Elamur Research⌕
☰
home / papers / 2026 / RacismoBR: A Manually Annotated Dataset for Racist Discourse Detection in Brazilian Portuguese

RacismoBR: A Manually Annotated Dataset for Racist Discourse Detection in Brazilian Portuguese

Paper · 2026 · João Vítor Vaz, Fabrício Benevenuto and Marcos André Gonçalves

Summary

Racism detection in Brazilian Portuguese has few dedicated resources. Existing Portuguese corpora usually fold racism into general hate speech, and their annotation guidelines are often borrowed from other cultural settings. The authors present RacismoBR, a small corpus of posts and replies from X. They collected it with targeted keyword sampling, and Black researchers labeled every item as racist or non-racist, so the labels rest on lived and sociolinguistic expertise.

On this data, they compared three families of classifiers under one controlled protocol:

  • classical machine learning,
  • fine-tuned Transformer Small Language Models,
  • instruction-tuned LLMs prompted with few-shot examples.

GPT-4.1 and BERTimbau reached the best Macro-F1. However, Wilcoxon signed-rank tests found no significant differences between models, largely because results varied a lot. Error analysis shows that implicit, euphemistic and context-dependent racism trips up all paradigms in similar ways. The authors conclude that careful, culturally informed annotation and dataset design matter more than larger architectures, and they argue for keeping humans in the moderation loop.

Key points

  • Data collection: 405 posts and replies were gathered from X via the paid API between September 2024 and January 2025. Keywords included "mulambo", "racismo", "racista", "candomblé", "africano/a" and "neguinho/a".
  • Annotation: eight Black researchers, working in four independent pairs, labeled the data in two classes (racist = 1, non-racist = 0). They followed a definition of racism adapted from Almeida (2019).
  • Results: GPT-4.1 and BERTimbau scored highest on Macro-F1, but Wilcoxon tests showed no statistically significant gap across paradigms.
  • Shared pattern: all models had higher precision on non-racist posts and higher recall on racist posts.
  • Availability and scope: the dataset is on GitHub (joaovitorvaz/RacismoBR-Dataset) under responsible disclosure. It covers single-turn texts and racism against Black people only.

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 · 18,614 worksArtificial Intelligence · 17,965 works

🧭 Topics 1Research subjects and languages the work is about.Discourse analysis · 471 works

Cite

Cite as João Vítor Vaz, Fabrício Benevenuto and Marcos André Gonçalves. 2026. RacismoBR: A Manually Annotated Dataset for Racist Discourse Detection in Brazilian Portuguese. Proceedings of the 17th International Conference on Computational Processing of Portuguese (PROPOR 2026) - Vol. 1.

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

@inproceedings{vaz2026racismobr,
  title = {{RacismoBR: A Manually Annotated Dataset for Racist Discourse Detection in Brazilian Portuguese}},
  author = {João Vítor Vaz, Fabrício Benevenuto and Marcos André Gonçalves},
  booktitle = {Proceedings of the 17th International Conference on Computational Processing of Portuguese (PROPOR 2026) - Vol. 1},
  year = {2026},
  url = {https://aclanthology.org/2026.propor-1.76/}
}

References

24 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Sílvio Almeida. 2019. Racismo Estrutural. Pólen Produções, São Paulo. find
  2. Gabriel Assis, Annie Amorim, Jonnatahn Carvalho, Daniel de Oliveira, Daniela Q. C. Vianna, and Aline Paes. 2024. Exploring Portuguese hate speech detection in low-resource settings: Lightly tuning encoder models or in-context learning of large models? In Proc. of the International Conference on Comp… find
  3. Stuart Hall, editor. 1997. Representation: Cultural Representations and Signifying Practices. SAGE Publications, London. find
  4. João Augusto Leite, Diego Silva, Kalina Bontcheva, and Carolina Scarton. 2020. Toxic language detection in social media for Brazilian Portuguese: New dataset and multilingual analysis. In Proc. of the 1st Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (AACL)… find
  5. Eloyna Augusta Mesquita Miranda. 2020. As religiães de matriz africana e o racismo religioso no Brasil. Master’s thesis, Universidade Federal da Bahia. find
  6. Mainack Mondal, Leandro Araújo Silva, and Fabrício Benevenuto. 2017. A measurement study of hate speech in social media. In Proceedings of the 28th ACM Conference on Hypertext and Social Media, HT ’17, page 85–94, New York, NY, USA. Association for Computing Machinery. find
  7. Cendra Devayana Putra and Hei-Chia Wang. 2024. Advanced bert-cnn for hate speech detection. Procedia Comput. Sci., 234(C):239–246. find
  8. Marcelo Augusto. 2021. Twitter analysis. GitHub repository. find
  9. Emily M. Bender, Timnit Gebru, Angelina McMillan-Major, and Shmargaret Shmitchell. 2021. On the dangers of stochastic parrots: Can language models be too big? In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, FAccT ’21, page 610–623, New York, NY, USA. Associat… find
  10. Su Lin Blodgett, Solon Barocas, Hal Daumé III, and Hanna Wallach. 2020. Language (technology) is power: A critical survey of “bias” in NLP. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 5454– 5476, Online. Association for Computational Linguistics.
  11. Paulo Henrique Caetano. 2020. A Palavra-Chave “Racismo” e suas Relações Lexicais no Discurso Brasileiro. Ph.D. thesis, Universidade Federal de Minas Gerais. find
  12. Bruno Cascalheira and et al. 2024. Annotation practices for socially situated language. In Proc. of the Brazilian Conference on Intelligent Systems (BRACIS). find
  13. William G. Cochran. 1977. Sampling Techniques, 3rd edition. Wiley. find
  14. Frantz Fanon. 2008. Pele Negra, Máscaras Brancas. EDUFBA, Salvador. find
  15. RD Station. 2025. As redes sociais mais usadas no Brasil e no mundo em 2025. Accessed: 2025-06-11. find
  16. Marcelo Augusto Araújo dos Reis. 2021. Predição de Comentários em Mídias Sociais sobre Discursos Racistas. Ph.D. thesis, Universidade de Brasília. find
  17. Brenda Salenave Santana, Aline Aver Vanin, and Leandro Krug Wives. 2022. Sexist hate speech: Identifying potential online verbal violence instances. In Proc. of the International Conference on Computational Processing of Portuguese (PROPOR), pages 177–187. find
  18. Xinyue Shen, Yixin Wu, Yiting Qu, Michael Backes, Savvas Zannettou, and Yang Zhang. 2025. Hatebench: benchmarking hate speech detectors on llm-generated content and hate campaigns. In Proceedings of the 34th USENIX Conference on Security Symposium, SEC ’25, USA. USENIX Association. find
  19. Leandro Silva, Mainack Mondal, Denzil Correa, Fabrício Benevenuto, and Ingmar Weber. 2016. Analyzing the targets of hate in online social media. In Proc. of the International AAAI Conference on Web and Social Media (ICWSM), pages 687–690. find
  20. Tarcízio Silva. 2022. Racismo Algorítmico: Inteligência Artificial e Discriminação nas Redes Digitais. Edições SESC, São Paulo. find
  21. Sebastião Rogério da Silva Neto. 2017. Uma abordagem computacional para identificação de indício de preconceito em textos baseada em análise de sentimentos. Master’s thesis, Universidade Federal de Alagoas. find
  22. Fábio Souza, Rodrigo Nogueira, and Roberto Lotufo. 2020. Bertimbau: Pretrained bert models for brazilian portuguese. In Proc. of the International Conference on Artificial Neural Networks (ICANN), pages 403–417. find
  23. Douglas Trajano, Rafael Bordini, and Renata Vieira. 2022. OLID-BR: Offensive language identification dataset for brazilian portuguese. Research Square Preprint. Preprint. find
  24. Teun A. van Dijk. 2008. Discourse and Power. Palgrave Macmillan. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/vaz-2026-racismobr-manually-annotated-dataset-racist/