The social intelligence research team of the social network has transferred translation services to artificial intelligence completely, according to an official blog post. The Facebook learning team has trained artificial intelligence to understand the best things like slang, typographical errors, etc. to provide a more accurate translation.
They use one type of neural network called the convoluntary neural-network (CNN), which has traditionally been relatively poor in this kind of thing. The common approach was to use a frequent neural network.
The fundamental difference, for us being not programmers for automated learning, is that the latter can only handle things in the system. CNN can process all data at once. This processor is obviously more powerful.
New translations have become remarkably better than before, instead of translating phrases, which may sometimes lead to meaningless words, they will now translate a word into a word and try to find a better approximation of what is meant. Even if there is no word from one language to another, it would be best to provide a readable translation.
0 Comments