Language to Code with Open Source Software

Abstract

With the development of deep learning, it has been applied in various field of computer science. Generating computer executable code from natural language descriptions is an urgent problem in the artificial intelligence. This paper proposed a solution based on deep learning for code generation. Encoder-Decoder model is used in our method to convert natural language description into target code. Because of the rapid development of information technology, all aspects of software resources have been greatly enriched. The deep learning model we designed takes the natural language description as input and generates the corresponding object code by extracting the code from the open source software library. We collected natural language descriptions of 20 problems that undergraduate students often encounter in their daily programming. Experimental results show that our method is practicable. Our approach also provides a good idea to extract useful code from open resource for code generation.