. . . well, at least one dog: Chaser, a female border collie born in 2004. Because of their marked inferiority to felids (the King of Pets), I don’t usually feature goggies on this website. But this bit of research, published in Behavioural Processes, was too good to pass up.
When I lived in Scotland, the one television show I never missed was the BBC’s “One Man and His Dog,” in which border collies and their owners would vie for a prize in sheepherding. Despite my indifference to dogs, I was fascinated at the skill with which these dogs herded errant packs of sheep using commands from their owners. I was sad to hear that the show was canceled, though Wikipedia says it’s still alive. (UK readers: is it?)
Border collies are clearly alert and intelligent beasts, and this new paper demonstrates it, showing that they have a stunning ability to learn and (supposedly) to combine nouns and commands, an ability to recognize that objects have names, and a talent for distinguishing different commands about how to deal with those objects.
The paper, bearing the turgid title of “Border collie comprehends object names as verbal referents” (free online, and you can see a summary/press release here), is by John Pilley and Alliston Reid, two psychologists at Wofford College in Spartanburg, South Carolina. I don’t want to produce a long summary, for the paper is well written and easily comprehensible to laypeople. Further, you can read the press release for a shorter take. I’ll just discuss the salient points here. Do consult the paper if you’re worried about controls, etc., since the experiments did seem well controlled.
Also, the link to the paper will take you to four videos (on the right) that you can play to see Chaser’s talents for yourself.
Over a period of three years, Pilley and Reid trained Chaser to recognize various objects: toys, stuffed animals, plastic items, etc., by telling the dog to “go to” that item and fetch it. (They eliminated the “clever Hans” effect by having the owner give orders when out of sight of the dog.) Once Chaser had learned to fetch a number of these items, they did further experiments. There were four in total.
- First, the learning of names. Chaser’s ability here was astounding: at the end of the training period, she had learned the names of 1,022 objects, and was able to reproduce them faithfully, almost without error. In one series of tests, for example, a group of 20 of the 1022 objects would be dispersed randomly on the floor. Chaser was then asked to select one item out of the 20. Then he would be asked to select another without replacement (order random, of course), and then another, until all 20 were gone. This was done in more than 50 successive trials, until all 1000-odd objects had been used. This meta-test often took many hours.
Amazingly, in no test—and there were many of them—did Chaser make more than two mistakes (in other words, she always got at least 18 objects correct). And she retained this ability to remember names for at least two years after training, as shown by retesting when she was five.
Here are some of the objects Chaser learned, with their names on the left (click to enlarge):
- Second, understanding new combinations of different words. What seems to be the Big Result of the paper, but one that doesn’t completely convince me, is the authors’ contention that Chaser “understood the separate meanings of proper-noun names and commands.” What they did was first train Chaser to perform three actions, apparently using objects that were not part of her previously-learned repertoire. These commands were “take” (i.e., fetch), “paw,” and “nose.” Then, once the commands were learned, Chaser was given combination commands using only three of the 1,022 objects combined with a requested action. For example, “take lamb,” or “nose lips” (this was an object resembling human lips), or “paw ABC” (a cloth cube with those letters on it). Note that the dog had learned the commands and the objects separately, and had never been given a directive that combined them. This was her first exposure to the two-word commands.
There were 14 trials (see Table 1 of the paper), and Chaser did the right thing all 14 times. The cumulative probability that this would happen by chance alone is 0.000000000000044. It would have been nice to do this with all 1,022 objects to get a better judgment on Chaser’s “combinatorial” abilities, but this is still telling. Chaser was obviously able to combine an action command with a noun command, demonstrating (to the authors) that she has “combinatorial understanding.” As the authors say, “She responded as though the commands and the proper-noun names were independent entities or morphemes Thus, in effect, Chaser treated phrases like ‘fetch sock’ as though the ‘sock was a sock and not a ‘fetch sock’—indicting [sic?] that her nouns referred to objects.”
This does demonstrate combinatorial abilities, which is pretty remarkable. My only quibble is whether Chaser understood the term “sock” as a noun by itself, or rather as a combination noun, “go to [i.e., “fetch”] sock”, which is the way she learned “sock”; and that this combination-noun was overriden by a third term, like “nose,” that preceded the object. In other words, while I’m convinced of Chaser’s combinatorial abilities, I’m not convinced that she learned that the objects really were nouns that weren’t attached to verb commands.
- Third, learning different noun categories. Chaser was trained to recognize not just the 1,022 names of objects, but also their classification under the rubric “toys.” She was also taught to distinguish these from a large number of other objects that she had not been allowed to play with. After training, 8 toys and 8 non-toys were strewn about and, with the investigator hidden, Chaser was asked to “fetch a toy.” She was then asked again, and the objects were not replaced after being fetched. She performed perfectly.
Beyond this broad category, Chaser was also trained to recognize those 116 objects that were balls under the name of “ball.” She was likewise trained to recognize 26 of her disk-like toys under the name of “frisbee.” She was again tested with 8 balls and 8 “non-balls”, and also with 8 frisbees and 8 “non-frisbees” (i.e., “fetch a frisbee”). In both cases she performed perfectly, even though she knew each of those objects not only as “ball” of “frisbee”, but also by their unique name and the general name of “toy.” Here is a sample trial showing 8 frisbees and 8 non-frisbees:
- Fourth, learning words by exclusion. In the last experiment, Chaser was asked to retrieve novel objects with novel names. There were 64 of these novelties, which differed from the 1022 objects whose names she had already learned. One of the novelties was placed with seven familiar objects. In the first two commands, Chaser was asked to fetch a familiar object. In the third, she was asked to bring a novel object with a novel name, one she hadn’t heard before. This forced her to discriminate by exclusion. She was successful eight times out of eight. Remarkable! However, this ability to remember novelty decayed quicky: in other tests, conducted immediately after the successful novelty trial, then ten minutes later, then 24 hours later, she was often unable to pick the novel named item out of a group of four novel and four familiar items. After 24 hours, her memory for the novel ones had decayed completely.
I think even caninophiles would be surprised by Chaser’s talents, which, of course, could probably be seen in other border collies—though not necessarily other dog breeds. Her abilities far exceeded anything necessary or useful in ancestral canids—dogs, after all, don’t have to remember names in the wild, though they certainly do have to recognize different conspecific individuals (but not 1022 of them!). Part of her performance is probably due to a co-option of brainpower used for other things (just like humans can learn to read music using neurons evolved for other reasons), and part to the fact that border collies are trained to recognize different commands. I’m not an expert on dogs, and I bet there are border-collie owners among the readers, so by all means recount your experiences or theories about the dogs.
Of course, any random cat could do exactly what Chaser did—and much more. It just wouldn’t want to!
h/t: Matthew Cobb
Pilley, J. W., and A. K. Reid. 201o. Border collie comprehends object names as verbal referents. Behavioural Processes, in press. doi:10.1016/j.beproc.2010.11.007