Goal: create a version of JMdict and JMnedict in JSON format with more comprehensible structure.
Principles:
- More regular file structure
- More clarity (mostly in naming conventions)
- Less custom things
Grab results here: Latest release
Note: There are two versions of the JMdict dictionary: full and "common"-only. Dictionary entries are considered common if /k_ele/ke_pri
or /r_ele/re_pri
elements in original file contain one of these markers: "news1", "ichi1", "spec1", "spec2", "gai1". Common-only distributions are much smaller. JMnedict has only one version.
Requirements: Bash and Zorba
Usage examples:
> ./build.sh help -- show this help message
> ./build.sh download -- download source dictionary files
> ./build.sh convert jmdict 1.2.3 -- convert only JMdict, version 1.2.3
> ./build.sh convert jmnedict 1.2.3 -- convert only JMnedict, version 1.2.3
> ./build.sh convert all 1.2.3 -- convert all, version 1.2.3
> ./build.sh archive -- create distribution archives
Workflow:
- Run
./build.sh download
to download source XML files into a build directory. - Run
./build.sh convert all 1.2.3
(use proper version, you can add modifiers like-dev
) to convert XML files into JSON versions. - Run
./build.sh archive
to create distribution archives once you're done. Make sure you set a proper version.
-
xref
(array of numbers/strings) := one of the following structures:- [kanji, kana, sense_index] - full format, which lists both kanji (or other non-kana characters) and kana writings which this sense applies to, as well as index of a sense (counting from 1)
- [kanji/kana, sense_index] - just one writing (kanji or kana) and sense index
- [kanji/kana] - just one writing (kanji or kana)
Examples:
["丸", "まる", 1]
,["○", "まる", 1]
,["二重丸", "にじゅうまる"]
,["漢数字"]
, etc. -
tag
(string) := all tags are listed in a separate section of the file, see the description of root JSON object
version
(string) := Semantic version of this projectdictDate
(string) := Creation date of JMdict file, as it appears in a comment with format "JMdict created: YYYY-MM-DD" in the original XML file headerdictRevisions
(array of string) := Revisions of JMdict file, as they appear in comments before DOCTYPE in the original XML file header. These only contain actual version (e.g., "1.08"), not a full comment. Original comments also mention changes made, but this is omitted in the resulting JSON filestags
(object) := all tags: parts of speech, names of dialects, fields of application, etc. All those things are expressed as XML entities in the original file. Keys of this objects are tags per se, values are descriptions, slightly modified from the original filewords
(array of objects) :=id
(number) := unique identifierkanji
(array of objects) := kanji (and other non-kana) writingscommon
(boolean) :=true
if this particular spelling is common. This field combines all the*_pri
fields from original files in a same way as [jisho.org][] and other on-line dictionaries do ("common" word markers). It gets rid of bunch of*_pri
fields which are not typically used. Words marked with "news1", "ichi1", "spec1", "spec2", "gai1" in the original file are treated as common, which may or may not be true according other sources.text
(string) := any non-kana-only writing, may contain kanji, kana, and some other characterstags
(array of tags) := tags applied to this writing
kana
(array of objects) := kana-only (with few exceptions) writings, typically considered as "readings", but these can be a word writings by themselvescommon
(boolean) := same as for kanji elementstext
(string) := kana-only writing, may only accidentally contain middle-dot and other punctuation-like characterstags
(array of tags) := same as for kanjiappliesToKanji
(array of strings) := list of kanji writings within this word which this kana version applies to."*"
means "all", empty array means "none"
sense
(array of objects) := senses (translations + some related data) for this wordspartOfSpeech
(array of tags) := all parts of speech for this sense. Unlike the original dictionary file, this field is never empty/missing. In the original file, part-of-speech from earlier sense elements may apply to following elements, in which case latter don't have defined part-of-speechappliesToKanji
(array of strings) := list of kanji writings within this word which this sense applies to."*"
means "all", empty array means "none"appliesToKana
(array of strings) := list of kana writings within this word which this sense applies to."*"
means "all", empty array means "none"related
(array of xrefs) := xrefs to related wordsantonym
(array of xrefs) := xrefs to antonyms of this wordfield
(array of tags) := fields of applicationdialect
(array of tags) := dialectsmisc
(array of tags) := other related tagsinfo
(array of strings) := other infolanguageSource
(array of objects) := source language info for borrowed words and wasei-eigolang
(string) := language code from the ISO 639-2 standardfull
(boolean) := indicates whether the sense element fully or partially describes the source word or phrase of the loanwordwasei
(boolean) := indicates that the Japanese word has been constructed from words in the source language, and not from an actual phrase in that language. See Wasei-eigotext
(string or null) := text in the language defined by alang
element, ornull
gloss
(array of objects) := translationslang
(string) := language code from the ISO 639-2 standardtext
(string) := a word or phrase
Notes:
- All fields in all objects are always present, none ever omitted
- Array fields are never
null
, only empty - The only place which allows
null
values issense->languageSource->text
field in word element
Same as for JMdict
version
(string) := Semantic version of this projectdictDate
(string) := Creation date of JMnedict file, as it appears in a comment with format "JMnedict created: YYYY-MM-DD" in the original XML file headerdictRevisions
(array of string) := Revisions of JMnedict file, as they appear in comments before DOCTYPE in the original XML file header. These only contain actual version (e.g., "1.08"), not a full comment. Original comments also mention changes made, but this is omitted in the resulting JSON filestags
(object) := all tags: parts of speech, names of dialects, fields of application, etc. All those things are expressed as XML entities in the original file. Keys of this objects are tags per se, values are descriptions, slightly modified from the original filewords
(array of objects) :=id
(number) := unique identifierkanji
(array of objects) := kanji (and other non-kana) writingstext
(string) := any non-kana-only writing, may contain kanji, kana, and some other characterstags
(array of tags) := tags applied to this writing
kana
(array of objects) := kana-only (with few exceptions) writings, typically considered as "readings", but these can be a word writings by themselvestext
(string) := kana-only writing, may only accidentally contain middle-dot and other punctuation-like characterstags
(array of tags) := same as for kanjiappliesToKanji
(array of strings) := list of kanji writings within this word which this kana version applies to."*"
means "all", empty array means "none"
translation
(array of objects) := translations + some related data for this wordstype
(array of tags) := name types, as specified in tagsrelated
(array of xrefs) := xrefs to related wordstranslation
(array of objects) := actual translationslang
(string) := language code from the ISO 639-2 standardtext
(string) := a word or phrase
Differences from JMdict format:
kanji
andkana
have nocommon
flag because in this dictionary priority data is missing (ke_pri
andre_pri
fields)translation
instead ofgloss
translation->translation->lang
seems to be always empty because the original XML file has no data in corresponding attributes, even though documentation says otherwise
Original XML files, JMdict_e.xml and JMnedict.xml are property of the Electronic Dictionary Research and Development Group, and are used in conformance with the Group's licence. Project started in 1991 by Jim Breen.
All derived files are distributed under the same license, as the original license requires it.
Source files of this project (excluding distribution archives containing JSON files) are available under Creative Commons Attribution-ShareAlike License v4.0. See LICENSE.txt