A Cognitive Model of New Data on Human Problem Solving
I. Project Description
Cognitive modeling is the creation of models which resemble and explain the way in which humans do things. What makes them so interesting to me is the process though which cognitive scientists go in order to create these models. Cognitive scientists often use a generative theory in creating such models. A generative theory is a theory that explains a set of empirical observations by actually generating them (as opposed to just summarizing them or characterizing them with equations or logic). Thus, a generative theory has to be executable, like a computer program or a "recipe".
The system on which I'm basing my work is named Cascade (VenLehn, Jones & Chi, 1991). Cascade was originally developed to explain the cognitive mechanisms involved in the self-explanation effect (Chi et al., 1989; Fergusson-Hessler & de Jong, 1990; Pirolli & Bielaczyc, 1989). Simplifying a bit, the effect shows that people learn more effectively by studying examples when they are careful to explain to themselves as many steps of the example as they can. Students who do not carefully explain worked out example steps do not perform as well on subsequent problems. Cascade models the potential learning mechanisms that cause this effect.
I now wish to apply the Cascade model to a new problem domain and a new set of psychological data. Originally, Cascade was written to solve problems in Newtonian physics, the domain used in Chi et al.'s study. Since Cascade was first created additional psychological research has been done in other problem domains. Due to the versatility of Cascade, applying the Cascade model to other problem domains would be beneficial. In the fall of 2000 I was able to learn more about the utility of fading examples (Renkl, Atkinson & Maier, 2000). In doing so these scientists shared some of their other research with Randy and have asked us to re-implement Cascade on their other domain which is probability. In doing so I would learn the existing Cascade model well in addition to gaining extensive knowledge about the new problem domain.
The first step in this project is a task analysis (Newell & Simon, 1972). In this step we take the problem in Renkl et al.'s study and figure out exactly what the correct procedure is in solving the problems. This is one of the most critical steps in doing this project as we must correctly identify each piece of the solution of the problem. One of the goals of a good cognitive model is to accurately solve the problem as a real person would solve it; we don't want to just solve it as best we can. The result of the CTA will be a target knowledge base summarizing expert knowledge for this domain.
While working on the task analysis we would also work on prototyping one of the examples. During prototyping we will provide Cascade with the knowledge to solve one of the problems so we can work out the kinks in our knowledge representation.