Neural networks are one of the hottest topics in tech right now; hardly a week goes by without another announcement of their use in a new futuristic-sounding application.
At SwiftKey, we’re using neural networks to enable our keyboards to understand language at a deeper level. For the first time since we launched SwiftKey Keyboard six years ago, we’ve rebuilt SwiftKey’s language engine from the ground up using the power of neural network technology – the first instance of neural networks being used locally on a smartphone. This gives you more accurate and useful next-word predictions, saving you time and (hopefully) adding a little fun to your typing. You can download or update now on Google Play!
In October of last year, we announced the release of SwiftKey Neural Alpha, the world’s first neural network-based keyboard. We released it as an early experimental prototype on SwiftKey Greenhouse, and were thrilled to see such an enthusiastic response from users.
With this update SwiftKey is now able to meaningfully capture the relationship and similarity between words. For example, having previously seen the phrase “Let’s meet at the airport”, the technology is able to infer that “office” or “hotel” are similar words which could also be appropriate predictions in place of “airport”.
Further, it understands that “Let’s meet at the airport” has a similar sentence structure to “Let’s chat at the office”. This intelligence allows SwiftKey to offer you the most appropriate prediction or autocorrection based on the sentence being typed.
Re-architecting the keyboard to use neural networks lays the groundwork for many exciting changes and improvements to come and we can’t wait to show you more of what we’re working on!
Initially, we’ll be rolling out neural networks in our US English and UK English language models, with more languages to come, so stay tuned! Users of these languages who have the latest version of SwiftKey will be upgraded automatically.
The SwiftKey Team
Wondering how neural networks work? Check out our infographic: