Retrieval Processes in Memory

views updated

RETRIEVAL PROCESSES IN MEMORY

The processes of learning and memory are often subdivided into stages of encoding (initial learning of information), storage (maintaining information over time), and retrieval (using stored information). Processes of encoding establish some representation of experience in the nervous system, which is referred to as an engram or memory trace. Memory traces certainly have physiological underpinnings, but cognitive psychologists use the construct as an abstraction to refer to the changed state of the cognitive system before and after some experience. Retrieval processes refer to the means of accessing stored information and can be affected by a variety of factors.

Retrieval is the key process in the act of remembering (Roediger, 2000). Most experiences of life are encoded and stored (at least briefly) but will never be retrieved and thus will have no real consequence for the individual. Encoding and storage are cheap, in the sense that all events leave their traces. Retrieval is the process by which latent information is actualized in ongoing behavior. A person may or may not realize that past events are being retrieved and are affecting current behavior. Explicit retrieval is referred to as remembering; when previous behavior affects ongoing performance outside of awareness, psychologists refer to it as priming of the behavior. The distinction corresponds roughly to the contrast between explicit and implicit memory processes (McDermott, 2000).

The division of processes into those affecting encoding, storage, or retrieval seems simple in concept but cannot be completely defended in practice. One reason is that it is impossible to distinguish cleanly between encoding and storage processes, because the two are inextricably connected. When does initial learning (encoding) end and maintenance of information over time (storage) begin? There is no clear answer to this question; suggestions provided by some theorists to cut this Gordian knot are relatively arbitrary. However, separation between the bundle of processes referred to as encoding and storage, on the one hand, and those involving retrieval, on the other, can be accomplished more directly.

The general logic of this separation is to hold conditions of encoding and storage constant and to manipulate only conditions of retrieval. For example, two groups of people could be presented with material (lists of words or sets of stories) to remember and could be treated identically until the time they are tested. Then one group of people might simply be given a blank sheet of paper and asked to recall all that they can of the material. Imagine that they recall 40 percent of the materials under these free recall conditions (so called because they are given no external cues to aid recall and are free to recall material in any order). This measure might be thought to reflect the amount of information that people have encoded and stored—the amount they know—but this conclusion would ignore the possibility that the bottleneck in performance is at the retrieval stage. Perhaps the people really have encoded and stored much more, but have simply failed to retrieve the extra material. This possibility can be examined by testing another group of subjects who are given retrieval cues to prompt recall of the material. Often appropriate cues can produce great benefits relative to free recall (Mantylä, 1986).

The advantage of cued recall over free recall indicates that more information is available (or is stored) in memory than is accessible (retrievable) on a particular test such as free recall (Tulving and Pearlstone, 1966). More generally, no test of memory provides a perfect measure of information stored in memory; the retrieval processes involved in any test filter the information. At best, people assess the information that can be produced under a particular set of retrieval conditions. Although no test or set of retrieval conditions can ever provide a perfect window on the contents of memory, study of retrieval processes can proceed meaningfully in many different ways.

Repeated Testing

One straightforward way to study retrieval processes is to test people repeatedly on the same material, under the same or differing conditions. For example, people might study sixty pictures of easily named objects and then be tested on the names of those objects under conditions of free recall for seven minutes (that is, with a large amount of time so they are not rushed). After a first test, they would be given a second and then a third test under identical conditions without intervening study of the material. An almost universal finding in such experiments is that people will recall items on the second and third tests that they did not recall on the earlier tests, a phenomenon called reminiscence (Ballard, 1913). Of course, some pictures recalled on the first test might be forgotten on later tests, but surprisingly the reminiscence or recovery between tests often outweighs the inter-test forgetting. When total recall improves over tests, this phenomenon is called hypermnesia (Erdelyi and Becker, 1974). Whereas reminiscence (recall of items on a later test that could not be recalled on an earlier test) almost always occurs in experiments, hypermnesia is observed more rarely and usually under free recall conditions (i.e., without retrieval cues). Under certain conditions the phenomenon is quite reliable, so the challenge is to specify the necessary conditions for its observation. One idea is that relational processing (associating the materials with one another) protects against forgetting across repeated tests whereas processing of individual items (providing features that distinguish them from others) leads to recovery of new items across tests (Burns, 1993). Whichever is the case—and some argue for a hybrid theory—the phenomena of reminiscence and hypermnesia point up again that a single test of retention provides a faulty assessment of the amount of information stored in memory (Roediger and Challis, 1989).

Testing with Retrieval Cues

The most popular method of studying retrieval processes is by manipulating the nature of the testing conditions, particularly the types of cues given to aid recollection. One precondition for studying explicit retrieval with cues is that the rememberer must be placed in what E. Tulving (1983) has called the retrieval mode. That is, the individual must be attempting to retrieve from his or her past. For example, you might see a ladder in your environment as you walk to work and it does not cue any memories; however, if you are asked, "Tell me about an experience from your past that involves a ladder," the query will catapult you into the retrieval mode and you will probably come up with a relevant memory. In studies of cued recall the retrieval mode is assumed by giving people explicit instructions to remember, which has led some theorists to ignore the concept (because it is a constant condition). However, retrieval mode is critical to understanding remembering.

In a typical cued recall paradigm, people are given a list of words to remember that belong to common categories. The list might be composed of words such as hawk, crow, goose, woodpecker, desk, dresser,couch, and footstool, representing birds and articles of furniture. After receiving a long list with many words and categories, some people are tested under conditions of free recall (recall the words in any order) and some under conditions of cued recall (the same instruction, but now the names of the categories are provided as retrieval cues). The typical finding is that people tested with category names as retrieval cues recall many more items than those tested without cues, showing again the disparity between information that is available in memory and that accessible on a particular test. The gains from cues are genuine and not due merely to guessing items belonging to the categories, because the items used are typically not the most likely to be guessed (robin and sparrow are avoided as study materials in favor of crow and woodpecker).

What causes retrieval cues to be effective? A primary consideration is what type of information was learned and how it was encoded—what information is stored in memory. The general principle governing retrieval of such stored information is called the encoding specificity principle: Retrieval cues are effective to the extent that they help reinstate or recreate processes involved in original learning (Tulving, 1983). The idea is that events are encoded in specific ways—people retain specific coded features of their experiences that may comprise the memory traces of these experiences. Retrieval cues are then effective to the extent that they match or overlap the specific encoded features. In addition, the match between cues and traces must be distinctive for provoking a specific memory; the cue should specify one event and not many events. A cue such as "remember the lecture" is ineffective in aiding recollection of a particular lecture because it is too general. You have been to many lectures.

Numerous laboratory experiments have confirmed the essence of the encoding specificity principle. If people are in the retrieval mode, retrieval cues that more precisely match, or recreate, the original features of the learning experience (and not other experiences) promote better recall (see Roediger and Guynn, 1996, for numerous examples). This is not to say that all retrieval phenomena are well accounted for, because empirical problems exist. For example, certain types of cues that seem as if they should be effective are not; in some cases, seemingly "good" retrieval cues actually hinder rather than help recall. One example is part-list cuing inhibition, wherein giving people part of a list of studied items hurts recall relative to control conditions (Slamecka, 1968). For example, if people are given lists of words belonging to common categories (such as the aforementioned examples) and then at test time either are given only category names as cues or are given category names plus two items from the categories, recall of the remaining items from the categories will be better when only category names are given as cues. Providing some items from the category in addition to the category names will recreate the learning situation better than just giving the category names, but in fact such item cues hurt recall.

Explaining this retrieval inhibition has proved to be a challenge; much research has established the validity of the finding and eliminated many artifactual possibilities for the results (Nickerson, 1984). One interpretation links the part-list cuing effect to a cueoverload principle, an idea embedded in the foregoing paragraphs but not named. The cue-overload principle states that a retrieval cue becomes less effective as more events are subsumed under the cue (Watkins, 1975) because each extra event makes the cue less distinctive with regard to any particular event. So, for example, a category name retrieval cue provides better recall of the studied members of a category if two items were given in each category of the list rather than six items. In the case of part-list cues, it may be assumed that presentation of the category members at test somehow adds to the number of items subsumed under the category name cue and thereby reduces recall. Numerous observations accord with the cue overload principle's interpretation of the part-list cuing effect, although other theories exist as well.

The cue overload principle complements the encoding specificity principle in making sense of the variable effectiveness of retrieval cues. Further, the cue-overload idea shows that both the compatibility of cue to the encoded trace and the distinctiveness of the process matters. That is, if one has had many experiences that leave similar traces, a retrieval cue may match too many of the traces to be effective.

State-Dependent Retrieval

Alcohol and other drugs having a depressing effect on the central nervous system are known to impair retention of information. The usual interpretation is that alcohol interferes with the neural processes that underlie encoding and storage of information, or the consolidation of information. This is likely true, but may not represent the whole story of drug-induced amnesia. Retrieval factors are at work, too. Clinicians working with alcoholic patients have observed that the patient may, for instance, hide a paycheck while drunk and then not be able to remember where it is hidden when sober. However, the next time the patient gets drunk, the check may be recovered. The phenomenon suggests that successful retrieval of memories may depend on matching the pharmacological states in which information is learned and used.

This phenomenon of state-dependent retrieval (better recall when pharmacological states of learning and testing match rather than mismatch) has been verified in laboratory experiments. In one case (Eich, Weingartner, Stillman, and Gillin, 1975) volunteer students smoked marijuana or a placebo cigarette before being exposed to a categorized word list. Four hours later the subjects again smoked either a marijuana cigarette or a placebo and then were tested on the material, first by free recall and then by cued recall in which category names served as the retrieval cues.

The results are shown in Table 1; first consider the free recall results, where the number of words recalled from the set of 48 is shown. If people were sober both when they studied the words and when they were tested on them, they performed best (11.5 words recalled). If they were under the influence of marijuana at study but sober at test time, they recalled fewest (6.7). This condition represents the usual case of drug-induced amnesia, when people experience events under the drug but are sober when tested. Is this effect due only to encoding and storage factors, or are retrieval factors at work, too? This question can be answered by examining the last row: When people were drugged at both study and test time, they recalled more words (10.5) than when they were drugged only during study time (6.7). Just as in the anecdote about the alcoholics related above, retention improved when the pharmacological state at test time matched that at study time. Do not conclude from this experiment that drugs improve memory, because they usually do not. (When people learned the information sober but were tested under marijuana, they performed worse than when tested sober.)

Although state-dependent retrieval is a real phenomenon, it usually occurs only under free recall conditions, as can be verified by examining the cued recall results. The category names served as good retrieval cues, because cued recall was better than free recall in all four conditions. However, the state-dependent retrieval effect (better recall in the drug-drug study and test condition compared with the drug-sober condition) has vanished. These results are broadly consistent with the encoding specificity hypothesis. Under conditions of free recall, a person's pharmacological state can serve as a retrieval cue, and if the cues match between study and test conditions, performance is enhanced. However, when powerful external retrieval cues are provided, they overshadow the weak "state" cues and render them ineffective. This account explains the common finding that state-dependent retrieval effects are rarely found on tests employing cued recall (Eich, 1989).

Do state-dependent retrieval phenomena exist with states other than drug states? The conditions most often investigated are moods, in studies where researchers induce happy or depressed moods in people by various means prior to study or test of material. The expected finding is that congruence of mood at study and test should produce better retention than when moods mismatch. This result has been reported in some studies, but there have been numerous failures to replicate it and the reasons for this state of affairs are not well understood at this point.

Transfer-Appropriate Processing

A viewpoint related to the encoding specificity hypothesis is transfer-appropriate processing, which emphasizes that all retention tests can be considered as cases of transfer of prior experience to the test situation. Depending on the nature of the task used to assess memory, some experiences will provide good transfer and others will provide poor transfer. Further, this approach emphasizes the relativity of memory tests: Some methods of learning may prove superior for one type of test but disastrous for another. The phrase transfer-appropriate processing was first used to explain some puzzling results obtained in the levels of processing tradition (Morris, Bransford, and Franks, 1977). Under many conditions, if people study events while focusing on their meaning, they retain the events better later than if they had focused on other aspects of the events, such as what they look or sound like, while studying them (Craik and Tulving, 1975). For words, retention is better on many tests after people have generated meaningful associations for the words (thus forcing attention to their meaning) than when rhyming words have been generated (causing attention to sounds or phonemes).

Most of the tests showing the superiority of meaningful encodings have been those, such as recall or recognition, that are thought to rely heavily on meaning (Roediger and Guynn, 1996). But suppose a test were given for the sound of words following study experiences encouraging attention to either the meaning or the sound of words; e.g., the word beagle was studied in a long list and the retrieval cue is, "Recall a word on the list that rhymed with legal." When such tests were constructed, the results came out largely as expected: Having people think about the rhyming aspects of words during study produced better performance on tests requiring knowledge of the sound of the words than did study experiences emphasizing the meaning of words (Morris, Bransford, and Franks, 1977). Therefore, the ways in which one studies events are not inherently good or bad for later retention; instead, whether study strategies are good or bad depends on their relation to the nature of the test. Learning experiences transfer well or poorly depending on the nature of the test and the type of knowledge it requires. The same idea is inherent in the encoding specificity principle.

The concept of transfer-appropriate processing has been applied to several different problems. One is the explanation of differences between explicit tests of memory (those in which people are told that their memories are being tested) and implicit tests of memory (those in which people are simply given a new task and retention is measured by how prior experiences transfer to the new task). Many implicit memory tests seem to involve perceptual components and to benefit from appropriate perceptual processing, whereas many explicit tests depend upon meaningful processing. Numerous experiments have confirmed that these two broad areas of experience (perceptual, conceptual) differentially affect certain tests in the predicted manner (Roediger, 1990).