Dogs with prior experience of a task still overimitate their caregiver

Abstract

Domestic dogs have been shown to copy their caregiver's actions, including ones which are causally-irrelevant to a physical goal-a behaviour called "overimitation". In a new overimitation task with a non-food reward, this study investigated "causal misunderstanding"-falsely assuming causally-irrelevant actions to have functional relevancy-as an explanation for dog overimitation (N = 81). By providing dogs with prior experience of the task to learn about the consequences of its irrelevant box-stepping and relevant bucket-opening action to obtain a toy-ball, we tested whether and when dogs would copy their caregiver's irrelevant-action demonstrations. Dogs with and without prior experience were compared to a third (control) group of dogs, who had neither prior experience nor caregiver demonstrations of the task. Results revealed that the timing of overimitation, rather than its frequency, was closely related to dogs' prior experience: dogs with prior experience attended to their reward first, then interacted with the irrelevant box later ("post-goal overimitation"), while dogs without prior experience first interacted with the irrelevant box ("pre-goal overimitation"). Our results suggest that, when action consequences are understood, dogs are overimitating for a secondary social goal that is clearly distinct from the task goal of obtaining a physical reward.

Keywords: Causal understanding; Domestic dogs; Overimitation; Prior experience; Social learning.

Fig. 1

The box-stepping overimitation task. The left picture depicts the task and its starting positions for the caregiver and dog (birds-eye view). The right pictures are the task’s actions: (a) causally-irrelevant box-stepping (before), (b) causally-relevant bucket-opening to obtain the ball reward, and (c) causally-irrelevant box-stepping (after). Example perfect scores of "4" for each action are depicted by a model caregiver and a model dog (Filou, male).

Fig. 2

The final IRR mixed model’s results: (a) the irrelevant-action scores per condition, with circle size depicting scoring frequency (i.e., larger circle = larger number of scores at that level), and the model’s confidence interval fitted values and upper/lower limits as line segments and bars; (b) the irrelevant-action score’s estimated marginal means for the interaction (condition:timing), calculated from the model are also shown, with ‘***’ representing a significant p-value of < .001.

Fig. 3

Baseline (control group) comparison barplots for: (a) irrelevant-action scores, (b) relevant-action scores. The horizontal lines denote a statistically significant t-test comparison (Bonferroni corrected) between groups, with ‘***’ representing a p-value < .001 and ‘**’ representing a p-value < .01.