Memory search and the neural representation of context

Abstract

A challenge for theories of episodic memory is to determine how we focus memory search on a set of recently learned items. Cognitive theories suggest that the recall of an item representation is driven by an internally maintained context representation that integrates incoming information with a long time-scale. Neural investigations have shown that recalling an item revives the pattern of brain activity present during its study. To link these neural and cognitive approaches, we propose a framework in which context is maintained and updated in prefrontal cortex, and is associated with item information through hippocampal projections. The proposed framework is broadly consistent with neurobiological studies of temporal integration and with studies of memory deficits in individuals with prefrontal damage.

Figure I

A schematic of the temporal context model (TCM) [8] at the end of a short study list: apple, cat, boat, dog. As each item is studied, the state of context is updated by incorporating features related to that item into the context representation; as new features are added the strength of the existing features is diminished. The currently studied item is then associated with the current state of context. The context representation can be used to guide memory search; by projecting back to the item feature layer it can provide support for recently seen items in a recall competition.

Figure 1

Patterns of category-related cortical activity gain strength before the recall of an item from that category during free recall [25] as shown by an event-related average of classifier estimates of category strength locked to the onset of recall verbalization. Time is indexed in terms of fMRI scans (each scan lasts 1.8 s and provides a full snapshot of brain activity). The black line corresponds to the average strength of the to-be-recalled category, centered on the set of recalled items for which there were no same-category recalls in the preceding 14 s. The red line corresponds to the average strength of any recently recalled categories, and the purple line corresponds to the average strength of any categories that had not been recently recalled; this serves as a baseline against which the emergence of the to-be-recalled category pattern can be assessed, which shows a significant increase 5.4 s before recall. Reproduced with permission from [25].

Figure 2

Brain oscillations associated with successful encoding are reinstated during correct retrieval [27]. The top row of brain maps contrasts gamma-band oscillatory activity during the two-second item presentation for items subsequently recalled and those that were forgotten. The bottom row contrasts gamma-band oscillations during the 500 ms preceding recall verbalization for correct items and for prior-list intrusions. In each map, red corresponds to regions where the contrast was significant, gray corresponds to non-significant contrasts and black indicates brain regions excluded from the analysis owing to insufficient electrode coverage. Adapted with permission from [27].

Figure 3

Prefrontal cortex (PFC) as the seat of temporal context. In this framework neural activity in temporal cortex represents the features of studied items; these update the state of a prefrontal representation of temporal context. Both item and context representations project to the hippocampus, where an episodic association is formed between the two. Direct connections between prefrontal and temporal cortices might also have a role in this process.