Statistical Keyboard Decoding
Document Type
Book Chapter
Publication Date
8-18-2022
Department
Department of Computer Science
Abstract
Text entry is a core task in our daily interaction with computers. Entering text using a keyboard remains the de facto standard due to its familiarity and efficiency. However, new interaction settings and devices make conventional text entry using a keyboard more challenging due to higher levels of uncertainty in detected user interaction events. For example, entering text on a smartwatch using a very small keyboard layout naturally results in less accurate touches than can be expected when entering text on a smartphone or physical keyboard. Fortunately, statistical keyboard decoding provides a technique for inferring the user's intended text from their noisy input. The approach leverages Bayes' Rule to help identify the most probable word given a model of the user's uncertain touch interaction and known language regularities. This chapter provides an overview of statistical keyboard decoding and examines the various design parameters which are known to dictate its performance. Two illustrative case studies are also presented which demonstrate how statistical keyboard decoding can enable efficient text entry in challenging interaction settings.
Publication Title
Bayesian Methods for Interaction and Design
Recommended Citation
Gaines, D.,
Dudley, J. J.,
Kristensson, P.,
&
Vertanen, K.
(2022).
Statistical Keyboard Decoding.
Bayesian Methods for Interaction and Design, 188-211.
http://doi.org/10.1017/9781108874830.009
Retrieved from: https://digitalcommons.mtu.edu/michigantech-p/16929