Models · Learning and Memory

Reconsolidation

A model of how established learning can sometimes become susceptible to modification and restabilisation after reactivation.

By Yona Ole Lobulu ·

Model9 min readD5.23

Topic
Learning and Memory
Read first
2 pieces should be read before this one
Reading time
About 9 minutes of reading
Difficulty
Late reading: this page sits at the far end of the Library, after many other pieces

The question

Under what conditions can retrieving established learning make it susceptible to modification before it becomes stabilised again?

Definition

A model of how established learning can sometimes become susceptible to modification after reactivation. Under appropriate conditions, reactivated learning may be destabilised, becoming temporarily susceptible to new information or interference before the affected learning is restabilised. Every transition in the sequence is conditional, and the outcome may be maintenance, strengthening, weakening or selective modification rather than erasure.

Imagine taking the same route through a city for months. You know each turn and landmark. Then one day, a familiar road is closed, and the journey no longer unfolds as expected.

When you travel there again, your remembered route may incorporate the obstruction and the alternative you found. Established knowledge remained stable enough to guide you, yet flexible enough to change.

Several processes could explain that change. You may have formed additional route knowledge, retrieved a different alternative or modified part of the established memory. The observation does not identify the mechanism.

The Reconsolidation Model addresses this question:

Under what conditions can retrieving established learning make it susceptible to modification before it becomes stabilised again?

Why stable memory must remain capable of change

Learning must persist to remain useful. Without stabilisation, previous experience could not reliably guide later recognition, expectation or action.

Yet permanent immutability would create a different problem. Environments and outcomes change, so information that was once accurate can become incomplete or outdated.

Memory therefore faces a stability–plasticity problem: how can learning remain stable enough to persist while retaining the capacity to change when circumstances require it?

As Consolidation Stabilises Learning establishes, consolidation supports persistence without permanently closing learning to later modification.

Reconsolidation represents one possible architecture through which established learning can become modifiable again.

What reconsolidation models

Reconsolidation most precisely refers to the restabilisation of learning after retrieval-dependent destabilisation.

The term is also commonly used for the wider sequence that may surround this restabilisation:

  • established learning is reactivated;
  • particular conditions may produce destabilisation;
  • aspects of the learning may become temporarily susceptible;
  • new information or interference may influence what persists;
  • affected learning is restabilised;
  • later expression provides evidence about the result.

Each transition is conditional.

If destabilisation occurs, the affected learning must be restabilised to persist in its susceptible form. What restabilises may be largely maintained, strengthened, weakened or modified.

Reconsolidation can therefore contribute to preservation as well as change.

Retrieval is not destabilisation

Retrieval makes established learning available under current conditions. It supports recognition, expectation, navigation and other uses of prior experience.

Retrieval can also be followed by several different consequences:

  • continued expression or strengthening;
  • additional encoding;
  • extinction or competing learning;
  • temporary susceptibility to modification;
  • no measurable lasting change.

This distinction builds directly on Context Controls Retrieval.

Context affects what becomes available, which expectations are active and how later behaviour expresses the learning. It can also shape whether the retrieval event introduces meaningful new information.

The changed route, for example, creates a discrepancy between what was expected and what occurred. That discrepancy may contribute to updating, but it does not establish that the existing route memory became destabilised.

Retrieval can create an opportunity for reconsolidation. It does not prove that reconsolidation occurred.

The Reconsolidation Model

Memory properties, learning history, context and timing affect the pathways the Model represents.

Reactivation may be followed by destabilisation, but it may instead support retrieval, strengthening, extinction, additional learning or no demonstrated lasting change. If destabilisation occurs, the memory becomes temporarily susceptible. New information or interference can then affect what is restabilised.

Later expression may be behavioural, physiological, subjective or neural. It provides evidence about what persisted, but it is not the complete memory itself.

The pathways represent conditional possibilities. They are not directly observed stages, compulsory transitions or a universal biological sequence.

When established learning becomes modifiable

No single boundary condition provides a universal test or trigger for reconsolidation. Outcomes depend on interactions among the properties of the memory, the structure of reactivation and what occurs afterward.

Properties of the memory

Memories differ in age, strength, learning history and type.

A recently acquired, weakly reinforced association may respond differently from older or repeatedly reinforced learning. Stronger or older memories can be more resistant to destabilisation or require different reactivation conditions.

Resistance does not mean immutability. It means that a procedure sufficient for one memory may not produce the same susceptibility in another.

Previous retrieval, extinction, contextual variation and earlier updating can also alter what becomes active and how it responds.

Conditioned fear, spatial learning, skills, episodic events and autobiographical memory should not be treated as equivalent. They differ in content, measurement and biological support.

Reactivation conditions

The duration, context and content of retrieval can alter its consequences.

A limited reminder may retrieve established learning without producing evidence of destabilisation. More extensive unreinforced exposure may support extinction, depending on memory strength, expectations, context and the wider procedure.

Timing matters as well. The intervals between acquisition, reactivation, intervention and testing can affect both the result and its interpretation.

There is therefore no reminder procedure or fixed "reconsolidation window" that can be applied universally across memories.

Prediction error and mismatch

Prediction Error Drives Updating explains how discrepancies between expected and actual events can signal that existing learning may need revision.

Several reconsolidation accounts propose that mismatch contributes to destabilisation. Human fear-conditioning studies have supported this possibility under defined conditions, while closely related studies have not reproduced the effect consistently.

Prediction error also sits alongside novelty, uncertainty and new contextual information. These ideas may overlap without being interchangeable.

The evidence supports a bounded conclusion: prediction error can contribute to destabilisation and updating under some conditions, but it neither guarantees reconsolidation nor functions as a universal requirement.

What can change?

Memory contains more than one type of information.

Relevant components may include:

  • remembered content;
  • context and expectancy;
  • emotional and physiological responding;
  • action tendencies and later interpretation.

These components can change differently.

Physiological responding may decrease while explicit knowledge remains. An emotional response may change without removing factual content. Behaviour may change in one context and return in another.

Evidence that one measured component changed does not show that the complete memory was replaced.

This is why "rewriting" can be a misleading general metaphor. It suggests that memory behaves like a single file whose previous contents are edited and saved over. Reconsolidation-compatible modification may instead be selective, incomplete and dependent on how later expression is tested.

Reconsolidation is not extinction

Reconsolidation and extinction can both follow retrieval, but they describe different explanatory processes.

Reconsolidation concerns restabilisation after retrieval-dependent destabilisation.

Extinction concerns learning that changes the expression of an established relation. As Extinction Is New Learning establishes, reduced responding after extinction does not generally mean that prior learning disappeared.

Reactivation duration can affect which interpretation becomes plausible, but duration operates together with memory strength, context, expectations, timing and the broader experimental procedure.

Researchers have tested whether extinction delivered after a reminder can alter later fear responding during a proposed period of susceptibility. Some studies report less return of fear. Other studies have produced weaker, different or unsuccessful results.

Even durable behavioural reduction can remain compatible with several explanations:

  • modification of established learning;
  • competing extinction learning;
  • altered retrieval;
  • context-dependent expression;
  • interacting processes.

Reconsolidation and extinction are conceptually distinct, although behaviour may not reveal a clean boundary between them.

How reconsolidation is inferred

Researchers infer reconsolidation from controlled patterns rather than observing the complete process directly.

A typical experiment:

  • establishes and stabilises learning;
  • reactivates it under specified conditions;
  • delivers a timed intervention or new information;
  • measures later expression against relevant comparisons.

A strong inference requires that learning and reactivation were established, intervention and timing were controlled, later change was selective and major alternatives were constrained.

Animal research can combine behavioural findings with pharmacological, cellular or molecular manipulation. Human studies rely more heavily on behavioural, physiological and subjective measures, making direct mechanistic identification more difficult.

Reduced responding can support a reconsolidation interpretation. It cannot identify that mechanism by itself.

Why absence of return is not proof of erasure

Researchers may test whether responding returns:

  • over time;
  • in another context;
  • after renewed exposure to an outcome;
  • during reacquisition.

These tests correspond to spontaneous recovery, renewal, reinstatement and reacquisition. They help distinguish durable change from temporary suppression.

If responding does not return, the conclusion must still remain bounded. Another context, response system or memory component might reveal residual learning.

This connects forward to Why Old Learning Returns and to Generalisation and Discrimination, which explain how expression can vary across cues and contexts.

The precise finding may simply be that the previous response did not return under the tested conditions and measures.

From laboratory memory to human change

Animal research provides the strongest mechanistic foundation for reconsolidation. Acquisition, reactivation, intervention and biological measurement can be controlled precisely. These studies support post-reactivation susceptibility and restabilisation in defined memory systems without establishing one universal molecular pathway.

Human laboratory research also supports post-retrieval modification, including in fear-conditioning and episodic-memory tasks. The findings are more procedurally sensitive. Differences in instructions, reminders, prediction error, timing, context, awareness and measurement can alter outcomes.

Some influential human effects have been reproduced or extended, while others have failed to replicate. A general phenomenon of retrieval-dependent plasticity should not be confused with guaranteed success from one reactivation procedure.

This is the protection supplied by Replication, Robustness and Scientific Confidence. Scientifically useful boundary conditions should be specified and tested rather than invoked only after a procedure fails.

Translation becomes more difficult for autobiographical and clinical memory.

Autobiographical memory may include repeated experiences, identity, bodily responses, narrative meaning, social context and many retrieval histories. Reactivating one component does not imply that the whole memory entered one susceptible state.

Clinical improvement after memory reactivation may involve reconsolidation, but it may also reflect exposure, extinction, expectancy, reappraisal, contextual learning, therapeutic interaction or several processes together.

The Model can guide questions about durable change. It does not establish that recalling an experience automatically opens a window in which a complex memory can be rewritten.

The Model is not the memory system

A behavioural change is an observation. Destabilisation and restabilisation are process-level inferences. The Reconsolidation Model organises relationships among those proposed processes.

At the biological level, distributed and overlapping mechanisms contribute to retrieval, plasticity and persistence. A neural correlate of retrieval or plasticity is not identical to a Model component such as destabilisation or restabilisation.

This is the permanent protection supplied by Scientific Models Are Tools: the Model is not the target system.

The Model does not represent a literal internal switch, one reconsolidation centre or a compulsory sequence implemented identically across memory systems.

A successful Model fit shows that a representation accounts for a measured pattern under stated assumptions. It does not prove that the memory system implements the Model literally.

Stable does not mean fixed forever

Stability gives memory continuity. Conditional plasticity gives it the capacity to adapt.

Established learning can sometimes become modifiable after reactivation.

Reconsolidation models this conditional possibility without implying that every retrieval destabilises, rewrites or erases memory.

Behind this page

The claims this model makes, the evidence behind them, and the limits it accepts.

Evidence status

Established

Well supported by a substantial, converging empirical literature.

Claims

  1. Established learning can remain stable enough to guide behaviour while retaining the capacity to change

    Established

    What this does not assert: The stability–plasticity problem the Model addresses.

  2. Retrieval makes established learning available under current conditions

    Established

    What this does not assert: Availability is not modification.

  3. Retrieval can be followed by continued expression, strengthening, additional encoding, extinction, temporary susceptibility or no measurable lasting change

    Established

    What this does not assert: The consequence is not fixed by the act of retrieval.

  4. Retrieval does not automatically destabilise memory

    Established

    What this does not assert: Central safeguard: retrieval ≠ destabilisation.

  5. Context affects what becomes available, which expectations are active and how learning is later expressed

    Established

    What this does not assert: Inherited from D5.8, Context Controls Retrieval.

  6. Context can shape whether a retrieval event introduces meaningful new information

    High confidence

    What this does not assert: A condition on destabilisation rather than a guarantee of it.

  7. A discrepancy between expectation and outcome may contribute to updating without establishing that established learning was destabilised

    Canonical inference

    What this does not assert: The route example is illustrative, not evidential.

  8. Retrieval can create an opportunity for reconsolidation

    High confidence

    What this does not assert: Opportunity is not occurrence.

  9. Memory properties, learning history, context and timing affect which pathways in the Model become possible

    Established

    What this does not assert: Interacting conditions, not a single controlling variable.

  10. Later expression may be behavioural, physiological, subjective or neural, and is evidence about what persisted rather than the memory itself

    Canonical inference

    What this does not assert: Measurement is not identity.

  11. The Model's pathways are conditional possibilities, not directly observed stages or compulsory transitions

    Canonical inference

    What this does not assert: Diagrammatic form must not be read as a biological sequence.

  12. An observed change in remembered content does not by itself identify the process that produced it

    Canonical inference

    What this does not assert: Additional learning, altered retrieval and modification remain competing accounts.

  13. No single boundary condition provides a universal test or trigger for reconsolidation

    Established

    What this does not assert: Boundary conditions interact.

  14. Memories differ in age, strength, learning history and type

    Established

    What this does not assert: These differences change what reactivation can do.

  15. Stronger or older memories can be more resistant to destabilisation or require different reactivation conditions

    Established

    What this does not assert: Reported across animal and human work; the boundaries are not fully specified.

  16. Resistance to destabilisation does not mean immutability

    High confidence

    What this does not assert: A procedure sufficient for one memory may not suffice for another.

  17. Previous retrieval, extinction, contextual variation and earlier updating alter what becomes active and how it responds

    Established

    What this does not assert: Retrieval history is part of the memory's state.

  18. Conditioned fear, spatial learning, skills, episodic events and autobiographical memory should not be treated as equivalent

    Canonical inference

    What this does not assert: They differ in content, measurement and biological support.

  19. The duration, context and content of retrieval can alter its consequences

    Established

    What this does not assert: Reactivation is a structured procedure, not a switch.

  20. A limited reminder may retrieve established learning without producing evidence of destabilisation

    Established

    What this does not assert: Absence of evidence here is about the procedure, not the capacity.

  21. More extensive unreinforced exposure may instead support extinction, depending on strength, expectation, context and procedure

    Established

    What this does not assert: The reactivation–extinction boundary is procedure-dependent.

  22. Intervals between acquisition, reactivation, intervention and testing affect both result and interpretation

    Established

    What this does not assert: Timing is a design variable, not a fixed constant.

  23. Consolidation supports persistence without permanently closing learning to modification

    Established

    What this does not assert: Inherited from D5.22, Consolidation Stabilises Learning.

  24. No reminder procedure or fixed reconsolidation window can be applied universally across memories

    High confidence

    What this does not assert: Central safeguard against a generic clinical protocol.

  25. Discrepancies between expected and actual events can signal that existing learning may need revision

    Established

    What this does not assert: Inherited from D5.9, Prediction Error Drives Updating.

  26. Several reconsolidation accounts propose that mismatch contributes to destabilisation

    Contested

    What this does not assert: A theoretical proposal with partial empirical support.

  27. Human fear-conditioning studies have supported a prediction-error contribution under defined conditions, while closely related studies have not reproduced it consistently

    Contested

    What this does not assert: The inconsistency is itself part of the evidential picture.

  28. Prediction error sits alongside novelty, uncertainty and new contextual information without being interchangeable with them

    Canonical inference

    What this does not assert: Overlapping constructs are not identical constructs.

  29. Prediction error neither guarantees reconsolidation nor functions as a universal requirement for it

    High confidence

    What this does not assert: Central safeguard: prediction error ≠ reconsolidation.

  30. Memory includes remembered content, context and expectancy, emotional and physiological responding, and action tendencies and interpretation

    Established

    What this does not assert: Components are measured separately and can change separately.

  31. Physiological responding may decrease while explicit knowledge remains

    Established

    What this does not assert: A recurring dissociation in human fear-memory studies.

  32. An emotional response may change without removing factual content

    Established

    What this does not assert: Change is often selective rather than total.

  33. Behaviour may change in one context and return in another

    Established

    What this does not assert: Expression is context-dependent.

  34. Consolidation does not imply immutability

    High confidence

    What this does not assert: Stability is a matter of degree, not a locked state.

  35. Evidence that one measured component changed does not show that the complete memory was replaced

    Canonical inference

    What this does not assert: Central safeguard: modification ≠ erasure.

  36. The rewriting metaphor misleads by implying that memory behaves like a single file edited and saved over

    Canonical inference

    What this does not assert: Modification may be selective, incomplete and measurement-dependent.

  37. Reconsolidation and extinction describe different explanatory processes

    High confidence

    What this does not assert: Both can follow retrieval; they are not the same account.

  38. Reduced responding after extinction does not generally mean that prior learning disappeared

    Established

    What this does not assert: Inherited from D5.6, Extinction Is New Learning.

  39. Reactivation duration operates together with memory strength, context, expectation, timing and procedure

    Established

    What this does not assert: No duration by itself settles the interpretation.

  40. Some post-retrieval extinction studies report less return of fear while others report weaker, different or unsuccessful results

    Contested

    What this does not assert: Replication of the human effect is uneven.

  41. Durable behavioural reduction remains compatible with modification, competing learning, altered retrieval, context-dependent expression or interacting processes

    Canonical inference

    What this does not assert: Behavioural durability underdetermines mechanism.

  42. Researchers infer reconsolidation from controlled patterns rather than observing it directly

    Canonical inference

    What this does not assert: The process is not observable in the way behaviour is.

  43. A strong inference requires established learning and reactivation, controlled intervention and timing, selective change and constrained alternatives

    High confidence

    What this does not assert: The four conditions the Library treats as minimal.

  44. Animal research can combine behavioural findings with pharmacological, cellular or molecular manipulation

    Established

    What this does not assert: This is why the mechanistic foundation is strongest there.

  45. Reconsolidation most precisely refers to the restabilisation of learning after retrieval-dependent destabilisation

    Established

    What this does not assert: The narrow, technical use of the term.

  46. Human studies rely more heavily on behavioural, physiological and subjective measures, making mechanistic identification harder

    Established

    What this does not assert: A methodological limit, not a claim about human memory.

  47. Reduced responding can support a reconsolidation interpretation but cannot identify the mechanism

    Canonical inference

    What this does not assert: Central safeguard: behavioural change ≠ unique evidence of reconsolidation.

  48. Spontaneous recovery, renewal, reinstatement and reacquisition tests help distinguish durable change from temporary suppression

    Established

    What this does not assert: Standard return-of-responding tests.

  49. Absence of return under tested conditions does not establish that the learning was erased

    High confidence

    What this does not assert: Central safeguard: absence of return ≠ proof of erasure.

  50. Another context, response system or memory component might still reveal residual learning

    Canonical inference

    What this does not assert: The bounded form the conclusion must take.

  51. Animal studies support post-reactivation susceptibility and restabilisation in defined memory systems

    Established

    What this does not assert: Without establishing one universal molecular pathway.

  52. Human laboratory research supports post-retrieval modification in fear-conditioning and episodic-memory tasks

    Established

    What this does not assert: The findings are more procedurally sensitive than the animal work.

  53. Instructions, reminders, prediction error, timing, context, awareness and measurement can alter human outcomes

    Established

    What this does not assert: Procedural sensitivity is the norm, not the exception.

  54. Some influential human effects have been reproduced or extended while others have failed to replicate

    Contested

    What this does not assert: Replication status differs by paradigm and measure.

  55. A general phenomenon of retrieval-dependent plasticity should not be confused with guaranteed success from one reactivation procedure

    Canonical inference

    What this does not assert: Phenomenon and protocol are different claims.

  56. The term is also commonly used for the wider sequence of reactivation, possible destabilisation, susceptibility and restabilisation

    Canonical inference

    What this does not assert: Usage varies across literatures; the Model states which sense is meant.

  57. Boundary conditions should be specified and tested rather than invoked only after a procedure fails

    High confidence

    What this does not assert: Inherited from D2.13, on replication and robustness.

  58. Autobiographical memory may include repeated experience, identity, bodily responses, narrative meaning, social context and many retrieval histories

    Established

    What this does not assert: Why translation from laboratory paradigms is difficult.

  59. Reactivating one component of an autobiographical memory does not imply that the whole memory entered one susceptible state

    Canonical inference

    What this does not assert: Component-level susceptibility is not whole-memory susceptibility.

  60. Clinical improvement after memory reactivation may reflect exposure, extinction, expectancy, reappraisal, contextual learning, therapeutic interaction or several processes together

    Canonical inference

    What this does not assert: Central safeguard: therapeutic change ≠ proof of reconsolidation.

  61. A laboratory effect does not by itself identify the mechanism operating in clinical change

    High confidence

    What this does not assert: Central safeguard: laboratory effect ≠ clinical mechanism.

  62. The Model does not establish that recalling an experience opens a window in which a complex memory can be rewritten

    Canonical inference

    What this does not assert: No procedure is implied or recommended by the Model.

  63. Destabilisation and restabilisation are process-level inferences rather than observations

    Canonical inference

    What this does not assert: Behavioural change is the observation.

  64. A neural correlate of retrieval or plasticity is not identical to a Model component such as destabilisation

    High confidence

    What this does not assert: Central safeguard: neural correlate ≠ model-component identity.

  65. Distributed and overlapping biological mechanisms contribute to retrieval, plasticity and persistence

    Established

    What this does not assert: No single region implements the Model.

  66. A successful Model fit shows that a representation accounts for a measured pattern under stated assumptions

    Canonical inference

    What this does not assert: It does not prove literal implementation. Inherited from D2.12.

  67. Each transition in the reconsolidation sequence is conditional

    Canonical inference

    What this does not assert: No step follows automatically from the one before it.

  68. Established learning can sometimes become modifiable after reactivation, without every retrieval destabilising, rewriting or erasing memory

    Canonical inference

    What this does not assert: The core thesis of this Model.

  69. If destabilisation occurs, the affected learning must be restabilised to persist in modified form

    Established

    What this does not assert: Restabilisation is what the term names most precisely.

  70. Reconsolidation can contribute to preservation as well as to change

    High confidence

    What this does not assert: What restabilises may be maintained, strengthened, weakened or modified.

Helpful beforehand

Useful context, but not required.

Where to go from here

Next published piece

What Reconsolidation Research Can Support

A research evidence audit of memory reconsolidation: what the findings support, and where stronger claims about destabilisation, erasure and clinical translation exceed them.

Continue through the Library →See where this sits in the graph →

Back to the Library →