Applying Weighted Finite State Transducers and Ripple Down Rules for Myanmar Name Romanization

Abstract

Romanization known as Latinization that refers to the representation of names with Roman (Latin) alphabets. The Romanization process is not trivial especially for Myanmar Language (Burmese) due to different Roman variations of a single Myanmar proper noun. The Myanmar Name Romanization project has so far developed a 55K Romanized word pairs of Myanmar personal names. We apply Weighted Finite State Transducers (WFST) and Ripple Down Rules (RDR) based tagging approaches to the task of Myanmar name Romanization and vice versa. We perform experiments on one closed test data set (5,000 of training data) and three open test data sets: 5,000 Myanmar personal names, 571 city and town names of Myanmar, and 292 Myanmar food names. The result shows that RDR approach gives better performance than WFST for open test data sets.