ε-extension Hidden Markov models and weighted transducers for machine transliteration

Abstract

We describe in detail a method for transliterating an English string to a foreign language string evaluated on five different languages, including Tamil, Hindi, Russian, Chinese, and Kannada. Our method involves deriving substring alignments from the training data and learning a weighted finite state transducer from these alignments. We define an ǫ-extension Hidden Markov Model to derive alignments between training pairs and a heuristic to extract the substring alignments. Our method involves only two tunable parameters that can be optimized on held-out data.