Modeling Cognitive Control in Mental Spatial Knowledge Processing

Human reasoning about spatial problems, like, for example, planning a route or designing a building, can be conceived as arising from—at least—three functionally distinct components: Long-term memory, working memory, and an externalization component. Long-term memory permanently retains all of the spatial knowledge a person has encountered and encoded previously. This knowledge may be retrieved from long-term memory (i.e., remembered), whenever the information already available regarding the current problem is not sufficient to solve it. In working memory information from long-term memory is combined with information obtained by inference and/or from the environment. The information residing in working memory is the basis for any inferences drawn during the problem solving process. In contrast to long-term memory, however, working memory is rather transient and can only hold a limited amount of information (e.g., Miller, 1956). Therefore, to facilitate spatial knowledge processing humans often use external knowledge representations like maps or blueprints to reduce the amount of information to be held in working memory (e.g., Verstijnen, Leeuwen, Goldschmidt, Hamel, & Hennessey, 1998). It is the function of the externalization component to realize the exchange of knowledge between external representations and working memory. Consequently, in developing a computational model of mental spatial knowledge processing all three components need to be realized. Furthermore, the implementation of the components has to be cognitively plausible, i.e., in accord with findings from psychological experiments. Yet, developing the single components, although crucial, is not enough. Rather, in addition, the interplay of the components has to be—somehow—controlled. Otherwise there would be no guarantee that the different components would work together, reasonably complementing each other, while some certain spatial problem is solved. Accordingly, to complete the computational model, some control mechanism has to be implemented. In particular, as with the components themselves, the control mechanism should be cognitively plausible. Thus, the question arises how such a control mechanism could be realized. The aim of my PhD work is to answer this question.