Alphabet Inc (NASDAQ:GOOGL) Google’s translation tool has developed a ‘language’ of its own. This was confirmed by Google’s Artificial Intelligence researchers, the brains behind Google Translate magic. The language dubbed ‘interlingua’ came about following the launch of the Google Neural Machine Translation – GNMT, back in September. The GNMT had been conceived to exploit deep learning for the purposes of producing better translations akin to human translators.
Google Translate Magic provides an answer to the big question
After GNMT went live, the question that emerged was whether the system could be used to translate languages it had not been taught directly. In other words, if the system could be taught to translate from English to Italian and from English to French, would the system be able to translate from French to Italian? The answer, it turned out, was a resounding yes. This then posed the question of whether the system had created its own language with which it was using to translate two languages that had not been directly linked previously. The researchers now think it has and this language has been dubbed ‘interlingua’.
The Google Translate Magic findings, which explain the believed existence of the artificial language, are now documented a research paper.
“We interpret this as a sign of existence of an interlingua in the network,” the team of researchers said regarding their findings.
Following the positive results, the multilingual GNMT has now been deployed on all of Google Translate. Additionally, Google has disclosed that in the translation efforts of 10 of the 16 language pairs that have been recently added, multilingual systems are now in use.
Google Brain, which is behind the Google Translate magic, has lately been in the news for all the right reasons. The team recently developed an encryption system that used artificial intelligence to do all the work. Three neural networks were created with each assigned a specific task. One was required to encrypt while the other was supposed to decode. The final one was supposed to decrypt without resorting to encryption keys. After some training, the neural networks succeeded in encrypting and decoding.
Receive News & Ratings Via Email - Enter your email address below to receive a concise daily summary of the latest news and analysts' ratings with our FREE daily email newsletter.