Skip to main content

Newtext analysis now works with AI —try it →

Sources

Sources and licenses

The Smart Vocabulary dictionary is a database reworked and edited from the text of English Wiktionary.

Each source, its authors, its license and the changes we made are listed below.

Where it comes from

  1. Word senses and transcription

    kaikki.org machine-readable English dictionary (Wiktextract data from English Wiktionary)

    The authors of the text are the English Wiktionary contributors.

    Tatu Ylonen brought the data into machine-readable form (Wiktextract). He is the author of the extraction program, not of the dictionary text — under the license, attribution as author goes to the Wiktionary contributors.

    LicenseCC BY-SA 4.0 (opens in a new tab)

    • Edits made before 7 June 2023 remain under CC BY-SA 3.0; the copy we took may contain edits of both kinds.
    • Under the Wikimedia terms this text is offered under the GFDL alongside CC BY-SA. We rely on CC BY-SA only and do not offer the GFDL on our own behalf.
    • The data has been modified — the full list is below, in the “What was changed” section.
    • The dictionary text we distribute also remains under CC BY-SA 4.0.
    • Word pages do not link to the original articles; the credit above names the authors of the whole dictionary text.
  2. CEFR levels (A1–B2)

    The CEFR-J Wordlist Version 1.5

    The CEFR-J Wordlist Version 1.5. Compiled by Yukio Tono, Tokyo University of Foreign Studies. Retrieved from http://www.cefr-j.org/download.html on 1/20/2020.

    Copyright: Tono Laboratory, Tokyo University of Foreign Studies (TUFS)

    LicenseNot Creative Commons — TonoLab/TUFS’s own terms (opens in a new tab)

    • This source is not under Creative Commons: the CC BY-SA terms on this page do not apply to it.
    • The terms permit free use for research and commercial purposes but require correct citation. The citation above is the form the publisher itself asks for.
    • Only the level labels were taken from the list and merged into the app’s own database; the original file is not distributed with the app.
    • Our copy was taken from the mirror repository below, and the citation is given in the form that repository states. The publisher’s own page distributes version 1.6 today.
    • TonoLab and TUFS have not reviewed, supervised or endorsed this app.
  3. Pronunciation recordings

    Wikimedia Commons

    Wikimedia Commons is the repository that holds the recordings, not their author. Each recording has its own author and its own license: CC BY-SA, CC BY, CC0 or public domain.

    LicenseSeparate for each file

    • There is no single license covering the whole collection, so no license name is given here on purpose: naming none is more honest than naming the wrong one.
    • The recordings are streamed from Commons unchanged: the app does not trim them, re-encode them or store them.
    • Each recording’s author and license are stated on its file page on Wikimedia Commons.
    • The app itself does not show the author or the license of each recording.
    • Wikimedia Commons, the Wikimedia Foundation and the authors of the recordings do not support or endorse this app.
  4. Computer voice

    Kokoro-82M

    The author of the model is hexgrad.

    LicenseApache License 2.0 (opens in a new tab)

    • Words and sentences that have no human recording are read by this synthetic voice. It is not a person’s recording.
    • The audio was generated in advance with the model; the model itself is not distributed with the app.
    • The model chooses the pronunciation itself, so for a word that can be said in two ways (record, read) the voice may not match the transcription on the word page.
  5. Topic groupings (derived, modified)

    Open English WordNet 2025

    The authors of the data are the Open English WordNet contributors.

    LicenseCC BY 4.0 (opens in a new tab)

    • The data has been modified: no WordNet text is used in the app. From its links between words (for example, a shirt is a kind of garment) we worked out which of 24 topics each of our words belongs to; only that grouping is stored.
    • The definitions in the topic lists come from English Wiktionary (the first entry above), not from WordNet.
    • The Open English WordNet authors do not support or endorse this app.

What was changed

  1. We modified the Wiktionary text. Before that, Wiktextract (kaikki.org) brought it into machine-readable form — that is a modification too, and it is recorded here.
  2. 29,557 words were selected from the English entries. Only nouns, verbs, adjectives and adverbs were kept; at most 4 senses were taken per word — 74,594 senses in total. Wiki markup was cleaned out and everything was moved into the app’s own schema.
  3. CEFR levels were taken from the other two sources and merged onto these words. The addresses of the pronunciation recordings are stored, not the files themselves.
  4. Words and dictation sentences that have no human recording were read aloud in advance with the Kokoro-82M model, and the audio is stored on our server. A human recording is always played first.
  5. For the topic dictionary, our words were grouped into 24 topics using the links between words in Open English WordNet 2025. Only the grouping (word, part of speech, topic) is stored, not WordNet’s own text.

Our license

The Smart Vocabulary dictionary database derived from the sources above is offered under CC BY-SA 4.0. CC BY-SA 4.0 (opens in a new tab)

  • The A1–B2 levels taken from CEFR-J are excluded: they are stored in a separate field and are governed by the TonoLab/TUFS terms — we cannot re-license them under CC BY-SA.
  • We do not place any additional terms on copying or reusing the dictionary text in the app.

Warranty and independence

  • All the source data listed above is provided “as is”, without warranties of any kind (CC BY-SA, Section 5).
  • The sources, their authors and their organizations do not endorse or sponsor Smart Vocabulary and are not affiliated with it.

Website

Fonts

The site uses your device's system fonts (San Francisco, Segoe UI, Roboto and others): no font file is downloaded and no request goes to an outside service.

Icons

Lucide — ISC license. The open book in the logo is Material Icons “auto_stories”, Apache License 2.0 (the same as in the Android app).

Images

The isometric picture on the home page (books and word cards) was drawn for Smart Vocabulary; it is not taken from another site.

29,557 words · 74,594 senses

Source copy: enwiktionary dump, 2026-09-02

Page revised: 23.10.2026