Cognitive Support during Object-Oriented Software Development: The Case of UML Diagrams

Abstract

The Object Management Group (OMG) accepted the Unified Modelling Language (UML) as a standard in 1997, yet there is sparse empirical evidence to justify its choice. This research aimed to address that lack by investigating the modification of programs for which external representations, drawn using the UML notations most commonly used in industry, were provided. The aim of the research was to discover if diagrams using those UML notations provided the modifying programmer with cognitive support. The application of the use of modelling to assist program modification was chosen as a result of interviews that were carried out in New Zealand and North America to discover how workers in the software industry used modelling, and if so, whether UML notation satisfied their needs. The most preferred UML diagrams were identified from the interviews. A framework of modelling use in software development was derived. A longitudinal study at a Seattle-based company was the source that suggested that program modification should be investigated. The methodology chosen for the research required subjects to modify two non-trivial programs, one of which was supplied with UML documentation. There were two aspects to the methodology. First, the subjects’ performances with and without the aid of UML documentation were compared. Modifying a program is an exercise in problem solving which is a cognitive activity. If the use of UML improved subjects’ performances then it could be said that the UML had aided the subjects’ cognition. Second, concurrent verbal protocols were collected whilst the subjects modified the programs. The protocols for the modification with UML documentation, for ten of the more successful subjects, were transcribed and analysed according to a framework derived from the literature. The framework listed the possible cognitive steps involved in problem solving where cognition could be distributed to and from external representations. The categories of evidence that would confirm cognitive support were also derived from the literature. The experiments confirmed that programmers from similar backgrounds varied widely in ability and style. Twenty programmers modified both an invoice application and a diary application. There was some indication that the UML diagrams aided performance. The analyses of all ten of the transcribed subjects showed evidence of UML cognitive support.