Figure 1. The temporal context model (TCM), a model of episodic recall based on contextual overlap, and computational models of semantic memory both predict transitive associations.

In all three panels, the figure shows the similarity of the representation of each item in a double function list of paired associates after training. a. Retrieved temporal context as defined by TCM shows transitive associations. The shading of each square codes for the similarity of the temporal context vector retrieved by the corresponding pair of items after ten trials of learning on the corresponding double function list. Vector similarity was assessed using the inner product. High values of the inner product are shaded dark. b. A representation generated using Latent Semantic Analysis (Landauer & Dumais, 1997) shows transitive associations. A singular value decomposition was computed for an item-context matrix corresponding to training on a double function list of pairs. Two dimensions were retained. Similarity of each pair of vectors was assessed using the cosine of the angle between them. High values of cosine are dark. c. The topic model (Griffiths, Steyvers, & Tenenbaum, 2007) was trained on a set of contexts simulating presentation of a double function list. The simulation used two topics and α = 0.1 and β = 0.1 (see Griffiths, Steyvers & Tenenbaum, 2007 for details). The similarity between each pair of items was estimated by comparing the Kullback-Leibler divergence of the distribution over topics induced by each item. Small values of divergence, corresponding to high similarity, are dark.