Proposing a training-free self-improving framework for zero-shot NER with LLMs significantly improves performance.
Providing definition and guidelines in the prompt can improve the performance and robustness of instruction-tuned language models for zero-shot named entity recognition, especially on unseen entity types.
SLIMER-IT, an instruction-tuning approach for zero-shot named entity recognition, leverages prompts enriched with definitions and guidelines to outperform state-of-the-art models on unseen entity types in Italian.