Uncertain Text Entry on Mobile Devices

Modern mobile devices typically rely on touchscreen keyboards for input. Unfortunately, users often struggle to enter text accurately on virtual keyboards. To address this, we present a novel decoder for touchscreen text entry that combines probabilistic touch models with a long-span language model. We investigate two touch models – one based on Gaussian Processes that implicitly models the inherent uncertainty of the touching process and a second that allows users to explicitly control the uncertainity via touch pressure. Using the first model we show that character error rate can be reduced by up to 7% over a baseline, and by up to 1.3% over a leading commercial keyboard. With the second model, we demonstrate that providing users with control over input certainty results in improved text entry rates for phrases containing out of vocabulary words.