Comparative Analysis between Google Neural Machine Translation and Human Translation of Martin Wickramasinghe’s Madol Doova (Chapter IV)

Abstract

Technology is replacing many duties that have been devoted to humans for decades. Translation too was not an easy task to complete. Because of that, experts in the field were interested in developing software for Translation. During the past years, these software were getting more and more effective and advanced. This research was conducted to find out the accuracy of machine translation in literary translation of Sinhala Language. The latest technology that is used in Machine Translation is Neural Machine Translation, which uses a large artificial neural network capable of deep learning and increases fluency and accuracy when translating documents. This research was conducted to find out how much neural machine translation is close to human brain translation. This research was limited to Sinhala and English translation of a literary document. For this research, the source text was Madol Doova, one of the greatest children’s novels in Sinhala literature, written by Martin Wickramasinghe in 1947. Its English translation by Professor Ashley Halpe was used as the human translation, and it was compared with the translated output of the Google Neural Machine Translation (GNMT), which is the software used in Google Translate, taking word choice, grammar, word division, word count and most importantly, sense into consideration. As a result, it was found that there are some translation errors in GNMT on word choice and delivering the sense of the original. However, GNMT has demonstrated better performances by relying so close to the source text word order. Although it is obvious that human translation still makes more sense in Sinhala literary translation, the accuracy demonstrated by GNMT confirms that it is not departing but rather approaching the quality of humantranslation.