The role of sleep in modulating representational change

The aim of this study is to test the NMPH model using a new intervention ? sleep ? in a statistical learning paradigm that has previously been shown to drive representational change (Kim et al., 2017). In this paradigm, a small number of exposures to an item in a particular temporal context (e.g., B follows A) creates an expectation that when A appears, B should follow. If this expectation is violated (A is followed by X), the memory for mispredicted item B is activated, causing it to weaken (i.e., be forgotten; Kim et al., 2014). However, if the mispredicted item is later restudied, this interleaved sequence of trials leads to differentiation of the A and B memory representations in the hippocampus compared to a nonviolation control condition (Kim et al., 2017). This study also used fMRI pattern similarity analysis to show that moderate (vs. weak) activation of B during violation trials led to more differentiation, in keeping with the NMPH. Interestingly, this differentiation effect was not immediately apparent, and was only observed when learning and the post-learning measurement of representational change occurred on separate days with an intervening night of sleep (unpublished observation). The key question we aim to address in this study is whether interleaved learning during wake is sufficient to cause differentiation, or whether sleep (in particular, REM sleep) is necessary to bring about these representational changes.