LANGUAGE WORD LIST NOTICES — WordAndLetters =========================================== These lists are compiled for practice unscramblers. They are NOT official academy or tournament dictionaries. English tools use the Standard and Extended open lists documented in WORDLIST-LICENSES.md. Counts below were measured from the files actually shipped in this build (2026-10-18), not from build-time estimates. SHIPPED LANGUAGE LISTS ---------------------- File Entries Language Hunspell-attested With definitions words-es.js 698,269 Spanish 652,164 178,150 words-de.js 740,907 German 325,623 102,417 words-it.js 467,975 Italian 494,596 119,630 words-fr.js 372,775 French 153,598 90,621 --------- total 2,279,926 THE LANGUAGE LOCK (why these lists are shorter than the raw sources) ------------------------------------------------------------------ The four lists are built from OpenSubtitles-derived frequency data. That corpus carries a large amount of material that is not vocabulary of the target language: personal names ("charlie", "peter", "morgan"), place and brand names, Japanese and Chinese proper nouns, romanised surnames, onomatopoeia, and — worst of all — English. Before this build the Spanish list contained 31,992 words that are also in the site's own English Standard list, among them `revealing`, `milky`, `beachfront`, `championship`, `counterfeit` and `straddle`. Every one of the four files is now filtered by a two-part rule. A word is shipped only if: 1. It is written entirely in that language's own alphabet, with no digits, quotation marks, hyphens, apostrophes or letters from another script; and 2. It is either attested in that language's Hunspell lexicon, or it ranks at least as high in that language's FrequencyWords list as it does in the English FrequencyWords list. A third rule trims the tail: a word of four letters or fewer that is not lexicon-attested must occur at least five times in the frequency corpus. The short tail of a subtitle corpus is keyboard mash and OCR noise — `xxb`, `ffc`, `wff`, `svcd`, `umd`, `adln` — and none of those can ever appear in a real rack. Long words are left alone on purpose: rare-but-real long forms are exactly what a player wants an anagram tool to find. Condition 2 is what removes the English imports. Using a language's own lexicon alone would be wrong in both directions: it would strip legitimate cognates (`total`, `animal`, `moral`, `canal` are English *and* Spanish; `war`, `gib`, `tut` are English *and* German; `non`, `di`, `fa`, `testa`, `vista` are English *and* Italian) while keeping the noise. Using the raw frequency list alone keeps all the English. The combined rule does neither. Net effect of the language lock in this build: Language Before After Removed Added Spanish 500,034 698,269 182,757 380,992 German 380,036 740,907 97,364 458,235 French 250,013 372,775 117,562 240,324 Italian 250,017 467,975 92,847 310,805 Almost everything removed is a proper noun or an English word: `glendon`, `farrell`, `preach`, `planning`, `newly`, `outcome`, `swore`, `morph`, `matchmaking`, `yuppie`, `albums`. A small residue is genuine vocabulary the frequency corpora do not use often enough, and a handful of real cognates lost their attestation; the trade is heavily in favour of correctness. SOURCES AND LICENSES -------------------- 1. hermitdave/FrequencyWords https://github.com/hermitdave/FrequencyWords - Code: MIT License, Copyright (c) 2016 Hermit Dave. - CONTENT: Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0). https://creativecommons.org/licenses/by-sa/4.0/ NOTE: the repository LICENSE file is MIT, but that MIT licence covers the generator CODE only. The upstream README states the generated word lists are "CC-by-sa-4.0 for content". Earlier revisions of this file described these lists as "free for any use", which was inaccurate. The Spanish, German, French and Italian lists are all built from this CC BY-SA 4.0 content. - This build reads content/2018//_full.txt for es, de, fr and it, and content/2018/en/en_full.txt as the language-lock reference. - IMPORTANT PROVENANCE NOTE: these word lists are generated from the OpenSubtitles.org subtitle corpus. The corpus is a compilation of film and television dialogue, and the source repository asserts CC BY-SA 4.0 over the derived word lists. Individual word spellings and word-frequency data are facts and are not copyrightable (see Words and Numbers, 9 F.3d 343 (2d Cir. 1993)), and no sentence, subtitle file, dialogue line or expressive content is reproduced here. We disclose the upstream chain so that any reviewer can retrieve the original and judge it for themselves. 2. words/an-array-of-spanish-words https://github.com/words/an-array-of-spanish-words MIT License, Copyright (c) Zeke Sikelianos. Derived from the Letterpress word list. ~636,000 forms considered and merged in full in this build. The MIT licence requires the copyright notice to be retained in copies, so it is reproduced on the Credits page. 3. LibreOffice / dictionaries — Hunspell https://github.com/LibreOffice/dictionaries Locales used: es_ES, es_MX, es_AR, es_CL, es_CO, es_US, es_UY, es_VE, es_PE (Spanish); de_DE_frami, de_AT_frami, de_CH_frami (German); fr_FR (French); it_IT (Italian). The es/ directory ships under MPL-1.1, LGPL-2.1, LGPL-3.0 and GPLv3 notices, and the upstream hunspell dictionaries are offered under MPL-2.0 / LGPL-2.1+ / GPL-2.0+ at the user's option. This build relies on the MPL-2.0 option. Why this is low risk for a shipped spelling list: the material taken from these files is individual word spellings. Short words and word lists of single terms are not independently copyrightable (see Words and Numbers, 9 F.3d 343 (2d Cir. 1993)). No .aff affix rules, no thesaurus, no hyphenation patterns, no dictionary prose and no images are redistributed. The affix tables were executed at build time to expand the shipped .dic stems into inflected forms; only the resulting spellings are redistributed. Obligations we still meet voluntarily: the upstream project, licence options and source location are disclosed here and on the Credits page so any reviewer can retrieve and compare the originals. SOURCES CONSIDERED AND REJECTED ------------------------------- GNU Aspell dictionaries (aspell-de, aspell-es, aspell-fr, aspell-it) were downloaded and evaluated as an attestation lexicon. All four are GPL-2+ / GPL. GPL data cannot be relicensed into a CC BY-SA 4.0 word list without creating a conflict, so they were excluded. The Hunspell dictionaries above are offered under MPL-2.0 as well and were used instead. SHAREALIKE HANDLING (CC BY-SA 4.0) ----------------------------------- The four language lists are adapted from CC BY-SA 4.0 content, so: - They are offered under CC BY-SA 4.0. https://creativecommons.org/licenses/by-sa/4.0/ - Attribution is given on the Credits page: FrequencyWords content by Hermit Dave, CC BY-SA 4.0. - A machine-readable notice is shipped in the header comment of each words-.js file, so the licence travels with the data. - ShareAlike attaches to those adapted word lists only. It does not propagate to the site's HTML, CSS, JavaScript, layout or branding. WordAndLetters original UI code is not relicensed by this notice. Scope note: no official RAE, Duden, Académie française or Zingarelli vocabulary was used. These are frequency-ordered open compilations. WIKTIONARY / KAIKKI (CC BY-SA 4.0) ----------------------------------- Bulk Wiktionary lexeme dumps are NOT compiled into the language unscrambler .js lists. The dictionary layer in assets/dict/ and assets/dict-lang/ does contain Wiktionary definitions and is labelled in the UI. Those definitions remain under CC BY-SA 4.0 with attribution on Credits and a per-sense source label in the dictionary interface. Material from English Wiktionary is used under CC BY-SA 4.0. "Wiktionary" and "Wikimedia" are trademarks of the Wikimedia Foundation. WordAndLetters is not affiliated with or endorsed by the Wikimedia Foundation. Definition layers shipped in this build, by language, with coverage of that language's word list: Language Headwords Senses Coverage of the word list Spanish 178,150 211,966 25.5% German 102,417 140,573 13.8% Italian 119,630 140,940 25.6% French 90,621 107,285 24.3% These glosses are written in English. They explain what a foreign word means; they are not translations. REVISIONS --------- 2026-10-18 Language lock applied to all four lists, plus a frequency floor on unattested words of four letters or fewer to strip the keyboard-mash tail of the subtitle corpus. English imports, other-language spellings, proper-noun runs and non-word tokens removed using Hunspell attestation plus a cross-corpus FrequencyWords ranking test. Counts replaced with measured shipped counts (every earlier figure in this file was wrong). Hunspell added for German, French and Italian and documented under MPL-2.0. an-array-of-spanish-words merged in full. OpenSubtitles provenance of FrequencyWords disclosed. GNU Aspell dictionaries recorded as considered and rejected on licence grounds. 100,894 broken "[object Object]" etymology values in the dictionary layer cleared. 2026-10-17 Corrected FrequencyWords content licence from "free for any use" to CC BY-SA 4.0; replaced contradictory per-language counts with measured shipped counts; added MIT attribution for an-array-of-spanish-words; documented LibreOffice hunspell provenance and the bare-spellings-only rationale; removed the stale "not bundled" claim.