Nurit Gronau¹, Rotem Avital-Cohen¹, Izhak Eliach¹, Keren Aloni-Carmeli¹; ¹The Open University of Israel

An object’s physical size is represented logarithmically relative to its real-world magnitude, a conceptual representation known as canonical size (Konkle & Oliva, 2011). We asked how this representation affects long-term memory (LTM) for specific episodic instances in which an object is viewed from a particular distance and in particular retinal dimensions. Specifically, does viewing an object in dimensions consistent with its familiar (conceptual) size enhance memory relative to viewing it in an atypical size? Prior work shows that visual LTM favors retinally large over small images (Masarwa et al., 2022). We tested whether retinal and conceptual size interact in LTM, hypothesizing that large presentations of conceptually large objects (e.g., a tree) would enhance memory more than large presentations of conceptually small ones (e.g., an apple), due to congruency between physical and conceptual size representations. Across a series of preregistered behavioral experiments, we replicated the LTM advantage for physically large items but found no interaction with conceptual size when probing memory for item identity (“Did you see an apple or a pear?”). However, when assessing memory for the actual physical size of stimuli (“Did you see a large or small apple?”), we observed a robust bias: participants tended to misremember objects as smaller than presented, for both targets and lures, and across different size scales. Notably, this bias was reduced for conceptually small objects shown in large dimensions, likely due to their increased salience and oddity. Thus, a conceptual-by-retinal size interaction emerged, but in an unexpected direction. These findings suggest distinct mechanisms for memorizing object identity versus physical size: identity memory is largely size-invariant, whereas physical-size memory is systematically reconstructed toward smaller (more distant) representations. We discuss how these results relate to object and scene memory reconstruction, offering new insights into the canonical size framework and the Boundary Extension phenomenon.