Putting a translator in the middle of a conversation

The first time I did this properly I felt absurd. There was a laptop with a person on it and a phone in my hand running a translation app, and I was moving between the two like somebody operating machinery. It is not an elegant setup. Nobody has ever made it look elegant. And yet a version of that clumsy arrangement has produced several of the better conversations I have had in these rooms, so it is worth describing honestly rather than pretending it is smooth.

The problem is not the network. It is the human lag, and it is far longer than you expect. Somebody says something. You look down. You type or paste. You wait. You read. You look up and construct a reply. You look down again. Even when you get quick at it, that is eight to twelve seconds per exchange, and for the whole of it there is a person on your screen watching you look at something that is not them.

That was the part nobody prepared me for, and it is the reason most people give up in the first three minutes. It is not the effort — it is their face while they wait. They can see you are doing something. From their side it is indistinguishable from you checking your phone mid-conversation, which is the universal signal for I am not really here. It reads as rudeness. It is the exact opposite of rudeness. But it looks the same, and nobody is going to give you the benefit of the doubt in the first ninety seconds with a stranger.

Which makes one sentence more valuable than every other technique on this page: tell them what you are doing. “One moment, I am using a translator” — typed, gestured, held up to the camera, however it lands. It converts the whole thing from apparent distraction into an obvious shared project. And more than half the time, the moment I have said it, the other person visibly relaxes and opens one of their own, and then two people are doing an awkward thing together instead of one person being weird at the other.

What comes through the machine and what does not is very uneven, and knowing the pattern in advance saves a lot of confusion. Facts survive intact. Questions survive. Places, jobs, weather, food, what time it is where you are, what you did today — all of that arrives essentially undamaged. What does not survive is everything that makes a conversation feel like a conversation: jokes, sarcasm, understatement, teasing, warmth. Those are carried in the gap between what a sentence says and what it means, and that gap is exactly what a translator flattens.

The specific effect is that machine translation makes everybody sound like formal correspondence. You send something light and it arrives as a sentence with the posture of an official notice. I once said something mildly teasing to a woman in São Paulo and watched her read it, pause, and answer with visible caution, because what she had received was the tone of a letter from a bank. We spent a few more attempts on this before agreeing, mostly in gestures, to stop trying to be funny — and the conversation got substantially better once we did. Not because humour was impossible, but because we stopped routing it through the machine and started doing it with our faces, where it works fine.

After about ten minutes the rhythm stops being a problem and becomes the format. The pauses turn from dead air into something more like the gaps in a conversation held over a long distance. Both of you slow down. You stop saying filler, because filler is not worth the eight seconds. What you get instead is fewer sentences, each of which somebody actually decided to send. I would not call it better than talking normally, but it is not worse in the way I assumed it would be, and it has a strange deliberateness that some evenings genuinely suit.

The practical rules are short and they all point the same way. One short sentence at a time. Long paragraphs come back mangled and you will not know they were mangled, which is worse than knowing. Idioms are the enemy — every language has expressions that translate into nonsense, and you cannot tell which of yours are idioms until one lands strangely. Say the literal thing. “I am tired” beats anything more colourful. Numbers, names and times are the most reliable cargo you can send, so lean on them when you need to be sure.

Setup makes more difference than anything else. Alt-tabbing on one screen is the worst arrangement, because every switch takes your face fully out of the frame and the other person loses you completely. A second device beside the screen is much better: your face stays roughly where it was, your eyes flick sideways rather than vanishing, and the interruption reads as glancing rather than leaving. Split screen on one machine is a reasonable middle. It sounds like a small distinction and it is the difference between the call surviving and not.

I should be straight about what the site does and does not do here, because I have watched people hunt for a button that does not exist. Lamp Whisper’s own interface runs in English, Spanish, Arabic, Portuguese and Turkish, and there is a switcher for it — but that changes the language of the site around the conversation, not the conversation itself. Nothing is translating what your match says to you. Whatever translator you use is yours, running in another window or on another device, and the site is not involved in it at all. Better to know that going in than to spend three minutes looking for a feature while somebody waits.

If you know in advance that a language gap is the whole point of the evening, text matching is worth considering over video for the first stretch. Text is a separate match type here, and it removes the hardest part of the problem, which is that speech has to be transcribed before it can be translated at all. Copying written words into a translator is quick and accurate. Catching a spoken sentence in a language you do not know, and getting it into a box correctly, is genuinely difficult, and it is where most attempts fall apart.

The discovery that surprised me, though, was the opposite of everything above. The best of these calls did not have more translation in them. They had less. Once you accept that precision is not happening, you start using the camera instead: turning it to the window so somebody in another climate can see what November looks like here, holding up what you are cooking, pointing at things, exaggerating your face like a bad actor. A woman in Ankara and I once spent close to twenty minutes on a functional total of about forty translated words and a great deal of pointing, and I remember it better than most conversations I have had entirely in English.

So: is it weird? For roughly ten minutes, yes, and there is no way around that opening stretch. After that it stops being weird and becomes a slightly slower way of talking to somebody you had no business being able to talk to at all. What is actually strange, once you have done a few, is how much gets through anyway — not through the machine, but around it.