Instant translation – no longer sci-fi
-
Loading... - J. Oliveira
- 23 Dec 2020
- 599 Views
- 4 Like
- 0 Comment
The idea that you could speak into a device in one language and it would emerge in another has long been a sci-fi fantasy. But this week that kind of automated translation came a step closer to reality when Skype launched the beta version of its Translator service.
For now it's an invitation-only trial and the only languages that it can handle are English and Spanish. I tried it out, talking to Maria Romero Garcia, a Spanish professor in Seville, who works with Skype.
What I found is that you have to use a good quality microphone and speak clearly in full sentences without pauses - otherwise the machine translation will kick in and interrupt you.
But the results were not bad at all. I asked Maria what she had been up to that morning. She replied in Spanish: "Esta manana ha estado trabajando un poco poco y concertando citas para ver a mis amigos esta tarde."
That came out in English as this: "This morning has been working a little bit and arranging appointments to see my friends this afternoon."
The technology does struggle at times - when Maria's cat wandered in front of the camera I asked what it was called and Translator decided I'd asked whether it was cold.
But there is a lot going on here, as Vikram Dendi, Microsoft's lead engineer on the project, explained on the line from the US. Live translation involves speech synthesis, voice recognition and machine translation - "each technology on its own is pretty complex, putting them together is a very difficult problem."
As someone who studied languages at university, spending many hours toiling my way through translation - and seeing friends go into the interpreting profession - I could see that teaching a computer a language was a huge challenge.
But it seems it is not a question of getting the machine to learn like a human. "It's not like someone who goes to school and learns a language by learning the rules," Vikram explains. "Computers use a different approach. They take large amounts of parallel texts - high-quality translated texts - and then use that text to build a probability base language model."
Computers, then, are living off the work of human linguists, scouring the web for examples of translated text. If this blogpost is translated into other languages, for instance, it could help feed translation engines of the future. That means the poor old human translators are helping to build the robots that could take their jobs, doesn't it?
Vikram Dendi says that is much too pessimistic a view - he believes that the explosion in internet use by people whose first language is not English will lead to a surge in demand for translation. "This will increase the amount of translation that will happen in the world - a portion of that will be done by technology and a portion by technology in conjunction with human translators."
In any case, Skype's Translator and its rivals have some way to go before they can match the abilities of a skilled human linguist. You would not want them involved in vital negotiations between world leaders for example. But over the next decade, you can expect to chat to friends whose language you don't share without stopping to flick through a dictionary.
