---
license: apache-2.0
task_categories:
- text2text-generation
language:
- zh
size_categories:
- 100M<n<1B
tags:
- chinese-spelling-check
- 中文
---

# Dataset for CSC

中文纠错数据集

# Dataset Description
Chinese Spelling Correction (CSC) is a task to detect and correct misspelled characters in Chinese texts.

共计 120w 条数据，以下是数据来源

|数据集|语料|链接|
|------|------|------|
|SIGHAN+Wang271K 拼写纠错数据集|SIGHAN+Wang271K(27万条)|https://huggingface.co/datasets/shibing624/CSC|
|ECSpell 拼写纠错数据集|包含法律、医疗、金融等领域|https://github.com/Aopolin-Lv/ECSpell|
|CGED 语法纠错数据集|仅包含了2016和2021年的数据集|https://github.com/wdimmy/Automatic-Corpus-Generation?spm=a2c22.12282016.0.0.5f3e7398w7SL4P|
|NLPCC 纠错数据集|包含语法纠错和拼写纠错|https://github.com/Arvid-pku/NLPCC2023_Shared_Task8  <br/>http://tcci.ccf.org.cn/conference/2023/dldoc/nacgec_training.zip<br/>http://tcci.ccf.org.cn/conference/2018/dldoc/trainingdata02.tar.gz|
|pycorrector 语法纠错集|中文语法纠错数据集|https://github.com/shibing624/pycorrector/tree/llm/examples/data/grammar|

其余的数据集还可以看

- 中文文本纠错数据集汇总 (天池)：https://tianchi.aliyun.com/dataset/138195
- NLPCC 2023中文语法纠错数据集：http://tcci.ccf.org.cn/conference/2023/taskdata.php


# Languages
The data in CSC are in Chinese.

# Dataset Structure
An example of "train" looks as follows:
```json
{
  "conversations": [
    {"from":"human","value":"对这个句子纠错\n\n以后，我一直以来自学汉语了。"},
    {"from":"gpt","value":"从此以后，我就一直自学汉语了。"}
  ]
}

```

# Contributions

[Weaxs](https://github.com/Weaxs) 整理并上传