Elamur Research⌕
☰
home / papers / 2026 / Exploring automatic terminology extraction from historical medical data

Exploring automatic terminology extraction from historical medical data

Paper · 2026 · Leonardo Zilio and Maria José Bocorny Finatto

Summary

Most terminology extraction tools are built for modern text and often struggle with historical writing, where spelling and specialized vocabulary were not yet standardized. Zilio and Finatto test how well current automatic terminology extraction (ATE) methods handle 18th-century Portuguese medical prose. For their test material, they take a single chapter (chapter 15 of part two) from the 1794 work known in short as Aviso a' Gente do Mar (roughly "Advice to Seafarers on Their Health"), which adapts G. Mauran's French handbook for sailors into Portuguese. A linguist annotated the chapter manually, and the authors broadened the notion of "term" to include named entities such as places, people and demographic details. They compared seven systems: two pattern-based tools, four prompted generative LLMs (each with four prompting strategies) and one BERT-like token classifier. Evaluation combined an automatic type-based f-measure against the annotation, a manual precision-only review, and a hybrid of the two. No single system was strong on its own. Combining models, however, pushed recall above 90%, and automatic output usefully extended the human annotation. This makes the approach a practical aid for Digital Humanities work on historical terminology.

Key points

  • The systems tested were TBXTools and SkE (pattern-based), DiSTER, Gemma, MedGemma and EuroLLM (generative), and MediAlbertina (BERT-like, trained for medical NER on Portuguese). BioBERTpt was also tried and performed poorly.
  • The corpus covers seven chapters, annotated in Label Studio by a single linguist without guidelines. The training portion has about 18,482 tokens and was used to derive TBXTools patterns and prompt examples. The test chapter has about 2,774 tokens and 923 types.
  • The best combination, DiSTER zero-shot + Gemma NER + TBXTools, reached an f-score of 0.5951. Pattern-based tools still achieved the highest recall and appeared most often in the top-performing combinations.
  • The LLMs showed several weaknesses:
    • Some hallucinated or copied material from the prompt; DiSTER in particular did this.
    • EuroLLM and the Gemma-based models sometimes modernized historical spelling.
    • On rare occasions, EuroLLM and MedGemma correctly merged discontinuous terms (e.g., "ventre dorido").
  • The authors note several limitations: the test set is a single chapter, there is only one annotator, and the evaluation is type-based rather than the token-based scheme used in TermEval.

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

Cite

Cite as Leonardo Zilio and Maria José Bocorny Finatto. 2026. Exploring automatic terminology extraction from historical medical data. 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{zilio2026exploring,
  title = {{Exploring automatic terminology extraction from historical medical data}},
  author = {Leonardo Zilio and Maria José Bocorny Finatto},
  booktitle = {Proceedings of the 17th International Conference on Computational Processing of Portuguese (PROPOR 2026) - Vol. 2},
  year = {2026},
  url = {https://aclanthology.org/2026.propor-2.37/}
}

References

26 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Laurence Anthony. 2004. Antconc: A learner and classroom friendly, multi-platform corpus analysis toolkit. pages 7–13. find
  2. M Teresa Cabré. 2010. Terminology and translation. Handbook of translation studies, 1:356–365. find
  3. Helena Cameron, Fernanda Olival, Renata Vieira, and Joaquim Santos. 2022. Named entity annotation of an 18th-century transcribed corpus: problems and challenges. In Proceedings of the Second Workshop on Digital Humanities and Natural Language Processing (2nd DHandNLP 2022) co-located with Internatio… find
  4. Lena De Pourcq, Marie Gregoire, and Leonardo Zilio. 2025. Exploring the power of generative artificial intelligence for automatic term extraction from small samples. In Electronic lexicography in the 21st century (eLex 2025) Intelligent Lexicography. Proceedings of the eLex 2025 conference, pages 11… find
  5. Maria José Bocorny [Org.] Finatto. 2025. Discursos Médicos no Século XVIII: genealogia de saberes e conhecimentos através da linguagem. Editora da ABRALIN. find
  6. Horace He and Thinking Machines Lab. 2025. Defeating nondeterminism in llm inference. Thinking Machines Lab: Connectionism. Https://thinkingmachines.ai/blog/defeating-nondeterminism-in-llm-inference/. find
  7. Kris Heylen and Dirk De Hertog. 2015. Automatic term extraction. Handbook of terminology, 1(01). find
  8. Matthew Honnibal, Ines Montani, Sofie Van Lan-deghem, and Adriane Boyd. 2020. spacy: Industrial-strength natural language. find
  9. Adam Kilgarriff, Vít Baisa, Jan Bušta, Miloš Jakubíček, Vojtěch Kovář, Jan Michelfeit, Pavel Rychlỳ, and Vít Suchomel. 2014. The sketch engine. Lexicography, 1(1):7–36. find
  10. Pedro Henrique Martins, João Alves, Patrick Fernandes, Nuno M Guerreiro, Ricardo Rei, Amin Farajian, Mateusz Klimaszewski, Duarte M Alves, José Pombal, Nicolas Boizard, and 1 others. 2025. Eurollm-9b: Technical report. arXiv preprint arXiv:2506.04079.
  11. Miguel Nunes, João Boné, João C Ferreira, Pedro Chaves, and Luis B Elvas. 2024. Medialbertina: an european portuguese medical language model. Computers in Biology and Medicine, 182:109233. find
  12. Antoni Oliver and Mercè Vàzquez. 2015. Tbxtools: A free, fast and flexible tool for automatic terminology extraction. In Proceedings of the international conference recent advances in natural language processing, pages 473–479. find
  13. Juri Opitz, Corina Raclé, Emanuela Boros, Andrianos Michail, Matteo Romanello, Maud Ehrmann, and Simon Clematide. 2026. Clef hipe-2026: Evaluating accurate and efficient person-place relation extraction from multilingual historical texts. arXiv preprint arXiv:2602.17663. — [arXiv:2602.17663](https://arxiv.org/abs/2602.17663) find
  14. Paulo Quaresma and Maria José Bocorny Finatto. 2020. Information extraction from historical texts: a case study. In Proceedings of the Workshop on Digital Humanities and Natural Language Processing, co-located with International Conference on the Computational Processing of Portuguese, DHandNLP@PROP… find
  15. João Rodrigues, Luís Gomes, João Silva, António Branco, Rodrigo Santos, Henrique Lopes Cardoso, and Tomás Osório. 2023. Advancing neural encoding of portuguese with transformer albertina pt. In EPIA Conference on Artificial Intelligence, pages 441–453. Springer. find
  16. 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
  17. Andrew Sellergren, Sahar Kazemzadeh, Tiam Jaroensri, Atilla Kiraly, Madeleine Traverse, Timo Kohlberger, Shawn Xu, Fayaz Jamil, Cían Hughes, Charles Lau, and 1 others. 2025. Medgemma technical report. arXiv preprint arXiv:2507.05201.
  18. Elena Senger, Yuri Campbell, Rob Van Der Goot, and Barbara Plank. 2025. Crossing domains without labels: Distant supervision for term extraction. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing: Industry Track, pages 1366–1378, Suzhou (China). Association fo… find
  19. Ayla Rigouts Terryn, Veronique Hoste, Patrick Drouin, and Els Lefever. 2020. find
  20. Termeval 2020: Shared task on automatic term extraction using the annotated corpora for term extraction research (acter) dataset. In Proceedings of the 6th International Workshop on Computational Terminology, pages 85–94. find
  21. Maxim Tkachenko, Mikhail Malyuk, Andrey Holmanyuk, and Nikolai Liubimov. 2020-2025. Label Studio: Data labeling software. Open source software available from https://github.com/HumanSignal/label-studio. ↗
  22. Renata Vieira, Fernanda Olival, Helena Cameron, Joaquim Santos, Ofélia Sequeira, and Ivo Santos. 2021. Enriching the 1758 portuguese parish memories (Alentejo) with named entities. Journal of Open Humanities Data, 7:20. find
  23. Leonardo Zilio, Maria Finatto, and Renata Vieira. 2022. Named entity recognition applied to Portuguese texts from the XVIII century. In Proceedings of the Second Workshop on Digital Humanities and Natural Language Processing (2nd DHandNLP 2022) co-located with International Conference on the Computa… find
  24. Leonardo Zilio, Rafaela R Lazzari, and Maria José B Finatto. 2024a. Can rules still beat neural networks? The case of automatic normalisation for 18th-century Portuguese texts. In Proceedings of the 16th International Conference on Computational Processing of Portuguese-Vol. 2, pages 83–92. find
  25. Leonardo Zilio, Rafaela Radünz Lazzari, and Maria Jose Bocorny Finatto. 2024b. NLP for historical Portuguese: Analysing 18th-century medical texts. In Proceedings of the 16th International Conference on Computational Processing of Portuguese-Vol. 1, pages 76–85. find
  26. A This query structure followed the one available on the model card on Huggingface. For EuroLLM and the Gemma-family models, the query was very similar, with just a slight change. Here is the query structure for EuroLLM as an example: Again, the “current_prompt” here contained the full prompt, such… find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/zilio-2026-exploring-automatic-terminology-extraction-historical/