The Dynamics of Categorization: Rapid Categorization Unraveled

Abstract

This dissertation explored a puzzle in visual object categorization: you usually spot the dog fastest; but in a glance, you spot the animal faster. The basic-level advantage (e.g., categorizing dog faster than animal or Golden Retriever) is a classic observation of categorization research. It is argued that basic-level categories follow the natural correlations and divisions of object features found in the environment and facilitated categorization at the basic level is an outcome of this representational advantage. A different story has recently emerged from studies of ultra-rapid categorization. With very brief stimulus presentations (<30ms), superordinate categorization (e.g., animal) is significantly faster and more accurate than basic-level categorization. Why does the relative timing of object categorizations vary so considerably under speeded category verification versus ultra-rapid categorization paradigms? This project explored two factors that affect the speed of categorization: stimulus exposure duration and the context of target categories. Brief exposures (25ms) and blocked target categories, typical of ultra-rapid categorization tasks, led to a superordinate-level advantage. Mapping the time course of perceptual processing suggested that brief exposure durations limit the amount of perceptual processing such that only relatively coarse visual properties of an image are encoded. This information is quickly encoded and favors an object’s superordinate-level category. With more time for encoding more detailed perceptual information, a basic-level advantage emerges. Target context, the order of target categories during an experiment, was found to have a very local effect on the speed of categorization. With only four repetitions of a superordinate categorization, the basic-level advantage in a standard categorization task was eliminated. The present research suggests that the basic-level advantage in standard categorization and superordinate-level advantage in ultra-rapid categorization are not in conflict. Rather, these paradigms characterize different aspects of the categorization process. The findings of this project offer a theoretical starting point for resolving the puzzle of how categorization decisions unfold over time.