Proposal of an Approach to Generate VDM++ Specifications from Natural Language Specification by Machine Learning

A natural language contains ambiguous expressions. The VDM++ is one of the methodotogies on the formal methods to write the specification without ambiguity. It is difficult to write a VDM++ specification, because VDM++ is written by strict grammar. This research proposes an approach to automatically generate the VDM++ specification by machine learning. This approach defines four data structures and has four processes. In this paper, variables and only real type in the VDM++ specification are generated automatically by this approach. In order to generate the variables and real type, it is necessary to extract the noun corresponding to the variable from the natural language specification. Consequently, our proposed approach can generate a VDM++ specification and we have confirmed that the generated VDM++ specification is grammatically correct.