How honeybees generalize visual patterns to their mirror image and left–right transformation

Abstract

The ability to generalize a familiar image to visual transformations like a mirror image or a left–right transformation may allow recognition of familiar images from a different viewpoint. As this problem applies to flower recognition by honeybees, Apis mellifera, we asked whether bees transfer acquired information about a previously rewarded pattern to its mirror image and/or its left–right transformation, and which are the mechanisms involved in such a transfer. Bees were trained either with a single pair of patterns or with six different pairs of patterns presented in a random succession. Within each pair one pattern was rewarded and the other not. All patterns had four quadrants, each displaying a different stripe orientation. In multiple-pattern training the six rewarded patterns shared a common configuration different from that of the six nonrewarded ones. After both kinds of training, the bees preferred the mirror image and the left–right transformation of the rewarded pattern (or rewarded configuration) to a novel pattern. They also preferred the left–right transformation to the mirror image. We explain this performance by: (1) matching with a retinotopic template of the trained patterns after training with a single pair of patterns; and (2) matching with a generalized pattern configuration after training with a randomized series of patterns. In the second case, orientations would be bound together in a specific spatial arrangement. Bees would associate a specific orientation with each retinal quadrant and approach the pattern provided that a particular quadrant contains a particular orientation. Although both strategies are based on comparison of an image currently perceived with one that has to be accessed from memory, they constitute different options as the former is less flexible while the latter allows for categorization of novel patterns.