An Approach to Extract a Conceptual Model from Natural Language Specifications Using NLP and IA

Abstract

Requirements engineering is a critical phase in software development. Errors committed in the requirements, if not found and corrected early in the engineering process, become costly problems later on. Artifacts described in natural language are the most suitable for both parties that participate in the process: experts of the domain and the software development team. Nevertheless, natural language can have some issues because the diversity of participants writing and reading specification. Thus, it is necessary to guide and assist the production of these artifacts. Kernel sentences is a linguistic definition about small sentences (with only one verb) written in active voice. We argue that kernel sentences are adequate to use in the acquisition of knowledge. Then, conceptual model are a good technique to summarize a large and complex specification to provide an overview to the whole group of participants. This paper proposes an approach to acquire specification collaboratively checking the writing style to conform kernel sentences, and synthetizing a conceptual model to summarize the specification to help participants to have an overview of the knowledge gathered in order to provide a consistent and coherent model. The process proposed has two main activities: acquisition of the knowledge checking its style and synthesis of the conceptual model. This paper also describes a prototype that uses natural language processing and artificial intelligence tools to support the process. Finally, the paper shows the result of a preliminary evaluation with promising results about the applicability of the process.