A basic tenet of principles of associative learning applicable to models of spatial learning is that a cue should be assigned greater weight if it is a better predictor of the goal location. Pigeons were trained to locate a goal in an acute corner of an isosceles trapezoid arena, presented on a slanted floor with 3 (Experiment 1) or 2 (Experiment 2) orientations. The goal could be consistently determined by the geometric shape of the arena; however, its position with respect to the slope gradient varied, such that slope position was not a good predictor of the goal. Pigeons learned to solve the task, and testing on a flat surface revealed successful encoding of the goal relative to the geometric shape of the arena. However, when tested in the arena placed in a novel orientation on the slope, pigeons surprisingly made systematic errors to the other acute—but geometrically incorrect—mirror image corner. The results indicate that, for each arena orientation, pigeons encoded the goal location with respect to the slope. Then, in the novel orientation, they chose the corner that matched the goal’s position on the slope plus local cue (corner angle). Although geometry was 2 times (Experiment 2) or even 3 times (Experiment 1) as predictive as slope, it failed to control behavior during novel test trials. Instead, searching was driven by the less predictive slope cues. The reliance on slope and the unresponsiveness to geometry are explained by the greater salience of slope despite its lower predictive value.
Nardi, Daniele; Nitsch, Kristian P.; and Bingman, Verner P., "Slope-Driven Goal Location Behavior in Pigeons" (2010). Faculty Research and Creative Activity. 21.