Ling Lee Chong¹ (), Yiwen Wang¹, Kara D. Federmeier¹, Diane M. Beck¹; ¹University of Illinois at Urbana-Champaign

Our prior work showed that the N300 amplitude is smaller in response to good scene exemplars than bad scene exemplars, reflecting a better match between good scene exemplars and our internal predictions. The frame-and-fill hypothesis proposes that coarse, low spatial frequency (LSF) information first generates initial predictions of an object’s identity, which are then refined by detailed high spatial frequency (HSF) information to improve object recognition. If the prediction described by the frame-and-fill model is the same one indexed by the N300, then LSF cues should be more effective than HSF cues in activating these predictions. To test this, we used a cueing paradigm in which participants viewed either scenes with only LSF or HSF information as cues that either matched or mismatched the subsequent full-spectrum target scene, which was either a good or bad scene exemplar. Participants indicated whether the filtered scene and full-spectrum scene belonged to the same category while EEG was being recorded. As in Kumar et al. (2021), we hypothesized that if the cue was effective, we would see a diminished difference between good and bad exemplars on mismatch trials relative to match trials. Our results showed that only the LSF cues were effective, as the LSF cue trials showed a significant modulation of match versus mismatch, with a larger N300 good/bad difference when the categories matched than when they did not. However, this effect was not present when HSF scenes were used as cues, as the N300 good/bad difference was similar across match and mismatch conditions. These findings suggest that LSF scenes were more effective at activating scene predictions than HSF scenes in keeping with the frame-and-fill hypothesis.