In May and June, the National Institute of Standards and Technology benchmarked the performance of 23 programs that translate Chinese or Arabic text into English. It found software written by Franz Och, a University of Southern California computer scientist, to be the most efficient.
Unlike software that uses the grammar rules of the foreign language to translate the text, Och's program translates by statistical analysis of word usage. To build the translation engine, Och created software that analyzed texts that already have been translated.
"One of the great advantages of the statistical approach is that most of the work goes into components that are language-independent," Och said. "As long as you give me enough parallel data to train the system on, you can have a new system in a matter of days, if not hours."