To document software and business requirements, it is desirable to use natural language text which can be processed by computer (as is done e.g. by RavenFlow software). We argue that requirements and their application domain model (including ontology) are interdependent, and that by using the domain model we can parse and semantically analyze the requirements more effectively. To this end we develop a type system for application domain entities that includes multiple inheritance, as well as specializable and negated types. We integrate it into the HPSG grammar formalism and extend the Minimal Recursion Semantics by adding new predications and constraints. We show how, by informing the parser with criteria to reject sentences on the basis of type mismatch, our system can less ambiguously translate natural language sentences into a Domain-Typed First Order Logic format, through pruning some semantically incorrect options within the search space.
Copyright is held by the author.
Member of collection