Dogs Rely On Visual Cues Rather Than On Effector-Specific Movement Representations to Predict Human Action Targets

The ability to predict others’ actions is one of the main pillars of social cognition. We investigated the processes underlying this ability by pitting motor representations of the observed movements against visual familiarity. In two pre-registered eye-tracking experiments, we measured the gaze arrival times of 16 dogs (Canis familiaris) who observed videos of a human or a conspecific executing the same goal-directed actions. On the first trial, when the human agent performed human-typical movements outside dogs’ specific motor repertoire, dogs’ gaze arrived at the target object anticipatorily (i.e., before the human touched the target object). When the agent was a conspecific, dogs’ gaze arrived to the target object reactively (i.e., upon or after touch). When the human agent performed unusual movements more closely related to the dogs’ motor possibilities (e.g., crawling instead of walking), dogs’ gaze arrival times were intermediate between the other two conditions. In a replication experiment, with slightly different stimuli, dogs’ looks to the target object were neither significantly predictive nor reactive, irrespective of the agent. However, when including looks at the target object that were not preceded by looks to the agents, on average dogs looked anticipatorily and sooner at the human agent’s action target than at the conspecific’s. Looking times and pupil size analyses suggest that the dogs’ attention was captured more by the dog agent. These results suggest that visual familiarity with the observed action and saliency of the agent had a stronger influence on the dogs’ looking behaviour than effector-specific movement representations in anticipating action targets.