This site uses cookies for analytics, personalized content and ads. By continuing to browse this site, you agree to this use.

Subscribe

Introducing the world’s first neural network keyboard

8th October 2015

Today, we’re excited to announce SwiftKey Neural Alpha – the world’s first smartphone keyboard software that uses an artificial neural network to predict and correct language.

This is a big step forward for mobile typing, offering a keyboard that makes smarter and more meaningful word suggestions in the context of your sentences. SwiftKey Neural’s underlying technology is based on the use of neural networks, a subfield of artificial intelligence inspired by the structure and operation of the human brain. It’s available now for smartphones on Android 4.4 (KitKat) and above as an early-stage project in SwiftKey Greenhouse.

Please allow cookies to watch this video

Since SwiftKey Keyboard launched on Android in 2010, it has defined the standard for predictive keyboards. It was the world’s first keyboard to introduce a simple, three-word suggestion bar above the keys that could accurately predict your next word. This was powered by our word sequence ‘n-gram‘ technology, an approach now used on more than a billion devices globally.

N-gram technology provides accurate predictions for common phrases and those learned from you. However, it has some limitations, as it can’t capture the underlying meaning of words and can only accurately predict words that have been seen before in the same word sequence. SwiftKey Neural’s intelligent understanding of sentence context introduces a more ‘human’ touch for mobile typing.

:: How does it work? See our neural network infographic ::

clusters

Neural networks ‘cluster’ words together

Through machine learning based on enormous amounts of language data, SwiftKey’s neural model is able to meaningfully capture the relationship between words. It understands word similarity, allowing it to compare words on the fly. Within the neural model, words can be visualized in ‘clusters’, located at varying degrees of proximity to one another.

This understanding allows SwiftKey Neural to predict words that have never been seen in a given sentence context during the learning phase. For example, having seen the phrase “Let’s meet at the airport” during training, 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 Neural to offer the most appropriate word based on the sentence being typed.

Comparison-1-6

SwiftKey ‘classic’ vs SwiftKey Neural – how the predictions compare

Until now, neural network language models have been deployed mostly on large servers, requiring significant computational resources. The launch of SwiftKey Neural Alpha is a breakthrough as it marks the first time this type of language model technology has been engineered specifically to operate locally on a smartphone keyboard – a huge challenge given the resource constraints.

:: Why Turing’s legacy demands a smarter keyboard – Read SwiftKey CTO Dr. Ben Medlock’s post on Medium ::

This is the first step in our exploration of neural networks for touchscreen keyboards. We envision that this powerful technology will fundamentally change the way we type on smartphones over the next five years. Rather than solely predicting your next word, we’re working hard on building a framework for the keyboard of tomorrow. It’s just possible that using this approach the keyboard of the future could know you so well, it’d accurately predict your entire messages for you, in your tone of voice, reflecting the events that go on in your daily life (see concept keyboard below).

Concept 'keyboard' of the future

Keyboard of the future? Neural nets could make message prediction possible (click to enlarge)

SwiftKey Neural Alpha provides an early example of this technology in action. It has already enabled the biggest leap in typing accuracy that we’ve seen since the original introduction of SwiftKey in 2010, which offered groundbreaking next-word prediction capabilities.

We’re only scratching the surface of what’s possible with this technology. We started SwiftKey because we believed a revolution in typing productivity on smartphones was long overdue. Today, we’ve launched the next step in our commitment to radically improving that experience, but it doesn’t end here. Give the new alpha a try and please send us feedback — it’ll help us refine what we’re building so we can continue to challenge the status quo.

Thanks,
The SwiftKey Team

neural-net-infographic