Statistical Machine Translation, Ripple Down Rules and Hidden Markov Model for Burmese Romanization

Burmese Romanization is the phonetic translation of Burmese (Myanmar) language into their phonetic Latin script. Burmese Romanization is aimed the reader who is unfamiliar with the Burmese script to pronounce the Burmese language reasonably accurately. We developed a 15K Burmese Romanization parallel corpus based on the practical usage of several Romanizations of native Myanmar people. The experiments were performed using different SMT approaches (PBSMT, HPBSMT, OSM), Ripple Down Rules (RDR) and Hidden Markov Model (HMM) approaches to the task of Burmese Romanization. The results show that the OSM approach achieves the highest performance in Burmese to Romanized Burmese direction. On the other hand, HMM approach gives the highest results in Romanized Burmese to Burmese.