Need private packages and team management tools?Check out npm Orgs. »


3.1.0 • Public • Published


This repository is a package for an npm/Node.js containing the data in Y. Tono, M. Yamazaki, and K. Maekawa, A Frequency Dictionary of Japanese (Routledge, 2013; publisher, Google Books), containing the top 5000 words in Japanese.

It contains a Node.js module which imports a JSON file tono.json which is generated from tono.tsv (a tab-separated file) by the tsv2json.js script.

The raw dictionary data is the property of Routledge and we beg that our use of this data be considered fair use. The small bits of code in this repository is released into the public domain.


$ npm install --save git://

Then, in JavaScript, var tono = require('tono-yamazaki-maekawa') will contain a 5000-long array of objects.

Meaning of TSV columns

Readings Hiragana or katakana or roumaji (English, there’s a few of these!) characters: how to pronounce the word. Multiple variants separated by /.

English Tono et al.’s notes in English on meaning, including parts of speech and glosses.

Kanji Optional: written form of the word using kanji (this is what’s provided by Tono, et al.). Multiple variants separated by /. Sometimes, one kanji may have multiple readings (same word), or the same reading can be produced by different kanji (also same word); for these cases, each written variant is separated by / and has furigana encoded in square-brackets like [kanji;reading]. Example: 大分 is read daibu and daibun, so it is listed as [大;だい][分;ぶ]/[大;だい][分;ぶん]. If there is no such ambiguity, i.e., one kanji, one reading, furigana is not given.

Roumaji Tono, et al.’s romanized pronunciation (the Readings column was automatically converted from this, so please notify me of errors).

Tono # Number between 1 and 5000.

Freq. per mil. Normalized frequency (occurrences per million words)

Dispersion Number between 0 and 1 indicating how universal the word is. Two words can have near-equal frequency of occurrence but have very different dispersions if one occurs in nearly all texts (dispersion close to 1) while another does not (lower than 1). Example: 事こと has normalized frequency 8747 and dispersion 0.96. Compare to えー (or ええ), an interjection meaning “whaaat?”, with normalized frequency 8636 and dispersion 0.07. Averaged over millions of documents, both occur approximately the same number of times but えー, a spoken word, is frequently omitted from many documents—it has low dispersion.

Register star Optional. The top fifty words in five registers are marked in this column, specifically:

  • BK: books
  • WB: web
  • OF: official documents
  • NM: newspapers and magazines
  • SP: spoken (only forty-seven, not fifty, words marked for this register—due to errors?)

JavaScript/JSON fields

“Readings” → readings

“English” → meaning

“Kanji” → kanjis

“Roumaji” → roumaji

“Tono #” → num

“Freq. per mil.” → freq

“Dispersion” → disp

“Register star” → register




npm i tono-yamazaki-maekawa

Downloadsweekly downloads






last publish


  • avatar
Report a vulnerability