Bootstrapping of a rule based English-Bangla machine translation system using work done for a sister language

Abstract

The idea of machine translation is not new and started from 17th century. The purpose of machine translation is to translate one human language to other human language. By saying human language it means- Bangla, English, Spanish, Hindi etc. Human language is ambiguas. So, a perfect machine translation system is very difficult to develop and yet to be done. Companies like Yahoo, Google etc. has their own research team researching on this issue to make machine translation more perfect. The whole idea of machine translation is not to do the perfect translation but to generate understandable translated sentences with the same meaning as the source sentences. English is the most researched language in this field and it is now possible to translate English to many other languages and vice versa. However, for Bangla no good translator is done yet. „Anubadok‟ was a good attempt but it lacks proper computational grammar for Bangla. So, I decided to work on a rule based machine translation system for English to Bangla machine translation. Recently „Center for Research on Urdu Language Processing (CRULP)‟ open sourced their English to Urdu machine translation system. As Urdu is a sister language of Bangla there is a lot of similarities between these two languages in both sentence structure and grammar. So, in my thesis semester I studied their system and figured out the way to use their system for English to Bangla translation. This report contains all steps for downloading and setting up the system, how that system works and how we can use this system for translating English to Bangla.