Tower of Babel Translator For Picking Up Hot Foreign Girls

This image was lost some time after publication.
This image was lost some time after publication.

Why limit yourself to knowing only one language when you pretend to know many more? That's what a bunch of U.S. scientists figured when they set about to create a "Tower Of Babel" translator. A series of electrodes connect to the user's neck and face to detect movements that are made when users mouth words. That is, by simply mouthing the word you want translated, rather than outright speaking them, the device can translate. There's two working prototypes right now, one that translates Chinese into English and one that translates from English into Spanish or German. Using a small vocabulary of about 100-200 words, scientists says they've attained an accuracy of 80 percent.

Advertisement

But think of the bigger picture: if scientists successfully made a device that translates all the world's languages, we'd be left without terrible Brad Pitt movies. And that, dear friends, would be a tragedy of unfathomable size and scope.

'Tower of Babel' translator made [BBC News]

Share This Story

Get our newsletter

DISCUSSION

Every few years, a product like this gets released - my field is computational linguistics, and as the size of the dictionary goes up, the quality of production will go down.

The technology simply isn't at the point where the dream of the translator from star trek is possible. We don't have a full understanding of the syntax of any language to produce a matrix that will accurately translate one language to another.

This is interesting in that it uses actual speech -> speech word based translation, as opposed to the canned phrase type translation used in most of the products out there - this alone makes the technology interesting, but until they get up to a dictionary of 10,000 words or so, with an accuracy of >80%, I don't see this technology being particularly practical. Even at that level of sophistication, there a are still going to be a LOT of errors, and a LOT of untranslatable words or phrases.

An N-gram type best-fit translation model would give the best translation rate, but even then, it would require a gigantic corpus of parallel speech, checked and possibly tagged by hand.

More power to them, though, I'd love to see someone get it right. It would revolutionize compling, and many other fields.