0:00:00.420,0:00:04.660 Google Translate is a free tool that enables you to translate sentences, documents and 0:00:04.660,0:00:07.680 even whole websites instantly. 0:00:07.680,0:00:12.099 But how exactly does it work? While it may seem like we have a room full of bilingual 0:00:12.099,0:00:17.470 elves working for us, in fact all of our translations come from computers. These computers use a 0:00:17.470,0:00:22.520 process called “statistical machine translation” -- which is just a fancy way to say that our computers 0:00:22.520,0:00:25.439 generate translations based on patterns found in large amounts of text. 0:00:25.439,0:00:30.539 But let’s take a step back. If you want to teach someone a new language you might 0:00:30.539,0:00:34.809 start by teaching them vocabulary words and grammatical rules that explain how to construct 0:00:34.809,0:00:39.909 sentences. A computer can learn a foreign language the same way - by referring to vocabulary 0:00:39.909,0:00:42.379 and a set of rules. 0:00:42.379,0:00:46.309 But languages are complicated and, as any language learner can tell you, there are exceptions 0:00:46.309,0:00:51.679 to almost any rule. When you try to capture all of these exceptions, and exceptions to 0:00:51.679,0:00:57.239 the exceptions, in a computer program, the translation quality begins to break down. 0:00:57.239,0:01:01.779 Google Translate takes a different approach. Instead of trying to teach our computers all 0:01:01.779,0:01:07.060 the rules of a language, we let our computers discover the rules for themselves. They do 0:01:07.060,0:01:10.619 this by analyzing millions and millions of documents that have already been translated 0:01:10.619,0:01:17.259 by human translators. These translated texts come from books, organizations like the UN 0:01:17.259,0:01:19.950 and websites from all around the world. 0:01:19.950,0:01:23.810 Our computers scan these texts looking for statistically significant patterns--that is 0:01:23.810,0:01:27.969 to say, patterns between the translation and the original text that are unlikely to occur 0:01:27.969,0:01:33.829 by chance. Once the computer finds a pattern, it can use this pattern to translate similar 0:01:33.829,0:01:38.710 texts in the future. When you repeat this process billions of times you end up with 0:01:38.710,0:01:43.700 billions of patterns and one very smart computer program. 0:01:43.700,0:01:48.399 For some languages however we have fewer translated documents available and therefore fewer patterns 0:01:48.399,0:01:53.049 that our software has detected. This is why our translation quality will vary by language 0:01:53.049,0:01:58.479 and language pair. We know our translations aren’t always perfect but by constantly 0:01:58.479,0:02:03.240 providing new translated texts we can make our computers smarter and our translations 0:02:03.240,0:02:05.539 better. 0:02:05.539,0:02:09.670 So next time you translate a sentence or webpage with Google Translate, think about those millions 0:02:09.670,0:02:15.349 of documents and billions of patterns that ultimately led to your translation - and all 0:02:15.349,0:02:17.650 of it happening in the blink of an eye. 0:02:17.650,0:02:19.260 Pretty cool, isn’t it? 0:02:19.260,9:59:59.000 Give it a try at translate.google.com.