Computational models of visual tagging

The studies reported in this chapter exemplify the experimental-simulative approach of the interdisciplinary research initiative on “Situated Artificial Communicators”. Two experiments on visual tagging strategies are described. In Experiment 1, participants were presented with random distributions of identical dots. The task was to look exactly once at each dot, with a starting dot specified. This setting allowed a quantitative analysis of scan path structures and hence made it possible to compare empirical scan paths to computer-generated ones. Five different scan path models were implemented as computer simulations, and the similarity of their scan paths to the empirical ones was measured. Experiment 2 was identical to Experiment 1 with the exception that it used items of varying color and form attributes instead of identical dots. Here, the influence of the distribution of colors and forms on empirical scan paths was investigated. The most plausible scan path models of Experiment 1 were adapted to the stimuli of Experiment 2. The results of both experiments indicate that a simple, scan path minimizing algorithm (“Traveling Salesman Strategy”; TSS) is most effective at reproducing human scan paths. We also found an influence of color information on empirical scan paths and successfully adapted the TSS-based model to this finding.