psychology, we analyze the processing of episodic memories --
really, any memories -- in terms of three stages:
The stage analysis of memory is based on the library
metaphor of memory, which in one form or another goes back
to Aristotle. In the library metaphor, processing a memory
is likened to processing a book in a library.
Thus, failures of memory are like failures to find a book in a library.
So let's begin at the beginning, with the processes by which a trace of experience is encoded in memory.
Perhaps the easiest way to construe the encoding stage is in terms of repetition.
Recall that Ebbinghaus (1885) first demonstrated that retention, measured by savings in relearning, was a function of the number of repetitions during the learning phase. This was consistent with the assertion by the philosopher J.S. Mill, that associations were stamped into the mind by means of repetition.
A little later, the importance of repetition was restated by Thorndike (1898) in terms of his Law of Use: associations which are repeated are strengthened, while those which are not repeated are weakened.
And much later, in the Atkinson and Shiffrin multistore model of memory, items were transferred or copied from the short-term store into the long-term store by means of rehearsal, which is another name for repetition.
The idea that repetition is critical for memory encoding has great intuitive appeal -- how often have we tried to memorize something -- a name, a telephone number, a chemical formula -- by repeating it over and over to ourselves. But it turns out that it isn't true.
A study by Craik and Watkins (1975) directly examined the effect of rehearsal, with a clever procedure designed to control rehearsal directly. Their subjects were instructed to listen to a series of word lists, and to be prepared to report the last word that began with a particular letter, such as P. The appropriate strategy in this task is to identify the first critical word, such as peas, and rehearse it until the the next one, potato, occurs, drop peas and pick up potato, and rehearse that word until the next critical word, punt, appears, repeating the cycle as each new critical word is presented.
Subjects are, of course, able to perform this task almost perfectly, with greater than 95% accuracy. the trick in the experiment is that the critical words are strategically spaced, so as to control the amount of rehearsal devoted to each one -- or, put another way, the amount of time each word resides in primary memory.
After subjects had performed this task on 27 lists of words, they were surprised with a test of recall -- for all items, critical or neutral, from all lists. The subjects had virtually no memory for the neutral items: this was to be expected, because these items had not been subject to any rehearsal at all. The main analysis looked at recall of the critical items as a function of the number of intervening items -- meaning the number of repetitions, or time in primary memory. The data was (sic) clear: retention did not vary as a function of rehearsal. There was no better memory for items rehearsed for 12 cycles than there was for items rehearsed for 1-2 cycles.
The Craik & Watkins experiment refuted one assumption of the Atkinson & Shiffrin multistore model of memory: that encoding in the long-term store occurred as a function of time in residence in the short-term store.
But more important, the experiment motivated a
distinction between two types of rehearsal:
The importance of elaborative rehearsal was dramatically illustrated by a series of experiments on incidental memory. In these studies, subjects were not asked to memorize a word list (that would be intentional memory). Rather, they were asked to perform some other task, irrelevant to memory, and then surprised with a test of memory -- that is, to retrieve memories that had been formed incidentally as a result of task performance.
This procedure was invented by Tresselt & Mayzner (1960) and revived by Hyde & Jenkins (1969), but it was popularized in a pair of now-classic studies by Craik & Lockhart (1972) and Craik & Tulving (1975).
in the Craik experiments, words were presented
under different orienting tasks.
These various orienting tasks were intended to control the "level" at which each item was processed. In theory, the orthographic and phonemic analyses were relatively shallow, in that they processed perceptual but not semantic features of the words. By contrast, the category and sentence tasks were relatively deep, because they required meaning and other linguistic analyses.
After going through the entire list, the subjects were surprised with a test of memory -- usually recognition, as levels of recall are usually quite low in incidental memory.
Craik & Tulving's Experiment 1 revealed a clear depth of processing effect: there was relatively poor memory for items subject to structural (orthographic) processing, better memory for items subject to acoustic (phonemic) processing, and best memory for items subject to semantic analysis (in the category and sentence conditions).
But this effect was equivocal, because the structural task does not require the subject to read the entire item. All he must do is to inspect the word for upper- or lower-case letters. Moreover, the various tasks obviously differ in difficulty. Because easier tasks can be performed in less time than more difficult tasks, there was an additional confound between the orienting task and the opportunity for rehearsal. And, in fact, the different orienting tasks did entail different response times -- meaning that the depth-of-processing effect might have been an artifact of rehearsal after all.
A later experiment unconfounded depth and response time, by devising a difficult structural task that required more time, and also required subjects to read the entire item. In this new orthographic task, the subjects were asked to classify words with respect to their pattern of vowels and consonants, such as CCVVC (as in brain) or VCCCV (as in uncle). A check on response latencies showed that, in fact, the structural task took much longer than the semantic task. Nevertheless, there was still an effect of depth of processing on recognition. Apparently, the depth of processing effect is not an artifact of rehearsal, and retention really is a function of the type of processing devoted to each item.
On the basis of results such as these, Craik and Lockhart (1972) underscored the importance for memory of depth of processing. In their view, memory is a byproduct of perception, during which events are analyzed at different levels -- first for their physical or sensory features, and later matched against stored abstract knowledge. For Craik and Lockhart, "deep" processing entails semantic or other "cognitive" analyses, as opposed to mere "perceptual" analyses. And they concluded that trace persistence is a function of the depth to which an item is analyzed.
The basic depth-of-processing effect has been replicated innumerable times (it can even be shown in groups in a classroom demonstration). However, the interpretation of the effect has undergone some changes. Originally, Craik and other theorists took "depth" literally. They viewed the various processing stages as ordered serially, so that subjects first performed a physical analysis, then an acoustic transformation, and finally a semantic analysis.
Note that this serial order preserves the essential features of the multistore model. But in Craik's view, these were three separate processes, not three separate stores. A major point of the depth-of-processing experiment was to refute the multistore model. For C&L, long-term retention was not a function of encoding a trace in a long-term as opposed to a short-term store. Rather, shallow processing allowed a trace to persist for only short intervals of time, while deep processing enabled the trace to persist over longer intervals.
Now, however, depth or level
is more of a metaphor, and other theorists have substituted
other terms to characterize what is going on.
The Elaboration Principle
For the purposes of this principle, elaboration
may be variously defined:
Depth of processing theory, and the concept of
elaboration represent a breakdown of the structural distinction
between short-term or primary memory and long-term or secondary
memory. In DoP theory, there is only one system for memory
storage, and "primary" or "short-term" memory simply refers to
that portion of a unitary memory store which is currently in a
high state of activation.
Notice that the elaboration principle applies to individual list items. Retention is improved if each new item is connected to previous knowledge. But it turns out that retention of individual items is also improved if we can connect them to each other -- according to what I will call the organizational principle. I already introduced the concept of organization in memory as an early manifestation of the cognitive revolution in memory theory and research, and now it is time to discuss it in more detail.
The principal evidence for organizational processing comes from the observation of systematic discrepancies between the order in which items are presented at the time of encoding and the order in which items are reported at the time of retrieval. In serial recall, such as employed by Ebbinghaus, these discrepancies are discouraged by the instructions given to the subjects. But in free recall, such as the tests used by the interference theorists, they are commonly observed.
For example, in the serial position effect, described in the lectures on the multistore model of memory, it is common for subjects to output the last items in the list first, as retrieval from so-called primary memory takes advantage of the final items high state of activation. Then subjects typically pick up the beginning of the list, retrieving items from so-called secondary memory, and proceed through the list.
If retrieval from primary memory is not an issue, there is a general tendency for subjects to recall a series of events in their original temporal order. (There are many exceptions to this rule, but it is a useful generalization).
However, serial organization can also be induced
by the method of incremental presentation introduced by
George Mandler and Dean (1969):
Serial organization is reminiscent of Ebbinghaus, and the classical principles of association, in that associations appear to be formed between contiguous items on a list. Mandler has argued that serial or temporal organization is based on what he calls pro-ordinate associations, in which each item serves as the cue for the next one, capitalizing on subjects' strong tendency toward forward association. Mandler argues that pro-ordinate association are not just a mater of association by contiguity. Rather, temporal organization can also reflect a kind of narrative structure in which memory preserves the temporal or causal relations among events (Mandler's wife, Jean, calls this form of organization schematic organization).
Serial organization is interesting, especially when it occurs spontaneously, because it is not simply a matter of association by contiguity. Still, the real interest in organizational activity comes from input-output discrepancies in which the recall of list items does not match the order of presentation
Apparently the first investigators to notice this kind of discrepancy were Jenkins and Russell (1952; this is the same Jenkins as in the Hyde and Jenkins experiment). J&R formed a list of words consisting of common associates, such as salt-pepper, black-white, dog-house, and ocean-blue, and then presented them to subjects in a pseudorandom order (pseudorandom, not perfectly random, because they did not permit associated items to appear in adjacent positions). Performance on the recall test was very good, with subjects recalling an average of 50% of the items after only a single study trial.
But J&R's primary interest was in organizational activity, and so they examined three types of re-pairings: forward (as in salt-pepper), backward (as in pepper-salt), and arbitrary (as in salt-house). They found that forward and backward pairs occurred with relatively high frequency, compared to arbitrary pairs. Thus, their subjects demonstrated associative clustering, or a tendency to reorganize the studied material such that semantically associated items were linked together in memory, even if they were not linked together in experience.
Thus, the list items were reorganized in memory, following an organizational scheme that reflected pre-existing associative structures. Following G. Mandler, we can call these associations co-ordinate associations, which reflect the horizontal associations among items at the same conceptual level.
Still another form of organization was discovered by Bousfield (1953), building on earlier hints from a study by Bousfield & Sedgewick (1944). They constructed a list of words consisting of 15 instances of each of 4 conceptual categories, such as animals (e.g., giraffe, baboon, zebra), names (e.g., Amos, Gerald, Byron), professions (e.g., milkman, typist, florist), and vegetables (e.g., eggplant, parsnip, garlic). These items were arranged in random order (meaning that adjacent items were occasionally from the same conceptual category).
Examining the organization of recall after a single study trial, Bousfield observed considerable clustering of recall by conceptual category, compared to a simulation of chance performance. That is, there were fewer "singletons" (recall of only a single category member) than would be expected by chance, and more "triples", "quadruples", etc. Furthermore, Bousfield found that the extent of clustering was correlated with the sheer amount of recall (r = .36, compared to r = -.04 in the simulation. In addition, he found that most intrusions in recall -- the incorrect recall of items not actually presented in the list -- came from the conceptual categories represented on the list -- the false recall of lion or George, for example, or doctor or broccoli. These intrusions, combined with the clustering itself, indicated that subjects picked up on the categorical organization of the list at the time of encoding, and used it to guide their recall.
An experiment by Bousfield and Cohen (1953), employing much the same procedure as Bousfield (1953) except with multi-trial free recall (i.e., 5 study-test trials as opposed to a single one), showed that category clustering increased as the subjects memorized the list. As the number of items recalled increased, so did performance on an index of category clustering.
It is important to understand the difference between category and associative clustering. Words like giraffe and zebra are not associatively related: even though they belong to the same conceptual category, subjects don't respond with zebra when presented with giraffe, the same way they respond to pepper when presented with salt. The items are not associatively related, but they are conceptually related, in that they are both instances of some superordinate concept. Following G. Mandler, we can say that category clustering is a reflection of sub-ordinate (as opposed to co-ordinate) associative structures, which reflect the vertical associations between category labels and their instances.
In temporal, associative, and category clustering, subjects pick up on structural relations between items that have been built into the list by the experimenter. But subjects can also impose their own, idiosyncratic, organization onto a list, even when the items are ostensibly unrelated to each other. This is known as subjective organization.
Subjective organization was first demonstrated by Tulving (1962), who created a word list in which there was no pre-existing structure. None of the items were associatively or conceptually related to each other (or, at worst, the relations were very remote). And Tulving precluded serial organization as well, by presenting the items for several study-test trials, presenting the items in a new random order on each trial.
Nevertheless, across trials individual subjects developed stereotypy in recall: each subject recalled the items in an increasingly constant list order, despite different random orders of presentation. This order of recall was different for each subject - -stereotyped, but idiosyncratic, but still capable of quantification. Often, the subjective organization took the form of a story, or perhaps a description of an image -- for example, a subject might remember a boy sitting at a table, with his dog at his feet, looking out a window through iron bars at the stars in the blue night sky (oh -- and there's also a pepper shaker on the table!).
And again, as with Bousfield and Cohen (1953), Tulving observed that recall and subjective organization rose together across trials.
Recall and organization clearly go together, but does organization cause recall to improve? After all, correlation does not necessarily imply causation. It might be that good recall causes organization to occur, or both recall and organization might be caused by some third variable.
G. Mandler (1967) confirmed the causal relationship between organization and recall in an experiment using an incidental learning paradigm. Subjects were presented with a list of words, each printed on a separate card, and they were asked to sort them into 2 to 7 categories, any categories, of their own choosing. This sorting task was then repeated over several trials. A control group was simply asked to read the items, one at a time, for the same number of trials as given to the experimental group.
In these lists, the items were not particularly related to each other: they were not categorized lists like those used by Bousfield (1953). Thus, the organizational activity displayed by these subjects was closer to Tulving's "subjective" organization.
Within each group, half of the subjects were told to anticipate a memory test, while the others were not. Thus, Mandler was able to compare incidental with intentional recall. With no sorting instructions, and no expectation of a memory test, recall was comparatively poor. In the other conditions, however, recall was pretty good, and in fact incidental memory by subjects in the sorting condition was superior to intentional memory by subjects in the non-sorting condition.
The subjects were free to select their own number of categories, between 2 and 7, and recall correlated with the number of categories. Recall was optimal with 5-7 categories, an outcome that is reminiscent of Miller's "magical number 7, plus or minus 2".
Evidently, then, organizational activity causes good memory, because organization improves memory even when the subject does not expect a future memory test. In addition, organizational activity appears to be part and parcel of intentional memorization: subjects who are trying to memorize also organize. Presumably, organizational activity takes place at the time of encoding. But it's clear that the categories can also serve as retrieval cues. Still, the categories must be noticed, or established, at the time of encoding so that they can be retained, and used at retrieval.
The Organization Principle
The accessibility of an item in memory is a function of the degree to which individual items are related to each other at the time of encoding.
Organizational activity reflects what Bartlett (1932) called the subject's "effort after meaning". It doesn't happen automatically. The subject must notice organization when it is present, and must impose organization when it is not. Either way, if no organization occurs, memory will be poor.
Organization can be variously defined in terms of the temporal associative, or conceptual relations among list items. But it can also be defined subjectively, in terms of a story or an image. Or just some idiosyncratic categories: for example, dog, child, pictures, jewelry, and bonds are all things that one would take out of a house in case of a fire..
Elaboration and organization are processes which operate at the encoding phase of memory processing, reflecting cognitive activity at the time an event occurs, which determines the ease with which that event will be remembered later.
classic experiment, Rogers, Kuiper, and Kirker (1977) employed a
variant on the standard levels-of-processing experiment,
employing personality trait adjectives (such as neurotic
and extraverted) as stimulus materials. Their
experimental design contained two standard LoP conditions, plus
a new one involving self-reference:
Rogers et al. observed a huge LoP effect, such that self-referent processing yielded far superior memory, even compared to the usual semantic processing. This is now known as the self-reference effect (SRE) in memory. Given the conventional interpretation of LoP in terms of the amount of contact between individual items and pre-existing knowledge, Rogers et al. inferred that the self, viewed as a schema, is perhaps the biggest knowledge structure in the human mind.
This interpretation of the SRE was very popular,
but there's a problem: the experimental design of Rogers et
al. confounds elaborative and organizational processing.
In a subsequent study Stanley Klein (the one at UC
Santa Barbara, not the one at UC Berkeley, and who discovered this confound),
who was the first to notice this problem, cleverly solved it by
creating a semantic task that also encouraged organization into
two categories, while at the same time creating a self-referent
task that used a different sentence frame for each
Klein's stimulus materials were body parts as opposed to trait adjectives, so the first question was whether you could get the traditional SRE with such materials. Analyzing just the traditional self-reference design, comparing unorganized semantic processing with organized self-referent processing, Klein obtained the traditional SRE.
But the really important question was answered by the full 2x2 design, which showed that organized semantic processing yielded memory that was just as good as organized self-referent processing, and that unorganized self-referent processing yielded memory that was as poor, relatively speaking, as unorganized semantic processing. Put another way, almost all the variance in memory was accounted for by organizational activity, and almost none of it was accounted for by self-referential processing.
Put another way, the SRE is an artifact of organization, and has nothing to do with self-reference.
Which is interesting in and of itself. But in the present context, Klein's finding is important because it shows that elaboration and organization really are different.
So, now we have discussed three ways in which
processing affects the encoding of a trace in memory.
The importance of elaboration and organization remind us that, in the final analysis, memory is a byproduct of perception, including the processing devoted to an event at the time it occurred. Encoding makes a trace more or less available in permanent memory. Elaboration and organization, operating together, make for a rich and distinctive memory trace -- one which is easy to retrieve, even after a long retention interval. If there is little elaborative or organizational activity -- if the subject does little beyond rote rehearsal - -the resulting trace will be difficult to retrieve, especially after long retention interval.
Now we turn to the retention interval itself, and examine the principles governing the storage phase of memory processing, and what transpires between encoding and retrieval.
This page last modified 10/04/2014.