教典備援網站｜台語辭典
A clean, standalone search site for the Ministry of Education's
Dictionary of Frequently-Used Taiwan Taiwanese (臺灣台語常用詞辭典).
Search and entry pages run entirely from static files — no dependency on the official site being up.
Live: https://taigikeyboard.tw/kautian-mirror/
Features
- One search box with live suggestions: Hanzi, Tâi-lô (tone marks, tone numbers, or toneless),
Pe̍h-ōe-jī, mixed Hanji and romanization (
bē赴 / be7赴 / 袂hù → 袂赴),
Zhuyin keyboard input, Taiwanese Phonetic Symbols, abbreviations, raw regex
- Entry pages: readings (TL + auto-converted POJ and 方音符號), senses, examples, alternative readings,
variant characters, 又見音 (same-hanzi entries), synonyms and antonyms, 語音差異 (accent readings),
詞彙比較 (per-accent vocabulary tables; both tables also show converted POJ / 方音符號), word audio
(fetched only when the speaker button is clicked)
- Entry actions: 收藏 (bookmark), copy the headword (
漢字 台羅), share the entry link (system share sheet, else
clipboard), and a word card — a PNG of the headword, readings and definitions with the source
credit, shared as a file where the browser supports it, otherwise downloaded
- Shareable URLs:
?q=<query> for a result list, ?id=<entry id> for an entry (same ids as the
official site)
- Left sidebar (collapsed by default; the top-left button docks it on wide screens, a drawer on phones):
外觀 (自動 / 淺色 / 暗色), 字型 (系統, 粉圓, 芫荽, 源樣明體, 源樣烏體 — the taigikeyboard roster,
CJK-subset by
scripts/subset-fonts.sh), 白話字 / 方音符號 switches, and the 收藏 list. All of it
lives in this browser's localStorage only
- SEO:
dist/sitemap.xml lists home + every entry URL (entry pages are client-rendered, so
crawlers can't reach them by links); result pages, missing entries and errors get noindex
The browser extension 教典鬥搜揣 lives in extension/ (see its README). Its search
engine, headword index and extension/kautian.ods are shared with the site, so one change covers
both; entry pages are built from the same kautian.ods.
Development
git clone https://github.com/taigikeyboard/kautian-mirror.git
cd kautian-mirror
make init # taigi-converter submodule
make install
make audio # optional: MOE word mp3 → dist/audio/ (300 MB download, cached in .cache/)
make sync-data # optional: kautian.csv + kautian.ods + word_unigrams.tsv ← taigikeyboard main
make build # dist/: search index, entry shards (data/entries/<id/500>.json), sitemap, app bundle;
# extension/dist/: extension bundle
make test # site + extension tests
make package # extension store zip → extension/kautian-extension.zip
make serve # http://127.0.0.1:8000
make serve rebuilds the JS bundle on change but copies index.html and styles.css only at
startup — restart it after editing those.
dist/ is a plain static site — deploy it to any static host. Run make audio before
make build so entry pages get play buttons; without it the site works, just silently.
Pushes to main deploy to GitHub Pages via .github/workflows/pages.yml (with audio).
extension/kautian.csv and extension/kautian.ods are copies of the cleaned data in
taigikeyboard (dictionary/sources/official/kautian/data/).
extension/word_unigrams.tsv is its corpus word counts (dictionary/shared/data/): search results
rank more used words first.
make sync-data fetches one commit from taigikeyboard main, rebuilds and tests locally.
It does not change the extension version, commit, push, or create a release.
Data updates
taigikeyboard's Kautian watch workflow is configured to check the MOE kautian.ods daily
at 08:00 Asia/Taipei, rebuild changed data and open a PR for manual merge after verification.
Merging the data PR closes its watch issue, preserving the source diff in the closed issue.
The site deployment checks taigikeyboard main daily at 09:00 Asia/Taipei, leaving one hour
for the upstream rebuild and verification. It builds from one upstream commit and publishes the site and
validated search JSON together. GitHub Actions can delay scheduled checks; these times
do not enforce a completion dependency. If the upstream update has not merged when the
site checks, it uses the last committed data and detects the update on a later run.
Scheduled checks compare the three source files' Git blob identities and the site commit with
the published descriptor. Unchanged inputs and site code skip downloading, building and
deployment. Pushes and manual runs still build the site; individual matching source files are
reused without downloading. Only a successful deployment advances the published comparison state.
From extension version 0.2.0, the extension uses its packaged dictionary on first use and checks
for compatible data updates during use, once per daily window starting at 09:00 Asia/Taipei,
matching the site schedule. If the browser is closed, it checks on the first use afterwards.
Updates are validated and stored
locally; a failed update keeps the existing dictionary. An updated dictionary is used on the
next page load. Pure data updates do not require a Chrome Web Store release; extension code
changes still do.
The format-specific descriptor is data/v6/latest.json; it identifies a JSON snapshot by SHA-256
and records its upstream commit on deployment. Older extension formats need their own compatible
data endpoint when a future format is introduced.
The ODS has the 詞彙比較 tables but not which entry shows which table (MOE curates the links,
sometimes across unrelated hanzi), so npm run build:comparisons reads them off every official
entry page into data/comparison-links.json. Rerun it after an ODS update (delete
.cache/comparison-pages.json first to refetch).
License
The source code is MIT licensed. Dictionary content originates from the Ministry of Education
dictionary (text and word audio; mp3 files are served unmodified) and is redistributed under
CC BY-ND 3.0 TW.
Han characters use jf open 粉圓 (SIL OFL 1.1),
the same typeface as taigi-converter.
This is an unofficial third-party site and is not affiliated with or endorsed by the Ministry of
Education.
