Elamur Research⌕
☰
home / papers / 2026 / AIWizards at MULTIPRIDE: A Hierarchical Approach to Slur Reclamation Detection

AIWizards at MULTIPRIDE: A Hierarchical Approach to Slur Reclamation Detection

Paper · 2026 · Luca Tedeschini and Matteo Fasulo

Summary

The paper addresses Subtask B in EVALITA 2026's MultiPRIDE shared task. The task is to decide whether an LGBTQ+-related slur in a tweet is used reclamatorily (in-group, affirming) or not, using both the tweet and the author's bio. The authors assume that community members tend to reclaim slurs more often, so they split the problem into two stages. First, DeepSeek-V3.2 weakly labels each tweet-plus-bio pair with a proxy "LGBTQ+ affiliation" signal. These labels train a BERT-like user encoder. Second, the encoder's latent space is merged through a gated fusion layer with a separate encoder pretrained for hate speech detection, and the combined model classifies reclamation. On Italian and Spanish data, the dual-encoder performs statistically on par with a strong class-weighted BERT baseline but does not surpass it. The authors present the work mainly as a modular framework that can absorb user-level sociolinguistic context without hurting accuracy, and that could later add network or multimodal signals.

Key points

  • The data is heavily imbalanced. Non-reclamatory tweets make up 879 of the Italian examples (80.95%) and 743 of the Spanish ones (84.83%), with 207 and 133 reclamatory cases respectively. The authors used stratified 70/15/15 train/validation/test splits.
  • The LLM proxy labels are more balanced: about 39.6% (Italian) and 41.3% (Spanish) of users were marked as affiliated. These labels serve only as weak supervision and are not used as a user-level prediction target.
  • The baselines are nickprock/setfit-italian-hate-speech for Italian and cardiffnlp/twitter-xlm-roberta-base for Spanish. Both take the tweet and bio joined with [SEP], and results are averaged over five seeds.
  • Italian scores clearly exceed Spanish ones. The authors attribute the gap to the Italian model's SetFit design and its prior hate speech fine-tuning, rather than to properties of the languages.
  • In the error analysis, some openly hostile Italian tweets were labeled reclamatory. The authors suspect that user-level signals can dominate the gating mechanism. Code is available at github.com/LucaTedeschini/multipride.

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

🎓ETH ZurichZurich · Switzerland72 works · 5,176 citations🎓University of BolognaBologna · Italy17 works · 1,821 citations

Also on the byline: Swiss Data Science Center.

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksSocial Sciences · 1,772 worksCommunication · 97 works

Cite

Cite as Luca Tedeschini and Matteo Fasulo. 2026. AIWizards at MULTIPRIDE: A Hierarchical Approach to Slur 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{tedeschini2026aiwizards,
  title = {{AIWizards at MULTIPRIDE: A Hierarchical Approach to Slur Reclamation Detection}},
  author = {Luca Tedeschini and Matteo Fasulo},
  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-1.19/}
}

References

10 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. R. Jeshion, Pride and prejudiced: On the reclamation of slurs, Grazer Philosophische Studien 97 (2020) 106 – 137. URL: https://brill.com/view/journals/gps/97/1/article-p106_106.xml. doi: 10. 1163/18756735-09701007 . ↗
  2. L. Filardo-Llamas, A. Roldán-García, Between “queer” and “cishet men”. lexical creativity, gender identities and sexualities in linguistic labels in tumblr, Journal of Language and Discrimination 9 (2025) 1–27. URL: https://doi.org/10.3138/jld-2025-0501. doi: 10.3138/jld-2025-0501 . arXiv:https://do… — [doi:10.3138/jld-2025-0501](https://doi.org/10.3138/jld-2025-0501) ↗
  3. C. Bianchi, Slurs and appropriation: An echoic account, Journal of Pragmatics 66 (2014) 35–44. URL: https://www.sciencedirect.com/science/article/pii/S0378216614000526. doi: https://doi. org/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) ↗
  4. B. Cepollaro, D. López de Sa, The successes of reclamation, Synthese 202 (2023). doi: 10.1007/ s11229-023-04393-y . find
  5. 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
  6. F. Cutugno, A. Miaschi, A. P. Aprosio, G. Rambelli, L. Siciliani, M. A. Stranisci, Evalita 2026: Overview of the 9th evaluation campaign of natural language processing and speech tools for italian, in: Proceedings of the Ninth Evaluation Campaign of Natural Language Processing and Speech Tools for I… find
  7. DeepSeek-AI, Deepseek-v3.2: Pushing the frontier of open large language models, 2025.
  8. L. Tunstall, N. Reimers, U. E. S. Jo, L. Bates, D. Korat, M. Wasserblat, O. Pereg, Efficient few-shot learning without prompts, 2022. URL: https://arxiv.org/abs/2209.11055. arXiv:2209.11055 . — [arXiv:2209.11055](https://arxiv.org/abs/2209.11055) ↗
  9. F. Barbieri, L. Espinosa Anke, J. Camacho-Collados, XLM-T: Multilingual language models in Twitter for sentiment analysis and beyond, in: Proceedings of the Thirteenth Language Resources and Evaluation Conference, European Language Resources Association, Marseille, France, 2022, pp. 258–266. URL: ht…
  10. T.-Y. Lin, P. Goyal, R. Girshick, K. He, P. Dollár, Focal loss for dense object detection, 2018. URL: https://arxiv.org/abs/1708.02002. arXiv:1708.02002 . ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/tedeschini-2026-aiwizards-multipride-hierarchical-approach-slur/