Encoding makes a memory trace available in memory, and elaborative and organizational processing at the time of encoding makes for a rich, distinctive memory trace that is easy to retrieve -- even after a long retention interval.
One implication of the three-stage analysis of memory, and of the library metaphor, is nothing. Storage is a passive stage, in which the encoded memory lies waiting to be retrieved -- just like a book, once cataloged, sits on the shelf until someone takes it down.
But it turns out that this view isn't quite right, and that memories receive further processing after encoding, and dynamically interact with each other even though they haven't been retrieved. Or, at least, they seem to do so.
Recall that there are three modes of processing
that occur at the time of encoding:
Memory is a byproduct of perception, and elaborative and organizational activity are critical to making a rich, distinctive memory trace available in memory over long retention intervals.
Turning to the retention interval itself, we know from Ebbinghaus that retention is a function of time. In perhaps his most famous study, Ebbinghaus learned lists of nonsense syllables to criterion, and then relearned them after retention intervals ranging from 19 minutes to 1 month. He found a progressive decline in savings that fit a logarithmic function reminiscent of Fechner's Law.
Results like this gave us the Time-Dependency Principle:
The Time-Dependency Principle
Memory diminishes with the length
of the retention interval between the time the trace is
encoded and the time that retrieval is attempted.
To be honest, though,
time-dependency is not a particularly satisfying principle, for
the simple reason that time is not a psychological
construct, and we'd really like our principles of memory to be
in stated in terms of basic mental processes. So, let's
look at some of the factors that mediate time-dependent
Perhaps the oldest hypothesis concerning
time-dependency is decay, which refers to an autonomous
process of degradation over time.
The decay hypothesis has generally found favor with respect to "short-term" memories, such as the sensory registers and primary memory (though even in the latter case, there has been vigorous debate over whether decay actually occurs, in the absence of displacement).
With respect to long-term secondary memory, however, it should be noted that memory does not always fade with time. Early in the scientific study of memory, Ballard (1913) gave repeated tests of children's memory for poetry after only a single study session, and found that they often remembered more material on later tests than they had on earlier ones. He contrasted this reminiscence effect with the more usual oblivescence, or forgetting.
Erdelyi (1974) revived Ballard's work, showing that reminiscence was more likely to occur for pictures than for words. He renamed Ballard's phenomenon hypermnesia, in contrast to the amnesia of ordinary forgetting. In Erdelyi's usage, reminiscence refers to the recovery of previously forgotten memories on later tests -- which may be matched or exceeded by inter-trial forgetting. In the typical Ebbinghausian case, inter-trial forgetting exceeds inter-trial recovery, resulting in a net loss of memory over time. In Erdelyi's usage, hypernnesia refers to the situation where inter-trialintertrial recovery exceeds inter-trialintertrial for(Erdelyi's asserte contrary notwithstanding, it is possible to observe hypermnesia with words as well as pictures, so long as the words are deeply processed at the time of encoding).
The classical disproof of the role of decay in secondary memory was offered by Jenkins and Dallenbach (1924), in an experiment that compared the effects of sleeping versus waking on memory, holding the retention interval constant. They observed much more forgetting when subjects stayed awake, compared to when they slept during the retention interval. If decay were an autonomous time-dependent process, then memory should have faded over the sleep interval too, which it did not. J&D observed the classic Ebbinghausian forgetting curve only in the waking trials. This observation led them to conclude that (p. 612):
"[F]orgetting is not so much a matter of the decay of old impressions and associations as it is a matter of the interference, inhibition, or obliteration of the old by the new".
After the Jenkins and Dallenbach study, the decay hypothesis was pretty much a dead letter. But even in the absence of empirical evidence, the decay hypothesis was unsatisfactory, for the simple reason that time is not a psychological concept. It doesn't refer to a mental structure or a mental process -- or even to behavior. McGeoch (1942; pronounced Mugyoo) offered an analogy to rust: metal rusts when exposed to the elements for a prolonged period of time, but rust is a product of oxidation, not time per se. So, the time-dependency of memory begs the question of What happens over time?
In fact, long before J&D, Muller and Pilzecker
(1900) offered an alternative: consolidation. They
argued that encoding is not instantaneous, but rather unfolded
over time as a result of two processes.
In M&P's view, interruption of the consolidation process destroys or degrades the memory trace. Put another way: what transpires over time is consolidation, and extended retention intervals provide more opportunity to interrupt the consolidation process.
Unlike Ebbinghaus' treatise, M&P's monograph, reporting some 40 experiments, has never been translated into English (hint, hint). However, very useful accounts of their research are available in secondary sources (McGaugh, 1999; Lechner et al., 2008).
M&P's research program differed from Ebbinghaus' in a number of respects:
Experiments 1-6 were essentially variants on Ebbinghaus, having to do with the effects of repetition and the retention interval. Experiments 38-40 covered relatively minor matters having to do with the effects of metric style and of association.
Experiments 7-27 were concerned with inter-associative interference. In these experiments, they taught their subjects (who were, mostly, their wives and themselves) a list of paired associates of the form A-B, and then another list of the form A-D, retaining the original cue but changing the response. They observed that incorrect B responses tended to intrude when subjects were attempting to recall the C responses; this tendency was greatest after a short interval, and disappeared after about 10 minutes. They interpreted this effect as reflecting the perseveration of the A-B list, even after it was no longer relevant.
Experiments 28-37 were concerned with retroactive inhibition (RI; M&P were apparently the inventors of this paradigm). Subjects first learned one set of paired associates, A-B, and then an entirely new set of pairs, C-D; then they were tested for their memory for the A-B pairs. Recall after the interpolated learning was only about 23% correct, compared to 48% in a control condition in which the subjects merely rested. They interpreted RI as reflecting the disruption of the consolidation process, degrading memory for the A-B pairs.
On the basis of experiments like these, M&P announced the Doctrine of Consolidation (pp. 196-197, quoted in Lechner et al., 2008):
"[A]fter reading a list of syllables certain physiological processes, which serve to strengthen the associations induced during reading of that list, continue with decreasing intensity for a period of time. These processes and their facilitating effects on these associations are being weakened to a greater or lesser extent if the... subject experiences further mental exertion immediately after reading a list...."
They also attempted to estimate the time
course of consolidation:
Therefore, they estimated that consolidation takes ~ 10 minutes.
Other researchers quickly noted a connection between M&P's arguments about consolidation and the pattern of the dissolution of memory noted by Ribot in his clinical investigation of the Diseases of Memory (1882) -- particularly the observation that new memories were lost before old ones. Moreover, in cases of the recovery of memory, it appeared that old memories were recovered before new ones. Ribot considered his Law of Regression to be a specific instance of a general principle articulated by J. Hughlings Jackson (1835-1911), a pioneering neurophysiologist who attempted an evolutionary analysis of cortical function. In Jackson's view, Dissolution reverses evolution: the last brain functions to evolve were the first ones lost to disease. But the general principle of "last in, first out" was also compatible with the new concept of consolidation. If a disease process impaired consolidation, very recent memories would be the ones most likely to be affected.
Of particular relevance here is the temporal gradient in traumatic retrograde amnesia induced by what Ribot called "cerebral shock" -- i.e., a concussive blow to the head.
In normal individuals, autobiographical memory shows a temporal gradient, such that people are more likely to remember more recent events than remote ones. This is true if you simply ask people to write down everything they remember from their lives, as Waldfogel (1949) did with a sample of college students (then he asked them to do it all over again, to assess the reliability of recall). Or you can systematically sample autobiographical memory as Crovitz (1970) and Robinson (1976) did, following the example of Sir Francis Galton, by presenting subjects with familiar words like car and horse and asking them to recall a personal experience related to the cue. Either way, you'll find that the overwhelming proportion of memories are of very recent vintage.
But when subjects are tested immediately following a concussive blow to the head (i.e., after which they lose consciousness), they frequently show a retrograde amnesia (RA) covering some period of time before the concussion. This RA is also temporally graded, such that the most recent memories are more likely to be lost than more remote ones. The interpretation was that the concussion interrupted the consolidation process, resulting in a loss of memory.
A clever study by Yarnell and Lynch (1973) confirmed this temporal gradient in American football players who had been "dinged" during play. When asked immediately after the incident what play they had been running, the players were usually accurate; but on a delayed test, they were often wrong. Y&D concluded that the memory had been lost due to a disruption of consolidation.
A similar temporal gradient is observed in the amnesia induced by electroconvulsive therapy (ECT), in which depressed patients receive a jolt of electric current through electrodes applied to the skull, which results in a convulsive seizure similar to that observed in epilepsy. When patients recover consciousness, they typically show a retrograde amnesia covering the period of time leading up to the treatment (ECT actually works, and there is less amnesia if it is administered unilaterally, as opposed to the traditional bilateral treatment).
Squire and his colleagues documented RA following ECT in a series of experiments employing two innovative tests of memory:
Squire & Chase (1975) employed a between-subjects design, comparing patients who received ECT (tested 6 hours after receiving their last treatment) and control patients who did not receive any ECT at all. The ECT patients showed a loss of memory covering the most recent two years.
Squire, Slater, & Chase (1975) employed a within-subject design in which a single group of patients was tested (with alternate forms of the tests) prior to their first ECT, and after their 5th dose. Again, ECT induced an RA extending back for about 2 years.
There is even an animal model of traumatic amnesia
known as electroconvulsive shock amnesia (ECS). A
typical experiment employs a procedure known as one-trial
step-down passive-avoidance learning:
Chorover & Schiller (1965) discovered that when ECS was applied to the rat within 10 seconds of receiving the foot-shock, the animal was rendered amnesic for the aversive experience - -as evidenced by the fact that, when put on the shelf again, it stepped right down as if nothing had happened. Again, the interpretation is that the ECS disrupts the consolidation process, resulting in the loss of recently acquired memory.
Results such as these led some investigators, such
as Squire (1992) and McGaugh (2000) to propose what has come to
be called the standard model of consolidation:
This standard model is very popular, and has led a number of neuroscientists to search for the physiological basis bases of consolidation in biochemical and structural changes at the synapse and neuronal locations.
At the same time, there are a number of observations that are troublesome for the standard model.
One problem is retrograde amnesia in
the amnesic syndrome resulting from damage to the
hippocampus and other structures in the medial-temporal lobe
Actually, in some cases, what appears as an RA may
be an AA instead.
Still, the fact that patients with the amnesic syndrome show RA as well as AA is problematic. If hippocampal damage causes amnesia due to a deficit in encoding and consolidation, then patients who suffered their hippocampal damage at a discrete time and place shouldn't show RA.
Further, there is the not-trivial problem of recovery from traumatic RA. As noted earlier, immediately after recovering consciousness concussed patients will show a temporally graded RA covering the period of time leading up to the loss of consciousness. However, continued testing typically reveals that some of these memories, initially lost, are subsequently recovered.
For example, in the study of ECT-induced amnesia by Squire & Chase (1975), the amnesia initially observed was entirely abolished on subsequent testing.
The problem here is that consolidation failure is supposed to result in a permanent loss of memory from storage. And ECT-induced RA evidently isn't permanent.
Actually, despite some recovery, there is always a final or residual retrograde amnesia covering the moments to minutes immediately prior to the loss of consciousness. This final RA is apparently permanent, and the memories covered by it are not subject to subsequent recovery.
Observations like these led proponents of the
standard model to distinguish between two kinds of
Thus, in the Yarnell & Lynch study of football "dings", the spared immediate memory was interpreted as indicating that short-term consolidation was unaffected by the concussion, but that long-term consolidation was disrupted.
Similarly, ECS-induced amnesia in rats, as in the Chorover & Schiller (1966) study, was taken as a laboratory model of the "Final RA" in traumatic retrograde amnesia.
The standard model of consolidation, with its distinction between short-term and long-term consolidation, has been enormously influential.
To take just one example, it has guided neuroscientific work, mostly at the cellular and molecular levels of analysis, on the biological basis for consolidation (e.g., Squire & Kandel, 1999, Chapter 7) -- new protein synthesis, and its underlying genetic mechanisms, cAMP-dependent protein kinase A, the formation of new synapses, late-phase long-term potentiation, and all the rest of that wet stuff.
This line of research, which won Eric Kandel the Nobel Prize for Physiology or Medicine, illustrates a basic point about the relationship between psychology and neuroscience: once psychology has identified and clarified a concept at the psychological level of analysis, then -- and only then -- does it make sense for neuroscientists to try to identify its biological mechanisms and substrates.
But research on the biological substrates of consolidation and memory doesn't have to be limited to the cellular and molecular levels of analysis. There is plenty of work that can be done at the psychological level as well. For example, much recent research on the effects of sleep on memory has been guided by the standard model of consolidation.
For example, Ekstrand and his colleagues revived J&D's work in a series of studies published in the late 1960s and early 1970s, which confirmed that sleep generally protects against forgetting -- what Ekstrand called the sleep effect -- but added some interesting details.
Forgetting is further reduced if the subject goes to sleep immediately after the study trial, compared to delayed sleep.
There is less forgetting if the retention interval covers the first half of a night's sleep, which is dominated by "slow-wave" NREM sleep, compared to the second half, which is dominated by REM sleep.
A short period of sleep prior to study actually increases forgetting.
Ekstrand's work was entirely at the behavioral level of analysis, but more recently other researchers have used our emerging understanding of the biological substrates of sleep as a vehicle for understanding the biological substrates of consolidation and memory. One of the leaders of this endeavor is UCB's Prof. Matt Walker.
Here is how Walker and Stickkgold (2006)
summarized what is currently known about sleep and memory.
All of these results suggest that sleep plays an important role in consolidation, and that the biochemical substrates of sleep are relevant to the biochemical mechanisms of consolidation.
Consolidation occurs after encoding, during the storage phase, and consolidation comes in two forms, short-term and long-term. It's a pretty good story. Unfortunately, there are two pieces of evidence that argue against it.
The first problem concerns retrograde amnesia in
the amnesic syndrome. The distinction between short- and
long-term consolidation is fine as far as it goes, but the
explanation only works if the RA has two related properties:
The problem is that RA in the amnesic syndrome can extend far, far back in the patient's pre-morbid life, 10 years, 50 years, in some cases covering the patient's entire life. This can't be explained by a disruption of long-term consolidation, because consolidation has to stop sometime.
A comprehensive review of the literature by Fujii,
Moscovich, and Nadel (2000) classified all published amnesic
patients (for whom the information was available) in terms of
the location of their brain damage (confirmed via autopsy or
brain-imaging) and the extent of their RA (all amnesic patients
suffer from a dense AA by definition).
So, apparently, RA occurs in the amnesic syndrome to the extent that there is (bilateral) damage to other areas of the medial temporal-lobe memory system.
Moreover, Fujii et al. showed that the RA
dissociated various aspects of declarative memory (which is just
another reason to be unhappy with Squire's use of the term).
In order to account for the pattern of RA observed in the amnesic syndrome, Nadel and Moscovitch (1997) have proposed a multiple-trace theory as an alternative to the concept of long-term consolidation. They have no objection to the concept of short-term consolidation. Their target is long-term consolidation. They don't think that the role of the hippocampus is to foster long-term consolidation. Rather, they think that the hippocampus plays a special role in binding the elements of an event together, along with information about the episodic context, into a cohesive episodic memory.
As noted, Nadel and Moscovitch have no objection to the idea of short-term consolidation mediated by the hippocampus -- the kind of process that persists after encoding, into the storage stage, for a matter of minutes. It's the disruption of short-term consolidation that accounts for the permanent "final RA" invariably observed in cases of traumatic RA, including RA induced by ECT in depressed patients and ECS in laboratory animals.
But wait! Even the final RA in ECS-induced amnesia may not be permanent. In a series of studies by Miller and his colleagues (e.g., Miller & Springer, 1975), the rats went through the one-trial step-down passive-avoidance learning procedure. Rats who received only the foot-shockfootshock took forever to step down, while those who received ECS within 0.5 seconds of the foot-shockfootshock stepped down as if nothing had ever happened to them. However, a third group of ratso received ECS, received a stressful "reminder" treatment prior to the passive-avoidance In one experiment, they received a brief tailshock delivered in another chamber. In another experiment, they were forced to swim in circulating icewater. These animals also failed the passive avoidance test, but they took significantly longer to jump down than those who had received the ECS. It was as if the tailshock or icewater treatment reminded them of the foot-shockfootshock, resulting in a partial return of the memory of the foot-shock, and the partial return of the passive avoidance response as well.
The point of the experiment, of course, is that reminders don't work if there's no memory to be reminded of. Therefore, even ECS delayed by only 0.5 seconds, which is supposed to be a laboratory model of the ostensibly permanent "final RA" induced by a disruption of short-term consolidation, doesn't produce a loss of memory from storage. The memory is still there -- otherwise, the reminder wouldn't have had any effect (yes, Miller did the obvious control experiments).
And if the memory is still there, what are we to make of short-term consolidation.
So the whole notion of consolidation has problems. Doubtless something goes on at the cellular and molecular level to "fix" a permanent record of an event in memory, but it's not clear that it's best described either by M&P's original idea of consolidation, or the standard model.
In fact, long before the standard model of consolidation was articulated, there were anomalous results from standard verbal-learning paradigms that cast the notion of consolidation in doubt, and suggested an entirely different way of thinking about the causes of forgetting.
First, and in a sense anticipating the later problems caused by extensive retrograde amnesia, it turned out that RI didn't disappear after just 6 minutes, as M&P thought it did. Indeed, significant RI effects have been documented even when the interpolated learning occurred as long as 6 weeks after the original learning. That's just too long to be accounted for by the interruption of the process of consolidation.
Second, and more important, it turned out that the type of interpolated experience mattered. M&P thought that "mental exertion" of any sort was sufficient to disrupt consolidation. But it turned out that the interpolated performance of a nonverbal task had little detrimental effect on verbal memories, and vice-versa. Indeed, the extent of RI proved to be a function of the similarity between the original material and the interpolated material.
Third, and perhaps most important, was the discovery of proactive inhibition (PI). In retroactive inhibition, subject who study an A-B list followed by a C-D list remember the A-B list more poorly than those who rested during the C-D phase. In proactive inhibition, subjects who study an A-B list followed by a C-D list also remember the C-D list more poorly than those who rested during the A-B phase. Consolidation failure can account for the retroactive effect of C-D on A-B, but it cannot account for the proactive effect of A-B on C-D.
Accordingly, some theorists, led by McGeoch (1942; pronounced Mugyoo) began to develop an alternative account of forgetting that accepted the phenomenon of retroactive inhibition, but abandoned the concept of consolidation. For these theorists, traces were never destroyed by consolidation failure. Instead, traces remained permanently available in storage, but suffered interference with accessibility at the time of retrieval. This accounted for both retroactive and proactive inhibition effects with a single principle, interference.
It wasn't just McGeoch, of course. There evolved a whole tradition of interference theory in the study of human learning and memory, in which theorists such as Benton J. Underwood (at Northwestern), Leo Postman (longtime UCB faculty member, and founder of the Institute of Human Learning, which became our present Institute for Cognitive and Brain Sciences), and Arthur Melton (at Michigan) were also involved. I have provided a fuller account of interference theory elsewhere. Here I only offer some examples of interference theory at work, to complete the story of storage and consolidation.
Link to a more extensive account of interference theory.
Click on "Interference" in the navigation bar above.
Interference theory is fairly represented by
McGeoch's (1942) response competition theory of
forgetting. McGeoch began with a distinction between
storage and retrieval, and argued that items in memory storage
competed with each other at the time of retrieval. This
competition was based on similarity. Thus, McGeoch was
able to give the same theoretical account for retroactive
inhibition as for proactive inhibition.
Note that there is a technical distinction between inhibition and interference. Inhibition is the phenomenon to be explained, and interference is the explanation of the phenomenon. These days, the psychology of memory is so thoroughly infused with the notion of interference that retroactive inhibition is used synonymously with retroactive interference, and proactive inhibition with proactive interference. But there is a technical difference between the phenomenon and its explanation. But now that you appreciate this difference, you can forget it.
generated a huge number of experiments devoted to understanding
the effects of item similarity on memory -- not only the
negative effects, as in RI and PI, but also the positive
effects, as in transfer effects. There emerged a number of
different paradigms for studying paired-associate learning, the
results of which are elegantly summarized by Osgood's (1949) transfer
In general, the interference theorists abandoned
serial learning for paired-associate learning, and savings in
relearning for % correct (or something like it). And then
they were particularly clever in adapting their memory
tests. Not only were there different list-learning
paradigms, but there were also different paradigms for testing
In this way, researchers in the interference tradition learned that the B targets were not erased from memory after all. Which, of course, goes against consolidation, and supports interference theory. That's not to say that consolidation doesn't occur. Obviously it does. Or, at least, we think it does. But consolidation failure doesn't account for retroactive inhibition -- or for much of any other forgetting that occurs after encoding.
Although cognitive neuroscientists, especially those involved in the search for the molecular and cellular bases of memory, have revived the notion of consolidation, and have set out on a quest to identify its molecular and cellular mechanisms, the concept of interference remains the dominant principle of forgetting to this day.
As an example of this, consider the fan effect (Anderson, 1974), which is a primary piece of evidence supporting Anderson's (1976) ACT -- which, in turn, is discussed below in the section on "Representation". In a typical experiment, Anderson taught his subjects short sentences about professionals in locations, such as:
|Notice in this example, that there is|
|The doctor is in the bank.||
|The firefighter is in the park.||
|The lawyer is in the park||
|The lawyer is in the church.||
Subjects memorize these so thoroughly that they never make mistakes in recognizing sentences that they have studied: the measure of memory here is reaction time to verify sentences as correct. The principle finding of this research is that response latency increases as a function of the number of facts memorized about a particular professional or a particular location. That is, if there are two facts about a profession (like the lawyer), it takes more time to verify either of them than it does to verify the single fact about the doctor; and if there are two facts about location (like the park), it takes more time to verify either of them than it does to verify the single fact about the bank.
This is called the fan effect because, Anderson views concepts like lawyer and park as represented as nodes in an associative network, with facts involving these concepts fanning out from the nodes. In Anderson's ACT model, retrieval can trace along only one of these associative pathways at a time. Thus, the more you know about a concept, the longer it takes you to retrieve any piece of knowledge about that concept. This is sometimes known as the paradox of interference. The fan effect is a particularly vivid demonstration of interference.
Interference theory has its problems too. In
the earliest versions of the theory, interference was generally
considered an automatic process (not that the theorists used
these precise words), that happened unintentionally. But
it turns out that interference is, to some extent, under the
control of the subject: for example, it can be ignored.
Still and all, interference theory provides the best account
available of why we forget things over time.
All of which suggests that the Time-Dependency
Principle might better be replaced by something different --
namely, an Interference Principle.
An Interference Principle (?)
Most forgetting occurs by virtue of interference from other items stored in memory.
again, not all time-dependent forgetting is mediated by
interference, so I think we should stick with the principle of
You don't have to be a psychologist to emphasize the role of interference in memory. Consider this passage from Arthur Conan Doyle's A Study in Scarlet (1887), the first "Sherlock Holmes" novel. Holmes has just met Dr. Watson for the first time, and mentioned that he intends to forget that the Earth revolves around the Sun. Watson doesn't understand, so Holmes explains.
"You see... I consider that a man's brain originally is like a little empty attic, and you have to stock it with such furniture as you choose. A fool takes in all the lumber of every sort that he comes across, so that the knowledge which might be useful to him gets crowded out.... Now the skillful workman... will have nothing but the tools which may help him in doing his work, but of these he has a large assortment, and all in the most perfect order.... Depend upon it: There comes a time when for every addition of knowledge you forget something that you knew before. It is of the highest importance, therefore, not to have useless facts elbowing out the useful ones.
This page last modified 10/04/2014.