Arabic word assets — source attribution and distribution notices

Ayaspell / Hunspell-ar
---------------------
Original spellchecker: Mohamed Kebdani, 2006–2008.
Arabic thesaurus attribution in upstream AUTHORS: Taha Zerrouki, 2006–2009.
OpenOffice package: Ahmad Farghal.

Original files: ar.aff, ar.dic, AUTHORS.txt, COPYING.txt, README_ar.txt.
Pinned LibreOffice dictionaries revision:
67032ee957631908098a72edf48dc81cac27e369
https://github.com/LibreOffice/dictionaries/tree/67032ee957631908098a72edf48dc81cac27e369/ar
Original project: http://ayaspell.sourceforge.net/

The preserved COPYING.txt offers GPL 2.0-or-later, LGPL 2.1-or-later, or
MPL 1.1-or-later as alternatives. This distribution uses the MPL 1.1 option
for the Ayaspell-derived data. The complete MPL 1.1 text is included at
arabic-source/MPL-1.1.txt. Original notices are preserved without alteration.

Local data modifications, 2026-10-25: parse the pinned aliases and rules;
retain compact stems through nine base letters for final three–seven-letter
forms; canonicalize the documented optional writing marks; preserve exact
exceptions for source spellings whose canonicalization affects affix
conditions; separate a finite candidate list from productive membership;
select a provisional everyday answer list; add separately cited literal
vocabulary omissions. Raw source files, the complete source for generating
these data, and the original spelling/review records are included. See
arabic-source/PROVENANCE.md and scripts/build-arabic-words.py.

wordfreq 3.1.1
--------------
Robyn Speer, wordfreq: https://github.com/rspeer/wordfreq
Version: https://pypi.org/project/wordfreq/3.1.1/
Data: Creative Commons Attribution-ShareAlike 4.0.
Code: Apache License 2.0. The runtime and builder do not import that code.

The original large_ar.msgpack.gz, its versioned package metadata and full
credits, the code-license notice, and the CC BY-SA 4.0 legal text are kept in
arabic-source/. The metadata preserves credits for the Leeds Internet Corpus,
Wikipedia, ParaCrawl, OPUS OpenSubtitles 2018 and SUBTLEX. Credit for SUBTLEX
belongs to Marc Brysbaert and collaborators; SUBTLEX remains freely available.
Preserve these original credits when redistributing derived frequency data.

Local corpus changes: canonicalize supported Arabic spellings, keep native
three–seven-cell tokens, and use source membership to select a finite list of
corpus-attested guesses. No frequency threshold chooses secret answers.
The wordfreq-derived selection/familiarity data retain CC BY-SA 4.0 terms.

The runtime JSON bundle collects morphology and finite candidates in separate
fields. Keep both source notices with the union candidate list and bundle;
packaging does not replace either component's license or change the game
application's license. Raw source and transformation records make the
components distinguishable and reproducible.

Project supplement and answer review
-----------------------------------
Nineteen literal supplemental spellings are separately identified in
arabic-source/supplement-ar.json. Its primary-publisher links establish word
usage; no articles, commercial dictionary definitions, images or tables are
redistributed. Concise English sense labels and editorial review notes are
original project annotations. The additions are not claimed to be Ayaspell.
Only ten supplemental nouns occur in the answer list; the other nine forms
are accepted guesses. Native editorial review is pending.

Provided without warranty, including a guarantee of exhaustive vocabulary,
native familiarity or fitness for a particular purpose. Upstream license
texts contain the applicable full terms.