From Natural Language Software Specifications to UML Class Models

Software specifications are typically captured in natural languages and then software analysts manually analyzed and produce the software models such class models. Various approaches, frameworks and tool have been presented for automatic translation of software models such as CM-Builder, Re-Builder, NL-OOML, GOOAL, etc. However, the experiments with these tools show that they do not provide with high accuracy in translation. Major reason of less accuracy reported in the literature is the ambiguous and informal nature of the natural languages. In this article, we aim to address this issue and present a better approach for processing natural languages and produce more accurate UML software models. The presented approach is based on Semantic Business Vocabulary and Rules (SBVR) recently adopted standard by OMG. Our approach works as the natural language software specifications are first mapped to SBVR rules representation. SBVR rules are easy to translate other formal representations such as OCL and UML as SBVR is based on higher order logic. A case study solved with our tool NL2UMLviaSBVR is also presented and the a comparative analysis of our tools research with other available tools show that use of SBVR in NL to UML translation helps to improve the accuracy.