Large-scale semantic indexing of Spanish biomedical literature using contrastive transfer learning.

BERTDeCS, a contrastive transfer-learning framework, achieves top performance in the MESINESP and MESINESP2 Spanish medical semantic indexing tasks within the BioASQ challenge.

• Why it matters: Biomedical literature growth makes automatic indexing essential, but non-English indexing lags due to insufficient high-quality Spanish training data.
• What they did: BERTDeCS adapts multilingual BERT (M-BERT) to Spanish biomedical text via contrastive learning and leverages enriched English MEDLINE annotations through transfer learning.
• The result: On Spanish datasets it outperforms state-of-the-art indexing methods, and preliminary Portuguese experiments suggest transfer to a closely related language.