Khmer-Chinese person name transliteration based on nonparametric Bayesian model and conditional random fields

Abstract

Transliteration is an important foundation of cross lingual information processing technology. In order to solve the transliteration problem of Khmer person name, a syllabification model of Khmer person name with Dirichlet process was proposed in the paper. Dirichlet process model was employed to syllabify the Khmer person name. We used Conditional random fileds (CRF) to prevent the possible bias in transliteration, and fused the contextual syllable features and contextual string features and the annotation transferring features. Syllabified Khmer person names and corresponding Chinese person names were used as training data for CRF transliteration model. The results showed that transliteration results based on CRF was better than that based on the Maximum Entropy, the accuracy reached to 46.25%.