Attention and improvement: social learning reveals two behavioural dimensions behind dog-human interactions.

Humans selected hundreds of dog breeds, which show different levels of dependence on human-provided cues. While robust grouping factors, such as genetic relatedness-based ancestry and function-based cooperativeness, denote useful frameworks to analyse behavioural differences in dogs, they may lack fine enough resolution for a more detail-oriented understanding of human-oriented behaviours. Here, we focused on behaviour-based analysis of the characteristic reliance of dog breeds on human behaviour. We utilized N = 187 adult purebred dogs' performance in a social learning task. Dogs observed a human demonstrator performing a detour around a transparent obstacle. With Principal Component Analysis, we established two behavioural dimensions: detour latency improvement and watching the demonstrator. Then, with K-means cluster analysis, we found six categories, based on different combinations of lower/higher values of the two dimensions. The clusters showed significant differences in both behavioural dimensions, highlighting attention combined with improvement types among purebred working dogs. Some extremes showed excellent improvement with only minimal attention to the human demonstrator, while others watched the human without improving their performance. We conclude that by finer analysis, it is possible to set up a dependence-based framework for the understanding of behavioural differences among dogs that go beyond functional breed-selection and ancestry.