Cognitive Modeling of Mental Rotation
Master's thesis of Fabian Joeres
This thesis aims to show that the cognitive processes of
three-dimensional mental rotation can be addressed with cognitive
models, using the cognitive architecture ACT-R. First, a conceptual
model of these processes was developed, based on a wide range of
previous findings on mental rotation.
Since the current version of ACT-R (ACT-R 6.0) does not offer the structures for modelling visuospatial cognitive processes, such a structure was developed and implemented in the architecture. The according module was named Spatial Module and contains the functions and structures necessary for modelling mental rotation processes. The abovementioned conceptual model was implemented in the extended architecture ACT-R. To fit models to experimental data, ACT-R offers several subsymbolic parameters. With the implementation of the spatial module, two new parameters were introduced. In order to define
valid default values for these parameters, the experiments of R. Shepard and J. Metzler (1971) and of Yuille and Steiger (1982) were simulated with the cognitive model. To evaluate the model itself, an experimental study with 27 participants was conducted. The experiment included a three-dimensional mental rotation task with Shepard Metzler objects. In each trial both stimuli needed for the comparison were presented simultaneously. A large number of previous studies have shown that the reaction time in mental rotation tasks
forms a linear function of the rotation angle. The slope of this linear function is called rotation rate. This study aimed to examine the effects of general training and of object familiarity on that rotation rate. Experimental results showed that rotation rate decreased significantly (i.e. rotation became ‘faster’) with practice. Furthermore, it was shown that significantly faster rotation occurred for familiar objects than for unfamiliar objects. These results could be predicted validly by the model. Therefore, the cognitive model and the extension of the architecture ACT-R that were developed in this thesis can be seen as a promising approach for modelling and simulation of visuospatial cognitive processes.