For hundreds of years, learning meant the formation of associations, and was considered the means by which society transmitted its acquired cultural capital. Learning was the cliché which lay behind almost every explanation in the social sciences. Increasingly, however, the study of learning has been transformed into the study of the human mind. When one speaks of learning one must speak of representations, of knowledge, of modularity, of innate and specific structures of mind. While the new view of learning is mostly restricted to the cognitive sciences—the social sciences as a whole have not been affected—one would expect that it will ultimately have a powerful impact on the social sciences.
In philosophy and psychology, learning has traditionally been regarded as a potential solution to the problem of knowledge. How is it that a human being comes to have knowledge of the world? In this context, the study of learning has long been central to the study of the human mind. Empiricist philosophers, such as Locke and Hume, conceived of knowledge as a system of association of ideas. Hume invoked principles which explained how these associations were formed. For example, the principle of contiguity said that if two ideas occurred near each other in time, then it was likely that an association was formed between these ideas. In the latter part of the nineteenth century these principles became the focus of experimental study. With the behaviourist revolution in American psychology in the twentieth century, association theory was modified so that it no longer was ideas that were associated. Rather, a stimulus was associated with a response. But the basic underlying notions of association theory persisted: that human knowledge is to be represented as a system of associations and that these associations are learned.
The study of learning thus became the experimental study of the learning of associations. To make these associations experimentally testable, the learning of arbitrary associations was studied, for example, the associations between nonsense syllables, like dax and gep. From this emerged the famous learning curve, showing the probability of individuals forming an association as a function of the number of times the associated items were shown to them simultaneously. Not surprisingly, it was discovered that the more practice that people have on an association, the better they learn it.
Grand theoretical schemes developed to explain learning, for example, those of Hull. These envisioned learning as a unitary phenomenon. In essence, human and animal learning were conceived to be the same thing, though there might be a quantitative difference. Within a species, what appeared to be different kinds of learning really were not. In short, the principles of learning remained unchanged across species and content of what was learned. A number of theoretical disputes arose over the precise character of these principles, but the different theories shared many underlying assumptions of breathtaking simplicity and elegance. These were that all learning was the same and that there were a few general principles of learning. The theory must have had simplicity and elegance, for what else could explain the fact that learning theory in this form lasted for so long? For the matter, plain and simple, was that learning theory did not work. If one actually considered real domains of human knowledge, it quickly became apparent that learning theory could not explain how that knowledge was acquired. The nonsense syllables of the laboratory, like the ‘ideas’ of the philosophers (or the ‘quality spaces’ of Quine), were abstractions which lost the essence of the problem.
The ideas which have replaced association theory are rich and interconnected. There is no way here even to hint at the extensive justification that has been developed for them. We shall simply list some of the themes of the new study.
Language and innate principles
Perhaps the major critique of the adequacy of association theory, as well as the most extensive development of an alternative theory for human learning, has come from the field of linguistics, namely from Chomsky (1965; 1975). Chomsky argues that the structure of language is such that there is no way that a human could learn language given any of the traditional notions of learning (call these ‘learning theory’). Since every normal person masters a natural language, learning theory could not possibly be correct. Chomsky argues that the only way that these structures could develop is if there is an innate basis for them in the human mind.
Conscious awareness
We are not aware of most of the knowledge that we possess. For example, most of the principles of language are beyond our conscious awareness. In broad terms, the modern innatist position is very much like that of the rationalist philosophers (Descartes, Leibniz). Probably the biggest single difference (besides the extensive detailed technical developments in the modern period) is that the philosophers generally seem to have believed that the principles of mind were available to conscious introspection. It also does not seem unreasonable to claim that even behaviourists would only invoke principles of explanation of which they were consciously aware, although they did not explicitly state this. According to the modern view, although principles of knowledge are not necessarily available to consciousness, they are still in the mind, and thus a matter of individual psychology. Giving up the assumption of the necessity of conscious awareness of principles of mind is a liberating force in learning theory, for it makes possible the development of theoretical constructs which traditionally would have been immediately ruled out on the basis of introspection.
Domain-specific and species-specific principles
The modern view violates the cardinal principle of the traditional view, that all knowledge except for some simple principles of association is learned. But it also violates the two subsidiary principles, that learning principles do not depend on the species or on the domain of knowledge. Principles of learning differ from species to species. Animals do not learn language, because the principles of language only occur in humans. Some scholars are quite willing to accept the innate character of principles of learning, but believe that these principles operate in all domains of knowledge. That is, the principles are some kind of complicated hypothesis-formation ability. But the modern view holds that there are different principles in different domains of knowledge. Thus those which underlie our ability to recognize objects are different from those which enable us to use language. This latter view has come to be called the modular view of cognition. In the entire history of learning theory, there are really only two general kinds of ideas about learning: the formation of associations, and some kind of hypothesis-formation. With the development of the modern view of domain-specific principles, it is possible to have a much more delineated learning theory with particular principles for particular domains of knowledge. Of course it is an open and empirical question whether more general principles of cognition and learning underlie the specific principles which have been discovered. On the evidence to date, it appears unlikely that the domains will be completely unified. For example, visual perception and language just look different.
Reinforcement
In many traditional views of learning, an organism could learn only if properly reinforced. For example, children were supposed to learn a response better if given a piece of candy when they made the correct response. The modern view, based on considerable evidence, is that reinforcement appears much less necessary for learning, although it may still be an effective motivator. Skinner (1957) argued that children learned to speak grammatically by being positively reinforced for correct sentences and negatively reinforced for incorrect sentences. But Brown and Hanlon (1970) and other investigators have shown that parents do not differentially reinforce grammatical and non-grammatical utterances of young children in the language learning period. Thus reinforcement does not appear to play a significant role in language learning, from the standpoint of giving information to the child. Its role as a motivator is more difficult to assess precisely.
Instruction
One of the surprising discoveries of the modern period is the degree to which children spontaneously develop cognitive abilities, with no special arrangements of the environment. The field of language acquisition, and cognitive development more generally, is replete with examples. It is clear that the rules of language, for example, are not taught to children. People in general do not know the rules, although they use them implicitly, so how could they teach them? Some scholars believe that although parents do not teach the rules of language, they nevertheless provide special instruction by presenting children with a particular simplified language that is fine-tuned to their levels of ability (Snow and Ferguson 1977). But the best evidence (Newport et al. 1977) seems to show that there is no such fine-tuning.
Certain abilities unfold naturally, with no special instruction. Language appears to be one of these. Principles of visual perception also follow this outline. The same may be true for certain basic principles of counting, although not for the learning of the names of numbers (Gelman and Gallistel 1978). Other abilities seem to stretch the ordinary limits of the human mind, and seem to demand instruction in the usual case. The learning of advanced mathematics, or many other subjects, seems to follow this pattern.
Learnability and feasibility
It has proved possible to define mathematically the question of the possibility of learning. Gold (1967) provided one of the first useful formalizations. Wexler and Culicover (1980) investigated the question of learnability for systems of natural language and showed that linguistic systems could be learned if language-specific constraints were invoked. They further investigated the problem of feasibility, that is, learnability under realistic conditions. Specific constraints can be invoked which allow for feasible learning systems, specifically, very complex systems which can nevertheless be learned from simple input. This would seem to mirror the situation for a child, who learns an essentially infinite system (say language) from exposure to only a fairly small part of the system.
Animal research
We have concentrated on learning in humans, especially the learning of language, the ability most centrally related to the human species. However, extensive modern research on animal learning also questions the traditional assumptions. It appears that traditional learning theory is not an adequate theory for animals.
Social implications
The assumption of innate principles of human cognition does not imply that there are innate differences between individuals or races. The central idea of the modern view is that the innate principles are part of the shared human endowment, just as the innate existence of a heart is. In fact, the existence of the innate principles of mind may be thought of as helping to define human nature, an old concept generally out of favour in the social sciences. Contemporary social scientists in general founded their theories of society and politics on a psychology closely associated with traditional learning theory. Thus, children are socialized into the values of their society. But it is conceivable that there are principles of mind (of human nature) which relate to the structure of society (or to interpersonal relations, or ethics). If so, the modern view of learning might one day be expanded to include these principles, and there may conceivably be a social science founded on the modern view of learning.
Kenneth Wexler
University of California
References
Brown, R. and Hanlon, C. (1970) ‘Derivational complexity and the order of acquisition of child speech’, in J.R.Hayes (ed.) Cognition and the Development of Language, New York.
Chomsky, N. (1965) Aspects of the Theory of Syntax, Cambridge, MA.
——(1975) Reflections on Language, New York.
Gelman, R. and Gallistel, C.R. (1978) The Child’s Understanding of Number, Cambridge, MA.
Gold, E.M. (1967) ‘Language identification in the limit’, Information and Control 10.
Newport, E., Gleitman, H. and Gleitman, L.R. (1977) ‘Mother I’d rather do it myself: some effects and non-effects of maternal speech style’, in C.E.Snow and C.A. Ferguson (eds) Talking to Children, Cambridge, UK.
Skinner, B.F. (1957) The Behavior of Organisms, New York.
Snow, C.E. and Ferguson, C.A. (eds) (1977) Talking to Children: Language Input and Acquisition, Cambridge, UK.
Wexler, K. and Culicover, P.W. (1980) Formal Principles of language Acquisition, Cambridge, MA.
Further reading
Piatelli-Palmarini, M. (ed.) (1980) Language and Learning, Cambridge, MA.
