Abstract
Studies on how individuals understand real-world scenes, and form predictions about them, have traditionally focused on taxonomic knowledge about the scene’s content (its structure and the objects it contains). Recently, functional knowledge, which represents the actions afforded by a scene, has been proposed as a fundamental dimension of scene processing. However, it is unclear how these two kinds of knowledge are related, and in particular whether functional scene understanding requires the mediation of object information. We examined how taxonomic (specifically object-based) and functional (action-based) rapid scene understanding use visual information about fine, local features and objects, conveyed by high spatial frequencies (HSF), and coarse, contextual features, conveyed by low spatial frequencies (LSF). In each trial across four experiments, we presented an HSF or LSF filtered scene and two object or action words, one highly consistent with the scene and the other inconsistent. Participants reported which word was consistent. In the first two experiments, the words were shown simultaneously as primes, followed by the scene image, which appeared until response (Exp.1, online) or for 150ms followed by a 100ms frequency-matched pink-noise mask (Exp.2, online). Exp.3 (online) reversed the word-scene sequence, using the visual scene as a prime, with the same presentation times as in Exp.2. Exp.4 (lab-based), used the paradigm of Exp.2, but added a bandstop filter including HSF and LSF. Responses were faster for action than object words, in all experiments except in the bandstop condition, which showed no differences. We also reported greater accuracy for action words across most experiments. These results suggest that, independently of whether it has been activated before the scene instance is encountered, functional knowledge has a primary role in scene understanding when only fine or coarse visual features are provided and does not require the mediation of object information.
