In Gavagai: Revista Interdisciplinar en Filosofia de los Lenguajes 2/1, 1986, 75-104.

Psychology of Language: a Field in Transition

Robert de Beaugrande

1. Regress, progress and epistemology in classical science.

During the 17th, 18th, and 19th centuries, numerous scientific discoveries merged into an unprecedented matrix of interrelated knowledge that enabled a new technological manipulation of the environment. This period might be called the classical period of science, during which the commonplace ideal of what science is and does was established for society at large and still holds force in our times. The dominant epistemology is based on observation and measurement and seems immune to skepticism because it is itself so emphatically skeptical.

Classical science maintains the optimism that we can explain any natural or historical object we choose, provided we are willing to expend the necessary research. From a psychological standpoint, this scheme constitutes a structure of regress, where ‘regress’ is defined as ‘relating the current state of an object to the conditions that brought it about’ (see now Beaugrande 1997). We may investigate the object in terms of temporality (how it evolved in time), causality (what caused it), complexity (what whole it is part of, or what parts it has), and exemplarity (what general law or principle it illustrates). Conversely, we may prefer to investigate the object in regard to its progress: how it will evolve in the future, what effects it will have, what new whole or parts it will form, or what new law or principle it will be subjected to.

This basic pattern of regress and progress can represent the two fundamental perspective for science to apply to an object.[1] At some point, regress and progress are terminated because they cannot be extended without end on any single occasion. The basic procedure of classical science is to select an object, to probe its regress, and then to envision its progress. Hence, explanation entails situating an object within a structure of regress and progress by demonstrating how the object came to be such and by predicting what will happen to it under certain conditions. The experiment is a procedure for intervening in the state of the object within a specific and controlled situation. A failed experiment is typically viewed as an exceptional interlude due mainly to inadequate equipment, unfavorable accidents, or badly formulated hypotheses. It is much less often considered that the experiment as such may be an inappropriate structure of regress and progress for certain objects.

On a large scale, we can define each science as a set of established procedures for devising explanations in the sense just described. Thus, ‘normal science’ as described by Thomas S. Kuhn (1970) is a domain in which regress and progress have been expressly restricted to certain authorized channels wherein they terminate at characteristic points. Moreover, the classical normal sciences form a general regress in their relationships to each other, at least in the view of the general public (see Beaugrande 1997: Ch. III for extensive discussion). Mathematics plus formal logic are commonly placed at the start, followed by physics plus chemistry, and then by biology. The facts of biology, that is, the natural laws, are guaranteed by the physical laws of physics, and the latter in turn by the axioms of mathematics.[2]

For mathematical explanations, such entities as probabilities, sets, and proportions are central. In physics, the mechanics of cause and effect are decisive, whereas biology centres its explanations on genetics and the environment.

What is not appreciated in the popular view is that this regress often entails a steady loss of content and richness in detail. Abstraction is usually performed via a certain loss of immediacy in respect to experienced reality. A biological explanation of the behavior of organisms in terms of their brute physiological and chemical construction is strongly reductive; and so is the referral of the physical world back to the principles of pure quantification, measurement, and identity. Only the advent of computer simulation, which combines a procedural version of mathematics and logic with a domain model of any degree of abstractness or concrete detail, has begun to counterbalance this reductionism.

During its formative period, each of the human sciences had to decide where it should situate its regress within this scheme. The status of its credentials depended critically on its declared alliances to the classical sciences. Psychology aligned itself with all three at once, the highest prestige often going to the mathematical branch. Linguistics tended to shift its allegiance among the three, though here too, the hope of referring the entire discipline back to mathematics and format logic periodically became the highest bid for authority: witness the ambitions of glossematics, generative grammar, Montague grammar, and so on.

Thus, psychological explanations combined statistics, causalities, and the structure of organisms, whereas in linguistics, the question of what constituted an explanation was never definitively answered, shifting each time the path of regress was changed. In both disciplines, the regress to the classical sciences entailed a characteristic reduction and loss of content, intensifying with the movement toward the purely mathematical or formal end. The danger thus impended that the most prestigious human theory with the most reductive point of regress might also be the theory that says the least about its human domain.

The optimism of classical science has gradually faded during the rise of the human sciences. The 20th century has witnessed, though often not acknowledged, a pervasive erosion of certainty as it has become apparent that the uniform progress of classical science was purchased at the price of a drastic reduction of epistemology as a human factor. In their rush to know, scientists failed to consider the status of knowing as such. The exaltation of the objective and the suppression of the subjective left a backlog of unresolved problems whose urgency and difficulty become evident as we undertake to formulate them.

The model of regress and progress proposed above becomes much more complicated for the objects of the human sciences. The mathematical, physical, and biological properties of the human being are at best boundary conditions and cannot provide by themselves a satisfactory explanation of the human situation. The classical channels of temporality, causality, complexity, and exemplarity do not encompass humans in a direct or simple way. Instead, such channels of regress and progress as cognition, motivation, and interpretation must be admitted. Human actions are seldom ‘caused’ in the same sense that applies to classical physical substances, e.g., the way heat ‘causes’ the expansion of matter. Human actions may be related to complex configurations of knowledge, beliefs, purposes, reasons, and so on. Therefore, the classical research paradigms are insufficient, though we have no consensus as yet about how they ought to be modified or replaced.

In some early psychological research, humans were regarded as ‘organisms’ whose behavior was directly controlled by ‘conditioning’ from the environment (e.g. Pavlov, 1927; Skinner, 1938). Manipulations of the environment were considered a sufficient causality to control and explain human behavior. Standard experimental designs paid little attention to people's prior knowledge and predispositions as ‘information processors’ (a term no one would have used in that early period). This attitude is easily understood in terms of the regress among disciplines as the reduction of the human to mathematical, physical, and biological factors.

More precisely, psychology was borrowing the channels of regress established in mathematics, physics, and biology, on the assumption that this regress was a necessary foundation for being ‘scientific’ at all. The loss of content was particularly acute. Motivation could be admitted only in terms of external stimulation by means of immediate rewards and punishments. Cognition and interpretation were discounted entirely, because they do not stand in a relationship of classical physical causality to an external situation.

The stance of early ‘classical psychology’ was a radical solution to what I would consider the most fundamental problem for human research of any kind. On the one hand, human action as an empirical fact is not found in a vacuum, but in natural contexts. Each action possesses not merely its external properties (e.g. motion, position, energy), but also its valence within the overall context of social action and interaction. On the other hand, a theory of human behavior must be developed by abstracting away from specific contexts to distill out what is typical, general, and repeatable. Thus, the main problem is: which aspects of context should in fact be leveled or discounted, and which should be retained in a theory?

In early research,[3] the answer to this question was a simple one: isolate language as far as possible from its natural contexts of communication. The most popular experimental material, nonsense syllables and word lists, bore at most a remote and superficial resemblance to ordinary discourse. Researchers hoped no doubt that factors apparently excluded from the design could safely be considered immaterial for the findings. The difficulty is that language is typically much more complex than it seems when people are using it, because each use presupposes much prior knowledge about how to communicate.

A comparable removal of communicative contexts was under way in ‘classical linguistics’ (cf. Beaugrande, 1984a). The famous divisions between ‘langue’ and ‘parole’ by Saussure (1916) and between competence' and ‘performance’ by Chomsky (1965) offer evidence of this trend. It is hence unlikely that language psychology could expect classical linguistics and psycholinguistics to offer significant help in solving the problem of context. On the contrary, the epistemology whereby a linguist can abstract the system of a language from any corpus of evidence or experience is as poorly explained so far as is the epistemology whereby a psychologist can classify a language event as a ‘stimulus’ or a ‘response’. If language categories remain undefined or at least underdefined until put to use in context, the removal of contexts might obscure the essentials of language as a communicative manifestation.

Saussure postulated an abstract system of language (‘langue’) viewed ‘synchronically’ in the perspective of the current state of the total language. The origin, evolution, and operation of the system were excluded from scientific inquiry. However, the abstract system is not an empirical entity; it can only be inferred from a set of specific language actions among human users. And these actions presuppose a complicated epistemology of language. The ‘synchronic system’ functioned as a termination point preventing regress into that epistemology. Accordingly, classical linguistics never attempted to explain how a theory of language could be constructed, that is, how the knowledge and experience with a language could be transposed into a theory or model of that language (see Beaugrande 1991 for detailed analysis).

A classical concept for terminating regress was ‘arbitrariness’. Lacking any obvious unified means to explain the relationship between language and meaning, researchers would simply take that relationship as given and declare it ‘arbitrary’. Here's Terence Hawks (1977: 24ff) following Saussure:

[What is] ‘arbitrary’ [is] self-contained and self-justifying: there is no appeal possible beyond it to some category of the ‘natural’. or the ‘real’ […] the arbitrariness of the linguistic sign is not ‘reasonable’, and so it cannot be discussed in the sense that we cannot profitably consider or debate its adequacy. The sign is simply there.

This tactic was necessary because the meaning and function of language are not explainable within the classical cause-effect paradigm of physics or the genetic/organic paradigm of biology. The quasi-mathematical principle of ‘arbitrariness’ was appropriated to suggest the randomness of language formation.

However, the relationship between language and meaning is by no means arbitrary or random if viewed within a properly epistemological paradigm. Different members of the same language community acquire and utilize the language in reliably comparable ways. Humans could not process meaning arbitrarily without impeding communication. The principle of textuality controls the actions people perform with the language (cf. Beaugrande & Dressler, 1981). By discounting ‘parole’ or ‘performance’, classical linguistics blocked the regress into the epistemology that might have explained how people know what they mean.

In Saussure's conception of the language system, ‘difference’ was also a main termination point. If we regress behind it, we attain the problematic thesis that all meaning presupposes a systematic process of differentiation. Yet classical linguistics did not explain how this process might work. Saussurian theory acknowledged that only selected differences are relevant for the system. In phonology, articulatory actions served to define ‘minimal features’ such as ‘voiced’ versus ‘unvoiced’. But when linguists set about to define ‘semantic features’ in a comparable way, no such easy point of orientation was possible, because the epistemology of meaningful actions had been excluded from linguistic investigation. What we need are not huge taxonomies of ‘semantic features’ imitating the classifications of logic, physics, or biology, but models of signification (including the making of taxonomies) as a psychological action.

‘Generative’ and ‘transformational’ linguistics sought to escape the dilemmas of the taxonomical orientation. Formal rules were devised for describing. the patterns of word sequences. However, it was assumed that these rules could be stated while abstracting away from meaning and purpose (Chomsky, 1957). This truncation had to be offset with an explosive proliferation of syntactic rules obviously unlike the actual procedures humans use in discourse. The later speculation that ‘grammar’ was ‘innate’ became necessary because rules seemed impossible for the average person to learn.

The only safeguard I can see against such dangers is for each science to develop an explicit epistemology of regress that clarifies the status of its termination points and offers them as opportunities for new theoretical progress via intensified regress. Psychology is both the study of human cognition and performance, and a cognitive and performative enterprise. Therefore, psychology needs theories that include its own research (cf. Neisser, 1982). Behaviorism fails completely, because it is pointless to say that the experimental findings are the ‘stimulus’ and the psychological theory is the ‘response’. No criteria are offered for preferring one theory over any other that might fit the same evidence. And the action of theory building cannot be sensibly analyzed into external behavior.

The situation of language psychology is similar: both the study of language and a meaningful communicative activity in its own right. This field might fill the gap left by the exclusion of epistemology from linguistics. The major issue would be how language is functionally coordinated with other aspects of human cognition and performance, such as action, planning, memory, schema formation, and so on (cf. surveys in Beaugrande, 1980, 1984a, 1997). The same framework could clarify how experimental settings influence the uses of language performed by monitored informant and the introspective investigator.

As it now stands, human language psychology is not merely an incomplete picture. Its entire status and validity as a human enterprise have yet to be clarified. We have assembled a large array of specific findings, some of them quite robust, but no comprehensive theory of discourse processing that could account for the wider significance of the findings, or that could reconcile disparate findings. Psychologists and psycholinguists are naturally language users themselves; as such, they have been decisively affected by their own experiences with language in context. This corpus of experience necessarily creates rich presuppositions about what language is and does. These presuppositions are in turn entailed, often without acknowledgement, in theories and in interpretations of experimental findings. Taken seriously, this fact means that all extant theories and findings need to be validated by research on the epistemology of cognitive and communicative predispositions.

2. Psychology and processing.

We can now return to elaborate the basic model of regress and progress. The framework I shall present implies that cognitive categories be projected as performative categories, a trend that psychology has only begun to indicate. That is, knowing is to be modeled as a complex event during which certain classes of processes are carried out. It might seem that the acknowledgement of cognition as a factor is still too recent for this next stage to be evident; but I predict this direction is the one we will see emerging in the coming years (cf. Beaugrande, 1980, 1980-81, 1984a, 1984b, 1997).

Our fundamental concept is significance, defined as ‘any symbolic relationship created between non-identical objects or events’. Significance arises whenever an organism recognizes that two or more events or objects are similar or different, or that they do or do not tend to occur together in experience. Thus, significance could be absent only if an organism would perceive and interact with the world in a totally random way -- a condition for which there is little evidence. Knowing can be defined as ‘having awareness of significance, whether consciously or not’. Knowing enables the preservation and hence the elaboration of significances into steadily more complex configurations. Meaning can be defined as ‘performing an action upon known significances’. Aspects of experience ‘have meaning’ or ‘are meaningful’ when humans give them significance and then apply that significance for a relevant purpose.

This basic scheme gives us the structure of regress and progress for a processing system, that is, a system that functions by performing actions of knowing and meaning upon significances. Knowing and meaning appear as the progress of significance and also have significance as their regress. Since meaning is always a mental event, not an object or property, it may affect the significances it uses. When people use a known concept to comprehend or manage a new experience, they may expand or adapt the concept, at least in some details, to fit the experience. Still, though the formation of a concept may never be totally definitive or completed, it is stabilized when it becomes a reliable termination point for regress. This same process is also the basis for a consensus about meaning: different systems attain comparable results by maintaining similar structures of regress.

However, the fact that the resources of the human processing system are limited is one of the most secure findings in psychology, as shown for instance by classical studies of performance, interference, distractor tasks, and so forth. Accordingly, regress and progress must be restricted and balanced. Terminating regress should be among the most decisive means for conserving the resources that can then be devoted to progress.

When complex significances become sufficiently well established through operational use, they function as customary points for the termination of regress. They are not projected back to their origins or components during the acts of cognition or communication. This process is presumably iterative and open-ended, depending on the nature of the organism. The higher-level the organism, the more capable it is of extending the process in principle, though not necessarily in practice. The termination of regress must be an essential precondition for the evolution of intelligence within a system of limited resources. Indeed, to say that a task has. been learned is to say that the skills it requires have reached an operational point of termination, so that performance does not require a heavy. load of resources.

The traditional problem of subject and object can be restated in this framework. The subject is the only agent capable of creating significances. The object is a termination point beyond which the subject does not pursue regress. At that stage, the object is granted independent status and its existence is no longer made to depend on the subject's knowing about it. Reality is the configuration of objects with that status, and is therefore also a complex termination point for regress. That is, the object and reality are means for truncating epistemology in order to reduce the operational load on the system. Of course, since reality cannot be known except through this activity in the subject, reality can be reopened to regress back into epistemology. Yet this process is likely to impose a heavy load on the system, especially in respect to objects that originated from a complex elaboration of significances.

Even so, important progress undoubtedly entails regress as well -- not a simple forward movement, but a looping back behind a state that was a termination point before. The reason is that reality cannot be changed as long as it is taken as an independent world we must observe and accept. Change can begin only when we comprehend the functioning of reality and the conditions that enable it, including our own contribution as subjects.

Presumably, the evolution of a processing system occurs by means of this progress through regress. The subject recognizes its current state as a termination point which is not simply inherent in the nature of things, but which has evolved from the function of processing. The state can thus be transposed to a higher level from which new progress can be undertaken.

In my opinion, intelligence is best regarded as the potential of a system for this mode of transposition, and not as an index of performance on specific tests on a given occasion. In principle, the level of intelligence is always mutable rather than fixed by the heredity or construction of the human organism. The myth of fixed intelligence arose by naively treating performance scores as termination points, encouraged by the rampant commercialization of psychometrics.

However, many practical factors can intervene to impede the development of the system. From the lower level, the higher level may appear unmanageable or incomprehensible. The evolution of the system would not be a simple linear progression from lower towards higher levels, but a complicated shifting up and down for purposes of reorganization; and this would be effortful and disorienting. The organism would have not merely to realize that its current level is inadequate and in need of revision; but also to envision and test the design of a higher level, probably near to overload conditions, since the higher level could not be managed from within the lower unless additional resources were made available, or unless only certain aspects were focused at any given moment.

Here, we may see why intelligence so often does not evolve: the lower levels fall to navigate the design of the higher because pursuing regress beyond one's current state consumes too many resources. Learning psychology has placed so much emphasis on the training of isolated tasks that insufficient consideration was given to the design of the learner's overall processing system. Findings indicate that learning demands a revision of old habits that had become termination points; but how this occurs in natural social settings is poorly explained so far.

It should be also mentioned that the dominant educational system stresses low-level performance, such as the repetition of facts, and typically neglects the conditions that would favor high-level skills, such as constructive and creative reasoning (cf. Papert, 1980). In particular, the constant stigmatization of errors discourages learners from attempting to move to higher, unfamiliar levels where a temporary increase in errors is natural if not essential. Evidently, educational theory has not properly recognized that major progress demands regress, and that this bi-directionality demands great effort and favorable circumstances for transcending habits of thinking, including those habits that the educational system itself is so emphatically devoted to inculcating.

In theory, the most efficient system should be the one that automatically selects and runs on the highest level (cf. Drewnowski & Healy, 1979; Marcel, 1980). The manipulation of complex units should require less effort than that of the total of their individual components, as is plain from the studies of ‘chunking’ originated by G.A. Miller (1956). Research shows that complex actions can be performed rapidly and efficiently only if the processor develops corresponding communities or packages of simple actions and does not analyze them during operation (cf. Schmidt, 1975; Schneider & Shiffrin, 1977; Shiffrin Schneider, 1977; Norman & Shallice, 1980). Or, in my terms, high-level skills require that processing terminate regress before the lower levels.

However, it may be very hard to decide just what the best level of complexity should be for future occasions. A totally novel situation might not be manageable in terms of an already elaborated high-level design. The system would then need to shift back to a lower level and try synthesizing the appropriate complexes on the spot; and this effort would dramatically increase the load of the system.

The conclusion would seem to be that flexibility rather than maximal efficiency should be the highest value in psychological theories of human processing. The system should possess the capacity to include its own design as an object that can be reopened to regress when new tasks and requirements are encountered. This principle holds both for everyday activities and for scientific inquiry. Still, the stabilization of reality, in ordinary life or in science, as an impassible termination point, is a constant danger, due to the very need of the system to conserve resources. Research needs to remain aware both of the epistemology of regress and of its own potential for regress into epistemology.

The linguistic sign is a meaningful entity that the processing system has endowed with the status of a termination point. In order to project forward from the sign to communicative act, the organism is not inclined to project backward through the levels of lesser complexity that preceded the origin of the sign. The supersign originated as regress is terminated at steadily greater levels of complexity. In view of the resource limitations expounded above, it is probably a general principle of language that regress is ordinarily possible only for selected signs at any one time, while the other signs within the communicative act are still treated as termination points. The Gestaltist principle that a figure must be perceived against a ground would be one articulation of this operational factor.

The information value attributable to a sign would be inversely proportional to the ease of referring its regress back from complex to simple. This conclusion may seem somewhat surprising, but the unusual supersign, while high in information, is also easy to focus and expound. For example, it takes less effort to define fairly uncommon words such as ‘ontogeny’ or ‘vicissitude’ than it does to define highly common words such as ‘of’ or ‘the’. Statistically frequent words are likely to be multi-functional, so that to open them to regress is to destabilize the system at multiple points. You'll probably see what I mean if you try delivering a lecture while forcing yourself to be aware of how you are using each definite article.

In actual communication, information is determined less by frequency than by unexpectedness or unpredictability in context (Beaugrande, 1980). Here, regress is terminated according to a principle we might call non-triviality: that the presented materials should not be totally obvious and well-known to all participants. However, thresholds of triviality can vary according to the people or situations involved. In a formal lecture, trivial material is far less acceptable than in an informal conversation among friends or family. A person of superior status condescends to inferiors by moving toward triviality, e.g. an adult addressing a child.[5]

From a processing perspective, a science is a domain people acquire as a system of supersigns. This factor is interesting when we consider that scientific inquiry is an enterprise of opening regress behind intuitive experiences or manifest facts and of treating what might seem trivial as if it were not so. A science is a system of high complexity and intends to reduce the complexity of its object domain by providing general explanations and classifications (Beaugrande 1997) However, the danger impends that. the science may rebuild the complexity of its objects as a mirror of its own complexity, so that the objects are now comprehensible, but only to the scientists themselves. Worse yet, the objects may be replaced by supersigns from the scientific framework. Moreover, since ‘normal science’ stipulates quite firmly how regress is done and at which points, theoretical blindness may be practiced regarding those termination points that are routinely left intact.

These dangers are very clearly illustrated in generative linguistics, The regress to formal logic emptied out the content of language as an object domain and substituted a set of specialized supersigns: ‘language’, ‘competence’, 'grammaticality’, ‘generate’ ‘'speaker/hearer’, and so on.

The theoretical definition of each supersign was at variance with its significances in terms of human factors (cf. discussion in Beaugrande, 1984a, 1998).[6] As a result, language was reconstructed as a mirror image of the theory, i.e., as an abstract formal axiomatics of linear strings. The theory thus became its own object, sanctioned by the traditional prestigious position of formal logic in classical science.

Human language psychology needs to remain aware of such dilemmas. Consider for example the major concepts I presented above: significance, knowing, and meaning. In commonsense usage, these terms are preferentially reserved for non-trivial instances, rather than for the most basic contents of consciousness. When people say something is ‘significant’ or ‘meaningful’, they intend to convey that it is not trivially so, i.e., that the significance or meaning is more elaborate or noteworthy than that we assign to everyday things. Similarly, ‘know’ is often used to designate certainty, not merely awareness.

Hence, my terms entail a regress back to the essential basis for these concepts. Such actions as knowing and meaning in the everyday sense now appear complex and thus in need of explanation, whereas they had formerly seemed simple and unproblematic. The question of how this new complexity might be modeled is still very open.

As I have already remarked, a major striving in earlier psychological research was to reduce contexts by a certain choice of experimental designs and procedures. To determine the status of the findings, we now need to consider how low-context tasks such as the recognition or recall of syllables and word lists, might be compared to high-context tasks such as the participation in discourse. Three postulates seem readily evident:

(1) A low-context task is less complex than a high-context one, because contextualizing increases complexity. Each contextual factor would add incrementally to the complexity of the task, so that a minimal context, e.g., working with simple nonsense syllables, would be the easiest. This postulate should be favored by classical language psychologists, because it suggests that their research was initiated on the most promising level.

(2) A low-context task is more complex than a high context one, because contextualizing decreases complexity by means of contextual constraints on the factors that need to be considered. Contexts tell people which of their already rehearsed and expertly skilled processing strategies should be used for the task. Such strategies would probably be less available or decidable for a novel, low-context task. This postulate should be unwelcome to classical psychologists, because it suggests that they made their own work far more complicated and difficult than was necessary. However, some current researchers in cognitive psychology seem to suggest such a postulate (e.g. Anderson, 1976; Kintsch, 1977).

(3) A low-context task can be either more or less complex than a high-context one, because contextualizing can either decrease or increase complexity, depending on the familiarity of the task in respect to one's prior knowledge and experience. Contextual constraints simplify a task if they correspond to the constraints of natural communication. But artificial, non-natural constraints, e.g. having to pair a real word with a nonsense word, make the task harder. This postulate, which I hold to be most plausible one, recommends an orientation of research toward experiments whose constraints are modeled closely on the familiar constraints of ordinary discourse (cf. Neisser, 1976; Spiro, 1977). Emphasis would be shifted away from skills or knowledge acquired during the experiment itself over to skills or knowledge acquired during people's prior history as social and psychological agents.

When processing entered the consideration of modern language psychology, the dominant approach was for a time close to the low-context frameworks of ‘stimulus/response’ and ‘minimal feature’ models. The ‘abstraction of traces’ from the input was believed to be the major operation of comprehension (cf. Gomulicki, 1956; E. Gibson, 1971). That is, the human would proceed by identifying and using the relevant features of the input without actively contributing or creating. The goal of research would be to find out how this selective and progressive abstraction occurs. Such an approach placed language research in proximity with investigations of sensory perception current at the time (e.g. J. Gibson, 1966). Context was construed as external to the subject.

Yet the same difficulty arose again that had impeded behaviorism and descriptive linguistics: the meaning of a word or text is not made available by external properties of sound or print. Thus, meaning cannot be abstracted from a language event in the same way as sensory data such as size or shape. Moreover, newer research indicated that sensory perception itself was highly constructive, matching the input against prior patterns of expectations (cf. survey in Rumelhart, 1977). These patterns were given the designation of schemas, a notion now returned to the center of discussion after long neglect.

Though ‘schema’ is certainly a complex concept, the nature of its complexity is far from established as yet. Researchers have not been able to agree about the quantity, specificity, origin and evolution of schemas people use (Beaugrande, 1984b). Schemas must contain rich expectations, but must also be adaptable enough to apply on many occasions. Also, it is not clear how schemas can best be explored in experimental research.

Presumably, language research differs from everyday language use in terms of the schemas people apply. Schemas of some order must be responsible for people's abilities to perform such classical experimental tasks as making ‘associations’ among items of word-lists, or using ‘tags’ or ‘cues’ during the recall of words. The least desirable hypothesis would be that people develop special, unique schemas for being test subjects in psychological experiments. However, this conclusion is quite implausible: people doing word tasks aren't going to forget that they use words in everyday communication as well. Still, participation in experiments could modify people's processing schemas to fit the occasion, especially when a large number of trials is performed.

Nonsense syllables were the simplest approximation of language widely used in experimentation. From a linguistic standpoint, nonsense syllables retain only the phonological and graphic aspects of language and, theoretically at least, repudiate the syntactic, semantic, and pragmatic aspects. The substantive aspect of the syllable is foregrounded at the expense of its normal communicative functions. The syllables constitute an artificial miniature language whose meaning and purpose are defined in terms of the experimental situation. Since that situation is novel and remote for the ordinary person, what the context in fact communicates is very unlikely to be the same for everyone.

Word lists at least admit semantics, since people are already habituated to associate words with meanings. Pragmatics is again recast to fit the goals of the task, usually, to recognize or recall which words were on the list. The conventional syntax of discourse is displaced by the artificial order of the list, especially when the order is randomized. This factor is important, since people can be shown to recall more words from sentences than from lists (Marks & Miller, 1964). Syntax evidently assists both the production and the reception of words (see also Tulving, Mandrel, & Bauman, 1964; Locker & Levine, 1975). Consistent performance with lists might be explained as a ceiling-effect: for the novel task, people normally. reach only a limited degree of proficiency within the time available. Discourse skills are likely to vary much more among the members of society, as we see for instance in public oratory.

All the same, the consistency of people's performance in the standard experimental designs using word lists made this line of research very popular for a time. The apparent generality of the findings made psychologists optimistic about the truth of their science. Yet these findings were not directly relevant to language as a communicative system. At least, no theory was brought forward to explain this relevance.

Compare the everyday situation where people have to report what was said on a previous occasion. This task vaguely resembles the recall of a word list, but with important differences. People are seldom expected to reproduce the exact discourse word for word; nor are they usually able to do so. Instead, the overall meaning will be stated, perhaps reusing some of the key words from the original. Hence, the various words assume different degrees of importance in this task, whereas their importance is uniform in a list-learning experiment where people are to recall as many words as possible. The hierarchical relations among words in discourse contrasts with the additive relations among words on a list. If we assume that schemas are memory organizers, then recall is being artificially limited by the experimental situation.

Now consider the design where people supply their own responses to a word in ‘free association’. Here also, subjects were often highly consistent. For instance, given the word ‘needle’, ‘thread’ and ‘pin’ were extremely common responses, whereas ‘pincushion’ and ‘diligence’ were extremely uncommon (Kent & Romanoff, 1910). Further studies showed that some ‘associations’ between word pairs are consistently stronger or weaker for most people (e.g. Jenkins, Mink, & Russell, 1958; Deese, 1961; Rothkopf & Coke, 1961). The stored schemas that stipulate what is or is not normally ‘associated’ ought to be active not just during word-list experiments, but also during ordinary communication, e.g., in a context involving the topic of ‘needlework’. The structure of conceptual associations determines which words will tend to have privileged functions in the recall of discourse. The documented ability of associates to act as distractors during experiments (cf. Anisfeld & Knapp, 1968; Underwood & Freund, 1968) indicates that the schemas become automatically activated, whether or not the context seems to require them. Schema theorists might derive the ‘strength-of-association’ phenomenon from the coherence of schemas. The relative frequency of associations across significantly large classes of contexts would be one factor determining the structure of a schema and thereby the strength of associations among its components. For instance, a great many people, including myself, have occasion to use a ‘needle’ with ‘thread’; but only skillful or habitual sewers are likely to own a ‘pincushion’ — I never had one, and I do not sew with ‘diligence’, but with clumsiness.

Within a schema, some components could be necessary, some could be merely typical and some could be purely accidental (Beaugrande, 1980). A necessary association would on the average elicit the most common responses, a typical one the next most common, and an accidental one the least common. For sewing, a needle is necessarily accompanied by thread; and for some tasks at least, pins must be used to hold the materials in place while they are sewed. But the act of sewing doesn't depend on owning a pincushion, or working with diligence.

Within this account, the list-learning findings are seen to regress back to the basic organization of significance, which is itself the forming of an association between two entities. The arbitrary pairing of nonsense syllables would resemble a very elementary act of significance and hence very low-level processing. When words are used, their prior significance to the test person acts as a termination point, and processing moves to a higher level. But due to the prior organization of knowledge and vocabulary, words are inconsistent psychological entities. Indeed, the more common a word is, the richer the associations it is likely to have in the experience of the individual person.

In normal life, words are of course learned in meaningful contexts. In discourse, known words function as channels of regress for learning new words. The original learning situation also provides regress by showing an occasion where the word was used. When the word is well learned, this regress can be terminated, and the word can function within the progressive action of being used in new situations. This whole process apparently requires little effort. A few encounters are enough to learn a new word, and sometimes just one. People must have a word-learning schema that renders this complex action easy.

In the experimental setting, the situation of encounter is very important if the person is expected to recall or recognize which words were on a list. Prior knowledge of these or other words is not very helpful and may actually distract the person by activating mental associations not needed for the task. The words will not be used for communicating, but for making a yes/no decision, itself often simplified into pressing a button. Thus, the everyday word-learning schema must be modified regarding those very aspects that make it efficient.

The fact that recognition can deal with far more words than recall can (Kintsch, 1977) suggests that people are learning just a few features of the word, enough to react when it appears again, but not enough to produce it without help. Discourse processing, on the other hand, attends to those features which contribute the most to coherent comprehension (cf. Just & Carpenter, 1981). The schema for the word-list experiment thus reflects an abnormal selectivity encouraged by the task. Such learning is fragmented and episodic, compared to the integrated and conceptualized learning from real discourse.

Now let us consider a counter-example where the task entails a higher context, as investigated in research on the story. In one paradigm, test persons hear or read a story and then try to retell it in their own words.[7] Unlike list-learning, this activity is quite common in everyday life situations in which people have to report what happened. Persons need no special training for the experiment. Indeed, giving them training before the experiment would be counter-productive, because the ordinary concept of the story is precisely the goal of inquiry.

The performance of a complex well-learned task makes this research fundamentally distinct from the classical experiment with the performance of a simple novel task. Still, story research has been widely acknowledged as a success because it has brought us some fairly general and robust findings about the effects of story structure upon comprehension and recall. In particular, if people are told a story that doesn't conform very well to conventional story structure, they tend to recall the story in a more conventional version (cf. Kintsch et al., 1977; Thorndyke, 1977; Mandler, 1978; Stein & Glenn, 1979; Johnson & Mandler, 1980). Hence, these experiments appear to activate out of everyday experience some prior knowledge structures (schemas) that can bring about systematic changes between input and output.

The ‘story-grammar’ paradigm was inspired by transformational grammar (see especially Johnson & Mandler, 1980). Hence, researchers were looking for the ‘rules’ which stipulate how stories are constructed. As in classical linguistics, the dominant issue is the arrangement of constituents. The order of sentences in a story is compared to the order of words in a sentence. Certain sequences are common, and others are not. But whereas linguists adopted the traditional word-classes (noun, verb, etc.) to label sentence constituents, psychologists disagreed noticeably about the nature and names for story constituents. Most often, the attempts of a character to perform an action and attain a goal were treated as the basic units.

However, we need to consider how far the stories being used in such experiments represent the stories people encounter in everyday life. In the studies most often cited, the stories were either invented by the experimenters (e.g. Mandler, 1978; Stein & Glenn, 1979) or adapted by greatly simplifying traditional tales (e.g. Rumelhart, 1975; Mandler & Johnson, 1977; Thorndyke, 1977). As I have recently argued (Beaugrande, 1982), these tactics allowed the researchers to project an unduly stark opposition between ‘well-formed’ stories versus ‘ill-formed’ ones. Yet the most famous ‘literary’ stories -- those that survive the test of time, like the Arabian Nights or the Decameron -- stimulate readers by both confirming and violating people's expectations (see also Beaugrande & Colby, 1979). The experimental stories lose some of this dynamics of natural storytelling in respect to the important question of how powerful and compelling prior schemas can be, and how departures and innovations occur in practice.

I would raise similar reservations toward the ‘story schema’ approach adopted by the Yale group. Here, the order of sentences in the story text is far less important as in the story-grammar approach. The focus is placed on expectations about content, i.e., the states and events to be mentioned. And the main procedure is not to experiment with humans, but to built story-understanding programs for computers. Here also, the stories were at first devised by the investigators (cf. Schank et al., 1975; Schank & Abelson, 1977). In later work, stories were taken from newspapers, but still as standardized types, such as reports about automobile accidents (Cullingford, 1978) or terrorist outbreaks (Schank, Lebowitz, & Birnbaum, 1978). Nobody has shown that these schemas would work for highly innovative texts, such as the later novels of James Joyce or William Faulkner.

What seems to be happening is what we might predict from the tradition of language psychology as I have described it. The high-context activity of story-telling has been reduced to a lower context in order to fit the classical psychological methodology. Investigators have hardly considered the role of story-telling as a common mode of human interaction. A skilled story-teller must adapt to the audience and motivate them to listen or read. The main question is not whether the presentation is story or a non-story, but whether or not it is a good, interesting story worth the effort. This kind of judgment is not made the same way by all humans on all occasions. Each occasion brings certain factors into prominence, factors that may be reshaped by the experimental setting.

The conclusions I would draw here might apply to the entire field of language psychology. The traditional trend of simplifying and lowering contexts should now be balanced by a trend of elaborating and raising contexts. Test materials should be made more realistic. The progress from nonsense syllables to word lists to single sentences to texts and stories should be extended to include real stories from the culture, such as folktales or literary novellas. Alternative versions of a story should be produced by skilled story-tellers rather than by psychologists. Qualitative factors which seem hard to capture in an experimental design should be measured by obtaining ratings from competent judges. For instance, such judges can be asked if a story is a good or interesting one, or which version of the story is more or less so. The differences among the versions can indicate which features influence the judges, for instance, hints and clues about possible outcomes; indications of character motives; triggers of emotional responses; relevance to human concerns; and so forth.

Experimental tasks should be devised that consolidate knowledge rather than isolating it. For example, test subject could be presented with a text and then asked to make up a title, a summary, continuation, and refutation, and so on. Such tasks refer not to single words or sentences, but to the overall import of the text. Or, the subject might read several texts and state the topics they do or do not have in common, or the human principle they all illustrate.

More attention should be given to individual cases. By concentrating on statistical generalities, experimenters are likely to overlook individual differences that may be very important for psychological development. For example, creativity is an action that modifies a prior system for a particular motive (Beaugrande, 1979). The specific outcome of this action is likely to be different for each person. Hence, many detailed individual studies will be needed before we can determine what is shared among creative actions in general.

Most important of all, an ethnography of psychological research should be elaborated. The roles of experimenters and test persons must be defined: how far the relationship is one of exploitation or co-operation; how test subjects interpret or infer what the experiment is intended to establish; what motivations or rewards influence behavior; what factors affect the evaluation of behavior as ‘correct’ or ‘intelligent’; and so forth. The findings should be compared to comparable situations in real life when people perform in schools, professions, and hobbies. We might then be better able to show the relevance of psychological research to everyday human situations. Comparisons like those I drew between studies of word-lists or artificial stories and realistic language use could be tested. It might become possible to determine how experimental conditions should be designed to encourage realistic and relevant results. Also, we could better decide what structures of regress and progress are appropriate for a particular object.

If, as I have contended, cognitive categories can be construed as performative ones, then cognition itself can be observed and tested, provided we have the proper methods for interpretation. Instances of ‘significance’, ‘knowing’, and ‘meaning’ (as I proposed to define these terms) are easy to find in all human activities. The question is how people invest these acts in respect to a certain intention and situation. Without a detailed, encompassing model of this investment, we will be more inclined to reduce rather than to elaborate the available data, or to accept external reality as a termination point beyond which we cannot regress.

Whereas my proposals would have seemed unreasonable and fantastic a few years ago, they have gained some cogency I recent years.. The concern voiced by Ulric Neisser (1976) for ‘ecological validity’ is now shared by increasing number of psychologists. Also, the conditions are more favourable for recognizing language as the major instrument for describing methods and interpreting findings, so that the psychology of language cab become a crucial domain for exploring scientific discourse within the wider context of social interaction. Explicit models could allow us to apply our research to our own enterprise, rather than keeping it from public view, as in classical psychology. We might thereby succeed in selecting productive channels of regress and progress that can reanimate theory and methodology for a wider spectrum of inquiry in the coming years.

NOTES

[1] An account of the epistemological origin of an ‘object’ will be offered in section 2.

[2] The preference for quantitative findings over qualitative ones is still strong in many research areas.

[3] My remarks will be directed mainly to American research.

[4] The expression ‘minimal’ clearly shows that these units are to be considered termination points.

[5] Wodak (1980) showed how courtroom judges patronize lower-class defendants by explaining simple things to them, such as how to steer an automobile.

[6] ‘Generate’ for instance normally means ‘produce'; in Chomsky's theory, it means ‘assign a structural description to’.

[7] A more traditional variant is trying to decide if the sentences on a subsequently presented list were in the story.

[8] In a recent paper, Robert Sternberg (1984) showed that a strong discrepancy is found between the kind of ‘intelligence’ measured by psychological tests and the kind recognized in social settings.[8] The same discrepancy might be found between good performance in laboratory experiments versus in everyday activities.

REFERENCES

ANDERSON, J.R. (1976), Language, memory and thought. Hillsdale, N.J.. Erlbaum.

ANISFELD, M., & KNAPP, M. (1968), ‘Association, synonymity and directionality in false recognition’. Journal of Experimental Psychology, 77, 171-179.

BEAUGRANDE, R. de (1979), ‘Toward a general theory of creativity’. Poetics, 8, 269-306.

---- (1980), Text, discourse, and process. Norwood, N.J.. Ablex.

---- (1980-81), ‘Design criteria for process models of reading’. Reading Research Quarterly, 16, 261-315.

---- (1982), ‘The story of grammars and the grammar of stories’. Journal of Pragmatics, 6, 383-422. .

---- (1984a), Text production. Norwood, N.j.: Ablex.

---- (1984c), 'Freudian psychoanalysis and information processing: Notes on a future synthesis’. Psychoanalysis and Contemporary Thought, 7/2, 147-194.

---- (1991) Linguistic Theory: The Discourse of Fundamental Works. London: Longman.

---- (1997), New Foundations for a Science of Text and Discourse. Stamford, CT: Ablex.

---- (1998), Performative speech acts in linguistic theory: The rationality of Noam Chomsky. Journal of Pragmatics 29, 1-39.

BEAUGRANDE, R. de, & COLBY, B.N. (1979), 'Narrative models of action and interaction’. Cognitive Science, 3, 43-66.

BEAUGRANDE, R. de, & DRESSLER, W. (1981), Introduction to text linguistics. London : Longmans.

CHOMSKY, N. (1957), Syntactic structures. The Hague: Mouton.

---- (1965), Aspects of the theory of syntax. Cambridge: MIT.

CULLINGFORD, R. (1978), Script application. New Haven: Yale diss.

DEESE, J. (1961), ‘From the isolated verbal unit to connected discourse’. In C.N. Cofer (ed.), Verbal learning and verbal behavior. New York. McGraw-Hill, 1961, 11-31.

DREWNOWSKI, A., & HEALY, A. (1977), ‘Detection errors on “the” and “and”’: Evidence for reading units larger then the word’. Memory & Cognition, 5, 636-647.

GIBSON, E. (1971), ‘Perceptual learning and the theory of word perception’. Cognitive Psychology, 2, 351-368.

GIBSON, JAMES J. (1966), The senses considered as perceptual system. Boston: Houghton Mifflin.

GOMULICKI, B. (1956), 'Recall as an abstractive process’. Acta Psychologica, 12, 77-94.

HAWKS, T. (1977), Semiotics and structuralism. Berkeley: Univ. of California Press.

JENKINS, J.J., MINK, W.D., & RUSSEL, W.A. (1958), ‘Associative clustering as a function of verbal association strength’. Psychological Reports 4, 127-136.

JOHNSON, N., & MANDLER, J. (1980), ‘A tale of two structures. Underlying and surface forms in stories’. Poetics, 9, 51-86.

JUST, M., & CARPENTER, P. (1980), ‘A theory of reading: from eye fixations to comprehension’. Psychological Review, 87, 329-354.

KENT, G.H., & ROSANOFF, A.J. (1910), ‘A study of association in insanity’. American Journal of Insanity, 67, 37-96.

KINTSCH, W. (1977), Memory and cognition. New York. Wiley.

KINTSCH, W., KOZMINSKY, E., STREBY, W., McKOON, G., & KEENAN, J. (1975), ‘Comprehension and recall of text as a function of content variables’. Journal of Verbal Learning and Verbal Behavior, 14, 257-274.

KUHN, T.S. (1970), The structure of scientific revolutions. Chicago: Univ. of Chicago Press.

LACKNER, J., & LEVINE, K. (1975), ‘Speech production: Evidence for syntactically and phonologically determined units’. Perception and Psychophysics, 17, 107-113.

MARCEL, A. (1980) ‘Phonological awareness and phonological representation’. In U. Frith (ed.), Cognitive processes in spelling. London: Academic, 1980, 373-403.

MANDLER, J. (1978), ‘A code in the node’. Discourse Processes, 1, 14-35. MANDLER, J., & JOHNSON, N. (1977), ‘Remembrance of things parsed: Story structure and recall’. Cognitive Psychology, 9, 111-151.

MARKS, L., & MILLER, G.A. (1964), ‘The role of semantic and syntactic constraints in the memorization of English sentences’. Journal of Verbal Leading and Verbal Behavior, 3, 1-5.

MILLER, G.A. (1956), ‘The magic number seven, plus or minus two’. Psychological Review, 63, 81-97.

NEISSER, U. (1976), Cognition and reality. San Francisco: Freeman.

---- (1982), ‘Cognitive psychology may at least establish a conception of human nature that is not self-contradictory’. Psychology Today, 16/5, 44-47.

NORMAN, D., & SHALLICE, T. (1980), Attention to action. Willed and automatic control of behavior. La Jolla: CHIP Report 99.

PAPERT, S. (1980), Mindstorms. New York: Basic Books.

PAVLOV, I. (1927), Conditioned reflexes. London.. Oxford.

ROTHKOPF, E., & COKE, E. (1961), ‘The prediction of free recall from word association measures’. Journal of Experimental Psychology, 62, 433-438.

RUMELHART, D. (1975), 'Notes on a schema for stories’. In Bobrow, D., & Collins, A. (eds.), Representation and understanding. New York: Academic, 1975, 211-236.

---- (1977), Introduction to human information processing. New York: Wiley.

SAUSSURE, F. de (1916), Cours de linguistique générale. Lausanne: Payot.

SCHANK, R., & ABELSON, R. (1977), Scripts, plans, goals and understanding. Hillsdale. N.J.: Erlbaum.

SCHANK, R., GOLDMAN, N., RIEGER, C., & RIESBECK, C. (1975), Conceptual information processing. Amsterdam, North Holland.

SCHANK, R., LEBOWITZ, M., & BIRNBAUM, L. (1978), Integrated partial parsing. New Haven. Yale CS-TR 143.

SCHMIDT, R. (1975), ‘A schema theory of discrete motor skill learning’. Psychological Review, 82, 225-260.

SCHNEIDER, W., & SHIFFRIN, R. (1977), ‘Controlled and automatic human information processing 1: Detection, search, and attention’. Psychological Review, 84, 1-66.

SHIFFRIN, R., & SCHNEIDER, W. (1977), ‘Controlled and automatic human information processing 2: Perceptual learning, automatic attending, and a general theory’. Psychological Review, 84, 127-190.

SKINNER, B.F. (1938), The behavior of organisms. New York. Appleton-Century-Crofts.

SPIRO, R. (1977), ‘Remembering information from text: The ‘state of scheme’ approach’. In Anderson, R., Spiro, R., & Montague, W. (eds.), Schooling and the acquisition of knowledge. Hillsdale, N.J.: Erlbaum, 1977, 137-165.

STEIN, N., & GLENN, C. (1979), ‘An analysis of story comprehension in elementary school children’. In R. Freedle (ed.), Advances in discourse processes. Norwood, N.J.: Ablex, 1979, 53-120.

STERNBERG, R. (1984), ‘The socialization of intelligence’. Paper at the 1984 Minnesota Symposium on Child Psychology, University of Minnesota.

THORNDYKE, P. (1977), ‘Cognitive structures in comprehension and memory of narrative discourse’. Cognitive Psychology, 9, 77-110.

TULVING, E., MANDLER, G., & BAUMAL, R. (1964), ‘Interaction of two sources of information in tachistoscopic word recognition’. Canadian Journal of Psychology, 18, 62-76.

UNDERWOOD, B., & FREUND, J. (1968), ‘Errors in recognition, learning, and recall’. Journal of Experimental Psychology, 77, 55-63.

WODAK, R. (1980), ‘Discourse analysis and courtroom interaction’. Discourse Processing, 3, 369-380.