부산대학교 도서관

Programmers who are new to a language find it difficult to unlearn the known language's syntax and learn the new one from scratch. To deal with this, there should be a system that translates the problem statement into a proper source code. The existing solutions for this problem are systems that works on Recurrent Neural Networks, long short-term memory architecture and other deep learning architectures whereas these technologies are inefficient as compared to the transformer architecture. Another problem with the existing solution is, it does not store the previously searched query which can be useful again in further work. To overcome the above-stated problems, Neural Machine Translation (NMT) can be used to convert the specific query from English to an equivalent Python code. Keeping this in mind, english language to python source code generation using transformers has been achieved in this work. Training the self-attention based transformer with the dataset[18] has been done. BLEU score of 0.78 has been achieved.