The question
Why can a previously reduced response return after time has passed, even when genuine new learning occurred?
Definition
Spontaneous recovery is the return of a previously reduced response after time has passed following extinction or comparable new learning.
A learned response can fade and later return.
A sound that once triggered fear may stop doing so after repeated safe encounters. An action that once produced an expected result may become less frequent when that result no longer follows. For a while, the change can look complete. Then time passes, and part of the old response reappears.
It is tempting to treat this as proof that nothing really changed. If the response came back, perhaps it was never truly gone.
But learning does not always work in such all-or-nothing terms. A response can become less available without the earlier learning being erased. New learning can be genuine even when it does not control behaviour permanently or under every condition.
The return of a previously reduced response after time has passed is known as spontaneous recovery.
What spontaneous recovery is
Spontaneous recovery is the reappearance of a learned response after a delay following extinction.
The defining sequence has four parts:
- A response is learned.
- The response meaningfully declines during extinction.
- A period of time passes.
- Responding increases again at a later test.
A response returning after time has passed is not enough by itself. There must first have been a meaningful reduction. If the response never declined, there is nothing to recover. The later increase must also be evaluated against an appropriate measure taken after extinction or following a shorter delay. Otherwise, ordinary variation could be mistaken for recovery.
In experimental learning, extinction occurs when a cue or action is repeatedly encountered without the outcome that previously followed it. A sound may have predicted an aversive event and therefore begun to evoke a defensive response. When the sound is repeatedly presented without that event, responding usually decreases.
If the sound is tested soon afterward, responding may remain low. If it is tested after a longer interval, part of the response may return. That delayed increase is spontaneous recovery.
The phenomenon has been studied across several forms of learning, although its expression and measurement differ between Pavlovian, instrumental and human experimental paradigms.
"Spontaneous" does not mean random or causeless. It does not mean that the response appeared independently of the earlier learning history. It identifies a return after time has passed without renewed training of the original contingency being required to produce the initial increase.
The retention interval is therefore not incidental. It is the defining feature of the procedure.
Extinction does not necessarily erase earlier learning
Spontaneous recovery matters because of what it reveals about extinction.
When responding declines, the simplest interpretation is that the original learning has been removed. The cue no longer produces the same response, so perhaps the earlier relation no longer exists.
But reduced responding is not proof of erased learning.
An extinguished response can return under several conditions: after time has passed, after the context changes, after a relevant outcome is encountered again or after the original contingency resumes. These return phenomena are difficult to reconcile with the idea that extinction necessarily deletes the original learning.
Extinction can instead establish additional learning that the cue or action now has different consequences under the current conditions. In some cases, this is described as learning that an expected outcome will not occur. In others, it may involve learning that another response is appropriate or that the relation has changed in this particular context.
The earlier and newer learning may therefore remain capable of influencing behaviour. What is observed at a given moment depends partly on which learning is retrieved and expressed under the conditions that exist then.
This does not mean that extinction always leaves the original learning entirely unchanged. Nor does it establish one universal extinction mechanism. Different accounts emphasise competing associations, contextual control, prediction updating, inference about changing situations or combinations of these processes.
The more precise conclusion is narrower:
Extinction changes responding without necessarily erasing what was learned before.
Performance provides evidence about what is controlling behaviour under the current conditions. It does not reveal everything the learning system retains.
How time can change what is expressed
Why can responding increase after a delay?
One leading explanation is that extinction learning is especially sensitive to context. Context includes more than physical location. It can include surrounding cues, internal state, recent events and temporal conditions.
Soon after extinction, the conditions at testing closely resemble those under which the newer learning occurred. The extinction experience is recent, and the cues that support its retrieval may be readily available.
As time passes, the conditions under which learning is retrieved and expressed can change. The delayed test may be less similar to extinction's temporal context. The newer learning may consequently exert less control, allowing the earlier learning to influence responding more strongly again.
From this perspective, spontaneous recovery resembles renewal. In renewal, responding returns when testing occurs outside the context in which extinction took place. In spontaneous-recovery procedures, elapsed time is the defining manipulation, although other contextual changes may accompany the delay.
Temporal context is a leading explanation, not a complete universal mechanism. Other compatible accounts describe the return through differential retention, changing accessibility, generalisation or competition between learning acquired at different times.
It is therefore too strong to say that time simply weakens a new memory or strengthens an old one. A delay changes the testing conditions in ways that can alter which learning controls current behaviour. The result depends on the organism's learning history, the extinction procedure, the length of the interval, the test conditions and the response being measured.
Spontaneous recovery is possible, not inevitable. Its magnitude varies, and under some conditions little or no recovery appears.
Return does not cancel the change that came before it
Spontaneous recovery creates an apparent contradiction:
- the response genuinely declined;
- the response later returned.
Both can be true.
New learning does not have to be false for earlier learning to regain influence. A return may instead show that the newer learning does not control behaviour equally well across all times and conditions.
The recovered response may also be partial. It may be stronger than it was at the end of extinction without reaching its earlier level. It may appear in one measure but not another. A person might still expect safety while showing increased physiological arousal, for example.
The system after extinction is therefore not necessarily identical to the system before extinction. Its learning history has changed, even when part of the earlier response becomes visible again.
This matters because recurrence is often interpreted as total reversal. An old reaction appears, and the conclusion follows that all progress has disappeared.
Spontaneous recovery shows why that conclusion can be mistaken. A returning response demonstrates that earlier learning remains capable of expression. It does not demonstrate that everything learned afterward has vanished.
The return is real and should not be dismissed. But it should not be inflated into proof that no change occurred.
Different returns have different conditions
Spontaneous recovery is one form of return after extinction. It is not a general name for every recurrence.
Renewal occurs when responding returns following a change in context. A response may remain low where extinction occurred but reappear under different conditions.
Reinstatement occurs after exposure to the relevant outcome or unconditioned stimulus following extinction. Its effect is often context-dependent and should not be reduced to the idea that any reminder restores the response.
Reacquisition occurs when the original contingency is presented again. If the cue once more predicts the outcome, responding may be learned again, sometimes rapidly.
Their defining conditions differ:
- spontaneous recovery is organised around a retention interval;
- renewal is organised around context change;
- reinstatement follows relevant outcome exposure after extinction;
- reacquisition follows renewed exposure to the original contingency.
Ordinary fluctuation is different again. Responses naturally vary. A temporary increase does not establish a specific return mechanism without the sequence and comparisons needed to identify it.
Relapse and recurrence are broader categories. They describe that something returned, not why it returned. Spontaneous recovery, renewal, reinstatement and reacquisition are possible contributors to that broader outcome.
In everyday life, several processes can operate together. Time passes while contexts, internal states and social conditions change. A person may encounter reminders, stressors or rewards that resemble earlier circumstances. The original contingency may partly return.
A recurrence observed after several weeks cannot therefore be classified as spontaneous recovery from timing alone.
The limit of the concept
The clearest evidence for spontaneous recovery comes from controlled learning experiments. These studies allow acquisition, extinction, delay and testing to be separated in ways that ordinary life rarely permits.
Complex human patterns involve more. Beliefs, goals, habits, physiological states, avoidance, reinforcement, relationships and personal meaning can all change together. A returning fear, urge or behaviour may resemble spontaneous recovery without being produced by that process alone.
The concept should therefore guide interpretation, not function as a diagnosis.
It shows that delayed return can occur without renewed training of the original contingency. But real-world recurrence may also involve additional influences that are difficult to observe.
Timing tells us that spontaneous recovery is possible. Timing alone does not tell us why a particular person's response returned.
What spontaneous recovery changes
Change is often judged from a single observation: Is the response present or absent now?
Spontaneous recovery shows why that question is insufficient. Current behaviour reflects both a learning history and the conditions under which parts of that history become influential.
Something that is not being expressed may remain available.
Something that returns may do so without eliminating what was learned in between.
The earlier learning may remain capable of expression. The later learning may remain real. Which one controls behaviour can change across time and conditions.
A returning response is therefore not automatic proof that change failed.
It is evidence that learning can persist beyond what behaviour reveals at any single moment.