Language is something that definitively falls within our domain. No matter where you are on the globe or the level of education you've achieved, there is always a basic method of communication and the possibility of translating its content. On the other hand, computers see language as one of their toughest challenges, and in these days of prediction and auto-correction, the difficulty multiplies. For the folks at SwiftKey, the solution is a new virtual keyboard backed by a neural network.
In the last decade, the traditional mobile phone gave way to the smartphone and lost most of its keys along the way. With the arrival of touch screens, the way we communicate with others also changed, and from a certain point of view, the process became a bit more complicated. However, part of the solution was inherited through the phones themselves: the predictive dictionary, which many of us knew as T9. Needless to say, its early versions were a disaster, and it was more comfortable for users to leave it disabled. Even today, the auto-corrections and predictions made by some keyboards challenge reason, giving rise to all kinds of jokes and absurd situations. In other words, there is plenty of room for improvement, and with that in mind, the people at SwiftKey believe the solution is a neural network.
A Keyboard That Thinks in Context
And so we meet the brand-new SwiftKey Neural Alpha, a new virtual keyboard for Android that bases its recommendations on the results of a dedicated neural network. Its developers recognize the special affinity the human brain has with language, and thus came to the conclusion that a neural network should be better equipped to handle the most natural problems, even though a neural network is insignificant compared to the human brain. The key is in the context and the indirect relationships that one or more words might have. As an example, SwiftKey mentions that a phrase like 'let's meet at the airport' can serve as a reference for understanding that other words are used in the same way, among them 'hotel' or 'office', and for detecting structures with similar meanings.
Looking Ahead: Whole Messages
With this kind of flexibility in its favor, the new SwiftKey Neural Alpha should be able to provide much more coherent and useful predictions for the user, but that's not all: SwiftKey is considering a future possibility, extending its predictions to whole messages. If a neural network accumulates enough knowledge to understand the full context of a message, creating an appropriate response shouldn't be that complicated. Of course... there's a long way to go. As its name suggests, the keyboard is in a very delicate alpha phase and is expected to make mistakes. The minimum support for SwiftKey Neural Alpha is Android 4.4, and so far, it only predicts in English.
