Journal article in Trends in Cognitive Science, Volume 21, Issue 8.

Access [free]: Antony et al. (2017) via ResearchGate

Editor's note: this week's paper is from a purely psychological setting, rather than a pedagogical setting. As such we've included some definition boxes, like this one, to help with reading the summary.

Scope of Paper

The paper proposes an explanation at the neural level for why retrieval practice improves a) memory, and b) the speed at which information can be stored to long-term memory.

Key Findings

Retrieval practice is a ‘rapid consolidation event’, which supports and speeds up the creation of hippocampal-neocortical representations of information (schema). This happens via the reactivation of associated memories.

Specifically, retrieval improves memory by integrating information with other neocortical knowledge and differentiating it from competing memories. This speeds up the transfer of new information into long-term memory, and makes specific memories easier to identify and access in future recall attempts.

More Detail

Retrieval practice: repeatedly and actively retrieving information from one's memory.

It’s well established that retrieval practice is a much more effective method of improving long-term retention of information than restudying. The benefits of retrieval practice are evident regardless of the topic being studied or the method of testing.

What’s less clear is why retrieval practice improves learning and memory from a neural perspective.

Hippocampus: A major part of the brain that is involved in consolidating information from short-term working memory to long-term memory.

Neocortex: A major part of the brain that is involved in cognition and long-term memory.

Evidence from studies performed on the brains of amnesia patients shows that when the brain suffers hippocampal damage, the creation of new memories is affected while long-term memories are left intact. There is a suggestion, therefore, that the hippocampus and the neocortex work together to allow new learning while preserving existing memories.

Specifically, the hippocampus learns quickly, and specialises in rapidly encoding and binding together new cortical associations. The neocortex learns slowly, and specialises in storing the statistical structure of experiences.

Cortical associations: The integration of new information with existing memories. In some cases, the new information may link multiple areas of existing memory.

Schema: Webs and patterns of information and the associations between them, held in long-term memory. All knowledge and memories are stored within schema.

The interaction of the two systems allows new information to slowly influence existing schema within long-term memory. However, if the hippocampus is damaged before enough hippocampal-neocortical interactions occur, the function of long-term memory will be impaired. This suggests that the interaction of the two systems is required to form stable, long-term memories. The process is known as systems-level consolidation.

When we sleep, memories undergo systems-level consolidation by the process of ‘offline’ reactivation, or replay. This occurs in hippocampal-neocortical circuits. The authors propose that retrieval practice promotes long-term retention in a similar way, triggering an ‘online’ process by prompting pattern completion.

Retrieval (and sleep) qualitatively transform memories in at least two distinct ways: by integrating new memories into existing schema, and by differentiating memories, which minimises competition between overlapping memories.

Both of these processes are explained by the tendency of retrieval to be imprecise. In the act of identifying the memory, retrieval simultaneously activates memories that related to the target memory.

Repeated imprecise retrieval gives the brain the same opportunity to cycle memories as occurs during sleep. Memories that are initially hippocampus-dependent are integrated into neocortical schema of long-term memories.

The coactivation of hippocampus and neocortex is a strong activation of a memory, which consolidates the integration of the memory. In comparison, restudy is a weaker activation of memory, and does not involve the coactivation of the hippocampus and neocortex. As such, memories are left isolated and distinct from other strongly integrated memories within schema.

Crucially, if retrieval rapidly embeds a memory in the neocortex, future attempts at retrieval can use both hippocampal and neocortical representations to identify the memory. As restudy does not activate these processes, recall attempts are not able to use both resources.

By invoking mechanisms in a similar way to sleep, retrieval practice stabilises memories and increases the strength with which they are stored. The authors note that sleep has other unique characteristics that are not shared with retrieval.


While most frameworks suggest that hippocampal-neocortical interactions store memories slowly over time, the authors suggest that retrieval specifically accelerates this processes.

It does so by invoking similar processes to the offline, systems-level consolidation that occurs between the hippocampus and the neocortex during sleep.