An Approach for Improving Thai Text Entry on Touch Screen Mobile Devices Based on Bivariate Normal Distribution and Probabilistic Language Model

Abstract

This paper presents an approach to improve the correctness for Thai text inputting via virtual keyboard on touch screen mobile phones. The proposed approach is to generate candidate character based on statistical model from bivariate analysis of pre-collected coordinate data and apply the character trigram model to each candidate character sequence. From user’s touch positions, a set of candidate characters with high position-based probability is generated. Then, the character trigram model is applied to each generated candidate characters sequence. For each character sequence, a probability is computed from the weighted combination of position-based and character trigram models. In the end, the character sequence with the highest probability is selected to be the most appropriate sequence. Experiments were conducted to compare the typing accuracy between an ordinary Thai virtual keyboard and our proposed algorithm using the same Thai keyboard layout. Results demonstrate that the proposed algorithm provides the improvement in the text entry accuracy in both character levels and word levels.