# Dictionary data licences

**The dictionary data in this directory is NOT covered by the repository's
Apache License, Version 2.0.** Each source keeps its original terms. The
compiled artifacts built from it — `dictionary/output/dictionary.bin`,
`dictionary.fst`, `association.bin`, `syllables.fst`, and the copies staged
into every application package — merge all of these sources into one index and
are **not separable** back into per-source rows.

Anyone redistributing this repository, or any application package built from
it, must satisfy every row below. `dictionary/docs/SOURCES.md` carries
publisher, origin URL, capture date, and pipeline notes per source.

## Status

Row counts are from `dictionary/output/dictionary.csv` (168,351 rows total).
"Exclusive" counts rows that no other source also supplies.

| Source | Rows | Exclusive | Licence | Verified |
| --- | --- | --- | --- | --- |
| `kautian` 教育部臺灣台語常用詞辭典 | 57,334 | 32,641 | **CC BY-ND 3.0 TW** | ✅ 2026-09-05, publisher's own 版權聲明 |
| `variants` 異用字 | — | — | inherits `kautian` | ✅ derived from `kautian`'s 異用字 sheet |
| `taijit` 台日大辭典 | 64,250 | 39,390 | **CC BY-NC-SA 3.0 TW** — non-commercial | ✅ 2026-09-05, ChhoeTaigi database README |
| `taihoa` 台華線頂對照典 | 57,477 | 28,873 | CC BY-SA 4.0 | ✅ 2026-09-05, ChhoeTaigi |
| `itaigi` 華台對照典 | 15,927 | 10,568 | **CC0** | ✅ 2026-09-05, ChhoeTaigi |
| `sitbut` 台灣植物名彙 | 1,351 | 540 | CC BY-SA 4.0 | ✅ 2026-09-05, ChhoeTaigi |
| `taigitv` 公視台語新詞辭庫 | 2,114 | 1,482 | CC BY 4.0 | ✅ 2026-09-05, publisher |
| `stti` 學科術語辭典 | 4,901 | 2,411 | 開放政府資料授權條款 1.0 | in-app credits screen; publisher statement not re-confirmed |
| `kungge` 台語工藝詞庫 | 994 | 668 | no data licence published; images-only CC BY-NC-ND | ✅ 2026-09-07, publisher's 圖像授權 page |
| `khiin` frequency data | 8,086 | 8,086 | **Unverified** | Khiin project (<https://github.com/khiin-pjh/khiin>); the specific file's terms not confirmed |
| `khpoo` 齒盤補充辭典 | 4,198 | 3,579 | **Unverified** | origin and rights holder not established |
| `lkk` 漢羅合用建議用字 | 58 | 58 | **Unverified** | origin and rights holder not established |
| `char_freq_merged.txt` | — | — | **Unverified** | corpus provenance not established |
| `dev` 開發者補充辭典 | 3,449 | 2,563 | Apache-2.0 | written by this project's maintainers |

The in-app credits screen (Home tab → 著作權) was the more accurate record for
`kautian`, `itaigi`, and `taijit`, where this pipeline's own notes had drifted.
It was **wrong** about `kungge`, which it labelled CC BY-NC 4.0 with no source;
that was corrected on 2026-09-07 against the publisher's own page. Neither
record is authoritative on its own — verify against the publisher.
`dictionary/LICENSE` is the single place the record lives.

## `kautian` — what is taken, and why CC BY-ND permits it

教育部's 版權聲明 (<https://sutian.moe.edu.tw/zh-hant/piantsip/pankhuan-singbing/>,
read 2026-09-05) releases the dictionary's 文字、音檔內容 under
**創用CC 姓名標示-禁止改作 3.0 臺灣 (CC BY-ND 3.0 TW)**.

There is **no NonCommercial term** — commercial use is permitted. What the
licence withholds is the right to distribute Adaptations.

What this project actually takes from that source is narrow. The pipeline's
`select` stage (`dictionary/sources/official/kautian/config.yaml`) reads
exactly two columns — **漢字 and 羅馬字** — from eight sheets, plus the
accent-difference sheet. It takes no 釋義, no 例句, no 華語對照, and no audio.
The shipped `dictionary.csv` has no definition or example column at all: every
other field is either a mechanical romanization transform this project computes
(`poj`, `tl_num`, `tps_num`, `*_notone`, `*_abbrev`), a frequency drawn from a
separate corpus, or a source bit.

That matters twice over:

- A pairing of a word with its pronunciation is a **fact about the language**,
  not 教育部's original expression. 著作權法 §10-1 extends protection to
  expression, not to the ideas, discoveries, or facts expressed.
- CC BY-ND 3.0 §3 expressly permits reproducing the Work, incorporating it
  into **Collections**, and making "such modifications as are technically
  necessary to exercise the rights in other media and formats". Assembling
  entries alongside eight other sources into a lookup index, and re-encoding
  that index as an FST, reads naturally as a Collection plus a format change
  rather than an Adaptation.

Attribution is required, and the application provides it on the Home tab →
著作權 screen.

This is the maintainer's reading, recorded so that it can be checked rather
than re-derived. It is not legal advice.

## `kungge` — what the publisher actually licenses

The publisher is **國立臺灣工藝研究發展中心 (NTCRI)**, not 教育部, and the site is
<https://kanggesu.ntcri.gov.tw/NTCRI_TaigiWebSite> (read 2026-09-07). The site has
three pages — 首頁, 關於本站, 圖像授權 — and the only licence statement on any of
them is the last one:

> 姓名標示-非商業性-禁止改作 — 僅允許使用者在下載授權者的著作之後，依原狀態分享
> 出去。使用者必須按照授權者指定的方式表彰其姓名，但不得對著作進行任何方式的改變，
> 或為商業目的之使用。

Three things follow, all of which contradict the record this file carried until
2026-09-07:

- The terms are **CC BY-NC-ND**, not CC BY-NC 4.0. No version is stated and the
  page links no CC deed.
- The page is titled **圖像授權 / `ImageLicense`**. It governs the 示意圖片, which
  this project does not take.
- There is **no licence statement covering the vocabulary itself**.

The 關於本站 page lists the glossary's components as 詞彙名稱（臺灣台語及華語對應
詞）、臺灣台語羅馬字拼音、真人發音音檔、示意圖片、語詞說明. The pipeline's `select`
stage keeps only the first two, and only their Taiwanese side: the shipped
`dictionary/sources/official/kungge/data/kungge.csv` has `hanzi` and `tl` plus
this project's own computed romanization transforms and a corpus frequency. The
raw scrape's `memo` (語詞說明), `audioPath`, `isHaveImage`, `provider`, and
`mainTypeName` fields are all dropped.

So the position is the same one recorded for `kautian` above, and rests on the
same ground rather than on a grant: a word paired with its pronunciation is a
**fact about the language**, and 著作權法 §10-1 extends protection to expression,
not to facts. The publisher's stated terms restrict the images and the prose,
neither of which this project redistributes.

Attribution is still given — Home tab → 著作權 credits 國立臺灣工藝研究發展中心 and
links both the site and its 圖像授權 page.

This is the maintainer's reading, recorded so that it can be checked rather than
re-derived. It is not legal advice.

## Non-commercial and share-alike obligations

`taijit` (CC BY-NC-SA 3.0 TW, 64,250 rows) is the **only** source carrying a
NonCommercial term over data this project ships. Because the compiled artifacts
merge every source into one inseparable index, **the compiled dictionary as
shipped must be treated as non-commercial**, regardless of the Apache-2.0
licence on this repository's source code. Neither `kautian` nor `kungge` is a
reason for that; `taijit` is.

`taihoa` and `sitbut` (CC BY-SA 4.0) and `taijit` (CC BY-NC-SA 3.0 TW) each
carry ShareAlike obligations. Whether one merged index can satisfy the
ShareAlike terms of two different licence versions at once has not been
analysed.

## What "Unverified" means here

Exactly what it says: **no redistribution licence has been identified for that
source.** It is not a claim that the data is public domain, open data, freely
reusable, or usable with attribution. A file being downloadable, or a dataset
being published by a government body, does not by itself grant a right to
reproduce, adapt, or redistribute it.

## Attribution

Applications built from this repository must credit at minimum:

- 中華民國教育部《臺灣台語常用詞辭典》 — <https://sutian.moe.edu.tw/>
- ChhoeTaigi 找台語 — <https://github.com/ChhoeTaigi/ChhoeTaigiDatabase>
- 公共電視文化事業基金會 / 公視台語台 — <https://www.taigitv.org.tw/taigi-words>
- 國立臺灣工藝研究發展中心《臺灣台語工藝詞庫》 — <https://kanggesu.ntcri.gov.tw/NTCRI_TaigiWebSite>
- 國家教育研究院
