Learning from Language Description: Low-shot Named Entity Recognition via Decomposed Framework

This is the implementation of the paper Learning from Language Description: Low-shot Named Entity Recognition via Decomposed Framework.

Overview

In this work we present SpanNER, which learns from natural language supervision to build Few-shot NER learner. At the same time, such a framework enables the identification of never-seen entity classes without using in-domain labeled data You can find more details of this work in our paper.

TO-DO

Release distantly-supervised checkpoint

Setup Environment

Install via pip:

1. create a conda environment running Python 3.7:

conda create -n SpanNER python=3.7
conda activate SpanNER

2. install the required dependencies:

pip install -r requirements.txt

Quick start

Run SpanNER

Training on CoNLL03
> bash ./scripts/run.sh

Notes and Acknowledgments

The implementation is based on https://github.com/huggingface/transformers
We also used some code from: https://github.com/facebookresearch/BLINK

How do I cite SpanNER?

@inproceedings{wang2021learning,
  title={Learning from Language Description: Low-shot Named Entity Recognition via Decomposed Framework},
  author={Wang, Yaqing and Chu, Haoda and Zhang, Chao and Gao, Jing},
  booktitle={Findings of the Association for Computational Linguistics: EMNLP 2021},
  pages={1618--1630},
  year={2021}
}