Proceedings of the Annual Meeting of the Cognitive Science Society

The concreteness–abstractness distinction of concepts faces challenges from proposals that conceptual representations occupy a continuous, multidimensional sensorimotor space. Additionally, there are concerns regarding psycholinguistic norms, arguing that presenting words in isolation introduces ambiguity and that the tendency to have high variability in mid-range ratings complicates the interpretability of judgments. Here, an artificial neural network (ANN) was used to model the relationship between concreteness and sensorimotor information by incorporating mean sensorimotor strength and inter-rater variability across 11 perceptual and action dimensions to allow for nonlinear interactions. The ANN predicted graded concreteness, indicating that sensorimotor information encodes structured regularities irreducible to additive effects. Prediction error analysis revealed that inter-rater variability provides informative structure rather than noise, and that performance is not driven by lexical ambiguity. These findings suggest that conceptual representations rely on systematically organized sensorimotor systems, supporting the concreteness–abstractness distinction as a graded, probabilistic construct grounded in sensorimotor experience.