A Robust Approach to Identify Banglish Words using Bangla Scripts

English word written in Bangla in the Bangla script is mentioned as “Banglish” word. Frequently use of Banglish words is degrading the standard of Bangla writing. There are many types of researches have been done on Bangla Natural Language Processing but detection of Banglish words from Bangla script still no remarkable works have been done. In this research paper, we have proposed a robust system and presented how to resolve the issue of identifying Banglish words. Moreover, our designed system could also be used to change these Banglish words into standard Bangla words using NER-BERT model. After cleaning the raw input data, the system parses it into individual sentences with unique ID then the sentences are split into several words. After getting each word, the system adds tags EN for English words, BN for Bangla words and UN for others. For training the system, two corpora are used. One corpus is used for word label tagging and another is used for dictionary. After identifying the Banglish words, the system changes automatically these into standard Bangla words based on the trained dictionary. The Dataset contains 700 sentences with various Banglish words. Our proposed system achieves overall accuracy of 80% for Banlish words identification and transformation into Standard Bangla words.