What Are JMdict and KANJIDIC2? Free Japanese Dictionary Data Explained
Hiroki Obara · Developer
The JMdict file available on November 7, 2026 has 219,036 entries. The KANJIDIC2 file from the same day has 13,108 kanji. Both are Japanese dictionary data published by the Electronic Dictionary Research and Development Group (EDRDG), and both are free to download. Compressed, the English-only JMdict is about 10.6 MB and KANJIDIC2 about 1.5 MB.
Unless noted, the numbers below come from counting those files. JMdict is rebuilt every day, so its count keeps moving.
It started in 1991
JMdict grew out of EDICT, a Japanese-English dictionary file. According to the project page, Jim Breen began the project in 1991, when an early DOS Japanese word processor called MOKE (Mark’s Own Kanji Editor) came out with a small version of EDICT inside. The file was expanded and edited over the following years.
By 1999 EDICT had about 60,000 entries. It was converted to an expanded format, and the first XML JMdict file came out in May of that year. Since mid-2006 the file has been generated daily from the source database. The project page credits the growth in entries largely to the many people who contributed entries and took part in editing.
KANJIDIC also began around 1991, as two files that were later merged. Its history section is a long list of contributors. Jeffrey Friedl did a major overhaul in 1992, adding the original frequency rankings among other things. Charles Muller provided Korean readings in 1996. KANJIDIC2 combines this material in XML and adds the kanji from JIS X 0213.
EDRDG was set up by Breen at Monash University in Australia in 2000, mainly to hold the copyright on the files and to receive funds for improving them. At first it was in effect just Breen. It is now independent of Monash and consists of Breen and collaborators around the world. JMdict is kept in an online database under an editorial board, and there is a way to propose corrections and new entries.
Commercial terms changed over time. Before 2000 the files were restricted to non-commercial use. A period of negotiated royalties followed. In February 2003 the files were opened to commercial use as well, on conditions mainly about acknowledgement.
What’s in a JMdict entry
An entry holds kanji headwords, readings and senses. The senses are translation equivalents, not explanations. 勉強 (べんきょう) has four.
| Sense | Part-of-speech tags | Glosses |
|---|---|---|
| 1 | noun, suru verb, transitive | study |
| 2 | noun, suru verb, intransitive | diligence, working hard |
| 3 | noun | experience, knowledge, lesson (for the future) |
| 4 | noun, suru verb, transitive, intransitive | discount, price reduction |
Entries also carry frequency tags. 勉強 has ichi1, news1 and nf03. The ichi tag means the word appears in the Ichimango goi bunruishuu, a vocabulary book from 1998. The news tag means it is in the first 12,000 words of a frequency file Alexandre Girardi compiled from the Mainichi Shimbun newspaper. The nf03 tag places it in the third set of 500 words in that file, ranks 1,001 to 1,500.
The tags that mark common words appear on 22,650 entries, about a tenth of the file. The project page says they “must be seen as only providing a crude indication” of how common a word actually is.
Usage notes are there too. 625 entries have the honorific tag and 1,342 have the onomatopoeic or mimetic tag. I wrote about using those in reading keigo in novels and looking up onomatopoeia.
Every entry has English glosses. The full file adds eight other languages.
| Language | Entries with glosses |
|---|---|
| German | 129,130 |
| Russian | 69,409 |
| Hungarian | 41,901 |
| Dutch | 41,827 |
| Spanish | 34,311 |
| French | 15,336 |
| Swedish | 14,518 |
| Slovenian | 8,776 |
The non-English glosses are matched in from other dictionary files by headword. They are not aligned sense by sense with the English.
What JMdict leaves out
Conjugated forms. Verbs appear in dictionary form. 食べる has an entry, and 食べました and 読んだ don’t. From conjugated form to dictionary form shows how to get back.
Suru verb forms. 勉強する is covered by the 勉強 entry and its suru verb tag.
Romaji. Readings are in kana. The project page says the files use no romanized Japanese, apart from proper names such as “Suzuki” and words like “ikebana” that have been adopted into English.
JLPT levels. JMdict has no JLPT field.
Most proper names. Those have their own EDRDG dataset, JMnedict.
What’s in KANJIDIC2
The 13,108 kanji are the 6,355 in JIS X 0208, the 5,801 in JIS X 0212 and 952 more from JIS X 0213. Here is part of the entry for 読.
| Field | Value |
|---|---|
| On readings | ドク, トク, トウ |
| Kun readings | よ.む, -よ.み |
| Name reading (nanori) | よみ |
| Meanings | read (English), lire (French), leer (Spanish), leia (Portuguese) |
| Grade | 2 |
| Stroke count | 14 |
| Frequency rank | 618 |
| Old JLPT level | 4 |
| Pinyin | du2, dou4 |
| Korean | dog, du |
| Vietnamese | Đậu, Độc |
In a kun reading, the dot marks where okurigana begin and the hyphen marks a prefix or suffix. The file also has index numbers for kanji dictionaries such as Nelson and Heisig, and lookup codes such as SKIP.
A few fields need a note.
Grade. 1 to 6 are the elementary school kanji, 1,026 in total. 8 is the rest of the jōyō kanji, 1,110. 9 and 10 are the jinmeiyō kanji approved for names, 651 and 212.
JLPT. This is the pre-2010 four-level test, not today’s N1 to N5. The project page notes that no official kanji lists exist for the new levels.
Frequency. Only 2,501 kanji have a rank, based on four years of the Mainichi Shimbun. The page warns that this leans toward newspaper vocabulary and that the last few hundred ranks are quite imprecise.
Coverage. 12,356 kanji have Japanese readings and 10,384 have English meanings, so not every character has every field.
For using this kind of information while you read, see Can You Learn Kanji by Reading?
The licence
The terms are in the EDRDG licence statement. The Japanese and English parts of JMdict, and KANJIDIC2, are under Creative Commons Attribution-ShareAlike 4.0 (CC BY-SA 4.0). Nothing needs signing. Using the files means accepting the conditions.
The main ones:
- Acknowledge the use and the source. A smartphone app has to do this on a screen reached from a menu, such as “About” or “Sources”. A mention on the launch page is not enough. There are sample wordings.
- Include the documentation and licence files, or link to them.
- Anything you alter or build on and then distribute goes under the same or a compatible licence, and you can’t claim copyright over the material.
- An app or server needs a procedure for regularly updating to the latest data. For web dictionary servers the example given is at least once a month.
- There is no warranty, and users take on the liability.
Commercial use is not restricted. You can bundle the files with software and sell it at any price, and the software doesn’t have to be open source. A donation is suggested when the files bring you a financial return.
Some parts are handled differently. The non-English glosses in JMdict are under separate copyright held by their compilers. In KANJIDIC2, contributors of material such as the pinyin and Korean readings keep their copyright and allow distribution under these conditions. SKIP codes have their own licence, which the KANJIDIC page gives as a non-commercial Creative Commons one.
Downloads are linked from the JMdict and KANJIDIC pages. To see the data without opening a file, EDRDG runs its own search server, WWWJDIC.
JMdict and KANJIDIC2 in Niho
I make Niho, a free reading app for iPhone and iPad with more than 17,000 Aozora Bunko works and a built-in dictionary. It has no ads and no in-app purchases.
Long-press a word and tap Look up. You get JMdict’s senses grouped by part of speech, with usage notes and a “common” badge. Below that, Kanji breakdown uses KANJIDIC2 to show each kanji’s meanings, on and kun readings, grade and stroke count. The JLPT level on a word comes from Jonathan Waller’s word lists.
The data ships inside the app, so lookups work offline. Words not in the dictionary can be checked with Jisho. The credits are under About in Settings. Getting Started with Niho walks through your first book.
