Automated Conversion of Natural Language Textual Specifications into Requirement Templates (RTs)

Abstract

The requirement gathering phase is the fundamental part of any software development project as it formally defines the project by generating the list of requirements that are extracted by interviewing and understanding all the stakeholders of the project. These extracted requirements are commonly specified using plain natural language text as it is universal, flexible and tends to be easier to understand. However, the usage of unrestricted natural language for specifying such requirements is proven to be ambiguous and inconsistent for deriving the actual software concepts. Therefore, requirement templates (RTs) are used to effectively specify system requirements and extract the actual concepts that need to be implemented in the project. Many such templates have been proposed in the existing literature which is helpful for effectively demonstrate the functional as well as non-functional aspects of the software system. In this thesis, the aim is to identify the way for preventing the Natural Language (NL) issues from propagating into the requirements and subsequently improve their quality. Our work started with the identification and analysis of RTs proposed in the existing literature that can help in specifying the textual requirements, followed by the selection of RTs for automated translation of requirements by applying NLP techniques. Our study revealed a total of 12 RTs and a selection of two RTs, EARS and RUPP, for the translation. We have also computed six quality metrics associated with the NL text (before translation to RTs) and the requirements translated into the RTs. Our results and transformation approach showed that the use of RTs help in improving the quality of requirements specification and useful in specifying the NL text into some restricted form based on the type of system requirements the application has.