adventures in machine translation
Posted by adiamondinsunlight on August 21, 2008
As I noted yesterday, I’ve been taking advantage of the mid-August slow-down to address long-neglected housecleaning tasks at my new organization, like the computer defragmenting I mentioned yesterday and backing up our shared files. Neither of which have been done before. Ever.
My colleagues have a fine collection of advanced degrees from well-regarded universities around the US and Europe – which is just another way of saying that we confirm the cliche that some of the brightest people have the least common sense.
Anyway. In between bouts of hausfrau’ing, I have been busily investigating the latest advances in machine translation. When I mentioned this online to our friend B, he responded in kind:
01010111011010000110000101, B wrote.
111001100111? he asked me.
I have absolutely no idea what he was saying – although I’m hoping it was something along the lines of:
Hi Diamond. I’ll be back in New York this weekend – let’s all try to meet up for another spicier-than-anything-nature-intended Punjabi dinner on Curry Row.
In any case, this wasn’t what I meant. Machine translation doesn’t refer to the way in which computers translate ones and zeros into human language – it refers to the way in which computers manually translate one human language to another. You’ve probably had some experience with this, whether through Google’s automatic page translation service or that first-generation standby, Babelfish.
Machine translation is usually a poor substitute for human translation – even when moving between relatively close languages like French and Spanish, or English and German. And when it comes to translating between Arabic and English, most manual translators give the would-be reader only a glancing sense of what the original text might say. Individual words translate well, but coherent phrases are relatively rare. (Try translating Al Jazeera’s homepage via Google and you will see what I mean.)
But, as can often be the case with mis-translations, some of the translations that the system I was testing out offered (English to Arabic and vice versa) made me think – like “plastic surgery”.
(This wasn’t a Freudian choice of phrase, honestly. We had been talking about plastic surgery the other day, and I was testing out technical terms – compound nouns or noun-adjective phrases that as a whole meant something different than the sum of their parts.)
The phrase in Arabic that I use for “plastic surgery” is:
I’ve never really thought about its literal meaning – so when the machine translator came up with
my initial reaction was to laugh. After all, a jur7 is a wound, and “plastique”, as in English, refers to explosives. So I thought that the site had erred, coming up with “explosive wounding” – which sounded like a pretty fine critique of plastic surgery to me. (And when I pulled out my dictionary, I learned that “jira7a” means surgery. Also an appropriate term, since surgery does involve making deliberate wounds in the human body.)
But when I thought further, I realized that the “normal” term for plastic surgery is a bit odd, too. 3amiliya al-tajmeel literally means “beautifying operation” – a phrase that lays bare the primary function of today’s plastic surgery procedures.
I’m still not sold on machine translation, but in this particular case, it gave me far more to think about than I had anticipated.