Translator's Blues

Slate.com ran an article the other day entitled Translator's Blues, in which the author frets that someday his translating job will be taken over by a computer. He starts by citing the scepticism most of his human colleagues hold toward machine translation ("MT"):

Anybody who's played around with translation software knows how bad the technology can be. Everyone in my office knows the hoary classic in which "The spirit is willing, but the flesh is weak," translated into Russian and back, comes out "The vodka is good, but the steak is lousy."

He then goes on to express his alarm at learning MT is a multi-billion dollar industry, and cites some software he tested out which at least gave better results than Babelfish. But we're not quite there yet:

The holy grail of MT is FAHQT: Fully Automatic, High-Quality Translation. For now, professional and amateur users content themselves with "gisting"—the practice of accepting 80 percent accuracy so as to get a general sense of a text's meaning. (Ninety percent accuracy leaves one error on every line.) Professionals who work with MT always do so in conjunction with human judgment, either by "pre-editing" to limit vocabulary or by "post-editing" to correct errors.

For now, he concludes, his (and my) job is safe.

Recently, someone on a translation-related mailing list pointed out the meaninglessness of expressing accuracy as a percentage--if a plural noun is translated as singular, is that the same percent wrong as it would have been had it been completely misrendered? What does "one mistake per line" mean?

I am not worried about machine translation. On the contrary, I think it opens up a new job possibility--MT software operator. Software capable of providing the "FAHQT" mentioned in the article would have to be pretty close to artificial intelligence, or use some pretty clever semantic indexing. Take another example from the article: "The con is in the pen." Software could figure out that this "pen" is the one which is the short version of "penitentiary" rather than the writing implement by its proximity to other, related words in the document.

That might lead to some pretty useful applications, but translation is still an art and not a science to me. I just came across a tough sentence in a job I am working on. The speaker grows coffee in Taiwan, and had a long period of hardship before becoming successful. He says:

「八年抗戰算什麼，我是走過王寶釧苦守寒窯十八年」

Literally, he says, "What was the eight-year war of resistance? I have gone through Wang Baochuan's difficult 18 years of keeping a cold oven." What would a machine do with that? A human has many options. If it were fiction, I might do something like "World War II was nothing--my struggle was more like the Hundred Years' War!" But these are the words of a real person and as such deserve a more faithful rendition that carry his way of expressing himself. But the reader needs the background behind the words. You could go with something like, "The eight-year War of Resistance [against Japan] was nothing--I went through 18 years of hardship like Wang Baochuan [the wife of a captured Tang Dynasty general]." and explain it in brackets, but that's not really elegant. You could even paraphrase it in third person, saying something like "He endured nearly two decades of struggle before finally tasting success."

It's a matter of judging your audience and deciding the style of the piece. No MT software is going to be doing that any time soon.