Numerical reproduction in pigeons.

Pigeons were trained to match the number of responses made during a production phase to the number of keylight flashes (2, 4, or 6) in a previous sample phase. In Experiment 1, there were 2 conditions in which the flashes were programmed to occur at a constant rate or within a constant overall duration. For both conditions, although accuracy was relatively low, responding increased linearly with flash number and coefficients of variation decreased. Positive transfer to novel numbers was obtained only when test and baseline trials had the same temporal organization, but multiple regressions revealed significant control by number independently of temporal cues. In Experiment 2, flashes were programmed to occur pseudorandomly to degrade the validity of temporal cues. Results were similar to in Experiment 1. A prototype response class model accounted for the major features of the data. According to the model, responses during the production phase are shaped into higher order units that are associated with different stimulus numbers and function as a rough category scale of numerosity.