We found that over the course of the training, story deviation increased and that the amount of neural conjunctive representation in DMN also tracked the amount of semantic conjunctive representation.

We also used a language model to compute how similar the story people came up with is from just the locus-item pair (“story deviation”), where higher story deviation means people added more novel details to the locus-item pair - a measure of semantic conjunctive representation.

In these DMN regions, the amount of neural conjunctive representation increased as participants gained more experience over the course of training.

We found that the conjunctive part of encoding representation is robustly and consistently reinstated during retrieval throughout the neocortex, especially in the DMN regions, much more than the locus- and item- alone representations.

We looked at the conjunctive part of the encoding representation that is not explained by a linear combination of locus and item representation (“encoding residual”), and used it (along with the locus- and item- alone representations) to predict retrieval representation.

We scanned participants three times through the course of the training, and we measured the neural representations of the items and the loci individually, and their combined representations during memory encoding and retrieval.

This strategy for forming strong memories involves pre-learning an imagined path through a sequence of anchors (loci) and then creating a meaningful association between each word in a list and a corresponding anchor.