Cognitive Psychology

Dogs can be taught new words in the same way as small children do

A 2004 research study by Kaminski et al. reported the case of Rico, a 9-year-old border collie that could recognize more than 200 items by their names. In two experiments, the dog demonstrated genuine acquisition of names (any non-verbal cues that could contribute to a successful retrieval of toys were eliminated) and exclusion learning (ability to identify a novel object among familiar ones by its unfamiliar name). Similar abilities to learn new words by exclusion were found by previous researchers to be present in normally developing 3 and 4-year-old children (Carey & Bartlett, 1978) and infants (Baldwin, 1993). However, the actual degree of similarity of Rico’s performance to that of small children was questioned by several other scholars. Markman and Abelev (2004) pointed out that the reward offered to Rico after a correct exclusion choice could have influenced the test’s outcome. Bloom (2004) doubted Rico’s ability to link words to objects in the same way as small children do. The dog could lack understanding that the word “sock” refers to a material item and perceive it simply as a command to fetch this sock. In this case, Rico’s language acquisition process could not be compared to that of small children as it was too different in nature, although outwardly similar.

The current research study by Pilley and Reid (2011) was conducted to address the concerns expressed by critiques of Rico’s experiment. The first goal was to find out whether a border collie could acquire a vocabulary of 1,000 words instead of 200, under the condition of intensive training. The second goal was to test the collie’s ability to distinguish between a word meaning an object and a command referring to that object. The third goal was to check whether the dog could learn several common nouns. The fourth goal was to measure the collie’s ability to inferential reasoning – identifying the name of a novel object by logical exclusion of familiar alternatives. Overall, Pilley and Reid (2011) aimed at answering the questions raised by a previous study of border collie language acquisition by Kaminski et al.

In their research study, Pilley and Reid (2011) tested four variables: a border collie’s ability to learn proper nouns, ability to learn independent meanings of nouns and commands, ability to learn common nouns, and ability to learn words by exclusion. A separate experiment was conducted for testing each variable.

The research hypothesis for the first experiment was that Chaser, a female border collie that was acquired by the authors at the age of 8 weeks and exhibited the usual characteristics of her breed, would be able to recognize at least 1,000 objects by names after 3 years of training. The research hypothesis for the second experiment was that Chaser would be able to recognize the name of an item independently from a command that involved that item. The research hypothesis for the third experiment was that Chaser would be able to learn three common nouns that stood for different categories of her toys, the proper names of which she already knew. The research hypothesis for the fourth experiment was that Chaser would be able to retrieve a novel object that was placed among familiar objects by its novel name.

In the first experiment, Chaser was required to identify and retrieve 50 random objects from among 1022 ones, the names of which she was taught in the preceding 3-year training. The experiment took place in a college auditorium, with 100 psychology students present as observers. All 1022 objects were placed randomly on the floor. The items Chaser had to bring were selected, according to a double-blind test methodology, by five students. Each student selected ten objects and altered their location on the floor as he or she wished, out of the experimenter’s sight. Next, each student wrote the names of chosen objects on paper and gave the sheet to the experimenter. Collecting all five lists, the experimenter called the names aloud in the order in which they had been written. After a name was called, Chaser had to find and bring a respective item. Prior to this public demonstration, the collie passed 145 formal monthly tests on object retrieval by name. The test results were used as an additional measure of Chaser’s ability to understand words as verbal references.

The second experiment consisted of 14 trials, in each of which the trainer gave Chaser a different combination of a command and an item name. There were three commands and three object names used in total. None of these noun-command pairs was ever presented to the collie before. To ensure a random choice, the experimenters assigned a specific number to each word and used a table of random numbers to select a command and an object for each trial.

In the third experiment, Chaser had to retrieve eight objects of a certain kind from a random array of 16 items, eight of which actually belonged to this category (exemplars) while other eight did not (non-exemplars). A separate testing session was conducted for each of three categories: “toy,” “ball,” and “Frisbee.” A random placement of exemplars and non-exemplars was created in an adjacent room out of the reach of the trainer so she could not provide any clues influencing the collie’s choice. During each session, Chaser was asked eight times to retrieve an exemplar, one at a time (e. g. “bring a ball” or “bring another ball”). None of the items used in the “toy” and “ball” testing sessions were available to the collie during pre-experiment generalization training, while some of the items used in the “Frisbee” session were.

The fourth experiment used a four-step procedure to test Chaser’s baseline novelty preference, inferential reasoning, and name retention. At the first step, Chaser was presented with a random array of eight familiar objects and two novel ones, which were not used in any previous sessions and the names of which she was not taught. The collie was asked to retrieve one familiar object by name until only the novel ones remained in the room. At the second step, the array included seven familiar and one novel object. The experimenter twice called the name of a familiar item, and then the novel name of a new object for Chaser to bring. Immediately, a retention test followed where the novel item from the previous stage was placed among three novel and four familiar items, all arranged randomly. Chaser had to retrieve two familiar objects by name, and then the novel object by its newly learnt name. This retention test was repeated twice (after 10 minutes and after 24 hours). In total, this four-step procedure was repeated eight times over eight days, each replication using different items.

The results obtained in this research study supported all four initial hypotheses by the authors

In the first experiment, Chaser correctly retrieved 46 objects out of 50, or 92% of the total number, with no less than 8 correctly identified objects in each set of 10. In all 145 formal tests preceding the experiment, the collie also demonstrated high accuracy, correctly retrieving no less than 18 items out of 20 regardless of the presence or absence of the trainer. In the second experiment, Chaser got the commands right in all 14 trials, clearly demonstrating her perception of noun and verb as separate semantic units. In the third experiment, the collie retrieved “toy,” “ball,” and “Frisbee” objects without an error in each trial. In the fourth experiment, Chaser made correct choices in all inferential reasoning and immediate retention tests. Meanwhile, after a 10-minute delay the collie could identify correctly only five out of eight novel objects, and after 24 hours this performance rate dropped to one out of eight. Overall, the results of Pilley and Reid’s experiments with Chaser provide strong evidence that border collies can learn the meanings of words and map them upon the referents in a way much similar to that that is normally found in small children.

The findings of Pilley and Reid (2011) can be used by dog owners who want to teach their pets to understand more advanced commands. For example, a dog can be taught to bring each family member his or her own slippers, provided that every pair of slippers is given its unique name and humans use it every time when asking the dog to bring this item. The research study demonstrates that a border collie can learn up to 1,000 names in total. Another possible practical implication is in circus performances, to produce a more spectacular effect. A dog can be taught to perform various activities, such as nosing, pawing, or biting, with each of several different items at the trainer’s request with 100% accuracy, as Chaser did in the second experiment.

This article expands on the conceptions of retention and reasoning. Firstly, this research study demonstrates the phenomenon of memory retention weakening over time in dogs as well as in humans. While the collie identified accurately 100% of novel objects right after she had been taught their names, the next day she could retrieve correctly only 12%. Secondly, the third experiment clearly demonstrates that dogs, and collies in particular, are capable of understanding set-inclusion relations – the inferred relationships among categories of objects (Revlin, 2012, p. 409). Chaser correctly determined whether an object belongs to a particular set or not in each of 24 trials.

Future research in this area can be conducted on dogs of other breeds to test whether they possess the same language acquisition skills as border collies. Besides, it might be interesting to begin a similar item recognition training with an older collie, to see to which degree the results are mediated by a subject dog’s age. Chaser was only 8 weeks when she started training and the last test was conducted when she was 5 years old, while Rico from an earlier experiment was 9 years old. Chaser’s ability to learn to recognize 5 times more words than Rico could result from the differences in their age as well as from those in their training processes, and future research is needed to determine the relative importance of both these factors.