DictateandTranslateonMac:OnePass,TwoLanguages

If you work across languages, you already have a routine: write the thing, open a translate tab, paste it in, copy the result, go back, paste again, reread it to make sure nothing got mangled. It takes maybe forty seconds. You do it fifteen times a day.

Translation-on-output collapses that routine into the act of speaking. You hold a shortcut, talk in whichever language you think in, and the text that appears at your cursor is already in the language your reader needs.

Why the tab workflow is worse than it looks

The forty seconds is not the real cost. The real costs are subtler:

• It taxes short messages hardest. The overhead is fixed, so a two-line Slack reply pays the same forty seconds as a long email. The predictable result is that people stop translating short messages and just send them in whatever language, or say less than they meant to.
• It breaks the thread of thought. Leaving the composer to visit a browser tab is a context switch, and coming back to a half-written message is measurably harder than continuing one.
• It launders formatting. Round-tripping through a web textarea strips structure and occasionally mangles punctuation, which you then fix by hand.

Removing the fixed overhead does not just save time — it changes what you are willing to write.

How translation-on-output works

Two stages run back to back. Your audio streams to a speech provider for real-time transcription in the language you spoke. That transcript is then sent for translation into your configured target language, and the translated result is what gets injected at your cursor. The untranslated intermediate never appears on screen.

The important design consequence: because translation runs on a clean transcript rather than on audio, it inherits the accuracy of the transcription step. Speak clearly and the translation is working from good input. Mumble a proper noun and the error propagates.

Set the target language deliberately

Translation-on-output has a failure mode worth knowing about: it is directional and stateful. If your target is set to German and you dictate a message intended for an English-speaking colleague, you will get German. There is no per-message prompt asking what you meant.

In practice most people settle into a pattern — target set to the language of their primary audience, toggled when they switch contexts. Worth building the habit of glancing at the setting before a long dictation rather than after.

Where it earns its keep

Team chat

This is the strongest case, because chat is where the fixed overhead hurts most. Replying in a colleague's language stops being a small decision and becomes the default. See dictating into Slack for the app-specific details.

Email to clients

Longer form, higher stakes, and a case where you should absolutely reread before sending. Translation-on-output gets you a solid draft in the target language; it does not remove your responsibility for what goes out under your name.

Documentation and support replies

Repetitive multilingual writing benefits enormously, because the overhead you are removing was being paid dozens of times a day. This is also where the selection assistant helps — translate an existing paragraph in place rather than rewriting it.

Where it does not help

Being honest about the boundaries:

• It is not interpretation. This translates text you dictate into a field. It does not translate a live call in either direction.
• Idiom and register are imperfect. Machine translation still flattens tone. For anything where phrasing carries weight — negotiation, discipline, condolence — draft it and have a human read it.
• Technical vocabulary drifts. Domain jargon and product names are the most common source of translation errors, in both directions.
• It requires a connection. Both transcription and translation are network calls. There is no offline path.

Frequently asked questions

How is this different from using Google Translate in a tab?

The tab approach costs you a context switch, a paste in, a copy out, and a paste back for every message. Live translation folds all of that into the dictation itself — you hold the shortcut, speak, and the translated text is what arrives in the field you were already writing in.

Which languages are supported?

Transcription covers 65 languages. Translation quality varies by pair: high-resource pairs like Spanish–English or French–English are strong, while lower-resource pairs are more variable and worth spot-checking before you rely on them for anything consequential.

Does translation add noticeable delay?

It adds a step, so it is not free. Transcription streams while you speak, and translation runs on the result, which means translated output lands slightly after untranslated output would. In practice it stays within conversational latency for short messages.

Can I translate text I have already written?

Yes. Select the text, trigger the selection shortcut, and ask for it in the target language. The selection is replaced in place, so it works in any app where you can select text.