Elamur Research⌕
☰
home / papers / 2026 / The F1 of Formula One: Applicability of Pre-trained NER Models to Brazilian TV Interview Transcripts

The F1 of Formula One: Applicability of Pre-trained NER Models to Brazilian TV Interview Transcripts

Paper · 2026 · João Pedro Gonçalves Munhoz, Luiz Felipe Guidorizzi de Oliveira, Isabella Belchior et al.

Summary

The authors want to build a large network of named entities from Roda Viva, a long-running Brazilian TV interview show. Its corpus holds 713 transcribed interviews. Before scaling up, they test whether off-the-shelf NER tools can handle this kind of text. Transcribed interviews are dialogic, fragmented and semi-spontaneous, which makes them harder than the written genres most Portuguese NER work targets.

As a pilot, they took six interviews with Brazilian Formula One drivers and extracted person mentions in three ways: manual annotation (the gold standard), a statistical-neural model from a standard NLP library, and an open-weight LLM. Both automatic methods reached similar F1 scores but made different errors:

  • Statistical model: relied heavily on capitalisation and known vocabulary. It produced mechanical false positives and missed lowercase names.
  • LLM: was more context-aware and tolerated transcription errors. Its false positives were mostly deictic kinship phrases and pronouns.

The authors prefer the LLM. They argue its errors can be reduced through better instructions and could later feed entity linking, which would support extracting social networks from Brazilian public discourse for Digital Humanities research.

Key points

  • Source corpus: 713 Roda Viva interviews from the Memória Roda Viva portal, made machine-readable by de Miranda Jr et al. (2024) and hosted in the LeGOS-UFSCar GitHub repository.
  • Pilot subcorpus: the six featured drivers are Ayrton Senna, interviewed in 1986; Nelson Piquet, in 1994; Emerson Fittipaldi, in 1995; Rubens Barrichello, in 1995 or 1996 (the paper is inconsistent); Christian Fittipaldi, in 1995; and Lucas di Grassi, in 2022. The authors transcribed the last two themselves.
  • Subcorpus size: 66 to 124 unique entities and 208 to 309 sentences per interview. Four UFSCar linguists produced the manual annotation.
  • Comparison: a dedicated neural NER model (the references cite spaCy) against an open-weight LLM. The excerpt says F1 scores were comparable but does not give the exact figures.
  • Shared limitation: both methods took only the bare name from phrases like "meu filho Luca". The LLM's tagging of phrases like "seu pai" is presented as raw material for future coreference and entity linking.

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

🧭 Topics 1Research subjects and languages the work is about.Named entity recognition · 200 works

Cite

Cite as João Pedro Gonçalves Munhoz, Luiz Felipe Guidorizzi de Oliveira, Isabella Belchior, Evandro Eduardo Seron Ruiz and Oto Araújo Vale. 2026. The F1 of Formula One: Applicability of Pre-trained NER Models to Brazilian TV Interview Transcripts. Proceedings of the 17th International Conference on Computational Processing of Portuguese (PROPOR 2026) - Vol. 2.

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

@inproceedings{munhoz2026f1,
  title = {{The F1 of Formula One: Applicability of Pre-trained NER Models to Brazilian TV Interview Transcripts}},
  author = {João Pedro Gonçalves Munhoz, Luiz Felipe Guidorizzi {de Oliveira}, Isabella Belchior, Evandro Eduardo Seron Ruiz and Oto Araújo Vale},
  booktitle = {Proceedings of the 17th International Conference on Computational Processing of Portuguese (PROPOR 2026) - Vol. 2},
  year = {2026},
  url = {https://aclanthology.org/2026.propor-2.40/}
}

References

8 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Hanne Kirstine Adriansen. 2012. Timeline interviews: A tool for conducting life history research. Qualitative Studies, 3(1):40–55. find
  2. Isaac Souza de Miranda Jr, Gabriela Wick-Pedro, Cláudia Dias de Barros, and Oto Vale. 2024. find
  3. Roda Viva Rafael Oleques Nunes, Dennis Giovani Balreira, André Suslik Spritzer, and Carla Maria Dal Sasso Freitas. 2024. A named entity recognition approach for Portuguese legislative texts using self-learning. In Proceedings of the 16th International Conference on Computational Processing of Portug… find
  4. Thierry Poibeau. 2024. Annotating References to Mythological Entities in French Literature. arXiv preprint arXiv:2412.18270. — [arXiv:2412.18270](https://arxiv.org/abs/2412.18270) find
  5. Diana Santos, Nuno Seco, Nuno Cardoso, and Rui Vilela. 2006. HAREM: An advanced NER evaluation contest for Portuguese. In Proceedings of the Fifth International Conference on Language Resources and Evaluation (LREC’06), Genoa, Italy. European Language Resources Association (ELRA). find
  6. Elisa Terumi Rubel Schneider, João Vitor Andrioli de Souza, Julien Knafou, Lucas Emanuel Silva e Oliveira, Jenny Copara, Yohan Bonescki Gumiel, Lucas Ferro Antunes de Oliveira, Emerson Cabrera Paraiso, Douglas Teodoro, and Cláudia Maria Cabral Moro Barra. 2020. BioBERTpt - a Portuguese neural langua… find
  7. Fábio Capuano de Souza, Rodrigo Nogueira, and Roberto de Alencar Lotufo. 2020. BERTimbau: Pre-trained BERT Models for Brazilian Portuguese. In Intelligent Systems, pages 403–417, Cham. Springer International Publishing. find
  8. Fábio Capuano de Souza, Rodrigo Nogueira, and Roberto de Alencar Lotufo. 2023. BERT models for Brazilian Portuguese: Pretraining, evaluation and tokenization analysis. Applied Soft Computing, 149:110901. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/munhoz-2026-f1-formula-one-applicability-pre/