Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

In-Context Learning for Few-Shot Nested Named Entity Recognition

Meishan Zhang, Bin Wang, Hao Fei, Min Zhang

TL;DR

The paper tackles the problem of few-shot nested NER by marrying in-context learning with a novel Entity Demonstration (EnDe) Retriever. EnDe uses contrastive representations across semantic, boundary, and label dimensions, augmented by POS tags and constituency structures, to select high-quality demonstrations for a four-part ICL prompt built on a frozen T5-base backbone. Across three nested and four flat NER benchmarks, and under $k$-shot settings, the approach achieves state-of-the-art results, with pronounced gains in low-shot regimes and when employing larger LMs such as GPT-3.5. The work demonstrates the effectiveness of targeted demonstration selection and boundary-aware prompting for complex NER tasks, offering practical benefits for data-scarce scenarios and guiding prompt design for few-shot NLP.

Abstract

In nested Named entity recognition (NER), entities are nested with each other, and thus requiring more data annotations to address. This leads to the development of few-shot nested NER, where the prevalence of pretrained language models with in-context learning (ICL) offers promising solutions. In this work, we introduce an effective and innovative ICL framework for the setting of few-shot nested NER. We improve the ICL prompt by devising a novel example demonstration selection mechanism, EnDe retriever. In EnDe retriever, we employ contrastive learning to perform three types of representation learning, in terms of semantic similarity, boundary similarity, and label similarity, to generate high-quality demonstration examples. Extensive experiments over three nested NER and four flat NER datasets demonstrate the efficacy of our system.

In-Context Learning for Few-Shot Nested Named Entity Recognition

TL;DR

The paper tackles the problem of few-shot nested NER by marrying in-context learning with a novel Entity Demonstration (EnDe) Retriever. EnDe uses contrastive representations across semantic, boundary, and label dimensions, augmented by POS tags and constituency structures, to select high-quality demonstrations for a four-part ICL prompt built on a frozen T5-base backbone. Across three nested and four flat NER benchmarks, and under -shot settings, the approach achieves state-of-the-art results, with pronounced gains in low-shot regimes and when employing larger LMs such as GPT-3.5. The work demonstrates the effectiveness of targeted demonstration selection and boundary-aware prompting for complex NER tasks, offering practical benefits for data-scarce scenarios and guiding prompt design for few-shot NLP.

Abstract

In nested Named entity recognition (NER), entities are nested with each other, and thus requiring more data annotations to address. This leads to the development of few-shot nested NER, where the prevalence of pretrained language models with in-context learning (ICL) offers promising solutions. In this work, we introduce an effective and innovative ICL framework for the setting of few-shot nested NER. We improve the ICL prompt by devising a novel example demonstration selection mechanism, EnDe retriever. In EnDe retriever, we employ contrastive learning to perform three types of representation learning, in terms of semantic similarity, boundary similarity, and label similarity, to generate high-quality demonstration examples. Extensive experiments over three nested NER and four flat NER datasets demonstrate the efficacy of our system.
Paper Structure (16 sections, 3 equations, 5 figures, 2 tables)

This paper contains 16 sections, 3 equations, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Task Definition
  4. Generative NER with In-context Learning
  5. Entity Demonstration (EnDe) Retriever
  6. Semantic Similarity Measurement
  7. Boundary Similarity Measurement
  8. Label Identication Measurement
  9. Experiments
  10. Datasets and Settings
  11. Results on Few-shot Nested NER
  12. Results on Few-shot Flat NER
  13. Effects of Shot Size
  14. Impacts of Using different LMs
  15. Conclusion
  16. ...and 1 more sections

Figures (5)

  • Figure 1: Illustration of nested NER.
  • Figure 2: Illustration of the prompts with in-context learning.
  • Figure 3: The framework of EnDe Retriever.
  • Figure 4: Results with k-shot samples on two datasets.
  • Figure 5: Performances of our system by employing LMs in different types and sizes.