SwiftKey is using neural networks to build a better predictive keyboard

  • By
  • October 12, 2015


Predictive keyboards can save you a lot of time when you’re tapping out messages, as long as their suggestions are on target. SwiftKey is working on a new way to up the accuracy. They’re using a neural network.

SwiftKey’s current predictive keyboard is already one of the best you’ll find in a mobile app store. From time-to-time, though, you’ve probably seen some suggestions that just don’t line up with the sentence you were typing. That’s because the current system isn’t always great at picking up on context.

As an example, SwiftKey suggests “moment” after “at the” has been typed. That’s because the current n-gram prediction system believes that moment is the word SwiftKey users choose most frequently. When you factor in the first part of the sentence, though — “I’ll meet you” — it just doesn’t fit.

[iframe id=”” align=”center” mode=”normal” autoplay=”no” maxwidth=”590″]

SwiftKey Neural aims to fix that. By leaning on a neural network to determine what words you’re likely to type next, suggestions become more natural. Instead of moment, SwiftKey Neural can tell you’re probably going to pick a location and it suggests words like office, airport, and hotel.

Eventually, SwiftKey Neural could be so good at predicting what you’re going to type next that a single word is all it needs to figure out an entire response. It may not even need that much help, and according to SwiftKey it may even be able to vary word choice to accurately capture your mood and tone.

Want to give it a try? There’s an alpha version available in Google Play that you can install right now. It’s very much a work in progress, but it’s quite usable… and an interesting look at the future of predictive keyboards.

No Comments Found

Leave a Reply