Anchor Objects Contribute to Scene Function Understanding.

Abstract

As we navigate everyday environments, we perform complex actions that are typically directed at objects, not entire scenes. While actions are strong predictors of scene category, it remains unclear whether scene function understanding depends more on object-level or scene-level information. We investigate if these actions drive the hierarchical structure of scenes by testing if anchor objects (i.e., the sink), which predict the location of local objects (i.e., the toothbrush), impact scene function understanding. In Experiment 1, participants explicitly matched action words and scenes in which the action-related anchor (REL), the action-unrelated anchor (UNREL) or a random, non-anchor object (RAND) were masked. Matching performance was significantly worse in REL scenes than in UNREL and RAND scenes. In Experiment 2, we measured scene function understanding by using the stimuli as primes in a lexical decision task on action words. Here, the REL condition elicited significantly slower response times than the RAND condition, suggesting that scene function understanding relies primarily on action-related, object-level information. A control experiment showed that these results could not solely be attributed to impaired scene categorization due to anchor-absence. We further showed that stimulus features such as visual realism and object clutter modulated task performance but did not alter the overall pattern of results. Together, our results suggest that scene function understanding depends mainly on the action-related anchor objects, whereas scene context is not always sufficient to understand a scene's functions.