Reinstatement

A response can return even when the cue that once produced it has not been directly retrained.

Imagine that a sound has repeatedly predicted an aversive event. The sound begins to evoke a defensive response. It is then presented safely until responding declines.

After extinction, the aversive event occurs again—but the sound is not present. Later, when the sound is tested, the defensive response returns.

The sound and the event were never paired again. The original cue–outcome relation was not directly retrained. Yet something that happened after extinction changed how the organism responded when the sound returned.

This pattern is called reinstatement.

What reinstatement is

Reinstatement is the return of responding to an extinguished cue after exposure to the relevant outcome or unconditioned stimulus following extinction.

An unconditioned stimulus is an event that can naturally elicit a response before learning—for example, an aversive event used during fear conditioning.

The standard reinstatement sequence has five parts:

1. A cue predicts an outcome, and a response is learned.
2. The cue is repeatedly presented without that outcome.
3. Responding meaningfully declines.
4. The relevant outcome is presented after extinction without the cue being directly retrained.
5. The cue is tested again, and responding increases.

Extinction is the procedure in which the cue is presented without the outcome that previously followed it. Successful response reduction must then be demonstrated before a later increase can be attributed to reinstatement.

A response appearing after an event is not enough. Researchers must show that responding was acquired, meaningfully reduced and then increased following the reinstatement manipulation relative to an appropriate earlier measure or control condition.

Without those comparisons, residual responding, ordinary fluctuation or another return mechanism could be mistaken for reinstatement.

In the standard procedure, the extinguished cue is absent when the relevant outcome occurs. Experimental variants require additional controls to show that the effect is reinstatement rather than direct retraining.

The defining event must also be relevant to the earlier learning structure. An unrelated stressor, an ordinary reminder or any unpleasant experience is not automatically a reinstating event.

The cue was not relearned directly

Reinstatement is sometimes described as if the old response were simply trained again. But that misses its defining feature.

In reacquisition, the original contingency resumes. The cue and outcome are paired again, and responding develops under renewed reinforcement.

In reinstatement, the relevant outcome occurs after extinction, but the cue is not directly paired with it. The event does not explicitly teach that the cue predicts the outcome again.

That does not mean that no learning occurs.

The outcome can give new predictive significance to the context in which it occurs. The setting may begin to signal danger, reward or another relevant consequence. When the extinguished cue is later encountered under those conditions, contextual expectation can affect how the cue is interpreted and which parts of the learning history influence behaviour.

The cue was not directly retrained. The conditions surrounding its later presentation changed.

Related reinstatement effects also occur after instrumental extinction. A former reinforcer can restore an extinguished action even when it is presented without directly reinforcing that action again. The procedural details differ, so instrumental and Pavlovian reinstatement should not be treated as identical.

Context can help restore responding

One of the strongest findings in reinstatement research is that the effect is often context-dependent.

If the relevant outcome occurs in one setting and the extinguished cue is later tested in that same setting, responding commonly returns. When testing occurs somewhere else, reinstatement may be weaker or absent.

A leading explanation is that contextual cues acquire or renew predictive significance during the reinstating event.

Before the event, the context may signal safety or carry little expectation of the original outcome. When that outcome occurs again, the surrounding conditions can become predictive of its possibility.

Later, the extinguished cue is encountered where the outcome has again become expected. Context-based expectancy may combine with preserved cue-based learning, making renewed responding more likely.

This is sometimes described as summation: influence from the context combines with the remaining influence of the cue.

Context may also affect retrieval. Extinction gave the cue an additional significance: under particular conditions, its earlier outcome does not occur. A reinstating event may make the earlier cue–outcome relation easier to retrieve or reduce the control exerted by extinction learning.

These explanations are compatible, and no single mechanism accounts for every reinstatement result. Changes in contextual inhibition, attention, arousal and retrieval may all contribute under different conditions.

Context also means more than physical location. It can include surrounding sensory cues, recent events, temporal conditions and internal state. Evidence is strongest for experimentally manipulated environments, so broader claims about internal or conceptual contexts require caution.

Matching the reinstatement and test contexts often strengthens the effect. But experimental findings do not support treating identical physical context as an absolute rule across every procedure and response measure.

Context is therefore central to many demonstrations of reinstatement without functioning as a universal switch that automatically restores an extinguished response.

Extinction learning can remain real

When responding returns, it is easy to conclude that extinction failed.

But reinstatement does not prove that the learning established during extinction was false or erased.

During extinction, the organism encountered the cue without its earlier outcome. Responding declined, and additional learning may have formed. The later outcome exposure does not remove that history. It adds another event to it and can alter the conditions under which different parts of the history influence behaviour.

Earlier cue–outcome learning may remain capable of expression. Extinction learning may also remain available. Which learning controls behaviour can depend on context, retrieval and the particular response being measured.

The return may be partial rather than complete. Responding may increase without reaching its original level. It may appear for one cue but not another, or in one setting but not elsewhere.

Different measures may also produce different results. Physiological arousal can increase without a matching change in explicit expectation or subjective fear. Each measure captures a different component of responding, and disagreement among them is itself informative.

The system after reinstatement is therefore not necessarily identical to the system before extinction.

Reinstatement shows that a relevant event can make earlier learning influential again. It does not show that everything learned during extinction has disappeared.

Different returns have different conditions

Reinstatement is one way that responding can return. It is not a general name for recurrence.

Spontaneous recovery is return after a retention interval following extinction. Time is the defining manipulation.

Renewal is return associated with a context change. Responding may remain low in the extinction context but reappear elsewhere.

Reacquisition occurs when the original contingency resumes. The cue and outcome are paired again, allowing direct renewed learning.

Direct retraining explicitly pairs the extinguished cue with its former outcome. This differs from the standard reinstatement procedure, in which the outcome is presented without the cue.

Generalisation occurs when responding extends across related cues or situations. It does not require the reinstatement sequence.

Ordinary fluctuation is response variation without sufficient evidence of a specific return mechanism.

Relapse and recurrence are broader outcome categories. They describe that something returned, not what caused it.

These processes can overlap outside controlled experiments. An adverse event may occur after a long delay, in a new context, while parts of the original contingency are also returning. A later response could reflect reinstatement, spontaneous recovery, renewal, reacquisition or several processes together.

The presence of one visible feature does not establish the complete mechanism.

A real-world resemblance is not a diagnosis

Reinstatement offers one way to understand how events occurring after change can affect later responding.

A person may become less reactive to a situation through repeated safe experiences. A later adverse event may then make the old reaction more available when the earlier situation is encountered again.

That sequence is only reinstatement-like unless the event is functionally related to the earlier outcome and competing explanations can be separated.

Human recurrence can involve beliefs, expectations, physiological state, avoidance, social reinforcement, changed circumstances and renewed exposure to the original contingency. The adverse event may also produce new learning, occur in another context or coincide with the passage of time.

A reminder is therefore not automatically a reinstating event. Nor is every stressful experience. "Being triggered" is not a technical synonym for reinstatement.

To identify reinstatement experimentally, the relevant acquisition, extinction, outcome-exposure and test sequence must be established. Ordinary life rarely allows those elements to be isolated with the same precision.

The concept should guide interpretation rather than diagnose why a particular person's fear, urge, habit or symptom returned.

What reinstatement reveals

Change is influenced not only by what happens while a response is being reduced, but also by what happens afterward.

A response can decline through extinction and later return after a relevant outcome is encountered. The cue does not need to be directly retrained for this to occur. The event can add new learning and alter the context, expectations or retrieval conditions under which the cue is later experienced.

Reinstatement is possible, not inevitable. It does not mean that extinction was unreal, that the previous state has been completely restored or that every later recurrence has the same cause.

It reveals something more precise:

Later events can add new learning and alter which parts of an earlier learning history control behaviour—even when the original cue has not been directly retrained.