# Word game data attribution

The Korean word vectors originate from fastText's Common Crawl and Wikipedia
Korean pretrained vectors, `cc.ko.300.vec.gz`, 2,000,000 entries and 300 dimensions.
They are distributed under Creative Commons Attribution-ShareAlike 3.0.

Source and license: https://fasttext.cc/docs/en/crawl-vectors.html

E. Grave, P. Bojanowski, P. Gupta, A. Joulin, and T. Mikolov,
*Learning Word Vectors for 157 Languages*, LREC 2018.
Paper: https://arxiv.org/abs/1802.06893

Changes made for this site: retain only NFC-normalized Hangul entries, remove
normalized duplicates, normalize vectors to unit length, store their values as
little-endian float32, and divide the files into private server-side shards.
No new training or inference API is used. The transformed vector data retain
the CC-BY-SA-3.0 license. Its full text is in `fasttext-CC-BY-SA-3.0.txt`.

The original daily-answer catalog and reference game implementation come from
NewsJelly's `semantle-ko`, a fork of Johannes Gätjen's German Semantlich,
itself based on the Semantle word-similarity game. The repository license is
GPL-3.0; its full text is in `semantle-ko-GPL-3.0.txt`.

Source: https://github.com/NewsJelly/semantle-ko
Commit: c265d3b6752d329b81559a1bcea562e9b70cf94f (2024-08-07)

The ranking pool and accepted fourteen-basic-jamo Korean words use `ko.dic` from
Hunspell Korean dictionary version 0.7.92. Built dictionary files are licensed
under GPL version 3 or later. Copyright (C) 2008-2019 hunspell-dict-ko
contributors; Copyright (C) 2017-2019 National Institute of Korean Language.
The original detailed notice and GPL text are preserved in
`hunspell-dict-ko-LICENSE.md` and `hunspell-dict-ko-GPL-3.0.txt`.

Source: https://github.com/spellcheck-ko/hunspell-dict-ko/releases/tag/0.7.92

The ranking pool differs from the historical NewsJelly deployment: this site
uses all normalized Hangul dictionary headwords with available vectors and
does not run the upstream transformer-based profanity filter. Guess vectors
retain the original Hangul-only input filter; no dictionary subset is imposed
on accepted guesses. All 4,650 original answers have vectors and are present
in the ranking pool.

## Fourteen-jamo answer selection

The expanded daily answer words and the separate `korean-fourteen-stdict.json` accepted-word dictionary are derived from the National Institute of Korean Language Standard Korean Dictionary, licensed CC BY-SA 2.0 KR. Original headwords, citation links, normalization and license details are preserved in [stdict-CC-BY-SA-2.0-KR.md](stdict-CC-BY-SA-2.0-KR.md). This selection data is separate from the GPL-3.0-or-later allowed-word dictionary and game code.