Tom Endejan¹, Linda Smith²; ¹Indiana University

The visual world is made up of a large variety of kinds of things – man-made and natural, living and nonliving, animate and inanimate, foods and tools, solids and substances. Growing research both in human neuroscience and vision indicates that people learn the predictive patterns of edges that distinguish these higher order classes. We asked 30 adults from 6 different locations around the world to draw 50 common things that would be known even by young children. The populations differ markedly in education and culture: From the Bora people in the amazon rain forest to metropolitan Japan. We measured the distributions within each sketch of line segment length, orientation, and symmetry across the medial axis. We find that the predictive edge patterns that broadly distinguish these higher order semantic categories are also generated by humans in their drawings independent of cultural background and level of education. For example, sketches of living objects exhibit lower proportions of long line segments and lower symmetry across the medial axis from sketches of non-living objects. Man-made sketches can be distinguished from natural by higher levels of horizontal and vertical segments and symmetry across the medial axis. The similarity in edge statistics between higher level semantic categories held despite the wide array of lower level sub-categories reported in the collected sketches. We also compared the edge statistics of our sketches to photographs of the same objects and found both selectivity and correspondences in the drawn images relative to photographs of their real world counterparts. The results strongly implicate universal representations of high-level predictive patterns and highlight the value of using drawings as a measure of human visual representation of high-level semantic categories.

Acknowledgements: The Cognitive Development Lab at Indiana University