NLG4RE: How NL generation can support validation in RE

Context and motivation: All too frequently functional requirements (FRs) for a (software) system are unclear. Written in natural language, FRs are underspecified for software developers; when written in formal language, FRs are insufficiently comprehensible for users. This is a well-known problem in RE. As long as this either/or dichotomy exists, FRs cannot be a “basis for common agreement among all parties involved”, as Barry Boehm puts it.

Question/problem: On the one hand, FRs should unambiguously specify the functional behaviour of the system to be written or adapted, and on the other hand be fully understandable by the customer that must agree with them. What is required to achieve this goal?

Principal ideas/results: A specification must describe the Statics as well as the Dynamics. In our approach it consists of a Conceptual Data Model (the data structure, i.e., the Statics) plus a set of System Sequence Descriptions (SSDs) representing the processes (i.e., the Dynamics). SSDs schematically depict the interactions between the primary actor (user), the system (as a black box), and other actors (if any), including the messages between them. We provide a set of rules to generate natural language expressions from both the Conceptual
Data Model and the SSDs that are understandable by the user (‘Informalisation of formal requirements’). Generating understandable representations of a specification is relevant for requirements validation tasks.

Contribution to validation: We introduce a form of Natural Language Generation (the NLG in the title) by defining a grammar and mapping rules to precise and unambiguous expressions in natural language, in order to improve understandability of the FRs and the data model by the user community.