Testing the fast consolidation hypothesis of retrieval-mediated learning

Abstract The testing-effect, or retrieval-mediated learning, is one of the most robust effects in memory research. It shows that actively and repeatedly retrieving information, compared to merely restudying it, improves long-term retention. Surprisingly, little is known about the neurocognitive mechanisms underlying this phenomenon. Attempting to fill this gap, a recent framework proposed that retrieval acts as a fast memory consolidation mechanism, stabilizing memories through online reactivation, similar to memory replay during offline (e.g. sleep) consolidation. In this fMRI study, we empirically tested the predictions derived from this framework. We predicted that reactivation during retrieval allows memories to become embedded in neocortex, creating an additional route to access the memory trace and rendering it less hippocampus-dependent. Participants encoded scene-object pairs and either retrieved or restudied the objects over two sessions, two days apart. We analysed univariate and multivariate changes in brain activity specific to retrieval but not restudy, and tested whether the predicted changes occur rapidly within a session, or evolve slowly, across the two days. Results showed that medial prefrontal cortex activation increased across retrieval trials within one session, consistent with a fast consolidation account. Hippocampal activity decreased across sessions, suggesting a slower mechanism. Moreover, Representational Similarity Analyses (RSA) showed that consecutive retrieval attempts strengthen both higher-level semantic and episode-specific information in parietal areas, both across but not within sessions. Our findings suggest that retrieval supports the online creation of a neocortical trace, which becomes increasingly relevant at long delays when hippocampus-dependent episodic details would otherwise have faded. Significance statement Repeated remembering strengthens memories much more so than repeated learning. The aim of this study was to shed light onto the poorly understood neurocognitive underpinnings of retrieval-mediated learning. We tested a novel framework proposing that a memory’s stabilization via retrieval relies on mechanisms akin to those involved in offline systems consolidation. Observing the retrieval-induced neural pattern changes across different timescales, we find that retrieval stabilizes semantic and episodic aspects of the original memories, and produces increases in prefrontal activity and decreases in hippocampal activity that are consistent with the consolidation view, but not necessarily with a fast acting mechanism. Our findings inform cognitive theories of the testing effect, suggesting that retrieval produces its benefits by interacting with hippocampal-neocortical consolidation mechanisms.