This repository is a mirror of the SudachiDict raw lexicon files, together with the supporting files needed to build them as a Lindera dictionary. It exists to pin the exact inputs of Lindera dictionary builds for reproducibility and availability, following the same convention as lindera/unidic-mecab and lindera/mecab-ipadic.
Current version: 20260723 (upstream tag v20260723).
| File | Origin | License |
|---|---|---|
small_lex.csv |
SudachiDict raw distribution | Apache-2.0 (UniDic-derived, see LEGAL) |
core_lex.csv |
SudachiDict raw distribution | Apache-2.0 (NEologd-derived, see LEGAL) |
notcore_lex.csv |
SudachiDict raw distribution | Apache-2.0 (NEologd-derived, see LEGAL) |
matrix.def |
SudachiDict raw distribution (byte-identical to the UniDic 2.1.2 matrix) | BSD/GPL/LGPL triple license (see COPYING) |
char.def |
unidic-mecab 2.1.2 source archive | BSD/GPL/LGPL triple license (see COPYING) |
unk.def |
unidic-mecab 2.1.2 source archive, modified (see below) | BSD/GPL/LGPL triple license (see COPYING) |
LEGAL, LICENSE-2.0.txt |
SudachiDict upstream (v20260723) |
— |
BSD, COPYING, GPL, LGPL |
unidic-mecab 2.1.2 source archive | — |
Large files (*.csv, matrix.def) are tracked with Git LFS.
Retrieved on 2026-08-25.
SudachiDict raw lexicons and connection matrix, from the official
distribution bucket
(https://sudachi.s3.ap-northeast-1.amazonaws.com/sudachidict-raw/):
| Archive | SHA-256 |
|---|---|
20260723/small_lex.zip |
b578ac9545899783d5d7e30d5d78d5d9dcf40b36965d4af0663ec2eb041c1093 |
20260723/core_lex.zip |
a2b39e1572adab08a649b1390b134517adc55f1d733c59358b113298788bf31c |
20260723/notcore_lex.zip |
a15edc5193b42acfebdc220825d8c7edf0059bf950169d09bcdf6c426a81ecd1 |
matrix.def.zip |
3ccaed582481b31be170c7200cf1cb2856f2d14c2fa601b3692d97dbad5a3bd1 |
char.def and unk.def are taken from the unidic-mecab 2.1.2 source archive
(https://lindera.dev/unidic-mecab-2.1.2.tar.gz, mirrored at
lindera/unidic-mecab), because
SudachiDict entries use UniDic 2.1.2 context IDs and character categories.
Checksums of the files as committed:
| File | SHA-256 |
|---|---|
small_lex.csv |
6cc54808a2359b48fe15105814ca3352dceb4c25305f6956b7a104f7f6611f00 |
core_lex.csv |
3a3c4b68cf0b64893dff6cc14ba690d79a99fefe1564dd751a418317db2e148d |
notcore_lex.csv |
6237df5751fea79e638dba54d088520ec3e2386209971e737c6ecc5d69f24950 |
matrix.def |
b20e0da1ed5472e0be2e5cb3c716672fd788e1701e08d54624b58ba52d73d927 |
char.def |
0790db12ab4f4d742b39b1737b9cce2a00c92220139a5bea9f0e8874898726e5 |
unk.def |
97fec17d4f477ee94cc25f1a1c318cec8ea4688491581c069d61677e9e2ad976 |
All files are unmodified from their upstream distributions, with one
exception: unk.def has a placeholder display-surface column inserted after
the four reserved columns, so that unknown words expose their morphological
details at the same column indices as SudachiDict's 19-column lexicon rows
(SudachiDict rows carry the display surface at column 4, 0-based; UniDic's
unk.def has its part-of-speech there).
The exact transformation applied to the original unidic-mecab 2.1.2
unk.def:
awk -F, 'BEGIN{OFS=","} {out=$1 OFS $2 OFS $3 OFS $4 OFS "*"; for(i=5;i<=NF;i++) out=out OFS $i; print out}' unk.defSee the SudachiDict recipe in the Lindera documentation: https://lindera.github.io/docs/sudachidict.html
- The SudachiDict lexicon files are distributed under the Apache License 2.0
(
LICENSE-2.0.txt). SeeLEGALfor the full attribution of the UniDic and NEologd data they incorporate. matrix.def,char.def, andunk.defare part of UniDic (unidic-mecab 2.1.2), which is triple-licensed under BSD/GPL/LGPL (BSD,GPL,LGPL; seeCOPYING).
This repository is not affiliated with or endorsed by Works Applications Co., Ltd. "Sudachi" and "SudachiDict" are the work of the WAP Tokushima Laboratory of AI and NLP.