---
license: apache-2.0
language:
- zh
pretty_name: Chinese-Braille
size_categories:
- 100K<n<1M
tags:
- chinese
- braille
---
# Chinese Braille Dataset
<p align="center">
  📃 <a href="https://arxiv.org/abs/2407.06048/" target="_blank">[Paper]</a> • 💻 <a href="https://github.com/AlanYWu/ChineseBrailleTranslation" target="_blank">[Github]</a> • 🤗 <a href="https://huggingface.co/datasets/Violet-yo/Chinese-Braille-Dataset-10per-Tone" target="_blank">[Dataset]</a> • ⚙️ <a href="https://huggingface.co/Violet-yo/mt5-small-ft-Chinese-Braille" target="_blank">[Model]</a> • 🎬 <a href="https://vision-braille.com/" target="_blank">[Demo]</a>
</p>

## Dataset Description
The `Chinese-Braille-10per-Tone` dataset addresses the scarcity of publicly available Chinese Braille datasets. The original Chinese text data was sourced from the publicly available *Leipzig Corpora Collection*. This dataset consists of one million discrete sentences collected from news media between 2007 and 2009.

The Chinese characters from the Leipzig Corpora Collection were converted into "fully toned" Braille using tools from The Braille Online Platform of China. The dataset is denoted as `Chinese-Braille-Dataset-Full-Tone` and can be found in https://huggingface.co/datasets/Violet-yo/Chinese-Braille-Dataset-Full-Tone. This tool labels the pronunciation of Chinese characters and converts them into Braille using rule-based appraoches. However, the tool sometimes returned misaligned data and garbled text. After cleaning the data and removing duplicates and misaligned sentences, the dataset size was reduced to about 600K sentences.

We randomly remove 90% tones from the `Chinese-Braille-Dataset-Full-Tone` dataset, resulting in the 
`Chinese-Braille-10per-Tone` dataset, as real-world usage of Braille omits about 90% of the tones accordindg to the research. To accurately remove tones from the Braille dataset without affecting numbers or punctuation marks, we followed the methoded described below:

Tones in Braille are represented as ⠁ (first tone), ⠂ (second tone), ⠄ (third tone), and ⠆ (fourth tone). Numbers are marked with ⠼ (e.g., 1 is ⠼⠁), and punctuation marks have specific patterns (e.g., period: ⠐⠆, question mark: ⠐⠄, exclamation mark: ⠰⠂). We split sentences into words using spaces. Because numbers start with ⠼ and punctuation marks have unique combinations, we could easily identify and exclude them from the removal process. This ensured that only the tone markers were removed.

We also created a `Chinese-Braille-Dataset-No-Tone` dataset by removing all tones from the `Chinese-Braille-Dataset-Full-Tone` dataset. This dataset can be found in https://huggingface.co/datasets/Violet-yo/Chinese-Braille-Dataset-No-Tone.

All the source code for the pre-processing is available on the [Github repository](https://github.com/AlanYWu/ChineseBrailleTranslation).

## Dataset Statistics
|           | \# Sample | Braille Len. (Mean/Median) String | Braille Len. (Mean/Median) Token | Chinese Len. (Mean/Median) String | Chinese Len. (Mean/Median) Token |
|-----------|:---------:|:--------------------------------:|:--------------------------------:|:---------------------------------:|:--------------------------------:|
| Training  | 525072    | 145/112                          | 149/115                          | 74/64                             | 59/51                            |
| Validation| 65634     | 117/96                           | 121/99                           | 71/62                             | 57/50                            |
| Test      | 65634     | 107/90                           | 111/93                           | 72/63                             | 58/50                            |

This table includes the original Braille lengths (mean and median) for both strings and tokens directly after the sample count.

## Resources
- Homepage: [Vision-Braille](https://vision-braille.com/)
- Repository: [Github](https://github.com/AlanYWu/ChineseBrailleTranslation)
- Paper: [arXiv](https://arxiv.org/)
- HuggingFace: [Dataset](https://huggingface.co/datasets/Violet-yo/Chinese-Braille-Dataset-10per-Tone), [Model](https://huggingface.co/Violet-yo/mt5-small-ft-Chinese-Braille)
  - [Full Tone Dataset](https://huggingface.co/datasets/Violet-yo/Chinese-Braille-Dataset-Full-Tone)
  - [No Tone Dataset](https://huggingface.co/datasets/Violet-yo/Chinese-Braille-Dataset-No-Tone)