A syllable-based name transliteration system
NEWS '09 Proceedings of the 2009 Named Entities Workshop: Shared Task on Transliteration
Hi-index | 0.00 |
This article describes the first trial on bidirectional Thai-English machine transliteration applied on the NEWS 2010 transliteration corpus. The system relies on segmenting source-language words into syllable-like units, finding unit's pronunciations, consulting a syllable transliteration table to form target-language word hypotheses, and ranking the hypotheses by using syllable n-gram. The approach yields 84.2% and 70.4% mean F-scores on English-to-Thai and Thai-to-English transliteration. Discussion on existing problems and future solutions are addressed.