nassimjp/english_historical_quotes_in_pashto

📚 English Historical Quotes in Pashto — Chat Format Dataset

A high‑quality, Pashto‑translated version of English Historical Quotes, converted into a chat‑style format suitable for training Pashto LLMs on quotation understanding, author attribution, and category‑based semantic reasoning.

This dataset transforms each quote into:

{
  "messages": [
    {"role": "user", "content": "<Pashto Quote>"},
    {"role": "assistant", "content": "لیکوال: <Author>\nکټګورۍ: <Categories>"}
  ]
}

🧩 Dataset Structure

Each row contains:

• —id — unique string identifier
• —messages — list of chat messages
• —user → Pashto quote
• —assistant → author + categories (Pashto)

Example:

{
  "id": "001493",
  "messages": [
    {"role": "user", "content": "د خوښۍ حق بنسټیز دی."},
    {"role": "assistant", "content": "لیکواله: انا پاولووا\nکټګورۍ: خوښۍ"}
  ]
}

🎯 Purpose

This dataset is designed for:

• —Pashto instruction‑tuning
• —Pashto assistant alignment
• —Quote‑based semantic reasoning
• —Author attribution tasks
• —Category‑based classification
• —Lightweight conversational fine‑tuning

It is NOT intended for:

• —Deep reasoning SFT
• —Chain‑of‑thought training
• —Multi‑turn conversation modeling
• —Long‑context LLM training

🛠️ Source & Processing

Original dataset: m-ric/english_historical_quotes

Processing steps:

1. 1.Quotes translated into Pashto
2. 2.Author names normalized
3. 3.Categories translated and cleaned
4. 4.Converted into HF chat format
5. 5.UTF‑8 normalization
6. 6.Deterministic JSONL generation

📦 File Format

Dataset is provided as:

• —train.jsonl — main chat dataset
• —Each line is a complete chat sample
• —Fully compatible with:
• —HuggingFace datasets
• —TRL
• —OpenChat / ChatML loaders
• —Talanda pipelines

🔍 Loading the Dataset

from datasets import load_dataset

ds = load_dataset("nassimjp/english_historical_quotes_in_pashto", split="train")

print(ds[0])

📈 Recommended Use Cases

• —Pashto assistant fine‑tuning
• —Cultural & historical quote understanding
• —Lightweight alignment tasks
• —Category‑aware semantic modeling
• —Author‑aware contextual reasoning

⚠️ Limitations

• —Quotes are short → not suitable for long‑context training
• —Assistant messages are metadata only → not reasoning
• —No multi‑turn dialogue
• —Not intended for CoT or deep instruction following

📜 License

This dataset follows the licensing terms of the original source dataset. Translations and chat formatting are provided by Spinzar Enterprises Inc. (Yūgen Kaisha).

🤝 Contributions

If you want to add:

• —More quotes
• —More categories
• —Better translations
• —Additional metadata

Feel free to open a PR or contact the maintainer.

🌟 Maintainer

Nassim Founder — Spinzar Enterprises Inc. (Yūgen Kaisha), Japan Pashto AI Research & Development Division