By Annalee NewitzPosted 12.28.2006 at 5:19 am 1 Comment
In a day of fantastic lectures and demonstrations at the Chaos Communication Congress, one of the most intriguing came from a computer science/artificial intelligence undergraduate at MIT named Christine Corbett Moran. She's been a very active contributor to an open source project called MOSES devoted to statistical machine translation (SMT). Although there are already a number of automatic translation software programs available -- many people are familiar with Google's translator and BabelFish -- few are open source and none are as robust as MOSES.
The advantage of making the program open source is that many people can implement it in various applications for an arbitrary number of languages. And the more that people implement it, the better MOSES gets. Moran joked that MOSES would be perfect for Finnish people who want to translate their writing into Klingon. But of course fast, automatic translation online is crucial to many people's daily lives -- not just Nordic Star Trek fans.
MOSES works sort of like a Bayesean spam filter, learning statistically which translations are "good" from vast quantities of language data. MOSES "learns" correct translations by poring over corpora of translations, the same way your spam filter "learns" when you mark some mail as spam. Moran said an excellent source of translations for MOSES are available from the European parliament, where speeches and discussions are translated into many languages at once. She urged the audience to test out the software, and add as many languages as possible to it.
A representative from Wikipedia seemed particularly excited by the possibilities, and vowed to test out MOSES on Wikipedia entries to see if it might work for mass translation on the huge community-edited encyclopedia. Moran thought that would be a great idea -- after all, the more correct translations MOSES sees, the better its translations become. As Moran fielded dozens of questions in the hall after her talk, it was obvious that open source translation programs are sorely needed. MOSES, or perhaps the next version of MOSES, may be what allows you to talk to people around the world in their languages -- instantaneously. --Annalee Newitz