Identification of disoriented objects: effects of context of prior presentation.

Half of the subjects in the training phase of Experiment 1 named objects shown in a number of orientations, whereas the other half named objects shown upright only. All subjects named objects seen in a number of different orientations in the transfer phase. Half of the objects in the transfer phase were the ones they had seen in the training phase (old objects), whereas the other half were objects they had not seen before (new objects). Mean naming time in the transfer phase increased more as the objects were rotated further from the upright for new objects than for old objects when the old objects had been seen in a variety of orientations. In contrast, a substantial and equivalent orientation effect on identification time was obtained for old and new objects when the old objects had been seen upright only. These results suggest that the extraction and use of orientation-invariant attributes to identify objects is not a "default" identification strategy employed by the human visual system. In Experiment 2, half of the objects named in the training phase were shown upright only, whereas the other half were shown in a number of orientations. Both types of objects (upright vs. rotated) were presented in a mixed fashion from trial to trial. The results revealed that prior naming of the objects in this context resulted in equivalent reductions in the magnitude of the orientation effect on identification time for both sets of objects (upright and rotated). Together, the results of these two experiments suggest that markedly different representations of objects are encoded, depending on the context in which objects are seen. Implications for models of pattern recognition are discussed.