elamur Research⌕
☰
home / papers / 2026 / SaFe Tweets at MultiPRIDE: From Multi-Model Ensembles to Expert LLMs for Reclamation Detection

SaFe Tweets at MultiPRIDE: From Multi-Model Ensembles to Expert LLMs for Reclamation Detection

Paper · 2026 · Sara Visconti and Federica Manzi

Summary

The paper describes one team's systems for the EVALITA 2026 MultiPRIDE shared task. The task asks whether terms in tweets that may be offensive to the LGBTQIA+ community are used by the author in a reclaiming way or not. The motivation is that automated moderation tools often flag community members' self-referential use of slurs as toxic, which can silence them. The authors entered subtasks A1 (Italian, text only), A3 (English, text only) and B1 (Italian, with author biography). For Italian, they combined a logistic regression classifier, SetFit and XLM-RoBERTa through majority voting. English has a much rarer reclaimed class and weaker lexical cues. For it, they built a two-stage pipeline: DeBERTaV3 fine-tuned with LoRA makes first-pass predictions, and LLaMa 4 Scout re-examines the low-confidence cases. Lightweight ensembles worked reasonably on Italian, but English stayed difficult even with a large LLM. The error analysis attributes this to subjective and inconsistent annotation, ambiguous tweets and scarce context. The authors conclude that clearer annotation guidelines may help more than simply adding model capacity.

Key points

  • Training data is imbalanced. Italian has 879 non-reclaimed and 207 reclaimed tweets (1086 total). English has 938 non-reclaimed and 88 reclaimed tweets (1026 total), so reclaimed tweets make up only about 8% of the English set.
  • In subtask B, author biographies were available only for Italian (B1). The English data had no biographies, which made the "author belongs to the community" criterion hard to apply.
  • The Italian system (A1 and B1) is a majority-vote ensemble of logistic regression, SetFit and XLM-RoBERTa.
  • The English system (A3) is a cascade. DeBERTaV3 with LoRA handles initial filtering, and LLaMa 4 Scout (109B total parameters, 16 experts, 17B active) refines uncertain predictions.
  • The error analysis shows tweets with plausible conflicting labels. The authors argue that LLM prompts need rules that match the annotators' guidelines in order to be effective.

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, topics and concepts

📚 Shelves 2The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 works

Cite

Cite as Sara Visconti and Federica Manzi. 2026. SaFe Tweets at MultiPRIDE: From Multi-Model Ensembles to Expert LLMs for Reclamation Detection. Proceedings of the Ninth Evaluation Campaign of Natural Language Processing and Speech Tools for Italian. Final Workshop (EVALITA 2026).

The paper itself, not this page. BibTeX from the library's record of the work.

@inproceedings{visconti2026safe,
  title = {{SaFe Tweets at MultiPRIDE: From Multi-Model Ensembles to Expert LLMs for Reclamation Detection}},
  author = {Sara Visconti and Federica Manzi},
  booktitle = {Proceedings of the Ninth Evaluation Campaign of Natural Language Processing and Speech Tools for Italian. Final Workshop (EVALITA 2026)},
  year = {2026},
  url = {https://aclanthology.org/2026.evalita-1.33/}
}

References

30 references, 8 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. L. Tunstall, N. Reimers, U. E. S. Jo, L. Bates, D. Korat, M. Wasserblat, O. Pereg, Efficient Few-Shot Learning Without Prompts, arXiv (2022). doi: 10.48550/arXiv.2209.11055 . — [doi:10.48550/arxiv.2209.11055](https://doi.org/10.48550/arxiv.2209.11055) find
  2. A. Conneau, K. Khandelwal, N. Goyal, V. Chaudhary, G. Wenzek, F. Guzmán, E. Grave, M. Ott, L. Zettlemoyer, V. Stoyanov, Unsupervised Cross-lingual Representation Learning at Scale, in: Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, Association for Computatio… find
  3. P. He, J. Gao, W. Chen, DeBERTaV3: Improving DeBERTa using ELECTRA-Style Pre-Training with Gradient-Disentangled Embedding Sharing, 2021. arXiv:2111.09543 .
  4. E. J. Hu, Y. Shen, P. Wallis, Z. Allen-Zhu, Y. Li, S. Wang, L. Wang, W. Chen, LoRA: Low-Rank Adaptation of Large Language Models, in: International Conference on Learning Representations, 2022. URL: https://openreview.net/forum?id=nZeVKeeFYf9. ↗
  5. Meta AI, The Llama 4 herd: The beginning of a new era of natively multimodal AI innovation, 2025. URL: https://ai.meta.com/blog/llama-4-multimodal-intelligence/. ↗
  6. T. Oliva, Fighting Hate Speech, Silencing Drag Queens? Artificial Intelligence in Content Moderation and Risks to LGBTQ Voices Online, Sexuality & Culture (2021). doi: 10.1007/ s12119-020-09790-w . find
  7. L. Draetta, C. Ferrando, M. Cuccarini, L. James, V. Patti, ReCLAIM Project: Exploring Italian Slurs Reappropriation with Large Language Models, in: Proceedings of the Tenth Italian Conference on Computational Linguistics (CLiC-it 2024), CEUR Workshop Proceedings, Pisa, Italy, 2024, pp. 335–342. find
  8. C. Ferrando, L. Draetta, M. Madeddu, M. Sosto, V. Patti, P. Rosso, C. Bosco, J. Mata, E. Gualda, MultiPRIDE at EVALITA 2026: Overview of the Multilingual Automatic Detection of Slur Reclamation in the LGBTQ+ Context Task, in: Proceedings of the Ninth Evaluation Campaign of Natural Language Processin… find
  9. E. Nossem, Queer, Frocia, Femminiell@, Ricchione et al. – Localizing ’Queer’ in the Italian Context., Gender/Sexuality/Italy 6 (2019) 1–27. doi: 10.15781/31yc-ys20 . — [doi:10.15781/31yc-ys20](https://doi.org/10.15781/31yc-ys20) find
  10. C. Bianchi, Slurs and Appropriation: An Echoic Account, Journal of Pragmatics 66 (2014) 35–44. doi: 10.1016/j.pragma.2014.02.009 . — [doi:10.1016/j.pragma.2014.02.009](https://doi.org/10.1016/j.pragma.2014.02.009) find
  11. S. McKinnon, “Building a thick skin for each other”: The use of ‘reading’ as an interactional practice of mock impoliteness in drag queen backstage talk, Journal of Language and Sexuality 6 (2017) 90–127. doi: 10.1075/jls.6.1.04mck . — [doi:10.1075/jls.6.1.04mck](https://doi.org/10.1075/jls.6.1.04mck) find
  12. D. Hartmann, A. Oueslati, D. Staufer, L. Pohlmann, S. Munzert, H. Heuer, Lost in Moderation: How Commercial Content Moderation APIs Over- and Under-Moderate Group-Targeted Hate Speech and Linguistic Variations, in: Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems, CHI ’25… — [doi:10.1145/3706598.3713998](https://doi.org/10.1145/3706598.3713998) find
  13. D. Nozza, A. T. Cignarella, G. Damo, T. Caselli, V. Patti, HODI at EVALITA 2023: Overview of the first Shared Task on Homotransphobia Detection in Italian, in: EVALITA, volume 3473 of CEUR Workshop Proceedings, CEUR-WS.org, 2023. find
  14. E. Zsisku, A. Zubiaga, H. Dubossarsky, Hate Speech Detection and Reclaimed Language: Mitigating False Positives and Compounded Discrimination, Proceedings of the 16th ACM Web Science Conference (2024). URL: https://api.semanticscholar.org/CorpusID:269228809. ↗
  15. E. W. Pamungkas, V. Basile, V. Patti, Do You Really Want to Hurt Me? Predicting Abusive Swearing in Social Media, in: International Conference on Language Resources and Evaluation, 2020. URL: https://api.semanticscholar.org/CorpusID:218974299. ↗
  16. A. T. Cignarella, M. Sanguinetti, S. Frenda, A. Marra, C. Bosco, V. Basile, QUEEREOTYPES: A Multi-Source Italian Corpus of Stereotypes towards LGBTQIA+ Community Members, in: Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-… find
  17. L. Veloso, L. Hirlimann, P. Wicke, H. Schütze, SLAyiNG: Towards Queer Language Processing (2025). doi: 10.48550/arXiv.2509.17449 . — [doi:10.48550/arxiv.2509.17449](https://doi.org/10.48550/arxiv.2509.17449) find
  18. P. He, X. Liu, J. Gao, W. Chen, DeBERTa: Decoding-enhanced BERT with Disentangled Attention, in: International Conference on Learning Representations, 2021. URL: https://openreview.net/ forum?id=XPZIaotutsD. ↗
  19. H. Liu, D. Tam, M. Muqeeth, J. Mohta, T. Huang, M. Bansal, C. Raffel, Few-Shot Parameter-Efficient Fine-Tuning is Better and Cheaper than In-Context Learning, arXiv (2022). doi: 10.48550/arXiv. 2205.05638 .
  20. N. Reimers, I. Gurevych, Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks, in: Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing, Association for Computational Linguistics, Hong Kong, China, 2019, pp. 3982–3992. doi: 10.18653/ v1/D19-1410 .
  21. J. Robert, J. Michael, N. Steven, H. Geoffrey, Adaptive Mixtures of Local Experts, Neural Computation 3 (1991) 79–87. doi: 10.1162/neco.1991.3.1.79 . — [doi:10.1162/neco.1991.3.1.79](https://doi.org/10.1162/neco.1991.3.1.79) find
  22. S. Mu, S. Lin, A Comprehensive Survey of Mixture-of-Experts: Algorithms, Theory, and Applications, 2025. URL: https://arxiv.org/abs/2503.07137. arXiv:2503.07137 . — [arXiv:2503.07137](https://arxiv.org/abs/2503.07137) ↗
  23. W. Fedus, B. Zoph, N. Shazeer, Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity, Journal of Machine Learning Research 23 (2022) 1–39. URL: http://jmlr.org/papers/v23/21-0998.html. ↗
  24. A. Q. Jiang, A. Sablayrolles, A. Roux, A. Mensch, B. Savary, C. Bamford, D. S. Chaplot, D. de las Casas, E. B. Hanna, F. Bressand, G. Lengyel, G. Bour, G. Lample, L. R. Lavaud, L. Saulnier, M.-A. Lachaux, P. Stock, S. Subramanian, S. Yang, S. Antoniak, T. Le Scao, T. Gervet, T. Lavril, T. Wang, T. L… find
  25. D. Dai, C. Deng, C. Zhao, R. X. Xu, H. Gao, D. Chen, J. Li, W. Zeng, X. Yu, Y. Wu, Z. Xie, Y. K. Li, P. Huang, F. Luo, C. Ruan, Z. Sui, W. Liang, DeepSeekMoE: Towards Ultimate Expert Specialization in Mixture-of-Experts Language Models, in: L.-W. Ku, A. Martins, V. Srikumar (Eds.), Proceedings of th… find
  26. F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, A. Müller, J. Nothman, G. Louppe, P. Prettenhofer, R. Weiss, V. Dubourg, J. Vanderplas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot, É. Duchesnay, Scikit-learn: Machine Learning in Python, 2018. URL: https://a… find
  27. Y. Liu, M. Ott, N. Goyal, J. Du, M. Joshi, D. Chen, O. Levy, M. Lewis, L. Zettlemoyer, V. Stoyanov, RoBERTa: A Robustly Optimized BERT Pretraining Approach, in: Proceedings of the 20th Chinese National Conference on Computational Linguistics, Chinese Information Processing Society of China, Huhhot,… — A Robustly Optimized BERT Pre-training Approach with Post-training
  28. T. Akiba, S. Sano, T. Yanase, T. Ohta, M. Koyama, Optuna: A Next-generation Hyperparameter Optimization Framework, 2019. doi: 10.48550/arXiv.1907.10902 . arXiv:1907.10902 .
  29. J. Liu, Contributors, Instructor: A library for structured outputs from large language models, 2024. URL: https://github.com/instructor-ai/instructor. ↗
  30. A. Yang, A. Li, B. Yang, B. Zhang, B. Hui, B. Zheng, B. Yu, C. Gao, C. Huang, C. Lv, C. Zheng, D. Liu, F. Zhou, F. Huang, F. Hu, H. Ge, H. Wei, H. Lin, J. Tang, J. Yang, J. Tu, J. Zhang, J. Yang, J. Yang, J. Zhou, J. Zhou, J. Lin, K. Dang, K. Bao, K. Yang, L. Yu, L. Deng, M. Li, M. Xue, M. Li, P. Zh… — Qwen3 Technical Report

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/visconti-2026-safe-tweets-multipride-multi-model/