Representing ordered associations in symmetric models of memory

Models of association memory make predictions about within pair order (AB vs. BA), either implying that order judgments of a retrieved pair should be at chance or perfect. Behaviour contradicts both predictions, when the pair can be recalled, order judgment is above chance, but still fairly low. We test two incremental modifications to symmetric, convolution-based models (which otherwise predict chance order judgment performance): 1) Encoding the item’s position as a subset of its features. 2) Position-specific permutations of item features. #1 achieved a close fit to order recognition data but compromised the well-known property of associative symmetry. #2 did not exhibit any reduction in symmetry but slightly overpredicted the dependence of order judgments on recall. In sum, simultaneously satisfying benchmark characteristics of association and order memory provides challenging constraints for existing models of association.