Google had been using its own translation system for Arabic, Chinese, and Russian translations, but now uses it for all 25 languages it translates.
The difference between Google’s system and other systems is the use of statistical learning techniques to massive amounts of text, rather than building a complex rules-based approach, according to the Google Translate FAQ.
“Google’s approach works better for some languages and worse for others, but at least Google can expand to other languages without having to know them and manually create models for each one,” Chitu writes.
At Google Blogoscoped, Philipp Lenssen compares Google Translation to Systran and a human translation of a German paragraph into English, and vice-versa. “I couldn’t see a clear winner yet (though I get the feeling Google’s results are slightly superior), but a lot of garbage results on both ends,” he writes.