I Pressed Shift Six Times Just to Type Three English Words in One Chinese Sentence

The other night I sent a quick line to a teammate in Slack: "明天 3pm 和 @alex review 一下 auth-service 的 PR，没问题就 merge."

Right after hitting Enter, I stopped and counted what my hands had just done. To type a single sentence that took two seconds to think of, my left pinky had hit Shift six times, and my eyes had dropped down to the floating Pinyin candidate bar five times. It struck me how absurd that habit is: in my head, the sentence was one continuous thought, yet my keyboard forced me to stop and negotiate with a popup every two syllables.

When I'm writing pure Chinese, I barely notice the candidate bar. My fingers type a phrase, my thumb taps the space bar, and the characters land on the page. Where the candidate bar starts driving me crazy is in everyday work messages, emails, and code reviews where Chinese and English live inside the same line.

When you step back and look at why that feels so tiring, the reason is simple. Traditional Pinyin input methods bundle two separate jobs into the same keystroke: getting your thought onto the screen and resolving Pinyin into specific Chinese characters.

As long as an entire paragraph stays in Chinese, coupling those two steps together works fine. Years of dictionary tuning make pure Chinese composition fast. The trouble starts as soon as a sentence mixes in English nouns, function names, or ticket numbers. Your brain is already composing the second half of the sentence, but your fingers have to stop in the middle of the first half to babysit the candidate window. Type one beat too fast and forget to press Shift before review, and the Pinyin buffer swallows your English letters into a nonsensical string of Chinese characters. You can't press Enter, you can't press Space, and you end up hitting Backspace ten times to start over.

There is another catch with word-by-word candidate bars. Spoken Mandarin has countless homophones. When you have only typed the first two syllables of a clause, the input method has no idea what the rest of your sentence is about, so its top candidate is often a guess. You have to stop typing, scan the numbered list (1, 2, 3), pick the right character, and only then move on.

For messages like these, I eventually switched to the opposite habit: I type the entire sentence in plain Latin letters first, and convert it once at the end.

To write that same message now, I stay on my regular English keyboard and type straight through from start to finish: mingtian 3pm he @alex review yixia auth-service de PR, mei wenti jiu merge. While I'm typing, nothing pops up underneath my cursor. The space bar is just a space bar, and the Enter key won't accidentally commit a candidate list. I don't have to pause and tell the keyboard which word is Pinyin and which word is English.

Once the sentence is sitting in the text box, I press UnimeType's Convert action (the exact trigger and selection rules are covered in Actions). Because the model reads the full sentence at once, it has the surrounding context on both sides of every word. It turns the Pinyin segments into Chinese characters and leaves review, auth-service, PR, and merge untouched. If I ever prefer a different character for a specific word, tapping Another cycles to the next valid candidate.

Separating drafting from conversion changes the rhythm of writing a bilingual message. Instead of stopping every two seconds to negotiate with a floating bar, I can get the whole sentence out of my head at normal English typing speed and glance at the converted result once before hitting send.

I still keep my system Pinyin input method around. When I sit down to write a long document in pure Chinese, or when I'm working completely offline without a local model running, traditional Pinyin with muscle-memory phrase selection is still the fastest way to write.

For the messages that actually fill up my day — quick replies in Slack, WeChat, Linear, and GitHub where almost every line mixes Chinese and English — I no longer want to pick characters word by word while I'm still thinking. Typing the whole thought first and converting once at the end is simply a lot quieter.