Graphemic Approximation of Phonological Context for English-Chinese Transliteration

Although direct orthographic mapping has been shown to outperform phoneme-based methods in English-to-Chinese (E2C) transliteration, it is observed that phonological context plays an important role in resolving graphemic ambiguity. In this paper, we investigate the use of surface graphemic features to approximate local phonological context for E2C. In the absence of an explicit phonemic representation of the English source names, experiments show that the previous and next character of a given English segment could effectively capture the local context affecting its expected pronunciation, and thus its rendition in Chinese.