Home Introduction Prescientific 19th Century Interference Cognitive Primary Learning Representation Encoding Storage Retrieval Implicit Memory Implicit Learning Neural Bases Modeling Development Emotion Personal Social Conclusion


The Cognitive Revolution in Memory

The date of the cognitive revolution in psychology is subject to some dispute.  

Other salient events were:

The cognitive revolution was consolidated with Neisser's Cognitive Psychology (1967), the field's first textbook, which provided a synoptic overview of the entire field, with an emphasis on perception, attention, and short-term memory, but including long-term memory and language as well. 

The cognitive revolution also spilled over into the study of animal learning:

But the roots of the cognitive revolution in the study of memory go back much further than 1967, or 1956, or even 1948.

The Onslaught of Behaviorism

In an important sense, psychology was "cognitive" at its very founding in the 19th century.  Psychology began as an attempt to understand mental life, especially cognition, and it initially focused on problems of sensation and perception precisely because it assumed that knowledge was acquired as a result of experience.

But then the behaviorist revolution hit: Psychologist lost interest in the mind, and focused its attention exclusively on observable behavior, ignoring private mental states like percepts and memories.  The agenda of radical behaviorism, as expressed by John B. Watson, its founder, and B.F. Skinner, its most famous proponent, was to relate behavior to stimulus conditions in the environment, without any reference to internal mental states or processes.

Consider, for example, how Watson construed memory in Behaviorism (1930, pp. 235-236):

"What the man on the street ordinarily means by an exhibition of memory is what occurs in some such situation as this: An old friend comes to see him, after many years' absence.  The moment he sees his friend, he says: "Upon my life!  Addison Sims of Seattle!  I haven't seen you since the World's Fair in Chicago.  Do you remember the gay parties we used to have in the old Windmere Hotel  Do you remember the Midway?  Do you remember _____", ad infinitum.  The psychology of this process is so simple that it seems almost an insult to your intelligence to discuss it, and yet a good many of the behaviorists' kindly critics have said that behaviorism cannot adequately explain memory.  Let us see if this is a fact.

"When the man on the street originally made the acquaintance of Mr. Sims, he saw him and was told his name at the same time.  Possibly he did not see him again until a week or two later.  He had to be re-introduced.  Again, when he saw Mr. Sims he heard his name.  Then, shortly afterwards, the two men became friends and saw one another every day and became really acquainted -- that is, formed verbal and manual habits towards one another and towards the same or similar situations.  In other words, the man on the street became completely organized to react in many habit ways to Mr. Addison Sims.  Finally, just the sight of the man, even after months of absence, would call out not only the old verbal habits, but many other types of bodily and visceral responses."

In other words:

"By 'memory', then, we mean nothing except the fact that when we meet a stimulus again after an absence, we do the old habitual thing (say the old words and show the old visceral -- emotional -- behavior) that we learned to do when we were in the presence of that stimulus in the first place."

This view of memory, such as it is, is quite compatible with that of associationism and interference theory.  Memories are associations, acquired through learning experiences.  The memory cue is essentially a stimulus that evokes an associated memory as a response.  It's just that simple.

This interpretation of memory as learning began with Ebbinghaus, and was consolidated (pardon the expression) with interference theory -- even though Ebbinghaus himself was not a behaviorist, and the behaviorist agenda was not announced until 1913.  But the later interference theorists -- McGeoch, Melton, Underwood, Postman (who worked here at UCB) and other leading figures in the "verbal learning" tradition were all influenced by behaviorism.

The behaviorist perspective also made a clear connection between human learning and memory and conditioning as studied in the animal laboratory.  Paired associates can be construed as stimulus-response pairs, where the subject learns to produce the target in response to the cue.  There are also explicit parallels to the vocabulary of conditioning, with response set analogous to discrimination, unlearning as analogous to extinction, and spontaneous recovery as the mechanism for interference.

The analogy between human verbal learning and animal conditioning evokes the arbitrariness assumption that was central to the stimulus-response theory of (animal) learning.  In animal conditioning, the stimuli and responses were selected arbitrarily, on the assumption that any stimulus could be connected with any response by virtue of the principles of association and reinforcement.  Similarly, verbal learning focused on meaningless materials: lists of nonsense syllables arbitrarily strung together and lists of paired associates making no inherent sense.  

The general idea was that the principles of memory were the principles of learning, and that human learning could be studied in the same manner as animal learning.


Behaviorism and its Discontents

In various forms, behaviorism dominated experimental psychology, especially in America, from 1920 to 1960 -- much as clinical psychology was dominated by psychoanalysis during the same period.  At the same time, there were some dissenters from the behaviorist point of view -- and these dissenters form the roots of the cognitive approach to memory.

F.C. Bartlett and Remembering

Chief among these was Frederick C. Bartlett (1886-1969), an extremely important figure in British psychology.  Bartlett had pursued his undergraduate and graduate studies in psychology at Cambridge University as a student of C.S. Myers, the founding director of the Cambridge Psychological Laboratory.  He was also a student of W.H. Rivers, a pioneering figure in social anthropology and social psychology who, during World War I, also did seminal work on traumatic war neurosis.  Bartlett eventually became the first professor of psychology at Cambridge.  Most British psychologists seek to tie themselves to Bartlett, in much the same way that American psychologists try to link themselves to William James -- he is that important.

Bartlett didn't honor disciplinary boundaries: trained in both experimental and social psychology, his major work, Remembering: A Study in Experimental and Social Psychology (1932), was a manifesto for a new kind of psychology that would look beyond the laboratory to real life.  As such, Bartlett was less obsessed than his peers with rigorous experimental control, and more concerned with practical applications.  

Bartlett was vigorously opposed to associationism and its evil twin, behaviorism.  He thought such "atomism" was the wrong approach for psychology, and that the important units for psychology were not associations but larger mental structures. 

The central concept in Bartlett's work was the schema (plural schemata or schemas, depending on whether you want to be correct), which he defined as an active organization of past experiences and behaviors.  

As an example of a schema, Bartlett referred to Head's analysis of the leg movements of a centipede, which must be organized in advances; if each leg movement were a response to the previous leg-movement, the centipede would never get anywhere.  

Another example would be Lashley's analysis of arpeggios in skilled piano-playing, in his essay on "The Problem of Serial Order in Behavior" (1951).  Again, each press of a finger on a key cannot be a response to the prior key-press, because the whole thing runs off too fast.  It all has to be organized in advance.

Bartlett was highly critical of Ebbinghaus' approach to memory.  Ebbinghaus has consciously emulated Fechner's psychophysical methods, using stimulus materials (the nonsense syllables) with no inherent meaning, and exercising strict control over the conditions of learning and remembering.  In Bartlett's view, Ebbinghaus made an error in enforcing "a perfectly automatic attitude of repetition in the learning", which does not represent remembering as it occurs in real life.

Bartlett understood what Ebbinghaus was trying to do, but just thought it wrong-headed.  As he wrote in Remembering:

"Once more [the first time was with Fechner] the remedy is at least as bad as the disease.  It means that the results of nonsense syllable experiments begin to be significant only when very special habits of reception and repetition have been set up.  They may, then, throw some light upon the mode of establishment and the control of such habits, but it is at least doubtful whether they can help us see how, in general, memory reactions are determined.

"I have dealt at this length with the nonsense syllable experiments, partly because they are generally regarded as occupying a supremely important place in the development of exact method in psychology, and partly because the bulk of this book is concerned with problems of remembering studied throughout by methods which do not appear to approach those of the Ebbinghaus school in rigidity of control.  But most of what has been said could be applied, with the necessary change of terminology and reference, to the bulk of experimental psychological work on perceiving, on imaging, on feeling, choosing, willing, judging, and thinking.  In it all is the tendency to overstress the determining character of the stimulus or of the situation, the effort to secure isolation of response by ensuring simplicity of external control."

Bartlett summarized his critique with the clarion call of the cognitive psychologist:

"The psychologist, of all people, must not stand in awe of the stimulus".

But if not the stimulus, then what?  The mind -- the mental structures and processes that mediate between stimulus and response.  For Bartlett, a psychology of memory must deal with memories and remembering -- how experiences are represented in the mind, and how memories are encoded and retrieved.  

Bartlett thought that a valid psychology could only be achieved by focusing on lifelike experiences and lifelike situations.  Therefore, in his experiments he told his subjects strange stories and asked them to repeat them, and he also examined remembering in different cultures.  In this way, he tried to discover the relevant mental processes.  

Bartlett construed perceiving and remembering as active effort after meaning: the perceiver, or rememberer, is trying to make sense of his experience and memory.  As such, Bartlett focused on the role of expectations, goals, and attitudes in remembering.

Such considerations were ruled out of bounds by the associationistic tradition in the study of memory, which discovered the role of language only very late in its run -- when Melton and his colleagues discovered that the forgetting of paired associates learned in the laboratory was caused by the spontaneous recovery of prior language habits, which had to be unlearned in order to learn the nonsense syllables.

Recall that Bartlett was writing all of this in 1932, the same year as McGeoch articulated his notion of response competition.  If Bartlett's call had been heeded, the psychology of memory wouldn't have been led down the garden path for almost 50 years.


The Gestalt Tradition

Dissent also came from other quarters in Europe, particularly on the Continent, in the form of Gestalt psychology.  Recall that the German word Gestalt refers to "the whole form", as expressed in the famous Gestalt aphorism that 

"the whole is greater than the sum of its parts".

Or, put more precisely, 

"The whole cannot be deduced from the characteristics of the separate pieces, but conversely; what happens to a part of the whole is... determined by the laws of the inner structure of its whole."

Gestalt psychology was based on the Law of Pragnanz, that psychological organization always tends toward good form -- or, at the very least, the best organizational form that the stimulus field will allow.  But this form is not created by the stimulus alone; it is also created by processes in the mind of the perceiver.  The Gestaltists' emphasis on the organizational activity of the mind, like Bartlett's emphasis on the importance of mental schemata, makes them, like him, a forerunner of the cognitive approach to memory.

Like Bartlett, though for very different reasons, the Gestaltists were opposed to both associationism and behaviorism.  They did not believe it was possible to analyze mental life into its elementary, or atomic, units, whether stimulus-response associations or anything else.  Rather, they argued that the mind imposes organization on stimulus material.

            (50874 bytes)The Gestalt viewpoint was applied principally to the study of perception, as in the formulation of the Gestalt laws of perception:

Each of these laws was a special instance of the Law of Pragnanz.

But the Gestaltists also had theories of other aspects of mental life:

In the Gestalt view of memory, perception is the input to memory -- and the percept is already an organized transformation of the stimulus.  The memory trace is the residue of perception, and activation of the trace revives the original perceptual process.  Individual traces are connected in memory, so that memory proceeds by activating one trace from another.  But this is not an associationistic process, because the stimuli and responses do not retain their individual identity.  Rather, the association fuses the elements, and becomes an entirely new thing -- again, the whole is greater than the sum of its parts.  Meaningful material is easier to learn and remember, precisely because it is more easily organized.  

The Gestalt approach to memory is exemplified by an experiment by Asch (1969) -- the same Asch who pioneered the study of conformity, and of impression-formation, in social psychology.  Asch had his subjects learn paired associates such as DAT-NIC.  When subjects pronounced the pairs as a fused unit, DATNIC, rather than as two separate syllables, DAT-NIC, the list was learned more easily and better retained.

The Gestalt theory of forgetting was that poorly organized memory traces eventually cohered into a better form, but the resulting memory was not an accurate representation of the original event.  Thus, coherent, well-organized events were better remembered.

And the Gestalt theory of retrieval was that cues to memory retrieved traces containing related material.  But the relationships were not associative in nature, but rather based on Gestalt laws of perception.

Wulf.JPG (46910 bytes)In a study by Wulf (1922), subjects studied simple line drawings, and then their memory was tested after 30 seconds or 24 hours.  The delayed reproductions showed evidence of systematic distortion: sharpening, which exaggerated the characteristics of the original; or leveling, which minimized the characteristics of the original.  Inspection suggested three mechanisms for leveling and sharpening:  normalizing, which made the figure look more like a familiar object; pointing, in which the reproduction emphasized features originally noted during perception; and autonomous change following the Gestalt laws of perception (e.g., introducing proximity, symmetry, or closure where there was none in the original).

Perkins.jpg (66647
            bytes)Autonomous changes were also observed by Perkins (1932), who presented line drawings lacking in balance or symmetry.  The subjects reproduced the drawings immediately, and also after delays of 2, 3, 9, 16, 30, and 49 days.  Over time, the reproductions gradually conformed to Gestalt principles.



Carmichael.jpg (58823
            bytes)Also along Gestalt lines, Carmichael, Hogan, and Walter (1932) examined the effect of suggestion on visual memory.  They presented ambiguous line drawings accompanied by a verbal label, and found that the subjects' reproductions conformed to the verbal label -- a phenomenon since termed verbal overshadowing by Loftus, Schooler, and others.  In the Gestalt view, however, the verbal label was incorporated into the perception of the stimulus, and thus into the resulting memory.


Aside from the phenomenon of verbal overshadowing, the Gestalt approach to memory was not of lasting importance.  But the demonstration of Gestalt effects was very important.  For present purposes, the Gestalt approach to memory is important as an alternative to associationistic behaviorism, and a precursor of modern cognitive psychology, because of its emphasis on mental processes that reorganize stimulus input.

The Organization of Memory

Bartlett's work was largely ignored in large because his experiments were poorly controlled.  The Gestalt psychologists were largely ignored because - -well, because they were European.  In many ways, the most important challenge to the associationistic view of memory came from within the verbal-learning tradition itself, with observations of the organization of free recall.  The usual procedure in verbal-learning experiments was to present a list of items for study, and then to test the subject's memory for the list.  According to theory, the underlying associations (i.e., between stimulus and response terms in paired-associate learning) were formed on the basis of contiguity. Thus, items should be recalled in the same order as they were presented -- with each recalled item serving as a stimulus for recall of the next item in the list as a response.  But this turns out not to be the case.

JenkinsRussell.JPG (41532 bytes)Jenkins and Russell (1952) were the first to observe associative clustering in free recall.   They constructed lists containing associatively related words such as black and white, high and low, and table and chair -- but these items were presented for study in random order.  They observed that their subjects' recall did not preserve the original order of presentation.  Rather, the list items were reorganized, and the words clustered together based on pre-existing associative relationships.  Clearly, memory was not preserving stimulus input.  Rather, it was reorganizing the input, so that the memory no longer resembled the original experience.

Actually, it turned out that this result was no challenge to associative theory, because the reorganization could represent proactive interference from prior learning.  If the associates had previously been encountered together, in phrases such as set it down in black and white, or high and low temperature, or pull the chair up to the table, then the subject comes into the laboratory with some associations already formed.  These prior linguistic habits must be unlearned.  The required unlearning is incomplete after one trial, and the spontaneous recovery of the previous habit, over the retention interval, gives the output the appearance of reorganization.

Bousfield.JPG (52410
            bytes)This interpretation of clustering was directly challenged by Bousfield (1953) who constructed his lists from category exemplars, such as animal (muskrat, panther, wildcat), occupations (blacksmith, baker, printer, chemist), names (Howard, Jason, Otto), and vegetables (radish, mushroom).  Again, he presented the items in random order, and again, he observed that recall did not preserve the input order.  conceptually related items were clustered together.  This is called category clustering.


This result is a direct challenge to association theory, because the category instances are not associatively related to each other.  Never, not in a million years, will the stimulus blacksmith elicit the response baker; nor will Howard elicit Jason.  Rather, recall has been reorganized according to a higher-order organizational principle, based on conceptual relationships.

Perhaps the icing on the cake was provided by Tulving's (1962) work on subjective organization.  His lists contained items that had no associative or conceptual relationships to each other: each item was, essentially, unrelated to every other item.  In a procedure known as multi-trial free recall, which involves repeated cycles of study and test, study and test (with each study list presented in a different random order), Tulving observed that the order of recall became more regular.  The order of recall was idiosyncratic -- not the same for every subject; but each subject, in his own way, developed a more regular order of recall -- usually reflecting a more-or-less coherent narrative, or the description of an image.

Again, the subject is not merely picking up on the structure of the stimulus list -- precisely because the stimulus list has no structure.  Rather, the subject is imposing structure on the list -- an active process not unlike Bartlett's "effort after meaning".

This contribution of the subject is what cognitive psychology is all about.  The subject is, in Jerome Bruner's wonderful phrase, going "beyond the information given" by the stimulus, actively trying to make sense of what is going on.  Memory is a reflection of this active process.

The Terminal Meta-Postulate

Another nail in the coffin of S-R behaviorism was provided by Bever, Fodor, and Garrett (1968), who were heavily influenced by Chomsky's earlier (1959) critique of Skinnerian behaviorism. They noted that, if taken seriously, S-R theory made the following assumptions, which collectively they dubbed the terminal meta-postulate (TMP; the following account is derived closely from "Contacts of Cognitive Psychology with Social Learning Theory" by G.H. Bower, an unpublished invited address given to the annual meeting of the American Association for the Advancement of Behavior Therapy, Atlanta, December 1977):

  1. The only elements required for a complete psychological explanation of behavior are isomorphic with potentially observable events, such as stimuli, responses, and their derivatives (such as mediating responses or response-produced stimuli).
  2. These elements become associated with each other if they occur close together in space and time (i.e., association by contiguity).
  3. All observable behavior can be explained by concatenating these associative links 

BF&G argued that the TMP can be refuted by any example of human performance which cannot be reproduced by machine restricted by the TMP.  Their counter-example was a mirror-image language, in which strings of letters, digits, or any other symbol are "grammatical" if the last half of the string is the mirror image of the first half -- for example, 

  1. CC
  2. DCCD

These strings can be generated by the following rewrite rules:

  1. S ==> cSc
  2. S ==> dSd
  3. S ==> nothing.

People can learn to discriminate grammatical from nongrammatical strings, so this is an aspect of human performance that ought to be reproduced by an S-R "machine" consisting of the four possible associations between the elements:

The problem is that while the machine can produce grammatical strings, it also produces ungrammatical strings like CCDDC.  In fact, there is no way to produce all and only grammatical strings using only the TMP.  The strings in a mirror-image language are "center-embedded", producing dependencies at arbitrary distances, such that if there is a C at the beginning of the string, there must be a C at the end of the string, no matter how long the string is.  What is required is a memory -- a structure for representing the elements in the string, and storing them until they are needed again later.  For example, a simple push-down stack memory, which stores tokens, or symbolic representations, of the two types of elements, C and D.  This memory, which operates on a principle of "last in, first out", will produce all grammatical strings, and only grammatical strings.

  1. To produce String 1, generate a C, put a token of the C in the memory and take it out again.
  2. To produce String 2, generate a D, put a token of the D in the memory; then generate a C and put a token of the C on top of the token of the D, then take the C token out out, and finally take the D token out.
  3. To produce String 3, put in D, then C, then D, then D, then C tokens as they are generated, then take out the C, then the D, then the D, then the C, and finally the D tokens.

The point is that in order to generate grammatical strings, you need a push-down stack to store tokens of the string elements, and neither the memory nor the tokens are stimuli or responses in the usual sense of these words  Rather, the tokens are representations of stimuli and responses. In order to account for people's ability to use a mirror-image language, we must postulate both a memory storage structure and a set of representations that are stored in it -- both of which are expressly prohibited by the TMP.  

The problem isn't limited to mirror-image languages.

Consider a familiar memory test, similar to that found in the Wechsler Adult Intelligence Scale (WAIS),  in which a subject has to repeat a string of digits forward, then backward.  Given the string 9327, the subject would have to resort to a push-down stack to store tokens of the string elements until they were needed in order to respond with 93277239.

Or consider a subject who learns two sequences of associations:

  1. BLUE ==> RED ==> GREEN
  2. GOLD ==> RED ==> BROWN.

The same element, RED, cannot be used in both sequences, because that would generate incorrect sequences such as BLUE ==> RED ==> BROWN and GOLD ==> RED ==> GREEN.  Instead, there have to be two tokens of RED, one associated with BLUE and GREEN, the other associated with GOLD and BROWN.  And these tokens, or copies, are neither stimuli nor responses.

So there's something wrong with the TMP.  And if there's something wrong with the TMP, then there's something wrong with the whole account of behavior offered by S-R behaviorism.

What's wrong with the TMP?  What's wrong is that, in order to manage things like learning and using a mirror language, which people can do easily, you need some kind of storage structure, and you need copies of stimuli and responses to put in that storage structure.  In other words, you need a memory, and you need memories.  You need to know something about the internal cognitive apparatus that mediates between stimuli and responses.

The Computer Model of the Mind

In the final analysis, the behavioristic view of human learning and memory was not brought down by any particular result or argument.  Rather, it was lost in a massive change in theoretical organization brought about by the cognitive revolution in psychology as a whole -- and especially by the computer metaphor of mind.

In the older, pre-cognitive metaphor, mind was likened to a telephone switchboard, with telephones representing items and wires representing the associations between them.  Once it was invented, the modern high-speed computer replaced the telephone switchboard.

The computer metaphor gave us the notion of cognition as information-processing.  In this metaphor, the mind extracts information from a stimulus, combines it with information stored in memory, transforms this information according to stored rules, and generates outputs in accordance with its programmed task.

In the computer metaphor of mind, brain is the hardware, and mind is the software.

AtkinsonShiffrin.JPG (41504 bytes)The computer metaphor was instantiated in the multistore model of memory, of which there were competing, but highly similar, proposals by Waugh and Norman (1965) and Atkinson and Shiffrin (1968).  This quickly became so popular that it was known as the modal model of memory (Murdock, 1967).  The figure at the left is the Atkinson-Shiffrin version.  For our purposes, the commentary that follows applies to either version, or any other version of the modal model.


(Atkinson began his career at Stanford, and moved to UCSD, where he rose to be Chancellor, served for a time as President Carter's science advisor, and finished his career as President of the entire UC system.  Norman also started out at Stanford, also moved to UCSD, and ended up back in Silicon Valley as a techno-guru at Apple.  Waugh became professor of psychology at Oxford, Shiffrin at Indiana.)

In the modal model of memory, knowledge is represented as information.  There is a set of storage structures that hold information in various forms.  And there is a set of control processes that pass information from one storage structure to another, transforming information along the way.


This page last modified 09/22/2008.