I’m a little behind on my stack of issues of Scientific American, and I’m currently wading through the December 2009 issue. There’s an article on World Changing Ideas, which is interesting in and of itself with 14 months of history between my reading and its publication, and it’s also available for free on Scientific American’s website. It’s also interesting because it features Data Mining algorithms.
One of the ideas is a Pocket Translator which could translate in real-time from one language to another. The primary driver is for use by soldiers to eliminate the military’s reliance on human translators.
It has a clever design, in that it uses Machine Learning and database look-up for common phrases. To wit:
The software works from a database of parallel texts—for example, War and Peace in two different languages, translated United Nations speeches, and documents pulled off the Web. Algorithms identify short matching phrases across sources, and the software uses them to build statistical models that link English phrases to Arabic ones.
But the real clincher is the punch-line:
A tourism-related translation app on a smart phone could help an American in Florence get directions from a non-English-speaking local, but they won’t chat about Renaissance art. “It is not going to work perfectly,” [John Makhoul, BBN’s chief scientist] says, “but it will do a pretty good job.”
In other words, close enough for government work.