Conflict Code Rate Reduction for a New Phoneme-based Chinese Input Method

Abstract

We have proposed a new phoneme-based Chinese input method with low conflict code rate, 24.7%, where the features we extract are all phonetic symbols. In this paper, we propose a modified phoneme-based Chinese input method to further reduce the conflict code rate significantly. We first propose a new reduction rule that will extract one feature phonetic symbol for each of the selected extended radicals in different manners. Two measurements will be considered to evaluate the performance of the modified input method. First, since users must recall these selected extended radicals all the time, it is required that the number of selected extended radicals must be small. Second, if we apply the new reduction rule on these selected extended radicals, the conflict code rate of the modified phoneme-based input method must be as low as possible. In the experiments, the number of selected extended radicals, n, is defined to be a small number(n=10). In the training phase, we apply an n-stage hill-climbing method to select these n extended radicals. At each stage, the hill-climbing method selects an extended radical under the constraint of minimizing the conflict code rate. After the training phase is finished, we construct a modified phoneme-based input method whose conflict code rate, 13.5%, is much lower than 24.7%.