nassimjp/pashto-cpt-lexicon

Pashto CPT Lexicon

A clean, normalized, deduplicated, and shuffled Pashto word‑level lexicon designed for CPT (Continual Pre‑Training), tokenizer calibration, linguistic analysis, and Pashto LLM development.

This dataset contains high‑quality Pashto tokens collected from curated text sources, normalized using consistent Pashto orthographic rules, and prepared for direct use in pretraining pipelines.

---

📌 Dataset Summary

• Language: Pashto (پښتو)
• Tokens: Word‑level (one token per line)
• Format: Plain text (.txt)
• Encoding: UTF‑8
• Normalization:
  • آ، أ، إ، ا → ا
  • ی، ې، ۍ → ی
  • Tatweel removed
  • Zero‑width joiners removed
• Processing:
  • Deduplicated
  • Shuffled
  • Cleaned
  • CPT‑ready

---

📂 Files Included

• final_pashto_words_shuffled.txt
  A single‑column lexicon containing all Pashto tokens, one per line.

---

🔧 Example Usage

with open("final_pashto_words_shuffled.txt", "r", encoding="utf-8") as f:
    words = [line.strip() for line in f]

---

📜 License

MIT License — Open for research and Pashto NLP development.

---

👤 Author

Nassim JP
Pashto AI Research & Development
Japan

---