Bridging Language Barriers in Coding: An NLP-Powered Tool for Programming Education in Regional Languages
Goutami Sooda, Arya Hariharan, Ahana Patil, Arya Vinod, S Chandana, L. B. Manvith, Nihar Mandahas, H. Siri, G Shobha, N. Deepamala, Jyoti Shetty · IEEE Access · 2025
In the digital age, coding has become a critical skill for improving career prospects and participating in the global economy. However, most educational tools primarily cater to English-speaking learners, limiting access for individuals in rural areas who are proficient only in regional languages, such as Kannada. This study presents an interactive web-based tool that leverages Natural Language Processing (NLP) to teach fundamental programming concepts in Kannada. The tool allows users to input algorithmic statements in Kannada, which are then translated into English and converted into Python code using a fine-tuned Salesforce CodeT5 model. The model, trained on a custom dataset, achieved strong performance with a BLEU score of 87.64 and an Exact Match score of 81.62%. The study also evaluated other models such as T5, PLBART, and GPT-Neo for the same task. Additionally, the platform offers tutorials, quizzes, and an embedded coding environment for learners to test the generated code. A user study involving Kannada-speaking students from diverse academic backgrounds was conducted to evaluate the usability and educational effectiveness of the tool. Results indicated that the majority of participants found the tool easy to use, the code generation accurate, and the overall experience beneficial in understanding programming logic. Notably, over 65% of students without prior programming experience expressed interest in pursuing programming careers after using the tool. By addressing language barriers, this research contributes to United Nations Sustainable Development Goal 4 (SDG 4) by fostering inclusive and equitable quality education, enabling learners to acquire programming skills in their native language.