Three deadlines, three target languages, and a call in between

A freelance translator's day rarely holds still for one project. This one carries a rush legal segment due at noon, a client call about a marketing line that doesn't survive a literal rendering, and a project note that switches languages mid-sentence — three different kinds of dictation inside one working day, none of which involve the app doing any of the translating itself.

8:50 AM — the rush segment, already translated, not yet typed

The source document arrived overnight: a delivery clause for a contract due back by noon. By the time the CAT tool's segment field is open, the French rendering is already decided — what's left is getting it typed without losing the morning to it.

Holding ⌥ Space and saying the sentence exactly as it was already decided lands it in the segment field through the accessibility API, the same way it would land in any other document editor. Nothing about the rendering changed between deciding it and seeing it on the page — a rush deadline is exactly when that gap is worth closing. The mechanics of the held key itself are the same regardless of which document editor or CAT tool ends up on the receiving end.

10:30 AM — a call about a line that won't translate literally

A marketing client wants a tagline moved into German, and the literal rendering reads flat in the target language — the kind of problem a translator solves by ear, saying candidate phrasings aloud before committing to one. None of that thinking is the app's; it only types whichever version gets said last.

After the call, the client note explaining why one phrasing won over another is prose, not a segment — typed into an email the same way any explanation would be, at whatever length the reasoning actually needs.

12:15 PM — a project note that switches languages mid-sentence

The afternoon's project manager update needs a line in Chinese followed immediately by an English aside about a delivery date, inside the same message. The default engine detects the spoken language itself across a single utterance and punctuates the Chinese half automatically, without a manual switch first — a narrower claim than full bilingual support for any language pair, but a real one for this specific one.

2:00 PM — a revision comes back, and the segment gets said again

The legal client's counsel wants one clause reworded before signature — not a rewrite of the whole document, just the delivery clause from the morning, with a narrower notice period. The translating happens again, briefly, the same way it did the first time: deciding the new French wording in the translator's head before touching the segment field at all.

Revisions are a normal part of a legal translation and this workflow treats this one the same way it treated the original: the app has no memory of the earlier version and does not compare the two, it only types whatever gets said into the segment that's currently focused.

3:30 PM — the document editor stalls mid-revision

Word freezes for a few seconds right after the revised clause is dictated — an ordinary annoyance on a machine running a CAT tool, a browser, and a video call all at once, and exactly the moment a rush afternoon can least afford to lose a sentence to.

Because the sentence was already written to local history the instant the key was released, before the editor did anything with it, the freeze costs nothing more than switching back and pasting the same text in again once Word responds — not redeciding the wording, and not saying it a second time.

Keeping a desktop and an on-site laptop from drifting apart

The morning's legal segments were dictated on the desktop where the CAT tool's translation memory lives; the client call happened on a laptop carried to their office for the meeting. A licence covers three devices, each installed separately, with no account and nothing synced between them — which means the laptop's history from the call and the desktop's history from the segments stay two separate logs rather than one merged timeline. For a translator moving between a home setup and client sites in the same week, that's a deliberate trade: nothing automatically reconciles across machines, so a habit of checking both histories before closing out a project is worth building rather than assuming.

What the day's history is worth at invoice time

A freelance translator billing by the hour has a use for the day's history beyond recovering a dropped sentence: every row carries a timestamp, so the gap between the first segment at 8:52 and the last project update at 12:16 is a real record of when work happened, even though it says nothing about how long any individual segment took to think through before it was said. It's not a time-tracking feature and nothing here totals the hours automatically — reading the timestamps by eye and matching them against an invoice is still a manual step, the same one it would be after typing instead of dictating. What changes is that the record exists at all, on the translator's own machine, without a separate app running in the background to produce it.

What the day's history actually holds

Every segment lands in local history the instant the key is released, before the transcription engine is asked to produce anything, and that history never leaves the machine it was said on. For a translator working from a client's source files under an NDA, that's the property that actually matters more than any single day's convenience.

Where the terminology has to be typed in by hand

The client's glossary has a preferred rendering for one repeated technical term, and the app has no way to learn it from context or from a previous project — it has to be typed into the dictionary once, used exactly as entered from then on. That's a real limit for a translator with a large in-house glossary and no patience for entering it term by term, worth weighing against a tool that learns vocabulary as it goes before committing a whole project's terminology to either approach.

For a project with only a handful of recurring terms — a company name, a product line, one piece of jargon the client repeats in every document — the entry cost is small enough to absorb in the first few minutes of the project and stops paying attention to it after. For a glossary running into the hundreds of entries, typing each one in by hand before the terminology comes out consistently is a real cost this workflow does not hide, and it's worth weighing against the size of the actual project before assuming the dictionary will keep pace with it.

Where the language list runs out

This app names ten languages explicitly and a second engine covering "dozens" more without a specific count. Wispr Flow publishes support for over a hundred, a materially longer list — worth checking against the actual pair a job requires before assuming either tool covers it, and covered at more length in the round-up of multilingual dictation apps. For a translator who only ever needs to switch between two named languages inside a single project, dictating in two languages walks through the mechanics in more detail than this one day can.

What a proofreading pass still requires

Saying a target sentence rather than typing it changes nothing about the obligation to check it before delivery. A mistranscribed word, a name the engine heard wrong, or a punctuation mark placed where the spoken pause happened to fall can slip into a dictated segment the same way a typo slips into a typed one — arguably more easily, since the eye doesn't pass over each character the way it does while typing. The read-back before delivery is not optional and this workflow does not treat it as such: every one of today's three deliveries got read once, in full, before it left the machine, the same review a typed draft would have needed regardless of how it was produced. What changed across the day was where the hour went — less of it spent on the mechanical act of typing an already-decided sentence, and, in principle, more of it available for exactly that read-back.

Why the CAT tool's own memory and quality checks still do their job

None of today's dictation touched what the CAT tool itself does with a segment once it's typed in. Translation memory still proposes a fuzzy match against an earlier project the same way it always has; the terminology checker still flags an inconsistent rendering the same way it would for a typed segment; the quality-assurance pass at the end of the project still runs unchanged. Dictation only changes how a fresh segment gets into the tool's text field when no match applies and the sentence has to be composed from nothing — everything the tool does with it afterward is exactly the software behaving as it always has, untouched by how the words arrived.

Where Apple's own Dictation already covers some of this, free

Apple Dictation is already installed, free, and supports several languages with a manual switch by clicking the abbreviation near the cursor. It doesn't translate either. What a dedicated hotkey mainly changes is when the words start appearing — the instant the key goes down, rather than after Apple's own short pause — which matters more across a day with three deadlines than it does for one occasional sentence.

Delivering three files without a shared connection

The legal segment, the marketing note, and the project update all left this machine the same way they arrived on it — locally, with nothing uploaded during drafting. The app works identically with the network off, which matters on a day that might otherwise be spent worrying about a client's confidentiality clause rather than the actual translating.

What the day costs

$19 once, three devices per licence, macOS 13 Ventura or later. A translator running a desktop for CAT-tool work and a laptop for on-site client meetings gets an independent history on each machine, with no account and nothing synced between them automatically. The thirty-day trial runs the complete app against a real project's actual deadlines rather than a single test sentence — the only fair way to judge whether the mechanism holds up against a day like this one.

Where the rest of a translator's day is covered

This one day covers three deliveries and a call; the fuller case for dictating a translation covers the mechanism itself in more depth — what it does and does not do, independent of any single project's deadlines.

What tomorrow's client note still needs typed by hand

Nothing about today's mixed-language message or the German phrasing call changes how the next unfamiliar term gets handled — it still needs a human decision the first time, and the dictionary entry that follows it. What changes, project after project, is only how much of the mechanical typing between those decisions gets replaced by saying the sentence once, in whichever language it was already decided in. Dictating email faster covers the correspondence side of the same day for anyone whose translating work is mostly project notes and client updates rather than segments inside a CAT tool.

A single rush day like this one is not really the point, on its own — the same three motions repeat across a full project, and then again across the next one: a source sentence gets read and translated silently, the target version gets said aloud instead of typed, and the read-back afterward catches whatever the engine got wrong. None of that sequence changes project to project. What changes is only how many hours of a working week end up spent on the middle step, typing sentences that were already fully decided before the key went down, versus how many hours are left over for the parts of the job — reading the source carefully, weighing a difficult phrase, catching an inconsistency the CAT tool's own checks missed — that were always the actual translating, and were never going to be replaced by any dictation tool, this one included.