MoDyS-Project VisACT

Software Visualization of ACT-R/PM Models

ACT-R/PM is a production system and a theory of the cognitive architecture. We apply ACT-R/PM for human-machine system modelling. Practical work turned out that ACT-R/PM and its development environment can be improved for software engineering purposes. We developed a visualization of ACT-R/PM models based on the statehart notation. A novel algorithm was developed to detect and geralize different working memory configurations. It was implemented as a plug-in for the integrated development environment eclipse. The purpose is to make development of complex models easier and less errornous. It also helps understanding and communicating about cognitive models of other developers.

Simulations of mental processes during the operation of technical systems can be a valuable source of information for system design, as part of support systems, for developing trainings, and to classify human errors. A current approach is to ground such human performance models on cognitive architectures like EPIC or ACT-R/PM. The founding of these architectures on established theories of cognition leads to comparatively fine-grained models. As an example ACT-R/PM offers detailed mechanisms in the areas of memory and perception/action. Thus cognitive models become very complex to develop and understand. An analysis of the production system paradigm that underlies most of the cognitive architectures revealed that the development complexity is very high because the control flow is not explicitly represented. Since there is currently no widely used diagram type for production system based cognitive models we propose a new visualization for these models that incorporates information on the control flow. It is based on statecharts.

The control flow is detected with a new algorithm for analyzing the interdependencies of ACT-R/PM productions. It mainly resembles the matching process of the production cycle. It does not include a detailed prediction of the outcome of the conflict resolution. This would result in a too fine grained chain of productions, that would not help understanding the purpose and structure of a model. The result of our algorithm is a directed graph. Nodes denote unique states of the declarative memory and productions which are applicable in these states are drawn as edges. States are generalized to reduce the complexity of the control flow. The graph is transformed into a statechart like visual representation. Goal oriented behavior with sub-goaling is considered with sub-graphs. The algorithm was implemented as a plug-in for the integrated development (IDE) environment eclipse.

• Leuchter, S., Nekrasova, L. & Urbas, L. (accepted). Software Engineering for Cognitive Modeling: Visualization of Production Systems. HCI International 2005, Las Vegas.