Optimal-constraint lexicons for requirements specifications

Constrained Natural Languages (CNLs) are becoming an increasingly popular way of writing technical documents such as requirements specifications. This is because CNLs aim to reduce the ambiguity inherent within natural languages, whilst maintaining their readability and expressiveness. The design of existing CNLs appears to be unfocused towards achieving specific quality outcomes, in that the majority of lexical selections have been based upon lexicographer preferences rather than an optimum trade-off between quality factors such as ambiguity, readability, expressiveness, and lexical magnitude. In this paper we introduce the concept of 'replaceability' as a way of identifying the lexical redundancy inherent within a sample of requirements. Our novel and practical approach uses Natural Language Processing (NLP) techniques to enable us to make dynamic trade-offs between quality factors to optimise the resultant CNL. We also challenge the concept of a CNL being a one-dimensional static language, and demonstrate that our optimal-constraint process results in a CNL that can adapt to a changing domain while maintaining its expressiveness.