Elamur Research⌕
☰
home / papers / 2026 / Parallia at #SMM4H-HeaRD 2026: ClinicalAligner26AM: A Cross-Lingual Aligner for Dataset Translation; Evidences from the MultiClinCorpus Shared Task

Parallia at #SMM4H-HeaRD 2026: ClinicalAligner26AM: A Cross-Lingual Aligner for Dataset Translation; Evidences from the MultiClinCorpus Shared Task

Paper · 2026 · François Remy

Summary

The paper addresses carrying entity annotations from Spanish clinical documents over to their post-edited translations. This is the MultiClinCorpus shared task at SMM4H-HeaRD 2026, with English, Italian, Dutch, Swedish, Romanian and Czech as targets. The author treats it as locating already-known mentions through word-level cross-lingual alignment, rather than detecting them from scratch.

ClinicalAligner26AM is initialized from ClinicalEncoder26AM, follows the general approach of AWESoME Align, and is trained on parallel clinical notes and conversations. Its training target comes from frozen teachers. Sentence-level (ClinicalEncoder26AM-L2), phrase-level (ClinicalEncoder26AM-L1) and token-level (XLM-Align) similarities are summed with a weak diagonal positional prior, then sharpened by Sinkhorn–Knopp optimal transport. The student learns to reproduce this matrix using plain cosine similarity between separately encoded texts.

At inference, source span scores are projected onto target tokens, and the longest well-supported span is decoded. The two submitted systems took the top two places, with character-level F1 mostly above 0.95. This suggests that an in-domain aligner can make building multilingual clinical corpora largely automatic.

Key points

  • Two systems were submitted: S2 = ClinicalAligner26AM and S3 = ClinicalAligner26AM-MCAI. The MCAI variant adds a positive teacher bias for token pairs inside manually mapped spans from the training split. They ranked 1st and 2nd across all languages and entity types (PROCEDURE, DISEASE, SYMPTOM).
  • Mean character-level F1 across the 6 languages:
    • MCAI: 0.96 / 0.97 / 0.96.
    • Base model: 0.95 / 0.96 / 0.95.
    • Strict F1 is about 0.80–0.85.
    • An NER-only CE26AM baseline scores clearly lower.
  • Training details:
    • The positional prior gives full weight within 50 tokens of the diagonal, decays up to 100 tokens, and forbids matches beyond that.
    • Sinkhorn–Knopp runs for 100 iterations.
    • The loss uses row- and column-wise cosine similarity instead of KL divergence.
  • Token-level teacher embeddings are smoothed with a 5-token context window, with neighbours discounted by a factor of 0.1. This reduces spurious matches on short tokens.
  • Data: about 1,200 training documents across 6+1 languages, 250 test documents, and about 30,000 background documents. At decoding time, a single gap token is allowed, and MultiClinNER predictions serve only as auxiliary support.

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 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 1Research subjects and languages the work is about.Multilingual & cross-lingual NLP · 828 works

Cite

Cite as François Remy. 2026. Parallia at #SMM4H-HeaRD 2026: ClinicalAligner26AM: A Cross-Lingual Aligner for Dataset Translation; Evidences from the MultiClinCorpus Shared Task. Proceedings of the 11th Social Media Mining for Health Research and Applications (SMM4H-HeaRD 2026) Workshop and Shared Tasks.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.smm4h-1.27

@inproceedings{remy2026parallia,
  title = {{Parallia at \#SMM4H-HeaRD 2026: ClinicalAligner26AM: A Cross-Lingual Aligner for Dataset Translation; Evidences from the MultiClinCorpus Shared Task}},
  author = {François Remy},
  booktitle = {Proceedings of the 11th Social Media Mining for Health Research and Applications (SMM4H-HeaRD 2026) Workshop and Shared Tasks},
  year = {2026},
  doi = {10.18653/v1/2026.smm4h-1.27},
  url = {https://doi.org/10.18653/v1/2026.smm4h-1.27}
}

References

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

  1. Abdulhalim Alhassan and 1 others. 2025. Discontinuous named entities in clinical text: A systematic literature review. JAMIA Open, 8(6):ooaf157. find
  2. Olivier Bodenreider. 2004. The unified medical language system (UMLS): integrating biomedical terminology. Nucleic Acids Research, 32(Database issue):D267–D270. find
  3. Jianlyu Chen, Shitao Xiao, Peitian Zhang, Kun Luo, Defu Lian, and Zheng Liu. 2024. M3-embedding: Multi-linguality, multi-functionality, multi-granularity text embeddings through self-knowledge distillation. In Findings of the Association for Computational Linguistics: ACL 2024, pages 2318–2335, Bang…
  4. Zewen Chi, Li Huang, Liyuan Liu, Xuan Bai, Heyan Huang, and Nigel Collier. 2021. Improving pre-trained cross-lingual language models via self-labeled word alignment. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference… find
  5. Alexis Conneau, Kartikay Khandelwal, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer, and Veselin Stoyanov. 2020. Unsupervised cross-lingual representation learning at scale. In Proceedings of the 58th Annual Meeting of the Association fo… find
  6. Kevin Donnelly. 2006. SNOMED-CT: The advanced terminology and coding system for ehealth. Studies in Health Technology and Informatics, 121:279–290. find
  7. Léo Jacqmin, Antoine Doucet, Emmanuel Morin, and Louis Martin. 2021. Spanalign: Projecting spans across languages by aligning subwords. In Proceedings of the Seventh Workshop on Noisy User-generated Text, pages 219–228, Online. Association for Computational Linguistics. find
  8. Salvador Lima López, Judith Rosell, Jan Rodríguez Miret, Fernando Gallego-Donoso, and Martin Krallinger. 2026. Multiclinai shared task training data. find
  9. Guillermo Lopez-Garcia, Jose Miguel Acitores Cortina, Jacob Berkowitz, Joey Chan, Ganesh Chandrasekar, Sumon Kanti Dey, Ivan Flores Amaro, Fernando Gallego, Lauren Gryboski, Ari Z Klein, Martin Krallinger, Salvador Lima-López, Tomohiro Nishiyama, Lisa Raithel, Ahmad Rezaie Mianroodi, Roland Roller,… find
  10. Sunil Mohan and Donghui Li. 2019. Medmentions: A large biomedical corpus annotated with umls concepts. In Proceedings of the 2019 Conference on Automated Knowledge Base Construction (AKBC 2019), Amherst, Massachusetts, USA. [Preprint]. find
  11. François Remy. 2026a. Diagnosable ColBERT: Debugging Late-Interaction Retrieval Models Using a Learned Latent Space as Reference. Preprint, arXiv:2604.19566.
  12. François Remy. 2026b. ClinicalEncoder26AM: A Multlilingual Diagnosable ColBERT Model; Evidences from the MultiClinNER Shared Task. Preprint, arXiv:2605.28521. — [arXiv:2605.28521](https://arxiv.org/abs/2605.28521) find
  13. Zi-Yi Dou and Graham Neubig. 2021. Word alignment by fine-tuning embeddings on parallel corpora. In Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume, pages 2112–2128, Online. Association for Computational Linguistics. find
  14. David Fraile Navarro, Pablo A. Gutiérrez, and 1 others. 2023. Clinical named entity recognition and relation extraction using natural language processing of medical free text: A systematic review. International Journal of Medical Informatics, 178:105182. find
  15. Fernando Gallego-Donoso, Salvador Lima-López, Judith Rosell, Eulàlia Farré-Maduell, and Martin Krallinger. 2026. The MultiClinAI Shared Task on Multilingual Clinical Corpus Construction and Concept Extraction: Systems, Evaluation, and Datasets. In Proceedings of the 11th Social Media Mining for Heal… find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/remy-2026-parallia-smm4h-heard-2026-clinicalaligner26am/