Hierarchical modeling to facilitate personalized word prediction for dialogue
Richard G. Freedman, Jingyi Guo, William H. Turkett, V. Paúl Pauca · 2013
The advent and ubiquity of mass-market portable computa-tional devices has opened up new opportunities for the devel-opment of assistive technologies for disabilities, especially within the domain of augmentative and alternative communi-cations (AAC) devices. Word prediction can facilitate every-day communication on mobile devices by reducing the phys-ical interactions required to produce dialogue with them. To support personalized word prediction, a text prediction sys-tem should learn from the user’s own data to update the ini-tial learned likelihoods that provide high quality “out of the box ” performance. Within this lies an inherent trade-off: a larger corpus of initial training data can yield better default performance, but may also increase the amount of user data required for personalization of the system to be effective. We investigate a learning approach employing hierarchical modeling of phrases expected to offer sufficient “out of the box ” performance relative to other learning approaches, while reducing the amount of initial training data required to facilitate on-line personalization of the text prediction sys-tem. The key insight of the proposed approach is the sepa-ration of stopwords, which primarily play syntactical roles in