Thursday, June 27, 2013

Machine Translation and the Future of Translation

There is a saying that has been going around the translation industry for the past few decades: “Machine translation is between 5 and 50 years away from perfection”. Although this is clearly a bit sarcastic and cynical, it also makes very good sense.

Compared with the machine translations of 10, 20, or 30 years ago, more contemporary automated translation engines – such as Google Translate – have really come a long way. However, they are still extremely far from perfect. Try, for example, entering an Italian or Spanish phrase into Google Translate, and translating it into English. Chances are the translation was pretty close if not perfect (depending on length, grammar, complexity, etc. of the phrase). Now try translating, say, a Japanese phrase into Estonian. The translation you get is probably barely even close.

Obviously, languages like English/Spanish/Italian which share many words and even grammatical structures are going to translate much more smoothly and naturally than completely unrelated languages such as Japanese/Estonian or Danish/Chinese. But sometimes, even Google Translate – used with pair of similar or even root-sharing languages – can surprise you. For example, with the many different ways to conjugate verbs in Spanish, even with a single, properly placed pronoun (I/you/we/he/she/it/they/etc.), the automatic translation you get can sometimes be laughably incorrect due to funky verb conjugations, dialect, even complicated punctuation.

Even so, automatic translation these days is much more advanced than it was even five or ten years ago. Does anyone else remember playing the “Babelfish game”? You’d put a simple phrase into Babelfish, translate it, translate it BACK, and repeat this process using two or more languages, eventually ending up with a bizarre, sometimes even psychedelic or surreal chunk of language that was almost invariably completely nonsensical and hilarious. Of course, you can still do the same thing with Babelfish or Google Translate these days, but I must admit that it takes a good number more clicks now than it did previously.

If you’ve ever seen Star Trek, with its “Universal Translator”, or listened to/read/seen/played any of various incarnations of the Hitchhiker’s Guide to the Galaxy series with its “Babelfish” (from which the Babelfish translation engine got its name), then you’re probably familiar with the direction in which a lot of people perceive the future of automatic/machine translation to be heading: a person says something – anything – in virtually any foreign language, and as the words leave their mouths, they are automatically translated by a machine (or an actual tiny fish, in “Hitchhiker’s” case), and heard by the listener in their native tongue. While this is obviously fantastical and inconceivable for a number of obvious reasons which I won’t bother going into here, one can’t help but wonder: just how close to that ideal can machine translation actually come? And, even more importantly: how will that affect us as translators?
Luckily for us, I really doubt that machine translation will ever get THAT far. However, it is true that the more precise and easy-to-use machine translation becomes, the harder it will be for translators to get work, and the less money will generally be available for these hardworking humans. I don’t presume to have the answers to the big “What do we do about the human/automatic translation battle?!” question, but I can offer two opinions I have on the matter:

1) As I stated earlier, I don’t believe it is possible that machine translation will ever reach the levels of science fiction-based examples of the “future of translation”; and 2) let’s all hope that, if machine translation IS ever perfected, it’s 50 years away, and not 5.