Neural Networking Being Used To Improve Google Translate

March 11, 2016 - Written By Daniel Fuller

The field of artificial intelligence has made some incredible strides over the years, from IBM’s Watson all the way up to Deepmind’s AlphaGO and the everyday A.I. behind products like Google Now. While it’s far from perfect, it has still experienced incredible breakthroughs over the years. One such breakthrough, called neural networking, basically allows an A.I. to use experience and a cooperation of different intelligent nodes to help predict outcomes that are generally thought of as unpredictable, as well as determine how best to react to a given situation. This type of cloud-sourcing allows advanced operations and faster processing that just wouldn’t be possible with a single terminal point working alone, using only data that was built or fed into it. Neural networking has proven useful in all sorts of situations, with translation now being looked at as a possible addition to that list, according to Googler Jeff Dean, part of the Google Brain team.

Speaking at the 2016 Structured Data conference in San Francisco, Dean said that his team is working with the team behind Google Translate to look into the feasibility of integrating neural networks into the product. Right now, the new data, inspired by a 2014 research paper, is in testing with Translate and has not quite reached a level of stability and quality that would allow it to go into the mainstream product for the moment. He later said that he predicted “good results down the line” and that, in the future, the Google Translate app may see an update that increases its use of neural networks and brings it a bit away from the crowdsourcing and more conventional methods that the app currently uses.

Thanks to the breakneck pace of advancements in the neural networking and machine learning fields of A.I., a researcher can output their findings and data and, within a couple of days or even hours, end up seeing their findings used in a wide range of products, both backend focused and consumer facing. With the field of A.I. continuously advancing like it has been in the past five years, the range of possible applications is quite overwhelming.