CMLI-NLP/Mongolian-pretrain-dataset

---
license: cc-by-4.0
language:
- mn
size_categories:
- 100K<n<1M
---

# Mongolian Pretraining Dataset

## Dataset Information

- **Language**: Mongolian (Traditional Mongolian script)
- **Size**: ~12GB
- **Format**: Plain text (.txt)
- **Use Case**: Language model pretraining

## Description
This dataset contains Mongolian text data for training language models on low-resource languages. The data uses Traditional Mongolian script and covers 45 core characters identified through frequency analysis.

**Code**: The Huffman transliteration framework implementation is available at [https://github.com/CMLI-NLP/HuffmanTranslit](https://github.com/CMLI-NLP/HuffmanTranslit)

## Technical Details

- **Character Set**: 45 core Traditional Mongolian characters
- **Encoding**: UTF-8
- **Script**: Traditional Mongolian script
- **Total Samples**: 933,941

## Usage

Suitable for:
- Continued pretraining of multilingual language models
- Cross-lingual transfer learning research
- Low-resource language processing

## Transliteration Framework Compatibility

This dataset is designed to work with the Huffman-based transliteration framework for reversible conversion between Traditional Mongolian script and Latin characters.

## Citation

If you use this dataset, please cite:

```bibtex
@inproceedings{zhuang2025enhancing,
title={Enhancing Cross-Lingual Transfer through Reversible Transliteration: A Huffman-Based Approach for Low-Resource Languages},
author={Zhuang, Wenhao and Sun, Yuan and Zhao, Xiaobing},
booktitle={Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
pages={16299--16313},
year={2025}
}
```