suwanpan/mongolian-script-text

Mongolian Script Text Dataset

Dataset Summary

This repository contains a curated, high-quality parallel text dataset pairing Modern Cyrillic Mongolian with Traditional Mongolian Script (Mongol Bicig). It is built specifically to support NLP applications targeting low-resource scripts, automated transliteration, orthographic mapping, and morphological analysis.

The dataset currently includes classic 20th-century Mongolian literature, poetry, and children's songs, and more to add soon.

---

Dataset Structure

File Hierarchy

The repository is organized by genre inside separate JSONL files:

• poems/minii_nutag.jsonl: B. Rinchen's poem Minii Nutag (My Native Land).
• poems/mongol_khel_poem.jsonl: D. Natsagdorj's Mongol Khel (Mongolian Language).
• children_songs/maamuu_naash_ir.jsonl: Traditional children's rhymes and songs.

Data Fields

Each sample line contains a clean JSON object with two primary keys:

• cyrillic (string): The text written in Modern Cyrillic Mongolian orthography.
• bicig (string): The aligned text in standard Unicode Traditional Mongolian Script.

Example Row

{
  "cyrillic": "Хэнтий, Хангай, Соёны өндөр сайхан нуруунууд",
  "bicig": "ᠬᠡᠨᠲᠡᠢ᠂ ᠬᠠᠩᠭᠠᠢ᠂ ᠰᠠᠶᠠ᠎ᠤ ᠥᠨᠳᠥᠷ ᠰᠠᠢᠬᠠᠨ ᠨᠢᠷᠤᠭᠤ᠎ᠨᠤᠭᠤᠳ"
}

---

Usage

Loading with Hugging Face Datasets

You can load all splits directly into Python using the datasets library:

from datasets import load_dataset

# Load the entire dataset
dataset = load_dataset("suwanpan/mongolian-script-text")

# Inspect a sample
print(dataset["train"][0])

---

Intended Tasks

• Transliteration & Sequence Mapping: Fine-tuning sequence-to-sequence models (e.g., ByT5, mT5, MarianMT) for Cyrillic Mongol Bicig conversion.
• Morphological Alignment: Learning archaic vowel retention, suffix harmony, and root integrity across Mongolian scripts.
• Text Normalization: Building spellcheckers and literary evaluation benchmarks.

---

License & Citation

This dataset is released under the Apache 2.0 License. Feel free to use, modify, and distribute it for research and open-source development.