Text prediction technology saves 100,000 years in mobile phone typing time

Credit: Meghan Schiereck/Unsplash

Research on how to program computers to process and analyse large amounts of natural language data has led to the saving of 100,000 years in mobile phone typing time and also enabled language learners to have feedback in less than 15 seconds.

Ted Briscoe and team’s research on language modelling and classification was critical to the development of SwiftKey, a keyboard app designed to pick up the patterns in people’s language use and, through machine learning techniques, accurately predict what they were most likely to write next.

Language data was gathered by the researchers and a technology and knowledge transfer company cofounded by Briscoe, iLexIR, provided the data to the 2009 start-up (co-founded by ex-Cambridge students Ben Medlock and John Reynolds) that became SwiftKey. The data was used to train the language prediction models, which says the co-founder and CTO of Switfkey “made a huge difference at a critical early stage”.

SwiftKey was first released as an exclusive for Android Market in July 2010, followed by an iOS release in September 2014 after Apple allowed third-party keyboard support.

By June 2014, users were estimated to have saved half a trillion keystrokes. By 2016, SwiftKey software was installed on more than 300 million devices and it was estimated that its users had saved nearly 10 trillion keystrokes, across 100 languages, amounting to more than 100,000 years in combined typing time.

The success of SwiftKey led to its acquisition by Microsoft in 2016 for $250 million, and the software was incorporated into Microsoft’s flagship mobile app, SwiftKey Keyboard for iPhone and Android (500 million installs by 2020).

The researchers’ expertise on automatic feedback and assessment of texts has also helped language learners and reduced the workload of teachers. Within a year of the launch of the free assessment software ‘Write & Improve’, it was used by over 650,000 people in 225 countries with well over a million pieces of writing submitted and checked. Learners obtain feedback on their texts in fewer than 15 seconds, helping to make language learning more efficient for learners and teachers.