Order of items within associations

Association-memory is a major focus of verbal memory research. However, experimental paradigms have only occasionally tested memory for the order of the constituent items (AB versus BA). Published models of associationmemory, implicitly, make clear assumptions about whether associations are learned without order (e.g., convolution-based models) or with unambiguous order (e.g., matrix models). Seeking empirical data to test these assumptions, participants studied lists of word-pairs, and were tested with cued recall, associative recognition and constituent-order recognition. Orderrecognition was well above chance, challenging strict convolution-based models, but only moderately coupled with association-memory. Convolution models are thus insufficient, needing an additional mechanism to infer constituent order, in a manner that is moderately correlated with associationmemory. Current matrix models provide order, but over-predict the coupling of orderand association-memory. In a simulation, when we allowed for order to be wrongly encoded for some proportion of pairs, order-recognition could be decoupled from cued recall. This led to the prediction that participants should persist with their incorrect order judgement between initial and final order-recognition, but this was not supported by the data. These findings demand that current models be amended, to provide order-memory, while explaining how order can be ambiguous even when the association, itself, is remembered.