Pigeons' performance in the number-left task: Associative or computational mechanism?

This study investigates the mechanisms that underlie pigeons' performance in the number-left task. After producing x light flashes, pigeons had to choose between a standard option that delivered reinforcement after a fixed number of additional flashes, S = 4, and a number-left option that delivered reinforcement after a variable number of additional flashes, L = 8 - x. In Experiment 1, pigeons were trained with forced and choice trials with 1 ≤ x ≤ 7. During testing, the number of choice trials was simply increased. In Experiment 2, pigeons were trained only with the anchor numerosities x = 1 and x = 7 and during testing unreinforced probe trials introduced the intermediate numerosities, x = 2, 3, 4, 5, and 6. Performance was similar in both experiments and consistent with a computational mechanism. To test whether performance in the previous experiments was due to the substantial overlap in the induced generalization gradients around the anchor numerosities, in Experiments 3a and 3b, we selected anchor numerosities that were farther apart (x = 5 and x = 50, with S = 12 and L = 53 - x). Yet, pigeons' performance remained similar. We discuss the implications of these findings for the mechanisms that underlie numerosity discrimination.