Swiftkey Releases Predictive Keyboard Built On A Neural Network

Building what amounts to a neural network inside a platform has long been one of the goals of computing and many a startup, but in the main they remain the preserve of university computing labs. They also tend to need a lot of computing power usually distributed via the Internet. ‘Smart keyboard’ startup Swiftkey hopes to change that today with the release of a new Alpha version of its Android keyboard which, they claim can almost literally think for itself about what you want to type next.

Until now Swiftkey’s keyboard has come packed with some preset algorithms which you can train over time. But this new version, they claim, uses artificial ‘neural networks’ to predict and correct what you are about to type in an even fuller, more accurate manner than before.

SwiftKey Neural Alpha is effectively a re-engineering of the Swiftly platform and an experimental project out of SwiftKey’s ‘skunkworks department’ called Greenhouse.

Neural networks are a ‘subfield’ of artificial intelligence inspired by the way the human brain works.

Swiftkey was the world’s first keyboard to introduce a three-word suggestion bar above the keys that could pretty accurately predict your next word, powered by its ‘n-gram’ technology which provides accurate predictions for common phrases. However, it has limitations because it can’t understand the underlying meaning of words and has to be taught.

No company has done a neural network on a keyboard before, but the keyboards available from Google and Nuance (Swype) both experiment with them for image recognition and translation. However, SwiftKey’s neural network model, they say, understands word similarity, allowing it to compare words on the fly.

Ben Medlock, Co-founder & CTO, SwiftKey, says this version is the culmination of some of the ideas put forward by British computer scientist and father of the field, Alan Turing.

SwiftKey Neural will be available in Google Play as of Thursday at 9am Pacific/12 noon Eastern here.