Catherine L. Harris
Keywords: cognitive revolution, cognitive linguistics, Chomsky, language acquisition, connectionism
Overview. Cognitive scientists have long debated whether language and cognition are separate mental faculties, or whether language emerges from general cognitive abilities.
What is the relationship between language and cognition? Do people who speak different languages think differently? Is a certain level of cognitive development required for language acquisition? These questions were of keen interest to thinkers in the early 20th century and have remained important today in anthropology, linguistics and psychology. However, the cognitive revolution of the 1950s brought a new question about the relationship between language and cognition: Is language fundamentally the same type of mental entity as other cognitive abilities, or is it fundamentally different?
A hallmark of modern cognitive science is the goal of developing a theory of cognition powerful enough to encompass all human mental abilities, including language abilities. A long- standing controversy concerns two ways of conceptualizing the architecture (or basic design) of cognition. One approach, called "Unified theories of cognition", proposes that general-purpose processes and mechanisms provide a foundation for all varieties of human intelligence. Examples of possible universal processes are the ability to induce a category from exposure to examples ("category induction"), and the ability to mentally complete a known pattern when confronted with a piece of it ("pattern completion"). Cognitive scientists generally force themselves to be specific about their proposed mechanisms by implementing them as computer algorithms which can be tested in artificial intelligence (AI) programs. Researchers have tried to show, using AI programs, that the same principles which can explain general problem solving can also explain aspects of language acquisition and processing.
The second basic way of conceptualizing human cognition emphasizes the differences between language and other abilities. A key idea is that many distinct domains of cognition exist and must be learned separately, using different mental mechanisms. This approach is referred to as the "modularity of cognition" or "mental Modules" approach. At first glance it may seem to go against the interdisciplinary spirit of cognitive science and against the possibility of a unified theory of cognition. However, the unifying theory precisely is the thesis of distinct mental modules, which are believed to have evolved to accomplish specific tasks relevant to mammalian evolution, such as visual exploration, or relevant to human evolution, such as language use. Much of the appeal of this approach comes from findings in neuropsychology showing that distinct areas of the brain serve distinct functions such as vision, language processing, motor coordination, memory, and face recognition. The interdisciplinary spirit is maintained because advocates of this approach reach out to biological scientists and evolutionary theorists. Those favoring modularity and embrace the principle of converging methodologies: a theory must have explanatory power in the distinct academic disciplines that compose the cognitive sciences.
These two approaches to the architecture of cognition developed out of different philosophical traditions and have evolved considerably during the half-century history of cognitive science. The next sections discuss how this evolution has shaped both research and debate on the language-cognition relationship, and the prospect for a synthesis.
At the dawn of the cognitive revolution, in the late 1950s, were two distinct ideas about the nature of mind. Discussed first are the views of linguist Noam Chomsky and the field of generative linguistics which he developed, because the underlying philosophy has remained fairly constant over the last 50 years.
Chomsky's major innovation was to conceive of language abilities as akin to a mental organ. On this view, children are born with a "language acquisition device" and with specific linguistic knowledge. This knowledge is thought to include the concepts of noun, verb, grammatical subject, and structures which constrain possible grammatical rules. In contrast to the views of the dominant psychological theory of the 1950s, behaviorism, Chomsky argued that children do not learn to speak by imitating adults. Key evidence was that children spontaneously use incorrect forms they could not have heard, like goed and breaked. Linguistic cverregularizations like these suggest that children are extracting rules from the language they hear, not merely imitating. Theorists at that time found it noteworthy that parents do not generally tell children that their utterance are ungrammatical. Because the language input to children is full of mistakes, stops and restarts, Chomsky felt that children could not learn language using general purpose problem solving or regularity-extraction skills. They needed to come to the task with a rich set of expectations about the nature of language. These expectations were believed to be specific to language, and thus did not share commonalities with other aspects of cognition. The name for this set of language-specific abilities has been variously called the language acquisition device (the historically early term) and universal grammar (a historically more recent term). Chomsky's approach to linguistics is called Generative Linguistics because an early goal was to describe mental structures which can generate all the grammatically valid sentences of a language.
Chomsky felt that the aspect of language which is most unique is syntactic ability. An example of specifically syntactic knowledge is illustrated by the sentence Colorless green ideas sleep furiously. Although the words in this sentence contradict each other and do not correspond to a possible reality (green is not colorless, ideas can not sleep furiously) speakers nevertheless recognize the sentence as having a correct grammatical structure. This example was used by Chomsky to argue that syntactic structure is not reducible to meaning, and thus represents a unique, independent human capacity, which is not derivative from other abilities.
Why did Chomsky feel so strongly that language was a unique mental faculty, different from general problem solving and different from other cognitive abilities? One of Chomsky's students, who later became a prominent critic of his views, is George Lakoff. Lakoff claims that Chomsky's training in the tradition of analytic philosophy is what led him to emphasize a sharp discontinuity between language and the rest of cognition. The analytical tradition holds that human reason is a unique capacity and works by its own rules and principles. In analytic philosophy, the ability to reason is the ability to manipulate representations according to formal rules independently of emotion and imagination. The highest form of human reasoning was believed to be logical and mathematical reasoning. Chomsky extended this basic philosophical idea to the realm of language abilities, proposing that the ability to use language was the ability to manipulate arbitrary symbolic representations. A key example was the idea of pure meaningless syntax, as illustrated by the syntactically well-formed but meaningless sentence given above (Colorless green ideas...). To Chomsky, syntax is the most creative part of the human mind, because it creates, from nothing external to itself, the structures of language upon which all human rationality is built. Because Chomsky's philosophy about language grew out of the analytic tradition, he required that language be an autonomous faculty of mind, independent of memory, attention, perception, emotion and gesture, social interaction and culture, contextual knowledge, needs of interpersonal communication
The innovativeness (and controversy) of Chomsky's ideas helped ignite the cognitive revolution of the 1950s, spurred decades of research and has had a foundational and continuing influence on cognitive science. A very different set of ideas were also at work during the dawn of the cognitive revolution, and these had different implications for understanding the relationship between language and cognition. With the birth of the field of computer science, a new way of conceptualizing human cognition arose using the metaphor of the brain as a computer and the mind as software. Thinking of mental operations as akin to steps in a computer program allowed psychologists and workers in the new field of artificial intelligence to begin to subdivide mental tasks, such as arithmetic or language comprehension, into a series of steps in a computer program. Psychologists like Alan Newell and John Anderson, and computer scientists like Roger Schank and Patrick Winston began to describe a range of human abilities -- from visual object recognition and general problem solving, to metaphor use and story understanding -- in terms of a set of internal representations and processes which transform those representations. A key aspect of this approach to cognition was that the theorist could write computer programs which would provide a formalizable theory of mental operations. A rigorous test of the theory could be performed by running the program and seeing if output matched human output.
Many of the early successes of this field involved language, including sentence comprehension, story understanding, and metaphor use. The history of successes and failures of the information processing movement are discussed elsewhere, but the key thing for the language and cognition relationship was that there was no reason to believe that language required special algorithms, knowledge or structures separate from other aspects of cognition. The new AI tradition and the information processing movement within psychology emphasized learning, particularly general purpose learning, and was thus opposed to the emphasis on innate knowledge structures that was part of Chomsky's new linguistics. Psychologists who were drawn to the information processing movement frequently felt inspired by the work of Swiss psychologist Jean Piaget, whose works began to be translated into English in the 1960s. Piaget also emphasized the commonalities between language and cognition, and proposed that language emerged out of the same broad cognitive changes that transform the sensory-motor processing of infants into the formal and logical mind of adults.
In the 1970s and 1980s, among those researchers who aligned themselves with the interdisciplinary field of cognitive science, there was a tendency for linguists to emphasize the specialness of language and cognition, and psychologists to emphasize commonalities between language and cognition. However, there was considerable variation in viewpoint within psychology and linguistics, which continues to this day. The next sections discuss three scientific developments during the 1980s and the 1990s which had implications for theories of the language- cognition relationship.
The "connectionist revolution" was launched in the mid 1980's. A new computational metaphor emerged based not on the Von Neuman computer, but on the idea that sophisticated computation emerge from massive networks of simple processing units, in which the units are akin to idealized neurons
In the early 1980s several groups of cognitive scientists became dissatisfied with the artificial intelligence systems being used to model cognition. These rule-governed systems employed databases containing knowledge of common situations, such as what one does at a restaurant. If-then rules specified the actions that an expert would take in specific situations. This expert knowledge thus constituted intelligent reasoning. Critics noted that these expert-systems were brittle and frequently failed to generalize outside of the bounds of their circumscribed database. Real human cognition seemed intuitively to involve not the application of fixed rules to a specific situation, but the satisfaction of many soft constraints, some of which could be ignored. Categories were not logical sets of necessary and sufficient conditions, but were graded groupings which shared "family resemblance," as had been argued earlier in the century by Wittgenstein. To the mathematical and cognitive psychologist David Rumelhart, flexible and creative reasoning seemed to similar to the motor process involved in reaching for a cup behind a pencil holder one one's desk -- or at least, more similar to this type of planning than to rule-governed algebraic reasoning.
Rumelhart and other cognitive scientists such as Terrence Sejnowski and Stephen Grossberg agreed that mental functioning involves computation, but asked what type of computation would be carried out by an organic structure like the brain, composed of massive numbers of simple processing units (neurons), linked together in a complex network of fiber tracts, with many units firing simultaneously. This movement was labeled connectionism because intelligent behavior was posited to emerge from large numbers of neuron-like processing units, connected together into networks in ways that fostered parallel processing. One early success story was a connectionist model which could learn English past-tense forms given the present-tense of a verb. The domain of English past-tense was provocative because it contained both rule- like behavior, (past-tense forms are generated by adding -ed to a present-tense form), and exceptions, which themselves congregated into patterns, such as the pattern that emerges from considering grow-grew, blow-blew, know-knew. Both connectionist networks and adult second language-learners are likely to make the error of generated "glew" for the past-tense of "glue", presumably on analogy to this sub-regularity within the past-tense system. Other connection language models explored how the ambiguous words are understood in their sentence context (such as bat in "The boy hit the bat"), and how abstract categories such as noun and verb can emerge from distributional regularities in text (i.e., the fact that nouns tend to occur in similar sentence positions). Because the basic system of processing units under these connectionist language models were the same as used to model visual and motor behavior, successes like the past-tense model were taken as support for a common computational architecture underlying both language and cognition.
A subset of linguists disagreed with Chomsky's (and the field's) emphasis on the uniqueness and specialness of language. The field of cognitive linguistics emerged in the late 1980's and helped initiate a flood of work connecting language and cognition.
One source for the cognitive linguistics movement came from an older tradition within linguistics called "functionalist linguistics." These linguists believed that constraints on the form of language (where form means the range of allowable grammatical rules) derive from the function of language. The goals of using language include serving efficient communication. Thus the function of diverse syntactic forms such as the subject vs. object distinction, and main clause vs. subordinate clause distinction, serve communicatory goals such as conveying which information is most important and which is background or context. But the goals of conveying the relevance and background/ foreground structure of the message are outside the language system. The proposal that the form of grammatical rules is influenced by communication is thus inconsistent with the autonomy of syntax hypothesis. That hypothesis specified that syntax was its own system, not shaped by the need for efficient processing or other exigencies of communication.
Functionalist linguists as well as other linguists were dissatisfied with the range of linguistic phenomena excluded by generative linguistics. They didn't accept the Chomskyan view that the most important aspect of language was a mechanical device for generating only the legitimate grammatical sentences. They wanted to understand language in all its diversity, including narrative, discourse, dialects, social cultural influences on language use and metaphor. Some functionalist linguists even proposed that rule use is just a minor aspect of language. Linguists like Dwight Bolinger and Charles Fillmore argued that speech utterances, not rules for generating utterances, are what is mentally stored. They noted that every language speaker has memorized huge numbers of odd coinages, colloquialisms, idioms and collocations, many of which share patterns. An example is the phrase "know by heart" which has the variation "learn by heart." Rules, partial regularities and rule- exceptions appear differ in degree, occupying a continuum from fully idiosyncratic, to partially regular, to fully rule- governed. This idea made researchers in this nascent linguistic movement sympathetic to the connectionist movement, although to date, the commonalities have not been fully explored.
Although trained in the generative grammar tradition, linguists like George Lakoff and Ronald Langacker noted that one couldn't describe which sentences are syntactically valid and which are invalid unless one made reference to nonlinguistic concepts. They pointed to the tendency for the same words describing movement in space to be used to describe movement in time ("This meeting runs until 3 o'clock.), and emphasized the necessity of incorporating the cognitive psychology of human category formation into linguistics.
This approach to linguistics came to be called cognitive linguistics, because aspects of general cognition, such as how we construe the meaning of a grammatical construction, were proposed to be important for describing linguistic structure. For example, one descriptive problem of grammar is to account for why some sentences, but not others, can undergo the passivization transformation. The sentence "John was hit by Mary" sounds fine, but "John was known by Mary" does not. To fully describe which transitive sentences can undergo passivization, one needs to invoke the notion that the subject and direct object are in dynamic interaction which each other. Thus, one cannot passivize a sentence like, "John left the auditorium," because John is not acting on the auditorium in a way that has consequences for it. But the sentence "John left the auditorium unguarded" can transformed into "The auditorium was left unguarded by John," presumably because John's action affects the status of the auditorium.
By 2000, the cognitive linguistics movement had grown into an enduring subfield within linguistics with considerable international members. However, it has remained outside the mainstream of linguistics and cognitive science. Proponents were intrigued by cognitive science research, such as the finding that abstract categories can be inductively learned by artificial neural networks. However, the majority of cognitive linguists have remained focused on specific linguistic questions, rather than addressing questions in an interdisciplinary way, as would cognitive scientists.
In the 1990s the field of cognitive neuroscience emerged from work in neuroscience and cognitive science. Cognitive neuroscience differs from basic neuroscience by having the goal of explaining complex cognitive abilities, but rejects the tradition from artificial intelligence (and much of cognitive science) that one can understand cognition abstractly, without reference to its neural underpinnings.
In the 1990s, some cognitive neuroscientists argued that basic questions about the language-cognition relationship, such as the autonomy of syntax hypothesis an the innateness and modularity of language can be evaluated from the neuroscientific point of view. Neurobiologists have noted the pluripotentiality of developing neural tissue, especially the regions of the brain which mediate language use. Like other aspects of cognition, language acquisition is heavily dependent on experience.
R-A Mueller has remarked that regional specialization in the brain is beyond doubt, but modularity of cognitive functions, including language, is highly debateable from the view of neurobiology and evolution. Functional specialization of brain areas most likely emerge because some brain areas are near to sensory input, such as sensory systems for vision and audition.
Scientists such as Jeffrey Elman, Elizabeth Bates and colleagues note that however closely one looks at the anatomy and physiology of the brain, there is no evidence of cortical structures unique to language or unique to humans. These researchers emphasize that language has an "epigenetic" not a "genetic" origin. There aren't enough genes in the human genome to code directly for outcomes such as the ability to use language. Saying that language areas develop epigenetically means that they are end products of complex chains of interactions with internal and external environments. These sequences of events are based probabilistically on genes rather than being rigidly determined by the genome. The concept of epigenesis dates back to psychologist Jean Piaget, who argues that cognitive abilities emerge from a biological structure which evolves, both before and after birth, in tandem with environmental forces.
The neurobiological evidence thus may run counter to what would be expected under the autonomy of syntax hypothesis. There is no known way that genes could encode for concepts like subject and verb. Thus the most parsimonious perspective is that language is similar to other aspects of cognition in terms of emerging out of a brain which evolved to have an oversized frontal cortex (relative to other primates) and an elongated period of childhood which privilege the role of learning.
These cognitive neuroscientists share a view of language which resonates with the cognitive linguists: they emphasize the codevelopment of language and perceptuomotor processes, with language acquisition understood to be semantically driven and embodied. The neurological representation of grammar is continuous with the representation of other language "components" and the neural substrate.
In the first half of the 20th century the main question about the relationship between language and cognition was whether the grammatical structure or vocabulary of our language influence thought processes. Cognitive science introduced a new question: Are language and cognition similar or distinct human abilities? The last 50 years have seen considerable controversy on this question, mirroring the development within cognitive science of two fundamentally different conceptions of the cognitive architecture. The tradition in artificial intelligence emphasized general purpose problem solving abilities, while the tradition from linguistics and philosophy led to an emphasis on distinct mental modules.
Table 1 summarizes four different theoretical perspectives on the language-cognition relationship. The view at the beginning of the 21st century appears to be best captured by the idea that cognition and language have complex similarities and differences, and both develop over the human lifespan from genetic factors being constrained by environmental input and cultural learning.
Table
Timeline Movement Main source of constraints Language/ cognition relationship
------------------------------------------------------------------ -----------------------------
'57-present Chomskyan linguistics innate language unique, unlike cognition
'60's-1990 Artificial intelligence learned subject to same principles
'80's-1990 Connectionism learned subject to same principles
'80's-present Modularity of mind innate language unique, unlike cognition
'90's-present Cognitive neuroscience dynamical interaction complex similarities and differences
New possibilities for synthesis continue to emerge, especially as cognitive scientists pay more attention to evolutionary, neurobiological and cultural factors. The field may currently be in a position to set aside the question of whether language is distinct from cognition and whether the brain is composed of distinct mental modules. The theorist Howard Gardner has noted a growing consensus about the importance of a new set of questions about how to divide up the grand areas of mind and brain. Scientists are emphasizing the distinction between areas of human ability that are available to all humans and played a role in evolution of our species (such as language and basic number use), compared to those requiring cultural elaboration (such as algebra and the ability to play musical instruments). The era of simplistic statements about the language- cognition relationship is drawing to a close, as cognitive scientists begin to deliver on the promise of a truly interdisciplinary approach to understanding the mind-brain.
VIII. Bibliography
Baars, B. J. (1986). The cognitive revolution in psychology. New York, NY: The Guilford Press.
Edelman, G.M. (1992). Bright air, brilliant fire. New York, NY: Basic Books.
Dromi, E. (Ed., 1993). Language and cognition: A developmental perspective. Norword, NJ: Ablex.
Elman, J.L., Bates, E., Johnson, M.H., Karmiloff-Smith, A., Parisi, D., and Plunkett, K. (1996). Rethinking innateness: A connectionist perspective on development. Cambridge, MA: MIT Press.
Gardner, H. (1985). The mind's new science: A history of the cognitive revolution. New York, NY: Basic books.
Gumperz, J. J., and Levinson, S. C. (1996). Rethinking linguistic relativity. Cambridge: Cambridge University Press.
Harris, C.L. (1990). Connectionism and cognitive linguistics. Connection science,
Lakoff, G., and Johnson, M. (1999). Philosophy in the flesh. New York, NY: Basic Books.
Langacker, R.W. (1987). Foundations of cognitive grammar: vol. 1. Theoretical prerequisites. Stanford, Ca: Stanford University Press.
MacWhinney, B. (Ed., 1999). The emergence of language. Mahwah, NJ: Erlbaum.
Morris, R.G.M. (Ed., 1989). Parallel distributed processing: Implications for psychology and neuroscience. Oxford: Oxford University Press.
Muller, R.-A. (1996). Innateness, autonomy, universality? Neurobiological approaches to language. Behavioral and Brain Sciences, 19, 561-610.
Newell, A. (1990). Unified theories of cognition. Cambridge, MA: Harvard University Press.
Pinker, S. (1994). The Language instinct. Cambridge, MA: MIT Press.
Rumelhart, D., and McClelland, J. (1986). Parallel distributed processing: Explorations in the microstructure of cognition, Volumes 1 and 2. Cambridge, MA: MIT Press.
Architecture of cognition#General term for the design of the human cognitive system. Specific proposed architectures may differ in their structure, such as the types of memory and processing structures.
Artificial Intelligence# The field of study dedicating to understanding how intelligence emerges from computational systems such as computer programs.
Behaviorism#The dominant paradigm with psychology from the 1930s through the 1950's. Claimed that understanding human and animal behavior can be achieved by analyzing the relationship between stimuli presented to an animal and its response.
Cognitive revolution#The 10-20 year period beginning in the 1950s when the then-dominant Behaviorist paradigm floundered and gave way to new ways of thinking about human psychology and the mind, including the use of the computational metaphor of mind, and the idea that language consists of a mental grammar which is a set of unconscious rules.
Epigenetic#Behavior results from a complex dynamic evolution of genes and environmental forces during prenatal and post-natal development.
Generative Linguistics#The dominant paradigm in linguistics from the 1960s through the beginning of the 21st century. The name refers to Noam Chomsky's proposal that the goal of linguistics should be to understand how novel sentences are generated.
Language acquisition device# Linguistic relativity hypothesis# The proposal that different languages are incommensurate systems of communication and thus utterances in one language can only be understood relative to the broader language system. Related to the Sapir-Whorf hypothesis, which is that forms of language influences thought. Named after the anthropological linguistics of the first half of the 20th century who wrote the uniqueness of different language and the necessity of evaluating a culture on its own terms (cultural relativity).
Regularity extraction#General term for inducing the dominant statistical tendencies in a set of data.
Information processing#A view of human cognition which emphasized analyzing large tasks into their subcomponent processes and information structures, on analogy to steps in a computer program.
Von Neuman computer#The original computer architecture consisting of registers in which arbitrary information bits could be stored, and a centralized processor which serially moved information between register locations.