📚 Where Shirabe gets its data

Shirabe stands on decades of open-source lexicography. Every entry, reading, and tag here originates from the projects below — please visit them, support them, and check their licences before redistributing.

Source versions#

Every dataset we currently have loaded, with its origin and the exact upstream build — re-importing a source updates its row here, so this stays an accurate reference. Licences and fuller attribution follow below.

Source	Origin	Licence	Version	Data date
JMdict — word entries	jmdict-simplified (EDRDG)	EDRDG	3.6.2	2026-06-22
JMnedict — names	jmdict-simplified (EDRDG)	EDRDG	3.6.2	2026-06-22
KANJIDIC2 — kanji	jmdict-simplified (EDRDG)	EDRDG	3.6.2	2026-06-22
KRADFILE — kanji → components	jmdict-simplified (EDRDG)	CC BY-SA 4.0	—	2026-06-29
RADKFILE — radical → kanji	jmdict-simplified (EDRDG)	CC BY-SA 4.0	3.6.2	2026-06-29
Example sentences (Tatoeba)	jmdict-simplified · Tatoeba	CC BY 2.0 FR	3.6.2	2026-06-22
Kanken levels (漢検)	mimneko/kanji-data	CC0 1.0	—	2026-06-26
Jōyō table (常用漢字表 本表)	mimneko/kanji-data	CC0 1.0	—	2026-02-10
Jōyō appendix (付表)	mimneko/kanji-data	CC0 1.0	—	2026-02-09
Radical names — Kanji alive	kanjialive	CC BY 4.0	—	2021-02-24
Pitch accents — UniDic	UniDic (NINJAL)	GPL/LGPL/BSD	202512	2025-12-31
Word frequency — jiten-global	jiten.moe	—	—	2026-07-12
Wikipedia abstracts	DBpedia	CC BY-SA 3.0	2016-10	2026-06-28
BabelStone IDS — kanji structure	BabelStone (Andrew West)	Public domain	—	2025-06-27
Stroke order — KanjiVG	KanjiVG (Ulrich Apel)	CC BY-SA 3.0	—	2025-08-16

Dictionary data#

JMdict — Japanese↔multilingual word entries

Compiled by the Electronic Dictionary Research and Development Group (EDRDG, James Breen et al.) and distributed under the EDRDG licence. We use the jmdict-simplified JSON conversion by Stanislav Petrov.

JMnedict — proper-noun (names) dictionary

Also from EDRDG, same licence terms; consumed via the jmdict-simplified JMnedict release.

KANJIDIC2 — kanji dictionary

EDRDG, same licence; sourced from jmdict-simplified's KANJIDIC2 all release, so kanji meanings come in English, French, Spanish and Portuguese — surfaced per language alongside the JMdict glosses.

JLPT levels (N5–N1)

The JLPT level shown on each kanji and word comes from Jonathan Waller's JLPT Resources (tanos.co.uk), licensed CC BY. Since the post-2010 JLPT publishes no official kanji or vocabulary list, these are the de-facto-standard unofficial N5–N1 lists (kanji via the KANJIDIC snapshot, words via Bluskyo/JLPT_Vocabulary).

KRADFILE — kanji component breakdown

The “parts” a kanji is made of, shown on each kanji page, come from EDRDG's KRADFILE / KRADFILE2 (Michael Raine, James Breen et al.), CC BY-SA 4.0, via the same jmdict-simplified JSON build (the inverse of the RADKFILE below).

BabelStone IDS — kanji structure

The positional Ideographic Description Sequences (⿰⿱⿴…) that show how a kanji is laid out come from Andrew West's BabelStone IDS data, released to the public domain.

Kanken (漢字検定) levels

The 配当 (assigned) Kanji Kentei level shown on each kanji comes from mimneko/kanji-data's 漢検漢字辞典 table, released under CC0 1.0.

Jōyō kanji table (常用漢字表) — official readings & examples

The 表内 / 表外 reading distinction and the example words shown beside each on/kun reading come from the 文化庁's 2010 常用漢字表, digitised by mimneko/kanji-data under CC0 1.0. The underlying table is a Japanese cabinet notification (内閣告示), which is not subject to copyright.

成り立ち — kanji formation (六書) & glyph origin

The formation type (象形・指事・会意・形声) and the Japanese glyph-origin (字源) notes come from ウィクショナリー日本語版 (Japanese Wiktionary)'s 字源 sections, used under CC BY-SA 4.0; shinjitai inherit their kyūjitai's 字源 (e.g. 国→國). A few jōyō gaps carry an AI-assigned classification, marked on the kanji page. Note: ja.wiktionary follows modern academic scholarship, which can differ from school-textbook classifications.

RADKFILE — search radicals → kanji

The 253 search radicals in the radical picker and the kanji built from each come from EDRDG's RADKFILE / RADKFILE2 (Michael Raine, James Breen et al.), licensed CC BY-SA 4.0; consumed via the jmdict-simplified JSON conversion.

Kangxi radical names & meanings

The Japanese reading, English gloss, stroke count, and positional category used to label the radicals are sourced from Kanji alive (kanjialive.com), licensed under CC BY 4.0.

Wikipedia abstracts (via DBpedia)

Lead-paragraph summaries of Wikipedia articles in every supported language come from the DBpedia project's long_abstracts dataset. The text remains the property of the Wikipedia contributors who wrote it and is dual-licensed under CC BY-SA 3.0 and the GNU Free Documentation License. Each abstract card on a word page links back to the source article.

Example sentences — Tatoeba

The example sentences attached to word senses come from the Tatoeba project (the JMdict examples set), licensed CC BY 2.0 FR, consumed via jmdict-simplified.

Pitch accents — UniDic

The pitch-accent (高低アクセント) data comes from the UniDic morphological dictionary by the National Institute for Japanese Language and Linguistics (NINJAL), triple-licensed GPL 2.0 / LGPL 2.1 / Modified BSD — free for commercial use. We read the accent (aType) from its lexicon.

Word frequency — jiten.moe

Rank ordering for the “frequency” word sort comes from the global frequency list published by jiten.moe.

Stroke order — KanjiVG

The animated stroke-order diagrams use KanjiVG by Ulrich Apel, licensed CC BY-SA 3.0.

Pronunciation audio — AivisSpeech:るな, AivisSpeech:TANAKA

The spoken pronunciations for words and example sentences are synthesized with AivisSpeech, an open-source Japanese text-to-speech engine, using the voices of るな (female) and TANAKA (male) under the Aivis Common Model License (commercial use permitted). Where no clip is available the browser's built-in speech synthesis fills in.

Software#

Shirabe is built on Ruby on Rails, Hotwire, and a stack of other open-source gems. The Japanese-language tooling is its own small ecosystem, much of it maintained alongside Shirabe:

kabosu — tokenisation

Ruby bindings for the Sudachi morphological analyser by Works Applications, with its SudachiDict full edition (the most complete — extra named entities and rare words) — this is how Shirabe splits and reads Japanese text. github.com/davafons/kabosu, Apache-2.0.

daidai — conjugation

Pure-Ruby Japanese verb and adjective conjugation, used to derive and explain inflected forms — github.com/davafons/daidai.

Typefaces#

Typeset in Inter Tight, Newsreader, Noto Sans JP, and JetBrains Mono.

Found something missing? Let us know.