A Novel Framework for Ancient Text Translation Using Artificial Intelligence

Abstract

Ancient script has been a repository of knowledge, culture and civilization history. In order to have a greater access to the valuable information present in the ancient scripts, an appropriate translation system needs to be developed keeping complexity and very less knowledge of the script available in consideration. In this study, a translation and prediction system has been implemented using Artificial Intelligence. The training has been developed using Sunda-Dataset and self-generated dataset, whereas the translation from ancient script viz. Sundanese script to English text is done using two layers Recurrent Neural Network. The technique used is compared with an existing translator called IM Translator. The results shows that the BLEU score is increased by 8% in comparison to IM Translator further WER is decreased by 10% in contrast to IM Translator. Furthermore, the N-Gram analysis results indicate 3% to 4% increase in 100% contrast value.