The Subtleties of Communication: How Dogs Decode Human Gestures and Facial Expressions

1. Introduction: Interspecies Communication

1.1 Learning or Something More

Whether dogs' use of human gestures goes beyond associative learning is one of the central questions in this field, not a settled starting point. Dogs follow human points with little or no specific training, and puppies a few weeks old with little human contact already do so (Hare et al., 2002), which is hard to explain by each individual's reinforcement history alone. That does not rule learning out. A dog can have learned that a moving human arm predicts where searching pays off, and the account discussed in chapter 7 gives lifetime experience a central role (Udell et al., 2010). Successful use of a cue does not by itself show an understanding of the human intention behind it (part of the broader picture of canine cognition). The interesting questions begin with keeping those possibilities apart.

1.2 What "Decoding" Means Here

The word is convenient and it carries more than it should. Decoding a signal, in the sense used throughout this article, means using a cue to locate something, to predict what a person will do next, or to adjust behavior accordingly. It does not mean understanding a language, recovering an intention, or representing what the person has in mind.

The same restraint applies to “communication.” Here it describes signals passing between human and dog that change what the other does. It does not assume that the dog treats a human gesture as a deliberate message, or that the dog’s own signals are sent as one; the dog’s side of the exchange is described by what it does with a signal, not by what it intends.

Those stronger readings are the ones this literature is still arguing about, and chapter 2.2 sets out why. Where this article says a dog reads a gesture, it means the dog used it — nothing more.

1.3 How to Read the Evidence

The abilities are well established; their explanation is not. This article flags which claims describe what dogs do (strongly evidenced) versus why they can do it (an active scientific debate between domestication-selection and lifetime-experience accounts). Keeping those separate is essential to representing the science honestly rather than picking a side by omission.

2. Referential Communication: Pointing and Intent

2.1 Following Human Points

One of the most robust findings in canine cognition is that dogs use human pointing — even subtle, momentary gestures — to locate hidden rewards in object-choice tasks (Hare et al., 2002). Dogs also adjust to a human's attentional state and integrate gesture with context, suggesting they process pointing within a broader framework of attention and prediction rather than tracking bare movement (drawing on the same inference about relationships between actions and outcomes seen in causal reasoning).

2.2 Do Dogs Grasp Intent?

Here the first honest caution enters. It is often said dogs read the communicative intent behind a point, but simpler accounts — local enhancement, learned associations with human hands and gaze, stimulus salience — can explain much of the behavior, and dogs' performance varies with the type of point and the dog's experience. Whether dogs understand pointing as a genuinely referential, intent-carrying signal (as human infants do) or solve the task through leaner mechanisms remains an open and actively researched question, not a settled fact.

3. The Multi-Lab Replication

3.1 Why This Study Was Needed

Point-following is the most cited finding in canine social cognition, and for two decades it rested on a large number of small studies run by individual laboratories with differing procedures. That is exactly the configuration in which an effect can look more robust than it is.

The ManyDogs Project was set up to address replicability in canine science directly, and its first study took the most reliable-looking phenomenon in the field and tested it across sites (ManyDogs Project et al., 2023).

3.2 The Design

Data were collected from 455 dogs at 20 research sites on two conditions of a two-alternative object choice task. In the Ostensive condition the experimenter made eye contact and said the dog's name before pointing to a baited cup; in the Non-ostensive condition the point was made without eye contact, preceded by a throat-clearing auditory control cue (ManyDogs Project et al., 2023).

The prediction was straightforward. If dogs treat pointing as a socially communicative act rather than a learned cue, the ostensive framing should improve performance.

3.3 The Result

Both conditions were significantly above chance, and there was no significant difference between them. The result was consistent across sites (ManyDogs Project et al., 2023).

Dogs followed the point either way. The social framing that the communicative interpretation predicts should matter did not measurably change what they did.

3.4 The Second Finding

The study also reported that dogs followed contralateral, momentary pointing at lower rates than earlier research had described, which the authors read as showing that there are limits to the robustness of point-following: not all pointing styles are equally likely to elicit a response (ManyDogs Project et al., 2023).

Contralateral means the experimenter points across the body to the further cup; momentary means the hand is withdrawn before the dog chooses. Both make the task harder, and both are common in the literature without being treated as a variable (why small changes to a task change performance).

3.5 What the Authors Conclude

Their own summary is that procedural details in study design matter, and that canine science needs replication work (ManyDogs Project et al., 2023). That is a methodological conclusion rather than a claim about dogs, and it is the more consequential of the two.

A field in which the same nominal task produces different rates depending on how the hand moves is a field in which cross-study comparison has been doing less work than it appeared to.

3.6 How to Hold This

None of it shows that dogs cannot read pointing communicatively. A null difference between conditions is not evidence that the conditions are equivalent to the dog, and the ostensive manipulation may simply have been too weak against a lifetime of everyday human contact.

What it does is remove a piece of support that was routinely treated as established, and it raises the standard of evidence any future claim in this area has to meet (a problem early testing runs into for the same reasons).

3.7 What the Object Choice Task Is

Nearly everything in this section comes from one paradigm. Two containers are baited out of the dog's sight, an experimenter indicates one, and the dog chooses. Correct choice is scored, usually across a fixed number of trials, and performance is compared against chance.

Its strengths are that it is simple, quick and comparable across species. Its weakness is that it compresses a rich question — what does the dog take the gesture to mean — into a binary outcome, and every conclusion in this literature has to travel through that compression.

3.8 Controls That Matter

Two controls separate gesture-reading from simpler explanations. An odor control establishes that the dog is not smelling the baited container, and a body-versus-point control establishes that the dog is not simply approaching whatever is nearest the human.

Studies that omit these are not measuring what they report measuring, and the presence or absence of such controls is a faster way to weigh a paper in this field than the size of its sample.

4. Facial Expression Processing

4.1 Discriminating Emotional Faces

Dogs can discriminate human facial expressions and attach emotional relevance to them, distinguishing happy from angry faces and responding differently (Müller et al., 2015). More strikingly, cross-modal work shows dogs matching a visual emotional cue to a congruent vocal one — orienting appropriately when a face and a voice carry the same emotion — which implies an integrated internal representation of emotional states rather than isolated feature detection (Albuquerque et al., 2016).

4.2 The Neuroimaging Picture

Neuroimaging adds support, with an accurate caveat about what was tested. Comparative fMRI revealed voice-sensitive regions in the dog brain, in locations analogous to humans', that respond to the emotional valence of vocalizations (Andics et al., 2014). This is strong evidence that dogs process human emotional voices with brain machinery resembling ours; it concerns vocal rather than facial emotion, and is best cited as such. Together with the behavioral face work, it builds a picture of dogs as sensitive processors of human emotional signals (the bond underlying which is shaped by oxytocin).

4.3 What Face Studies Can and Cannot Show

The distinction to hold is between responding to an emotional cue and understanding an emotion. Discrimination is the outcome measure in most of this work: the dog reliably distinguishes one class of face from another. Discrimination establishes that the dog detects a difference and uses it, which is a real finding and a modest one.

It does not establish that the dog experiences the expression as an emotion, or attributes a state to the person. Those are inferences, and the paradigms are generally not built to test them (as the judgment-bias work handles the same inference problem).

4.4 The Stimulus Problem

Most face studies use photographs. A photograph is static, two-dimensional, cropped, evenly lit and detached from body, voice, movement and odor — that is, it removes almost everything a dog would normally use.

Performance on photographs may therefore understate what dogs do with real faces, or overstate it by removing conflicting channels. Which of the two is happening has not been established, and it is a live limitation on the whole subfield.

5. Gaze Following and Social Attention

5.1 What Is Established

Gaze is a central communicative channel, and dogs are highly sensitive to human gaze direction, using it to locate objects, anticipate actions, and gauge whether they are being watched (a form of prediction about what a human will do next). Dogs follow human gaze into distant space, distinguish attentive from inattentive people, and adjust behavior to whether they are observed (a sensitivity to others' attentional states related to metacognitive monitoring). Mutual gaze has also been linked to physiology: in a laboratory study, longer mutual gaze between dogs and their owners was associated with rises in urinary oxytocin in both partners, which the authors proposed as a positive feedback loop between attention and bonding (Nagasawa et al., 2015); 5.3 sets out what that result does and does not establish.

5.2 What Gaze Following Does Not Establish

Following a gaze into distant space is a robust finding and it admits more than one explanation. An animal that orients where a conspecific orients gains information whether or not it attributes attention to the other individual, and orienting is the observable while attribution is the inference.

The distinction matters because the popular version of this research runs the inference automatically, and the experimental designs are generally not built to separate the two.

5.3 The Oxytocin Loop and Its Limits

The reciprocal gaze finding is among the most widely repeated results in the field (Nagasawa et al., 2015), and it is worth stating what it establishes: an association between mutual gaze and urinary oxytocin in both partners, in a laboratory setting, with the usual difficulties of measuring a peptide in urine.

It is a real result about a real physiological system. It is not a demonstration that the bond consists of that loop, and reading it that way flattens a relationship into a single hormone (where the relationship itself becomes the variable).

5.4 Eye Contact Has More Than One Function

Sustained gaze at a person is frequently read as a request — the dog is asking for help, or for the door, or for the food. It sometimes is, and the same behavior also occurs as orientation toward the most informative thing in the room, as a learned response that has reliably produced attention, and as part of waiting for an expected outcome.

Which of those is operating is not readable from the gaze itself, and the practical consequence is that a dog looking at a person is telling you where its attention is, not what it wants (reading attention without inferring intention).

5.5 Attention Is Not Only Visual

Studies of attentional state overwhelmingly manipulate what the human can see. A dog's assessment of whether a person is available also draws on posture, movement, sound and, plausibly, odor, none of which the standard paradigms vary systematically.

What the literature therefore describes is sensitivity to visual availability specifically, which is narrower than sensitivity to attention (as the free-ranging literature describes for dog-dog life).

6. Emotional Contagion and Bidirectional Signaling

6.1 The Two-Way Channel

Communication runs both ways. Dogs do not only decode human signals; they are influenced by human emotional states through emotional contagion — the transfer of affect between individuals. Work on emotional contagion reports associations between human stress and stress measures in dogs, although most of it is correlational and the direction of influence is rarely settled (explored in depth in the work on human stress and canine behavior). If those associations reflect influence in both directions, human–dog communication is partly a shared physiological and emotional system rather than a one-way transmission from human to dog.

6.2 What Bidirectional Means Here

Bidirectional is often read as symmetrical, and the evidence does not support that. Humans shape the dog's environment, decide what happens next and control access to nearly everything the dog cares about. The influence runs both ways; the power does not.

That asymmetry belongs in any account of the relationship as a communicative system, because it determines who bears the cost when the communication fails.

6.3 Contagion Is Not Empathy

Affect transfer between individuals is a low-level process documented across many species, and it does not require understanding another's state. Empathy in the fuller sense adds a representation of what the other is experiencing, which is a much stronger claim and a much harder one to test.

Treating the first as evidence of the second is the recurring error in this area, and it survives because the stronger reading is the more appealing one.

7. The Domestication Debate: How Did Dogs Get This Way?

This is where the article most needs to correct a common oversimplification. The origin of dogs' human-directed social skills is one of the liveliest debates in the field, and it is not resolved.

7.1 The Domestication-Selection Hypothesis

The influential early account holds that gesture-reading is a derived trait produced by domestication. In the founding study, dog puppies with little human contact followed human points while human-reared wolves did not, suggesting selection during domestication for social-cognitive skills specialized for communicating with humans (Hare et al., 2002). Related work argued dogs are even sensitive to human ostensive-communicative cues in ways reminiscent of human infants (Topál et al., 2009).

7.2 The Ontogeny Counter-Account

That narrative was directly challenged. Given the right rearing and daily human interaction, wolves can use momentary distal pointing to find food without training — and, in the same study, dogs tested outdoors or living in a shelter did not follow the same points (Udell et al., 2008). This reverses the tidy "wolves can't, dogs can" story and points instead to a two-stage account in which lifetime socialization, experience, and testing conditions — not a fixed domestication-given module — largely determine performance (Udell et al., 2010). On this view, the dog's skill is real but emerges through ontogeny, and is not evidence of a hard-wired, uniquely canine faculty.

7.3 Toward a Synthesis

The field has moved toward integration, and the most recent evidence complicates the picture. Range and Virányi (2015) argue that the social attentiveness and cooperative tendencies dogs direct at people have roots in wolf social life, which would make the human-directed skills a modification of an inherited capacity rather than something created from nothing. The puppy comparison in chapter 8 adds that the modification is visible early: dog puppies with limited human contact read human gestures more skillfully than wolf puppies raised with intensive human care, and no wolf puppy succeeded at the individual level (Salomons et al., 2021). The success of socialized wolves in Udell et al. (2008) and the failure of wolf puppies in that comparison are not necessarily in conflict, because the two studies tested animals under different rearing and testing conditions (which is exactly where the early sensitive period does its work). The defensible summary is that both domestication and individual development contribute, that a dog–wolf difference is already visible in puppies, and that its size in any given comparison depends on rearing, testing conditions and analytic choices the field still disputes (Hansen Wheat, van der Bijl & Wynne, 2023; Salomons et al., 2023). The skills are present across dog breeds, and breeds also differ in them: in citizen-science data from 1,508 dogs of 36 breeds, a communication factor showed an among-breed heritability of 0.39 (Gnanadesikan et al., 2020), which does not make breed a good predictor of any individual dog (consistent with how weakly breed predicts individual behavior).

7.4 Why the Debate Has Lasted Twenty Years

The two accounts are not straightforwardly rival predictions about the same observation. The selection account expects the skill to appear early and with little experience; the ontogeny account expects it to track exposure. Both are compatible with a dog that follows points, and separating them requires either animals with unusual rearing histories or datasets large enough to model age and experience independently.

Both kinds of evidence are expensive, which is why the same handful of studies keeps being cited by both sides — and, as the exchange in the next chapter shows, sometimes the same dataset (where what domestication changed is more contested than usually presented).

7.5 What Would Actually Decide It

The decisive design is a longitudinal one: the same animals tested repeatedly from weaning onward, with rearing conditions held equivalent across species and enough subjects in each age band to make an age effect visible.

Nothing on that scale exists. Wolf puppies raised to research standards are rare, expensive and hard to test, and the ethical and practical constraints are not going to ease.

It is worth saying plainly what follows from that. This question will probably stay open for structural reasons rather than because nobody has thought of the right study, and any account that presents it as settled — in either direction — is describing a preference rather than a finding. That applies to the confident domestication story sold in popular books and equally to the confident learning story offered as its correction.

8. The Wolf Comparison and the Exchange That Followed

8.1 The Study

The sharpest recent test of the domestication account compared 44 dog puppies with 37 wolf puppies aged five to eighteen weeks on a battery of temperament and cognition tasks (Salomons et al., 2021).

The rearing was deliberately asymmetric in the wolves' favor. Wolf puppies received human care from ten to eleven days after birth — caregivers available for constant contact, hand-feeding, sleeping with the litter through the testing period — while the dog puppies stayed with their mothers and littermates with relatively limited human interaction (Salomons et al., 2021).

8.2 What Was Found

Dog puppies were more attracted to humans, read human gestures more skillfully and made more eye contact with humans than the wolf puppies. Both species performed similarly on memory and inhibitory control tasks (Salomons et al., 2021).

The pattern is what makes the result interesting: the species difference appeared in the social domain and not in the non-social one, which is what the domestication hypothesis predicts and what a general intelligence difference would not.

8.3 The Reanalysis

The dataset was published, and it was reanalyzed. Hansen Wheat, van der Bijl and Wynne argued that when environmental factors are adequately controlled, wolves and dogs perform similarly, and that rearing condition together with willingness to approach a stranger explains the difference in point-following (Hansen Wheat, van der Bijl & Wynne, 2023).

Their two specific questions were whether dog puppies outperformed wolf puppies before being placed individually in raisers' homes, and whether the tendency to approach a stranger predicts gesture success better than species does.

8.4 The Reply

The original authors ran further analyses. Comparing puppies still living with littermates who had not yet gone to raisers' homes (mean age 7.6 weeks, N = 13) against same-age wolf puppies with more human exposure (N = 5), the puppies were above chance on the pointing task at 70.5% correct while the wolves at 60% were not; on the marker gesture the figures were 85.9% against 56.7% (Salomons et al., 2023).

They also report that these littermate puppies did not differ significantly from older puppies already living in raisers' homes on the pointing task, and that no individual wolf of any age performed above chance on the combined gesture measure (Salomons et al., 2023).

8.5 The Approach-a-Stranger Objection

On the second question the reply argues that the design already handles it: both a correct and an incorrect choice in the two-way object choice task require approaching a bowl equidistant from the experimenter, all subjects passed a memory test requiring the same approach behavior, and in a body-versus-point control the wolves did not avoid the experimenter (Salomons et al., 2023).

They add that species and willingness to approach a stranger co-vary strongly, so that model comparison cannot separate their contributions in this dataset (Salomons et al., 2023). That is a concession as much as a rebuttal: the two variables cannot be told apart with the data available.

8.6 What a Reader Should Take From It

The exchange is worth reading in full rather than picking a side, because it demonstrates something the popular version never shows: the wolf sample at the youngest ages is five animals, the analytic choices materially change the conclusion, and both teams are working from the same numbers.

The defensible summary is that dog puppies with little human exposure use human gestures at rates wolf puppies with much more human exposure do not, that the size of that gap depends on how age and rearing are handled, and that the underlying samples are small enough for those choices to matter.

9. Sources of Miscommunication

9.1 Where It Breaks Down

For all their skill, dogs do not read us perfectly, and miscommunication arises when signals are ambiguous, inconsistent, or mismatched to canine perception. Common culprits include anthropomorphic assumptions about what a dog understands, conflicting channels (cheerful words with tense posture), and poorly timed cues. Inconsistent communication raises uncertainty and cognitive load, which can engage stress pathways and, over time, impair learning and stability (the physiological cost of chronic uncertainty).

9.2 The Cue Is Not What You Think It Is

Dogs learn what actually predicts the outcome, which is often not the element the handler believes is doing the work. A verbal cue delivered with a consistent hand movement may be functioning as a hand cue, and this only becomes visible when one channel is removed.

Testing it is simple and rarely done: give the signal without the accompanying movement and see what happens.

The same applies in the other direction. A dog that appears to respond to a spoken word may be responding to the moment the handler stops moving, to a shift in posture that precedes the word, or to the position the handler has taken in the room. None of that is a failure on the dog's part. It is a description of what the dog was actually given to work with, and it is recoverable only by changing one channel at a time. Whether dogs learn spoken words as names for objects is a separate question (word learning in dogs).

9.3 Context Is Part of the Signal

A cue learned in one room, at one distance, with one person, is a cue tied to those conditions until it has been trained otherwise. Failure in a new setting is usually not disobedience but a signal the dog has not yet met in that form.

This is the single most common source of the impression that a dog understands and chooses not to comply.

9.4 Human Faces Are Difficult Objects

Human facial expression is a channel dogs demonstrably use, and it is also the channel humans most overestimate. Expressions are fast, individually variable, and frequently at odds with the rest of the body.

Where a handler wants to be understood, posture and movement are the more legible instruments, and they are the ones under conscious control for longer.

10. Multimodal Integration: Combining Channels

10.1 Reading Several Channels at Once

Dogs' interpretation of human communication rests on integrating multiple channels — gesture, gaze, facial expression, and tone of voice — into a coherent read of intent, rather than responding to any single cue in isolation. This multimodal combination is what lets a dog reconcile a word, a look, and a posture into one interpretation, and it suggests processing beyond simple single-cue responses (a capacity built on social learning from humans). It is also why coherence across channels matters so much in practice.

10.2 Integration Has a Cost

Combining channels takes processing, and processing is affected by arousal, fatigue, pain and age. A dog that reads a handler accurately in a quiet room may not do so in a busy one, and the difference is not motivational.

Communication failures cluster in exactly the conditions where integration is hardest, which is a more useful diagnosis than assuming the dog knows better.

10.3 What Would Advance This

The multimodal claim is plausible and thinly tested. What is missing are designs that put channels into conflict systematically — gesture pointing one way, gaze another, voice a third — and measure which one the dog follows, across enough dogs to see the spread.

Until that exists, multimodal integration is a reasonable description of what dogs appear to do rather than a mechanism that has been demonstrated.

11. What This Means in Practice

11.1 A Point Is Not a Cue Until It Is Trained

Dogs follow points above chance without instruction, which chimpanzees tested on the same kind of task did not match (Hare et al., 2002). That is not the same as a reliable cue in a working context, where the pointing style, the distance, the distractions and the dog's prior history all move the success rate.

The replication data make this concrete: contralateral momentary pointing produced lower rates than the literature had led people to expect (ManyDogs Project et al., 2023). A handler whose signal is not working may have a signaling problem rather than a dog problem.

11.2 Same Side, Held Until the Dog Chooses

The two features that lowered success in the replication data, pointing across the body and withdrawing the hand before the dog chooses, can both be avoided. Pointing on the same side as the target and holding the gesture until the dog has chosen removes those two difficulties. Whether distance to the target adds a further cost has not been isolated in the studies reviewed here, so keeping it short is a precaution rather than a finding.

Where clarity matters — teaching a new behavior, working at distance, an older dog with reduced vision — moving toward the easier version costs nothing and removes a variable (what a dog can actually see).

11.3 Do Not Assume the Dog Knows You Mean It

The ostensive manipulation in the multi-lab study did not change performance (ManyDogs Project et al., 2023). Whatever the dog is doing when it follows a point, adding eye contact and its name did not measurably improve it under those conditions.

In practice that argues against relying on a felt sense that the dog understood the intention behind a signal. The observable question is whether the behavior occurred, at what rate, under which conditions.

11.4 Channel Conflict Costs More Than Any Single Cue

Because dogs integrate gesture, gaze, expression and voice, incoherence across those channels is more disruptive than a weak signal in one. Cheerful words delivered with a tense body are not a mixed message to a dog; they are two messages, and the postural one tends to be the older and more reliable channel.

Handlers who report inconsistent responses often have consistent hands and inconsistent bodies (practical tips for clearer everyday communication).

11.5 Individual Variation Is Not Noise

Group means in this literature conceal individual spread. The object choice task scores performance against chance, and a group result above chance is compatible with some individuals performing at chance; in the puppy comparison, half of the dog puppies succeeded at the individual level (Salomons et al., 2021). Assuming a given dog reads gestures as well as the average is an assumption, not an observation.

Testing it takes minutes: a few trials with a baited container and a clear point will tell you where an individual sits.

11.6 What Not to Conclude About Your Dog

A dog that does not follow a point is not thereby unintelligent, untrained or disconnected. Non-social task performance was comparable between species in the puppy comparison (Salomons et al., 2021), and within dogs the same dissociation is plausible.

Reading gesture is one narrow ability among many, and it is the one that happens to have been studied most.

12. Summary at a Glance

Point-following is real but not uniform — Across 455 dogs at 20 sites, performance was above chance, but contralateral momentary pointing produced lower rates than earlier work reported (ManyDogs Project et al., 2023).

Ostensive framing made no measurable difference — Adding eye contact and the dog's name before the point did not significantly change performance compared with a non-ostensive control (ManyDogs Project et al., 2023).

Dog puppies outperformed human-raised wolf puppies on gestures — 44 dogs against 37 wolves aged five to eighteen weeks, with the wolves receiving far more human contact (Salomons et al., 2021).

The species difference was social, not general — Memory and inhibitory control performance was comparable between the two species (Salomons et al., 2021).

The dataset was reanalyzed and contested — A commentary argued that rearing condition and willingness to approach a stranger account for the difference (Hansen Wheat, van der Bijl & Wynne, 2023).

The original authors replied with further analyses — Puppies not yet placed in homes were above chance where same-age wolves were not, though the wolf sample at that age was five animals (Salomons et al., 2023).

Species and temperament cannot be separated in that dataset — The reply concedes that the two variables co-vary too strongly for model comparison to parse them (Salomons et al., 2023).

Procedure is a variable, not a detail — The multi-lab authors conclude that study design details shape the result and that canine science needs replication work (ManyDogs Project et al., 2023).

13. Research Gaps and Critical Appraisal

The evidence is strong on abilities and genuinely open on mechanism.

The abilities are robust. Point-following, emotional-face discrimination, cross-modal emotion matching, and gaze use are well replicated in dogs (Hare et al., 2002; Müller et al., 2015; Albuquerque et al., 2016).

Origin is unresolved. Whether these skills are domestication-derived, ontogeny-driven, or both is actively debated. Socialized wolves followed momentary distal points in one study (Udell et al., 2008), while wolf puppies with intensive human care did not match dog puppies on gesture tasks (Salomons et al., 2021), and the size of that difference depends on how rearing and age are handled (Hansen Wheat, van der Bijl & Wynne, 2023). Claims that wolves categorically "cannot" read human gestures are not supported, and neither are claims that the dog–wolf difference has been explained away.

Intent is uncertain. Whether dogs grasp communicative intent or use leaner mechanisms (local enhancement, learned cues) is not settled, and rich mentalistic interpretations outrun the data.

Method sensitivity. Performance depends heavily on rearing, testing environment, and task design, so results generalize cautiously (a recurring measurement challenge in behavior research).

Individual variation. Socialization history and the specific relationship shape a given dog's skill, so population findings map loosely onto any one dog.

The central finding did not replicate as expected. The largest multi-site study found no difference between ostensive and non-ostensive pointing, and lower rates for contralateral momentary pointing than earlier work reported (ManyDogs Project et al., 2023). Summaries written before 2023 read more confidently than the current evidence supports.

The wolf samples are small. In the exchange over the puppy comparison, the youngest wolf group numbered five animals (Salomons et al., 2023). Conclusions about species differences at that age rest on a sample where a single individual moves the percentage substantially.

Species and temperament cannot be separated in the available data. The original authors state that the two co-vary too strongly for model comparison to parse them (Salomons et al., 2023). That is an open question rather than a settled one in either direction.

Almost everything rests on one paradigm. The two-alternative object choice task carries the weight of this literature. A finding that appears across many studies using one task is less robust than the same finding appearing across several different tasks.

14. Conclusion

Human–dog communication is a dynamic, adaptive system that integrates perception, cognition and emotion, and dogs are active interpreters of human signals rather than passive recipients — following points, discriminating emotional faces, matching faces to voices, using gaze, and responding to human emotional states. These abilities are robustly documented, and the largest test of point-following to date confirmed the core finding across 20 sites while showing that pointing style matters and that ostensive framing made no measurable difference (ManyDogs Project et al., 2023). What remains open is why dogs have these skills. The early account in which domestication alone created a gesture-reading faculty absent in wolves was challenged by socialized wolves that followed human points (Udell et al., 2008); that challenge was in turn qualified by dog puppies that outperformed intensively socialized wolf puppies on gesture tasks while matching them on memory and inhibitory control (Salomons et al., 2021); and that result is disputed on analytic grounds, with wolf samples at the youngest ages small enough for those choices to matter (Hansen Wheat, van der Bijl & Wynne, 2023; Salomons et al., 2023). The current evidence supports a contribution of both selection and lifetime experience without settling their relative weight. For training and welfare, the practical lesson does not depend on that debate: communication is not a set of commands issued at a dog but a shared system, and coherent, well-timed, non-contradictory signals work with the capable but imperfect interpreter on the other end of the leash.

Key Insights (Takeaways)

• Dogs demonstrably use human communicative signals: they follow human pointing, above chance across 455 dogs at 20 sites (ManyDogs Project et al., 2023), discriminate happy from angry faces (Müller et al., 2015), match emotional faces to voices cross-modally (Albuquerque et al., 2016), and process emotional human voices with brain regions analogous to ours (Andics et al., 2014). How reliably a dog follows a point depends on how it is made: contralateral, momentary pointing produced lower rates than earlier work reported.

• Why they can is debated. The early account that domestication created gesture-reading absent in wolves (Hare et al., 2002) was challenged when socialized wolves followed human points and some dogs tested outdoors or at a shelter did not (Udell et al., 2008). Dog puppies then outperformed wolf puppies with far more human contact on gesture tasks, with no individual wolf puppy succeeding (Salomons et al., 2021), and a reanalysis argues that rearing condition and willingness to approach a stranger explain the difference (Hansen Wheat, van der Bijl & Wynne, 2023). Both selection and experience appear to contribute; their relative weight is unresolved.

• Whether dogs grasp communicative intent or use leaner mechanisms (learned cues, local enhancement) is unresolved, and adding eye contact and the dog's name before a point did not measurably change performance (ManyDogs Project et al., 2023), so mentalistic claims should be held cautiously. The skills are present across breeds, though breeds also differ: a communication factor showed an among-breed heritability of 0.39 in 1,508 dogs of 36 breeds (Gnanadesikan et al., 2020).

• Communication runs both ways: longer mutual gaze was associated with rises in urinary oxytocin in dogs and owners in a laboratory study (Nagasawa et al., 2015), and work on emotional contagion links human and canine stress measures, mostly correlationally — so the human's state is plausibly part of the signal rather than outside it.

• Practically, miscommunication comes from conflicting channels, mis-timed cues, and anthropomorphic assumptions. Coherent, consistent, well-timed multimodal signals work best; treat communication as a shared system, not a stream of commands.

References

Albuquerque, N., Guo, K., Wilkinson, A., Savalli, C., Otta, E., & Mills, D. (2016). Dogs recognize dog and human emotions. Biology Letters, 12(1), 20150883. https://doi.org/10.1098/rsbl.2015.0883

Andics, A., Gácsi, M., Faragó, T., Kis, A., & Miklósi, Á. (2014). Voice-sensitive regions in the dog and human brain are revealed by comparative fMRI. Current Biology, 24(5), 574–578. https://doi.org/10.1016/j.cub.2014.01.058

Gnanadesikan, G. E., Hare, B., Snyder-Mackler, N., & MacLean, E. L. (2020). Estimating the heritability of cognitive traits across dog breeds reveals highly heritable inhibitory control and communication factors. Animal Cognition, 23(5), 953–964. https://doi.org/10.1007/s10071-020-01400-4

Hansen Wheat, C., van der Bijl, W., & Wynne, C. D. L. (2023). Rearing condition and willingness to approach a stranger explain differences in point following performance in wolves and dogs. Learning & Behavior, 51(2), 127–130. https://doi.org/10.3758/s13420-022-00544-2

Hare, B., Brown, M., Williamson, C., & Tomasello, M. (2002). The domestication of social cognition in dogs. Science, 298(5598), 1634–1636. https://doi.org/10.1126/science.1072702

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

Müller, C. A., Schmitt, K., Barber, A. L. A., & Huber, L. (2015). Dogs can discriminate emotional expressions of human faces. Current Biology, 25(5), 601–605. https://doi.org/10.1016/j.cub.2014.12.055

Nagasawa, M., Mitsui, S., En, S., Ohtani, N., Ohta, M., Sakuma, Y., Onaka, T., Mogi, K., & Kikusui, T. (2015). Oxytocin-gaze positive loop and the coevolution of human–dog bonds. Science, 348(6232), 333–336. https://doi.org/10.1126/science.1261022

Range, F., & Virányi, Z. (2015). Tracking the evolutionary origins of dog–human cooperation: The "Canine Cooperation Hypothesis." Frontiers in Psychology, 5, 1582. https://doi.org/10.3389/fpsyg.2014.01582

Salomons, H., Smith, K. C. M., Callahan-Beckel, M., Callahan, M., Levy, K., Kennedy, B. S., Bray, E. E., Gnanadesikan, G. E., Horschler, D. J., Gruen, M., Tan, J., White, P., vonHoldt, B. M., MacLean, E. L., & Hare, B. (2021). Cooperative communication with humans evolved to emerge early in domestic dogs. Current Biology, 31(14), 3137–3144. https://doi.org/10.1016/j.cub.2021.06.051

Salomons, H., Smith, K. C. M., Callahan-Beckel, M., Callahan, M., Levy, K., Kennedy, B. S., Bray, E. E., Gnanadesikan, G. E., Horschler, D. J., Gruen, M., Tan, J., White, P., vonHoldt, B. M., MacLean, E. L., & Hare, B. (2023). Response to Hansen Wheat et al.: Additional analysis further supports the early emergence of cooperative communication in dogs compared to wolves raised with more human exposure. Learning & Behavior, 51(2), 131–134. https://doi.org/10.3758/s13420-023-00576-2

Topál, J., Gergely, G., Erdőhegyi, Á., Csibra, G., & Miklósi, Á. (2009). Differential sensitivity to human communication in dogs, wolves, and human infants. Science, 325(5945), 1269–1272. https://doi.org/10.1126/science.1176960

Udell, M. A. R., Dorey, N. R., & Wynne, C. D. L. (2008). Wolves outperform dogs in following human social cues. Animal Behaviour, 76(6), 1767–1773. https://doi.org/10.1016/j.anbehav.2008.07.028

Udell, M. A. R., Dorey, N. R., & Wynne, C. D. L. (2010). What did domestication do to dogs? A new account of dogs' sensitivity to human actions. Biological Reviews, 85(2), 327–345. https://doi.org/10.1111/j.1469-185X.2009.00104.x