A necessary step for modelling how users learn to perform a new computer task is to develop an environment that will routinely allow cognitive models and subjects to interact with interfaces. We have created a version that uses Soar as the cognitive modelling architecture and Garnet as the simulation language. They are tied together using an interprocess communication mechanism based on UNIX sockets. This environment solves two difficulties: (1) The interaction between the cognitive model and the simulation can be set up routinely because the general and robust underlying sockets have been made easy to use; and (2) Tasks are instrumented to record user and model actions for later model validations. This environment also includes a simple air traffic control task. We found that our initial cognitive model spends a large amount of time interacting with the task, particularly looking. This suggests a new avenue in which to explore interaction in HCI. An early version, while waiting for the world to respond, would continue problem solving and then be interrupted by the world. This suggests that mental models are important for learning about interactive tasks and that interruptibility may be critical, even in relatively static tasks.
[Appeared as : Ong, R., & Ritter, F. E. (1995). Mechanisms for routinely tying cognitive models to interactive simulations. In HCI International '95. Osaka, Japan: July 1995]