Stepwise conceptualization in pigeons.

Abstract

We explored the relation between concept learning and the number of training exemplars, programmed in a stepwise fashion. Eight pigeons (Columba livia) were trained via differential food reinforcement to peck 1 of 2 simultaneously displayed color images: benign or malignant human breast tissue samples. In each session, only 1 exemplar from each category was trained. If the learning criterion was achieved in 1 session, then those 2 exemplars were discarded and 2 new exemplars were trained in the next session. For the consistent group the reinforced category (benign or malignant stimuli for different birds) was maintained throughout all 24 training pairs, but for the inconsistent group the reinforced category was pseudorandomly changed from pair to pair. Pigeons in the consistent group evidenced striking gains in accuracy during the 1st few trials of training on new pairs (exceeding 90% correct over Pairs 10-12), whereas pigeons in the inconsistent group evidenced reliably weaker gains in initial categorization accuracy (rising above only 70% correct after comparable training). Stepwise training involving as few as a dozen exemplars per category was thus sufficient to support robust concept learning in pigeons.