System and Software Requirements Validation through Inspections:
Constructive Reading and Mining Requirements from Natural Language Requirements
Documents
Article type: Research Article
Authors:
Affiliations: Bridgewater College, Department of Computer Science,
Bridgewater, VA 22812, USA
Abstract: Defects introduced early in the effort to engineer a system due to
poorly identified requirements are generally seen as a major factor leading to
high system and software costs, especially if the defective requirements are
undetected until later development phases in the lifecycle of engineering a
system. In software development, inspection methods have been particularly
successful when applied to code inspections, but have often been substantially
less effective when they are applied to natural language requirements
specifications. Yet, the ultimate savings due to error detection, diagnosis,
and correction before a trial system is produced are generally great. This
paper addresses the problem of improving requirements inspections by exploring
foundational issues, such as the ability of inspectors, the degree that skill
is present as a stable and determining factor, and whether defect detection is
influenced by intrinsic differences in difficulty of detection among defects. A
new inspection technique, denoted Constructive Reading Inspection Process
(CRIP), was developed and used to explore requirements inspection, which
involves extracting the conceptual entities and their interrelationships as
opposed to looking solely for defects. Inspections of phase products in
software development is a best-practice in systems and software engineering,
however reading inspection of natural language requirements is less productive
than the same practices applied to code. Our study of reading inspections
reveals issues largely overlooked in the past which suggest that low yields can
be partly accounted for by cognitive factors, lack of uniform expectations, and
the character of ordinary reading. The Constructive Reading Inspection Process
focuses on revealing and illuminating the conceptual entities and inspecting
those; thereby shifting the implied operational metaphor from fishing (looking
for defects) to mining (exhuming and polishing entities). This paper describes
the process, its use, and provides a preliminary evaluation of its
effectiveness.