The Alex Studies

Cognitive and Communicative Abilities of Grey Parrots

Irene Maxine Pepperberg

• Publisher: Harvard University Press
• Year: 2002
• ISBN:
  9780674255463
• Topic: Ciencias biológicas
• Subtopic: Ciencias en general

Can a parrot understand complex concepts and mean what it says? Since the early 1900s, most studies on animal-human communication have focused on great apes and a few cetacean species. Birds were rarely used in similar studies on the grounds that they were merely talented mimics--that they were, after all, "birdbrains." Experiments performed primarily on pigeons in Skinner boxes demonstrated capacities inferior to those of mammals; these results were thought to reflect the capacities of all birds, despite evidence suggesting that species such as jays, crows, and parrots might be capable of more impressive cognitive feats.Twenty years ago Irene Pepperberg set out to discover whether the results of the pigeon studies necessarily meant that other birds--particularly the large-brained, highly social parrots--were incapable of mastering complex cognitive concepts and the rudiments of referential speech. Her investigation and the bird at its center--a male Grey parrot named Alex--have since become almost as well known as their primate equivalents and no less a subject of fierce debate in the field of animal cognition. This book represents the long-awaited synthesis of the studies constituting one of the landmark experiments in modern comparative psychology.

1

INTRODUCTION

In Search of King Solomon’s Ring

As Holy Scripture tells us, the wise King Solomon, the son of David “spake also of beasts, and of fowl, and of creeping things, and of fishes” (I Kings IV.33). A slight misreading of this text . . . has given rise to the legend that the king was able to talk the language of animals, which was hidden from all other men . . . I am quite ready to believe that Solomon really could do so, even without the help of the magic ring which is attributed to him by the legend.

Konrad Lorenz, King Solomon’s Ring (1952:xiii)

In the very earliest times, when both people and animals lived on earth, a person could become an animal if he wanted to, and an animal could become a human being. Sometimes they were people and sometimes animals and there was no difference. All spoke the same language. That was the time when words were like magic.

Nalunglaq, a Netsilik Eskimo (Rasmussen 1972:45)

The wish to “talk” with animals and understand their lives is not a recent phenomenon. According to legend, King Solomon possessed a ring that enabled him to communicate at will with all the birds and beasts in his realm. Similarly, some Native Americans supposedly had the ability to change into various animals and thus share their lives. And our literature—although primarily that for children—abounds with stories in which the protagonists, human and animal, freely “speak” with each other (e.g., Lofting 1948; Sleigh 1955; Clark 1963; King-Smith 1984). As with most legends and stories, these represent deeply held human dreams and desires.

In the twentieth century, researchers began to test whether these dreams and stories might be turned into reality. Scientists resolved to teach animals to communicate using human speech. Some researchers raised infant apes in their homes, thus providing their subjects with the same experiences as those of a human infant, including a language-rich environment (e.g., Furness 1916; Yerkes 1929; Hayes and Nissen 1956/1971; Kellogg 1968; note Laidler 1978). Despite the scientists’ hard work and dedication, the apes never learned to use more than a few words of human speech. The apes’ failures caused emphasis to shift toward studies designed to understand an animal’s natural communicative behavior. Researchers examined various species, particularly birds, vervet monkeys (Cercopithecus aethiops), chimpanzees (Pan troglodytes), and marine mammals. Thus scientists, lacking access to Solomon’s ring or Native American shapeshifting, developed experimental techniques to achieve insight into nonhuman communication. And after decades of work, they did indeed gain some knowledge about why birds sing (reviews in Kroodsma and Miller 1982, 1996), how vervets use alarm calls (Cheney and Seyfarth 1990), the differences that exist between affiliative and aggressive vocalizations in chimpanzees (Goodall 1986), and about dolphins’ use of their whistles (Tursiops truncatus; Tyack 1986; Sayigh et al. 1990; McCowan and Reiss 1995; review in McCowan and Reiss 1997). Interestingly, at about the same time as these studies began, the field of psychology was shaken by the aptly named “cognitive revolution”—the radical notion that levels and types of intelligence in nonhumans formed a continuum with those of humans. This notion inspired researchers to study a wide range of behavior—including communication—in various species (e.g., Hulse, Fowler, and Honig 1968). Nevertheless, at that time most scientists believed that truly interactive interspecies communication was no more a reality than King Solomon’s ring (e.g., Chomsky 1966). Even so, several dedicated researchers set out to prove not only that such communication was possible, but that it could also be used to examine intelligent behavior. They took the then-unacceptable stance that previous failures were simply a consequence of requiring apes to learn to communicate vocally, and developed new training paradigms (Gardner and Gardner 1969; Premack 1971; Rumbaugh et al. 1973). These pioneers set the stage for an entirely new field, and their successes led other researchers to begin working with marine mammals (e.g., Herman, Richards, and Wolz 1984; Schusterman and Krieger 1984). Their paradigms often challenged not only the accepted procedures for training, but also conventional techniques for testing animals and the existent doctrines of what should be tested. A very brief review of the history of several traditions in the scientific community at that time will help clarify why these new techniques constituted such a paradigm shift.

In the 1970s, the behaviorist tradition, best exemplified by Skinner (1974), still represented mainstream, laboratory-based behavioral science. In reaction against studies centered on anecdotal evidence and the introspective, “mentalist” approach that characterized the late 1800s (e.g., Romanes 1883/1977; review in Burghardt 1984), behaviorists emphasized experimental controls and eschewed discussions of thought or mental representations, information processing, or intentional actions. Scientists canonized the cautionary tenets of Morgan (1894) and the strictures of Watson (e.g., 1929). Models of learning were based on “associationist” principles: Simple laws were formulated to explain how external sensory input caused observable behavior (e.g., stimulus-response associations, reviews in Macphail 1982; Rescorla 1985; Roitblat 1987). Thus, for example, an animal saw an environmental stimulus, such as a green-lit button in one location; at some point the animal’s subsequent random behavior included an action such as pressing a nearby green-lit button and that particular action was rewarded. A weak association was thus formed between stimulus and action; reward controlled the formation of the association. If the animal instead hit a nearby red-lit button, no reward was given and the action would be, in the parlance of the behaviorists, “extinguished.” The next time the lit-button scenario appeared, the correct action was more likely, because it had been strengthened by the previous presence of a reward; thus behaviorists explained “match-to-sample” learning. Behaviorist researchers posited that even complex behavior patterns could be explained by breaking down these patterns into simpler, component parts that each have as their basis the same type of paired association. They argued that all behavior could be reduced to such laws, that the number of these laws was few, and that there was no need to examine cognitive skills, which, in part, include “learning, remembering, problem solving, rule and concept formation, perception, recognition” (Roitblat 1987:2). And, according to Skinner (1938), one needn’t study a wide variety of animals, because none would react any differently from a pigeon or a rat: The rules of learning were universal.

Despite initial successes, behaviorists found themselves faced with situations their laws could not explain. At first their general reaction was not unlike that of the classical physicists at the turn of the twentieth century, who were able to explain all but a few physical phenomena and who believed that their current paradigms would eventually provide appropriate answers; later, such physicists realized that explaining these phenomena required the entirely new paradigm of quantum physics (Kane and Sternheim 1988). Similarly, the apparently anomalous activities of animals (for examples, see Breland and Breland 1961; Garcia and Koelling 1966; Smith and Roll 1967; Williams and Williams 1969; reviewed by Roitblat 1987), whose natural responses to stimuli could not be reshaped by behavioristic training, forced at least some researchers to consider a new paradigm in which animals were seen as active processors of information (Kamil 1984, 1988; e.g., Pepperberg 1990c). This need was made even clearer by behavioral ecologists, whose data could be explained only by positing mechanisms such as selective attention and long-term memory (e.g., Schoener 1991; Pyke, Pulliam, and Charnov 1977; Kamil and Sargent 1981; reviewed by Roitblat 1987).

But even this cognitive stance, which argued that behavior is best explained in terms of mental representations and information processing, still often retained the physical techniques of behaviorism. Researchers in many cases continued to examine animals in isolation, in sterile laboratory conditions, and using operant conditioning techniques. In operant procedures, the researcher typically holds an animal at 80% of free-feeding weight and places it in a small box (an “operant chamber”) containing devices that deliver a stimulus (such as a colored light), provides a response mechanism (such as a lever to press), records these responses (via a computer link), and delivers food rewards (perhaps via an automatic pellet dispenser). The animal is isolated from almost all stimuli other than those controlled by the experimenter; this isolation extends to all interactions with a natural environment, including interactions with other organisms. The experimental stimuli, usually tones, colored lights, or line patterns, generally have little to do with the animal’s natural experiences. Also, an animal cannot communicate directly with the experimenter, so no explicit transfer of information exists about the task to be learned. Thus the animal must determine, through trial and error, the question that the experimenter wishes to ask and to which it must respond—the specific nature of the task (Pepperberg 1992)—as well as the appropriate response. The rationale is that such procedures examine learning proclivities so basic that they exist in the absence of environmental influences.

Researchers determined, using such procedures, that animals did indeed actively process information and use mental representations to solve problems (see, e.g., review in Roitblat 1987; Roitblat and von Fersen 1992); why, therefore, might one need to engage an animal in direct conversation to determine its cognitive capacities? Two major reasons exist. One is basically procedural, the other philosophical; they are, however, closely intertwined.

The procedural reason for using interspecies communication becomes apparent if one compares the experimental design of a representative task—the study of concepts of color, shape, similarity and difference—from operant and communicative perspectives. In a typical operant setting, a subject is given a red triangle, a red square, and a blue square, and is rewarded for choosing blue when the objects are backed in white and for choosing triangle when the backing is black (see Thomas 1980). A successful subject may learn something about similarity and difference (choose “odd color” versus “odd shape”), but only with respect to the specific objects rather than the concept of same versus different (Premack 1978; Pepperberg 1987a). A concept, in contrast, can be extended to any modality. In general, reports of how well the animal transfers its knowledge to novel operant situations (which would demonstrate some understanding of the concept) do not include its responses to the very first presentation of novel objects; rather, researchers report “learning curves” that show merely that the animal takes less time to learn to respond correctly on the novel task than to learn the task in the first place (Thomas and Noble 1988). Moreover, an animal in such a study need not determine specifically what, if anything, is same or different.¹ Contrast this study with one in which an animal learns to communicate with humans by using some form of symbolic code (e.g., vocal labels, American Sign Language, abstract computer symbols). Here the animal first acquires labels for various colors, shapes, and materials, category labels for these attributes, and comprehension of the phrases “What’s same?” and “What’s different?” The animal is then shown a red wooden triangle and a red wooden square, and asked, “What’s different?” The animal must, based on its understanding of the question, first determine whether the experimenter is interested in similarity or difference. The animal must then observe the two objects, determine what attribute they do not have in common (the “odd” attribute), encode that information, and tell the experimenter the category label of this attribute, “shape.” If the experimenter then asks “What’s same?” the animal can respond either “color” or “matter.” The animal can be given any set of objects, familiar or novel, and still respond appropriately on the very first trial (see Chapter 5). Clearly, both techniques explore an animal’s cognitive abilities, but the communicative approach enables researchers to ask more complex and advanced questions—which is the philosophical reason for using interspecies communication.

Specifically, possession of (or the ability to acquire) human language or at least some form of a human-based communication code has been posited as a necessary precondition for an organism to organize and process information for certain complex cognitive tasks (for details, see, e.g., Macphail 1982, 1985, 1987; Premack 1983, 1986).² A brief summary of the position is that without some form of language-like code, an animal is capable only of remembering and working from images (such as mental pictures of the location of items in space), which restricts its capacities. An animal that has been taught a human-based code, in contrast, can supposedly work from abstract representations (e.g., not simply learn that “red” and “green” are associated with particular hues, but understand that these labels form a category known as “color” and that this knowledge of the category can be used to solve problems). Thus, according to this argument, a language-trained animal can understand and use a concept it would otherwise be unable to learn. One refutation of this position is that an animal trained in this manner simply has more experience in categorizing its world according to the experimenter’s specific demands (see, e.g., Bronowski and Bellugi 1970; Marshark 1983; Miller 1983). The most interesting refutations, however, come from studies of human children. Reese (1972), for example, showed that labeling seems not to be a prerequisite for understanding concepts of same and difference because even children with normal language skills (grades K-2) had somewhat more difficulty labeling objects than denoting if pairs are same or different. Rice (1980), moreover, showed that language training does not improve a child’s ability to learn concepts such as “color” (as opposed to the labels for individual colors) if the child has not already reached a certain stage of development. Subsequent studies (review in Mandler 1990) support the idea that children form many concepts well before they acquire language.³ What is not easily resolved, however, is how complex a concept must be before its comprehension might indeed require a form of language learning: Animals that have been trained to use a human-based language code do better than untrained animals on some tests, such as those requiring understanding of analogies, whereas little difference exists between trained and untrained animals on tasks such as those involving spatial relations (Premack 1983).

Evaluating whether language training enables animals to succeed on complex tasks is also tricky because it is almost impossible to separate procedural and philosophical effects. The reason is that interspecies communication is a particularly powerful means of studying animal cognition. As I have previously noted (e.g., Pepperberg 1987c), interspecies communication (1) enables researchers to communicate directly to their subjects the precise nature of the questions being asked (the animals need not determine the nature of the question through trial and error); (2) takes into account research showing that social animals respond more readily and often with greater accuracy when placed within a social context at least somewhat similar to their field situation (see Menzel and Juno 1982, 1985); and (3) enables researchers to compare data not only between humans and other animals but among various nonhuman species. Interspecies communication also provides an open, arbitrary code that can create an enormous variety of signals; such variety allows researchers to examine the nature, and not just the extent, of the information perceived by an animal. Two-way communication also allows rigorous testing because animals can be required to choose a response from their entire repertoire rather than from a subset relevant only to the topic of a particular question. Moreover, an animal that learns such a code may also respond in novel and possibly innovative ways that may demonstrate even greater competence than that required by the responses of operant paradigms. Interspecies communication may therefore simply allow researchers to demonstrate more readily the inherent capacities of their animal subjects, rather than enable the animals to learn more complex tasks. Of course, an animal must first be taught the communication code, and here again major contrasts exist between an operant and an interactive protocol.

In the late 1960s and early 1970s, the several research laboratories studying interspecies communication used three quite different approaches. Several reviews of these projects are available (e.g., Ristau and Robbins 1982; Roitblat, Herman, and Nachtigall 1993; Hill 1995), and thus a detailed description of the projects and their results are not necessary here. I will instead focus on studies with chimpanzees, and briefly describe how the different training approaches led to strikingly different results (for details, see the review in Pepperberg 1997).

Although the extent to which chimpanzees can acquire a communication code isomorphic with that of human language is unresolved (see, e.g., Terrace et al. 1979; Savage-Rumbaugh et al. 1980a,b; Greenfield and Savage-Rumbaugh 1991), the animals studied did learn to label, request, refuse, and categorize objects and some actions, and often learned to comprehend some form of human speech (see, e.g., Fouts, Chown, and Goodin 1976; Savage-Rumbaugh et al. 1985; see Williams, Brakke, and Savage-Rumbaugh 1997 for a recent review). The conditions under which chimpanzees were taught human-based codes ranged from computer-driven systems with limited social interaction (Rumbaugh 1977), to a system in which the ape was trained to use magnet-backed plastic chips to communicate with humans in a laboratory setting (Premack 1971), to cross-fostering programs in which the ape was raised much like a human child (Gardner and Gardner 1969, 1989). Even when the actual code being taught in two laboratories—American Sign Language (ASL)—was identical, techniques used for training often differed radically (Gardner and Gardner 1969; Terrace 1979b), and the different projects obtained different results. All these projects subsequently were harshly critiqued (primarily with respect to claims of language abilities; see, e.g., Lenneberg 1971; Terrace et al. 1979), but the severest criticism was reserved for projects with the most naturalistic settings; the arguments were that these projects lacked experimental controls (Sebeok and Umiker-Sebeok 1980). Interestingly, animals in programs that relied most heavily on training related to operant conditioning (and often controls, e.g., Premack 1976; Rumbaugh 1977; Terrace 1979a) demonstrated behavior that was far less flexible and less “language-like” (note Terrace 1979b) than that of apes trained in ways that more closely resembled the experiences of young children (Gardner and Gardner 1969, 1989).⁴ Those who criticized the more naturalistic, communicative approach eventually adopted the primary elements of this paradigm (Savage-Rumbaugh 1991; Williams et al. 1997), but only the criticisms were voiced in the late 1970s.

In any case, the accepted dogma in the mid-1970s was that interspecies communication studies were possible, but that subjects must be animals with a close phylogenetic relationship to humans, such as chimpanzees (see, e.g., Sarich and Cronin 1977),⁵ or at least with large brains, like dolphins (see, e.g., Russell 1979; note, however, Morgane, Jacobs, and Galaburda 1986). Thus when I decided to study parrots, colleagues generally expressed extreme skepticism. Numerous studies had already shown that birds—represented by the laboratory pigeon—were greatly inferior to mammals on standard tests of complex behavior, for example, on tasks such as reversal learning (Bitterman 1975) and insight detour problems (Krushinskii 1960). Researchers had also shown that birds lacked, to any great extent, cerebral cortex—the so-called mammalian organ of intelligence (see, e.g., Jerison 1973). Finally, studies using mynahs and various parrot species had attempted—and failed—to teach birds to communicate with humans (see, e.g., Mowrer 1950; see Chapter 2). The overall perception was t...

Table of contents

• Cover
• Title
• Copyright
• Contents
• Preface
• 1 Introduction: In Search of King Solomon’s Ring
• 2 Can We Really Communicate with a Bird?
• 3 Can a Parrot Learn Referential Use of English Speech?
• 4 Does a Parrot Have Categorical Concepts?
• 5 Can a Parrot Learn the Concept of Same/Different?
• 6 Can a Parrot Respond to the Absence of Information?
• 7 To What Extent Can a Parrot Understand and Use Numerical Concepts?
• 8 How Can We Be Sure That Alex Understands the Labels in His Repertoire?
• 9 Can a Parrot Understand Relative Concepts?
• 10 What Is the Extent of a Parrot’s Concept of Object Permanence?
• 11 Can Any Part of a Parrot’s Vocal Behavior Be Classified as “Intentional”?
• 12 Can a Parrot’s Sound Play Assist Its Learning?
• 13 Can a Parrot’s Sound Play Be Transformed into Meaningful Vocalizations?
• 14 What Input Is Needed to Teach a Parrot a Human-based Communication Code?
• 15 How Similar to Human Speech Is That Produced by a Parrot?
• 16 How Does a Grey Parrot Produce Human Speech Sounds?
• 17 Conclusion: What Are the Implications of Alex’s Data?
• Notes
• References
• Glossary
• Credits
• Index