Social Learning in Dogs: Observation and Demonstration

1. What Social Learning Is

1.1 The Definition

Social learning is any process in which the behavior, presence, or products of one individual influence what another acquires or expresses. Its defining feature is that the observer need not interact with the environment directly: the demonstrator's actions, or their consequences, carry the information.

The advantage is obvious. Social learning is faster than individual trial and error and avoids the costs of discovering by experiment which options are dangerous. Bandura's (1977) social learning theory established the framework in human psychology; the comparative literature has since spent decades establishing that the label covers several distinct mechanisms.

1.2 Why Dogs Are an Interesting Case

Dogs have spent their evolutionary history in proximity to humans, and their sensitivity to human social signals is unusual among non-primates. That makes them a natural test case for whether selection on social attentiveness also amplifies learning from observation — and it makes them a species where the temptation to over-read behavior as sophisticated cognition is correspondingly high (as canine cognition research generally illustrates).

1.3 Why the Term Covers Too Much

Social learning ranges from a dog noticing that a place is worth investigating because another dog was there, to reproducing a specific novel action after seeing it once. Those are different capacities with different evidential requirements, and one term covers both.

The hierarchy in the next chapter exists precisely to separate them, and most disputes in this area come from a claim at one level being defended with evidence from another (which is what operationalizing a behavioral claim is for). A headline reporting that dogs imitate is frequently built on a study demonstrating something several steps below imitation.

1.4 Why It Matters Practically

If a dog is learning by local enhancement, then another dog's presence draws attention to a location and nothing more is transferred. If it is imitating, then a handler could teach through demonstration.

The two prescribe entirely different uses of a demonstration dog, and the distinction is invisible from the outcome alone.

2. The Mechanism Hierarchy

These are ordered by cognitive demand, and the distinctions do real work: the same experimental result can support very different conclusions depending on which mechanism is assumed.

Local enhancement. The demonstrator's presence or activity draws the observer's attention to a place. The observer learns where to look, not what to do.

Stimulus enhancement. The demonstrator's interaction with an object increases the observer's attention to that object. The observer learns what is worth investigating, not how.

Social facilitation. A behavior already in the observer's repertoire increases in frequency or intensity because a conspecific is performing it. No new behavior is acquired.

Emulation. The observer learns about the outcome another's actions produce — that pushing the box reveals food — without copying the technique. More demanding than enhancement, less than imitation, and closely tied to how flexibly the animal can recombine what it knows (the flexibility literature).

Imitation. The observer reproduces the demonstrator's specific actions, including motor patterns not previously in its repertoire, guided by having watched. Whiten and Ham's (1992) standard is the useful one: imitation requires reproducing the topography of the action, not just its result.

The practical stakes are real. If dogs mostly emulate, demonstrations should maximize the salience of the outcome. If they imitate, the precision of the demonstrator's movements is what matters (and separating the two is a measurement problem before it is a theoretical one).

2.1 Why the Hierarchy Is the Most Useful Part

Ordered from least to most demanding — local enhancement, stimulus enhancement, emulation, imitation — the hierarchy provides the standard test for any social learning claim: which is the least demanding mechanism that would produce this observation?

That question does most of the work in evaluating the studies below, and it is the question popular accounts skip (where causal reasoning is examined at length). Applying it is not a way of dismissing findings; it is how the field decides what a finding shows.

2.2 Why the Simplest Account Wins by Default

Where two mechanisms predict the same behavior, the less demanding one is preferred until something distinguishes them. That is a working rule rather than a claim about what is true, and it cuts against imitation in most designs.

It also cuts the other way. Preferring the simpler account is not the same as having shown that the richer one is false, and the field has a long history of treating the first as the second. Both errors are available and only one of them gets called credulous.

3. Learning from Conspecifics

3.1 The Detour Paradigm

The most productive line of dog-to-dog work came from Pongrácz and colleagues using a detour task: reaching food behind a V-shaped fence. Pongrácz et al. (2001) established that human demonstration significantly improved dogs' detour performance, and Pongrácz et al. (2003) extended the paradigm to ask what dogs preferentially copy, finding that they favor unambiguous demonstrations.

What this establishes with reasonable confidence is that observing a demonstrator complete a goal-directed task improves subsequent performance. What it does not cleanly establish is how much action-level information transfers. Detour tasks have been criticized in comparative cognition precisely because ruling out enhancement mechanisms is difficult — a directional cue plus a motivational boost can produce the same result as route learning.

3.2 The Two-Action Test

The stronger design manipulates how the demonstrator solves the task. Range, Virányi and Huber (2007) had dogs observe a trained conspecific pull a handle using either paw or mouth. Observer dogs showed a preference for the demonstrator's effector.

This effector-matching result is the most frequently cited canine evidence for action-level copying. It is also interpretable as a sophisticated form of body enhancement — the observer's attention drawn to a body part rather than a motor program being copied. The distinction has not been resolved, and any design permitting imitation also permits emulation.

3.3 What Rank Does

Social learning between dogs is not rank-neutral. Pongrácz and colleagues have reported that dominance status affects both individual and social learning performance in dogs, with higher-ranking animals learning better from a human demonstrator in a two-action test. This is a genuine effect worth knowing about — and worth keeping separate from the popular dominance model, which it does not support (as the dominance literature makes clear).

3.4 What the Detour Paradigm Can and Cannot Show

A dog watching another dog solve a detour and then solving it faster has demonstrably learned something, and the paradigm is among the oldest in the canine literature. What it learned is the open question: the route, that the fence can be circumvented, or simply that the far end of the fence is worth approaching.

The design that distinguishes these varies the demonstrated route and checks whether the observer copies the specific path rather than the general solution (because transfer across contexts is not automatic). Where a dog takes the shorter route regardless of which was demonstrated, it learned about the fence rather than about the demonstration.

3.5 Why the Two-Action Test Is the Stronger Design

Presenting two equally effective ways to achieve the same outcome, and checking which one the observer uses, isolates copying of the action from learning about the outcome. That is what makes it the standard test for imitation across species.

Where a study uses it and finds action-matching, the claim is much harder to reduce to a simpler mechanism than where it does not. Much of the disagreement about imitation in animals reduces to whether a given study used this design or something weaker.

4. Learning from Humans

4.1 Goal Sensitivity

Miklósi et al. (1998) showed that dogs observing a human manipulate a rod attached to a food container were more likely to solve the task than controls, and that their responses were sensitive to the direction and apparent goal of the demonstrator's action rather than to the object alone. Attending to what the demonstrator appears to be trying to do is a meaningfully different thing from attending to what the demonstrator touches (and sits alongside other higher-order canine capacities).

4.2 Ostensive Cues

Range et al. (2009) found stronger social learning when demonstrations were accompanied by direct social address — eye contact, verbal cues, object-directed gesture — than when physically identical actions were performed without them. Whether this reflects selective attention, differential encoding, or a learned rule about when human behavior is informative has not been determined.

The applied implication is nonetheless clear: a demonstration performed at the dog is processed differently from one performed near it.

4.3 The Wolf Comparison

Miklósi et al. (2003) compared dogs and human-socialized wolves on using human pointing to locate hidden food. Dogs outperformed wolves, and extensively socialized wolves still lagged behind pet dogs with no special training — a result interpreted as evidence for an evolutionary rather than purely experiential component.

The interpretation is influential and contested. Rearing history and social experience are difficult to equate fully across species even under controlled conditions, and the comparison rests on that equivalence (the domestication debate in more detail).

4.4 Social Referencing

Merola, Prato-Previde and Marshall-Pescini (2012) found that dogs facing a novel ambiguous object looked toward the owner's face before approaching or avoiding, and adjusted behavior according to the owner's expressed emotional valence.

This matters for how demonstration works in practice. Dogs do not observe neutrally; they read the social and emotional framing of the demonstration as part of the event (which connects directly to emotional contagion).

4.5 Social Referencing Is a Different Capacity

Looking at a person when something ambiguous appears, and adjusting behavior according to that person's reaction, is not learning an action from a demonstration. It is using another individual as a source of information about a situation (Merola, Prato-Previde & Marshall-Pescini, 2012).

Both are social learning in the broad sense and they call for different training applications: one supports demonstration, the other supports the handler's own reaction being worth managing (where arousal is treated as its own variable). The second is the more practically useful of the two, because it applies in every ambiguous situation rather than only where something is being taught.

4.6 What the Wolf Comparison Establishes

Comparisons between dogs and wolves on human-directed tasks are informative about defaults rather than capacities, and they carry the confounds that attach to all such work: research wolves are few, intensively hand-raised and repeatedly tested, while pet dogs are numerous and naive.

Whichever direction a comparison comes out, testing experience differs systematically between the groups (with individual variation doing more work than group means suggest). The comparisons remain worth making; what they measure is not purely a species difference.

5. The Ostensive-Cue Question

5.1 What the Claim Says

Ostensive cues — eye contact, the dog's name, a raised tone directed at it — are held to signal that what follows is meant for the dog, and to improve what it takes from a demonstration. Dogs have been reported to be more likely to copy a demonstration preceded by such cues (Range et al., 2009).

The claim matters because it treats the dog as receiving communication rather than merely observing an event, which is a considerably stronger reading of what social learning is (where gesture comprehension is treated in detail).

5.2 The Large Test

A related prediction was tested at scale in a different paradigm. Across 20 research sites, data from 455 dogs compared pointing preceded by eye contact and the dog's name against pointing preceded by a non-social auditory cue (ManyDogs Project et al., 2023).

Both conditions were significantly above chance and there was no significant difference between them, consistently across sites (ManyDogs Project et al., 2023). The consistency across sites is the part that matters: this was not one laboratory failing to find an effect.

5.3 How the Two Results Relate

They are not the same experiment. One concerns copying a demonstrated action, the other following a point, and a dog could plausibly use ostensive framing for one and not the other.

What the large study does is remove the assumption that ostensive cues generally enhance what a dog takes from a human. That assumption underwrites a good deal of training advice, and it now needs to be defended for each paradigm rather than treated as established (which is what operationalizing a behavioral claim is for).

5.4 Why a Null Result Is Not Nothing

A null difference does not show that the conditions are equivalent to the dog. The manipulation may simply have been too weak against a lifetime of everyday human contact, in which almost everything a person does is already directed at the dog. A laboratory contrast between two ways of getting a dog's attention may be smaller than the contrast the dog experiences at home every day.

What it does is shift the burden. A claim supported by one laboratory and unsupported by twenty is not refuted, and it is no longer something a reader should accept without asking which study produced it.

5.5 What This Leaves Standing

That dogs learn from watching people is not in question, and nothing here touches the demonstration findings themselves. What is in question is the further claim that the social framing of a demonstration is doing specific work, over and above the demonstration itself.

For practice the difference is small: a handler who makes contact before demonstrating loses nothing. For the theoretical account of dogs as receivers of human communication, it matters a great deal, and that account is what most popular writing on the topic is built from.

6. The "Do as I Do" Paradigm

6.1 The Design

The Do as I Do method, systematized by Claudia Fugazza, teaches a dog a generalized rule linking a cue to reproducing whatever the human has just done. Once the rule is established, the dog can be tested on novel actions it has never been trained to perform.

The logic is that a generalized imitation rule should let the dog reproduce a new action immediately, which neither local nor stimulus enhancement can explain (they carry no action-specific information) and which operant shaping cannot explain (it requires a reinforcement history for each behavior).

6.2 Deferred Imitation

Fugazza and Miklósi (2014) tested whether dogs could reproduce demonstrated actions after a delay. Dogs performed above chance on familiar actions across the intervals tested and showed some capacity for deferred reproduction of novel actions at shorter delays, with novel-action performance declining more steeply as delay increased.

6.3 Recall After Incidental Encoding

Fugazza, Pogány and Miklósi (2016) tested 17 dogs trained on Do as I Do in an unexpected recall test — dogs had been retrained to simply lie down after watching an action, so they had no reason to anticipate being asked to reproduce it. When unexpectedly cued to imitate, they could, with performance declining as the delay increased from one minute to one hour.

The authors interpret this as episodic-like memory: incidental encoding of a specific observed event, spontaneously retrieved. The interpretation remains debated against stricter criteria for what-where-when memory, and the result applies to dogs already trained on the protocol. What it supports more modestly is well worth having: dogs retain information about observed actions beyond the moment of observation (and sleep is part of how that consolidates).

6.4 The Critiques

The training history problem. Establishing the rule involves extensive reinforcement of copy-like behavior, which makes it hard to exclude sophisticated reinforcement-based generalization rather than imitation in a demanding sense.

Action selection. The actions used are ones dogs can physically perform and observers can reliably code, which may constrain generalizability.

Replication. Cross-laboratory replication has been limited, partly because the protocol requires substantial training investment. Whether the capacity is species-typical or a trainable-but-not-universal skill is open.

For applied purposes the philosophical question matters less than the practical finding: with training, dogs reliably reproduce human-demonstrated actions.

6.5 What the Method Requires

Do as I Do depends on the dog first learning a general rule — that a cue means reproduce what was just demonstrated — which takes substantial prior training. That training is part of the finding rather than a preliminary to it.

A dog that has been taught the rule and then copies a novel action has demonstrated something real, and the achievement belongs to the dog rather than to the training alone. Whether an untrained dog would do the same is a separate question the paradigm does not address (which is what operationalizing a behavioral claim is for). It is also the question most readers have when they encounter the finding.

6.6 The Deferred and Incidental Results Are the Strongest

Of the findings in this chapter, two are hardest to reduce to simpler mechanisms. Copying after a delay requires something retained rather than immediately mirrored (Fugazza & Miklósi, 2014), and recall of an action the dog was not trying to memorize removes the possibility that it was practising for a test (Fugazza, Pogány & Miklósi, 2016).

Both come from the same research group, which is the limitation running through this chapter rather than a criticism of the designs. The designs are careful; what is missing is anyone else running them.

7. What Modulates Social Learning

7.1 Developmental Stage

Early social experience shapes the conditions under which later social learning occurs — this is well established for socialization generally (during the sensitive period). Whether there is a specific sensitive window for observational learning as a mechanism has not been demonstrated with comparable rigour, and claims that a narrow age range optimizes social learning capacity should be treated cautiously. Adult dogs remain substantial social learners.

7.2 Breed

Breeds developed for close human cooperation tend to perform better on human-directed social learning tasks than those selected for independent work. The pattern fits the domestication account. Within-breed variation is large enough that breed-level generalizations map poorly onto individuals (as breed-behavior research consistently finds), and individual temperament predicts engagement better than breed label (the temperament and coping-style evidence).

7.3 Emotional State

A fearful dog may fail to engage with a demonstrator not through cognitive incapacity but because defensive processing has taken the attentional resources (as is routinely the case in reactive dogs). A highly aroused dog may orient briefly and encode nothing. Keeping the dog below its individual threshold is a precondition for demonstration-based work, not a refinement of it (the arousal constraint in full).

7.4 Relationship and Familiarity

Dogs orient more readily and persistently to familiar partners than to strangers. Whether attachment quality modulates observational learning independently of general familiarity effects has not been resolved (the attachment framework in detail).

7.5 Task Complexity

For simple tasks, observation provides a robust advantage. For multi-step tasks the advantage narrows: dogs may acquire the goal through observation but still need individual reinforcement to build the sequence. Demonstration is most useful for directing attention and establishing a first approximation, which reinforcement then refines (through the ordinary prediction-error machinery).

7.6 Why Modulators Matter More Than the Headline

The variables in this chapter — age, breed, emotional state, relationship, task difficulty — are usually reported as qualifications to the main finding. In practice they are the finding, because they determine whether social learning is available in a given situation at all.

A dog that is aroused, unfamiliar with the demonstrator or faced with a task beyond its level will learn nothing from a demonstration however good the demonstration is (where arousal is treated as its own variable). Checking those conditions before attributing a failure to the dog's aptitude is the practical use of this chapter.

7.7 Breed Effects Are Smaller Than Assumed

Breed differences appear in this literature and are frequently overread. Variation within a breed is large, the samples in these studies are small, and what is measured is performance on one task rather than a general aptitude for social learning (with breed explaining less than is usually assumed).

A dog from a breed with no reputation for biddability may learn from a demonstration perfectly well, and the reverse. Testing the individual dog takes one session and settles the question for that animal.

8. Practical Application

8.1 Demonstration Dogs

Using a trained dog to accelerate acquisition in a novice is long-standing practice and rarely evaluated systematically. The conditions that the evidence supports: the demonstrator should be well trained, calm, and genuinely reinforced rather than performing under pressure; the observer should be below arousal threshold and positioned to see clearly.

8.2 Do as I Do as a Training Tool

The advantage is speed for complex physical actions that are awkward to capture or shape. The cost is the upfront investment in establishing the rule. For most applications it complements conventional reinforcement rather than replacing it.

8.3 Working Dog Programs

Observational components are most useful during initial acquisition — establishing what category of outcome the dog is working toward — and less so during deployment, where independent action is required. Early observational learning followed by individual reinforcement for robust independent performance is the sequence most consistent with the evidence.

8.4 Group Housing and Welfare

In shelters and multi-dog households, conspecifics are a continuous source of social information. Distress can spread through social transmission of affective state — not imitation in the technical sense, but consequential. Calm demonstrator dogs are used in some enrichment programs on the same logic in reverse, with the obvious caveat that an already-distressed observer exposed to a distressed demonstrator gets worse rather than better (and training method shapes the whole picture).

8.5 What a Demonstration Dog Can Realistically Do

The defensible use is drawing attention: another dog approaching an object, entering a space or interacting with an apparatus makes the observing dog more likely to do the same. That is local or stimulus enhancement and it is genuinely useful.

What a demonstration dog cannot reliably do is teach a specific action, and building a class around the assumption that it can produces disappointment for everyone involved. Attention-drawing is the modest version and it is the one that works.

8.6 Where Social Learning Works Against You

The same mechanisms transfer behavior nobody intended. A dog that watches another bark at the gate is more likely to approach the gate, and in a multi-dog household the effect compounds.

Managing what dogs observe each other doing is a legitimate part of a behavior plan and it is rarely discussed alongside the positive applications (because transfer across contexts is not automatic). Separating dogs during a trigger event is not only about the dog reacting; it is also about the one watching.

9. Where This Literature Comes From

9.1 A Concentrated Field

Almost every canine study in this article comes from two research groups. The detour paradigm, the rank work, the ostensive-cue and human-cue studies and the Do as I Do development originate largely in Budapest (Miklósi and colleagues; Pongrácz and colleagues; Fugazza and Miklósi), and the selective imitation and ostensive work in Vienna (Range, Virányi and Huber).

That is normal for a young field and it has consequences a reader should know about (with individual variation doing more work than group means suggest). It is also a reason the field advanced quickly, since two groups with a shared programme can build on each other without waiting for a wider consensus.

9.2 What Concentration Does

Two groups sharing methods, apparatus, populations and assumptions will produce results that agree with each other, and that agreement is not the same as independent replication. Where a systematic feature of the shared approach influences the outcome, it influences all of them together.

It also means the dogs are drawn from particular populations — Hungarian and Austrian pet dogs, often recruited repeatedly — and testing experience is itself a variable in cognitive work. A dog that has been through several studies is not a naive subject.

9.3 What the Multi-Site Study Adds

The value of a study run across 20 sites is precisely that it removes this confound (ManyDogs Project et al., 2023). Whatever it finds cannot be attributed to one team's procedure or population.

That is why its results carry disproportionate weight here, including the parts that are inconvenient for the established picture. Multi-site studies of this kind are new in canine science and there are very few of them.

9.4 The Theoretical Frameworks Are Older and Broader

The mechanism hierarchy that organizes this article comes from comparative work across the animal kingdom (Whiten & Ham, 1992), and the social learning framework itself from human psychology (Bandura, 1977).

Both are frameworks rather than findings, and neither was built on dogs. They earn their place by sorting the phenomena, not by supporting any particular claim about this species. Citing Bandura in a canine article establishes a vocabulary, not a result.

9.5 What Would Strengthen the Field

Direct replication of the imitation findings by an unconnected group, in a different population, using the published protocol. Nothing exotic, and it has not been done for most of the results described here. The multi-lab model that produced the pointing study shows it is organizationally possible.

Until it is, the honest description is that a small number of laboratories have produced a coherent and interesting body of work, and that its robustness outside those laboratories is largely untested (which is what operationalizing a behavioral claim is for). That is a description of a young field rather than an accusation, and the same was true of most areas covered in this collection twenty years ago.

10. Social Learning at a Glance

The mechanism hierarchy — Local enhancement (attention to a place); stimulus enhancement (attention to an object); social facilitation (more of an existing behavior); emulation (learning the achievable outcome); imitation (reproducing the action itself). Surface similarity between observer and demonstrator behavior does not identify which is operating.

Conspecific learning — Detour studies establish that observation improves performance (Pongrácz et al., 2001, 2003). The two-action test provides the strongest current case for action-level copying via effector matching (Range et al., 2007), which remains interpretable as body enhancement.

Human-directed learning — Dogs attend to the apparent goal rather than only the object (Miklósi et al., 1998), learn more from demonstrations framed as communication (Range et al., 2009), and use owners as information sources about ambiguous stimuli (Merola et al., 2012).

Do as I Do — The strongest experimental case for reproducing novel demonstrated actions (Fugazza & Miklósi, 2014), including recall after incidental encoding in 17 trained dogs (Fugazza et al., 2016). Results apply to dogs trained on the protocol and should not be generalized to untrained dogs observing casually.

Practical use — Demonstration and individual reinforcement are complementary. Demonstration supplies a template and a motivational scaffold; reinforcement refines and stabilizes it. Both require the dog below threshold and attending to the right thing.

11. Research Gaps and Critical Appraisal

The imitation–emulation distinction is unresolved. Any design permitting imitation also permits emulation, and the strongest current canine evidence — effector matching — has a plausible enhancement-based reading.

The evidence base is concentrated in one research network. This deserves stating plainly. The empirical core of canine social learning research comes from the comparative ethology tradition at Eötvös Loránd University and closely collaborating groups — Miklósi, Pongrácz, Range, Fugazza and colleagues. The work is rigorous and heavily cited. It also means that the more specific claims — effector matching, deferred imitation under Do as I Do, ostensive cue effects — have largely not been reproduced by groups with different theoretical commitments, subject populations, and laboratory cultures. Independent findings that exist are broadly consistent, but this is a structural limitation on generalizability rather than a minor caveat.

Replication is thin. Several key findings, including aspects of the Do as I Do results, lack independent cross-laboratory replication with full procedural transparency. Pre-registered replication would materially strengthen the field.

Sample sizes are small. Most studies use tens of dogs, frequently recruited through the same channels, in laboratory settings with constrained object-manipulation tasks.

Anthropomorphism is a live risk. Dogs' social attunement invites reading simpler mechanisms as sophisticated cognition. Parsimony — exhausting enhancement and emulation accounts before invoking imitation — is a methodological requirement, not conservatism.

The neurobiology is entirely unexplored. Evidence for mirror-neuron-like mechanisms in dogs is non-existent rather than weak: no study has identified neurons with that response profile in this species. The framework is worth investigating and should not be presented as an established component of canine cognition (as with canine neurobiology generally).

Laboratory findings may understate the phenomenon. Controlled paradigms constrain what dogs can display, and field observation often suggests richer social learning than laboratory tasks capture. Bridging the two remains an open methodological problem.

The literature comes from two research groups. Nearly all canine social learning studies cited here originate in Budapest or Vienna, so agreement between them reflects shared methodology as much as independent confirmation.

The ostensive-cue assumption has been challenged at scale. A multi-site study of 455 dogs found no difference between ostensive and non-ostensive framing in a pointing paradigm (ManyDogs Project et al., 2023), which does not settle the imitation case and does remove a general assumption.

Direct replication by unconnected groups is largely missing. Most of the imitation findings have not been reproduced outside the laboratories that produced them.

Prior training is part of the Do as I Do finding. The paradigm requires teaching a general copying rule first, and what an untrained dog would do is not addressed by it.

12. Conclusion

Dogs extract information from what conspecifics and humans do, and they do so through several mechanisms rather than one. At the lower end, a demonstrator directs attention toward a place or an object — useful, and cognitively cheap. At the upper end, and under specific training conditions, dogs reproduce actions they have never been reinforced for, retain them across delays, and recall them when unexpectedly asked. Between those poles sits most of everyday training, where demonstration establishes an approximation and reinforcement does the rest. The honest summary of the evidence is that the lower mechanisms are well established, the upper ones are supported by careful work whose central results have not been widely replicated outside the network that produced them, and the neural basis is unstudied. For practice, the durable conclusions do not depend on resolving the imitation debate: demonstrate with clear communicative framing, use a calm and genuinely reinforced model, keep the observer below threshold, and treat what the dog takes from the demonstration as a starting point to be refined rather than a finished behavior.

Key Insights (Takeaways)

• "Learning by watching" covers at least five mechanisms of very different cognitive demand, from attention being drawn to a location up to reproducing a novel action. Behavior that resembles the demonstrator's is not by itself evidence of imitation, and separating the mechanisms is a design problem that most paradigms do not fully solve.

• The strongest current canine evidence for action-level copying is effector matching in the two-action test (Range et al., 2007) — dogs preferring the body part the demonstrator used. It remains interpretable as attention drawn to a body part rather than a motor program being copied.

• Demonstrations framed as communication work better. Dogs learn more when the action is accompanied by eye contact and verbal address (Range et al., 2009), attend to the demonstrator's apparent goal rather than only the object (Miklósi et al., 1998), and consult the owner's expression about ambiguous stimuli (Merola et al., 2012).

• Do as I Do provides the strongest current case for imitation, among tested dogs trained on this protocol, and the narrowest claim. Seventeen dogs trained on the protocol recalled and reproduced human actions when unexpectedly asked, with memory decaying over an hour (Fugazza et al., 2016) — a result about protocol-trained dogs, not about dogs watching casually.

• The evidence base has a structural limitation worth knowing. Most canine social learning research comes from one collaborating research network, cross-laboratory replication is thin, and no neural mechanism has been identified in dogs at all. The practical recommendations do not depend on resolving any of this.

References

Bandura, A. (1977). Social learning theory. Prentice Hall.

Fugazza, C., & Miklósi, Á. (2014). Deferred imitation and declarative memory in domestic dogs. Animal Cognition, 17(2), 237–247. https://doi.org/10.1007/s10071-013-0656-5

Fugazza, C., Pogány, Á., & Miklósi, Á. (2016). Recall of others' actions after incidental encoding reveals episodic-like memory in dogs. Current Biology, 26(23), 3209–3213. https://doi.org/10.1016/j.cub.2016.09.057

ManyDogs Project, Espinosa, J., Stevens, J. R., et al. (2023). ManyDogs 1: A multi-lab replication study of dogs' pointing comprehension. Animal Behavior and Cognition, 10(3), 232–286. https://doi.org/10.26451/abc.10.03.03.2023

Merola, I., Prato-Previde, E., & Marshall-Pescini, S. (2012). Social referencing in dog–owner dyads? Animal Cognition, 15(2), 175–185. https://doi.org/10.1007/s10071-011-0443-0

Miklósi, Á., Kubinyi, E., Topál, J., Gácsi, M., Virányi, Z., & Csányi, V. (2003). A simple reason for a big difference: Wolves do not look back at humans, but dogs do. Current Biology, 13(9), 763–766. https://doi.org/10.1016/S0960-9822(03)00263-X

Miklósi, Á., Polgárdi, R., Topál, J., & Csányi, V. (1998). Use of experimenter-given cues in dogs. Animal Cognition, 1(2), 113–121. https://doi.org/10.1007/s100710050016

Pongrácz, P., Miklósi, Á., Kubinyi, E., Gurobi, K., Topál, J., & Csányi, V. (2001). Social learning in dogs: The effect of a human demonstrator on the performance of dogs in a detour task. Animal Behaviour, 62(6), 1109–1117. https://doi.org/10.1006/anbe.2001.1866

Pongrácz, P., Miklósi, Á., Timár-Geng, K., & Csányi, V. (2003). Preference for copying unambiguous demonstrations in dogs (Canis familiaris). Journal of Comparative Psychology, 117(3), 337–343. https://doi.org/10.1037/0735-7036.117.3.337

Pongrácz, P., Vida, V., Bánhegyi, P., & Miklósi, Á. (2008). How does dominance rank status affect individual and social learning performance in the dog (Canis familiaris)? Animal Cognition, 11(1), 75–82. https://doi.org/10.1007/s10071-007-0090-7

Range, F., Heucke, S. L., Gruber, C., Konz, A., Huber, L., & Virányi, Z. (2009). The effect of ostensive cues on dogs' performance in a manipulative social learning task. Applied Animal Behaviour Science, 120(3–4), 170–178. https://doi.org/10.1016/j.applanim.2009.05.012

Range, F., Virányi, Z., & Huber, L. (2007). Selective imitation in domestic dogs. Current Biology, 17(10), 868–872. https://doi.org/10.1016/j.cub.2007.04.026

Whiten, A., & Ham, R. (1992). On the nature and evolution of imitation in the animal kingdom: Reappraisal of a century of research. Advances in the Study of Behavior, 21, 239–283. https://doi.org/10.1016/S0065-3454(08)60146-1