Earlier this month I wrote an article in Global By Design on Google’s ambitious machine translation (MT) plans. While Google is still very much in pre-beta stage with its home-brewed statistical MT (SMT) software, early indications are that Google is headed in the right direction.
The National Institute of Standards and Technology just published the results of an evaluation of more than a dozen MT solutions. The tests were designed only to look at how well the software translates from Chinese and Arabic into English. In the four tests that Google participated in, it beat the competition by a wide margin. It beat out Systran, IBM, Sakhr Software, and a number of universities.
Oddly missing from this list of competitors is Language Weaver, which also makes a SMT solution. I suspect that Language Weaver would have given Google a run for its money. Also note that this test does not imply that Google is ready to launch its software to the world just yet; I hear that their technology is not yet ready to tackly high-bandwidth, multi-user situations. But the lesson here I think is clear: Google is going to be a major player in MT in the years ahead and SMT in particular is going to be a major force.