Google promises better translations by using neural network
Google is an artificial neural network committed to translations in Google Translate to improve. That should be the number of translation errors greatly reduce.
Until now translated Google Translate separate phrases and taped it back behind the other. Now get Google to self-learning algorithms that at one time a whole sentence can view and translate.
The neural vertaalnetwerk (GNMT, for Google Neural Machine Translation) should always be better, to the extent there are more sentences to be translated, and feedback is given.
In different language pairs approaches the new system, the translation quality of people, claims Google in a blog post about the renewal. The number of errors by 55 to 85 percent can be reduced.
Starting this week, all translations from Chinese to English via the new system. In the coming time, the neural network was rolled out to the 10,000 other language pairs that Google Translate supports.
“Machine translation is absolutely not solved”, write researchers from Google. “GNMT may still have significant errors that a human translator would never make, such as dropping of words and the wrong translation of nouns, or unusual terms, and the translation of sentences instead of the view of the context in a paragraph or page.”
Google uses smart machine learning systems in more and more areas, such as the recognition of images and sound. Recently demonstrated by a research team from Google have how a neural network written text could convert it to natural sounding speech.