/ConfusionCluster

An Analysis Tool to Models for Chinese Spell Checking Released on ACL2023.

Primary LanguagePython

ConfusionCluster

分析及CCCR Probe 指标对于ACL 2023 (Findings): Investigating Glyph Phonetic Information for Chinese Spell Checking: What Works and What's Next

训练纠错模型的代码及训练数据见另一github仓库:SpecialEdition 论文中的Probe实验一见github仓库:

Environment setting

python >= 3.7

conda create -n ctcSE python=3.7
conda activate ctcSE

pip install -r requirements.txt

Quick Start

将main.py中路径设置为你要测试的纠错模型的路径。

bash run.sh

⚠️注意最好该模型是huggingface对象输出能够返回logits用于分析,否则请参考本项目中对ReaLiSe的处理,提前运行一次inference将模型logits保存到文件再读入。