Differential Sensitivity to Human Communication in Dogs, Wolves, and Human Infants

See allHide authors and affiliations

Science  04 Sep 2009:
Vol. 325, Issue 5945, pp. 1269-1272
DOI: 10.1126/science.1176960


Ten-month-old infants persistently search for a hidden object at its initial hiding place even after observing it being hidden at another location. Recent evidence suggests that communicative cues from the experimenter contribute to the emergence of this perseverative search error. We replicated these results with dogs (Canis familiaris), who also commit more search errors in ostensive-communicative (in 75% of the total trials) than in noncommunicative (39%) or nonsocial (17%) hiding contexts. However, comparative investigations suggest that communicative signals serve different functions for dogs and infants, whereas human-reared wolves (Canis lupus) do not show doglike context-dependent differences of search errors. We propose that shared sensitivity to human communicative signals stems from convergent social evolution of the Homo and the Canis genera.

Convergent findings indicate that human infants pay special attention to various nonverbal communicative signals directed at them [such as eye contact, gaze-shifts, and pointing (15)]. This skill may provide the basis for preverbal infants’ early emerging competence to engage in triadic communicative interactions with others (68). Recent evidence suggests that signals conveying manifestation of intention to communicate induce a learning attitude in infants, which enables them to acquire knowledge from observation of adults’ demonstrations (9). In brief, infants are biased to assume that the information manifested by the adult can be generalized to other situations. We have shown (10) that this bias can account for young infants’ perseverative search error (11) in the A-not-B object search task. In our study with 10-month-olds (10), participants observed a human demonstrator repeatedly hiding (and then allowing the infant to retrieve) an object at one of two potential hiding locations: four times at location A and then 3 times at location B, either in a communicative context or without any social cues. In the communicative context, we replicated the standard finding of a strong tendency to search (erroneously) for the object at its previous hiding location (A) during the B trials. This perseverative bias has been substantially reduced, however, when no communicative cues accompanied the object-hiding actions. The robust association between the ostensive-communicative context of the hiding actions and the perseverative search error supports the “natural pedagogy” hypothesis (9), according to which the perseverative error is in large part due to a pragmatic misinterpretation of the experimenter’s hiding actions as constituting a communicative teaching demonstration rather than being just a hide-and-search interactive game.

Humans are not the only species that show special sensitivity to human ostensive-referential signals. Recent results indicate a functionally similar sensitivity and preference in dogs for certain nonverbal cues of human ostensive and referential communication (12). Unlike great apes (13), dogs exhibit some understanding of human referential intentions expressed in communicative gestures, such as pointing, as shown by their success in solving the so-called object choice tasks (14, 15).

To investigate the functional nature of dogs’ sensitivity to ostensive-referential cues in a comparative manner, we used the A-not-B object search paradigm that had been used to demonstrate the influence of communicative cues on human infants’ perseverative search errors (10). In the first experiment, we tested whether communicative signals would have a notable effect on dogs’ tendency to perseverate in a search task. Three groups of adult dogs (12 in each) participated in a task involving searching for an object that they saw being hidden behind one of two identical screens. In the first phase, the dog was allowed to fetch a toy repeatedly from behind the screen where it was hidden (four A trials). In the test phase, the experimenter repeatedly hid the toy object behind the alternative B screen (three B trials). Each dog participated in one of the following three conditions:

In the social-communicative (SocCom) condition, the hider attracted the dog’s attention by ostensive addressing signals (“[dog’s name] + Watch!”). Then she picked a rubber ball from the floor while establishing eye contact and addressing the dog (“Watch!”) and walked to screen A with the toy in her hand being constantly visible to the dog. As she placed the ball behind screen A, she displayed gaze shifts looking back and forth between the hiding location and the dog (A trials). During the B trials, she passed behind screen A with the ball visibly held in her hand and moved on to screen B, placing the object behind it. After showing her empty hands to the dog, the subject was allowed to choose which screen to go behind to retrieve the toy.

In the noncommunicative (NonCom) condition, the experimenter performed the same object-hiding manipulations as in the SocCom condition with her back turned toward the dog. Thus, neither eye contact nor facial cues were displayed while the experimenter held the object in her hand clearly visible to the subject. In addition, she did not talk to the dog but attracted the dogs’ attention by clapping her hands and making a conspicuous noise with the toy.

In the nonsocial (NonSoc) condition, the experimenter remained still next to the dog while another experimenter, who was invisible to the dog, made the ball move behind the screens by pulling a transparent string (invisible to the dog) to which it was attached. No communicative signals were displayed toward the dogs.

Dogs’ search responses were categorized as correct (selecting the baited location scored 1), ambiguous (score of 0.5) or incorrect (selecting the nonbaited location scored 0) (16). Dogs fetched the object reliably during the A trials in all three conditions (mean percentage of correct choices were 94% in SocCom and 98% in NonCom and NonSoc groups). However, during the B trials we found striking context-dependent differences in the number of dogs committing the A-not-B error (table S1), and a one-way analysis of variance (ANOVA) on the response scores also showed highly significant differences (F2,33 = 10.436, P < 0.001). Post-hoc pairwise comparisons (Tukey–Kramer test) revealed that dogs in the SocCom condition searched at the baited screen (B) less often than those in the NonCom condition (P < 0.05) or in the NonSoc condition (P < 0.001). In addition, dogs in the SocCom condition displayed a search bias toward the empty (A) screen because they performed well below the success rate expected by random search (t11 = 3.576, P = 0.004). In contrast, dogs in the NonSoc condition were significantly more successful than chance during the B trials (t11 = 3.867, P = 0.003) (Fig. 1).

Fig. 1

Scores of correct responses (mean + SE) in the B trials as a function of the hiding context. The dogs (n = 12 for each condition) received four A trials followed by three B trials. The SocCom condition was that the human experimenter repeatedly hid a toy object behind screen A and then behind screen B using ostensive-communicative signals. The NonCom condition was that the experimenter performed the same object-hiding manipulations as in the SocCom but without ostensive-communicative signals. The NonSoc condition was that the experimenter remained still next to the dog while the object moved behind the screens without any perceivable human manipulation. **P < 0.01, one-sample t test, in comparison with the success rate expected by random search (0.5 times three B trials).

These results clearly indicate that, similarly to human infants (10), the communicative context induced in dogs a tendency to perseveratively (and erroneously) search for a hidden object at a previously repeatedly baited location (A) even when they observed the object being hidden at a different location (B). This error, however, has been eliminated when the hiding events were not accompanied by communicative signals. Thus, contrary to previous accounts (17, 18), the perseverative search tendency found in both species cannot be explained as stemming from an inability to locate hidden objects or to inhibit a search response at a previously rewarded location (A). Moreover, if the social-communicative signals simply had a distracting effect, one would expect random search and not an explicit bias to the empty location (A), which we found in the SocCom condition. Therefore, we propose that search error in dogs and infants may be indicative of their shared social competence that involves preparedness for learning from humans through communication.

One intriguing question is whether dogs’ sensitivity to human communicative signals is the evolutionary consequence of domestication. It is increasingly assumed that, during their evolution in an anthropogenic environment (and paralleled by the divergence from the wolf), dogs have become selected to display increased sociality (19, 20), cooperability (21), and communicability (22, 23). This preparedness enables the dog to become sensitized to human communicative cues (12) if the individual is properly socialized to humans (24).

This account predicts that only dogs but not human-reared wolves would respond differentially to communicative versus nonsocial hiding contexts in a search task. We tested this prediction in experiment 2, in which we compared the performance of a different group of naïve pet dogs (n = 12) to that of 10 extensively socialized, hand-reared wolves [for more details on socialization of wolves, see (19-21)] in the SocCom and NonSoc conditions. The procedure was similar to that which was used in experiment 1 except that the bait was a piece of food in a small plastic cup, and dogs and wolves participated in both the communicative and the nonsocial hiding conditions (within-subject design) (16).

Dogs and wolves selected the baited location reliably during the A trials in both conditions (mean percentage of correct choices were 95% and 88% for wolves and 94% and 92% for dogs in the SocCom and NonSoc conditions, respectively). However, the performance during the B trials differed markedly between species and contexts (Fig. 2 and table S2). A two-way ANOVA on the response scores for hiding context and species as factors revealed more correct responses in the NonSoc than in the SocCom condition (F1,20 = 15.003, P = 0.001), more correct responses by the wolves than by the dogs (F1,20 = 4.675, P = 0.043), and a significant interaction between these factors (F1,20 = 13.027, P = 0.002). This interaction was due to the fact that, similar to experiment 1, dogs selected the baited location on B trials less frequently in the SocCom condition (t11 = 5.043, P < 0.001), whereas no such effect was found in wolves (t9 = 0.208, P = 0.840).

Fig. 2

Comparison of dogs and wolves in the A-not-B task. Scores of correct responses (mean + SE) in the B trials as a function of the hiding context. # # # indicates P < 0.0001 in a paired t test.

The robust A-not-B error in the communicative hiding context in dogs, which was absent in extensively socialized wolves, represents a striking interspecies difference, which could be best explained by assuming that selective processes in the course of domestication of dogs led to sensitivity to human ostensive and referential signals (21, 25). However, the fact that dogs, like human infants, commit the perseverative search error when (and only when) the repeated hiding events are presented in a communicative context does not necessarily imply that this effect is mediated by the activation of the same type of interpretive bias that the ostensive cues were hypothesized to trigger in human infants (9, 10). Evidence suggests that dogs’ response to human communication is primarily driven by a motivation to satisfy ostensively cued human imperatives even when the human’s action demonstration conveys an inefficient or mistaken solution to goal approach (26), food choice (27), or object choice (28). These findings, along with the results presented in experiments 1 and 2, raise the question of whether human ostensive and referential signals serve the same communicative functions in dogs that they do in human infants. If human communication is functionally interpreted as imperatives by dogs, it might be tied to the situational context, whereas infants, whose primary motivation is to learn from ostensive demonstrations, would attempt to generalize the communicative content to new situations.

Because one of the crucial components of the A-not-B task is the identity of the person they interact with, in experiment 3 we investigated how dogs and 10-month-old human infants react if, after the A trials, the identity of the hiding person is changed and a new experimenter continues the hiding during the B trials in the SocCom condition. If the ostensive hiding action is interpreted as an imperative order associated with a specific “instructor,” we could expect the perseverative search bias to diminish during the B trials, which would represent a different imperative to act on because it is given by a different person. In contrast, if the ostensive hiding action is interpreted (or misinterpreted) as conveying some generalizable information about the type of object hidden or the function of the hiding location (9) that is not related to the identity of the particular demonstrator, switching the experimenter should not reduce the tendency to commit A-not-B error.

For naïve infants (n = 12), we applied the same procedure used by Topál et al. (10) in the ostensive-communicative hiding context, and the procedure for a group of naïve dogs (n = 12) was similar to that which was used in the SocCom condition of experiment 1. However, after the experimenter had repeatedly hidden the toy in the A trials, she left and another familiar person continued to demonstrate the hiding actions during the B trials (16). During the A trials, dogs fetched the object reliably from behind screen A (mean percentage of correct choices was 98%), showing a performance similar to that found in the SocCom condition of experiment 1 (94%). Infants also searched for the toy correctly in the majority of A trials (82%), replicating the success rate (88%) reported by Topál et al. (10) in the same ostensive-communicative hiding context. However, infants and dogs responded to the new experimenter in the B trials differentially (Fig. 3 and table S3). Infants displayed a perseverative search bias to reach toward location A, and their success rate was significantly below chance level (t11 = 2.932, P = 0.014). In contrast, dogs did not show a significant search bias toward the empty A location (t11 = 0.103, P = 0.920), suggesting that they did not generalize to the new situation in the B trials what they had learned during the A trials.

Fig. 3

Dogs and 10-month-old infants respond differentially to the switch of the experimenter in the ostensive communicative hiding context. Left (white) columns indicate correct responses (mean + SE) with the same experimenter [data from experiment 1 (dogs) and from Topál et al. 2008 (infants)] in B tests. Right (gray) columns indicate the performance during the B trials in the experimenter-switch condition (experiment 3). **P < 0.02, one-sample t test, in comparison with the success rate expected by random search (0.5 times three B trials).

These results show the differential influence of changing a basic stimulus parameter (the identity of experimenter) on dogs’ and human infants’ tendency to commit the perseverative search error. The finding that, as compared with the SocCom condition in Experiment 1, dogs did not perseverate after switching the experimenter is consistent with the hypothesis that dogs anchor communication to the specific situation and especially to the specific communicator who is ostensively addressing it to them. At the same time, and unlike in the NonSoc condition of Experiment 1, dogs were not always successful in finding the object hidden by the new experimenter. Their random search pattern indicates that the unchanged aspects of the situation (such as same object, same room, and same screen) carried sufficient cues of the previous context to confuse them and suggests that it is the overall similarity of the test situation to the training situation that determines whether dogs extend the scope of the learned imperative to the new context. Crucially, the human who ostensively communicates toward the dog forms an indispensable element of the context. For infants, however, it does not seem to matter who performs the ostensive hiding demonstration, and they readily generalize their erroneously learned object-finding action to the new-person context. This result provides further validation for the natural pedagogy hypothesis (9) as an explanatory framework of the Piagetian A-not-B error in infants (10).

In summary, these results show an apparent behavioral analogy between human infants and dogs. In both species, one of the most important causal factors leading to perseverative search errors is the communicative ostensive-referential context. The seemingly mistaken response, called A-not-B error, is not (or at least not only) due to insufficient attentional and motor functioning but paradoxically may be indicative of sophisticated social competence in both dogs and human children. However, the precise function of the cognitive-interpretive mechanisms elicited by communication differs between dogs and humans. For infants, ostensive and referential communicative signals serve a primarily epistemic function by indicating an opportunity to acquire culturally shared knowledge about referent kinds (9). Dogs’ sensitivity to these signals is parasitic on human communication by exploiting them for a different function: to give orders to perform some specific action at a referentially indicated particular location in the presence of (and for) a specific person presenting the imperative.

The lack of a similar sensitivity to human ostensive-communicative signals in extensively socialized wolves supports the view that this is an evolutionary novel skill in the Canis genus, providing a typical case for convergent social evolution (as the consequence of domestication) between man and dog. In addition to human-ape comparisons, the study of the behavioral convergence between dogs and humans (29) offers a comprehensive framework for understanding the evolutionary emergence of human social behavior.

Supporting Online Material

Materials and Methods

Tables S1 to S3


Movies S1 to S6

References and Notes

  1. Materials and methods are available as supporting material on Science Online.
  2. We thank Z. Sümegi, B. Belényi, D. Újváry, M. Tóth, and Á. Kovács for their assistance in data collection. Funded by the Országos Tudományos Kutatási Alap (grant T049615) and the Bolyai Foundation of the Hungarian Academy of Sciences.
View Abstract

Navigate This Article