Cognitive Constraint Modeling: A Formal Approach to Supporting Reasoning About Behavior

摘要: Cognitive Constraint Modeling: A Formal Approach to Supporting Reasoning About Behavior Andrew Howes (HowesA@cardiff.ac.uk) School of Psychology, Cardiff University, Cardiff, Wales, UK CF10 3YG. Alonso Vera (avera@mail.arc.nasa.gov) NASA Ames Research Center, MS 262-3 Moffet Field, CA 94035. Richard L. Lewis (rickl@umich.edu) Department University Michigan, Ann Arbor, MI 48109-1109. Michael McCurdy (mmccurdy@arc.nasa.gov) Abstract asymptotic bounds on skilled behavior. The specific objectives the paper are: (1) To introduce hypothesis that behavior is optimal solution a constraint satisfaction problem defined by architecture, task environment, and knowledge constraints. (2) formal modeling approach, called CCM, directly supports reasoning about By using deductive inference algorithms, CCM computes necessary consequences constraints imposed strategic knowledge, cognitive architecture. These may determine, for example, which environmental processes can execute in parallel have sequential dependencies. (3) specify two ontologies provide alternative information processing vocabularies theory, resulting descriptions first straightforward formalization temporal dependencies, implicit existing work based CPM- GOMS. second richer ontology permits specifying sets communicating processes, where both inter-process communication channels buffers are subject resource This framework has much common with McClelland's cascade model (McClelland, 1979). Both formally set declarative axioms part specification. following structure. We background our then describe tool, CORE (Constraint-based Optimal Engine). application CORE, dependency axioms, dual experiment reported Schumacher, Lauber, Glass, Zubriggen, Gmeindl, Kieras, Meyer (1999) subsequently modeled Byrne Anderson ACT- R/PM (Byrne Anderson, 2001). In doing so we show flexible enough support implications central peripheral bottleneck theories performance. requires 42 simple, universally quantified, statements Modeling (CCM) an approach provides investigating objective, environmental, architectural constraints, derives predictions from specifications dependency-based cascade-based expressing relationships between processes. software tool demonstrates potential advantages described. be used partially automate generation behavioral given specification explore dual-task data previously EPIC ACT-R. Introduction When people acquire skill they able adapt their as incrementally improve value some utility function. With practice, scope improvement attenuates performance asymptotes. It asymptote at level consistent environment or perhaps determined brought task. instead human More plausibly, combination including stochastic profiles cognitive, perceptual, motor systems. bounded multiplicity (Simon, 1992). There course been aimed its acquisition (e.g. Lebiere, 1998; 1997; Taatgen 2003). purpose current initial demonstration how models generated derivation multiple particular, this