"Reading" a visual scene: The effect of spatially grounded expectations on memory

Inferences and expectations evoked by an experience support interpretation and are sometimes misattributed to the experience itself-a constructive error observed in multiple domains (e.g., memory for sentences, stories, and events). Boundary extension (BE), memory beyond the edges of a view, may reflect this type of error (multisource model; Intraub, 2010). In contrast, it may simply reflect a spatial distortion involving good continuation of visible content (Bainbridge & Baker, 2020a). To explore this, we addressed an understudied aspect of BE: its content. In Experiment 1 (n = 156), participants predicted the unseen surroundings beyond close-ups of objects on continuous backgrounds and close-ups of objects cropped to eliminate a surrounding container. Expected-container and expected-continuation pictures were selected based on agreement across predictions. These were presented for 15 s each to naive participants, and the expanse and content of BE was assessed in recognition memory (Experiments 2a and 2b, total n = 394; Experiment 3, n = 184) and free recall (drawings: Experiment 4, n = 66). Expected containment elicited unprecedented false memory for unique content in BE and instances of alternative "readings" of the same views. Remembered expanse was constrained by expected containment, with this constraint being "released" when conditions favored false memory for the container rather than expectation alone (Experiment 2b). Results suggest that prior knowledge (e.g., learned contingencies, object-to-context associations) is leveraged during scene perception, contextualizing a view and simulating anticipated surroundings. These outcomes would have adaptive value for the observer during the dynamics of natural scene perception. (PsycInfo Database Record (c) 2025 APA, all rights reserved).