SMT Algorithms for Indian Languages - A Case Study of Moses and MT Hub for English-Maithili Language Pair

Abstract

Around 34 million people worldwide speak Maithili. Due to lack of digital content, this language is considered resource poor in the technology and internet space. A vast majority of Maithili speakers cannot access internet due to unavailability of Maithili content on internet and also due to the fact that there is no English to Maithili Machine Translation (EMMT) system available. Creating such useful resource requires sizeable aligned parallel text corpora, divergence research between the source and target language and suitable Statistical Machine Translation (SMT) algorithms. This paper while developing the required linguistic resource for a statistical EMMT, compares the two popular SMT algorithms - Microsoft Translator Hub (MTHub) and Moses for their suitability for the EMMT system, and documents the experiments carried out on these platforms.