SMT Versus NMT: An Experiment with Punjabi–English

In this paper, a comparison of the Statistical machine translation (SMT) and neural machine translation (NMT) for Punjabi to English in the fixed domain of health and tourism is provided. We have tried to answer does NMT perform equivalent well or better with respect to the SMT system? We have developed the three base models viz., SMT-based model using the Moses toolkit, followed by long short-term memory (LSTM) model and bidirectional LSTM model using the OpenNMT toolkit. All three models used the Punjabi–English parallel corpus of TDIL health and tourism. Finally, the quality of translation is validated using the automatic parameter like the BLEU score and the TER score. We observed that bidirectional LSTM performs better than simple LSTM an SMT system.