Posts

The most common question I get at events like parties, when I tell people that I train parrots to communicate with humans in English, is, “But do they really understand what they are saying?”

I have 40-plus years of data and probably over a hundred publications to demonstrate that such is indeed the case. Sometimes, however, it is the anecdote — the once-in-a-blue-moon, not-possible-to-reproduce, I-can’t-believe-that-just-happened-incident — that transforms the general public into believers. These incidents can never hold up to scientific scrutiny, and at most can be the impetus to begin a serious scientific study, but they are consistently interesting and valuable.

Alex Chimes In

It is always exciting when the incident does lead to a series of experiments. Such was the case when we were training Griffin to respond to auditory numbers — sets of clicks. At the time, we made sure that the inter-click intervals were of random lengths and that the lengths of the sounds constituting the clicks were also random so that Griffin could not merely respond to the amount of sound or stimulus duration, but rather had to attend to the exact number.

We were working with “two,” and Griffin simply was not interested. I had clicked twice, asked, “How many?” and Griffin ignored me. I started over: “Griffin, pay attention! Listen! (two clicks) How many?” Again, Griffin continued to preen. Behind me, Alex, from his cage, piped up with a very clear FOUR. I told him he was wrong, that the answer was TWO, and gave Griffin one more chance. Alex then very clearly produced SIH (his label at the time for “six”).

DUH…I realized that Alex was adding the clicks! We quickly designed a study, based on work my colleague Sally Boysen had done with her apes (Boysen & Berntson, 1989), and showed that Alex could indeed sum small quantities of objects, and then Arabic numerals (Pepperberg, 2006, 2012).

Without any training, Alex also transferred his use of the label “none” from indicating a lack of similarity or difference in a pair of objects to the absence of a specific quantity. The details are in Pepperberg & Gordon (2005); basically, he manipulated me into asking him to identify a null set — so that he could use “none”— and that led to our study of his understanding of a zero-like concept.

Choosing What To Follow Up

Sometimes we intended to follow up on an incident, as with Alex’s famous, untrained, and very unexpected spelling out of N-U-T during a demonstration of what was supposed to be a simple identification of sound-letter associations (Pepperberg, 2008). Unfortunately, we never got around to testing him on other possible combinations.

And sometimes the incident is definitely something that we do not care to replicate, as when a student forgot to give Athena fresh water one morning…and she very clearly and, apparently intentionally, uttered WATER. Water deprivation is clearly a situation that we certainly do not want to happen in our lab deliberately! Athena has never again said WATER, even when she is trying to get us to do something like extend her shower. She had, however, made her point.

Griffin Makes His Thoughts Clear

Recently, Griffin added to our anecdote list. Suzanne, our long-time lab manager, who is also a post-doctoral fellow with us this year, was giving the birds a break from sessions. She had Griffin on one hand, was tickling his head with the other, but wasn’t giving him her full attention as she was also trying to read a journal article. Griffin kept saying TICKLE; she would stop reading briefly, completely attend to his needs for a short while, then go back to reading. He kept repeating his request. After a few such repetitions, he blurted out an extremely loud PAPER while he reached over, grabbed the article, threw it to the floor, and then stared at her directly and demanded TICKLE! It was quite clear what he thought of her behavior!

As I said at the outset, I’ve spent over 40 years collecting data and publishing papers to convince my scientific colleagues that my parrots really do have symbolic reference. Formally, the definition is that it “involves semantic and pragmatic use of noniconic symbols — be they auditory/vocal, manual, or lexical — to stand for (but not be limited to) items such as physical objects and their attributes, various concepts, relations among these items and concepts, actions that can be done to or with these items, and comments about these items (e.g., relating to past/ future/hypothetical issues).” (Pepperberg, 2021).

But sometimes it is the simple anecdote that helps people understand the intelligence of these parrots.

Glossary

Semantic: Involving the meanings of symbols

Pragmatic: Involving practical use

Noniconic: An arbitrary relationship. For example, an iconic representation of a dog by a child is the use of the term “bowwow”; noniconic representations are the labels “dog,” “chien,” etc. that have nothing canine-like about them

Lexical: Generally refers to something written

References

Boysen, S. T., & Berntson, G. G. (1989). Numerical competence in a chimpanzee (Pan troglodytes). Journal of Comparative Psychology, 103, 23–31.

Pepperberg, I. M. (2006). Grey parrot (Psittacus erithacus) numerical abilities: addition and further experiments on a zero-like concept. Journal of Comparative Psychology, 120, 1–11.

Pepperberg, I.M. (2008). Alex & Me. New York: Harper-Collins.

Pepperberg, I. M. (2012). Further evidence for addition and numerical competence by a Grey parrot (Psittacus erithacus). Animal Cognition, 15, 711–717.

Pepperberg, I.M. (2021). Nonhuman and nonhuman-human communication: Some issues and questions. Frontiers in Psychology, 12:647841. Doi: 103389/fpsyg.2021.647841.

Pepperberg, I. M., & Gordon, J. D. (2005). Numerical comprehension by a Grey parrot (Psittacus erithacus), including a zero-like concept. Journal of Comparative Psychology, 119, 197–209.

A journal article, published this August (Schnell et al., 2021), showed that a species of European jay demonstrates clear distress when their expectations of finding a treat are violated. The experimenters showed the birds that a very desirable food was being hidden under one of two cups, and then, without the birds knowing, the experimenters swapped it for something the birds didn’t like nearly as much. Not all birds reacted strongly, but those that did check under other cups, or squawked and flew away, and usually ignored the less-desired food they had found.

Parrots Dislike Trickery Too!

I wasn’t at all surprised by their results, having observed the same behavior in my parrots (Pepperberg et al., 1997) decades ago! We performed the study as part of an experiment on object permanence—the notion that an object does not cease to exist (is ‘permanent’) even if it is out of sight. The very first studies of this type were performed by a Swiss psychologist, Jean Piaget, who was interested in studying the development of various types of intelligence in children. He started testing babies when they were only a few months old and kept presenting more and more complicated types of tasks as the children got older (Piaget, 1936/1952).

The easiest task just checked to see if the baby could track the object as it was moved around; then the tasks got harder: The object was partially hidden under something like a blanket, then fully hidden; next came tracking when two possible hiding places existed, then three, and so forth. Other researchers (e.g., Uzgaris & Hunt, 1975) looked at even more complicated tasks—actions like hiding the treat in the experimenter’s hand and then moving the hand among several cups before hiding the treat under one cup and showing an empty hand.

When we did the study in question, Alex had already passed all the Uzgiris and Hunt tasks (Pepperberg & Kozak, 1986); here we wanted to track Griffin from a very early age to see if his development paralleled that of children (it did, but he progressed more quickly)—plus, our concern was that our parrots might be smart enough to track something of interest, but not necessarily remember what it was that they were tracking.

One way to check this possibility was to do the ‘treat substitution’ trick. Other researchers (e.g., Tinklepaugh, 1928, 1932) had examined how nonhuman primates reacted to being tricked with “reat substitution,” so we thought that would be an interesting task for our birds. And they reacted similarly—Griffin turned over the cup under which he thought he would find a nut, didn’t find the expected cashew, then turned over all the other cups and, when that didn’t work, ran over to the experimenters!

Alex reacted similarly, but also stared at the experimenters with slitted eyes—an aggressive stance—and added some emphatic banging of his beak on the table, which for him was a signal of displeasure and frustration. At the time we did these experiments, we couldn’t videotape each trial with a tiny camera or a mobile phone as we do now (the experiment was actually done in 1996!), so we don’t have a cinematic record—just our notes (thus no pretty pictures this month).

Of course, this behavior makes a lot of sense when one thinks about the lives of nonhumans that are usually preyed upon in the wild—they form expectations of where food should be over time, and if it is not there, they will have wasted a lot of precious foraging time and effort and possibly exposed themselves to predators for naught. And, of course, if they take the time to break open a particular type of nutshell, they definitely expect to find that type of nut! In general, novelty in the wild is more likely to indicate some form of danger (e.g., an unexpected shadow could be a hawk looking for a meal) rather than something fun, which is why most parrots are somewhat neophobic (i.e., shy away from new objects or situations, and dislike surprises).

Thus, although these findings in parrots and jays may not have violated our expectations, as scientists we are always interested in testing our hypotheses. And we are always excited when we find data to show that those hypotheses were correct!

References

Pepperberg, I.M. & Kozak, F.A. (1986). Object permanence in the African Grey parrot (Psittacus erithacus). Anim. Learn. Behav. 14:322‑330.

Pepperberg, I.M., Willner, M.R., & Gravitz, L.B. (1997). Development of Piagetian object permanence in a Grey parrot (Psittacus erithacus). J. Comp. Psychol. 111:63–75.

Piaget, J. (1952). The origins of intelligence in children (M. Cook, Trans.). New York: International Universities Press. (Original work published 1936)

Schnell, A.K., Maria Loconsole, M., Garcia-Pelegrin, E., Clive Wilkins, C., & Nicola S. Clayton, N.S. (2021). Jays are sensitive to cognitive illusions. R. Soc. Open Sci. 8: 202358. https://doi.org/10.1098/rsos.202358

Tinklepaugh, O. T. (1928). An experimental study of representative factors in monkeys. J. Comp. Psychol. 8:197-236.

Tinklepaugh, O. T. (1932). Multiple delayed reaction with chimpanzees and monkeys. J. Comp. Psychol. 13:207-243.

Uzgiris, I. C., & Hunt, J. (1975). Assessment in infancy: Ordinal scales of psychological development. Champaign: University of Illinois Press.

Our model-rival (M/R) technique is almost always successful in training our parrots to acquire novel labels for objects, actions, attributes, and concepts. (Of course, whether they will subsequently use the labels is a matter of choice: Alex did so constantly, whereas Athena will talk only when absolutely necessary—e.g., when she knows Griffin will get all the green beans unless she pipes up!) When success is elusive, however, some theories and experiments with children often explain the reasons. An interesting example was when Griffin, unlike Alex, couldn’t seem to learn to use his color labels to describe objects. The issue was something called mutual exclusivity.

Mutual Exclusivity in Children

The term describes the behavior of children who, in the process of learning language, may persist in exclusively using the first label that they learn for a given object, and thus have difficulty acquiring additional labels for this item (Liittschwager & Markman, 1991, 1994). They function under the assumption that every object has a label—and only one. Having learned, for example, to refer to their pet as a doggie, such children will often refuse to use any other label for that creature; that is, they will actually insist, “NOT animal, doggie” (see Merriman, 1991). Specifically, they have problems understanding how categories can overlap. Although this may seem like a serious issue, at that stage in acquisition, it actually helps them learn new labels. Thus, given two objects, one whose label is known (e.g., block) and another whose label is not (e.g., tongs), a child using mutual exclusivity will, after hearing the novel label tongs, exclude it from the familiar, already labeled item and assign this new label to the new, unlabeled item (e.g., Tomasello & Akhtar, 1995).

This process is also sometimes called fast mapping (see Carey, 1978), although it often takes the child a while to use the new label appropriately. Eventually, as children develop additional language skills this process is extended to other types of label acquisition (see Markman, 1994); for example, if children are given two balls—that they easily label ball—one red and one an odd greenish-yellow color, and are told to pick up the chartreuse thing, they quickly figure out that the new weird label is a color.

A Parrot’s Grasp of Mutual Exclusivity

These findings made us think about the slightly different ways in which African greys Alex and Griffin were taught their object and color labels, and how mutual exclusivity could explain the differences in their subsequent behavior. First, let’s look at Alex. We started with objects that had no real color—for example, white strips of paper, pieces of wood like tongue depressors, silvery keys. He learned to identify them respectively, as “paper,” “wood,” and “key.” A clothespin was labeled “peg wood.” After he could accurately label these objects, as well as “hide” (pieces of rawhide chew toys), we introduced the color “green,” using it not as an alternative label, but as an additional label—so we modeled “green wood” and “green key.”

If the object had no color, he was to respond “wood” or “key,” but if it were colored, he had to produce both labels, “green wood” or “green key.” We used two different objects so that he would recognize that the attribute and label could be used with multiple items. It took some time for him to learn to produce the new sounds (although he had the “eee” from “key”), but he then quickly learned to add the label when asked about the newly tinted items and even transferred the label fairly quickly to “hide”: Although he initially called it “hide,” he attained an accuracy of 60% correct (“green hide”) after only 5 min with the M/R procedure. That is, out of seven presentations immediately following training, he correctly labeled the object “green hide” four times and incorrectly labeled it “hide” three times (Pepperberg, 1981). His accuracy improved rapidly, and he subsequently learned other color names, also as additional labels, without any problem. He did not engage in mutual exclusivity.

Griffin, however, was taught very differently (Pepperberg & Wilcox, 2000). Griffin’s experiences with objects were designed to resemble more closely those of small children. Instead of the uncolored items Alex first encountered, Griffin’s objects—wool pompons, bits of paper, toy cars (called truck), and so forth—varied in color and size; only corks were uncolored. Thus, he had a more complicated task than Alex had been given. For example, during initial training, he had to recognize that truck was the common term to describe variously colored and shaped toy cars.

His color training also differed from Alex’s. He was first trained on color labels (rose, green, yellow, orange, blue) by two humans using a computer display (colored circles on a white background) as part of a separate study on how the inability to interact with physical objects might affect label acquisition. And, unlike Alex, he was not required to produce both color and object labels for a colored item; rather, he had to respond with a single label to “What color?” versus “What toy/matter?” The goal was to see whether these differences affected his pattern of learning and eventual success. Once he had acquired several color and object labels, we then queried him about the colors of objects he could already label (“wool” pompons) and items he was simultaneously learning to label (a plastic “cup”). Might he use labels differently in the two cases? And if he did, would he exhibit mutual exclusivity. That is, refuse to use the color label for the object he already knew as “wool” but not for the one he could not quite yet label as “cup”?

And that is exactly what he initially did (Pepperberg & Wilcox, 2000). Although he eventually did learn to use color labels (and later shape labels) appropriately for objects for which he had initially learned material or toy labels, he took many sessions and considerable extra training to do so. The reverse was true for objects for which he first learned color labels, although, again, he eventually learned their object labels as well. What is also of interest is that children who are taught multiple labels are a bit more like the way in which Alex was taught, where multiple labels are introduced in a complementary manner (“This is a kitty; a kitty is a kind of animal” do not exhibit mutual exclusivity (see Gottfried & Tonks, 1996).

Many reasons have been proposed for why any organism might engage in mutual exclusivity (see references in Pepperberg & Wilcox, 2000). The interesting point is that both parrots and humans, separated by over 300 million years of evolution, can exhibit very similar learning strategies!

References

Carey, S. (1978). The child as word learner. In Linguistic Theory and Psychological Reality; ed. M. Halle, J. Bresnan, and A. Miller, pp. 264-293. Cambridge, Mass.: MIT Press.

Gottfried, G. M., & Tonks, J. M. (1996). Specifying the relation between novel and known: Input affects the acquisition of novel color terms. Child Development, 67, 850-866.

Liittschwager, J. C., & Markman, E. M. (1991, April). Mutual exclusivity as a default assumption in second label learning. Paper presented at the biennial meeting of the Society for Research in Child Development, Seattle, WA.

Liittschwager, J. C., & Markman, E. M. (1994). Sixteen- and 24-montholds’ use of mutual exclusivity as a default assumption in second-label learning. Developmental Psychology, 30, 955-968.

Markman, E. M. (1994). Constraints on word meaning in early language acquisition. Lingua, 92, 199-227.

Merriman, W. E. (1991). The mutual exclusivity bias in children’s word learning: A reply to Woodward and Markman. Developmental Review, 11, 164-191.

Pepperberg, I. M. (1981). Functional vocalizations by an African Grey Parrot (Psittacus erithacus). Zeitschrifl fiir Tierpsychologie, 55, 139-160.

Pepperberg, I. M., & Wilcox, S. E. (2000). Evidence for a form of mutual exclusivity during label acquisition by Grey parrots (Psittacus erithacus)? Journal of Comparative Psychology, 114, 219-231.

Tomasello, M., & Akhtar, N. (1995). Two-year-olds use pragmatic cues to differentiate reference to objects and actions. Cognitive Development, 10, 201-224

The purportedly simple task called “match-to-sample” seems straightforward and forms the basis of many more advanced types of studies. In the picture to the right, the subject first sees a red and a green circle (a). After a short delay, that screen disappears, and another with only one circle appears (b). After another short delay, that screen is replaced by the original stimulus. If the subject taps the matching color (c), it gets a reward. Sounds simple, right?

Well, not very. Before anything even begins, the experimenter must make sure that the subject—say, you or your parrot—can distinguish the two colors. Some nonhumans (and humans!) are somewhat color-blind, others—like parrots—see in the ultraviolet. Thus, for the parrots, the “red” and “green” might not actually look like what you and I see as red and green, but the two colors will look different enough for the purposes of the experiment. That isn’t true for all subjects, so some knowledge of the subjects’ visual system must be obtained initially.

Experimenters also have to make sure that the subject doesn’t like one color more than the other so that the subject doesn’t come into the experiment with some kind of preference, which would skew the results. Ditto for a side preference—if for some reason (say, you are left-handed), and you like to tap the left circle more, that would also skew the results.

Everyone Likes Rewards

Subjects will also probably have to be taught to tap something for a reward: If I sat you in front of a computer screen with those two circles on it, and gave you no direction—no mouse, no keyboard, just the screen—would you know what to do? At first, I’d have to wait however long it took until you chose to a tap on a circle before I could give you a reward; then, in order not to skew the results, I’d have to reward you for learning to tap both colors and make sure that you are tapping both equally (again, to avoid any preferences)—but of course that teaches you to tap both, which is not what I want you to do eventually. But let’s assume that you are now tapping both colors equally and we start the experiment: After seeing both circles, your screen blanks and the single circle appears, which then disappears and the original screen returns…what should you do? As it turns out, it takes most subjects at least a few trials to figure out what is being tested—to tap the correct “match” and to unlearn the need to tap both circles equally.

Note that even if subjects guess correctly on the first trial and get a reward, it doesn’t mean that the subjects will act the same way on the second trial—they might want to try to find out what happens if they make several different choices. Now, in my lab, we would skip most of these steps by simply demonstrating what we want our birds to learn—by having a bird sit on a perch while two trainers sit in front of the screens that appear and disappear, showing what should be tapped—but we still have to worry about color and side preferences. And that is just for the basic step in many tasks! You can begin to see what is involved.

For some tasks, like studies on object permanence, where we are testing whether subjects understand that something still exists even if they can no longer see it, we hide things in one of several places and then ask them to find it—and thus have to make sure that they can’t tell where the object is through inadvertent cues. One such cue is scent. Even though we think that parrots’ sense of smell is about as good as ours (i.e., not very good), we can’t be sure, so we have to control for possible scent cues. This issue would be especially important for subjects like dogs, that are really good at scent detection.

We also have to make sure that we touch all the different possible sites after we hide the object and before we let the birds make a choice, to ensure that they aren’t just going to—or avoiding—the last place we touched, as another cue. And we have to make sure that we aren’t cuing them by gazing at the correct spot. Parrots don’t track our eye movements at a close distance, so that’s not usually a problem, but they can track our head movements, so we have to be careful to stare downwards or straight ahead when they make their choices. Of course, other species might be able to track eye movements. Again, you start to see some of the possible subtle issues…and even how the species that is being tested affects the issues.

For some tasks, we have to find two different treats—whether food or toys—that the birds like equally well—something that is not always easy to do. Occasionally that is just impossible, so we show the birds that they can ask for some favorite treat if they’ll play along and choose or identify something that isn’t particularly attractive or fun for them. It often takes a long time for them to agree to work under that condition.

Then there are issues such as “expectation cuing”—something that can occur when the subjects experience many trials on the same topic in a row, so they “expect” that the answers will involve a particular type of response, like a number or color label, and therefore ignore all the other possible answers, which significantly simplifies the task. That is why we do lots of different types of tests all at the same time.

I’ve tried to provide just a sampling of just some of the types of issues that are involved when one is trying to determine the various abilities of nonhumans. Many others exist, but even from this short essay, I hope you can see that the task of designing an experiment is never simple!