Tomoshi Japanese Dictionary — Open Data Layer
The open data layer of Tomoshi, a Japanese dictionary for Chinese- and English-speaking learners, developed by Y1Z (individual developer). Mirror of the GitHub releases at tomoshi-app/tomoshi-dict-data; this page, the GitHub release and the Zenodo archive (https://doi.org/10.5281/zenodo.22672054) carry the same file and the same SHA-256.
This dataset publishes every table of the Tomoshi dictionary database that is derived from share-alike-licensed sources (JMdict, KANJIDIC2, Japanese Wiktionary, KanjiVG and others), together with Tomoshi's own additions built on top of them: complete Chinese gloss translations, kanji-variant usage notes, zh/ja contrast notes, English→Japanese word-choice cards, synonym/register analysis cards and transitive/intransitive pair cards — in Simplified Chinese, Traditional Chinese and English.
这是 Tomoshi 日语词典的开放数据层：所有由 share-alike 许可来源（JMdict 等） 衍生的表，连同构建于其上的中文释义、写法辨析、译法辨析、近义辨析、自他配对 等内容（简体、繁体、英文），按下述许可开放提取与再分发。
Files
| File | Contents |
|---|---|
tomoshi-dict-open.db.zst |
The whole dataset as one zstd-compressed SQLite database (86 MB compressed, 650 MB raw). Decompress with zstd -d tomoshi-dict-open.db.zst or any zstd implementation |
SHA256SUMS |
Checksums of the compressed and the decompressed file |
LICENSE.md, NOTICE.md |
Per-table licensing summary and attribution notices |
Current version: 2026-09-02 (matches the data pack shipped in the Tomoshi app on that date; export format v2). Versions are tagged the same way on GitHub.
import sqlite3, zstandard
with open("tomoshi-dict-open.db.zst", "rb") as f, open("tomoshi-dict-open.db", "wb") as out:
zstandard.ZstdDecompressor().copy_stream(f, out)
db = sqlite3.connect("tomoshi-dict-open.db")
print(db.execute("SELECT table_name, license FROM table_licenses").fetchall())
print(db.execute("SELECT cn, summary FROM cn_contrast_en WHERE cn = 'abandon'").fetchone())
What's inside
| Table | Contents |
|---|---|
entries, forms |
JMdict entries (~217k) and surface-form index. Entry JSON keeps glosses, POS, priority, JLPT level, variant notes; example sentences are stripped (see below) |
jpn_defs |
Japanese definitions: Japanese Wiktionary (via kaikki.org Wiktextract) + LLM-generated definitions derived from JMdict glosses |
zh_defs / zh_defs_zhtw |
Chinese translations of JMdict glosses (~217k, Simplified / Traditional). No _en variant: the English glosses are the JMdict originals inside entries |
zh_variant_note (+_zhtw, _en) |
Kanji-variant usage notes (写法辨析): which spelling of a word to use when |
cn_contrast (+_zhtw) |
zh/ja false-friend contrast notes (译法辨析), keyed by Chinese word |
cn_contrast_en |
English→Japanese word-choice cards (~9.9k): for an English word with several Japanese renderings, which one to pick and why. Keyed by English word; same schema as cn_contrast |
word_relations_analysis (+_zhtw, _en) |
Synonym/register analysis cards (近义・语体辨析) |
verb_pairs, verb_pairs_note (+_zhtw, _en) |
Transitive/intransitive verb pairs and explanation cards (自他配对) |
kanji, kanji_gloss (+_zhtw) |
KANJIDIC2 kanji data + Chinese kanji glosses derived from it. No kanji_gloss_en: English meanings are the KANJIDIC2 originals inside kanji |
kanji_strokes |
Stroke-order data from KanjiVG |
vocab_jlpt |
Community JLPT level estimates (not official data) |
freq_rank, kanji_words |
Frequency ranking and kanji→word mapping derived from JMdict priority markers |
meta, table_licenses |
Provenance and per-table licensing metadata |
Language suffixes: no suffix = Simplified Chinese, _zhtw = Traditional
Chinese, _en = English. The English tables keep their source names
(cn_contrast_en, zh_variant_note_en) so the schema stays identical across
languages.
Not included (proprietary or third-party layers of the app, unrelated to the share-alike obligation): the reverse zh→ja and en→ja lookup indexes, pitch-accent data with its human verification layer, grammar points and their notes, conversational frequency ranking, WordNet-derived relation links, and example sentences (Tatoeba text plus Tomoshi's Chinese translations — stripped from the entry JSON here).
How the derived content was made
The derived layers (Japanese definitions, Chinese gloss translations, variant/contrast/synonym/transitivity notes, English→Japanese word-choice cards) were generated with LLM assistance using JMdict/KANJIDIC2 gloss text as model input, then reviewed by a separate model. They are not hand-verified; errors are best reported through the Tomoshi app's one-tap report feature, or to support@tomoshi.app.
Licensing
This dataset is not single-license. Each table's license and required
attribution are recorded in the table_licenses table inside the database and
summarized in LICENSE.md:
- Most tables: CC BY-SA 4.0 (JMdict/KANJIDIC2 lineage, © EDRDG; Tomoshi's derived additions © Y1Z, same license).
kanji_strokes: CC BY-SA 3.0 (KanjiVG, © Ulrich Apel).
Full attribution notices are in NOTICE.md. If you reuse this data:
- Attribute the EDRDG for JMdict/KANJIDIC2-derived content per their licence statement, and credit "Tomoshi (Y1Z)" for the derived Chinese- and English-language layers;
- License your derivative of BY-SA tables under the same (or a compatible) license, and state your modifications;
- The name "Tomoshi" and its logo are trademarks/brand identifiers and are not licensed by this dataset — don't use them in a way that suggests your product is Tomoshi or endorsed by it.
Modifications statement (CC BY-SA §3(a)(1)(B))
Relative to the upstream sources, this dataset: restructures JMdict/KANJIDIC2 XML into SQLite; adds LLM-assisted derived content (Japanese definitions, Chinese gloss translations, variant/contrast/synonym/transitivity notes in Chinese and English, English→Japanese word-choice cards — generated with JMdict/KANJIDIC2 gloss text as model input, then reviewed by a separate model); derives frequency rankings from JMdict priority markers; and removes example sentences.
Citation
@misc{tomoshi_dict_data,
author = {Y1Z},
title = {Tomoshi Japanese Dictionary — Open Data Layer},
year = {2026},
howpublished = {\url{https://github.com/tomoshi-app/tomoshi-dict-data}},
doi = {10.5281/zenodo.22672054},
note = {Version 2026-09-02. Derived from JMdict/KANJIDIC2 (EDRDG), Japanese Wiktionary and KanjiVG; CC BY-SA}
}
Sustainability pledge
If Tomoshi is ever permanently discontinued, the proprietary curated data layer excluded from this dataset will be re-released under a Creative Commons license, so work built on Tomoshi data is never stranded.
- Downloads last month
- 20