I think one of the weakest links in the Android experience is the keyboard. There are various alternative keyboards out there and you’ve probably seen the Samsung commercials promoting Swype. Well UK-based TouchType‘s SwiftKey is more compelling.
The SwiftKey keyboard itself is nothing special; the action is all behind the scenes in its text-prediction engine. It has been available in beta for several months and, without any publicity, has seen about 250,000 downloads. A new version is out this morning.
In an email to me yesterday CMO Joe Braidwood explained the research and guts behind the text-prediction capability:
We partnered with a group of Cambridge engineers (one who is now our CTO and co-founded TouchType) who [used the] Large Hadron Collider’s supercomputer grid to . . . analyze the relationship between words. From this, we generated a powerful statistical language processing engine that understands the probability of different words occurring one after another, and added a very dynamic machine-learning layer to it, to understand individual versus generic language habits. The marriage of these two technologies we call our Fluency prediction engine, and it’s what drives SwiftKey.
So, when you or anyone picks up SwiftKey on their phone and downloads a language module, they’re using a very refined product that started life as an enormous amount of data collected by the LHC’s supercomputers. We think that’s pretty cool.
I’ve just used it for the past 24 hours and it works well. In some cases you don’t even need to start typing letters on the keypad, it simply offers up the next words in a predicted sequence. This is kind of an amazing event when it happens.
Here’s a video demo of how it works:
There’s a larger story here, which isn’t going to show up in most of the posts that will appear this morning lauding the Android keypad experience. That larger story is how the technology behind SwiftKey, the Fluency engine, can be used across devices and appliances. For example, this could be integrated into Connected TV (Apple TV, Google TV) or in-car systems to complement voice commands.
It’s the database and analysis in the background which is the real story not the particular Android keypad implementation.