The role of memory reinstatement during the online encoding and retrieval of episodic event sequences in humans
We experience our life in a continuous manner, but despite that, such continuous stream of information is segmented by the brain into distinct episodes that we can later recall. Interestingly, events experienced just once might be remembered even a lifetime. The Event Segmentation Theory (EST) was developed to explain how such segmentation occurs. The EST posits that our brain makes sense of the ongoing experience by activating ‘event models’, which consist in active representations of ‘what is happening now’ based on previously stored experiences. At some points, it is thought that errors in predictions about what will happen next might start to increase. Such moments are known –and perceived- as ‘event boundaries’ and involve relevant changes in the ongoing experience, either perceptual or conceptual. The current literature shows that event boundaries have a decisive influence on the memory for episodic sequences. First, event boundaries studied in experimental settings seem to weaken associations between items presented across-boundaries. Second, event boundaries improve subsequent memory for the resulting episodes. Such sequence episodes are defined as items sharing the same context (for instance, an ‘event model’). Thus, a plausible hypothesis is that the encoding of event boundaries may trigger a neural mechanism that helps form a bound and coherent memory representation of the previous episodic sequence items. The exact neural mechanism supporting such event boundary function is currently unknown. While theoretical models, as the Temporal Context Model, explain how the sequence items sharing the same context can become associated as a sequence unfolds, in a way that the sequence order is remembered, they do not explain whether and how event boundaries may play a role in episodic memory segmentation and formation of memory episodes. Interestingly, Ben Yakob and Dudai found in 2011 a hippocampal signal time-locked to the offset of an episode. The authors concluded that such signal could act as a neural mechanism that helps ‘wrap-up’ the previously encoded episode. In the current thesis, we examined the possibility that event boundaries could trigger a memory reinstatement of the just encoded episode and that this would act as such ‘wrap-up’ mechanism that promotes memory formation at long-term.
In this thesis, in a series of EEG studies we assessed if memory reinstatement of a just experienced episodic sequence occurred online, time-locked to the onset of an event boundary (Study 1 and Study 2), as well as during retrieval, time-locked to the presentation of a sequence cue item (Study 3). We developed a Spatiotemporal Pattern Similarity (STPS) analysis to assess memory reinstatement. Such analysis is based on the idea that if a memory representation is reinstated at a given moment, the resulting brain activity patterns should be more similar to the original experience encoding brain patterns than if there was no memory reinstatement.
In Study 1, in order to assess if memory reinstatement occurs at event boundaries after an episodic sequence, we adapted a paradigm designed by Dubrow and Davachi (2013), which consists of sequences of faces and objects: a simplification of real-life experiences, as contexts were defined by just two item categories. Event boundaries were defined as category switches (from object to face, or face to object), and episodes were defined as same-category trains of different items. We found memory reinstatement at event boundaries that were presented immediately after an episodic sequence, but not at event boundaries after non-episodic sequences of items. Moreover, such reinstatement was behaviourally relevant, as it correlated with subsequent temporal order memory for within-episode items. Therefore, memory reinstatement had a direct impact on the memory for the previously experienced episode.
We next wondered if memory reinstatement would occur at event boundaries in a more naturalistic experimental paradigm. In order to assess this question, in Study 2 participants were confronted with more naturalistic episodic sequences depicting daily life actions. After each episodic sequence, an incongruent or congruent item was presented, which were conceptualized as expected or unexpected inputs in the unfolded experience that could potentially resemble mechanisms of event boundaries as in Study 1. We hypothesized that incongruent items might elicit a mismatch signal, given that its information does not fit with the behavioural expectations derived from prior knowledge activated during encoding (i.e., schema) which may lead to the reactivation of the just encoded event episode. Our findings revealed a memory reinstatement of the previously presented episode triggered only by incongruent item presentation. Interestingly, the extent of memory reinstatement at incongruent item’s presentation correlated with recognition memory for such items, suggesting that reinstatement was behaviourally relevant in a more naturalistic paradigm.
Finally, we wondered if episodic memory representations are reinstated when appropriately cued at retrieval, such that the temporal structure generated during encoding was preserved at retrieval. Study 3 consisted of sequences depicting everyday activities and participants had to create a different narrative for each sequence. During the retrieval phase, participants were confronted with the first picture of each learned episodic sequence and had to verbally recall the whole sequence. We found that the memory representations of the episodic sequence items reinstated time-locked to the presentation of the cue item. Interestingly, the extent of memory reinstatement of each item correlated with subsequent verbal recall memory for that item. Moreover, recall dependencies were found, such that successful recall of item 2 resulted in a better recall of the following items. This suggests that the temporal structure formed during encoding would be preserved at retrieval. Interestingly, memory reinstatement at retrieval could guide upcoming behaviour, as stored episodic information relevant for the current experience would be immediately available.
All in all, our findings provide the first evidence of online memory reinstatement of the just encoded episodic sequences at event boundaries in humans. Moreover, our results add novel data about the possibility that online memory reinstatement of the just encoded episodic sequence may be triggered when unexpected inputs take place in more naturalistic settings. Finally, we show that the temporal order memory structure representation derived from sequential encoding is rapidly reactivated upon cued-memory retrieval preceding memory recall. Our results support the notion that the rapid memory reinstatement during online encoding as well as during retrieval is a relevant neural mechanism supporting memory formation for sequence event episodes and for the preservation of their temporal structure in long-term memory.