Current - Issue

Linguistic diversity in India with 22 official languages and hundreds of dialects possess difficulty to access digital education. Although due to advances in machine translation there is quite improvement in multilingual communication but existing tools such as Google Translate often fail in the educational domain where accuracy and semantic precision are critical. Given the growing need of digital education, it is imperative to spot the difficulties faced by students of different diversities. In order to improve multilingual communication in education domain, a comparative evaluation of Transformer based multilingual models mainly mBERT, IndicBERT and M2M-100 for translating educational texts from English into Indian regional languages has been presented in this study. To evaluate model performance we have curated a domain specific bilingual dataset derived from NCERT and state level textbooks. The models were fine tuned and assessed using BLEU, METEOR and TER metrics supplemented by human evaluation of fluency and semantic correctness. Evaluation done in this paper will lead to the design of AI driven multilingual e-learning systems that can enhance educational accessibility in rural and underserved systems.