Discrimination and representation of relative numerosity in a bisection task by pigeons

We trained 4 pigeons in a numerical bisection task to discriminate between pairs of keylight flashes with a ratio of 1:3 (2 vs. 6, 4 vs. 12, and 8 vs. 24) that were presented in a sample phase. Responses to the blue key were reinforced after a sequence of a larger number of flashes, and responses to the white key were reinforced after a sequence of a smaller number of flashes. The intervals between flashes in the sample phase were randomized to attenuate the covariation of temporal cues with flash number. Pigeons responded accurately in each of the discriminations, with typically 85%-90% correct responses. Transfer tests showed that the proportion of large responses increased with number and performance generalized to larger values outside the training ranges. Psychometric functions superposed when plotted on a relative scale, and estimates of Weber fractions were approximately constant, suggesting that variability was scalar. However, contrary to previous research in nonhumans, bisection points were located at the arithmetic, not geometric, mean. Hierarchical logistic regressions confirmed significant control over responding by number beyond that attributable to temporal cues. These results show that pigeons are able to respond accurately in a relative numerosity discrimination with successively presented visual stimuli, although the nature of the numerical representation and response rule remains unclear.