The IIIT-H Gujarati-English Machine Translation System for WMT19

Abstract

This paper describes the Neural Machine Translation system of IIIT-Hyderabad for the Gujarati→English news translation shared task of WMT19. Our system is basedon encoder-decoder framework with attention mechanism. We experimented with Multilingual Neural MT models. Our experiments show that Multilingual Neural Machine Translation leveraging parallel data from related language pairs helps in significant BLEU improvements upto 11.5, for low resource language pairs like Gujarati-English