Data sources & licences
Where Word Holder's dictionary content comes from, what we change, and the licence each source is under.
Wiktionary, via Kaikki.org
Every sense in Word Holder — the meanings, their labels (archaic, regional…), pronunciations, noun plurals, autocomplete suggestions and verb conjugation tables — starts from Wiktionary, written by its contributors and extracted into machine-readable form by Kaikki.org. We use the English Wiktionary and, for words it lacks, the Wiktionary edition in the word's own language.
No entry, no definition. When neither edition has the word, Word Holder does not write a definition for it. A conjugation table for a regular verb Wiktionary has no table for is generated by rule; it is labelled “Generated” and is not Wiktionary's.
What we change. We select and rank the senses (choosing the one that fits the sentence you are reading), merge and drop duplicates, and use AI to write the gloss in the learner's language, the definition and the example sentences, grounded on the Wiktionary senses. Translations of examples are ours too. Under every definition a line links to the exact Wiktionary entry the senses came from, so you can compare.
Wiktionary is licensed under CC BY-SA 4.0. The definitions, examples, suggestions and conjugations we derive from it are made available under the same licence, with the changes described above. Attribution: “Wiktionary, via Kaikki.org. Modified by Word Holder.” The published dictionary data and its notice are at gs://wordholder-dicts/NOTICE.txt.
Leipzig Corpora Collection
The “Example sentences” panel under a definition shows real sentences, mostly from news text, taken from the Leipzig Corpora Collection and selected automatically for readability. Their English translations are machine translations. The corpora are licensed under CC BY 4.0. A review card that uses one of these sentences credits it on the card.
Word-frequency lists
How common a word is — used to order suggestions and rank words — is counted from Wikipedia article text and from the FrequencyWords lists (OpenSubtitles). Both are share-alike sources; the derived lists are published under CC BY-SA 4.0, with changes indicated. Attribution: “Wikipedia. Modified by Word Holder.”
Spell checking
“Did you mean…?” suggestions use word lists from the LibreOffice / Hunspell dictionaries. Each dictionary keeps its own upstream licence (GPL, LGPL, MPL, BSD and others); they are not ours to relicense.
Reading passages
Each reading passage carries its own source and licence, shown in the reader. They are not covered by the licences above.
Written by Word Holder
Every definition starts from the word's Wiktionary entry. The senses, labels and English glosses come from there, via Kaikki.org. Word Holder's AI then writes each sense's explanation, the translations and extra example sentences. It works from that entry and does not add meanings of its own. A second AI pass checks every sentence it writes, and any that fail are dropped. Examples from Wiktionary or the Leipzig Corpora Collection are real usage, not generated.
Glosses in your language and the quizzes are written by Word Holder with the help of AI language models too, grounded on the sources above. Mistakes in them are ours, not the sources'. Spotted one? Tell us.