Description
The trouble with vocabulary apps is that you have to set aside time to open one, drill, and close it — separate from the typing you do all day anyway. Qingjian takes the other route: you already type hundreds of times a day, so it quietly places a translation in the language you are learning next to each pinyin candidate, and you pick up a word while you type — no interruption, no quiz, no forced recall.
It is a good pinyin IME first: written in Rust, whole-sentence input where you type the pinyin in one go and a local model re-ranks the candidates after a pause, entirely offline. On top of that it adds part-of-speech and a gloss to each candidate: with Chinese as your primary language and English as the one you are learning, "开发" shows development and "编译" shows compile; switch to Japanese and the same candidates show 開発 and コンパイル. The rule is one auxiliary language per candidate — clean, not English-Japanese-Korean-German piled on at once. A gloss you have seen a few times fades and eventually disappears.
Pinyin, dictionary, sentence model, learning history and glosses all live on your machine and are never uploaded. macOS and Windows, GPL-3.0.
Gloss beside candidates: each candidate carries a translation and part of speech, while word frequency, a personal n-gram and typo correction all learn locally so it fits you over time.
One learning language at a time: set a primary and a learning language, switch to Japanese or Korean for a different gloss set, and never more than one auxiliary language per candidate.
Whole-sentence input: type a full sentence of pinyin and a local model re-ranks the candidates, alongside abbreviated pinyin, fuzzy sounds, double pinyin, an English mode and emoji.
Fade as you learn: unfamiliar glosses are marked orange, then fade and drop away once you know them, so it stops nagging.
Fully local: pinyin, dictionary, sentence model, learning data and glosses stay offline, with optional cloud suggestions off by default.
It is a good pinyin IME first: written in Rust, whole-sentence input where you type the pinyin in one go and a local model re-ranks the candidates after a pause, entirely offline. On top of that it adds part-of-speech and a gloss to each candidate: with Chinese as your primary language and English as the one you are learning, "开发" shows development and "编译" shows compile; switch to Japanese and the same candidates show 開発 and コンパイル. The rule is one auxiliary language per candidate — clean, not English-Japanese-Korean-German piled on at once. A gloss you have seen a few times fades and eventually disappears.
Pinyin, dictionary, sentence model, learning history and glosses all live on your machine and are never uploaded. macOS and Windows, GPL-3.0.
Features
Gloss beside candidates: each candidate carries a translation and part of speech, while word frequency, a personal n-gram and typo correction all learn locally so it fits you over time.
One learning language at a time: set a primary and a learning language, switch to Japanese or Korean for a different gloss set, and never more than one auxiliary language per candidate.
Whole-sentence input: type a full sentence of pinyin and a local model re-ranks the candidates, alongside abbreviated pinyin, fuzzy sounds, double pinyin, an English mode and emoji.
Fade as you learn: unfamiliar glosses are marked orange, then fade and drop away once you know them, so it stops nagging.
Fully local: pinyin, dictionary, sentence model, learning data and glosses stay offline, with optional cloud suggestions off by default.
