ParaBART: A Prompt-based Method with Parabiotic Decoder for Few-shot Named Entity Recognition

Anonymous

Keywords: few-shot learning, prompt learning, named entity recognition

Abstract: Prompt-based methods have been widely used in few-shot named entity recognition (NER). We first conduct a preliminary experiment and observe that what really affects prompt-based NER models is the ability to detect entity boundaries. However, previous prompt-based NER models neglect to enhance the ability of entity boundary detection. To solve the issue, we propose a novel method, ParaBART, which consists of a BART encoder and the Parabiotic Decoder we design. Parabiotic Decoder includes two BART decoders and a conjoint module. The two decoders are responsible for entity boundary detection and entity type classification respectively and share the well-learned knowledge through the conjoint module, which replaces unimportant tokens’ embeddings in one decoder with the average embedding of all tokens in the other decoder. Moreover, we propose a novel boundary expansion strategy to enhance the ability of entity type classification. Experimental results show that ParaBART can achieve significant performance gains over previous state-of-the-art methods. For reproducibility, all datasets and codes are provided in the supplementary materials.

Paper Type: long

Research Area: Information Extraction