European starlings unriddle the ambiguous-cue problem

The ambiguous-cue problem is deceptively simple. It involves two concurrently trained simultaneous discriminations (known as PA and NA trials), but only three stimuli. Stimulus A is common to both discriminations, but serves as non-reinforced stimulus (S-) on PA trials and as reinforced stimulus (S+) on NA trials. Typically, animals' accuracy is lower on PA trials-the ambiguous-cue effect. We conducted two experiments with European starlings (Sturnus vulgaris) using Urcuioli and Michalek's (2007, Psychon B Rev 14, 658-662) experimental manipulations as a springboard to test the predictions of two of the most important theoretical accounts of the effect: the interfering cue hypothesis and value transfer theory. Both experiments included two groups of birds, one trained with a regular ambiguous-cue problem (Group Continuous) and another trained with partial reinforcement on PA trials (Group PA-Partial). The experiments differed only in the number of sessions (18 vs. 36) and daily trials (360 vs. 60). As previously observed, we found faster acquisition on NA trials than on PA trials in both experiments, but by the end of training PA performance was surprisingly high, such that no ambiguous-cue effect was present in Group Continuous of either experiment. The effect was still present in both PA-Partial groups, but to a smaller degree than expected. These findings are inconsistent with the literature, in particular with the results of Urcuioli and Michalek (2007) with pigeons, and question the aforementioned theoretical accounts as complete explanations of the ambiguous-cue effect. In our view, to achieve such high levels of accuracy on PA trials, starlings must have attended to configural (i.e., contextual) cues, thus differentiating stimulus A when presented on PA trials from stimulus A when presented on NA trials. A post hoc simulation of a reinforcement-based configural model supported our assertion.