Shifts in Context Affect Hippocampal Activity and the Sequence of Recall

People often find that memories linked to the same place or period come to mind together. Visiting a former workplace can prompt recall of events that had not been thought about for years, such as conversations or group activities. Yet a change of setting can also impair access to recent information; for example, when a person enters a room and forgets what they intended to do there. Temporal context models (Howard and Kahana, 2002) provide a framework for understanding these phenomena. They propose that internal context changes gradually, and this drifting state becomes linked to each event. When experiences share a similar context, they are more likely to be retrieved together. However, changes in context can increase the rate of internal neural drift, separating experiences that occurred close together in time. The hippocampus is thought to maintain these representations by integrating spatial and temporal information.

Research across neural and behavioral domains indicates that the organization of memory reflects both gradual contextual change and event boundaries. Studies have shown gradual changes in hippocampal activity as time passes during an event, consistent with neural drift (Folkerts et al., 2018). Additionally, hippocampal pattern resets have been observed at event boundaries, such as moving between rooms (Baldassano et al., 2016). These event boundaries segment experience into discrete units, which support the chunking of information in memory (Radvansky and Zacks, 2017). Behavioral work has demonstrated that context changes during learning can disrupt temporal clustering, suggesting that shifts in context weaken temporal associations (DuBrow and Davachi, 2013; Horner et al., 2016). However, prior neural work has not tested whether experimentally induced context changes alter the rate of hippocampal drift. As a result, there is no direct evidence connecting experimentally induced context change, hippocampal drift, and the organization of recall.

Rait et al. (2025) provide evidence for this missing link by examining whether changes in external context influence the rate of hippocampal drift and whether such changes in drift affect temporal organization. The authors hypothesized that rapid changes in external context would increase drift rate in the hippocampus and that this increase would be reflected in reduced temporal clustering during recall. To test this, Rait et al. had participants complete eight study-test blocks, each consisting of 24 unique words. During each block, a scene image preceded every word. For the No Switch condition, the same scene preceded all words. For the other conditions, two scenes alternated, switching once (Low Switch), three times (Medium Switch), or eleven times (High Switch) during encoding. Each condition was administered twice using different scenes and words. Following each list, participants were asked to report all the words they remembered, in any order; this allowed the authors to measure both recall accuracy and temporal clustering. Temporal clustering was assessed by measuring the order in which words were recalled and comparing the recall order to the original study order, irrespective of scene change.

Behaviorally, Rait et al. demonstrated that frequent scene switching disrupted the temporal organization of recall. Importantly, recall accuracy was similar across conditions, indicating stable memory strength. In contrast, temporal clustering was highest in the low and medium switch conditions and lowest in the high switch condition, where the context changed more often. These results suggest that a more stable external context helps preserve the temporal structure of memories without impacting the number of items remembered.

At the neural level, Rait et al. found that hippocampal drift varied with the rate of context change. Drift was measured using lag-1 autocorrelation, whereby each voxel's signal at one time point was correlated with its value at the next. In this analysis, higher values indicate slower drift. They found that autocorrelation decreased as scene switching increased, which indicates faster drift under rapid context change. Across participants, higher hippocampal drift was associated with less temporally organized recall, supporting a link between hippocampal drift rate and memory structure. Overall, these results support the view that hippocampal dynamics reflect an internal temporal context that shapes memory organization; frequent contextual change increases hippocampal drift and weakens temporal order during recall.

Rait et al. provide compelling evidence that hippocampal drift correlates with the organization, rather than the strength, of episodic recall. This work offers a critical link between behavioral support for temporal context models and direct neural evidence about how hippocampal states shape recall order. Additionally, the authors separate memory strength from memory structure, showing that context affects order but not recall quantity; this dissociation strengthens the interpretation that hippocampal drift specifically underpins memory organization.

While supporting models that incorporate a drifting context representation, the findings also align with related theoretical accounts. Event segmentation theory (Radvansky and Zacks, 2017) proposes that the brain organizes continuous experience into discrete events, with each boundary triggering a reset of ongoing representations. Boundary-driven reorganization in the hippocampus has been observed in prior studies. For example, Ben-Yakov and Dudai (2011) found that hippocampal activity increased immediately after a film clip ended, and this offset-locked response predicted whether the clip was later remembered. This change in neural activity indicates that event boundaries trigger hippocampal updating. Within this framework, each scene change in Rait et al. represented an event boundary that prompted hippocampal updating. As scene changes became more frequent, they created many boundaries that triggered repeated hippocampal updating, which in turn limited the integration of information over time. Therefore, this boundary-based account predicts the same pattern observed by Rait et al., as repeated resets would lower temporal autocorrelation in the hippocampal signal and disrupt the temporal structure of recall.

Given the above, one potential problem with interpreting the Rait et al. findings is related to the autocorrelation approach. Lag-1 autocorrelation captures the amount of change in the hippocampal signal between time points, but it does not identify the cause of that change. Both gradual drift and boundary-driven resets reduce autocorrelation, so the observed effects cannot be linked to drift specifically. A boundary-locked analysis would allow direct testing of transient reconfiguration, as in Ben-Yakov and Dudai (2011). Postboundary peaks in the hippocampal BOLD signal would support an event segmentation account, indicating discrete reconfiguration at boundaries. An absence of such peaks would support neural drift accounts consistent with temporal context models. This distinction is necessary because the observed reductions in autocorrelation could reflect drift, boundary resets, or both.

Both models plausibly account for the data because the two processes are not necessarily mutually exclusive. Indeed, DuBrow et al. (2017) hypothesized that gradual drift and boundary-driven resets function together. Within an event, a gradually evolving internal context signal drifts across time. At event boundaries, reorganization of context occurs producing a new internal state, which then begins to drift again. Applied to the Rait et al. findings, this means that drift and boundary-driven change could both contribute to reduced autocorrelation. On this basis, distinguishing the relative contribution of each process may be difficult.

One possible way to disentangle the contributions of boundary resets and drift could involve representational similarity analysis (RSA). Using this approach, the hippocampal activation pattern for each studied word can be extracted and correlations computed for every possible pair within a list. These correlations can then be grouped into pairs that occur within the same scene and pairs that span a scene change. This separation allows direct comparison of how stable the hippocampal representation remains within a context and how sharply it changes when the context switches. Event segmentation theory predicts relatively stable similarity within a scene and a pronounced decline at the boundary. In contrast, temporal context models predict a gradual decline in similarity with increasing temporal distance, followed by an additional decline at a scene boundary, while still preserving some similarity across that boundary. These predictions differ in the form of representational change, with one emphasizing step-like resets and the other continuous drift. The difference between the patterns may be small, and both accounts predict reduced similarity across boundaries, which make the two theoretical accounts difficult to distinguish empirically. Nonetheless, a method with higher temporal resolution may be needed to determine the timing and form of these changes in hippocampal representation. Combining OPM-MEG with the RSA approach outlined could allow for detection of rapid boundary-locked responses that occur too quickly for fMRI. Evidence of a brief, boundary-driven shift in representational patterns would support an event segmentation account, whereas the absence of such transient responses would be more consistent with continuous drift.

Overall, Rait et al. provide valuable evidence linking hippocampal dynamics to temporal memory organization, allowing future research to reveal the mechanisms that drive these dynamics. They demonstrate that environmental structure influences hippocampal context signals during encoding, and these signals in turn predict the order of subsequent recall. However, autocorrelation alone cannot determine whether reduced stability reflects gradual drift or event segmentation. Analyses using boundary-locked approaches or RSA could clarify when and how hippocampal changes occur. Combining high-temporal-resolution methods with RSA could help identify whether hippocampal signal changes are continuous or boundary-driven.