UniWise: LetterWise pour Unistroke, la prédiction de texte pour améliorer la reconnaissance de geste

Abstract

Nowadays, linguistics and language processing represent an important part of the text entry domain. Prediction and/or automatic correction systems are often included in mobile devices with text entry capabilities. However, in gesture based text entry, few such systems are known. We propose UniWise, a system which combines Unistrokes, a gesture alphabet, and LetterWise, a disambiguation system. Our purpose was to study the fusion of the results from the gesture recognizer and results of the prediction system. Our system was tested with several fusion functions. The best function showed a theorical improvement of the recognition error rate of 17,23%. An experiment was conducted to measure the impact on user performance. The result was an improvement of 17,26% in the recognition error rate in the three last sessions and an improvement of 1,04 for the position of the correct character in the recognition result list. According to user feedback, the LetterWise component was transparent to the users.