Completeness of Natural Language Requirements: A Comparative Study of User Stories and Feature Descriptions.

Checking the completeness of requirements is critical for software validation, as incomplete requirements can adversely affect the delivery of high-quality software within budget. Many existing methods rely on the domain model that defines the correct and relevant constructs, against which the requirements completeness is checked. However, building accurate and updated domain models requires considerable human effort, which is often challenging in practical settings. To operate in the absence of domain models, we propose to measure a textual requirement's completeness based on a universal linguistic theory, namely Fillmore's frame semantics. Our approach treats the frame elements (FEs) associated with a requirement's verb as the roles that should participate in the syntactic structure evoked by the verb. The FEs thus give rise to a linguistic measure of completeness, through which we compute a requirement's actual completeness. Using our linguistic-theoretic approach allows for a fully automatic completeness check of different real-world requirements. The comparisons show that our studied feature descriptions are more complete than user stories.