EXPRESS: How error rates scale with chance rate among dogs following human pointing gestures, and why it matters

Humans use pointing gestures to communicate with dogs. However, efforts to study how dogs receive this communicative signal have produced ambiguous results. Prior work presenting a choice between two options has shown that dogs choose the pointed-to option at above chance levels, but just barely. With only two options, the high probability of choosing correctly by chance weakens the signal. After replicating the 2-option task, we changed the task’s parameters by presenting dogs with six options. We compared two models that make divergent, quantitative predictions on how the success rates will relate on the two tasks. One model stresses that multiplying the options should reduce the number of correct choices made by chance at a higher rate than instances of genuine communication. A second model that involves no understanding of referents, but just a systematic shift in attention, predicts the effect size to remain constant irrespective of the number of options. Dogs chose the location pointed to at barely above chance rates in both the 2-choice and 6-choice tasks. This pattern discourages the view that dogs understand the referential aspect of human pointing. Additional patterns that become visible with more than two options bolster this interpretation. Dogs chose the correct side of the arena and succeeded on second choices disproportionately more often. Dogs succeeded with far and near locations at similar rates. These results challenge the view that dogs interpret human pointing as referring to a specific object, which has implications for claims that point-following evinces intentional capacities among dogs.