Computer scientists have wrung nearly all of the performance they can from this approach and can't do much better than they are, even as hardware continues to get faster. This method involves not parsing sentences the way you learned to do in grade school but rather comparing lots of phrases with their existing translations (which Google did by downloading stacks of translated documents, like those from the United Nations). It's as good as it is because of advances in statistical translation. I use Google Translate almost every day, usually to see what the rest of the world is thinking. Here, for example, is a sentence from a World Cup report from a Madrid newspaper: "Passing Switzerland, Spain is now in the hands of the video and suture."
Frequently the system has the fluency of a barely competent human translator, one who happens to be both distracted and drunk. Spend more time with the software, though, and you realize while Google Translate is often spot-on perfect, especially with simple sentences from structurally simple languages like Spanish, it can produce puzzling sentences.
That, at least, is the impression you get as a new user. When you first encounter it, there is a whiff of sci-fi make-believe about this tearing down of ancient language barriers. Give it a Web page from a Spanish newspaper and in about a second the text is converted to English. Google Translate is a free Google service that translates between scores of languages almost as quickly as Google returns search results. Which is quite an admission, because Och is responsible for one of the most amazing computer programs in the world: He is head of the division at Ask Franz Josef Och the same question, though, and he says that even with a machine a thousand times more powerful than today's his program wouldn't run significantly better than it does right now, as far as most people could tell.
0 kommentar(er)
