Numerical Competence in an African Gray Parrot (Psittacus erithacus)

An African gray parrot (Psittacus erithacus), Alex, trained to label vocally collections of 1-6 simultaneously presented homogeneous objects, correctly identified, without further training, quantities of targeted subsets in heterogeneous collections. For each test trial Alex was shown different collections of 4 groups of items that varied in 2 colors and 2 object categories (e.g., blue and red keys and trucks) and was asked to label the number of items uniquely defined by the conjunction of 1 color and 1 object category (e.g., "How many blue key?"). The collections were designed to provide maximal confounds (or distractions) and thus replicate the work of Trick and Pylyshyn (1989) on humans. Humans count rather than subitize under such conditions. Alex's results (83.3% overall accuracy) are therefore discussed in terms of their relation to human numerical competence, particularly with respect to counting. Several studies have shown that a wide range of vertebrates recognize numerical quantities (e.g., reviews in Boysen & Capaldi, 1992; Davis & Perusse, 1988). Such studies are also often proposed as a basis for cross-species comparisons of general cognitive processing abilities (see Case, 1985; Gelman & Gallistel, 1986; Lenneberg, 1971; cf. Starkey, Spelke, & Gelman, 1990). The rationale is that numerical competence indicates an ability to generalize abstract categories across domains: Number is not an inherent attribute of an object, as is color, shape, or material, but is rather a descriptor that is applicable to any discrete collection of entities. Whether numerical competence indicates general abstract cognitive processing abilities remains, however, to be fully tested. Central to this issue is what constitutes numerical competence. Unfortunately, for quantities most often studied (1-8), two enumeration processes appear to exist: Subitizing, a supposedly fast, effortless, perceptual apprehension of numbers usually up to 4 that uses preattentive mechanisms (Kaufman, Lord, Reese, & Volkmann, 1949; Stevens, 1951; Taves, 1941; Wolters, van Kempen, & Wijlhuizen, 1987), and counting, a slow, effortful process generally for numbers greater than 4 that requires spatial attention (Aoki, 1977; Atkinson, Campbell, & Francis, 1976; Oyama, Kikuchi, & Ichihara, 1981). Note, however, that some researchers argue that counting does not begin until 6-7 (Mandler & Sheba, 1982; Miller, 1956; see Trick & Pylyshyn, 1989, 1991). During subitizing, number loses its abstract nature as a descriptor to become a holistic attribute, like color or shape (Pepperberg, 1988c). In contrast, counting is seen as a set of abstract steps: One (a) produces a standard sequence of number tags, (b) applies a unique number tag to each item to be counted, (c) remembers what already has been counted, and (d) knows that the last number tag used tells how many objects are there (Fuson & Hall, 1983). Despite arguments to the contrary (e.g., Beckmann, 1924; Gelman & Gallistel, 1986; Mandler & Sheba, 1982), subitizing is generally viewed as the simpler process (see Cole & Scribner, 1974; Davis & Perusse, 1988, and commentaries). Too, the cognitive overlap of subitizing and counting proposed by Gallistel and Gelman (1991, 1992) seems countered by data that has shown that the overlap is a form of subitizing (Davis & Perusse, 1988), that is, a post-counting process, distinct from precounting perceptual recognition, that through practice allows exceptionally speedy labeling of fewer than 4-6 items. This division retains counting as the purported process for larger quantities. Whether different processes are used to enumerate small versus large quantities and the relative complexity of such processes must be known if numerical competence can be used to compare cognitive abilities across species. Precounting subitizing, for example, does not require subjects to handle abstract entities, an ability that may indicate more general abstract processing capacities. Unfortunately, even studies in which animals selected or labeled quantity used collections within subitizing range (≤6; e.g., Boysen & Berntson, 1989; Davis, 1984; Matsuzawa, 1985; Pepperberg, 1987a). Questions thus still exist as to whether (a) processes used by animals to recognize quantities up to 8 involve anything more than simple perception (e.g., Mandler & Shebo, 1982), (b) animals can use for larger amounts the same enumeration process in which adult humans ultimately obtain proficiency, and (c) overall cognitive competence can be judged by numerical competency on small numbers if a simple perceptual mechanism can be used to recognize smaller collections. Clearly, animals that are to be compared with humans must be tested so as to demonstrate the more complex counting procedure. So far, few nonhumans have met the criteria for human counting. Some data on nonhumans suggest a capacity for counting (e.g., rats, Capaldi & Miller, 1988; cf. Davis & Albert, 1986), but only chimpanzees (Pan troglodytes) have shown abilities similar to young children (Boysen & Berntson, 1990). A chimpanzee, Sheba, demonstrated ordinality (Boysen & Berntson, 1988) and labeled, with a card depicting an Arabic numeral, the sum of two arrays separated in time and space (Boysen & Berntson, 1989). She was, however, tested only on sets up to 4. Might other animals show such capacities-or more advanced ones-with proper training? If capacities necessary for numerical competence are used for other complex cognitive tasks, then the answer is likely affirmative. Specifically, an African gray parrot (Psittacus erithacus) that can categorize, comprehend concepts of same-different, absence of information, and relative size, and show certain stages of numerical competence (Pepperberg, 1983, 1987a, 1987b, 1988a; Pepperberg & Brezinsky, 1991) will be a good candidate for such a study. This bird, Alex, can already vocally label collections of 26 simultaneously presented homogeneous or heterogeneous (e.g., corks and keys) objects (Pepperberg, 1987b). Tests have ruled out the possibility that the parrot uses cues, such as mass, brightness, surface area, odor, object familiarity, or canonical pattern recognition, but have not ruled out his use of a noncounting strategy, such as clumping or chunking-a form of subitizing (e.g., perception of 6 as two groups of 3; see Jevons, 1871; Mandler & Shebo, 1982; von Glasersfeld, 1982)-to obtain the correct answer. Choosing a Task to Study Numerical Competence in an Avian Subject Demonstrating avian counting behavior is challenging. Birds are extremely sensitive to quantifiable sequential auditory patterns (Seibt, 1982) and, unlike some species (e.g., rats; Davis & Albert, 1987), may transfer this ability to visual patterns. Homogeneous sets of external events may thus simply be judged as to whether they match or fill up specific perceptual patterns (Meek & Church, 1983), even for quantities that humans count. Crows (Corvus brachyrhynchos), for example, likely use various numbers of caws with different temporal patterning to identify each other in flocks (Thompson, 1968, 1969); each crow thus judges several different additive acoustic patterns at least to some degree with respect to quantity. Other avian species may recognize particular sets of repetitions of different neighbors' vocalizations so as to respond appropriately (European blackbirds, Turdus merula; Wolfgramm & Todt, 1982; wood peewees, Contopus virens, Smith, 1988; carduelid finches and their hybrids, Giittinger, 1979). Such behavior is not unlike human determination, without necessarily counting, of how many times to repeat fa Ia Ia in a familiar Christmas carol. Birds, however, seem particularly likely to memorize many such perceptual sets. This behavior is often termed serial subitizing (Burns, 1988, p. 581; note Davis & Perusse, 1988). Given the possibility of avian auditory-to-visual transfer, the sequential addition process used for the chimpanzee Sheba (Boysen & Berntson, 1990) may not be valid for testing a bird. An inferential test to distinguish perceptual recognition from counting has, however, been suggested by recent work on humans (Trick & Pylyshyn, 1989). The system is based on visual processing mechanisms and involves distractors (Trick & Pylyshyn, 1991). Subjects enumerate items in a field of distractors under two conditions, (a) either white or vertical lines among green horizontals or (b) white vertical lines among both green verticals and white horizontals. Subitizing was evident for 1-3 only for the first condition. Subitizing thus seemed to fail when visual items required conjunctive attentive processing for recognition, that is, when subjects must distinguish among various objects defined by a collection of features (e.g., color and shape). Such findings are consistent with the suggestion (Glanville & Dallenbach, 1929) that the number of items that can be apprehended simultaneously decreases as the amount to be perceived about them increases. Similar tests can be given to the parrot Alex because he can already use a conjunctive condition to identify a single object within a collection (e.g., a red key within a collection of colored keys and other red items; Pepperberg, 1992). He can now be asked to label the quantity of a subset of items similarly defined. Success may demonstrate that Alex's competency, if not necessarily his strategy, is equivalent to that of humans.