dataflare/arabic-dialect-corpus

Arabic Dialect Corpus

A comprehensive collection of Arabic dialectal text, standardized for Natural Language Processing (NLP) model training, evaluation, and linguistic analysis. This corpus has been meticulously processed to ensure high-quality tokenization and consistent metadata.

Dataset Statistics

Metric	Value
Total Records	127,180
Total Tokens	5,802,324
Average Tokens per Record	45.62
Dialect Categories	5

Changelog

Version 1.0 (January 2026)

This release establishes the baseline for the corpus with strict quality controls:

• Token Count: Validated 5.8M+ tokens using cl100k_base (GPT-4 standard).
• Data Density: Optimized average record length to ~45 tokens for efficient training.
• Dialect Coverage: Confirmed distribution across 5 distinct dialect categories.
• Quality Assurance: Zero empty records and standardized metadata schema.

Dataset Structure

Each record in the dataset contains the following fields:

• text (string): The raw Arabic text content.
• topic (string): The semantic category or topic of the text.
• utterance_type (string): Classification of the utterance (e.g., statement, question).
• dialect (string): The regional dialect name: Masri (Egyptian), Khaleeji (Gulf), Levantine, Maghrebi (North African), or Iraqi.
• tokens (int): The precise token count calculated using cl100k_base encoding.

Usage

Loading the Dataset

The dataset is hosted on the Hugging Face Hub and can be loaded directly using the datasets library.

from datasets import load_dataset

# Load the complete dataset
dataset = load_dataset("dataflare/arabic-dialect-corpus")

Detailed Methodology

Collection and Processing

The data was aggregated from diverse sources including transcribed media and public archives. The processing pipeline involved:

1. Normalization: Text normalization to remove noise while preserving dialectal features.
2. Segmentation: Splitting long passages into training-ready chunks.
3. Token Counting: Rigorous token counting using tiktoken to assist in curriculum training and length bucketing.

Citation and License

This dataset is released under the MIT License.

If you rely on this corpus for your research or application, please cite it using the following BibTeX entry:

@dataset{arabic_dialect_corpus,
  title={Arabic Dialect Corpus},
  author={Dataflare},
  year={2026},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/dataflare/arabic-dialect-corpus}
}