On November 14th, 2016, researchers Melvin Johnson, Mike Schuster, Quoc V. Le et al. published an academic paper entitled "#Google’s Multilingual Neural Machine Translation System: Enabling Zero-Shot Translation." In the paper, the latest achievements of Google's Neural Machine Translation (NMT) #Artificial Intelligence platform are discussed. According to Google's top #machine learning computer scientists, the Neural Machine Translation platform has developed its own internal language.
Google's A.I. poised to change life as we know it
For the first time, the artificial intelligence at the core of the platform has been able to make connections between languages with which it shares no common linkages. To discover this, the scientists analyzed what would happen if they trained their platform to translate from English to Japanese, Japanese to English, English to Korean, Korean to English. They found that the system was better than existing platforms at translating between seemingly very different language traditions built of dissimilar roots. To take things further they wanted to see if the system would be able to translate between Japanese and Korean, without using English as a linkage between them. They describe this type of A.I process "zero-shot translation" and suggest for it to be possible, the system must have developed an internal wholly separate sense of logic based meaning making, a language or dialectic framework.
To be able to establish how to correctly interpret the nuanced and complex differences between unconnected languages, the NMT platform has to have an internal way to make comparisons between words, phrases and the subtle differences between their meanings in context. According to Google's top research staff, these findings suggest their learning system has the ability to learn how to solve problems it was not originally designed to be able to. They don't quite understand the specific mechanics of how this has happened but it seems clear the project has made a lot of progress since it was first launched on September 27, 2016.
Could Neural Machine Translation be our Skynet?
Full downloadable PDF text of the original paper describing Google's Neural Translation platform called "Google's Neural Machine Translation System: Bridging the Gap between Human and Machine Translation" along with the most recent paper are available on the open source arVix database.