Google reported significant progress for its instant translation service through the adoption of a deep learning neural network. Compared to its current application helpful, this new tool reduces the error rate to 85% and would be almost indistinguishable from a human translation. For now, he is able to translate Mandarin Chinese into English.
Instant translation is one of the areas in which the artificial intelligence (AI) has to play a major role.This is already the case with applications for and computer, including Microsoft or Translate who perform feats and allow us to consult websites or interact with people in a foreign language.
But significant progress, particularly in terms of deep learning and neural networks, let us glimpse the day not so long ago when polyglots machines will do as good performers as humans. Google gives us an overview with a new AI which make very significant progress in its instant translation system.Called (GNMT), it is already used by the Google Translate ( Google Translate ) to switch from Mandarin to English, which would represent 18 million daily translations.
The artificial intelligence of Google better understand the meaning of a sentence
According to the US giant, its reduced from 55 to 85% the error rate compared to the current version of the Google translation algorithm. The new deep learning method that is used is called long short-term memory (MTSA). She works on the model of human memory, giving AI the ability to retain short and long-term information. Once at the end of a sentence that analysis, the system is able to remember the beginning. It takes into account all the words, the sentence as a whole and its context, which gives a much more effective result in meaning.
So far, Google used a method called sentence-based machine translation (PBMT) that was the word for word translation. The results were quite uneven, and sometimes very bad when it came to Asian languages. It is also no coincidence that the new translation system was applied to the Chinese.
The MTSA technique is not new, but Google managed to make it work much faster, enough to what can be used for an online translation service. How? Using its own processors specifically designed for artificial intelligence and multilayer neural networks. According to Google, the same sentence processed by the MTSA method that took 10 seconds to translate do today requires 300 milliseconds.
Furthermore, researchers have found a way to optimize the collaboration between the different layers of the neural network. Thus, while a layer is doing its analysis, the next begins work without waiting for the first to finish. Google says it plans to expand the use of its new IA to other languages supported by Google Translate in the coming months.
Relaxnews original article published on 01/15/2015
The Google Translate app, also called Google Translate, is seen with two new features. One, Word Lens, allows translation of inscriptions on the panels. The other recognizes the languages and automatically transcribes content through oral conversation mode. What turn ainto a real instant translator.
Currently,can translate text into 90 languages and even receive voice translations for some of them. The company achieved an update of the application to recognize whether a person speaks a language relatively popular and instantly make a and especially oral.
The application is also able to translate some information boards present in the street. This is only possible in some languages only.
How helpful? This video will explain it. © Google Videos, YouTube
Vs. Google Translation Skype Translator
Microsoft’s system is based on machine learning technology, which means that the more the system is used and it is perfected, hence the need to have many registered testers, including for the next available languages as french, Chinese, Arabic, Urdu, Russian or German.