MeSHClass-ES and AnatEM-ES: Open Resources for Spanish Biomedical NLP

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Science & Research — Life Sciences & Biology, Artificial Intelligence & Machine Learning, Mathematics & Computational Sciences · Depth: Advanced, medium

Summary

Santiago Martinez Novoa and colleagues introduce MeSHClass-ES and AnatEM-ES, two new open Spanish biomedical NLP corpora. MeSHClass-ES is a 29,063 abstract bilingual corpus translated from PubMed using Opus-MT. AnatEM-ES is an anatomical entity corpus translated with a chunk-level LLM-based pipeline, preserving BIO annotations across 13,849 entity mentions. Both corpora achieved a mean COMET score of 0.73. Benchmarking nine encoder models, RigoBERTa-2.0 led both tasks with a micro-F1 of 0.69 for classification (tied with SciBETO-large) and an F1 of 0.66 for Named Entity Recognition (NER). Domain pretraining and model capacity significantly improved performance, with a 4-point spread for NER and a 3-point spread for classification. XLM-RoBERTa-large proved a competitive multilingual baseline. An evaluation of four open-weight decoders (7-9B) showed QLoRA-adapted Gemma-2-9B reached 88% of the best encoder's classification performance (micro-F1 0.61 vs 0.69), but decoders performed below 40% of the best encoder F1 for NER. The corpora, translation pipelines, and evaluation code are publicly available on HuggingFace Hub and GitHub.

Key takeaway

For NLP Engineers developing Spanish biomedical applications, these new corpora offer crucial resources. You should prioritize encoder models like RigoBERTa-2.0 for tasks such as Named Entity Recognition, as decoders currently show significantly lower performance (below 40% F1). Leverage the released MeSHClass-ES and AnatEM-ES datasets on HuggingFace Hub to enhance your model training and evaluation, especially when domain-specific pretraining is critical for performance gains.

Key insights

Spanish biomedical NLP resources, MeSHClass-ES and AnatEM-ES, bridge a critical language gap, with encoders outperforming decoders significantly for NER.

Principles

• Domain pretraining boosts biomedical NLP.
• Model capacity improves task performance.
• Encoders excel over decoders for NER.

Method

Corpora were created by translating PubMed abstracts (Opus-MT) and AnatEM entities (chunk-level LLM pipeline) while preserving BIO annotations. Benchmarking involved nine encoders and four decoders.

In practice

• Use RigoBERTa-2.0 for Spanish biomedical tasks.
• Consider XLM-RoBERTa-large as a baseline.
• Access corpora on HuggingFace Hub.

Topics

• Spanish NLP
• Biomedical Corpora
• Named Entity Recognition
• Text Classification
• Encoder Models
• Decoder Models
• HuggingFace Hub

Best for: Research Scientist, NLP Engineer, AI Scientist, Machine Learning Engineer

Related on AIssential

• Analysing Lightweight Large Language Models for Biomedical Named Entity Recognition on Diverse Ouput Formats — Lightweight LLMs offer competitive biomedical NER performance, with specific output formats proving more effective than multi-format tuning.
• A Dataset of Brazilian Portuguese Clinical Notes for Anaphylaxis Detection — A new Brazilian Portuguese clinical note dataset enables NLP-driven anaphylaxis detection for research and health record analysis.
• Domain Fine-Tuning FinBERT on Finnish Histopathological Reports: Train-Time Signals and Downstream Correlations — Domain fine-tuning FinBERT on medical text can be predicted by observing embedding geometry changes.
• Combining Real and Synthetic Speech for ASR Adaptation in Brazilian Portuguese — Domain-guided synthetic speech effectively augments real data for ASR adaptation in specialized, low-resource scenarios.
• MMed-Bench-IR: A Heterogeneous Benchmark for Multilingual Medical Information Retrieval — MMed-Bench-IR reveals severe cross-lingual performance gaps in medical information retrieval, highlighting the need for benchmarks that assess multilingual and biomedical expertise jointly.
• EDEN: A Large-Scale Corpus of Clinical Notes for Italian — EDEN provides the largest Italian clinical note corpus and a CRF-filling benchmark for LLM-based information extraction.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.