Extinction in Dog Behavior: Why Learned Behavior Disappears – and Returns

1. What Extinction Is

1.1 A Procedure, Not a Fading

Extinction is the procedure of no longer delivering the reinforcer that previously followed a behavior, with the result that the behavior declines in frequency, intensity, or duration.

It is not forgetting. Forgetting implies passive decay of a memory over time. Extinction is active learning: the dog acquires new information — that the previous contingency no longer holds. A dog that has stopped counter-surfing because the counter is now always empty has learned something new, not lost something old.

1.2 What the Dog Ends Up With

Bouton (2002) states the modern position: extinction does not destroy the original learning but produces new learning that is stored alongside it. The cue or action ends up with two available meanings, and which one governs behavior depends on what the current context retrieves — the same way an ambiguous word means different things in different settings.

This is why every subsequent section of this article exists. If extinction erased learning, relapse would need a separate explanation. Because extinction competes rather than deletes, relapse is the expected default whenever retrieval conditions favour the original memory (the same competitive-memory logic governs fear memories).

2. Inhibitory Learning, Not Erasure

2.1 What the Neuroscience Says

The circuit-level account is consistent and well characterized — in rodents. Extinction learning is associated with prefrontal regions, particularly the infralimbic cortex, which is thought to inhibit amygdala output rather than remove the original association. Extinction consolidation depends on NMDA receptor activation, and blocking those receptors prevents extinction while enhancing them pharmacologically can accelerate it.

2.2 What That Means for Dogs

None of this has been measured in dogs. There is no canine work on infralimbic function during extinction, no receptor pharmacology of canine extinction learning, and no imaging of a dog forming an inhibitory memory. The mammalian conservation argument makes the extension reasonable; it does not make it evidence (the same caveat applies to frontal control claims generally).

What is worth taking from the neural literature is the qualitative point rather than the anatomy: extinction is an active, biologically costly learning process, not the absence of one. That has a practical consequence — anything that impairs learning impairs extinction specifically.

2.3 The Prediction-Error Link

Extinction is driven by negative prediction error: the reward fails to arrive when it was expected, and the value of the preceding cue or action is revised downward (Schultz, Dayan & Montague, 1997), a dip below baseline in the dopaminergic teaching signal (the neurochemistry of that signal). This is also why extinction is only effective when the dog actually notices the change — an animal that never expected reinforcement on this trial generates no error to learn from (the prediction-error mechanism in detail).

3. Why Context Decides Everything

Original conditioning generalizes across settings reasonably well. Extinction does not. This asymmetry is the defining property of extinction learning rather than an incidental complication (Bouton, 2002), and it is the single most useful thing a trainer can know about the process.

The canine demonstration is unusually clean. Hall (2017) reviews evidence that professionally trained explosives detection dogs whose searching went unreinforced along one route showed rapidly declining search performance on that route — while performance on a different route, where targets were still being found, remained intact. Extinction attached itself to the specific context in which it occurred and did not spread.

In everyday terms: a dog that has learned the doorbell no longer predicts anything in your hallway has learned that about your hallway. The inhibitory memory does not travel on its own (which is the same transfer problem that affects trained behavior generally).

4. Extinction in Operant and Classical Learning

4.1 Operant Extinction

Withholding the reinforcer that previously followed a behavior. The dog jumps and receives no attention of any kind — no eye contact, no verbal response, no touch — and jumping declines. The dog surfs the counter and finds nothing, because the counter is now reliably empty.

Two side effects are commonly reported. Extinction-induced variability is the most useful: the dog tries other behaviors in an attempt to produce the old outcome, which creates opportunities to capture and reinforce an alternative. Extinction-induced frustration is the one to plan for, and it is more likely when reinforcement stops abruptly after a history of continuous reinforcement.

4.2 Classical Extinction

Presenting the conditioned stimulus without the unconditioned stimulus until the conditioned response declines. The clicker sounds and no food follows; the anticipatory response fades. A person who previously predicted something unpleasant appears repeatedly without anything unpleasant happening, and the fear response declines.

The difference matters clinically. A classically conditioned response is not a voluntary behavior but an emotional and physiological reaction. Extinction of fear is slower, more context-bound, and considerably more prone to renewal than extinction of an operant response (which is why reactivity needs the emotional layer addressed).

4.3 Why They Cannot Be Separated in Practice

The two run together. Ignoring a jumping dog is an operant procedure, but it also changes what visitors predict, which is classical. Reducing fear of a trigger through classical procedures reduces the operant avoidance that fear was driving.

The failure mode is specific and common: extinguishing an overt behavior without touching the emotional conditioning underneath produces a dog that complies and is still afraid.

5. The Four Routes Back

Bouton (2002) treats all four as expressions of the same thing — retrieval competition between an intact excitatory memory and a context-bound inhibitory one — rather than as four separate mechanisms.

5.1 Spontaneous Recovery

The extinguished behavior reappears after a period without training, with no new reinforcement. The usual account is that the inhibitory memory becomes less accessible with elapsed time, letting the original briefly dominate. Longer intervals make recovery more likely, and successive recoveries are typically weaker and extinguish faster — a pattern worth recording rather than judging by impression (which is what operationalizing behavior is for).

Practically: expect it, do not punish it, and simply run the procedure again.

5.2 Renewal

The behavior returns because the context changed. The standard designations are worth stating correctly, because they are frequently garbled.

ABA renewal — learning in context A, extinction in context B, test back in A. Behavior returns. This is the most robust form.

ABC renewal — learning in A, extinction in B, test in a novel context C. Behavior returns, generally somewhat less strongly.

AAB renewal — learning and extinction both in A, test in a different context B. Behavior returns. This case is the important one for trainers, because it shows that simply leaving the context where extinction happened is enough to bring the behavior back — even when acquisition and extinction occurred in the same place.

5.3 Resurgence

The original extinguished behavior returns when a newer, reinforced replacement behavior is itself no longer reinforced. A dog taught to settle on a mat instead of counter-surfing will tend to return to counter-surfing if the mat behavior stops paying. This is the strongest argument for maintaining the replacement behavior indefinitely rather than assuming it has taken over.

5.4 Reinstatement

A single unpaired encounter with the original reinforcer restores the extinguished response. One guest who greets the jumping dog, one successful counter-surf after a visitor left food out, and the behavior is back — not because the dog has "learned that persistence works," but because the reinforcer itself has re-established the expectation.

6. Bursts, Frustration, and What Actually Happens

6.1 Extinction Bursts Are Less Universal Than Assumed

The initial increase in behavior when reinforcement stops is usually presented as inevitable. It is not. Lerman and Iwata (1995) analysed 113 sets of extinction data and found bursting in 24% of cases — and, critically, in 36% of cases where extinction was used alone against only 12% where it was combined with other procedures such as reinforcing an alternative behavior.

That dataset comes from applied behavior analysis with human participants, so the percentages should not be transferred to dogs literally. The direction is what matters: bursts are a minority outcome, and giving the animal an alternative that still pays roughly cuts their frequency.

6.2 The Emotional Cost

Unexpected non-reinforcement produces frustration. Dogs with low frustration tolerance may escalate, redirect, or disengage, and for these individuals extinction used alone is a poor tool regardless of how theoretically sound it is (the neurobiology of canine frustration in detail), and frustration tolerance itself varies with temperament (as the coping-style literature describes).

This is a welfare consideration and not only a technical one. A procedure that works but leaves the dog reliably frustrated has costs that belong in the decision.

6.3 Stress Impairs Extinction Specifically

Because extinction is new learning, anything that degrades learning degrades extinction. Chronic stress impairs prefrontal function in rodents, primates, and humans (Arnsten, 2009) — extrapolated to dogs, not measured in them — and the behavioral corollary is well supported: a dog carrying a high baseline stress load is a poor candidate for a procedure that depends on acquiring new inhibitory learning. Address the stress first (the full account of chronic stress in dogs).

7. What Makes Extinction Work — and When Not to Use It

7.1 The Reinforcement History Beforehand

Behavior maintained on an intermittent schedule is more resistant to extinction than behavior maintained on continuous reinforcement. This is the partial reinforcement extinction effect, and it is one of the most reliable findings in the operant literature.

The common training conclusion drawn from it — therefore train on variable schedules from the start — does not follow, and has been tested in dogs. Cimarelli et al. (2021) clicker-trained naïve dogs with either continuous reinforcement or reinforcement on 60% of clicks. Partial rewarding did not improve learning speed, and the partially rewarded dogs subsequently showed a more pessimistic bias in a cognitive bias test. Build with continuous reinforcement; introduce variability once the behavior is fluent (schedules and their effects in full).

7.2 Expectancy Violation

Craske et al. (2014), working on human exposure therapy, argue that the effective ingredient is not the disappearance of the response but the strength of the mismatch between what was expected and what occurred. Extinction trials work best when the change in contingency is unmistakable to the learner.

Applied to dogs, this means partial, hesitant, or inconsistent extinction is worse than no extinction: it teaches that reinforcement is sometimes absent, which is precisely the schedule that makes behavior harder to extinguish later. The framework is human and clinical, and its transfer to dog training is an extension rather than a finding.

7.3 Differential Reinforcement Is Not Optional

Combining extinction with reinforcement of an alternative, incompatible behavior does three things at once: it reduces the burst rate (Lerman & Iwata, 1995), it lowers frustration by giving the dog a route that still pays, and it builds the behavior you actually want rather than merely subtracting one you do not.

The caveat from §5.3 applies: keep reinforcing the replacement, at least intermittently, indefinitely.

7.4 Context Work — With a Caveat

The standard advice is to run extinction across multiple contexts to reduce renewal. It is reasonable and follows directly from the context-dependence of extinction. It is also less certain than usually presented: Bouton and colleagues have reported that extinction in multiple contexts does not necessarily reduce vulnerability to relapse, and the conditions under which it helps are not fully settled.

Practical reading: vary contexts, and do not treat having done so as proof against renewal. Plan for relapse either way.

7.5 When Extinction Is the Wrong Tool

Fear-based behavior, including fear-based aggression. Withholding reinforcement does not address the underlying emotional state and may worsen it. The indicated approach is systematic desensitization and counterconditioning, which has controlled canine support (Butler, Sargisson & Elliffe, 2011; Shnookal et al., 2024) and treats the anxiety rather than its outward expression (the neurobiology of canine anxiety) (and the fallout of getting this wrong is well documented).

Separation-related behavior. Howling and destruction on departure are typically panic-driven rather than maintained by owner attention. Ignoring the dog on return does not treat panic (the neurobiology of separation-related distress).

Compulsive and self-reinforcing behavior. Tail chasing, flank sucking, chasing shadows, barking at echoes — where the behavior generates its own consequence, there is no external reinforcer to withhold. Hall (2017) notes that the reinforcement contingencies maintaining canine stereotypies are frequently not the ones owners assume.

Anything where the burst would be dangerous. If escalation carries bite risk, extinction alone is not an acceptable plan.

8. Summary: Extinction and Relapse at a Glance

Operant extinction — Procedure: withhold the reinforcer that followed the behavior. Result: behavior declines; variability increases; frustration possible. Outcome: not erasure — a competing inhibitory memory forms.

Classical extinction — Procedure: present the conditioned stimulus without the unconditioned stimulus. Result: conditioned emotional response declines gradually. Outcome: original association intact; strongly context-bound; renewal likely.

Spontaneous recovery — Trigger: elapsed time since the last extinction session. Mechanism: the inhibitory memory becomes less accessible. Response: repeat the procedure; each recovery is typically weaker.

Renewal — Trigger: change of context, including leaving the context where extinction occurred (AAB). Mechanism: extinction is context-bound, the original association is not. Response: vary contexts, and still expect relapse.

Resurgence — Trigger: the replacement behavior stops being reinforced. Mechanism: removing the competitor lets the original re-emerge. Response: keep paying the replacement indefinitely.

Reinstatement — Trigger: a single unpaired encounter with the original reinforcer. Mechanism: the reinforcer restores the expectation. Response: manage the environment; brief visitors and family explicitly.

9. Research Gaps and Critical Appraisal

The neural account is entirely extrapolated. Infralimbic inhibition, NMDA-dependent consolidation, and the pharmacology of extinction enhancement are rodent findings. Nothing comparable exists in dogs.

The behavioral phenomena are better established than the canine specifics. Spontaneous recovery, renewal, resurgence, and reinstatement are robust across species, but most parameter estimates — how long until recovery, how many contexts are enough — come from rodents, pigeons, and human clinical samples.

There is a real canine literature, and it is small. Hall (2017) is the substantive review of persistence and resistance to extinction in dogs, covering behavioral momentum, detection-dog applications, and breed differences (which fits the wider picture on breed and behavior).

Extinction bursts are the exception. They appeared in 24% of analysed cases, and in 12% when combined with other procedures (Lerman & Iwata, 1995) — human applied behavior analysis, not canine.

Multiple-context training is not a guaranteed fix. Bouton and colleagues have reported that extinction across several contexts does not necessarily reduce relapse vulnerability, which qualifies advice that is usually given without hedging.

The inhibitory learning model is a human clinical framework. Craske et al. (2014) address exposure therapy for human anxiety disorders. Its principles map plausibly onto dog training; that mapping has not been tested in dogs.

Individual variation is substantial. Frustration tolerance, persistence, and baseline arousal differ enough between dogs that the same procedure produces markedly different results (as arousal research would predict).

10. Conclusion

Extinction is not forgetting and not unlearning. It is the addition of a second memory that competes with the first for retrieval, and that competition is decided largely by context — which is why behavior returns after a pause, in a new place, when a replacement stops paying, or after a single accidental reward. Read that way, relapse stops being evidence of failure and becomes a predictable event with a known response: repeat the procedure calmly, keep the alternative behavior reinforced, prevent accidental reinforcement, and expect each recovery to be weaker than the last. The limits are worth holding alongside the framework. The circuit-level story is borrowed from rodents, the parameter estimates come mostly from other species, and the canine literature is genuinely thin. But the core practical implications do not depend on the neuroscience: build behavior with continuous reinforcement before thinning it, never use extinction alone on fear, always pair it with something that still pays, and treat durable change as something maintained rather than achieved.

Key Insights (Takeaways)

• Extinction does not erase the original learning. It creates a second, context-bound inhibitory memory stored alongside the first, leaving the cue with two available meanings whose expression depends on what the current context retrieves (Bouton, 2002).

• Context-dependence is the defining property, not a detail. Detection dogs whose searching went unreinforced on one route lost performance on that route while remaining accurate on another (Hall, 2017) — and behavior returns simply on leaving the context where extinction happened, which is what AAB renewal describes.

• Extinction bursts are a minority outcome. They occurred in 24% of analysed cases, and in only 12% when extinction was combined with reinforcing an alternative behavior (Lerman & Iwata, 1995). Differential reinforcement is the practical lever, not a nicety.

• Do not thin reinforcement early to build durability. Intermittent schedules do produce greater resistance to extinction, but in naïve dogs partial rewarding neither sped learning nor left affect untouched (Cimarelli et al., 2021). Build continuously, vary later.

• Extinction is the wrong tool for fear, panic, and self-reinforcing behavior. Fear-based aggression and separation-related distress need desensitization and counterconditioning (Butler et al., 2011; Shnookal et al., 2024); compulsive and self-reinforcing behavior offers no external reinforcer to withhold.

References

Arnsten, A. F. T. (2009). Stress signalling pathways that impair prefrontal cortex structure and function. Nature Reviews Neuroscience, 10(6), 410–422. https://doi.org/10.1038/nrn2648

Bouton, M. E. (2002). Context, ambiguity, and unlearning: Sources of relapse after behavioral extinction. Biological Psychiatry, 52(10), 976–986. https://doi.org/10.1016/S0006-3223(02)01546-9

Bouton, M. E. (2004). Context and behavioral processes in extinction. Learning & Memory, 11(5), 485–494.

Butler, R., Sargisson, R. J., & Elliffe, D. (2011). The efficacy of systematic desensitization for treating the separation-related problem behaviour of domestic dogs. Applied Animal Behaviour Science, 129(2–4), 136–145. https://doi.org/10.1016/j.applanim.2010.11.001

Cimarelli, G., Schoesswender, J., Vitiello, R., Huber, L., & Virányi, Z. (2021). Partial rewarding during clicker training does not improve naïve dogs' learning speed and induces a pessimistic-like affective state. Animal Cognition, 24(1), 107–119. https://doi.org/10.1007/s10071-020-01425-9

Craske, M. G., Treanor, M., Conway, C. C., Zbozinek, T., & Vervliet, B. (2014). Maximizing exposure therapy: An inhibitory learning approach. Behaviour Research and Therapy, 58, 10–23. https://doi.org/10.1016/j.brat.2014.04.006

Hall, N. J. (2017). Persistence and resistance to extinction in the domestic dog: Basic research and applications to canine training. Behavioural Processes, 141(Pt 1), 67–74. https://doi.org/10.1016/j.beproc.2017.04.001

Lerman, D. C., & Iwata, B. A. (1995). Prevalence of the extinction burst and its attenuation during treatment. Journal of Applied Behavior Analysis, 28(1), 93–94. https://doi.org/10.1901/jaba.1995.28-93

Pavlov, I. P. (1927). Conditioned reflexes: An investigation of the physiological activity of the cerebral cortex (G. V. Anrep, Trans.). Oxford University Press.

Rescorla, R. A. (2001). Experimental extinction. In R. R. Mowrer & S. B. Klein (Eds.), Handbook of contemporary learning theories (pp. 119–154). Lawrence Erlbaum.

Schultz, W., Dayan, P., & Montague, P. R. (1997). A neural substrate of prediction and reward. Science, 275(5306), 1593–1599. https://doi.org/10.1126/science.275.5306.1593

Shnookal, J., Tepper, D., Howell, T., & Bennett, P. (2024). Counterconditioning-based interventions for companion dog behavioural modification: A systematic review. Applied Animal Behaviour Science, 276, 106305. https://doi.org/10.1016/j.applanim.2024.106305

Skinner, B. F. (1938). The behavior of organisms: An experimental analysis. Appleton-Century.