Statistical machine translation gains respect
Don Geer · Computer · 2005
Relatively few researchers have worked on approaches that compare and analyze documents and their already-available translations to determine statistically, without prior linguistic knowledge, the likely meanings of phrases. These statistical systems use this information to translate new documents. For years, because processors were not fast enough to handle the extensive computation these systems require, many experts considered statistical systems inferior to rule-based systems. However, when the Speech Group of the US National Institute of Standards and Technology's Information Access Division tested 20 machine translation technologies, a statistical system developed by Google finished in first place. The NIST test results' significance is that Google and other organizations will invest more time, money, and talent into researching this approach. Meanwhile, faster processors and other advances are making statistical translation technology more accurate and thus more useful. However, the approach must still clear several hurdles - such as still inadequate accuracy and problems recognizing idioms - before it can be useful for mission-critical tasks.