Bayanat Labs Open Arabic Books
Arabic books published by the Hindawi Foundation, from Andreas Hallberg's The Arabic E-Book Corpus, rebuilt into chapters: 1,545 of the corpus's 1,745 books, more than fifty-four million words of body text. Front matter and licence pages are set apart and footnotes are split out with their positions. Every row in the dataset has an attribution line and the book's own licence wording; the chapter and book rows also hold raw and normalised text. Each book's colophon was read, and a book is included only when its own statement supports the release. For many books the CC BY statement covers the design, and the text rests on the publisher's public-domain statement; copyright terms differ by country, so check the law of your own jurisdiction. Dataset under CC BY 4.0, builder code under Apache-2.0.
| Licence | CC BY 4.0 (dataset and page text), with the rights of each book as its colophon states them; Apache-2.0 (builder code) |
|---|---|
| Version | v0.1.0 · Built from The Arabic E-Book Corpus, version 1, with colophons read on 2026-11-04; the builder is tagged v0.1.0 on GitHub and is not on PyPI. |
| Published | |
| Updated | |
| Who it is for | Teams and researchers working on Arabic language models, long-document reading or summarisation who want whole modern books by chapter, with credits and each book's own licence wording. |
| Level | Intermediate |
Key numbers
Counted by script from the released files. Each key number is read from stats/stats.json or a file in qa/ of the Hugging Face dataset at revision 39876c0e7f28c2d16887c7c139800e0df9f87263 and pinned to its SHA-256. Code at tag v0.1.0, commit 259f92797634829a0b7c9d0186305df65ae43ded. MANIFEST.json lists the configs and their files, and the dataset's checksum file gives the SHA-256 of every file.
Downloads and identifiers
| Where | What |
|---|---|
| Hugging Face | Dataset on Hugging Face |
| GitHub | Builder on GitHub |
| huggingface.co | Counts (stats/stats.json) |
| huggingface.co | Text comparison (qa/text_comparison.json) |
| huggingface.co | Licence re-reading (qa/licence_recheck.json) |
| huggingface.co | Reading check (qa/reading_check.json) |
| huggingface.co | Sources and licences (LICENSE-sources.md) |
| bayanatlabs.com | Arabic Wikisource Public-Domain Works, published separately |
| bayanatlabs.com | arabic-normalize, the policy behind the normalised text |
Get started
Load a config with the datasets library. The builder installs from GitHub at the v0.1.0 tag; the package is not on PyPI.
from datasets import load_dataset
chapters = load_dataset("Bayanat-Labs/open-arabic-books", "chapters", split="train")
books = load_dataset("Bayanat-Labs/open-arabic-books", "books", split="train")
metadata = load_dataset("Bayanat-Labs/open-arabic-books", "metadata", split="train")
pip install "git+https://github.com/Bayanat-Labs/open-arabic-books@v0.1.0"What is released
| Part | Licence | In this release |
|---|---|---|
| chapters (Parquet): one chapter, preface, appendix or similar section per row, with raw and normalised text, footnotes, attribution and licence wording | CC BY 4.0 | Yes |
| books (Parquet): one whole book per row, with text, footnotes, front matter and licence page | CC BY 4.0 | Yes |
| metadata (Parquet): one book per row without its text: credits, dates, categories, licence evidence and source hash | CC BY 4.0 | Yes |
| Counts in stats/ and checks in qa/, with the file list and checksums | CC BY 4.0 | Yes |
| Builder code on GitHub; the repository holds no text from any book | Apache-2.0 | Yes |
| Text of this record | CC BY 4.0 | Yes |
Attribution: Authors, translators, editors, compilers and foreword writers as each book credits them; Hindawi Foundation, publisher; Andreas Hallberg, The Arabic E-Book Corpus, version 1, University of Gothenburg, 2024, https://doi.org/10.5878/7rbh-gy93, CC BY 4.0; processed by Bayanat Labs, Open Arabic Books, https://huggingface.co/datasets/Bayanat-Labs/open-arabic-books.
Intended uses
- Read or process whole Arabic books by chapter, with footnotes kept apart and tied to their positions.
- Build language-model or summarisation data, after checking the rights statement against the law of your own jurisdiction.
- Filter by each book's licence decision, original-text status and category with the metadata config.
Out of scope
- Treating the publisher's public-domain statement as legal advice or as settled for your country: copyright terms differ by country.
- Reusing the text without credit: use the attribution line in each row or the credit chain on this page.
- Treating the release as affiliated with or endorsed by the Hindawi Foundation, Andreas Hallberg or the University of Gothenburg: it is neither.
Rights at a glance
Training, display, redistribution and synthetic-voice use are separate permissions.
| Use | Status | Note |
|---|---|---|
| Training | Allowed | Dataset under CC BY 4.0, with credit. For many books the CC BY statement covers the design and the text rests on the publisher's public-domain statement; copyright terms differ by country, so check the law of your own jurisdiction. |
| Display | Allowed | With credit: authors and others as each book credits them, the Hindawi Foundation as publisher, Andreas Hallberg's corpus and Bayanat Labs. |
| Redistribution | Allowed | Dataset under CC BY 4.0 with the same credit, code under Apache-2.0. Cover art and images are not included. No trademark rights are granted, and no endorsement by any party named here is implied. |
| Synthetic voice | Not applicable | No audio. |
Limitations
- Scope: 1,545 of the 1,745 books in the corpus. The other 200 are not in this release: for 187 translations the colophon does not state the licence of the Arabic translation, for 12 it states a different licence or reserved rights, and for 1 the author's own licence contradicts its public-domain statement.
- What CC BY 4.0 covers: each book's colophon was read from the corpus file. For 1,212 books it says the original Arabic text is in the public domain and the book design is under CC BY 4.0, so the CC BY statement covers the design and the text rests on the publisher's public-domain statement. For 332 translations it puts the Arabic translation and the design under CC BY 4.0 and the original text in the public domain; for 1 it puts the original text under CC BY 3.0. Each row quotes the wording in license_quote.
- "Public domain" is the publisher's statement on the book. Copyright terms differ by country, and many of these authors died within the last seventy years: Wikidata death years show 356 books whose matched author died after nineteen fifty-five, and many authors have no match. Some works are also recent enough to remain protected where terms run from publication. Check the law of your own jurisdiction before relying on it.
- Where a colophon says the original text is in the public domain and the corpus year is recent, the book was read with the author's death year where known: 169 such books are kept.
- Median word recall against the corpus plain-text edition is 0.9995, and the lowest is 0.973. A separate check re-read every source hash and licence quotation, with 0 mismatches. Samples from 100 random books were read, a sample check rather than a proofread, and chapter boundaries were correct in 100. Spelling and conversion errors inside chapters outside the samples were not checked.
- Images and SVG figures, including formulas drawn as SVG, are not in the text; columns count what was left out. Duplicate chapters are marked, not removed.
- Corpus values such as the original author, year and title are the corpus's own, unchanged, typos and odd years included.
- Book credits name people, some of them living. They appear only as the book credits them. To ask for a name or record to be corrected or removed, write to hello@bayanatlabs.com.
Sources and licences
| Source | Licence | Redistributed here |
|---|---|---|
| The Arabic E-Book Corpus, Andreas Hallberg, University of Gothenburg, version 1, 2024; books published by the Hindawi Foundation | CC BY 4.0; files matched the SHA-256 values of the official record. The rights of each book are stated in its colophon, quoted in license_quote | Yes |
| Wikidata, for identifiers and death years | CC0 1.0 | Yes |
Credit: authors, translators, editors, compilers and foreword writers as each book credits them; Hindawi Foundation, publisher; Andreas Hallberg, The Arabic E-Book Corpus; processed by Bayanat Labs, who split the books into chapters, separated front matter, licence pages and footnotes, normalised the text and added identifiers. Not affiliated with or endorsed by the Hindawi Foundation, Andreas Hallberg or the University of Gothenburg. The books that left the release are listed with their reason in the dataset's qa folder. Source details
How to cite
Cite as:
Bayanat Labs (2026). Open Arabic Books, version 0.1.0 (from The Arabic E-Book Corpus, version 1, Hallberg, 2024). Bayanat Labs. https://huggingface.co/datasets/Bayanat-Labs/open-arabic-books
No DOI. Cite the URL in the citation above and the date you read it.
@misc{bayanatlabs2026openarabicbooks,
author = {{Bayanat Labs}},
title = {Open Arabic Books, version 0.1.0 (from The Arabic E-Book Corpus, version 1, Hallberg, 2024)},
year = {2026},
month = nov,
howpublished = {\url{https://huggingface.co/datasets/Bayanat-Labs/open-arabic-books}}
}
Changelog
| Date | Version | Change |
|---|---|---|
| v0.1.0 | Chapters, books and metadata on Hugging Face, with counts and checks; builder on GitHub. |
Corrections and removal requests
If something in this release should not be there, or is wrong, email hello@bayanatlabs.com. We handle removal requests within 30 days, and platform requests within 24 hours. How removal requests work
Related product: Bayanat Knowledge