DIA2-Arabic/DIA2-Dataset-Non-Diacritized

---

DIA2: A Comprehensive and Diverse Diacritized Modern Standard Arabic Corpus

DIA2 is a large-scale, natively sourced, and diacritized Modern Standard Arabic corpus designed for NLP research and LLM development. It is curated from 28 diverse Arabic sources including books, news articles, encyclopedic content, and poetry, and explicitly avoids machine-translated content.

This repository contains the Non-diacritized version of DIA2. The full DIA2 release consists of three datasets:

• DIA2-Dataset-Diacritized
• DIA2-Dataset-Non-Diacritized
• DIA2-Tashkeela-Gold

Preprocessing Pipeline

The preprocessing pipeline includes rigorous text cleaning, URL- and document-level deduplication, and automatic diacritization. Full preprocessing code is available at: https://github.com/fatimadekmak/DIA2-arabic-preprocessing

Citing this Dataset

If you use DIA2 in your research, please cite:

@inproceedings{dekmak2026dia2,
  title     = {DIA2: A Comprehensive and Diverse Diacritized Modern Standard Arabic Corpus for Large-Scale NLP Research},
  author    = {Dekmak, Fatima and Elbassuoni, Shady and Shaban, Khaled and Hajj, Hazem and El Hajj, Wassim and Abu Adla, Yasmine and Alabrash, Buthaina},
  booktitle = {Proceedings of the OSACT7 Workshop at LREC-COLING 2026},
  year      = {2026}
}