# Third-party notices

The generated dictionaries combine data sources. The Kotlin program is MIT, but the
generated dictionary data keeps the license and notice requirements of its
upstream data sources. Do not treat the complete release archive as a single
MIT-licensed work.

Repository license scope is described in [LICENSE-DATA.md](../../../LICENSE-DATA.md).
The upstream documents in `third-party/LICENSE-DATA.md` and `third-party/NOTICE.md`
are preserved JapaneseCorpus notices: references to "this repository" in those
documents describe JapaneseCorpus, not this converter project.

## Project source code

The Kotlin conversion program source files are distributed
under the MIT License. The complete project license is in the repository
[LICENSE](../../../LICENSE).

## JapaneseCorpus / JMdict English data

The hiragana-to-English source combines the JapaneseCorpus direct-loanword
asset `mozc-english-reading-unigram-00000.txt.zst` with the curated full-reading
pairs in `src/main/english/reading-supplement.tsv`. The CI resolves one latest
published [JapaneseCorpus release](https://github.com/KazumaProject/JapaneseCorpus/releases)
and records its tag and dictionary checksum; all downloaded inputs come from
that same release. The manifest records the CMUdict pronunciation source.

The initial supplement was checked against JMdict_e-20261012 from
[JapaneseCorpus v2026.1012.19](https://github.com/KazumaProject/JapaneseCorpus/releases/tag/v2026.1012.19),
SHA-256 `5004ad9c777db2188baaa51554d4bb3e1ef3781725c4b38467723886b5d47d9e`.
Each supplemental row records its JMdict entry ID or official reference URL
and the reason for accepting the complete reading/surface pair.

JMdict is copyright James William BREEN and the Electronic Dictionary Research
and Development Group. The JMdict-derived source, curated supplement, and
adapted database data are licensed under
[CC BY-SA 4.0](https://creativecommons.org/licenses/by-sa/4.0/).
The authoritative attribution and license notice is preserved in
[`third-party/JMDICT-LICENSE.html`](third-party/JMDICT-LICENSE.html),
with current [EDRDG terms](https://www.edrdg.org/edrdg/licence.html).
This data license is separate from the MIT-licensed Kotlin program.

Modifications include selecting direct English loanword pairs, converting
Katakana readings to hiragana, adding individually checked full-reading pairs
and attested variants, and applying the defensive quality filter documented in
[`docs/english-dictionary.md`](../../../docs/english-dictionary.md).
Only `primary` candidates are emitted. Source explanations and unapproved
compounds remain in audit reports rather than being trimmed automatically.
Redistribution of the JMdict-derived portion must preserve attribution and
ShareAlike requirements. The binary-only assets ZIP does not include these
notices; downstream distributions must provide the required source and license
information, for example through the application's source/license screen.

## NGSL and CMUdict general English supplement

General English headwords were prioritized using the New General Service List
(NGSL) 1.2 by Charles Browne, Brent Culligan, and Joseph Phillips, licensed under
[CC BY-SA 4.0](https://creativecommons.org/licenses/by-sa/4.0/).
The [official NGSL site](https://www.newgeneralservicelist.com/new-general-service-list)
provides the frequency list. The raw list and frequency statistics are not
included in the binary assets; its version and checksum are recorded in the
dictionary creation documentation. Frequency rankings are used for selection,
not as measured IME candidate costs.

Some supplemental kana readings are editorial adaptations of full US English
pronunciations from [CMUdict](https://github.com/cmusphinx/cmudict), commit
`74790861f652b15e4ac49015a90074ad62a27690`.
CMUdict is Copyright (C) 1993-2015 Carnegie Mellon University. All rights reserved.
Its redistribution conditions and disclaimer are preserved verbatim in
[`third-party/CMUDICT-LICENSE`](third-party/CMUDICT-LICENSE); source and binary
redistributions of the derived material must reproduce the required notice.
Rows distinguish these reviewed pronunciation-derived readings from readings
attested in Japanese sources, and record the original ARPAbet pronunciation.
The curated combined supplement remains distributed under CC BY-SA 4.0, with
the CMUdict notice retained for its portion. The Kotlin code remains MIT.
The CC BY-SA [legal code](https://creativecommons.org/licenses/by-sa/4.0/legalcode.en)
applies to the project-owned editorial contributions and the relevant adapted
data; this does not replace the original CMUdict conditions or claim ownership
of the upstream material. Changes include selection, hiragana normalization,
reviewed pronunciation-derived kana, costs, de-duplication, and binary encoding.

The binary-only ZIP does not include notice files. Downstream distributions must
include the CMUdict notice alongside the existing JMdict attribution.

## Google Mozc resources

The Japanese dictionary and related resources are derived from
[google/mozc](https://github.com/google/mozc). Mozc's code license and binary
redistribution notice are in its
[`LICENSE`](https://github.com/google/mozc/blob/master/LICENSE). The bundled
`src/data/dictionary*` resources are mixed-license data. The upstream notice is
preserved in this repository as `src/main/resources/third-party/MOZC-LICENSE`,
alongside `src/main/resources/third-party/IPADIC-NOTICE`. The upstream source
notice is also available at
[`src/data/dictionary_oss/README.txt`](https://raw.githubusercontent.com/google/mozc/master/src/data/dictionary_oss/README.txt).

The Mozc dictionary notice includes, among other conditions:

- NAIST/IPAdic copyright and attribution requirements;
- the ICOT Free Software no-warranty and notice requirements; and
- Okinawa dictionary public-domain terms.

Those upstream conditions continue to apply to the corresponding generated
dictionary assets in this release. Google/Mozc names are not used to endorse
this project.

## Practical redistribution rule

When redistributing the generated assets, preserve the attribution and license
information required by each data source. The binary-only assets ZIP does not
carry notice files. This repository preserves them under
`src/main/resources/third-party/`; downstream packages must provide the required
information or links in their own distribution. `LICENSE-DATA.md` and `NOTICE.md`
in that directory provide additional JapaneseCorpus data provenance.

## 漢方・医療辞書

`src/main/specialist` のKNApSAcK KAMPO由来データと `kampo` 辞書の加工部分は
[CC BY-SA 2.1 日本](https://creativecommons.org/licenses/by-sa/2.1/jp/) に従います。
KNApSAcK Family © 金谷 重彦 (奈良先端科学技術大学院大学)。
万病辞書 MANBYO_202106 と百薬辞書 HYAKUYAKU_202107 由来データおよび
`medical` 辞書の編集・読み補完部分は [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/) で提供します。
作成者の表示は各提供元の保存した公式文書を維持します。

加工は表記・読みの抽出と正規化、重複除外、空欄へのGemini推定読みの追加、
品詞・コストの割当、バイナリ化です。原本の自動読みと推定読みの正確性は全件確認していません。
出典・版・取得ZIPハッシュ・各行の原本対応・加工状態は
[専門辞書の文書](../../../docs/specialist-dictionaries.md) と `src/main/specialist` に記録します。
出典・著作権・条件・加工表示を保持し、漢方の継承条件を守ってください。
配布時はライセンス全文を含む別添 `DICTIONARY-LICENSES.txt` を維持してください。