Third-party keyboards on Android have pretty much followed a standard approach for predicting what you’re going to type next all these years. The usual method involves analyzing the last couple of words you typed or your saved history and suggest matched results based on that. However, Swiftkey, the six-year-old company which now works under Microsoft, is about to change how these applications function. The latest Swiftkey update on Android has replaced the outdated probability-based algorithm with artificial neural networks.


Back in last year October, Swiftkey unveiled an experimental application that was entirely developed through artificial networks that were designed to predict what you’re going to type next based on the context instead of what you typed a second ago. Now, after months of testing and changes, the main Swiftkey app has been reengineered by implementing those networks at the core for providing smarter and more importantly, natural recommendations. To be slightly more precise, the ditched N-gram probability model read the last two words you wrote, scrutinized the massive database, and displayed the three best-suited words. That two words margin can create some substantial shortcomings as it isn’t reading the entire sentence.


For instance, if you typed “I’ll meet you at the”, through the probability architecture, Swiftkey would come up with “moment” or “end” because it is inferring only “at the”. With the neural approach, it will show results such as “airport” or “office”. Think of it as nerves that run in your brain. The problem could’ve been partially solved by increasing the number of words the app examined, however, the database grows exponentially which could lead to significant performance issues. As far as personalized suggestions ( saved history, social media links ) are concerned, Swiftkey hasn’t entirely removed the n-gram engine, hence, it will still come into play whenever required. The update is now live on the play store free of cost, so go ahead and give it a whirl.

Was this article helpful?