Your best friend and worst enemy.
We’ve all been guilty of using online translators at one time or another. Maybe you were just translating some Japanese you found online, or trying to figure out what that Jpop song just said. Or maybe you were using online translators to finish your Japanese homework. (Don’t worry, I won’t tell your sensei.)
And even though we’ve written about how you really shouldn’t use online translators and how they can go terribly wrong, I won’t judge you. I definitely have used Google Translate late at night for my homework for Japanese class. But have you ever wondered about how online translators actually work, and why you shouldn’t use them?
There are two main ways that computers translate one human language to another: one based on rules, and one based on something a little more complicated.
Rules, Rules, Rules
The first and older one is based on rules. A computer is programmed with the basic rules of a language and is given a dictionary. Then, when somebody puts in some text, the computer translates the text according to those rules and gives you a rough translation.
However, that’s proven to be a really crappy way to translate things because pretty much every single language in the world has tons of exceptions to its rules and a lot of the time, a translation will just end up with something that’s garbled and nonsensical.
Friends don’t let friends use Babel Fish.
A great example of this is Google’s early version of Google Translate. In the early days of Google Translate, Google founder Sergey Brin got an email written in Korean from a Google fanboy. But when Brin plugged in the email to Google Translate, he got the translation “The sliced raw fish shoes it wishes. Google green onion thing!” Not quite what the author had in mind.
Lies, Damned Lies, and Statistics
The other, more recent way that computers translate one human language to another is with huge databases of official, human translations. That means that these programs are given translations from places like the United Nations and the European Union and use those to make translations. This kind of translation is called statistical machine translation.
Statistical translation is what Google Translate currently uses, which is why it sometimes seems better, more natural than other translators. (I’m lookin’ at you, Babel Fish.) And statistical translation definitely works well with Google’s way of doing things: statistical translation requires lots of disk space (for the databases) and computing power, which Google has in spades.
Why Online Translators Suck
But there are big problems with both methods. There’s a lot of nuance in language that’s hard for a machine to catch, machines have problems with metaphors, and there are things like slang and different dialects that even a native speaker might have a hard time with. So while machine translations have come a long way since Sergey Brin heard about Google’s “green onion thing,” there’s still a long way to go before us humans are rendered obsolete. (Humans: 1, Computers: 70,136,459,345.)
And what lies ahead for translation tools? It’s hard to say at the moment. At this point, computer scientists are trying hard to make statistical translation better and better by adding more and more information to pull from. But like I pointed out above, this method has its limits.
It’s hard to think of a way that translation tools can stay ahead of the curve, so it looks like for the foreseeable future, human translation will reign supreme.
If you’re interested to read more, check out these two New York Times articles: “Google’s Computing Power Refines Translation Tool,” and “I, Translator.”
P.S. In point, in the continuation of Twitter.
P.P.S. As our for the sake of, the favorite, Facebook.