Elamur Research⌕
☰
home / papers / 2026 / PlainMedScale: A Corpus of Multi-Level Simplified Medical Texts in German and English

PlainMedScale: A Corpus of Multi-Level Simplified Medical Texts in German and English

Paper · 2026 · Bruno Brocai, Ilaria Papagno and Mayumi Ohta

Summary

Most corpora for medical text simplification pair an expert text with a single lay version. This binary setup hides the fact that simplification spans a range of difficulty levels. PlainMedScale addresses this with a document-aligned corpus in German and English that has four difficulty tiers per language:

  • MSD Manuals, professional edition
  • MSD Manuals, consumer edition
  • Gesund.Bund
  • a certified plain-language tier: Apotheken Umschau (Einfache Sprache) for German, NHS condition pages for English

Articles are linked through a pipeline with three stages: lexical matching against medical vocabularies, embedding re-ranking, and an LLM judge validated on a manually labeled gold set. A qualitative reading ties each tier to one communicative role: reference, explanation, decision support, or access.

Two pilot studies use the alignments. First, many readability metrics that separate two registers fail to change monotonically across all four tiers. Second, when an open-weight LLM is asked to write Plain Language, its output gets close to the intended level overall, yet traces of the source text's difficulty remain.

The resource lets researchers stress-test readability measures and evaluate simplification against several targets.

Key points

  • Corpus size: both languages have 1594 MSD Prof. and 1594 MSD Consumer articles, plus 354 Gesund.Bund articles. The plain-language tier has 180 Apotheken Umschau articles (German) and 500 NHS articles (English).
  • Alignment pipeline:
    • Stage 1 does lexical lookup of article titles in ICD-10, MeSH, German MeSH, the Human Disease Ontology (DO) and Pschyrembel. Entries are lemmatized with spaCy and stored in SQLite, and identifiers are propagated through DO cross-references.
    • Stage 2 re-ranks candidates with Qwen3-Embedding-8B. It also runs retrieval for unmatched articles, accepting a match only if a reverse title lookup resolves back to the same article.
    • Stage 3 filters the links with an LLM judge.
  • LLM judge: two annotators with medical backgrounds labeled 956 pairs as correct, specialization, generalization or incorrect. GPT-5.4-mini (about 80% overall accuracy) was chosen over GPT-5.4, Claude Haiku 4.5 and Qwen3-30B-Instruct. A re-check of 100 pairs gave 92% precision on "correct" (Wilson CI 67–99) and at least 88% on the other classes.
  • Readability in German: only five measures shift monotonically across all tiers: Wiener Sachtextformel, adjective ratio, negation ratio, term frequency and jargon density. For example, Wiener Sachtextformel rises from 6.80 (Apotheken Umschau) to 12.23 (MSD Prof.), and jargon density from 0.043 to 0.148.
  • Simplification experiment: Qwen3-30B-A3B-Instruct-2507 outputs keep the source-difficulty ordering on every testable monotonic metric except the negation ratio. This residue is weak on average, and the outputs mostly land near the Apotheken Umschau level.

Interesting ideas

  • I1. A difficulty measure should be validated as moving in the same direction across several ordered levels, not just as separating two endpoints, because a gap between two corpora may not hold elsewhere on the scale.
  • I2. A "bleed-through" probe asks whether a property of the input still predicts that property in the output after a transformation that should erase it; an ideal system would leave no correlation.
  • I3. Linking two documents through shared concept IDs in external ontologies is made safer by a round-trip check: an embedding-retrieved match counts only if looking up the concept's title leads back to the same item.
  • I4. The judge labels a pair not just as match or non-match but also as narrower or broader than the anchor, since many links are partial topic overlaps rather than errors.
  • I5. Precision is bracketed between two estimates with known opposite biases: a reviewer who sees the LLM's label likely overestimates it, while noisy original labels likely underestimate it.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

  • I1 (learn): Mova's evaluation could require that a confidence or complexity score rise monotonically across graded text sets, used as a sanity check alongside negative controls.
  • I2 (back): when Mova regenerates text from the meaning graph, the same test could check whether surface traits of the source (sentence length, jargon) leak into the output instead of being fully re-decided.
  • I3 (world): entity linking to WordNet/Wikidata could keep only links that survive a reverse lookup, cutting noisy grounding in the knowledge graph.
  • I4 (world): ontology linking could record hyponym/hypernym relations as separate outcomes, so "paralysis" vs "vocal fold paralysis" becomes a typed edge instead of a false match.
  • I5 (learn): when auditing LLM-teacher silver labels, reporting both an anchored review and a blind one gives an honest range instead of one inflated number.

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

🎓Heidelberg UniversityHeidelberg · Germany38 works · 41,356 citations

Also on the byline: Heidelberg University · Center for Applied Linguistics.

 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 worksHealth Professions · 69 worksGeneral Health Professions · 14 works

🧭 Topics 1Research subjects and languages the work is about.English · 2,263 works

Cite

Cite as Bruno Brocai, Ilaria Papagno and Mayumi Ohta. 2026. PlainMedScale: A Corpus of Multi-Level Simplified Medical Texts in German and English. arXiv:2608.01158.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2608.01158 · arXiv 2608.01158

@misc{brocai2026plainmedscale,
  title = {{PlainMedScale: A Corpus of Multi-Level Simplified Medical Texts in German and English}},
  author = {Bruno Brocai, Ilaria Papagno and Mayumi Ohta},
  year = {2026},
  doi = {10.48550/arxiv.2608.01158},
  eprint = {2608.01158},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2608.01158}
}

References

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

  1. Apotheken Umschau. 2020. Apotheken Umschau in Einfacher Sprache. https://www.apotheken-umschau.de/einfache-sprache/. ↗
  2. Dennis Aumiller and Michael Gertz. 2022. Klexikon: A German Dataset for Joint Summarization and Simplification. In Proceedings of the Thirteenth Language Resources and Evaluation Conference, pages 2693–2701, Marseille, France. European Language Resources Association. find
  3. Richard Bamberger and Erich Vanecek. 1984. Lesen– Verstehen–Lernen–Schreiben. Jugend und Volk, Wien. find
  4. J. Allen Baron, Claudia M. Sánchez-Beato Johnson, Michael A. Schor, Dustin Olley, Lance Nickel, Victor Felix, Susan M. Bello, Carol Greene, Richard Lichenstein, Katharine Bisordi, Rima Koka, Cynthia Bearer, Melissa Clarkson, Todd Detwiler, Lynn M. Schriml, and 1 others. 2026. The DO-KB knowledgebase 2026 update: Expanding programmatic and language access. Nucleic Acids Research, 54(D1):D1425– D1436. find
  5. Chandrayee Basu, Rosni Vasu, Michihiro Yasunaga, and Qian Yang. 2023. Med-easi: Finely annotated dataset and models for controllable simplification of medical texts. Preprint, arXiv:2302.09155. ↗
  6. Alessia Battisti, Dominik Pfütze, Andreas Säuberli, Marek Kostrzewa, and Sarah Ebling. 2020. A corpus for automatic readability assessment and text simplification of German. In Proceedings of the Twelfth Language Resources and Evaluation Conference, pages 3302–3311, Marseille, France. European Language Resources Association. ↗
  7. Bettina M. Bock. 2015. Barrierefreie Kommunikation als Voraussetzung und Mittel für die Partizipation benachteiligter Gruppen. Ein (polito-)linguistischer Blick auf Probleme und Potenziale von «Leichter» und «einfacher Sprache». Linguistik Online, 73(4):115–137. Florian Borchert, Christina Lohr, Luise Modersohn, Jonas Witt, Thomas Langer, Markus Follmann, Matthias Gietzelt, Bert Arnrich, Udo Hahn, and Matthieu-P. find
  8. Schapranow. 2022. GGPONC 2.0 - the German clinical guideline corpus for oncology: Curation workflow, annotation policy, baseline NER taggers. In Proceedings of the Thirteenth Language Resources and Evaluation Conference, pages 3650– 3660, Marseille, France. European Language Resources Association. find
  9. Bundesinstitut für Arzneimittel und Medizinprodukte. 2024. ICD-10-GM Version 2025. Internationale statistische Klassifikation der Krankheiten und verwandter Gesundheitsprobleme, 10. Revision, German Modification. https://www.bfarm.de/ DE/Kodiersysteme/Klassifikationen/ICD/ ICD-10-GM/_node.html. Bundesinstitut für Arzneimittel und Medizinprodukte. n.d. DRKS - Deutsches Register Klinischer Studien. https://www.drks.de/search/de. ↗
  10. Council of Europe. 2023. Guide to health literacy. contributing to trust building and equitable access to healthcare. Steering Committee for Human Rights in the fields of Biomedicine and Health (CDBIO). https://rm.coe.int/ inf-2022-17-guide-health-literacy/ 1680a9cb75. ↗
  11. Ashwin Devaraj, Iain Marshall, Byron Wallace, and Junyi Jessy Li. 2021. Paragraph-level simplification of medical texts. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 4972–4984, Online. Association for Computational Linguistics.
  12. Brigitte Fischer-Wagener, Dietrich Rebholz-Schuhmann, Gabriele Wollnik-Korn, Miriam Albers, Roman Baum, Leyla Jael Castro, and Lukas Geist. 2023. Deutscher MeSH. https://repository.publisso.de/resource/frl%3A6440319. ↗
  13. Gesund.Bund. 2025. Gesund.Bund. https://gesund.bund.de/. ↗
  14. Dirk Goldhahn, Thomas Eckart, and Uwe Quasthoff. 2012. Building Large Monolingual Dictionaries at the Leipzig Corpora Collection: From 100 to 200 Languages. In LREC, volume 29, pages 34–43.
  15. Matthew Honnibal, Ines Montani, Sofie Van Lan-deghem, and Adriane Boyd. 2020. spaCy: Industrial-strength Natural Language Processing in Python. find
  16. Koki Horiguchi, Tomoyuki Kajiwara, Takashi Ninomiya, Shoko Wakamiya, and Eiji Aramaki. 2025. Multimsd: A corpus for multilingual medical text simplification from online medical references. In Findings of the Association for Computational Linguistics: ACL 2025, pages 9248–9258, Vienna, Austria. Association for Computational Linguistics. find
  17. Inclusion Europe. n.d. Informationen für alle. Europäische Regeln, wie man Informationen leicht lesbar und leicht verständlich macht. Brussels. https://easy-to-read.inclusion-europe.eu/wp-content/uploads/ 2014/12/DE_Information_for_all.pdf. ↗
  18. Sarah Jablotschkin, Elke Teich, and Heike Zinsmeister. 2024. DE-Lite - a New Corpus of Easy German: Compilation, Exploration, Analysis. In Proceedings of the Fourth Workshop on Language Technology for Equality, Diversity, Inclusion, pages 106–117, St. Julian’s, Malta. Association for Computational Linguistics. find
  19. Roman Jakobson. 1960. Closing statement: Linguistics and poetics. In Thomas A. Sebeok, editor, Style in Language, pages 350–377. MIT Press, Cambridge, MA. find
  20. Sebastian Joseph, Kathryn Kazanas, Keziah Reina, Vishnesh Ramanathan, Wei Xu, Byron Wallace, and Junyi Jessy Li. 2023. Multilingual simplification of medical texts. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 16662–16692, Singapore. Association for Computational Linguistics. find
  21. David Klaper, Sarah Ebling, and Martin Volk. 2013. Building a German/simple German parallel corpus for automatic text simplification. In Proceedings of the Second Workshop on Predicting and Improving Text Readability for Target Reader Populations, pages 11–19, Sofia, Bulgaria. Association for Computational Linguistics. find
  22. Philippe Laban, Tobias Schnabel, Paul N. Bennett, and Marti A. Hearst. 2021. Keep it simple: Unsupervised simplification of multi-paragraph text. In Association for Computational Linguistics, pages 6365–6378. ↗
  23. Chen Lyu and Gabriele Pergola. 2024. Society of medical simplifiers. In Proceedings of the Third Workshop on Text Simplification, Accessibility and Readability (TSAR 2024), pages 61–68, Miami, Florida, USA. Association for Computational Linguistics. ↗
  24. Christiane Maaß. 2020. Easy Language – Plain Language – Easy Language Plus. Balancing Comprehensibility and Acceptability. Frank & Timme, Berlin. find
  25. MSD Manuals. 2025. MSD Manuals. https://www.msdmanuals.com/de/. National Library of Medicine. n.d. PubMed. https://pubmed.ncbi.nlm.nih.gov/. ↗
  26. NHS. 2025. NHS: Health A to Z. https://www.nhs.uk/health-a-to-z/. ↗
  27. Thomas Proisl and Peter Uhrig. 2016. SoMaJo: State-of-the-art tokenization for German web and social media texts. In Proceedings of the 10th Web as Corpus Workshop (WAC-X) and the EmpiriST Shared Task, pages 57–62, Berlin. Association for Computational Linguistics. find
  28. Pschyrembel. 2025. Pschyrembel Online. https://www.pschyrembel.de/. ↗
  29. Qwen Team. 2025. Qwen3 technical report. Preprint, arXiv:2505.09388. ↗
  30. Karen Scholz and Markus Wenzel. 2025. Evaluating Readability Metrics for German Medical Text Simplification. In Proceedings of the 31st International Conference on Computational Linguistics, pages 6049–6062, Abu Dhabi, UAE. Association for Computational Linguistics. find
  31. Kristine Sørensen, Stephan Van den Broucke, James Fullam, Gerardine Doyle, Jürgen Pelikan, Zofia Slonska, Helmut Brand, and (HLS-EU) Consortium Health Literacy Project European. 2012. Health literacy and public health: A systematic review and integration of definitions and models. BMC Public Health, 12(1):80. find
  32. Regina Stodden, Omar Momen, and Laura Kallmeyer. 2023. DEplain: A German parallel corpus with intralingual translations into plain language for sentence and document simplification. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 16441–16463, Toronto, Canada. Association for Computational Linguistics. find
  33. Vanessa Toborek, Moritz Busch, Malte Boßert, Christian Bauckhage, and Pascal Welke. 2023. A new aligned simple German corpus. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 11393–11412, Toronto, Canada. Association for Computational Linguistics. ↗
  34. U.S. National Library of Medicine. 2025. Medical subject headings (MeSH). https://www.nlm.nih.gov/mesh/. ↗
  35. William Xia, Ishita Unde, Brian Ondov, and Dina Demner-Fushman. 2025. Jebs: A fine-grained biomedical lexical simplification task. Preprint, arXiv:2506.12898. ↗
  36. Yanzhao Zhang, Mingxin Li, Dingkun Long, Xin Zhang, Huan Lin, Baosong Yang, Pengjun Xie, An Yang, Dayiheng Liu, Junyang Lin, Fei Huang, and Jingren Zhou. 2025. Qwen3 embedding: Advancing text embedding and reranking through foundation models. arXiv preprint arXiv:2506.05176. ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/brocai-2026-plainmedscale-corpus-multi-level-simplified/