InnovationResearchScience

Google AI Creates Its Own Universal Language For Smart Translation

As News Editor of Silicon UK, Roland keeps a keen eye on the daily tech news coverage for the site, while also focusing on stories around cyber security, public sector IT, innovation, AI, and gadgets.

Follow on:
Google + Linkedin Subscribe to our newsletter Write a comment

Google’s artificial intelligence (AI) has created its first own language through the use of machine learning

Google’s artificial intelligence (AI) has created its first own language through the use of machine learning, allowing it to improve how it translates languages.

The search giant fired up its Google Neural Machine Translation (GNMT) back in September to aid the development of the Google Translate service.

Google found that be training the system to learn to say translate English to French and French to Italian it could create a means for the system to translate English to Italian even if it has not studied the language before.

Machine learning meets translation

google-ai-languageThe system works by analysing whole sentences of language rather than individual words then inferring similarities between them and other sentences in different languages to seek out common patterns in the languages. By doing this the system can translate a language into another even if it has not seen a translation between the languages before.

It achieves this by making use of a neural network, effectively an artificial take on how the human brain works, to spot patterns in languages that allow it to carry out translation of other languages it has no prior knowledge of. Google dubbed this process “zero-shot translation”.

The inference is that languages have phrases with similar meanings that the GMNT can pick up on called interlingua.

Looking at the geometry of the data the GNMT worked with, Google said it spotted patterns in the data that show the GNMT drawing the same meanings from three different languages.

“This means the network must be encoding something about the semantics of the sentence rather than simply memorising phrase-to-phrase translations. We interpret this as a sign of existence of an interlingua in the network,” the research team said.

It is difficult to understand how exactly the GNMT has achieved this as machine learning modules mean underlying code is dynamic and evolves with the data its fed, so spotting parts of the code that allow it to carry out this translation wizardry is difficult.

google-ai-translatorGoogle’s theory is the system effectively created a ‘universal language’ based on the common patterns if finds in languages.

Effectively, the GNMT has created its own form of system to aid its understanding of languages it has not been trained to understand. On a philosophical standpoint, this is notable as we are seeing smart computers effectively taking their own measures to understand the human race.

Google is making significant headway with its various AI development, having created AI-powered neural networks that learn to create their own encryption and create a DeepMind AI that can navigate the London Underground.

The technology industry is characterised by rapid change and populated by colourful figures. New developments are often so transformational they seem hard to believe… and in some cases natural scepticism is justified. But can you spot the fake stories from the real ones?