Neural Machine Translation System for Pakistani Languages

Neural machine translation (NMT) is the most widely used machine translation framework for bilingual machine translation and it relies on parallel corpus which is not available for low resource languages which include majority of Pakistani languages. This research entails the creation of a large scale Pakistan Multilingual Corpus (PMC) with 19 million parallel sentences for Punjabi(Shahmukhi), Punjabi(Gurmukhi), Pashto, Sindhi, Urdu and English. This study uses NMT to create sophisticated bidirectional and multilingual NMT systems for Pakistani languages. We have built bidirectional Gurmukhi to Shahmukhi NMT models (gur<->shah), bilingual and an English-centric many-to-one NMT model for Pakistani languages including Punjabi Shahmukhi as there is no NMT model exists for Punjabi Shahmukhi. Comprehensive testing and evaluation process is done to gauge the system's efficacy and accuracy.