Machine Translation System Using Deep Learning for Punjabi to English

Machine Translation (MT) is ongoing research from the last decades. Research started with a simple word to word replacement from source (e.g. English) to target language (e.g. Hindi). Then research moved to statistical-based machine translation (SBMT) which is based on the parallel corpus. Now from the last few years, deep learning is used to develop an MT system. In this paper, an artificial neural network-based machine translation (ANMT) system is trained and tested for Punjabi to the English language. To train the proposed system a parallel corpus of Punjabi-English language is prepared and based on this corpus three models for Punjabi to English NMT system have been developed. In this work, the BLEU score is used to evaluate the performance of the system. The proposed system had shown a BLEU score of 36.98 for smaller sentences, 34.38 for medium sentences and 24.51 for large sentences using model 1, BLEU score of 36.62 for smaller sentences, 35.51 for medium sentences and 26.61 for large sentences using model 2, BLEU score of 60.68 for smaller sentences, 39.22 for medium sentences and 26.38 for large sentences using model 3.