Language Research: Parrots
A review of language-centered studies on parrots and implications for understanding parrot cognition.
Language-based animal studies have previously focused on teaching language to nonhuman primates in an attempt to understand how language evolved in humans. However, they eventually branched out, using language as a tool to understand cognition in other animals such as dolphins and birds (e.g., Hayes and Nissen 1971; Herman 1980; Pepperberg 1981). Several bird species have demonstrated an ability to imitate human speech, including corvids, songbirds, and, notably, parrots. Parrots (Psittacidae sp.) are a family composed of approximately 390 species. Language research on parrots has focused on African grey parrots (Psittacus erithacus), particularly the work of Dr. Irene Pepperberg, which used their ability to interface with humans to explore their cognitive ability. This work paved the way for other researchers to continue examining parrots and their cognition through language. These studies have allowed for an understanding of how parrots learn and how this ability compares to other species, including humans.
Parrots are a suitable study species for language-based research for several reasons. The main reason is parrot’s ability to imitate human speech, allowing for limited communication between researcher and subject. This ability has been observed in various parrot species, all of which were found to learn quickly (e.g., Lashley 1913). Additionally, parrots are social animals and display nonsexual affiliative vocalizations with other group members, making them more willing to engage with researchers and build relationships with them (Colbert-White et al. 2014). There is also evidence that parrots learn through social interactions with others, as seen with orange-chinned parakeets (Power 1966). Several neurological features also make them suitable for language research. Despite having small brains compared to other animals used in language studies, parrots have a larger relative sized brain than expected for their body size (Péron 2012). Similar to humans, they also show brain asymmetry and a bias for communication related structures (e.g., Bottjer and Arnold 1985). Together these features make parrot species ideal for examining cognition through language ability.
Irene Pepperberg’s Studies with Alex the Parrot
The most notable subject in language research using parrots was Alex the African grey parrot, who was trained and studied by psychologist Dr. Irene Pepperberg, from 1977 until his death in 2007. Unlike early language studies on apes, these studies with Alex attempted to develop an understanding of animal cognition rather than human language development. Pepperberg’s work focused on testing cognitive ability using language as a tool and teaching Alex to communicate by training him to use English speech.
This training was accomplished by using a Model/Rival technique, which used social interactions with his trainers to demonstrate desired behaviors (Pepperberg 1981). For example, Alex would view one researcher asking another researcher questions about an object of interest, discussing the object name, size, shape, or color. When these questions were answered, the other researcher would be praised or scolded appropriately. Additionally, these roles would be reversed in order to show that either participant in an interaction may ask questions, encouraging Alex to do so himself. When being tested, if Alex correctly identified an item, he was rewarded with either the item itself (helping to intrinsically reinforce the name with the object) or a desirable treat that was requested (e.g., a piece of banana after saying “I want banana”) (Pepperberg 1988).
To ensure accuracy while testing Alex’s understanding of language, several precautions and controls were employed during testing procedures. To avoid Alex forming an association between certain trainers and certain types of questions, trainers would not repeat the same question topic during their trials (Pepperberg 1981). Additionally, objects and questions used in training were randomly ordered by a researcher who would subsequently not administer the training to avoid cuing from trainers and to prevent Alex from preempting certain questions for a topic (e.g., Pepperberg 1990). Questions of the same type were also never repeated within a session, unless answered incorrectly (e.g., several “how many” questions). Randomized order and no repetition of question types also helped to maintain consistency across sessions in terms of Alex’s attention, avoiding behaviors such as preening, interruptions, or discontinuing engagement (Pepperberg 1999). In the case of the trainer mistakenly judging a response as incorrect, Alex was trained to repeat his response, allowing for a second confirming judgment (Pepperberg 1999). Additionally, accuracy is further ensured by another trainer repeating any answer given by Alex, reducing the chance that a similar to correct response is accepted (Pepperberg 1981). A control was also implemented to dissuade Alex from asking repeatedly for an item rather than correctly identifying the one he is presented with. In these situations, he was allowed four attempts at a correct identification, before moving on, and with no rewards being given for incorrect answers (Pepperberg 1981).
Model/Rival training allowed researchers to test Alex’s ability to label and request items, understand categories, comprehend similarities and differences, as well as understand absence, numbers, and size of objects relative to each other. For example, Alex could successfully identify 100 objects, as well as 7 colors and 5 shapes in approximately 80% of trials (Pepperberg 1999). Additionally, Alex also performed well in recursive tasks in which he was tested on several categories at once (e.g., “which object is shape-b?”) and even when a conjunctive condition was added (e.g., “what object is color-a and shape-b?”) (Pepperberg 1990). Alex also used the term “none” to indicate the absence of an item, despite not being taught this label (Pepperberg and Gordon 2005). Outside of his training sessions, Alex was observed engaging in sound play, often related to labels that he was being taught in training (Pepperberg et al. 1991). Overall, these studies and observations allowed for the examination of African grey parrot’s cognitive abilities in detail.
Alongside research on Alex, additional studies were carried out on three other African grey parrots, named Griffin, Kyaaro, and Alo, all previously untrained and varying in age. The aim of studying these individuals was to determine how using a Model/Rival technique was effective for training Alex in order to gain insight into how he learned. This was achieved by breaking the technique down into its major components and then removing one or more to examine their importance (Pepperberg 1994). The components were defined as referencing (e.g., labelling objects and rewards for successful use of labels), context and function (e.g., motivating use of a label through tasks such as requests), and social interaction (e.g., interaction with the subject, which guides the learning experience) (Pepperberg 1994). Removal of all three components or removal of social interaction, context, or referencing resulted in no learning being observed in experiments with the three subjects, with successful learning only resulting from the presence of all three components (Pepperberg 1994).
Further studies by Pepperberg and colleagues examined the parrots’ understanding of labels and, if they assume mutual exclusivity, the belief that a label defines an entire object rather than a specific feature such as color or shape. To investigate this, comparisons were made between how Alex and Griffin were taught labels for objects. While Alex was told features of objects alongside the object names themselves (e.g., ball, red ball), Griffin was taught these labels separately (e.g., ball, red) (Pepperberg 1981; Pepperberg and Wilcox 2000). Consequently, Alex showed no mutual exclusivity when using labels, and Griffin did, responding to questions only with the first label he learned for items.
Another study examined the categorization abilities of two African grey parrots, Shango and Zoe (Giret et al. 2009). They were taught words that either belonged to food (e.g., chickpea), object (e.g., fork), or neutral (e.g., “hello”) categories, and then the parrots recorded during and after training sessions to examine their spontaneous vocalizations. Researchers found that Zoe used spontaneous food labels more when faced with food items, and Shango did so for food and object items, suggesting that they grouped items based on edibility.
Giret et al. (2010) next examined how effective different training methods would be when learning labels across ten parrots. Initially, three parrots were trained with the Model/Rival method used in Pepperberg’s studies and the other seven used variants upon this. They were also exposed to an Intuitive training method in which items were handled by a trainer while saying their label, a Repetition/Association method, in which items were produced when the subject says a label they’ve been exposed to, and a Diffusion/Association method, where recordings of labels were played to the parrots (Savage-Rumbaugh and Lewin 1994). They found that the Repetition/Association method was most effective for learning, while Diffusion/Association failed. However, the researchers noted that Model/Rival was likely less effective due to differing housing conditions and individual circumstances with the subjects compared to Pepperberg’s studies.
Other studies have described how African grey parrots are capable of substituting words with similar meanings and uses within contexts, such as “street” and “road” (Kaufman et al. 2013). Vocalizations of an untrained subject, Cosmo the pet African grey parrot, were found to display such substitution in sentences such as “Cosmo wanna,” in which she used words such as “talk,” “go,” and “cuddle.” In this situation, Cosmo understood that “Cosmo wanna” symbolizes a request and the words following represent the requested action. Another study on Cosmo found that she used vocalizations depending on context (Colbert-White et al. 2011). For example, when she was separated from her owner, Cosmo would emit vocalizations discussing where she or her owner where (e.g., “where are you”). Outside of studies focusing on Alex, several other researchers have conducted investigations on parrots and language.
Implications and Similarities to Other Species
The findings of these studies offer insight into the cognitive capabilities of African grey parrots. For example, the ability to understand and switch between categories of labels such as colors and shapes demonstrates a higher-order class concept and an understanding of abstract concepts (Hayes and Nissen 1971). An understanding of how items can be the same or different from each other may suggest a simple understanding of language, as labels must be understood in order to describe relationships between items (Premack 1978). In addition to this, an understanding of absence suggests advanced cognitive development (Brown 1973). This is supported by the ability to compare items relative to each other (e.g., “which color bigger?”) in which information about two features are attended to simultaneously. Kaufman et al. (2013) suggest that Cosmo’s use of substitution may also suggest complex cognition. Studies removing elements of Model/Rival training highlight the essential aspects of learning for African grey parrots: referentiality, contextual use, and social interaction. Colbert-White et al. (2011) also demonstrate the importance of social interaction for learning. However, further studies demonstrated that mutual exclusivity occurs when subjects learn labels purely in a context-dependent manner, similar to that of young children (Liittschwager and Markman 1994). Similarly, the use of perceptual features to categorize objects seen in Shango and Zoe resembles the categorization ability of children under 1 year (Madole et al. 1993). Another parallel with language development in children can be seen in Alex’s use of sound play outside of training sessions (Pepperberg 1991). As well as similarities to language development in humans, African grey parrots also display cognitive similarities with other species. Nonhuman primate, cetacean, and pinniped species have also demonstrated an ability to comprehend recursive and conjunctive tasks and the relationships between items, as well as use labels and make requests (e.g., Herman 1987). Pepperberg (1999) proposes that parrot cognitive ability is similar to that of primates, cetaceans, and pinnipeds due to their shared long-life history and complex social system.
Controversy and Limitations of Parrot Language Studies
While language studies on parrots have uncovered much about their cognition, they also have limitations and have been criticized. For example, Hebert Terrace, who previously conducted studies on chimpanzees and language, proposed that Alex the parrot’s responses were simply based on external guidance from trainers rather than a real understanding of the task (Smith 1999). However, there were measures taken during the experiments with Alex to avoid such cuing from trainers (e.g., Pepperberg 1990). One frequent limitation of these studies is a low sample number, often only with one subject (e.g., Kaufman et al. 2011). While this is true for studies that aim to compare between groups, researchers have argued that if one individual can perform a task, this suggests that the species is generally also capable of doing so (Triana and Pasnak 1981). However, repeated experiments and additional subjects can be useful for highlighting issues with methodology, such as in Giret et al. (2010) in which the Model/Rival method failed in one lab involved, but not the other. Other issues may occur due to individual features such as personality, motivation, age, or sex (Cussen 2017).
Despite limitations and criticisms, language-based studies of parrots have uncovered much about African grey parrot cognition, showing that, contrary to previous beliefs, they demonstrate complex cognition. Examples of their capabilities include referential labelling, making requests, categorization, word substitution, and an understanding of context, many of which are comparable to those of human children. While language studies on nonhuman primates decreased over time, studies of parrot cognition using language are still being conducted today.
- Bottjer, S. W., & Arnold, A. P. (1985). Cerebral lateralization in birds. In S. Glick (Ed.), Cerebral lateralization in nonhuman species (pp. 11–38). Orlando: Academic.Google Scholar
- Colbert-White, E. N., Corballis, M. C., & Fragaszy, D. M. (2014). Where apes and songbirds are left behind: A comparative assessment of the requisites for speech. Comparative Cognition & Behaviour Reviews, 9, 98.Google Scholar
- Hayes, K. J., & Nissen, C. H. (1971). Higher mental functions of a home-raised chimpanzee. In A. M. Schrier & F. Stollnitz (Eds.), Behaviour of non-human primates (Vol. 4, pp. 57–115). New York: Academic.Google Scholar
- Herman, L. M. (1980). Cognitive characteristics of dolphins. In L. M. Herman (Ed.), Cetacean behavior (pp. 408–409). New York: Wiley.Google Scholar
- Herman, L. M. (1987). Receptive competencies of language-trained animals. In J. S. Rosenblatt, C. Beer, M.-C. Busnel, & P. J. B. Slater (Eds.), Advances in the study of behaviour (Vol. 17, pp. 1–60). New York: Academic.Google Scholar
- Pepperberg, I. M. (1999). The Alex studies. Cambridge, MA: Harvard University Press.Google Scholar
- Péron, F. (2012). Language-trained animals: A window to the “black box”. International Journal of Intelligence Science, 2(4A), 149–159.Google Scholar
- Premack, D. (1978). On the abstractness of human concepts: Why it would be difficult to talk to a pigeon. In S. H. Hulse, H. Fowler, & W. K. Honig (Eds.), Cognitive processes in animal behaviour (pp. 421–451). Hillsdale: Erlbaum.Google Scholar
- Savage-Rumbaugh, E. S., & Lewin, R. (1994). Kanzi: The ape at the brink of the human mind. New York: Wiley.Google Scholar
- Smith, D. (1999). A thinking bird or just another birdbrain? The New York Times, pp. A00001.Google Scholar