On-screen keyboards are never quite right, and while auto-correct errors are lessening, sometimes they just don’t connect with you, but what if the engine was modelled on a human brain?
One of the leaders in virtual keyboards is trying that very concept with the creation of an algorithm based on a neural network to predict how you will type.
Not sure what that is? No worries, let’s explore.
Regular word guessing keyboards work by watching your words as you type them, and usually hinge on the last few that have been entered. For instance, “to be or” might yield “what”, “not” or “the” to be the next words picked in a typical keyboard test, with the engine using learned logic and common phrases to guess this.
Neural networking, however, is an approach made by understanding how our brains put two and two, essentially making an algorithm that works in a similar approach to our brains, hence the whole “neural” side of things.
For a keyboard, this means understanding words and how they fit together, and specifically targeting the level of individuality every writer and typist brings to the table.
We’re not quite on the specifics of how SwiftKey’s Neural app works, but in alpha, it’s a preview of the way a keyboard could behave if it were mimicking the way our brains work, so does it do our brains justice?
In the first few days of using the app, not really, though that’s because you have to give SwiftKey Neural the chance to get to know you.
One could argue that those first few days of using the app are a little like taking the app out on a date, doing the whole thing as if it were a romantic shindig where you were trying to get to know it, because that’s what it feels like.
After those few days have passed by where you get to know the neural keyboard, it starts to pick up on how you talk to people, guessing suggested words a little better, and allowing you to press the suggestions a little faster than you might otherwise.
Interestingly, while SwiftKey first clued us onto the idea of connecting our Google accounts, Twitter, and other social networking systems into its network to work out how we write, SK Neural pulls back on that, letting its artificial brain do the math and viewing more than just the last few words, but the overall structure of how you put your sentences and phrases together.
Just like its less-brain-like sibling (SwiftKey Classic), SwiftKey Neural supports the gesture typing, so you can drag your finger over letters quickly and get the words written faster, and there’s even support for Emoji if you need it.
Keyboard sizes can also be adjusted, which is handy, and Google Voice support is also there, letting you speak your words and have the Google engine take them and spin them into text.
But the keyboard is the most interesting part of this working checking out, particularly because its suggestion system is modelled on how we think, and as time goes on, this could be handy for coming up with your responses before you even have to write them in.
Right now, it’s not the keyboard to replace your favourite on-screen set of keys, but if you like seeing a glimpse of the future, SK Neural is definitely worth the effort, especially since, you know, it’s free.