Copyright © Françoise Herrmann
There is a new kid in town for machine translation! And it’s bin nominated for the 2013 EPO Oscars!
This model suggests that machine translation recognize phrases instead of words. And it comes as a novel attempt at resolving the now almost 75-year challenge of machine translation.
Evidently, there’s no rush… or obstacle on the path to the solution.
We all know that context matters, and that “The baby is in the pen” makes perfect sense, and only nonsense in machine translation, where in 100 % of cases the machine translation for “pen” comes back as “stylo” (in French) instead of “parc” (still in French).
So the new solution is to parse whole phrases, for which it is claimed that there is more context-independence, compared to single words, which are so context-dependent for meaning.
Great computational solution, with no detours or bifurcations!
The patent applicant is the University of Southern California. It was awarded to the team of computational linguists: DANIEL MARCU, KEVIN KNIGHT, WILLIAM WONG and PHILIPP KOEHN.
This is also a DoD (Department of Defense) DARPA (Defense Advanced Research Projects Agency) project, and a model in use by Google Translate!
Indeed, we have all already had an opportunity to use it with the EPO automated translation engine for patent abstracts, and patent descriptions,haven't we?
|
EP1488338 - PHRASE-BASED JOINT PROBABILITY MODEL FOR STATISTICAL MACHINE TRANSLATION
|
No comments:
Post a Comment