ML-Jonibek/Uzb-Eng-Translation-1

📖 English ↔ Uzbek Literary Translation Dataset

---

🌟 Dataset Summary

This dataset contains parallel sentence pairs translated between English (EN) and Uzbek (UZ), extracted from the classic literary work One Thousand and One Nights (Ming bir kecha / مینگ بیر کیچه).

The dataset is ideal for:

• 🤖 Training and fine-tuning Neural Machine Translation (NMT) models for the EN↔UZ language pair
• 📊 Benchmarking translation quality on literary text
• 🔍 Linguistic research on English–Uzbek language structure
• 📚 Low-resource language NLP development (Uzbek)

---

📂 Dataset Structure

Translation1_en_uz.jsonl

Each line is a JSON object with two fields:

Field	Type	Description
en	string	English source sentence/segment
uz	string	Uzbek translated sentence/segment

🔎 Example Entry

{
  "en": "A good story belongs to the whole world.",
  "uz": "Yaxshi hikoya butun dunyoga tegishli."
}

{
  "en": "The stories known as the Thousand and One Nights are very old.",
  "uz": "\"Ming bir kecha\" nomi bilan tanilgan hikoyalar juda qadimiy."
}

---

📊 Dataset Statistics

Property	Value
Format	JSONL
Total pairs	~2,245
Source language	English (en)
Target language	Uzbek (uz)
Domain	Literary / Classic
Text type	Prose, narrative
Script (UZ)	Latin (Uzbek Latin alphabet)

---

📖 Source Material

🏛️ One Thousand and One Nights — an ancient collection of Middle Eastern folk tales compiled during the Islamic Golden Age.

Property	Details
Title	One Thousand and One Nights / Ming bir kecha
Origin	9th century CE
Publisher	The Penn Publishing Company, Philadelphia, 1928
Source	en.wikisource.org
Illustrator	Virginia Frances Sterrett
Download	www.aliceandbooks.com
License	Public Domain (CC0 1.0)

---

🚀 Loading the Dataset

Using 🤗 HuggingFace Datasets

from datasets import load_dataset

dataset = load_dataset("ML-Jonibek/Translation1_en_uz", split="train")
print(dataset[0])
# {'en': 'A good story belongs to the whole world.', 
#  'uz': 'Yaxshi hikoya butun dunyoga tegishli.'}

Manual loading with Python

import json

data = []
with open("Translation1_en_uz.jsonl", "r", encoding="utf-8") as f:
    for line in f:
        data.append(json.loads(line))

print(f"Total pairs: {len(data)}")
print(data[0])

---

🎯 Intended Use

✅ Suitable for

• Fine-tuning multilingual models (mBART, NLLB, mT5, etc.) on EN↔UZ
• Creating Uzbek language corpora for NLP research
• Training sequence-to-sequence translation models
• Evaluating BLEU/chrF scores on literary Uzbek text

⚠️ Limitations

• Text is literary in style — may not generalize well to news, technical, or conversational domains
• Uzbek translations may reflect older translation conventions
• Some segments are sentence fragments due to literary segmentation

---

💡 Example Use Case: Fine-tuning with transformers

from transformers import MarianMTModel, MarianTokenizer

model_name = "Helsinki-NLP/opus-mt-en-uz"
tokenizer = MarianTokenizer.from_pretrained(model_name)
model = MarianMTModel.from_pretrained(model_name)

texts = ["A good story belongs to the whole world."]
inputs = tokenizer(texts, return_tensors="pt", padding=True)
translated = model.generate(**inputs)
print(tokenizer.decode(translated[0], skip_special_tokens=True))

---

📜 License

This dataset is derived from public domain source material and is released under the Creative Commons Zero (CC0 1.0) license.

You can copy, modify, distribute and perform the work, even for commercial purposes, all without asking permission.

---

🙏 Citation

If you use this dataset in your research, please cite:

@dataset{translation1_en_uz_2024,
  title     = {English-Uzbek Literary Translation Dataset (One Thousand and One Nights)},
  language  = {en, uz},
  source    = {One Thousand and One Nights, Penn Publishing Company, 1928},
  license   = {CC0 1.0 Universal},
  url       = {https://huggingface.co/datasets/your-username/Translation1_en_uz}
}

---

🤝 Contributing

Contributions, corrections, and improvements are welcome!
Feel free to open an issue or pull request on the dataset repository.

---

Made with ❤️ for the Uzbek NLP community 🇺🇿