Preprint reports multilingual clinical annotation gains from constrained LLM workflow

A preprint reports a constrained LLM workflow for transferring Spanish clinical annotations into six languages while preserving target text and reconstructing character offsets. The study says direct LLM projection produced the strongest results among the tested approaches: GLM 5.2 reached a mean strict F1 of 0.9201 across 18 language-entity combinations, while locally deployable Gemma4:31B reached 0.9133. The authors report 55,416 grounded mentions with reconstructed offsets.