Meetings in more than one language

Everything Jaha produces falls into one of two kinds, and telling them apart answers almost every question about languages.

A meeting has one voice channel and one chat thread. Ten people, one stream. Nothing in that stream can be personalised — whatever Jaha says out loud, everybody in the room hears the same thing, so something has to decide which language it is in.

Everything else is rendered for exactly one reader: the summary you open, the digest in your inbox, the notification when somebody asks for you, the dashboard itself. Those simply follow your own language, with no compromise involved.

Ask in Hindi and the spoken answer comes back in Hindi — in a room of six Arabic speakers, and regardless of what anyone has set in their profile.

The obvious alternative is to follow the room’s majority, and it is worse than it sounds: it means the person who actually asked receives an answer they cannot read, in a language chosen by five people who did not ask. Nobody would pick that deliberately, but “use the room’s language” arrives there by accident.

When the recording is not sure which language a question was in — a short question gives it very little to go on — Jaha falls back to what you have said you speak, then to your organisation’s default, then to English. It never guesses silently: the choice it made and the reason are recorded against the meeting.

The spoken answer goes out immediately, in the asker’s language. A moment later the same answer appears in the meeting chat in the other languages actually present in the room — one message, each block labelled with the language’s own name.

Two things make that safe to rely on. The translations never delay the spoken answer: they catch up behind it, because a pause after somebody asks a question is the difference between an assistant and an obstacle. And a translation renders the answer that was already given rather than answering again, so the written version cannot contradict the spoken one on a number or a date.

A meeting where everybody shares a language gets no translations and costs nothing extra. The additional work happens only in rooms that need it.

In Settings, “The language I speak” is a fact about you rather than an instruction to Jaha. It does three jobs: it helps the recogniser hear you correctly, it decides the language of anything written for you alone, and it breaks the tie when a spoken question is too short to identify.

Leaving it unset is a real answer, not a missing one. It means your organisation’s default applies and what Jaha actually hears takes the lead — which is often the better behaviour for somebody who moves between languages.

Separately, the language of the dashboard itself is yours alone and changes nothing about meetings: an administrator can read the product in French while their organisation meets in Arabic.

The summary, decisions and action items are written once, from the recording, in the language the meeting was actually held in. Every other language is a translation of that one.

This matters more than it sounds. Producing each language independently would let two colleagues open the same meeting in two languages and be shown different figures, with nothing on the page to say which was right. Instead every rendition traces back to a single source, and the page tells you which one it is.

Quotes are the exception that proves it: a highlighted quote is a thing a person actually said, so the original wording is kept exactly and the translation sits alongside it. Jaha will not put translated words in somebody’s mouth.

Jaha works end to end in five languages: English, Arabic, French, German and Hindi. It recognises speech, speaks aloud, and writes answers and recaps in each of them.

Other languages are not offered, and that is a deliberate choice rather than a gap we hope you will not notice. A sixth language would need a voice, a recogniser and a translation path we can stand behind, and listing one before all three exist would mean transcribing a meeting into the wrong language while the settings screen claimed otherwise.

There is also no private audio channel. Jaha cannot speak one language into one person’s ear and another into somebody else’s — a Teams call has a single stream, so the written renditions in the meeting chat are the mechanism, not a lesser substitute for one we have not built.