asce3/open-arabic-text-corpus

Open Arabic Text Corpus

A curated Arabic text corpus designed to support research, experimentation, and language-model pretraining.

Overview

This dataset contains Arabic-language text collected from documented sources and processed for use in natural language processing and language-model research.

Release: 1.0.0

Primary language: Arabic

Format: JSONL

Dataset contents

Each record represents one text document.

Field	Description
id	Stable document identifier
text	Extracted document text
source	Original dataset or source
url	Source URL, when available
language	Detected language
language_score	Language identification score, if available

Fields may be absent depending on the source.

Processing

The intended pipeline includes:

• Removing empty or invalid documents
• Filtering by text length
• Checking language metadata
• Deduplicating repeated content
• Preserving source provenance

Document which steps have actually been completed for this release.

Intended uses

• Arabic language-model pretraining experiments
• NLP research and text analysis
• Language identification and corpus statistics
• Development of Arabic-language applications

Limitations

• Coverage may be uneven across dialects and topics.
• Language classifiers can misidentify dialectal Arabic.
• Web-derived text may contain errors, duplication, bias, personal information, or low-quality content.
• The corpus should not be treated as a verified factual reference.

Licensing and provenance

The source datasets and their associated terms must be reviewed individually. Consult the original sources and document the applicable redistribution conditions.

Do not assume that all underlying content can be freely redistributed simply because it is publicly accessible.

Loading the dataset

from datasets import load_dataset

dataset = load_dataset(
    "asce3/open-arabic-text-corpus",
    split="train",
    streaming=True,
)

for example in dataset.take(3):
    print(example["text"][:500])

Citation

If you use this dataset, cite the version and source information published with this release.