This talk will describe a COGENT reimplementation of Norman & Shallice's ``Contention Scheduling'' theory of routine action selection. The theory provides an account of the sequencing of low-level actions within routine or well-learned tasks, such as changing down the gears when driving towards a red traffic light, or preparing a cup of coffee. The implementation is based closely on an earlier Sceptic implementation of the theory. In this earlier implementation the main functional components of the theory were implemented as distinct rules. Within the COGENT implementation the principal functional components are implemented within distinct processes and buffers, further clarifying the structure of the theory.
The COGENT reimplementation is of interest for a number of reasons. First, the Contention Scheduling theory was originally specified in graphical form over 15 years before the COGENT graphical language was developed. We are thus able to compare the original graphical representation with that of the COGENT reimplementation. Second, the central computational mechanism with the theory is that of interactive activation. The reimplementation shows how standard COGENT box types can be used to implement interactive activation processes. There are a number of drawbacks to this form of implementation: calculating and updating activations is very inefficient, and there is no support for the display of changing activation values. The talk will therefore end by discussing how interactive activation processes might be encapsulated into a new general-purpose COGENT object.