Decomposed Meta-Learning for Few-Shot Named Entity Recognition

TL;DR

This paper introduces a decomposed meta-learning framework for few-shot NER, addressing span detection and entity typing separately with MAML-based techniques, leading to improved performance on benchmarks.

Contribution

It proposes a novel decomposed meta-learning approach combining MAML for span detection and MAML-ProtoNet for entity typing in few-shot NER tasks.

Findings

Achieves superior performance over prior methods on benchmark datasets.

Effectively adapts to new entity classes with limited labeled data.

Demonstrates the effectiveness of decomposed meta-learning in NER.

Abstract

Few-shot named entity recognition (NER) systems aim at recognizing novel-class named entities based on only a few labeled examples. In this paper, we present a decomposed meta-learning approach which addresses the problem of few-shot NER by sequentially tackling few-shot span detection and few-shot entity typing using meta-learning. In particular, we take the few-shot span detection as a sequence labeling problem and train the span detector by introducing the model-agnostic meta-learning (MAML) algorithm to find a good model parameter initialization that could fast adapt to new entity classes. For few-shot entity typing, we propose MAML-ProtoNet, i.e., MAML-enhanced prototypical networks to find a good embedding space that can better distinguish text span representations from different entity classes. Extensive experiments on various benchmarks show that our approach achieves superior performance over prior methods.