The current scenario of Software Development evolves from single programming language development framework (vis Java, C#, Php, Python, Java Scripts etc) to multi language micro services based architecture. For Software Products there are plethora of low level and high level programming languages. The challenge is to create a single software defect predictor model which is common to language as well as project. The Idea of conceptualizing this work emphasized that, Similar to natural language, every programming language also have linguistics characteristics like syntax, semantics, pragmatics and grammars. The technique of Natural Language Processing (NLP) is one of the oldest areas of machine learning research and is employed in significant fields such as machine translation, speech recognition, sentiment analysis, and various other text processing (Kumar & Singh, 2020). In this paper we have leveraged the concept of NLP in (Deep Learning Network) DLN to convert the Source code into sequence of lexer and parser classes based on the defined grammar, fixed length feature vectors classes are passed into embedded layer of Advance Long short-term memory (A-LSTM) Network. This Network can learn the linguistic pattern in source code, then it is used to predict the defective modules in the projects regardless of programming language. The results outperformed as compare to hand crafted software matrices based DLN in identification of buggy modules in software projects.

Software Defect, Natural Language, LSTM, Software Metrics, Software Quality