EXPRESS: Object features hierarchically guide eye-movement and short-term memory for naturalistic scenes with preserved spatial layout across the lifespan.

Abstract

Visual working memory (VWM) operates within structured environments, yet it remains debated whether representations are guided primarily by the global scene configuration (gist) or the intrinsic features of objects. This question extends to cognitive ageing, where declines in VWM co-occur with a relative preservation of global information over fine-grained visual details. To disentangle these influences, younger and older adults detected changes to an object's identity, location, or both, while its semantic consistency with the scene was manipulated. Unlike our previous work (D'Innocenzo et al., 2022), where location changes disrupted the spatial layout, here we preserved the layout by swapping the critical object, thereby isolating memory for object-location binding from sensitivity to global disruptions. Across age groups, conjunctive changes were detected more accurately than single-feature changes. Crucially, detecting location changes was significantly more difficult when the layout was preserved (swaps) than when it was disrupted (displacements). This demonstrates that while layout disruptions provide salient cues, recalling an object's location from a stable configuration requires retrieving its intrinsic features. This is further supported by a detection advantage for semantically inconsistent objects (e.g., a torch in a bathroom), which was observed specifically in younger adults, suggesting an age-related decline in the strategic use of contextual violations. Eye-tracking during retrieval revealed longer fixations for inconsistent objects, indicating increased effort in integrating them. In contrast, consistent objects were fixated faster, a novel retrieval-phase finding likely driven by memory from their initial encoding as inconsistent. While core attentional mechanisms were preserved with age, our results argue that when global structure remains unaltered, VWM is guided by hierarchical representations of object features, in which semantic meaning plays a central, organising role.