WordBonk

Sources & licences

Last updated:

The English version is the official, legally binding text. Translations into other languages are being prepared and are provided for convenience only.

WordBonk is built on the work of many open projects and volunteers. This page names the third-party data and content you see in the app, what we use it for, the licence it is under and what we change.

  • Machine translation and AI content are marked as such. Translations made by a machine-translation service and text generated by AI are not reviewed by the sources listed here.
  • No endorsement. None of these projects, publishers or authors endorses WordBonk or is responsible for how we present their material.
  • Share-alike. Where a source is under a share-alike licence (CC BY-SA), the entries we build from it — definitions, glosses, stroke diagrams — are available under the same licence.
  • Corrections. If a credit is missing or wrong, or you believe material here infringes your rights, please tell us through the copyright notice form.

Dictionaries & translations

  • Wiktionary — definitions, translations, synonyms, word forms and pronunciation for every language, by the Wiktionary contributors.
    Licence: CC BY-SA 4.0 and GFDL (older revisions CC BY-SA 3.0)
    Credit: Wiktionary contributors; each entry's history is on its Wiktionary page.
    Our changes: shortened, filtered, combined with other sources and, where no gloss exists in your language, machine-translated.
  • Kaikki.org (wiktextract) — machine-readable Wiktionary extracts (English, Russian, German, Spanish, French, Portuguese and Japanese editions) for glosses, translations and related words.
    Licence: same as Wiktionary — CC BY-SA 4.0 and GFDL
    Credit: Tatu Ylonen, “Wiktextract: Wiktionary as Machine-Readable Structured Data”, LREC 2022; data from Wiktionary.
    Our changes: as for Wiktionary.
  • DBnary — Wiktionary translation links in RDF (GETALP, Université Grenoble Alpes), used to fill translations into Hindi and Arabic.
    Licence: CC BY-SA 3.0
    Credit: Gilles Sérasset, DBnary: Wiktionary as a Lemon-based multilingual lexical resource in RDF (Semantic Web Journal, 2015).
    Our changes: shortened and filtered.
  • Free Dictionary API (dictionaryapi.dev) — English definitions, phonetic transcriptions and links to pronunciation recordings.
    Licence: Wiktionary content, CC BY-SA 3.0 as stated in each API response
    Our changes: shortened and combined with other sources.
  • WordNet 3.0 — English lemmas and parts of speech (through NLTK), which decide word forms and hints on cards.
    Licence: WordNet 3.0 licence
    Credit: WordNet 3.0 Copyright 2006 by Princeton University. All rights reserved.
  • Google Translate — machine translation of words, sentences and image-search queries when a dictionary has no entry in your language.
    Licence: a Google service, used under Google's terms
    Our changes: machine translation — not reviewed by a human.

Example sentences

  • Tatoeba — English example sentences on cards, chosen from an offline copy of the corpus.
    Licence: CC BY 2.0 FR; some sentences also CC0 1.0. We only ship sentences whose author is known or that are CC0.
    Credit: each Tatoeba example on a card shows its sentence number and author (“Tatoeba #1234 · author”), linked to the sentence on tatoeba.org.
    Our changes: sentences filtered for unsuitable content and ranked by difficulty; the text is not changed.
  • Example sentences in other languages, and in English when Tatoeba has none, are generated by AI — see AI-generated content.

Word frequency & spelling

  • wordfreq — word-frequency estimates for all languages, used to order cards and pick easier example sentences and gaps. The lists themselves are not shown.
    Licence: data CC BY-SA 4.0, code Apache License 2.0
    Credit: Robyn Speer, wordfreq; built from Google Books Ngrams, OpenSubtitles 2018 (OPUS), the SUBTLEX lists (Brysbaert, New, Keuleers, Cai and others), the Leeds Internet Corpus, Wikipedia and ParaCrawl.
  • OpenAI Whisper — English spelling normaliser — pairs of British and American spellings, so “colour” counts as “color” in dictation.
    Licence: MIT License, Copyright (c) 2022 OpenAI
    Our changes: none (copied unchanged).

Japanese

  • JMdict — Japanese word definitions, readings and parts of speech, retrieved through the Jisho.org API.
    Licence: EDRDG licence (CC BY-SA 4.0)
    Credit: This site uses the JMdict dictionary files. These files are the property of the Electronic Dictionary Research and Development Group, and are used in conformance with the Group's licence.
    Our changes: shortened and combined with Wiktionary data and machine translation.
  • KANJIDIC2 — kanji meanings, readings, stroke counts and levels in the kanji breakdown, retrieved through kanjiapi.dev.
    Licence: EDRDG licence (CC BY-SA 4.0)
    Credit: This site uses the KANJIDIC2 dictionary file. This file is the property of the Electronic Dictionary Research and Development Group, and is used in conformance with the Group's licence.
  • Jisho.org and kanjiapi.dev — the free services through which we query JMdict and KANJIDIC2 (kanjiapi.dev).
    Licence: the data is EDRDG's (above).
  • JLPT word lists by Jonathan Waller — the JLPT level shown on Japanese cards, as supplied by Jisho.org.
    Licence: Creative Commons Attribution (CC BY)
    Credit: JLPT data from Jonathan Waller's JLPT Resources.
  • Noto Emoji Animation — the animated emoji in news and messages (shown as Lottie animations from our own server).
    Licence: CC BY 4.0
    Credit: Animated emoji by Google (Noto Emoji).
    Our changes: none; the animations are played as published.
  • KanjiVG — the stroke-order diagrams on Japanese cards.
    Licence: CC BY-SA 3.0
    Credit: KanjiVG is copyright © 2009–2026 Ulrich Apel.
    Our changes: none; the diagrams are drawn as published.

Pronunciation

  • Supertonic 3 — AI speech for words and sentences, synthesised on our own servers.
    Licence: model OpenRAIL-M, code MIT License; Copyright (c) 2026 Supertone Inc.
    Our changes: none to the model; the audio is machine-generated.
  • Microsoft Edge neural voices — AI speech when our own voice is unavailable, through the open-source edge-tts client.
    Licence: a Microsoft service; the audio is machine-generated.
  • Google Translate text-to-speech — basic AI speech through the open-source gTTS library (older app versions).
    Licence: a Google service; the audio is machine-generated.
  • Pronunciation recordings (Wikimedia Commons) — recordings of real speakers linked from Wiktionary entries via dictionaryapi.dev, played when a word has one.
    Licence: each recording's own free licence (mostly CC BY-SA), named on its Commons page.
    Credit: the speakers and uploaders named on each file's Commons page.

Images

Card images are found by a search you start yourself; the picker shows each picture's source and author. We link to the image where its source hosts it. When you choose a picture, its author, licence and source page are saved with the card and shown under the image (“author · licence · source”; Unsplash photos as “Photo by … on Unsplash”). Icons under MIT or Apache 2.0 need no on-screen credit and get none; images added before credits were saved, or pasted as a link, show none.

  • Wikimedia Commons — photos and drawings.
    Licence: each file's own free licence (CC0, CC BY, CC BY-SA or public domain), named on its Commons page.
  • Openverse — openly licensed images, limited to licences that allow commercial use.
    Licence: each work's own Creative Commons licence or public-domain mark, named on its Openverse page.
  • Pixabay — photos and illustrations.
    Licence: Pixabay Content License
  • Unsplash — photos.
    Licence: Unsplash License
    Credit: Photo by the photographer named in the picker, on Unsplash.
  • Icons via Iconify — flat icons and emoji for card pictures, from these sets: Fluent Emoji and Fluent UI Color (Microsoft, MIT License), Streamline color sets (Streamline, CC BY 4.0), IconPark (ByteDance, Apache License 2.0), Marketeq (MIT License), Twemoji (Twitter, CC BY 4.0), Noto Emoji (Google, Apache License 2.0), OpenMoji (CC BY-SA 4.0) and Flat Color Icons (Icons8, MIT License).
    Licence: per icon set, as listed.
    Credit: CC BY and CC BY-SA icons (Streamline, Twemoji, OpenMoji) show the set's author and licence under the image.

Videos

  • YouTube — videos for listening practice and “watch in context” clips, played in YouTube's own embedded player (youtube-nocookie.com), together with their subtitles.
    Licence: videos and subtitles belong to their creators and are shown under the YouTube Terms of Service. We do not host or download the videos.
    Our changes: subtitles are split into phrases and cleaned (sound tags, speaker labels) for dictation; the creator's words are not changed.
  • Filmot — a search index of YouTube subtitles that finds clips where a word is spoken.
    Licence: a third-party service used through its API.

Library books

The bundled library holds only public-domain texts: every author and translator whose work we keep died in 1925 or earlier, and the work was first published by 1930. Introductions, notes and illustrations by later editors are removed.

  • Project Gutenberg — English, German and Spanish books.
    Licence: Public domain
    Our changes: the Project Gutenberg header, licence and images are removed, so these are not Project Gutenberg editions; transcribers' notes and tables of contents are removed. We thank Project Gutenberg and its volunteers for the transcriptions.
  • Wikisource — books in Russian (ru.wikisource.org), French (fr), Portuguese (pt), Spanish (es), Arabic (ar), Hindi (hi) and German (de), transcribed by Wikisource volunteers.
    Licence: Public domain (the author's text). Wikisource's own additions are CC BY-SA 4.0; we do not use them.
    Our changes: only the author's text is kept — Wikisource templates, editors' notes, footnotes, page numbers and links removed; German, Spanish and Portuguese spelling mechanically updated to later rules.
  • Aozora Bunko (青空文庫) — Japanese books, from the modern-orthography files, with their furigana.
    Licence: Public domain
    Credit: each book ends with Aozora Bunko's colophon (source edition, input and proofreading volunteers) in a chapter “このファイルについて”.
    Our changes: converted to e-book format; the inputter's notes and illustrations removed.

AI-generated content

Explanations, the tutor and book chat, some example sentences, glosses and corrections are generated by Google Gemini and, as a fallback, by open-weight models (gpt-oss) hosted by Groq. AI output is marked in the app, is not reviewed by the dictionaries listed above and may be wrong. Object recognition in “scan to card” uses Gemini; the background is removed on our servers with the U²-Net model through rembg (Apache License 2.0). See the AI & Content Disclaimer.

Estonian course (pilot)

  • EKI ühendsõnastik — Estonian dictionary data for the Estonian course, which is being prepared.
    Licence: CC BY 4.0
    Credit: EKI ühendsõnastik © Eesti Keele Instituut, via Ekilex/Sõnaveeb.
    Our changes: shortened, combined with other data and machine-translated; not reviewed or endorsed by EKI.

Fonts

Self-hosted, all under the SIL Open Font License 1.1: Inter (The Inter Project Authors), Commissioner (The Commissioner Project Authors; instanced to a subset of its axes), Literata (The Literata Project Authors), Noto Sans Devanagari (The Noto Project Authors) and IBM Plex Sans Arabic (IBM Corp.). Flags are our own simplified drawings.

Open-source software

WordBonk runs on open-source software, each under its own licence, among them React, Vite, Tailwind CSS, Lucide icons, FastAPI, SQLAlchemy, PostgreSQL, Redis, NLTK, simplemma, pykakasi (Japanese readings), num2words, word_forms, deep-translator, gTTS, edge-tts, youtube-transcript-api, yt-dlp, PyMuPDF, ONNX Runtime, rembg and FSRS. We thank their authors.