Low Resource and Domain Specific English to Khasi SMT and NMT Systems

Abstract

Machine translation systems for low resource languages face challenges in terms of quality and understanding. Our work focus on the translations for English to Khasi using two epitomes of translations using statistical and neural machine translation approaches. As part of this system development, we built an English-Khasi parallel dataset from existing domain specific literature. The quality of translations of statistical machine translation (SMT) and neural machine translation (NMT) systems for low resource and domain specific setting are substantially analyzed considering automatic and subjective evaluation techniques.