Textual Entailment with Dynamic Contrastive Learning for Zero-shot NER

Anonymous

Abstract: In this paper, we study the problem of zero-shot NER, which aims at building a Named Entity Recognition (NER) system from scratch. It needs to identify the entities in the given sentences when we have zero token-level annotations for training. Previous works usually use sequential labeling models to solve the NER task and obtain weakly labeled data from entity dictionaries in the zero-shot setting. However, these labeled data are quite noisy since we need the labels for each token and the entity coverage of the dictionaries is limited. Here we propose to formulate the NER task as a Textual Entailment problem and solve the task via Textual Entailment with Dynamic Contrastive Learning (TEDC). TEDC not only alleviates the noisy labeling issue, but also transfers the knowledge from pre-trained textual entailment models. Additionally, the dynamic contrastive learning framework contrasts the entities and non-entities in the same sentence and improves the model's discrimination ability. Experiments on two datasets show that TEDC can achieve state-of-the-art performance on the task of zero-shot NER.