Transfer Performance with Novel Displays of each Display Type

The most important result from this part of the experiment was the clear evidence of above chance levels of transfer to novel exemplars of all four display types (filled bars).The results for the texture display type replicate Cook et al.=s (1995) with a new set of birds, while the results for the geometric, feature, and object displays are new. These transfer results specifically rule out the already unlikely possibility that this learned discrimination was based on learning stimulus-specific response associations or rules for the entire set of training displays (e.g., Carter & Werner, 1978). Likewise, the results also rule out the possibility that the original discrimination was based only on judgements of the relative novelty of the Same and Different stimuli (e.g., Macphail & Reilly, 1989; see Cook et al., 1995 for a detailed discussion of this issue).

What was the basis for this positive transfer? Because it was observed across a wide variation in display types and configurations, it suggests a rule with a fair degree of scope and flexibility. As argued above, no single or simple perceptual attribute shared in common across the displays seems sufficient to account for these transfer data. Rather the broad extent of the transfer across the display types argues for a more abstract property, one shared in common across the dissimilar display types. We think the most likely candidate is one based on the global identity and non-identity of the element relations in the displays. As such, we propose that the most parsimonious interpretation -- of both the equivalent rate of discrimination learning across the display types and the breadth of positive transfer found in these experiments -- is that the pigeons developed and used a single generalized Same-Different rule applied to all displays simultaneously.

Mean choice accuracy on Baseline (B) and non-reinforced Novel Transfer (T) trials for each display type. The dotted reference line represents chance performance in the task.