Building object representations from parts: Tests of a stochastic sampling model

In 2 experiments, the perceptual processing assumptions of the extended generalized context model (EGCM) of categorization were tested. The model assumes that object parts are sampled independently and in parallel in the earliest stages of categorization. In both experiments, participants categorized images of artificial objects under various levels of time pressure. Each participant also categorized partial objects, which were obtained by removing object components. The EGCM assumes that categorization judgments under time pressure can rely on incomplete object representations. The model successfully predicts performance in the speeded categorization task from the classification of partial objects. The results provide strong support for the EGCM, which performed consistently better than 2 alternative models.