An Extensive Approach of Converting Chittagonian to Standard Bangla

Abstract

The demand of language translation is growing rapidly. Modern technology is offering the translation service for several languages. But the translation of different Bengali dialects is not very well explored filled in NLP. The goal of our work is to develop a proper conversion system which will convert the local form of Bangla spoken by the people of Chittagong to the standard Bangla language. The dialect of Chittagong district is one of the most different dialect in Bangladesh. We will use Natural Language Processing and Machine Learning to build an efficient system to convert the dialect to the standard form of Bangla language. So that a person who is not from Chittagong can easily communicate with the locals of Chittagong. To build this system we have used an LSTM based encoder decoder model. This model consists of LSTM, Embedding, Repeat Vector, Time Distributed dense layers. We have achieved 99.3% model accuracy for training and 72.5% model accuracy for testing sets. The accuracy we have achieved from our system is quite good. But the accuracy can rise if we can reduce our data loss more. As the data loss and the accuracy of our work is totally depending on the volume of the dataset. With time we will increase the volume of our existing dataset.