Danu, Manuela-Daniela, Rodríguez-Ortega, Miguel, Becerra-Tome, Alberto, Arancibia, Pablo, Gallego-Donoso, Fernando, Pratesi, Lorenzo, Bai, Xi, Dinescu, Andrei-Damian, Lima-López, Salvador, Guala, Andrea, Haitjema, Saskia, Mack, Jonas, Miranda, Michele, Nabki, Wesam Al, Rodríguez-Miret, Jan, Marica, Vasile George, Kincl, Vladimír, Dostálová, Kateřina, Andrej, Michael, Ujil, Alicia, Allaart, Corinne, Boonstra, Machteld, Overmars, Malin, Shah, Anoop, Ceroni, Giovanna, Kossoff, Jonathan, D'Amario, Domenico, Laborante, Renzo, Iaconelli, Antonio, Fumarulo, Isabella, Lenkowicz, Jacopo, Lilli, Livia, Weiss, Emma, Scafa, Alex, Lindberg, Felix, Gamage, Milou, Farré-Maduell, Eulàlia, Bratières, Sébastien, Krallinger, Martin, van Es, Bram · Nov 5, 2026
CardioLM: A multilingual suite of small language models for the cardiology domain
Markdown · JSON · Original source
The authors of CardioLM introduce a multilingual cardiology corpus and models for identifying medically relevant entities in text. Their CardioCCC corpus contains 508 cardiology case reports first published in Spanish and extended to six other languages through machine translation, annotation projection, and clinician review. The authors report that specialty-specific substantially improved pseudo-perplexity, a measure used in language modeling, but made little difference to strict span-level F1, a measure of exact-match identification. By contrast, jointly trained multilingual models achieved the best strict F1 in nearly every language and entity type, outperforming the strongest monolingual system by up to 0.13 F1. The authors also report that clinicians rated CardioNER above a rule- and dictionary-based baseline. Their central argument is that multilingual joint training, rather than specialty-specific pretraining alone, delivered the strongest entity-recognition results in their tests and offered a simpler deployment path for multi-country clinical information extraction. The results concern a corpus built from case reports; they do not establish how the models perform on other clinical documentation.
Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.
Original source
cs.AI · Article
Source details
Original source date: Nov 5, 2026 (2026-11-05).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-08T06:04:06.648Z. First observed: 2026-11-06T05:03:36.124Z. Retrieved: 2026-11-08T06:03:21.364Z.
Referenced by
Recorded citations and public file attachments. A reference alone does not establish agreement or influence.
No public references recorded here yet.
Limits
The corpus is based on cardiology case reports originally published in Spanish and extended into other languages through machine translation, annotation projection, and clinician review. The supplied abstract reports model results on the corpus but does not establish performance on clinical documentation beyond these case reports.
Open question
Would the reported multilingual-model advantage hold on clinical documentation beyond the case reports used to build and evaluate the corpus?
Source-stated
The authors introduce CardioCCC, a cardiology case-report corpus extended into six languages, and CardioLM, a suite of cardiology language models.
The first is CardioCCC, a corpus of 508 cardiology case reports originally published in Spanish and extended to Czech, Dutch, English, Italian, Romanian, and Swedish
Source-stated
The authors report that specialty-specific pretraining improved pseudo-perplexity substantially but had little effect on strict span-level F1.
Cardiology-specific pretraining reduced pseudo-perplexity by up to a factor of eight, yet changed strict span-level F1 by at most 0.04
Source-stated
The authors report that jointly trained multilingual models achieved the best strict F1 in nearly every language and entity type, outperforming the strongest monolingual system by up to 0.13 F1.
The jointly trained multilingual XLM-RoBERTa-large models achieved the best strict F1 in nearly every language and entity type, exceeding the strongest monolingual system by up to 0.13 F1.
Source-stated
The authors report higher clinician ratings for CardioNER than for a rule- and dictionary-based baseline.
Clinicians rated CardioNER well above a rule- and dictionary-based baseline, with the mean rating across entity types rising from 26% to 73%.
Escape Meridian interpretation
The reported corpus consists of case reports, so these results alone do not establish how the models perform on other clinical documentation.
a corpus of 508 cardiology case reports originally published in Spanish
Source and contributor context
Danu, Manuela-Daniela, Rodríguez-Ortega, Miguel, Becerra-Tome, Alberto, Arancibia, Pablo, Gallego-Donoso, Fernando, Pratesi, Lorenzo, Bai, Xi, Dinescu, Andrei-Damian, Lima-López, Salvador, Guala, Andrea, Haitjema, Saskia, Mack, Jonas, Miranda, Michele, Nabki, Wesam Al, Rodríguez-Miret, Jan, Marica, Vasile George, Kincl, Vladimír, Dostálová, Kateřina, Andrej, Michael, Ujil, Alicia, Allaart, Corinne, Boonstra, Machteld, Overmars, Malin, Shah, Anoop, Ceroni, Giovanna, Kossoff, Jonathan, D'Amario, Domenico, Laborante, Renzo, Iaconelli, Antonio, Fumarulo, Isabella, Lenkowicz, Jacopo, Lilli, Livia, Weiss, Emma, Scafa, Alex, Lindberg, Felix, Gamage, Milou, Farré-Maduell, Eulàlia, Bratières, Sébastien, Krallinger, Martin, van Es, Bram · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Pre-training
A model's initial training on a large dataset, before further adaptation for particular uses.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Editorial provenance
Version 1. Publication ID: af59c372-7671-4b27-8840-778494e44608.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-08T06:04:06.648Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 4a3626a747b3c99878f8d8a0240c40fdc1569963e0ecbbb527fd86e0bb4c0424. Source text hash: 0da061dfe54f5b558ea3433a4fb075d1938ee43c75b689773e65c855745c9305.
Revision and source-check history
Version 1 · 2026-11-08T06:04:06.648Z
CardioLM: A multilingual suite of small language models for the cardiology domain
The authors of CardioLM introduce a multilingual cardiology corpus and models for identifying medically relevant entities in text. Their CardioCCC corpus contains 508 cardiology case reports first published in Spanish and extended to six other languages through machine translation, annotation projection, and clinician review. The authors report that specialty-specific substantially improved pseudo-perplexity, a measure used in language modeling, but made little difference to strict span-level F1, a measure of exact-match identification. By contrast, jointly trained multilingual models achieved the best strict F1 in nearly every language and entity type, outperforming the strongest monolingual system by up to 0.13 F1. The authors also report that clinicians rated CardioNER above a rule- and dictionary-based baseline. Their central argument is that multilingual joint training, rather than specialty-specific pretraining alone, delivered the strongest entity-recognition results in their tests and offered a simpler deployment path for multi-country clinical information extraction. The results concern a corpus built from case reports; they do not establish how the models perform on other clinical documentation.
Original source
cs.AI · Article
Source details
Original source date: Nov 5, 2026 (2026-11-05).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-08T06:04:06.648Z. First observed: 2026-11-06T05:03:36.124Z. Retrieved: 2026-11-08T06:03:21.364Z.
Limits
The corpus is based on cardiology case reports originally published in Spanish and extended into other languages through machine translation, annotation projection, and clinician review. The supplied abstract reports model results on the corpus but does not establish performance on clinical documentation beyond these case reports.
Open question
Would the reported multilingual-model advantage hold on clinical documentation beyond the case reports used to build and evaluate the corpus?
Source-stated
The authors introduce CardioCCC, a cardiology case-report corpus extended into six languages, and CardioLM, a suite of cardiology language models.
The first is CardioCCC, a corpus of 508 cardiology case reports originally published in Spanish and extended to Czech, Dutch, English, Italian, Romanian, and Swedish
Source-stated
The authors report that specialty-specific pretraining improved pseudo-perplexity substantially but had little effect on strict span-level F1.
Cardiology-specific pretraining reduced pseudo-perplexity by up to a factor of eight, yet changed strict span-level F1 by at most 0.04
Source-stated
The authors report that jointly trained multilingual models achieved the best strict F1 in nearly every language and entity type, outperforming the strongest monolingual system by up to 0.13 F1.
The jointly trained multilingual XLM-RoBERTa-large models achieved the best strict F1 in nearly every language and entity type, exceeding the strongest monolingual system by up to 0.13 F1.
Source-stated
The authors report higher clinician ratings for CardioNER than for a rule- and dictionary-based baseline.
Clinicians rated CardioNER well above a rule- and dictionary-based baseline, with the mean rating across entity types rising from 26% to 73%.
Escape Meridian interpretation
The reported corpus consists of case reports, so these results alone do not establish how the models perform on other clinical documentation.
a corpus of 508 cardiology case reports originally published in Spanish
Source and contributor context
Danu, Manuela-Daniela, Rodríguez-Ortega, Miguel, Becerra-Tome, Alberto, Arancibia, Pablo, Gallego-Donoso, Fernando, Pratesi, Lorenzo, Bai, Xi, Dinescu, Andrei-Damian, Lima-López, Salvador, Guala, Andrea, Haitjema, Saskia, Mack, Jonas, Miranda, Michele, Nabki, Wesam Al, Rodríguez-Miret, Jan, Marica, Vasile George, Kincl, Vladimír, Dostálová, Kateřina, Andrej, Michael, Ujil, Alicia, Allaart, Corinne, Boonstra, Machteld, Overmars, Malin, Shah, Anoop, Ceroni, Giovanna, Kossoff, Jonathan, D'Amario, Domenico, Laborante, Renzo, Iaconelli, Antonio, Fumarulo, Isabella, Lenkowicz, Jacopo, Lilli, Livia, Weiss, Emma, Scafa, Alex, Lindberg, Felix, Gamage, Milou, Farré-Maduell, Eulàlia, Bratières, Sébastien, Krallinger, Martin, van Es, Bram · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Pre-training
A model's initial training on a large dataset, before further adaptation for particular uses.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Editorial provenance
Version 1. Publication ID: af59c372-7671-4b27-8840-778494e44608.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-08T06:04:06.648Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 4a3626a747b3c99878f8d8a0240c40fdc1569963e0ecbbb527fd86e0bb4c0424. Source text hash: 0da061dfe54f5b558ea3433a4fb075d1938ee43c75b689773e65c855745c9305.
uncertain · 2026-11-08T06:04:06.648Z · anthropic/claude-opus-5.5
Publication: af59c372-7671-4b27-8840-778494e44608
attribution — pass: The reading consistently attributes the corpus, models, results and conclusion to the paper's authors with phrasing such as 'The authors report'. It does not present the reported results as established fact or as Escape Meridian's view, and the case-report limitation is clearly separated as interpretation.
We present two contributions.
summary — uncertain: The reading tracks the abstract, but central terms are unexplained for a general reader. CardioNER appears only in the clinician-rating sentence and is never defined as the NER models fine-tuned on CardioCCC. The entity types (diseases, symptoms, procedures, medications) and 'pretraining' are also unexplained. The stakes are missing: clinical text is in local languages, but NLP resources are concentrated in English. Smallest repair: define CardioNER, list the entity types, add that motivation, and note that pretraining gains were concentrated in languages whose backbones lacked biomedical coverage. The 26%-to-73% rating could also be kept.
clinical natural language processing (NLP) resources remain concentrated in English
date — pass: The displayed original publication date of 2026-11-05 matches the arXiv v1 submission date. The feed date of 2026-11-06 is correctly not used.
[Submitted on 5 Nov 2026]
support — pass: Each source-stated claim matches its excerpt. 'Substantially improved pseudo-perplexity' fairly reflects a reduction of up to a factor of eight, and 'little difference' fairly reflects an F1 change of at most 0.04. The interpretive limitation and the question are conditional and grounded in the corpus being built from case reports. They do not overstate the evidence or invent benefits or costs.
changed strict span-level F1 by at most 0.04, with benefits concentrated in languages whose backbones lacked biomedical coverage
media — pass: The selected media is the verified direct arXiv abstract page for this paper. It is the most direct way for readers to access the source.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.
A model's initial training on a large dataset, before further adaptation for particular uses.
Definition referencearXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivA model's initial training on a large dataset, before further adaptation for particular uses.
Definition referencearXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXiv