Abstract:

Automatically extracting diagrams from natural language requirements is the challenge of all researchers in the field of Requirements Engineering (RE). Since software engineering projects are becoming more complex and stokeholders are more demanding, the need for modeling is obvious. In the context of requirements engineering, we need to model textual requirements. However, there are several gaps in existing approaches, which reduce their effectiveness. In this paper, we will discuss all the drawbacks of existing approaches and we will focus on reducing the gap between unstructured natural language requirements and formal representations, particularly representation with Unified Modeling Language (UML) diagrams. With our proposed solution, we aim to improve the quality of the treated requirement, because it strongly affects the quality of generated diagrams. We will develop a requirement processing based on two steps: i) A classic Natural Language Processing (NLP) combined with syntactic rules; ii) A deep processing that generates a typical textual representation based on a sentence template named GGFS template (Generic Grammatical Form Sentence template) and heuristic rules. The typical requirement is a set of sentences having the same grammatical form inspired by the user stories format. It has the role of a userinterface, which facilitates the intervention of stokeholders to validate the obtained requirements. We expect to have a smooth and effective transformation of requirements into UML diagrams using typical requirements.