In general, the ability to attribute attention to others seems important: it allows an animal to notice the presence of other individuals (whether conspecifics, prey, or predators) as well as important locations or events by following the body orientation or eyegaze of others. We've spent a lot of time here at The Thoughtful Animal thinking about how domestication has allowed dogs to occupy a unique niche in the social lives of humans. They readily understand human communication cues such as eye-gaze and finger-pointing, and capitalize on the infant-caregiver attachment system to have their own needs met. There are several explanations for the emergence of these abilities in dogs:

(1) domestic dogs inherited these abilities from their wolf ancestors;

(2) dogs learn social communication cues by associative learning, simply by sharing physical and social space with humans; and

(3) through the process of domestication, the ability to read human social-communicative cues has emerged due to selection pressures and convergent evolution.

There is pretty solid evidence that dogs far outperform wolves in these tasks, so for now we can safely place that explanation to the side. There is evidence that various species, following intense experience with humans, display improved abilities in reading human social cues, including apes, dolphins, seals, ravens, and parrots. However, in dogs this skill emerges incredibly early in development (before there's been the same sort of intense enculturation that other species require), and does not appear to change significantly with age (remember the thing about finding developmental signatures?) That domestication appears to be the best explanation for the emergence of these abilities in dogs suggests that other domesticated species may show similar skills.

Indeed, there are other animals that have been domesticated, and not just as companion animals or pets. The five classic "barnyard" mammals are all considered domesticated: cows, pigs, sheep, goats, and horses. And horses, in particular, operate within a really interesting framework in which they are both considered work animals as well as, in some cases, companion animals. They don't live in our homes, as dogs do, but neither are they bred for production, like pigs, cows, sheep, and goats.

Whatever their exact role, it seems clear that they would benefit from the ability to comprehend human social communicative cues.

Horses (both domestic and wild) generally communicate visually (though vocal signals also play a role), using body language, and in particular, by small coordinated movements of their heads, ears, and eyes. So, horses have a natural predisposition to detect subtle facial and head-related cues, have been domesticated by humans for several thousands of years, and most individuals are fairly well enculturated. Taken together, this may allow domestic horses to read human communicative gestures particularly well. However, with the exception of the famous case of Clever Hans (which deserves a post of its own), there hasn't been an extensive investigation into the ability of horses to comprehend human social cues - until recently, when several researchers from the Center for Mammal Vocal Communication and Cognition Research at the University of Sussex began investigating horses.

In all, fifty-two horses participated in their experiment, with 36 included in the final analysis, ranging in age from 10 months to 38 years, all living socially in herds, some of whom were occasionally ridden. While the extent of human enculturation was not known for most of the horses, all had at least some exposure to humans.

First, the horses participated in the warm-up phase to introduce them to the general idea and to the experimenters. The subject was held at the release point (R) while the two experimenters approached the horse from the center (C) with their hands out in front of them, so that they could give the horse a food reward. The experimenters then withdrew back to point C, so that the handler could then lead the horse to follow and receive a second food reward. The reward was delivered jointly by the two experimenters (who also rotated positions) so that the horses would not develop a preference for one individual or the other. The idea was to shape a response across 10 trials in which, when released at point R, the horse would approach the experimenters at point C to receive a reward. This was done with attentive experimenters, who faced and made eye contact with the horses, and also with inattentive experimenters, who displayed body and face positions that were (as you might expect) not directed towards the horses.

Figure 1: Experimental set-up.

The test trials included four cued trials (which did not include food rewards), punctuated by reinforcement trials (which included food rewards, to maintain motivation). The experimenters took their positions (E). During each trial, one experimenter was inattentive, and one experimenter was attentive. Experimenters returned to point C, at which time the horses were released from point R to retrieve their reward. Subjects were then led in a figure eight around the test area (either to the right or to the left) before beginning the next trial.

Four different types of cues were tested. The attentive person stood facing forwards and maintained eye contact with the horse as it approached. The inattentive person stood with his back towards the subject, or with his body forwards, but his head turned away, or with body and head forwards, but with eyes closed. During the fourth "mixed" condition (to control for the other three), the attentive person stood with his body turned away from the subjects but his head turned towards the subject to maintain eye contact, while the inattentive person stood with his body forwards but his head turned away.

Figure 2: Still images of the four conditions.

The key variable, as you might expect, was which experimenter the horse approached. The differences between the conditions would allow the experimenters to determine which - if any - of the attentional cues (body position, head position, eye open/closed) the horses used in determining which experimenter to approach. A choice was considered correct if the horse approached and stood within one meter of the attentive experimenter within sixty seconds of release.

The horses chose to approach the attentive person significantly more often than the inattentive person in the body, head, and eye conditions, but not in the fourth "mixed" condition. Scores were not correlated with age, suggesting that younger horses did as well as the older horses, so that can be considered as evidence that the results are not due to intensive enculturation - that there is possibly a genetic component, or that horses are born prepared to quickly learn these cues. Further studies with foals (infant horses) could help to clarify this issue.

Figure 3: Horses chose the attentive individual more than expected by chance in the body, head, and eye conditions.

These results indicate that horses are indeed skilled at reading human attentional cues, and are capable of perceiving both subtle cues (eye gaze) as well as grosser cues (such as body and head orientation) to do so. Since they were equally accurate across all three conditions, it appears as if they are equally able to utilize each different type of social information. The experimenters also collected response time information, and it turned out that response times were longer for incorrect trials, compared to correct trials, for the head and body conditions. They interpret this data such that the horses were less sure of their decision on these occasions. For the eye condition, however, there was no difference in response time. Given the above interpretation for the head and body conditions, this would seem to indicate that the horses did not have any underlying uncertainty.

Figure 4: Response time data for each condition.

I find this interpretation to be highly speculative, though the difference in response time between correct and incorrect trials is intriguing, and particularly the difference in incorrect response time between conditions. The incorrect response time for the head and body conditions, while statistically different from the correct response time, also has also a lot more variance (they don't indicate what the error bars represent, but I'm guessing they are standard error). I'm at a loss to think of alternative explanations, however. What is also particularly interesting is that four of the eight subjects who approached the inattentive person in the body condition moved around to the front of the person, potentially to enter the experimenter's field of vision (again, highly speculative interpretation). They also used additional gestures such as nudging the inattentive experimenters to seek a food reward, suggesting that they may have been trying to get the attention of the experimenter.

One final result was that females were significantly more accurate than males in choosing the attentive experimenter, but all the experimenters provide is more speculation:

...long-lasting bonds between female horses are commonplace and in this sense horse societies have been described as matriarchal (Goodwin 1999; Jensen 2002; Wells and Goldschmidt-Rothschild 1979). This result would suggest that females may possess enhanced social discriminative abilities as has been reported in other matriarchal societies (McComb et al. 2000, 2001). However, recent research has not shown

domestic female horses to be more skilled than males at recognizing close associates (Proops et al. 2009), neither has this sex difference been reported in previous studies of

human attention attribution conducted with other species.

All of these are interesting speculations, but lots more research is necessary to pick apart these hypotheses. It is also unfortunate that the full extent of the horses' experience with humans was unknown. It will also be important to compare domestic horses with their wild cousins. So far, the only species in which these comparisons have been conducted are with dogs/wolves, and foxes.

Although the factors which give rise to this ability are likely to vary across taxa, the high numbers of correct responses on first trials, the lack of correlation between age and performance, and the accurate performance of some very young subjects in this study suggest that horses, like dogs, may have a predisposition to be highly sensitive to human attentional cues.

Despite its few shortcomings, this study, which is among the first conducted for horses, still provides a wealth of information, as well as a handful of new hypotheses. I am continually amazed by the wealth of information - and new questions - that can be generated based on simple behavioral data from within a cleverly designed experiment.

Proops, L., & McComb, K. (2009). Attributing attention: the use of human-given cues by domestic horses (Equus caballus) Animal Cognition, 13 (2), 197-205. DOI: 10.1007/s10071-009-0257-5