Elamur Research⌕
☰
home / papers / 2026 / An Enhanced Training-Free Pipeline for Entity Recognition and Linking: A Low-Resource Case Study – 20-th Century Historical Medical Texts

An Enhanced Training-Free Pipeline for Entity Recognition and Linking: A Low-Resource Case Study – 20-th Century Historical Medical Texts

Paper · 2026 · Phu-Vinh Nguyen and Vera Danilova

Summary

Biomedical entity linking (EL) systems are usually trained on large modern corpora, so they struggle with 20th-century medical writing, where terms have shifted in form and meaning or fallen out of use. The authors propose CbEL (confidence-based entity linking), a training-free pipeline built on an LLM. A NER model first detects mentions. The LLM then proposes search keywords, and candidates are retrieved from a knowledge base with fuzzy and n-gram matching. Next, the LLM picks a candidate and assigns it a confidence score. Links that pass a threshold are accepted and added to the context to help resolve harder mentions over several iterations. Links that never become confident are discarded. The authors also release MedHistory, a set of British Medical Journal texts annotated by experts against the Human Disease Ontology. CbEL was compared with the few-shot OneNet framework. It performs slightly better on historical texts, about the same on NCBI, and clearly better on news-domain benchmarks, while producing far fewer spurious candidates. The work offers a way to connect historical medical vocabulary to current ontologies without labelled training data.

Key points

  • MedHistory dataset: 514 BMJ texts sampled across the 20th century and digitised with DeepSeek-OCR. Eight volunteer experts (five physicians, three medical historians) annotate links to the Human Disease Ontology through a custom web tool that includes a "historical term" flag. So far, 235 unique entities are annotated, and 54 of them form the historical-only subset MedHistory-hs.
  • Setup: Both pipelines run on Qwen-3-8B quantized to 4 bits, and they share a single pre-trained NER model. The confidence threshold τ is 0.75. Evaluation reports pipeline-level NER F1, disambiguation accuracy, and end-to-end F1.
  • Benchmarks: MedHistory, MedHistory-hs, the NCBI Disease Corpus, AIDA-CoNLL, MSNBC, and KORE50. CbEL's largest advantage over OneNet appears on the news-domain sets.
  • Error analysis:
    • CbEL produces roughly 90 candidates on average, about 12% of which end up correctly linked, with only around 3 wrong assignments.
    • OneNet generates about 290 candidates, recovering only ~3 extra correct links while adding ~5 errors.
    • Both systems miss obsolete terms such as "Bright's disease".
  • Availability: Code is on GitHub (ActDisease/CbEL) and data on Hugging Face (npvinHnivqn/MedHistory). The project is funded by the ERC ActDisease project at Uppsala University.

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

🎓Uppsala UniversityUppsala · Sweden64 works · 13,164 citations

 Fields, topics and concepts

📚 Shelves 6The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksBiochemistry, Genetics and Molecular Biology · 234 worksMolecular Biology · 231 works

🧭 Topics 2Research subjects and languages the work is about.Named entity recognition · 200 worksLow-resource languages · 267 works

Cite

Cite as Phu-Vinh Nguyen and Vera Danilova. 2026. An Enhanced Training-Free Pipeline for Entity Recognition and Linking: A Low-Resource Case Study – 20-th Century Historical Medical Texts. Proceedings of the 1st Workshop on Linguistic Analysis for Health (HeaLing 2026).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.healing-1.8

@inproceedings{nguyen2026enhanced,
  title = {{An Enhanced Training-Free Pipeline for Entity Recognition and Linking: A Low-Resource Case Study – 20-th Century Historical Medical Texts}},
  author = {Phu-Vinh Nguyen and Vera Danilova},
  booktitle = {Proceedings of the 1st Workshop on Linguistic Analysis for Health (HeaLing 2026)},
  year = {2026},
  doi = {10.18653/v1/2026.healing-1.8},
  url = {https://doi.org/10.18653/v1/2026.healing-1.8}
}

References

23 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Nitish Gupta, Sameer Singh, and Dan Roth. 2017. Entity linking via joint encoding of types, descriptions, and context. In Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 2681–2690, Copenhagen, Denmark. Association for Computational Linguistics.
  2. Oshin Agarwal and Daniel M. Bikel. 2020. Entity linking via dual and cross-attention encoders. ArXiv, abs/2004.03555. — [arXiv:2004.03555](https://arxiv.org/abs/2004.03555) find
  3. Marco Basaldella, Fangyu Liu, Ehsan Shareghi, and Nigel Collier. 2020. COMETA: A corpus for medical entity linking in the social media. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 3122–3137, Online. Association for Computational Linguistic… find
  4. Richards Hays. 2024. The relevance of medical history to current practice. Australian Journal of General Practice, 53(3):157–160. find
  5. Johannes Hoffart, Stephan Seufert, Dat Ba Nguyen, Martin Theobald, and Gerhard Weikum. 2012. KORE: keyphrase overlap relatedness for entity disambiguation. In 21st ACM International Conference on Information and Knowledge Management, CIKM’12, Maui, HI, USA, October 29 - November 02, 2012, pages 545–… find
  6. Hannah Bast, Matthias Hertel, and Natalie Prange. 2022. ELEVANT: A fully automatic fine-grained entity linking evaluation and analysis tool. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing: System Demonstrations, pages 72–79, Abu Dhabi, UAE. Association for… find
  7. Johannes Hoffart, Mohamed Amir Yosef, Ilaria Bordino, Hagen Fürstenau, Manfred Pinkal, Marc Spaniol, Bilyana Taneva, Stefan Thater, and Gerhard Weikum. 2011. Robust disambiguation of named entities in text. In Proceedings of the 2011 Conference on Empirical Methods in Natural Language Processing, pa… find
  8. O. Bodenreider. 2004. The Unified Medical Language System (UMLS): integrating biomedical terminology. Nucleic Acids Research, 32(Database issue):D267– D270. find
  9. Marta Boscariol, Luana Bulla, Lia Draetta, Beatrice Fi-umanò, Emanuele Lenzi, and Leonardo Piano. 2025. Evaluation of llms on long-tail entity linking in historical documents. Preprint, arXiv:2505.03473. — [arXiv:2505.03473](https://arxiv.org/abs/2505.03473) find
  10. Nikolaos Kolitsas, Octavian-Eugen Ganea, and Thomas Hofmann. 2018. End-to-end neural entity linking. In Proceedings of the 22nd Conference on Computational Natural Language Learning, pages 519–529, Brussels, Belgium. Association for Computational Linguistics. find
  11. Samuel Broscheit. 2019. Investigating entity knowledge in BERT with simple neural end-to-end entity linking. In Proceedings of the 23rd Conference on Computational Natural Language Learning (CoNLL), pages 677–685, Hong Kong, China. Association for Computational Linguistics.
  12. Xukai Liu, Ye Liu, Kai Zhang, Kehang Wang, Qi Liu, and Enhong Chen. 2024. OneNet: A fine-tuning free framework for few-shot entity linking via large language model prompting. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 13634–13651, Miami, Florida,… find
  13. Erin Connelly, Charo I. del Genio, and Freya Harrison. 2020. Data mining a medieval medical text reveals patterns in ingredient choice that reflect biological activity against infectious agents. mBio, 11(1):10.1128/mbio.03136–19. — [doi:10.1128/mbio.03136–19](https://doi.org/10.1128/mbio.03136–19) find
  14. Daniel Loureiro and Alípio Mário Jorge. 2020. Medlinker: Medical entity linking with neural representations and dictionary matching. In Advances in Information Retrieval, volume 12036 of Lecture Notes in Computer Science, pages 230–237, Cham. Springer International Publishing. find
  15. Sunil Mohan and Donghui Li. 2019. Medmentions: A large biomedical corpus annotated with umls concepts. ArXiv, abs/1902.09476. — [arXiv:1902.09476](https://arxiv.org/abs/1902.09476) find
  16. Parth M. Patel and Sukumar P. Desai. 2014. A clinician’s rationale for the study of history of medicine. Journal of Education in Perioperative Medicine, 16(4):E070. find
  17. Lynn M. Schriml, J. Allen Baron, Claudia Marie Sánchez-Beato Johnson, James B. Munro, Elvira, Sue Bello, Melody Swen, Anu, Becky Jackson, Andra Waag, Chris Mungall, and James A. Overton. 2025. Diseaseontology/humandiseaseontology: Do november 2025 release (v2025-11-25). OBO Foundry ontology, DOID. find
  18. Ozge Sevgili, Artem Shelmanov, Mikhail Arkhipov, Alexander Panchenko, and Christian Biemann. 2020. Neural entity linking: A survey of models based on deep learning. Semantic Web, 13:527–570. find
  19. Qwen Team. 2025. Qwen3 technical report. Preprint, arXiv:2505.09388.
  20. Paul Thompson, Riza Theresa Batista-Navarro, Georgios Kontonatsios, Jacob Carter, Elizabeth Toon, John McNaught, Carsten Timmermann, Michael Worboys, and Sophia Ananiadou. 2016. Text mining the history of medicine. PLOS ONE, 11(1):e0144717. find
  21. Haoran Wei, Yaofeng Sun, and Yukun Li. 2025. Deepseek-ocr: Contexts optical compression. ArXiv, abs/2510.18234.
  22. Hongyi Yuan, Zheng Yuan, Ruyi Gan, Jiaxing Zhang, Yutao Xie, and Sheng Yu. 2022. BioBART: Pretraining and evaluation of a biomedical generative language model. In Proceedings of the 21st Workshop on Biomedical Language Processing, pages 97–109, Dublin, Ireland. Association for Computational Linguist… find
  23. Shibingfeng Zhang and Giovanni Colavizza. 2025. Named entity recognition of historical text via large language model. Preprint, arXiv:2508.18090. — [arXiv:2508.18090](https://arxiv.org/abs/2508.18090) find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/nguyen-2026-enhanced-training-free-pipeline-entity/