Play Signals in Dogs: How Dogs Communicate "This Is Play"

1. What a Play Signal Is

1.1 The Problem Play Has to Solve

Social play between dogs borrows almost all of its movements from other contexts. A dog that chases another dog uses the motor patterns of pursuit. A dog that grabs the neck of its partner and shakes its head uses movements that belong to fighting and to killing prey. A dog that stands over another dog uses a posture that, in another setting, would be a threat. If a dog read these movements at face value, play would be impossible: every chase would be an attack, every bite a provocation.

That most of these interactions end without injury implies that something separates play from the real thing. The classic answer is that dogs use special signals to frame what follows as play. The interesting question is how much of this has been tested, and what the tests found.

This article is about communication. Why dogs play at all and what play may do for development are covered elsewhere (what is known about the neurobiology and functions of play). Here the question is what a dog does to tell another dog, or a person, that this is play, and whether the partner responds.

1.2 Metacommunication: A Message About the Message

The underlying idea is called metacommunication. A play signal does not carry a message of its own in the way a growl does. It is meant to change the meaning of the actions around it: the bite that follows a play signal is supposed to be read as a play bite. In the terms used in recent play research, the metacommunication hypothesis proposes that a signal such as the play bow serves to change the meaning of a subsequent pattern and to improve communication between players (Maglieri, Zanoli, Mastrandrea & Palagi, 2023).

The hypothesis makes a testable prediction. If a signal mainly reframes risky actions, it should appear close to them: just before or after a play bite or a pounce. If it appears mostly at other moments, for example during pauses, its main job may be something else. This prediction has been tested several times.

1.3 The Usual Inventory

Ethological descriptions of social play, going back to Bekoff and Byers (1981), use a recurring set of markers. In the operational definition used in our library, play is scored when at least two of them occur: the play bow (forelegs extended, elbows lowered, hindquarters raised), exaggerated bounding locomotion, self-handicapping (rolling over, allowing the partner on top), rapid role reversal within about three seconds, and a relaxed open mouth without facial tension (how dog behavior is defined and measured in research).

These markers fall into two groups. Some are discrete displays, such as the play bow. Others are properties of the whole interaction: looser and larger movements, switching roles, a stronger partner holding back. Most research on play signals examines the discrete displays; the structural features have been studied more as patterns than as signals.

1.4 Signals in Sequence

A single behavior is rarely informative on its own. As with so-called calming signals, a bow means little until it is read in the sequence of what came before, what follows and how the partner responds (why single signals have to be read in context). The play bow is one of the few canine displays for which context, receiver and outcome have been examined together.

2. The Play Bow: Three Hypotheses Tested

2.1 A Stereotyped Display

The play bow is the most recognizable play signal in dogs. The front of the body is lowered, the elbows are on or near the ground and the hindquarters stay raised. It appears in related canids as well. An early study measured the variability of the bow in infant coyotes, wolves, wolf–dog hybrids, beagles and adult free-ranging dogs and found that both the duration and the form of the bow showed marked stereotypy (Bekoff, 1977). The author argued that this uniformity had been favored by selection because it helps the bow communicate play intention.

A uniform form makes a display easy to recognize, but it does not show what the signal does. For that, one has to look at when it is used and what happens next.

2.2 Punctuation

The influential functional account came from Bekoff (1995). In a sample of ten adult dogs plus four infants each of dogs, wolves and coyotes, play bows occurred in association with behaviors that could be misinterpreted as aggressive, particularly bites accompanied by head shaking. Bekoff proposed that bows clarify playful intent, functioning like punctuation that marks how the surrounding actions are to be read (Bekoff, 1995; as summarized in Byosiere, Espinosa & Smuts, 2016a).

This is the metacommunication idea in its clearest form, and the account most training literature still repeats. The sample was small and mixed young animals of three species with adult dogs, and whether partners understood the bow as clarification was not tested.

2.3 Adult Dogs: Restarting Play

The most detailed test in adult dogs analyzed 414 play bows by 16 adult pet dogs during 229 bouts of dyadic play, coding the behavior of both dogs immediately before and after each bow (Byosiere et al., 2016a). The results did not support the punctuation account for adult dogs. No bites with head shaking were observed around the bows, the bowing dogs did not show more offensive behavior than their partners, and after a bow it was the partners, not the bowers, that became more offensive.

What the study did find was a different, consistent pattern. Bows most often followed a brief pause in play. After the bow, both dogs were more active, synchronous behaviors increased, and the bowing dog often ran away while the partner gave chase. The authors concluded that in adult dyadic play, bows mainly serve to re-initiate play after a pause, and suggested that the clarifying function Bekoff described may play a larger role in young animals (Byosiere et al., 2016a).

The study has clear limits. It relied on archived video, and four pairs supplied about three quarters of the data. The findings describe how a small number of dyads used the bow, but the central result recurs in other studies.

2.4 Puppies and Wolves

A companion study applied the same hypotheses to puppies (Byosiere, Espinosa, Marshall-Pescini, Smuts & Range, 2016b). It examined ten dog puppies aged two to five months and wolf puppies aged 2.7 to 7.8 months, all hand-reared in peer groups at a research center. Young dogs used play bows much like adult dogs: bows most often occurred after a brief pause and were followed by highly active play. In the dog puppies, it was again the partners rather than the bowers that showed more offensive behavior after a bow.

The wolf puppies differed. Their bows were about as frequent relative to observation time, but they did not succeed in eliciting further play activity from the partner. The authors concluded that the bow may function differently in wolf puppies with regard to re-initiating play. Two further details are worth noting. The puppy data gave no support to the idea that bows position the bower strategically for an attack, and very few synchronous behaviors followed bows in either puppy sample, unlike in adult dogs (Byosiere et al., 2016b).

These are research animals with a particular rearing history, and the groups were small, so the study does not show how family puppies use the bow.

2.5 A Replication in Wolfdogs

A study of Czechoslovakian wolfdogs tested the metacommunication hypothesis explicitly (Maglieri et al., 2023). Across 118 play sessions of 24 dogs, 76 play bows were recorded. All 76 bows were performed within the playmate's visual field. Bows were mainly performed during a short pause in an ongoing session, and the bow triggered the playmate's reaction again. The study found no evidence for the metacommunicative function: bows did not necessarily precede contact offensive patterns such as play biting or pushing, and their use was not affected by how asymmetric the session was.

The authors described the bow as a signal that restores the partner's motivation to play, and suggested that a mismatch in the number of bows two dogs perform may indicate different motivations to play. The sample is a single breed and the number of bows modest, but three independent analyses now tie bows to pauses and restarts rather than to reframing risky actions.

2.6 Where the Hypotheses Stand

Taken together, the evidence on the play bow points in one direction. The bow is a uniform display (Bekoff, 1977). In the data available for adult dogs, puppies raised in a research setting and one breed of wolfdog, it is used mainly after pauses and is followed by renewed activity (Byosiere et al., 2016a, 2016b; Maglieri et al., 2023). The punctuation account (Bekoff, 1995) has not been confirmed in these later samples. It has not been refuted for all contexts either: its original sample included young animals of three species, and no study has yet tested directly whether partners interpret bows as clarification.

In short, the bow behaves like an invitation to resume, not like a disclaimer attached to a bite. That still requires the sender to monitor the interaction and the partner's attention.

3. Attention: Signals Are Sent to Someone

3.1 Attention to Attention

A signal only works if it is perceived. An observational study of dyadic rough-and-tumble play examined whether dogs take this into account (Horowitz, 2009). The sequential behaviors and head direction of both dogs were recorded throughout the bouts. Play signals were sent nearly exclusively to partners that were facing forward. Attention-getting behaviors were used most often when the partner was facing away, and before a dog signaled an interest in playing.

Stronger attention-getters were used when the partner was looking away or distracted, less forceful ones when it faced forward or to the side. The author concluded that these dogs attended to, and acted to manipulate, their partner's attention (Horowitz, 2009). The abstract does not state the number of dogs or bouts.

3.2 Converging Data From Bows

In the adult pet dogs, 409 of 414 bows were given while the two dogs were within each other's visual field. Of the four remaining bows that could be assessed, the bowing dog barked in three (Byosiere et al., 2016a). In the wolfdogs, all 76 bows were performed within the playmate's visual field (Maglieri et al., 2023). This pattern, signals given where they can be seen and attention-getting behavior added where they cannot, is one of the clearer indications in canine research that a display is directed at a specific receiver (what the research shows about canine body language).

3.3 What This Does and Does Not Show

The finding shows that dogs adjust when and how they signal to whether the partner can see them. It does not show that dogs represent what the partner thinks, or that every dog does this equally well; individual differences have not been examined in detail.

In practice: When a dog repeatedly bows, barks or nudges a partner that is turned away or sniffing elsewhere, and the partner does not respond, this often looks less like play than like one dog trying to recruit another that is not interested. This is an observation from training practice, not a finding from controlled studies. The research above suggests what to look for: whether the partner turns toward the signal and joins in.

4. The Play Face and Rapid Mimicry

4.1 The Relaxed Open Mouth

The second classic play signal is facial: the relaxed open mouth, often called the play face. Unlike the bow, it can accompany long stretches of an interaction.

A study of 203 play sessions in an off-leash dog park in Palermo examined where the two signals occur (Cordoni, Nicotra & Palagi, 2016). Play bows were performed independently of the kind of play the dogs were engaged in, whether locomotor play such as chasing or contact play such as wrestling. The relaxed open mouth, by contrast, was used preferentially during contact play, when face-to-face interactions were more likely. In 82 percent of cases, play bites did not follow a relaxed open mouth. The authors read this as evidence that dogs place the display in contexts where it is easy to detect, and that an open mouth in play is not simply a bite about to happen.

4.2 Rapid Mimicry

A study of 49 dogs filmed during free play in a dog park found rapid mimicry of play signals, defined as reproducing the partner's display within one second (Palagi, Nicotra & Cordoni, 2015). Dogs showed a stronger and faster mimicry response after perceiving a play bow or a relaxed open mouth than after perceiving jumping or biting, two play patterns that resemble the signals in their motor form. Sessions punctuated by rapid mimicry lasted longer than sessions punctuated only by signals. And mimicry was more frequent between dogs with closer social bonds: the stronger the bond, the higher the level of mimicry.

The study was observational and limited to one park, but the control comparison matters: dogs copied the play signals more than similar-looking movements.

4.3 Mimicry Is Not Proof of Shared Emotion

The authors linked rapid mimicry to emotional contagion, a basic form of sharing another individual's state. That link is an interpretation: the study shows a fast, selective, bond-dependent motor response, but whether the dogs shared an emotional state was not measured (what is known about emotional contagion in dogs).

A recent review of social play across species makes a related point. Play has typically been associated with positive emotional states, but the review argues that it may relate to different emotional systems, from attachment to conflict, and that it oscillates between competition and cooperation (Cordoni & Norscia, 2024). A play face therefore tells us that a dog is signaling play. It does not guarantee that the interaction is pleasant for both partners at every moment (why behavior is not a direct readout of emotion).

5. Self-Handicapping, Role Reversal and Asymmetry

5.1 The Popular Rule

Advice on dog play often contains a rule of thumb: good play is balanced. The dogs take turns on top and in the chase, and the stronger dog holds back so that the weaker one can win. Self-handicapping and role reversal do occur and are among the markers used to define play (Bekoff & Byers, 1981). The question is whether balance is the norm.

5.2 The 50:50 Rule Tested

Bauer and Smuts (2007) analyzed the play of 55 dog dyads in a dog park and tested what they called the 50:50 rule, the expectation that play continues only if the partners balance their roles. The result ran against the expectation. Only 5.4 percent of the dyads showed a balanced ratio of winning and losing roles, and 21.8 percent played in a completely asymmetric way, with one dog consistently in one role. Older dogs and dogs classified as higher-ranking showed more offensive play elements and less self-handicapping than their partners. They did not give up their advantage to keep the game going.

The study is often cited as evidence that role reversal is the rule. It shows the opposite. Because welfare was not measured, it shows that asymmetry is common, not that it is always harmless.

5.3 Littermates

A longitudinal study followed play in four litters of puppies from 3 to 40 weeks of age (Ward, Bauer & Smuts, 2008). In mixed-sex pairs, males initiated play more often than females, showed more offensive play behavior toward females, and also self-handicapped more, that is, they more often adopted the weaker role or restrained themselves. In same-sex pairs, females initiated play with females across all periods, whereas males initiated play with males more only in the latest period. Play bows did not differ between the sexes.

Self-handicapping appeared, but depended on the composition of the pair. With four litters, the study does not show which combinations of adult dogs live together more peacefully (what the research shows about sex differences in dog behavior).

5.4 Asymmetry and Session Length

The Palermo study adds a dynamic view (Cordoni et al., 2016). Play asymmetry was high in the 203 sessions, and it did not differ between dogs that were strangers and dogs that were familiar with each other, which the authors took as a sign that play has a limited role in forming dominance relationships. Asymmetry was, however, negatively related to session length: the more lopsided a session, the shorter it lasted. More players were associated with longer sessions, and the number of play bows exchanged may also have influenced how long sessions lasted.

The authors suggested that dogs cope with asymmetry partly by ending sessions sooner, which may limit the risk of escalation, and partly by exchanging play signals. Escalation itself was not measured.

5.5 Rolling Over Is Not Submission

Rolling onto the back in play is often read as submission. A study of videotaped play between dog dyads tested this directly (Norman, Pellis, Barrett & Henzi, 2015). The frequency of rollovers was determined primarily by how long a bout lasted. A size difference between the partners had no effect on whether rollovers occurred, and there was no evidence that smaller dogs rolled over more often or stayed on their backs longer. The supine phase was mostly short.

Most rollovers were either defensive, such as evading a bite to the nape, or offensive, launching an attack from below. None could be categorized as submissive. The authors concluded that asymmetries in rollovers cannot be assumed to reflect asymmetries in the relationship between the partners.

5.6 Why a Dog Would Handicap Itself

If balanced play is rare, why does self-handicapping occur at all? One comparative hypothesis proposes that play rehearses recovery from sudden loss of control, such as falls, unexpected positions and momentary disadvantage, and that animals sometimes put themselves into such positions deliberately (Špinka, Newberry & Bekoff, 2001). On this view, self-handicapping is not mainly a courtesy to the partner but a way of creating the unexpected. The hypothesis comes from a review across mammals, not from dog studies.

6. Fair Play: A Hypothesis

6.1 The Argument

Marc Bekoff has argued that social play in group-living animals can reveal the evolutionary roots of fairness and social morality (Bekoff, 2004). Play, in this view, depends on a kind of social contract: individuals keep playing because it benefits both, and numerous mechanisms have evolved to facilitate the initiation and maintenance of play. Signals that announce play, restraint in bite strength and force, and the willingness to take the weaker role are read as elements of cooperative fair play, and the paper asks whether animals can forgive.

The argument is influential, but it is a synthesis, not a test, and its central claim, strong selection for cooperative fair play, is a hypothesis about evolution.

6.2 What the Data Would Need to Show

For fairness to be more than a metaphor, play would need to be fairly balanced on average or break down when it is not, and signals would need to appear when play risks being misread. The data fit this only partly. Balanced play was rare in the dog park sample, and older and higher-ranking dogs did not give up their advantage (Bauer & Smuts, 2007). Asymmetry shortened sessions but did not prevent them (Cordoni et al., 2016). And the play bow was tied to pauses rather than to risky actions (Byosiere et al., 2016a; Maglieri et al., 2023).

None of this rules out that dogs are sensitive to how a partner behaves in play. It does mean that "dogs play fair" is not an established finding. The same caution applies to related claims that dogs have a sense of fairness in other contexts, where the experimental evidence shows sensitivity to unequal outcomes but leaves its interpretation open (what the inequity aversion experiments show).

7. Pauses and Transitions

7.1 Pauses Are Part of Play

Pauses run through all of the play bow studies. In adult dogs, puppies and wolfdogs, bows most often came after a brief pause and were followed by renewed activity (Byosiere et al., 2016a, 2016b; Maglieri et al., 2023). A pause may give both dogs a moment to check whether the partner still wants to continue and to change roles, with the bow as the restart.

The play bow studies did not measure arousal, so they cannot show that pauses lower it (how arousal regulation works in dogs). They show that pauses and restarts are a recurring element of play between dogs.

7.2 Shaking Off

Another behavior seen in social settings is shaking, a rapid side-to-side movement of the body or head. An observational study of 96 dogs in a public dog run and a private daycare in New York City recorded 120 shaking episodes and the behaviors before and after them (Bryce, Nurkin & Horowitz, 2024). Shaking most often occurred between two different activities or behavioral categories, which supports the idea that it marks a transition. There was no significant change in postures related to affect after shaking, and no change in the number of dogs nearby.

The study did not single out play. It supports a modest reading: a shake often marks a transition between activities.

In practice: In group play, pauses, shakes and brief disengagements that are followed by both dogs voluntarily coming back together often look like a well-regulated game. Long sequences without any pause, in which arousal climbs steadily, more often precede the moment when one dog has had enough. This is an observation from practice and not a validated criterion.

8. Play Between Dogs and People

8.1 A Different Kind of Game

Dogs are unusual among carnivores in that adults routinely play, with other dogs and with people. According to a review of adult play, play between dogs is structurally different from play between dogs and people and may be motivationally distinct (Bradshaw, Pullen & Rooney, 2015). Dogs often compete over objects when they play with other dogs, but are usually more cooperative when the partner is human.

The same review notes that the majority of dogs do not seem to treat competitive games with a person as contests for dominance; winning possession of an object during a game appears simply rewarding. The authors suggest that the frequency and form of play may indicate the quality of the dog–owner relationship, a suggestion rather than a tested measure (Bradshaw et al., 2015).

8.2 Human Play Signals

If dogs use play signals with each other, do they respond to play signals from people? A two-part study addressed this (Rooney, Bradshaw & Robinson, 2001). In the first part, 21 owners were filmed playing with their dogs, and 35 actions used by three or more owners to start play were identified, including postures, vocalizations and physical contact. How effective each action was at starting play varied widely and was not related to how often owners used it.

In the second part, composites of common signals were tested with 20 Labrador retrievers. A human "bow" and a "lunge" both increased play. The lunge significantly increased both the number of play bouts and their mean duration. Adding play vocalizations made both postural signals more effective. The authors concluded that specific human actions communicate a playful context to dogs and can be considered interspecific play signals (Rooney et al., 2001).

The second part used a single breed and a standardized procedure, but the study shows that the play signals owners use most are not necessarily those that work best. Dogs' broader skill at reading human gestures is a separate research area (how dogs read human gestures).

8.3 Commands Are Not Play Signals

People often talk to their dogs during play, and much of that talk consists of commands: come, get it, give, drop. A study examined commands given by 21 familiar and 17 unfamiliar people who played with a dog, and the dog's responses (Mitchell, 2021). Dogs more often ignored than satisfied most commands during play, and more often played than not after a command. Dogs that played after a command were as likely to have satisfied it as to have ignored it, which suggests that commands do not serve as play signals. The author concluded that commands were more likely a response to what the dog was doing, spoken to direct or encourage the dog, than a signal that framed the interaction as play.

8.4 What People Experience

A citizen science project coded 187 dog–human play bouts from owner videos (Horowitz & Hecht, 2016). Dog affect was overwhelmingly positive, and physical contact, owner movement and closeness were highly correlated with positive owner affect. Owner-submitted videos probably overrepresent successful sessions.

8.5 Winning, Losing and Misread Signals

A persistent belief holds that letting a dog win tug games makes it dominant. An experiment with 14 Golden Retrievers tested this with 20 tug sessions that the dogs were allowed to win and 20 that they lost (Rooney & Bradshaw, 2002). Confidence, the factor closest to what is often called dominance, was unaffected. Obedient attentiveness was higher after play, whether the dogs won or lost. The ten most playful dogs showed more playful attention seeking after winning than after losing. The authors suggested that the effects of games may depend on play signals, and that outcomes may matter more when those signals are absent or misread (how play works as a reward in training).

In practice: In dog–human play, many dogs respond more to a person's posture, movement and tone than to words. A brief pause, a bow or a quick step back often restarts a game more reliably than a command. This is an observation from practice and fits the findings above, but it has not been tested as a training method.

9. Telling Play From Escalation

9.1 People Find It Hard

Whether two dogs are playing or fighting is a question every owner faces sooner or later, and the research suggests that humans are less reliable at answering it than they think. In a video study, 92 adults either categorized the context of short clips of interactions between children, dogs and macaques as aggressive, playful or neutral, or predicted their outcome (Epperlein, Kovacs, Oña, Amici & Bräuer, 2022). Participants performed above chance overall, but they were particularly poor at assessing aggression between dogs. Aggressive dog interactions were categorized at chance level, and predictions of how they would end were significantly below chance. Performance was poor for every aggressive dog clip.

An older study points the same way. Sixty observers, including owners, veterinarians, trainers and people without dog experience, described the behavior of nine dogs in video clips, and experience with dogs made no statistically significant difference; the differences lay between individuals across all groups (Tami & Gallagher, 2009). Both studies used short clips of unfamiliar dogs, which limits how far the results transfer to an owner watching their own dog. They nonetheless caution against confidence based on experience alone (what the research shows about aggression in dogs).

9.2 Criteria With Some Research Support

No study has validated a checklist that separates play from escalation in real time. Several features, however, are supported by the research reviewed above as typical of play between dogs:

• Signals directed at a partner that can see them, and responses to those signals (Horowitz, 2009; Byosiere et al., 2016a; Maglieri et al., 2023).

• Pauses followed by renewed activity, often with a bow (Byosiere et al., 2016a, 2016b).

• Rapid mimicry of bows and relaxed open mouths, which was associated with longer sessions (Palagi et al., 2015).

• Rollovers that are part of the action rather than an end to it (Norman et al., 2015).

What the research does not support is the use of balance as a test. Strongly asymmetric play was common in the dog park sample (Bauer & Smuts, 2007), and asymmetry did not differ between familiar and unfamiliar partners (Cordoni et al., 2016). A game in which one dog is mostly on top is not, on its own, evidence of a problem.

9.3 Criteria From Practice

Practitioners rely on additional signs that have not been tested as predictors but follow from the logic of the studies above.

In practice: The most useful single question is whether both dogs keep choosing to continue. If the dogs separate and both come back, the game is likely mutual. If one dog repeatedly tries to leave, hides behind a person or under a bench, or stops returning signals while the other keeps pursuing, the interaction has probably stopped being play for that dog, even if the pursuer still bows and still shows a relaxed open mouth. A short hands-off test, briefly holding the more active dog back and watching whether the other approaches again, can make this visible. This is an approach from practice, not a validated procedure.

Other signs used in practice include a stiffening body, a closed mouth with tense lips, sudden stillness, a change in vocalization and the disappearance of pauses. Each is better read as a change from how the dogs were playing a moment before, and a sequence is more informative than any one element (how to read single signals in sequence).

9.4 Mismatched Partners

Play between dogs depends on both partners. Differences in size, age, play style and motivation shape what happens. The research gives some anchors. Play bows did not differ between the sexes in littermates, but self-handicapping and offensive play did depend on the sex composition of the pair (Ward et al., 2008). Older and higher-ranking dogs held back less (Bauer & Smuts, 2007). A size difference did not affect rollovers (Norman et al., 2015). And a mismatch in the number of bows two dogs perform has been proposed as a sign of different motivations to play, a suggestion from one breed that has not been tested as a predictor (Maglieri et al., 2023).

Mismatch is not the same as asymmetry. A large dog playing with a small one may suit both if the small dog keeps returning, while two similar dogs may be a poor match if one wants to wrestle and the other to run. Free-ranging dogs and dogs in groups that can move away from each other have more opportunities to regulate such mismatches than dogs confined together (what research on free-ranging dogs shows about social behavior), and in households with several dogs the question of whether a partner can avoid play becomes central (what the research shows about multi-dog households).

9.5 When to Step In

Research does not provide a threshold at which play should be interrupted. A reasonable rule follows from the findings: step in when one dog no longer seems to choose the interaction, when signals stop being returned, when pauses disappear and arousal keeps climbing, or when you can no longer read what is happening. A short, calm break in play is usually less risky than overlooked overload; even so, any interruption should fit the situation and the safety of everyone involved.

How to interrupt matters as well. A sharp correction, a shout or a physical intervention can stop the behavior in the moment. The question is what the dogs learn from it and at what cost. A dog that is startled or hurt in the middle of play may associate the experience with the partner, the place or the person, and play that has been suppressed this way has not been resolved. The broader evidence on aversive methods shows that they carry measurable costs to welfare and to the dog–human relationship (what aversive methods do to dogs).

In practice: A calm, predictable interruption usually works better: recalling the dogs from a distance, moving away with them, or calmly leading one dog away while keeping enough distance from the other, and allowing a short break before deciding whether play continues. Do not step between highly aroused dogs or dogs that are already fighting in order to separate them. If the dogs come back together voluntarily and relaxed, play can resume. If one of them avoids the other, the game is over for now. This is an approach from practice, not a tested protocol.

9.6 Play Is Not Automatically Good

The leading review on dog play concluded that it is neither a reliable nor a generalizable welfare indicator, and that early experience, life history and the context of the interaction determine what any given episode means (Sommerville, O'Connor & Asher, 2017). A broader review noted that play often disappears first when an animal is under pressure, but that it can also increase under load, rebound after deprivation and occur under conditions no one would call good (Held & Špinka, 2011). Together with the view that play spans cooperation and competition (Cordoni & Norscia, 2024), this means that play signals tell us that a dog is framing an interaction as play. They do not tell us, on their own, that the interaction is good for both dogs.

10. Summary at a Glance

Play signals are behaviors that dogs use to start, frame and maintain play (Bekoff & Byers, 1981). The play bow is a uniform display (Bekoff, 1977) that was thought to clarify that risky actions are meant playfully (Bekoff, 1995). Later studies in adult pet dogs, research-reared puppies and wolfdogs did not confirm that specific interpretation of bows as clarifications around potentially ambiguous actions; bows mostly followed pauses and were followed by renewed activity, and they were almost always given within the partner's view (Byosiere et al., 2016a, 2016b; Maglieri et al., 2023). Dogs direct play signals at partners that are facing them and use attention-getters when they are not (Horowitz, 2009).

Dogs mimic each other's play bows and relaxed open mouths within a second, more so between closely bonded dogs (Palagi et al., 2015). Balanced play is rare: only 5.4 percent of 55 dyads balanced their roles, and older and higher-ranking dogs held back less (Bauer & Smuts, 2007). Asymmetry shortened sessions but did not prevent them (Cordoni et al., 2016). Rolling over in play was defensive or offensive, never classified as submissive (Norman et al., 2015). Fair play is a hypothesis (Bekoff, 2004), not an established finding.

Play between dogs and people differs structurally from play between dogs (Bradshaw et al., 2015). Human bows and lunges can start play, while commands do not act as play signals (Rooney et al., 2001; Mitchell, 2021). People are poor at recognizing aggression between dogs in short videos (Epperlein et al., 2022), and experience alone does not guarantee better reading (Tami & Gallagher, 2009). Play is not automatically a sign of good welfare (Sommerville et al., 2017; Held & Špinka, 2011).

11. Research Gaps and Critical Appraisal

11.1 Small and Concentrated Samples

The central studies are small: 16 adult dogs, with four pairs supplying about three quarters of the data (Byosiere et al., 2016a); ten dog puppies and a group of hand-reared wolves (Byosiere et al., 2016b); 24 wolfdogs of one breed with 76 bows (Maglieri et al., 2023); ten adult dogs and four infants each of three canid species (Bekoff, 1995); four litters (Ward et al., 2008). The abstracts available to us for the attention study and the supine posture study do not report sample sizes (Horowitz, 2009; Norman et al., 2015).

11.2 Observational Designs

Almost all dog–dog studies are observational. They show which behaviors occur together and in which order, not that a signal causes the partner's response: a partner that becomes active after a bow may simply be ready to resume after a pause. Experimental tests are largely missing; the dog–human study with composite signals is an exception (Rooney et al., 2001).

11.3 Settings and Populations

Most dog–dog data come from dog parks, daycare settings and research centers. Dog parks offer many partners and the chance to leave, and research-reared puppies have a different history from family puppies. Transfer to family dogs is plausible for the basic patterns but not established for the details.

11.4 Receiver Understanding

The most important open question is how partners interpret play signals. Signals are directed at attentive partners, followed by renewed activity and mimicked, but no study has tested directly whether a dog reads a bow as clarification of a later action, the core of the metacommunication hypothesis.

11.5 Escalation and Welfare

Few studies follow play to the point where it becomes a conflict. Escalation was not measured directly in the asymmetry study (Cordoni et al., 2016), and welfare was not measured in the role balance study (Bauer & Smuts, 2007). The human-perception data come from short video clips of unfamiliar animals (Epperlein et al., 2022; Tami & Gallagher, 2009). There are no validated criteria for when play should be interrupted, and the criteria used in practice have not been tested as predictors.

11.6 Interpretation Beyond the Data

Two interpretations in this field go beyond their data. The link between rapid mimicry and emotional contagion is plausible but was not measured (Palagi et al., 2015). The fairness account is a theoretical argument (Bekoff, 2004), and the observed asymmetries do not straightforwardly support it (Bauer & Smuts, 2007; Cordoni et al., 2016). Both ideas are valuable as hypotheses. Neither should be taught as fact.

12. Conclusion

Dogs clearly communicate about play. They use a stereotyped display, the play bow, send it to partners who can see it, recruit attention when they cannot, and respond to each other's play signals with fast and selective mimicry. What the research has not confirmed is the most popular version of the story: that the bow mainly works as a disclaimer attached to risky actions, and that good play is balanced and fair. In the data available, the bow behaves like an invitation to resume after a pause, and asymmetric play is common rather than exceptional.

For people watching dogs play, the practical lesson is to watch the structure of the interaction rather than decode single signals: whether signals are returned, whether pauses are followed by both dogs coming back, and whether each dog still chooses to continue. People misread dog interactions more often than they expect, so a calm interruption when in doubt is a reasonable default. Play signals tell us how a dog frames the interaction. Whether the game works for both dogs is something we have to observe.

Key Insights (Takeaways)

• The play bow is a uniform display, but in later samples of adult pet dogs, research-reared puppies and wolfdogs it mainly followed pauses and restarted play rather than clarifying risky actions (Bekoff, 1977; Bekoff, 1995; Byosiere et al., 2016a, 2016b; Maglieri et al., 2023).

• Dogs send play signals to partners that are facing them and use attention-getters when the partner looks away; 409 of 414 bows in one study were given within the partner's view (Horowitz, 2009; Byosiere et al., 2016a).

• Dogs mimic play bows and relaxed open mouths within a second, more than similar-looking movements and more between closely bonded dogs; whether they share an emotional state was not measured (Palagi et al., 2015).

• Balanced play is rare: only 5.4 percent of 55 dyads balanced their roles, so asymmetry alone is not a sign of a problem (Bauer & Smuts, 2007).

• Rolling onto the back in play was defensive or offensive, not submissive, in the dyads studied (Norman et al., 2015).

• Fair play is a hypothesis, not an established finding (Bekoff, 2004).

• People are poor at recognizing aggression between dogs in short videos, so whether both dogs keep choosing to continue is a more useful guide than a single signal (Epperlein et al., 2022).

References

Bauer, E. B., & Smuts, B. B. (2007). Cooperation and competition during dyadic play in domestic dogs, Canis familiaris. Animal Behaviour, 73(3), 489–499. https://doi.org/10.1016/j.anbehav.2006.09.006

Bekoff, M. (1977). Social communication in canids: Evidence for the evolution of a stereotyped mammalian display. Science, 197(4308), 1097–1099. https://doi.org/10.1126/science.197.4308.1097

Bekoff, M. (1995). Play signals as punctuation: The structure of social play in canids. Behaviour, 132(5–6), 419–429. https://doi.org/10.1163/156853995X00649

Bekoff, M. (2004). Wild justice and fair play: Cooperation, forgiveness, and morality in animals. Biology and Philosophy, 19(4), 489–520. https://doi.org/10.1007/sbiph-004-0539-x

Bekoff, M., & Byers, J. A. (1981). A critical reanalysis of the ontogeny and phylogeny of mammalian social and locomotor play: An ethological hornet's nest. In K. Immelmann, G. W. Barlow, L. Petrinovich, & M. Main (Eds.), Behavioral development: The Bielefeld interdisciplinary project (pp. 296–337). Cambridge University Press.

Bradshaw, J. W. S., Pullen, A. J., & Rooney, N. J. (2015). Why do adult dogs 'play'? Behavioural Processes, 110, 82–87. https://doi.org/10.1016/j.beproc.2014.09.023

Bryce, A., Nurkin, P., & Horowitz, A. (2024). Shake it off: Investigating the function of a domestic dog behavior in social contexts. Animals, 14(22), 3248. https://doi.org/10.3390/ani14223248

Byosiere, S.-E., Espinosa, J., & Smuts, B. (2016a). Investigating the function of play bows in adult pet dogs (Canis lupus familiaris). Behavioural Processes, 125, 106–113. https://doi.org/10.1016/j.beproc.2016.02.007

Byosiere, S.-E., Espinosa, J., Marshall-Pescini, S., Smuts, B., & Range, F. (2016b). Investigating the function of play bows in dog and wolf puppies (Canis lupus familiaris, Canis lupus occidentalis). PLOS ONE, 11(12), e0168570. https://doi.org/10.1371/journal.pone.0168570

Cordoni, G., Nicotra, V., & Palagi, E. (2016). Unveiling the "secret" of play in dogs (Canis lupus familiaris): Asymmetry and signals. Journal of Comparative Psychology, 130(3), 278–287. https://doi.org/10.1037/com0000035

Cordoni, G., & Norscia, I. (2024). Nuancing 'emotional' social play: Does play behaviour always underlie a positive emotional state? Animals, 14(19), 2769. https://doi.org/10.3390/ani14192769

Epperlein, T., Kovacs, G., Oña, L. S., Amici, F., & Bräuer, J. (2022). Context and prediction matter for the interpretation of social interactions across species. PLOS ONE, 17(12), e0277783. https://doi.org/10.1371/journal.pone.0277783

Held, S. D. E., & Špinka, M. (2011). Animal play and animal welfare. Animal Behaviour, 81(5), 891–899. https://doi.org/10.1016/j.anbehav.2011.01.007

Horowitz, A. (2009). Attention to attention in domestic dog (Canis familiaris) dyadic play. Animal Cognition, 12(1), 107–118. https://doi.org/10.1007/s10071-008-0175-y

Horowitz, A., & Hecht, J. (2016). Examining dog–human play: The characteristics, affect, and vocalizations of a unique interspecific interaction. Animal Cognition, 19(4), 779–788. https://doi.org/10.1007/s10071-016-0976-3

Maglieri, V., Zanoli, A., Mastrandrea, F., & Palagi, E. (2023). Don't stop me now, I'm having such a good time! Czechoslovakian wolfdogs renovate the motivation to play with a bow. Current Zoology, 69(1), 50–58. https://doi.org/10.1093/cz/zoac013

Mitchell, R. W. (2021). You talkin' to me? An assessment of commands as play signals during dog–human play. Animal Cognition, 24(2), 329–339. https://doi.org/10.1007/s10071-021-01469-5

Norman, K., Pellis, S., Barrett, L., & Henzi, S. P. (2015). Down but not out: Supine postures as facilitators of play in domestic dogs. Behavioural Processes, 110, 88–95. https://doi.org/10.1016/j.beproc.2014.09.001

Palagi, E., Nicotra, V., & Cordoni, G. (2015). Rapid mimicry and emotional contagion in domestic dogs. Royal Society Open Science, 2(12), 150505. https://doi.org/10.1098/rsos.150505

Rooney, N. J., & Bradshaw, J. W. S. (2002). An experimental study of the effects of play upon the dog–human relationship. Applied Animal Behaviour Science, 75(2), 161–176. https://doi.org/10.1016/S0168-1591(01)00192-7

Rooney, N. J., Bradshaw, J. W. S., & Robinson, I. H. (2001). Do dogs respond to play signals given by humans? Animal Behaviour, 61(4), 715–722. https://doi.org/10.1006/anbe.2000.1661

Sommerville, R., O'Connor, E. A., & Asher, L. (2017). Why do dogs play? Function and welfare implications of play in the domestic dog. Applied Animal Behaviour Science, 197, 1–8. https://doi.org/10.1016/j.applanim.2017.09.007

Špinka, M., Newberry, R. C., & Bekoff, M. (2001). Mammalian play: Training for the unexpected. The Quarterly Review of Biology, 76(2), 141–168. https://doi.org/10.1086/393866

Tami, G., & Gallagher, A. (2009). Description of the behaviour of domestic dog (Canis familiaris) by experienced and inexperienced people. Applied Animal Behaviour Science, 120(3–4), 159–169. https://doi.org/10.1016/j.applanim.2009.06.009

Ward, C., Bauer, E. B., & Smuts, B. B. (2008). Partner preferences and asymmetries in social play among domestic dog, Canis lupus familiaris, littermates. Animal Behaviour, 76(4), 1187–1199. https://doi.org/10.1016/j.anbehav.2008.06.004