Pii: S0376-6357(01)00200-5

Two experiments tested a bottlenosed dolphin’s ability to match objects across echolocation and vision. Matching was tested from echolocation sample to visual alternatives (E–V) and from visual sample to echolocation alternatives (V–E). In Experiment 1, the dolphin chose a match from among three-alternative objects that differed in overall (global) shape, but shared several ‘local’ features with the sample. The dolphin conducted a right-to-left serial nonexhaustive search among the alternatives, stopping when a match was encountered. It matched correctly on 93% of V–E trials and on 99% of E–V trials with completely novel combinations of objects despite the presence of many overlapping features. In Experiment 2, a fourth alternative was added in the form of a paddle that the dolphin could press if it decided that none of the three-alternatives matched the sample. When a match was present, the dolphin selected it on 94% of V–E trials and 95% of E–V trials. When a match was absent, the dolphin pressed the paddle on 74% and 76%, respectively, of V–E and E–V trials. The approximate 25% error rate, which consisted of a choice of one of the three non-matching alternatives in lieu of the paddle press, increased from right to center to left alternative object, reflecting successively later times in the dolphin’s search path. A weakening in memory for the sample seemed the most likely cause of this error pattern. Overall, the results gave strong support to the hypothesis that the echolocating dolphin represents an object by its global appearance rather than by local features. © 2002 Elsevier Science B.V. All rights reserved.