Columbia psychologists Elizabeth Brannon and Herb Terrace
Two Columbia University psychologists have taught monkeys to discriminate computer-generated images containing as many as nine objects and to respond to them in ascending order, with a success rate well above what would be predicted by chance.
The work is the strongest evidence so far of numerical ability in non-human primates, said the researchers, Herbert S. Terrace, professor of psychology at Columbia and professor of psychiatry at the New York State Psychiatric Institute, and Elizabeth Brannon, a Columbia graduate student in psychology. Their work appears in the January issue of the Journal of Experimental Psychology: Animal Behavior Processes.
The research challenges the prevailing view, which dates to Descartes, that non-human primates cannot think because they cannot use language. It also challenges the views of B.F. Skinner, the noted behaviorist and Terrace's mentor at Harvard, who held that all examples of animal intelligence were simply conditioned behavior that didn't require cognitive explanations. Terrace and Brannon believe that cognitive processes -- thought -- are needed to explain the kind of complex behavior they are studying. They hope to show that human intelligence, like other human attributes, can be traced to animal origins.
"We have ample evidence that animals can think without language," said Terrace, who heads Columbia's Primate Cognition Laboratory. "In our current and previous research, we have shown that animals solved complex problems without help from external cues."
Added Brannon, "Though monkeys do not recognize the word 'two' or the symbol '2,' they share with humans the capacity to master simple arithmetic, on at least the level of a two-year-old child. We don't have direct evidence yet, but it seems likely that these monkeys, and other non-human primates, can calculate."
In a recent series of experiments, Brannon and Terrace trained two male rhesus monkeys, Rosencrantz and Macduff, by presenting them with 35 sets of images on a touch-sensitive video screen. Each picture contained a different number of different objects from one through four, for example, one triangle, two bananas, three hearts and four apples. The stimuli appeared in random positions on the screen, to prevent the monkeys from learning the required sequence as a series of fixed motor movements. Other features of the pictures unrelated to number, such as size, surface area, shape and color, were also varied randomly.
When the monkeys touched the pictures in ascending order, i.e., one square, two trees, three ovals and four flowers, they received a banana-flavored food pellet. If they made an error, the screen turned black for several seconds and a new trial began with different pictures. This "video game" paradigm, which the monkeys now clearly enjoy, trained them to perform cognitive serial tasks without verbal instructions -- without language, the researchers said.
"It's like using your password to get money from a cash machine, but it's actually much harder for the monkeys," Terrace said. "The pictures, and their position on the screen, change each time they try for another pellet of food. When you go to a cash machine, you don't have to deal with the numbers being in strange positions each time. We ask a lot, cognitively speaking, of our non-human primate subjects."
Over the course of learning 35 different training sets, the monkeys got better and better at responding in the ascending numerical order, one to four. The two psychologists then tested Rosencrantz and Macduff on 150 test trials, in which a new stimulus set, showing numbers of objects from five to nine, was presented on each trial. The monkeys performed just as well as they had on the original 35 training sets.
"There was no way the monkeys could have done this, unless they had learned some numerical rule for ordering the contents of the pictures," Brannon said.
To test whether the monkeys understood the ordinal relations between non-consecutive numbers -- that, for example, five is greater than three -- Brannon and Terrace gave the monkeys a new set of problems in which they were shown up to nine objects. The task was to first touch the picture containing the smaller number of objects, then the one with the larger number of objects. For example, if a monkey was shown one picture with five large circles and another containing seven small circles, the correct order was five, then seven. Rosencrantz and Macduff responded correctly even when the number of objects in the pictures exceeded four.
"This finding is important because it shows that monkeys know things about number that we haven't taught them," Brannon said.
Brannon and Terrace believe that arithmetic and language evolved separately, and that number skills preceded human speech. "Language is a complex social skill, whereas counting can be learned by the individual," Terrace said. "Counting is useful in foraging for food, assessing a group of predators or ordering the number of dominant males in one's group."
With further studies, the research team hopes to answer other questions on language and numerical abilities in non-human primates. Do monkeys, like humans, immediately identify small numbers of objects without counting? Could a non-human primate learn to match two different pictures of three objects, one containing, for example, three circles, the other, three triangles? Can monkeys learn symbols that represent numbers? How hard would it be for a monkey to learn a nonconsecutive numerical sequence, for example, three-one-four-two? Brannon and Terrace believe that these, and similar experiments on the numerical ability of animals, will provide a unique window into the evolution of human intelligence.
Terrace is best known for Project Nim, in which he tried to teach a chimpanzee named Nim Chimpsky to communicate with sign language. Nim, named for the philosopher and linguist Noam Chomsky, was raised and socialized like a human infant in the mid-1970s. Nim appeared to learn American Sign Language, and eventually mastered a 125-sign vocabulary. However, Terrace was not convinced that the chimp had demonstrated the ability to create unique sentences, the hallmark of language. After analyzing more than 20,000 different combinations of signs produced by the animal, Terrace concluded that Nim signed mainly to obtain particular rewards and that most of his signed combinations were unoriginal imitations of those uttered by his human teachers.
Brannon was awarded a National Science Foundation Graduate Fellowship to pursue a Ph.D. studying the cognitive abilities of monkeys. She joined the Primate Cognition Laboratory because she saw the potential for using Terrace's serial task, which he developed to study the precursors of language in non-human primates, to ask questions about natural abilities that monkeys might use to represent their environment, such as counting. She was recently awarded a fellowship from the National Institutes of Health to continue her experiments on monkey counting, the topic of her Ph.D. dissertation.
The work was supported by a grant to Terrace from the National Institutes of Mental Health.