Technical Comments

Comment on “Differential Sensitivity to Human Communication in Dogs, Wolves, and Human Infants”

See allHide authors and affiliations

Science  09 Jul 2010:
Vol. 329, Issue 5988, pp. 142
DOI: 10.1126/science.1187748


Topál et al. (Reports, 4 September 2009, p. 1269) showed that dogs, like infants but unlike wolves, make perseverative search errors that can be explained by the use of ostensive cues from the experimenter. We suggest that a simpler learning process, local enhancement, can account for errors made by dogs.

Many studies have recently shown the remarkable sociocognitive abilities of dogs (13), including their sophisticated understanding of human communicative signals (47). Topál et al. (8) extended these results by further showing that dogs, like infants, are more likely to commit the so-called A-not-B error when presented with an experimenter using ostensive-communicative cues: In the presence of such cues, dogs commit search errors on 75% of trials, whereas in a comparably social but noncommunicative situation, perseveration falls to 39%. The authors concluded that “in both species, one of the most important causal factors leading to perseverative search errors is the communicative ostensive-referential context” (8). In our view, technical problems in the experimental setup involving dogs call the results and their interpretation into question.

In social-communicative (SocCom) trials described by Topál et al. (8), the experimenter attracted the dog’s attention by using her voice (saying “watch”) together with other ostensive cues (looking at the dog directly) when picking up the toy placed on the ground adjacent to the A barrier. This action occurred in both A and B trials. However, whereas in A trials the toy is then immediately hidden behind the A barrier, in B trials the toy is hidden behind the B screen further away, but the experimenter does not recall the dog’s attention by using her voice before hiding it; rather, she simply holds the object up and looks at the dog. The result is that in both A and B trials, the spot immediately adjacent to the A barrier, where the object is located to begin with, is made more salient by the experimenter’s use of both behavioral and vocal stimuli [see movies S1 to S6 in (8)]. This alone may explain why dogs continue to prefer the A location.

In the noncommunicative (NonCom) trials, a different procedure was followed: The object was again picked up from the spot adjacent to the A barrier, and by squeezing it a sound was produced. The same sound was produced again above the A barrier in A trials, and, crucially, above the B barrier in B trials. Thus, unlike in the SocCom condition, the B location was made salient by both behavioral (holding up the toy) and sound stimuli. In this condition, dogs made significantly fewer A-not-B errors.

Local enhancement refers to a learning mechanism by which a particular location is made more salient to an individual by various means (9), which may involve the presence of a subject, actions, and/or sounds (10). Taking into account the above observations, it is possible that the dogs’ A-not-B errors reported by Topál et al. are a result of having overly enhanced the A location in the B trials of the SocCom condition (by using both actions and sounds only in the A location). In contrast, no such bias occurred in the NonCom trials because both A and B locations were made equally salient (with both actions and sounds).

To test this hypothesis, we carried out a modified version of both the social-communicative trials (mod SocCom) and the noncommunicative trials (mod NonCom). In the mod SocCom condition, we followed Topál et al.’s procedure with one crucial difference: In the B trials, just before hiding the object behind the barrier, the experimenter not only looked at the dog but also said “watch.” If dogs made no errors in B trials, it is likely that previous results could be explained by the fact that the A location throughout testing was made more salient (local enhancement) than the B location by using multiple cues (voice and body gestures). In the mod NonCom condition, we instead chose to examine whether we could “induce” an A-not-B error by using a sound only when picking up the toy in the spot just in front of the A barrier in all trials [thus, we specifically overemphasized the initial location of the object close to the A barrier as was done in the SocCom condition in (8)]. In all other respects, procedures (11) of the mod NonCom condition were identical to the NonCom trials of Topál et al. (8). Confirmation of the role of local enhancement would come from seeing that dogs in the mod NonCom group make significantly more errors than dogs in the mod SocCom group.

Our results (Fig. 1) show that in A trials, 95% of dogs in the mod SocCom and 98% of dogs in the mod NonCom condition chose the correct location. In B trials, in the mod SocCom condition (i.e., with the A and B location equally enhanced), dogs chose the correct location on 98% of trials (compared with 25% in Topál et al.’s SocCom condition). In the mod NonCom condition (where only the A location was enhanced), dogs chose the correct location on 76% of trials. Thus, dogs made significantly more errors in B trials in the mod NonCom condition than in the mod SocCom (Mann-Whitney, N14,14 z = 2.19, P = 0.03).

Fig. 1

Scores of correct responses (mean + SE) in the B trials in the different conditions. The dogs (n = 14 for each condition) received four A trials followed by three B trials. In the mod SocCom condition, the experimenter repeatedly hid a toy object behind screen A and then behind screen B using ostensive-communicative signals. However, in contrast with (8), the same signals were used before hiding in A and B (thus equally enhancing the two locations). In the mod NonCom condition, the experimenter purposely enhanced the A location more (by squeezing the toy before hiding it in that location only), thus mirroring the extra sound enhancement that was given to the A location in the SocCom trials in (8). The difference between the two modified conditions was significant (Mann-Whitney test). *, P < 0.05. #, Data from the SocCom condition and NonCom condition reported in (8) is presented here only for comparison.

Overall, our results suggest that the A-not-B error shown by dogs in the study of Topál et al. (8) may be explained by the fact that the A location was given greater emphasis in all trials by the researcher using both her voice and bodily cues to attract the dogs’ attention. Once the A and B location are enhanced equally, correct responses soar from 25 to 98%.

Procedures followed in studies with human infants did not differentially enhance the two locations (12). It thus seems premature to conclude that “search error in dogs and infants may be indicative of their shared social competence that involves preparedness for learning from humans through communication” (8). Nevertheless, given their long history of domestication (13, 14) and intensive contact with people in their everyday life, we do not exclude that dogs may be particularly sensitive to human communicative cues.

References and Notes

  1. The toy was hidden behind screens with lateral wings; thus, even if the dog chose to walk down the midline, there was no chance of seeing the object until the dog made a definite choice of location. Scores were either 1 (correct choice) or 0 (incorrect choice). Coding of the dog’s choices was carried out by S.M.-P. and a second coder blind to the underlying hypothesis. There was 100% agreement.
  2. This research was supported by funds from the University of Milan to E.P. and a doctoral grant to C.P. We thank all the owners and dogs that participated as volunteers. This research complies with the current Italian laws on animal welfare.
View Abstract

Navigate This Article