Communication by text -- e.g., email, SMS messaging or social media -- is a large and increasing part of how people interact every day, which places demands on how quickly and comfortably text can be entered on devices. In this talk, I present some text entry scenarios that are somewhat more challenging than standard typing in a Latin alphabet, and discuss how natural language modeling can improve input efficiency and/or accuracy. First, I will discuss providing text entry (and text-to-speech synthesis) for individuals with severe motor disabilities, when text entry becomes the principal or sometimes sole communication modality. In particular, I will cover some earlier work on using Huffman codes to improve the efficiency of text input when using a single binary switch and a visual scanning interface, as well as more recent applications in brain-computer interfaces. Next, I will discuss text entry on mobile devices, and present some recent work on challenges in mobile text entry in Asian languages, including methods to support romanized input (e.g., using QWERTY keyboard) of languages with other native writing systems, such as the Devanagari script for Hindi. In all of these scenarios, statistical models of natural language sequences provide effective solutions to tricky computer interface problems.
Brian Roark is a research scientist at Google since 2013.
More information: http://www.lanzaroark.org/brian-roark