When Google announced that its Google Translate app would be getting an update, the technology press went into a bit of a swoon. First with the story was the New York Times:
"The idea of a universal translator—a device that can seamlessly translate between languages—has been a longtime fixture in science fiction."
"Technology hasn’t quite gotten there, even on Earth, but Google has come one step closer with an upgrade of the Google Translate application, which is being released on Wednesday."
As word spread, the headlines got headier: “Google could be about to destroy the language barrier,” said the Telegraph. Google contented itself with saying it was “one step closer” to breaking the language barrier. But the casual reader could be forgiven for focusing on the more excitable responses.
The dream has transfixed science-fiction fans for decades. "Star Trek" had its universal translator, and Douglas Adams’s satirical "Hitchhiker’s Guide to the Galaxy" series had its Babel Fish. What if technology (or, in Adams's case, a super-evolved, ear-insertable fish) really could end all language barriers? In an optimistic scenario, world peace would be all but certain. In Adams’s satire, perfect understanding would cause “more and bloodier wars than anything else in the history of creation.”
Whichever your view, don’t throw away the language textbooks yet. Google Translate’s update does two nifty things. One is that when you point your camera at a foreign-language text (a book cover, sign or menu), optical-character-recognition and translation software instantly work together to try to render the text into your language—on your camera screen, so you can see the words in context. The second update is to make the voice-recognition and voice-synthesis parts of Google Translate recognise languages, instantly convert spoken words into text, translate the text, and then say the words in the target language. This looks closer to our Universal Translator than anything yet devised.
Alas, reality is different. What Google has done (in steps, not revolutionary leaps) is truly impressive. But anyone relying on it in a sensitive situation is likely to confuse lots of other people. At worst, it may annoy or insult them.
Johnson tested the app with a German colleague. We had the following exchange, both of us speaking our native languages slowly and carefully, pretending that we were trying to reserve a restaurant table. The text at the top is the original language as spoken; the text below is the words Google heard and transcribed in the original language. The third line of text is the translation (which was then synthesised into speech in the target language).
Me: Good morning, how are you?
Recognised: Good morning, how are you
Translated: Guten Morgen, wie geht es Ihnen?
Colleague: Guten Morgen, wie geht es dir?
Recognised: Guten Morgen, wie geht es dir
Translated: Good morning, how are you?
Perfect so far. But then:
Me: I’d like to reserve a table for 4 please
Recognised: I’d like to reserve a table for 4 please
Translated: Ich möchte einen Tisch für 4 behalten Sie bitte
This is word salad; the German reads something like “I’d like a table for four you keep please”. Understandable; if you were a willing and friendly restaurant host, you’d get it.
Colleague: Wir haben leider keinen Tisch im Moment. [Unfortunately we don’t have a table at the moment.]
Recognised: Wir haben leider keinen Tisch moments.
Translated: We did not get a table moments.
Now we have gone off the rails. Here the German speech recognition has failed, leading to a nonsensical English translation. Since I knew what my colleague meant, I went on:
Me: That’s a shame. I’ll come back later.
Recognised: That’s a shame I’ll come back later
Translated: Das ist schade, dass ich komme später wieder
This is worse than nonsense: I have now told her, in somewhat broken German, “It’s a shame that I’ll be back later.” We have the opposite of the Babel Fish. Perfect understanding would have meant a nice friendly interaction. Instead, I have insulted the restaurant. And for most people, such mishaps would be unwitting ones: the problem with using Google Translate with a language you don’t speak is that you don’t know when it has made a whopper of an error.
The camera-based optical recognition works rather better: picking letters out of an image is relatively straightforward, and translating carefully worded writing is far easier than doing the same with speech. That the character-recognition now works instantly and on the screen—even replacing the German text with English words in the same colour, size and font—is the most science-fictional thing about the new Google Translate.
But for the instant-interpreter functionality, most of my tests have gone roughly as well as the one above. Speech recognition is vastly harder than optical-character recognition. Speech is full of false starts and mumbles, throwing much more garbage into the translation engine. As engineers say, garbage in, garbage out. In another test, a mention of “Siemens” turned into “demons” and, thus, Dämonen in German.
The technology should continue to get incrementally better. And as people get used to the pitfalls as well as the possibilities, they will make sure to speak extra slowly and carefully, with the simplest possible language. The app is still useful, if not perfect—just like Star Trek’s Universal Translator. According to the "Star Trek" fan wiki, the gadget was still experimental at the time of the launch of the Enterprise NX-01. Despite its being able to translate alien languages in relatively short order, due to the UT's experimental nature, the use of a skilled linguist—in Enterprise's case, Hoshi Sato—was still required
And that in 2151.
Read the original article here.