Testing and training geometrical knowledge in baboons through a conceptual categorization task

Sensitivity to geometric regularity has been proposed as a signature of human cognitive singularity. Baboons can detect an intruder shape among quadrilaterals but, unlike humans, show no advantage for regular shapes such as squares and rectangles, suggesting they are not sensitive to geometric regularity. Here we asked whether, with symbolic training, baboons may show an enhanced attention to geometric features and their compositions. Using a symbolic match-to-sample task with frequent introduction of new generalization stimuli, we trained baboons and human controls to categorize stimuli as representing a given concept (A) or not (not-A) across two non-geometric domains (image category and line color) and three geometric domains (line perpendicularity, parallelism, and length equality). Baboons learned to perform above chance with all categories and generalize them to novel positive and negative exemplars (except for length equality). These findings provide evidence that baboons can extract geometric properties and associate them to arbitrary symbols. Crucially, this explicit geometry training did not result in the emergence of human-like performance in the quadrilateral intruder task: overall performance improved uniformly across shapes, but the geometric regularity effect failed to emerge. Thus, in baboons, recognizing individual geometric properties is insufficient to trigger the compositional perception of shapes that characterizes human cognition.