Tonghe zhuang^1,2,3*, Selma Akarsu^1*, Malin Styrnal^1,2,3, Martin N. Hebart^1,2,3; ¹Computational Cognitive Neuroscience and Quantitative Psychiatry, Department of Medicine, Justus Liebig University Giessen, Germany, ²Vision and Computational Cognition Group, Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig, Germany, ³Center for Mind, Brain and Behavior, Universities of Marburg, Giessen, and Darmstadt, Germany * equal contribution

Objects in everyday life differ in how much they matter for our cognition and behavior. For example, we may more often look at rocks on the ground than babies, yet in our mental and neural representation of objects, a baby will matter more, highlighting that occurrence frequency is not sufficient to capture the relevance of objects. We term this graded importance representational relevance: the degree to which an object representation is likely to matter directly or indirectly for downstream cognition and behavior. Here we aim at operationalizing representational relevance for vision and cognitive science. We hypothesize that representational relevance will be related with, yet not identical to traditional measures of mental access, including lexical decision times, familiarity, or word frequency. To measure representational relevance (RR), we developed a rank-ordering “Family ”Feud”-style task in which participants arranged object words according to how often they believed other people would spontaneously name them (e.g., which is mentioned more frequently, “lion” or “burrito”?). This yielded an RR score for each concept. Split-half analyses over 1,000 random partitions showed that RR scores were highly reliable (Spearman–Brown corrected reliability = .957). RR showed strong good construct validity, correlating with lexical decision times (LDT, r = .706), familiarity (r = −.558), and frequency (r = −.685). To externally validate RR, we linked scores for 601 object concepts to a large-scale fMRI dataset (Hebart et al., 2023). Voxel encoding models revealed reliable RR effects in high-level occipitotemporal cortex, and RR scores were more predictive than word frequency and familiarity and led to comparable effects of LDTs, underscoring the usefulness of RR. Together, this work introduces representational relevance as a construct and highlights its potential usefulness for improved understanding of the relative weighting of objects beyond their frequency of occurrence.

Acknowledgements: MNH was supported by the ERC Starting Grant COREDIM, a LOEWE Start Professorship of the Hessian Ministry of Higher Education, Research, Science and the Arts, and the Deutsche Forschungsgemeinschaft, under the Collaborative Research Center “Cardinal Mechanisms of Perception” and Germany’s Excellence Strategy.