Every once in a while, we get a sneak peek into the new technology that companies are creating that will ultimately make something better, faster, or [adjective here]. Swiftkey recently launched the latest project from Swiftkey Greenhouse: Swiftkey Neural Alpha. This is the first keyboard on a smartphone that uses artificial neural networks to fix mistakes and predict words. Swiftkey currently utilizes n-gram technology to do this by looking for patterns and common phrases.

clusters-400x288

Neural Network Clusters

While n-gram technology does use context to create predictions, this new neural network-based engine goes one step further toward truly understanding what you mean. Swiftkey's blog post has a full rundown of how the technology works, and there are many videos explaining neural networks and machine learning. At the end of the day, there is only one thing you probably care about - how well does it work?

Remember that your experience may vary, but the short answer is: pretty darn well. I have used Swiftkey Keyboard almost exclusively for the last 4 years and I've gotten to the point where it's slower to use the predictions than to tap out words. While I still do this out of habit on Neural, I have found that the suggestions are accurate enough that I end up using them fairly often.

To show how the different keyboards predict words, I did a fresh install of both applications onto an old device to ensure there was no typing history or learned phrases. Here are a couple different examples I found interesting:

Note: Swiftkey Keyboard is on the left and will always be red. Neural Alpha is on the right and will always be blue.

Screenshot_2015-10-14-20-15-06 Screenshot_2015-10-14-20-15-22

This first example shows how the context of a sentence still plays a role in the n-gram technology. Both apps have very similar suggestions a good chunk of the time, but sometimes it's the little things that make you see the differences. In this next comparison we see that both keyboards have completely different words for the same sentence. Swiftkey's predictions aren't necessarily bad, but I think that Neural Alpha's are far more relevant.

Screenshot_2015-10-14-20-12-44 Screenshot_2015-10-14-20-11-36

The third example really shows how words are connected by relationships. The current keyboard thinks that swimming pools are full of either people or love. While the first option of "fun" on the Neural keyboard isn't quite right, it does give the correct answer as one of the three predictions. Hint: the correct answer is hopefully "water". If you have a pool full of anything else, you might want to get that checked out.

Screenshot_2015-10-14-20-13-20 Screenshot_2015-10-14-20-13-05

I think this last example is one of the best phrases I could find. The difference here is pretty obvious, so I'll keep it short and let you check out the screenshots.

Screenshot_2015-10-14-20-14-01 Screenshot_2015-10-14-20-13-43

There are a few changes between Neural Alpha and regular Swiftkey. The only language that is currently available in the new application is English, and there is no personalization from your SMS or other apps. There is also a notable lack of themes, but Carbon works great for me. Some of the little settings such as keyboard size and long-keypress duration are still included, so those who like to tweak their settings won't lose arguably deal-breaking features. It's best to think of Neural Alpha as an experimental engine that will hopefully show up in the main keyboard someday.

I have only been using this new application for a week now, which is not enough time to determine if it truly revolutionizes the way I type on mobile. However, my experience so far has been positive, and I will continue to use it moving forward. I am very interested to see how predictions and performance withstand the test of time, but for now, it looks like Swiftkey is on to something. If you haven't tried it yet, check out the download link below and let us know how it works for you.