An Approach for Improving Thai Text Entry on Touch Screen Mobile Phones Based on Distance and Statistical Language Model

Abstract

This paper introduces an approach for improving Thai text inputting via virtual keyboard on touch screen mobile phones. We propose a technique for character candidate selection to choose the most proper character sequence. The proposed approach consists of two main parts i.e. candidate character generation based on touch area, and trigram model for candidate selection. The candidate generation is used to acquire more than one character per one touch. We define the area nearby the touch point as a touch area. The area is defined based on the statistics collected from various users. The characters within the touch area are considered as a set of candidate characters. The candidates are weighted based on their distances from the touch point. Then, the character trigram model is applied to select the combination of characters with the highest probability suggested to user. In order to evaluate the performance of the proposed technique, we evaluate the word-level accuracy of the words generated by the proposed technique and the words generated by combining the characters with the shortest distances from the touch point. Our technique yields better performance. It produces the results with the accuracy of 93.29% in character level and 66.67% in word level, while the accuracy is 83.48% in character level and 48.33% in word level when the nearest characters are selected.