Korean looks like an easy language to transcribe. The writing system is alphabetic, the sounds map cleanly to syllable blocks, and there's none of the homophone density that makes some Asian languages so unforgiving. And yet Korean dictation fails in a way native readers spot in half a second — not on the words themselves, but on two things a raw transcript almost never gets right: where the spaces go, and how polite the sentence is. Both are decisions about the whole sentence, not the sounds passing through it. A tool that transcribes what it hears and stops there hands a Korean writer text that is technically correct and socially wrong. Here's what separates Korean dictation you can send from Korean dictation you have to rewrite.
띄어쓰기: spacing is a grammar decision, not a pause
Korean puts spaces between words, but where a "word" ends is governed by rules that don't line up with where a speaker pauses for breath. Particles attach to the noun before them with no space; auxiliary verbs, dependent nouns, and counting units all have their own spacing conventions. Speech doesn't announce any of this — people run whole clauses together and pause in the middle of grammatical units. So a tool that inserts spaces where it hears silence will scatter them in the wrong places, and Korean spacing errors are not subtle. A misplaced space reads to a native speaker the way a missing one does in English: instantly, glaringly off.
The problem is that correct 띄어쓰기 can't be recovered from the audio. The same string of sounds is spaced differently depending on the grammatical role each chunk is playing — whether that syllable is a dependent noun or part of the verb, whether this is a particle or a standalone word. You have to parse the sentence to place the spaces. That's why spacing is the clearest single tell that a Korean transcript came out of sound-matching alone: it will be close, and close is exactly what a reader notices.
Honorifics: the same sentence, four different registers
Korean grammar bakes politeness into the verb. "I'll send it" is not one sentence — it's 보낼게, 보낼게요, 보내드리겠습니다, and more, each carrying a different level of deference, and each correct only in the right setting. The distinction isn't decoration; using a casual ending with a senior colleague reads as rude, and using the most formal ending in a team chat reads as stiff and strange. Native speakers switch registers constantly and mostly without thinking, but the register is a property of who you're writing to, not of the words you spoke.
This is where transcription alone runs out of road. When you dictate a message, you might speak it in a middle register, or half-form it, or say the polite ending out loud only because you're thinking about the recipient. A raw transcript freezes whatever came out of your mouth. But a Slack DM to a peer and an email to a director are not the same document, even when the content is identical — the ending has to shift, the sentence-final particles have to shift, sometimes the whole verb changes. Getting that right means deciding the register for the destination and reshaping the sentence to match, which is a rewrite, not a recording.
Why the rewrite pass matters more in Korean than in English
In English, a clean transcript is often close to a finished sentence. Capitalize the first word, drop the filler, add a period, and you can usually send it. Korean doesn't offer that shortcut. The transcript can capture every syllable perfectly and still be wrong on spacing and wrong on register — two things you cannot fix by transcribing more accurately, because they aren't in the audio to begin with. They're decisions about grammar and audience that have to be made after the sounds are known.
That's the case for treating dictation as two steps, not one. First recognize the speech; then rewrite the result into text a Korean reader would accept — spacing resolved by parsing the sentence, register set for who the message is going to, the half-finished spoken phrasing tightened into something written. In English the second step is a nicety. In Korean it's the difference between output you send and output you retype, which is why post-recognition rewriting isn't a polish layer here — it's the part that makes Korean dictation work at all.
외래어: the English running through every sentence
Sit in a Korean product meeting and the sentences are full of English. Loanwords and borrowed terms sit inside Korean grammar as a matter of course — 미팅, 프로젝트, 데드라인, 피드백 — and alongside them people say the English words outright: "이번 스프린트 데드라인까지 리뷰 끝내주세요." Some of these have settled into Korean spelling; some are still spoken as English and expected back as English. A writer knows which is which, and expects the text to match: the settled loanword written in Hangul, the live English term written in the Latin alphabet.
A tool that commits to one language gets the other wrong in both directions. Lock onto Korean and the English comes back as approximate Hangul that no one would type; lock onto English and the Korean falls apart around it. And these borrowed terms still take Korean particles, so getting them right isn't just spelling the word — it's fitting an English term into a Korean grammatical slot and spacing it correctly against the particle that follows. That only happens if the tool is reading the whole sentence, not classifying it as one language and transcribing accordingly.
Why "supports Korean" doesn't tell you much
A tool can list Korean, transcribe a slow clean sentence in a quiet room, and still hand you scattered 띄어쓰기, a register that fits a chat but not the email you were actually writing, and English terms mangled into approximate Hangul. The language count on the feature list won't warn you about any of it. One real session will. Dictate the way you actually work — a message to a colleague, then the same content as a note to someone senior, with your usual mix of Korean and English — and read what comes back. Is the spacing something you'd leave alone? Does the politeness match the person you're writing to? Would you send it, or fix it first? That question, not the presence of Korean on a list, is what tells you whether Korean voice typing will save you work or just move it. See how we think about it at Sageio Type.