Gemini and Claude Prompting Achieves Top Named Entity Recognition in Latin
August 6, 2026
Prompt engineering with Gemini-2.5-Pro and Claude-Sonnet-4.5 secured first place in the EvaLatin 2026 NER task for Classical Latin. The method utilizes cross-lingual transfer learning to handle both coarse-grained (11 classes) and fine-grained (28 classes) entity recognition in underrepresented ancient languages.
HOW THIS AFFECTS YOU
●
researcherThis demonstrates that high-performing NER is possible for low-resource languages through strategic prompting of frontier LLMs.