The use of local features and global spatial context for object recognition in a visuomotor task in the Mongolian gerbil

Abstract

Mongolian gerbils were trained to jump across gaps of randomly varying width in order to obtain a food reward. During training, gerbils learned to jump to each of two landing platforms differing in width. These landing platforms were associated with unique spatial contexts (Experiment 1), local features (Experiment 2), or both (Experiment 3). In test sessions, a landing platform was substituted that was intermediate in width between the two training platforms of novel width, in order to determine whether gerbils could use retinal image size to calibrate distance. Experiments 1 and 2 suggested that gerbils used spatial context but not local feature information to identify the target platforms. In Experiment 3, when the probe platform was presented with the same local feature information and in the same context as seen in training sessions, gerbils over- or underjumped in a manner predicted if they were using the retinal image size of the target to calibrate the distance across the gap. When the probe trials contained a mismatch between context and local feature information, no systematic over- or under-jumps were seen, and jumping accuracy was decreased. Taken together, these results suggest that, in this task, objects are identified primarily on the basis of the spatial context in which they are seen and not on the basis of their local features.