Unwilling or unable? Using three-dimensional tracking to evaluate dogs' reactions to differing human intentions

Abstract

The extent to which dogs (Canis familiaris) as a domesticated species understand human intentions is still a matter of debate. The unwilling-unable paradigm has been developed to examine whether nonhuman animals are sensitive to intentions underlying human actions. In this paradigm, subjects tended to wait longer in the testing area when presented with a human that appeared willing but unable to transfer food to them compared to an unwilling (teasing) human. In the present study, we conducted the unwilling-unable paradigm with dogs using a detailed behavioural analysis based on machine-learning driven three-dimensional tracking. Throughout two preregistered experiments, we found evidence, in line with our prediction, that dogs reacted more impatiently to actions signalling unwillingness to transfer food rather than inability. These differences were consistent through two different samples of pet dogs (total n = 96) and they were evident also in the machine-learning generated three-dimensional tracking data. Our results therefore provide robust evidence that dogs distinguish between similar actions (leading to the same outcome) associated with different intentions. However, their reactions did not lead to any measurable preference for one experimenter over the other in a subsequent transfer phase. We discuss different cognitive mechanisms that might underlie dogs' performance in this paradigm.

Keywords: canine cognition; comparative cognition; goal understanding; intentions; theory of mind.

Figure 1.

Experimental setup. (a) The experimenter gives food reward to the dog through a hole in the polycarbonate panel during a motivation trial. (b) The holes in the panel are covered by an additional polycarbonate panel during the blocked condition. Informed consent was obtained for publication of identifying images (photo credits: Thomas Suchanek/Vetmeduni).

Figure 2.

Dot and box plots of dogs' performance in the test trials. (a) Proportion of time away from experimenter, (b) Proportion of trials looking away from the experimenter, (c) Proportion of time sitting or lying down, (d) Proportion of time lateral approach response. The dots represent individual mean values; the grey lines connect the values of the same individuals across conditions. On the right side of each graph, a box plot is shown (blue: blocked condition; green: teasing condition; orange: clumsy condition).

Figure 3.

(a) Plot showing the dogs (n = 48) roaming pattern across the four conditions (based on the head-centre keypoint). The purple dots indicate areas visited by the dogs; darker purple areas were visited more frequently. The small, grey square indicates the location of the caregiver's chair, the large, yellow square the location of the fenced area (where the experimenter offered food). The black rectangle indicates the interest area around the caregiver's location. The red concentric lines show the contours of a two-dimensional kernel density estimation highlighting frequently visited areas. (b–d) Dot and box plot of dogs' performance in the test trials based on the three-dimensional tracking data. (b) Mean proportion of the room area visited. (c) Mean proportion of trials in which the caregiver interest area was visited. (d) The angle of the tail relative to the tail-base–head-centre axis (positive values: rightward deflection, negative values: leftward deflection). The dots represent individual mean values; the grey lines connect the values of the same individuals across conditions. On the right of each plot, a box plot shows the median and the quartiles (blue: blocked condition; green: teasing condition; orange: clumsy condition).