Google's new translation trick, called "zero-shot" translation, can automatically translate many new language pairs.
Google Neural Machine Translation (GNMT) – first released in September – is a new system that supports Google Translate, built based on neutral network. Neutral network is a type of computer system that simulates the human brain, can "learn" from past actions to solve new problems without having to be programmed specifically.
Although Google has been updated about translation capabilities, it has improved the accuracy thanks to Machine Learning, but the new Google Translate even translates from this language in the other language even though both languages have not been grafted yet. previous pair.
The article on the Google Research blog last November 22 said:
"We have a language system with examples from Japanese to English and Korean to English … [GNMT] share the parameters to translate between these 4 different language pairs … It can pay on reasonable translations from Korean to Japanese, although GNMT has not been previously instructed. According to what we know, this is the first time that transfer learning is active in Machine Translation ”
TopDev Techtalk # 52: Technology of the Future: What Breakthrough from Machine Learning?
* Ho Chi Minh: 18:00 – 21:00 on December 2, 2016.
In fact, this is just one of Google's steps to get closer to Artificial Intelligence. Earlier this year, similar to Apple's Siri and Amazon's Alexa,