The effects of extended reinforcement schedules on the receiver operating characteristic (ROC) of an echolocating bottlenose dolphin (Tursiops truncatus)

Abstract

A bottlenose dolphin was conditioned to report via two manipulanda the presence or absence of a solid steel sphere suspended 80 cm beneath the water's surface, 7 m (minimum) from his position. The dolphin's biosonar target detection and trained binary reporting behavior were tested utilizing variable primary reinforcement schedules (VR) from VR1 (fish reward every correct response) to VR100 (fish reward on one trial, randomly selected from 100 trials). Tests were also conducted with a fixed reinforcement schedule (FR) of FR100 (6.8 kg of fish reward after correctly responding on the last trial of a 100-trial session). The probability of target presentation was 0.50 for all schedules tested and the number and amount of reinforcement was kept equal for “present” and “absent” responses during all tests. The dolphin's ROC remained essentially unchanged and Neyman-Pearson at all reinforcement schedules. However, if reinforcement was kept at the more extended schedules (VR20 to VR100 and FR100) for more than eight consecutive 100-trial sessions all responses became “target-absent.” Response topography clearly showed the difference between detection and response bias.