Instructional Representation in Cognitive Modeling

Our previous research demonstrated how a cognitive architecture can inform the design of instructions. This abstract describes continuing research into instructional representations in both human instruction and cognitive architectures. In our past study (Taatgen et. al, submitted), two different forms of instructions were devised. These were dubbed “list-style” and “operator-style.” As the name list-style implies, procedural information is represented as a list of steps. These are common in cognitive models and human instruction. The list-style instructions for this experiment were taken directly from a major airlines training manual. The operator-style instructions take the basic format of “if you are in this state: then do this: and the result will be this” (if/then/result). Further, the “if” part of the operator was assumed to follow Taatgen’s (in press) minimal control principle and rely as much as is practical on external state information. See Taatgen, Huss & Anderson (submitted) for more elaboration on these instruction styles. Predicted benefits of operator style instructions were their generalizability and error recovery. We tested this, by providing our subjects with trials (referred to as “hard” trials) that required extrapolation from their initial instruction. In other words, if the subjects simply followed the instructions as they were given, they would not successfully complete these tasks. Also predicted was that the operator condition would better handle trials that were partially completed. This is based on the claim that operator instructions are biased towards reliance on external state information and should better be able to pick up a trial that is mid-task. This claim was not previously tested and forms the basis for this experiment.