Abstract
We have previously shown that meaning is correlated with where people look in scenes across a variety of tasks. To test for a causal relationship between scene meaning and attention we used a scene inversion paradigm. Inverting a scene makes a scene harder to identify and its object properties/relationships more difficult to extract (Kelley, Chun, & Chua, 2003; Epstein et al., 2006). Therefore, scene inversion should make it harder to extract scene meaning and we should observe a weaker meaning-attention association in inverted scenes relative to normally viewed scenes. The present study tested this hypothesis by examining how attention to meaningful scene regions is modulated by viewing real-world scenes from a normal and inverted orientation. Participants' (N=35) eye movements were recorded while they viewed 102 real-world scenes for 6 seconds each while performing a scene memorization task. In a within-subjects design, each participant saw half the scenes normally and half the scenes inverted, counterbalanced across participants, and presented in a random order. The spatial distribution of local semantic features for each scene was represented using human ratings of the informativeness of isolated scene patches (‘meaning map’; Henderson & Hayes, 2017). We then applied a logistic general linear mixed effects model to examine how a scene region’s meaning map value was related to its likelihood of being fixated in the normal and inverted scene conditions, with subject and scene treated as random effects. The results showed a strong dissociation with both an increased likelihood to look at low meaning scene regions and decreased likelihood to look at high meaning scene regions in the inverted scene condition relative to the normal scene viewing condition. This finding provides evidence for a causal relationship between scene meaning and attention and suggests scene meaning extraction is impaired by non-canonical viewpoints.