Abstract
Picture stories are composed of discrete events based on goal-directed behaviors. When perceiving an event, a working-memory representation of that event is created, called an event model, which makes predictions about what may occur next in the story. Studies have shown that changes in the setting (the scene gist) and character action often indicate that a new event model is needed, but it is unknown which of these two features is processed first when constructing an event model. We examined the time-course of scene gist and action categorization to determine which is categorized earliest in scene perception. Images were categorized according to their superordinate scene gist (Indoor vs. Outdoor), basic level scene gist (e.g., Park vs. Yard), or action (e.g., Raking vs. Mowing).
Rosch et al.’s (1976) basic level theory predicts that basic level scene categories should be recognized prior to those at the superordinate level. However, recent studies (e.g., Loschky & Larson, 2010) have shown the opposite to occur early in scene perception. Alternatively, eye-movement studies show an attentional bias to fixate people in scenes, which predicts that actions may be categorized prior to scene categorization. We tested these three competing hypotheses by randomly assigning participants to one of three categorization tasks (superordinate scene, basic scene, or action). Scenes were presented for 24 ms, with masking SOAs of 23, 60, 106, 200, and 376 ms, and participants then responded "Yes" or "No" to a valid or invalid category label.
The results showed that image categorization occurred in a global-to-local fashion. Namely, at early SOAs (e.g., 23 ms), superordinate scene gist categorization showed the greatest sensitivity, followed by the basic level, and then actions. Thus scene gist is categorized prior to actions, suggesting that scene gist is the first stage in perceiving events, constructing event models, and comprehending picture stories.
Meeting abstract presented at VSS 2012