What is a Neural Network for a keyboard?

Artificial neural networks are inspired by the original neural network - the human brain. Neural net language models are able to capture the meaning of words, versus word sequence (‘n-gram’) models, which simply look at historical usage and frequency of word use.

While SwiftKey currently uses n-grams (NLP and machine learning) within the keyboard, it’s this understanding of meaning and fuller context that sets neural network language models apart from their more primitive forms.

Since they capture more data about how words are used together, neural networks models are able to deduce the relative similarity of words - for example, Monday-Tuesday-Wednesday, home-office-gym, etc., whereas n-gram models simply observe sequences of words that have been ‘seen’ before, usually in 2-3 word chunks. The proximity observed by neural models can be visualized as clusters of words that have a certain likeness.

SwiftKey Neural Alpha demonstrates an early example of this technology in action, and we’ve already seen this lead to the biggest increase in typing accuracy since SwiftKey’s original launch.

Watch this video on the world's first neural keyboard.


Get the app!

  • To find out more about SwiftKey Neural Alpha, share feedback or request help, please check out our SwiftKey Neural VIP Forum
Was this article helpful?
1 out of 4 found this helpful
Free xml sitemap generator
Powered by Zendesk