an interdisciplinary research cluster that seeks to account for intelligent activity, whether exhibited by living organisms (especially adult humans) or machines. Hence, cognitive psychology and artificial intelligence constitute its core. A number of other disciplines, including neuroscience, linguistics, anthropology, and philosophy, as well as other fields of psychology (e.g., developmental psychology), are more peripheral contributors. The quintessential cognitive scientist is someone who employs computer modeling techniques (developing computer programs for the purpose of simulating particular human cognitive activities), but the broad range of disciplines that are at least peripherally constitutive of cognitive science have lent a variety of research strategies to the enterprise. While there are a few common institutions that seek to unify cognitive science (e.g., departments, journals, and societies), the problems investigated and the methods of investigation often are limited to a single contributing discipline. Thus, it is more appropriate to view cognitive science as a cross-disciplinary enterprise than as itself a new discipline.
While interest in cognitive phenomena has historically played a central role in the various disciplines contributing to cognitive science, the term properly applies to cross-disciplinary activities that emerged in the 1970s. During the preceding two decades each of the disciplines that became part of cogntive science gradually broke free of positivistic and behavioristic proscriptions that barred systematic inquiry into the operation of the mind. One of the primary factors that catalyzed new investigations of cognitive activities was Chomsky’s generative grammar, which he advanced not only as an abstract theory of the structure of language, but also as an account of language users’ mental knowledge of language (their linguistic competence). A more fundamental factor was the development of approaches for theorizing about information in an abstract manner, and the introduction of machines (computers) that could manipulate information. This gave rise to the idea that one might program a computer to process information so as to exhibit behavior that would, if performed by a human, require intelligence.
If one tried to formulate a unifying question guiding cognitive science research, it would probably be: How does the cognitive system work? But even this common question is interpreted quite differently in different disciplines. We can appreciate these differences by looking just at language. While psycholinguists (generally psychologists) seek to identify the processing activities in the mind that underlie language use, most linguists focus on the products of this internal processing, seeking to articulate the abstract structure of language. A frequent goal of computer scientists, in contrast, has been to develop computer programs to parse natural language input and produce appropriate syntactic and semantic representations.
These differences in objectives among the cognitive science disciplines correlate with different methodologies. The following represent some of the major methodological approaches of the contributing disciplines and some of the problems each encounters.
Artificial intelligence. If the human cognition system is viewed as computational, a natural goal is to simulate its performance. This typically requires formats for representing information as well as procedures for searching and manipulating it. Some of the earliest AI programs drew heavily on the resources of first-order predicate calculus, representing information in propositional formats and manipulating it according to logical principles. For many modeling endeavors, however, it proved important to represent information in larger-scale structures, such as frames (Marvin Minsky), schemata (David Rumelhart), or scripts (Roger Schank), in which different pieces of information associated with an object or activity would be stored together. Such structures generally employed default values for specific slots (specifying, e.g., that deer live in forests) that would be part of the representation unless overridden by new information (e.g., that a particular deer lives in the San Diego Zoo). A very influential alternative approach, developed by Allen Newell, replaces declarative representations of information with procedural representations, known as productions. These productions take the form of conditionals that specify actions to be performed (e.g., copying an expression into working memory) if certain conditions are satisfied (e.g., the expression matches another expression). Psychology. While some psychologists develop computer simulations, a more characteristic activity is to acquire detailed data from human subjects that can reveal the cognitive system’s actual operation. This is a challenging endeavor. While cognitive activities transpire within us, they frequently do so in such a smooth and rapid fashion that we are unaware of them. For example, we have little awareness of what occurs when we recognize an object as a chair or remember the name of a client. Some cognitive functions, though, seem to be transparent to consciousness. For example, we might approach a logic problem systematically, enumerating possible solutions and evaluating them serially. Allen Newell and Herbert Simon have refined methods for exploiting verbal protocols obtained from subjects as they solve such problems. These methods have been quite fruitful, but their limitations must be respected. In many cases in which we think we know how we performed a cognitive task, Richard Nisbett and Timothy Wilson have argued that we are misled, relying on folk theories to describe how our minds work rather than reporting directly on their operation. In most cases cognitive psychologists cannot rely on conscious awareness of cognitive processes, but must proceed as do physiologists trying to understand metabolism: they must devise experiments that reveal the underlying processes operative in cognition. One approach is to seek clues in the errors to which the cognitive system is prone. Such errors might be more easily accounted for by one kind of underlying process than by another. Speech errors, such as substituting ‘bat cad’ for ‘bad cat’, may be diagnostic of the mechanisms used to construct speech. This approach is often combined with strategies that seek to overload or disrupt the system’s normal operation. A common technique is to have a subject perform two tasks at once – e.g., read a passage while watching for a colored spot. Cognitive psychologists may also rely on the ability to dissociate two phenomena (e.g., obliterate one while maintaining the other) to establish their independence. Other types of data widely used to make inferences about the cognitive system include patterns of reaction times, error rates, and priming effects (in which activation of one item facilitates access to related items). Finally, developmental psychologists have brought a variety of kinds of data to bear on cognitive science issues. For example, patterns of acquisition times have been used in a manner similar to reaction time patterns, and accounts of the origin and development of systems constrain and elucidate mature systems.
Linguistics. Since linguists focus on a product of cognition rather than the processes that produce the product, they tend to test their analyses directly against our shared knowledge of that product. Generative linguists in the tradition of Chomsky, for instance, develop grammars that they test by probing whether they generate the sentences of the language and no others. While grammars are certainly germane to developing processing models, they do not directly determine the structure of processing models. Hence, the central task of linguistics is not central to cognitive science. However, Chomsky has augmented his work on grammatical description with a number of controversial claims that are psycholinguistic in nature (e.g., his nativism and his notion of linguistic competence). Further, an alternative approach to incorporating psycholinguistic concerns, the cognitive linguistics of Lakoff and Langacker, has achieved prominence as a contributor to cognitive science.
Neuroscience. Cognitive scientists have generally assumed that the processes they study are carried out, in humans, by the brain. Until recently, however, neuroscience has been relatively peripheral to cognitive science. In part this is because neuroscientists have been chiefly concerned with the implementation of processes, rather than the processes themselves, and in part because the techniques available to neuroscientists (such as single-cell recording) have been most suitable for studying the neural implementation of lower-order processes such as sensation. A prominent exception was the classical studies of brain lesions initiated by Broca and Wernicke, which seemed to show that the location of lesions correlated with deficits in production versus comprehension of speech. (More recent data suggest that lesions in Broca’s area impair certain kinds of syntactic processing.) However, other developments in neuroscience promise to make its data more relevant to cognitive modeling in the future. These include studies of simple nervous systems, such as that of the aplysia (a genus of marine mollusk) by Eric Kandel, and the development of a variety of techniques for determining the brain activities involved in the performance of cognitive tasks (e.g., recording of evoked response potentials over larger brain structures, and imaging techniques such as positron emission tomography). While in the future neuroscience is likely to offer much richer information that will guide the development and constrain the character of cognitive models, neuroscience will probably not become central to cognitive science. It is itself a rich, multidisciplinary research cluster whose contributing disciplines employ a host of complicated research tools. Moreover, the focus of cognitive science can be expected to remain on cognition, not on its implementation. So far cognitive science has been characterized in terms of its modes of inquiry. One can also focus on the domains of cognitive phenomena that have been explored. Language represents one such domain. Syntax was one of the first domains to attract wide attention in cognitive science. For example, shortly after Chomsky introduced his transformational grammar, psychologists such as George Miller sought evidence that transformations figured directly in human language processing. From this beginning, a more complex but enduring relationship among linguists, psychologists, and computer scientists has formed a leading edge for much cognitive science research. Psycholinguistics has matured; sophisticated computer models of natural language processing have been developed; and cognitive linguists have offered a particular synthesis that emphasizes semantics,