Prompts in Few-Shot Named Entity Recognition

Abstract

The article explores various methods for generating questions (prompts) in question-answer approaches to named entity recognition. In such methods, recognition is understood as answering questions about named entities. The system must return valid named entities of the required type that are referenced in the sentence. In particular, we study the effect of prompts on the nested named entity recognition in few-shot setting. We test different prompt-based approaches against data from the RuNNE-2022 nested named entity extraction competition.