The role of anchor objects in scene function understanding

Throughout every day, we perform actions, and action information has been suggested to inform scene categorization. Here we hypothesise that actions also drive the hierarchical structure of many scenes, where anchor objects (e.g., stoves) predict the presence and position of local objects (e.g., pots) by dividing a scene in functionally distinct 'phrases'. Specifically, we test whether the presence of anchor objects informs scene function understanding. In Experiment 1, participants matched an action word and a scene from which we either removed an action-related anchor object (REL), an action-unrelated anchor (UNREL) or a non-anchor object (RAND). Matching performance was impaired in REL compared to UNREL and RAND. Experiment 2 measured scene function activation more implicitly by priming a lexical decision task (LDT) on action words with the same stimuli (including an inconsistent condition: INCON, e.g., "cooking" in a bathroom). LDT performance was impaired after INCON and REL compared to RAND and UNREL primes. A control experiment showed that this effect was partly but not solely due to scene categorization. The results imply that understanding scene function is most closely tied to anchor objects directly relevant for actions whereas contextual scene information is not always sufficient to give rise to this understanding.