Applying Word Pair Model to the Chinese Syllable-to-Word Problem

Syllable-to-word (STW) conversion is a main task of Chinese Language Processing and a fundamental to syllable/speech understanding. The two major problems of STW conversion are syllable-word segmentation and homophone selection. This paper presents a word pair model (WPM) that can effectively perform homophone selection and syllable-word segmentation to improve Chinese input systems. The STW experimental results show that: (1) the WPM is able to achieve tonal (syllables input with four tones) and toneless (syllables input without four tones) STW accuracies of 99% and 93%, respectively, among the converted words; (2) the WPM is able to cover 97% (in average) tonal and toneless STW conversions of poly-syllabic words for the testing syllables; and (3) while applying the WPM as an adaptation processing, together with the Microsoft Input Method Editor 2003 (MSIME) and an optimized bigram model (BiGram), the average tonal and toneless STW improvements are 33% and 31%, respectively.