Is memory a storehouse or an action? There is no consensus on a single and unified theory of memory or even on a single general metaphor for memory (Roediger, 1980; Koriat & Goldsmith, 1996). The classical metaphor of memory describes it as a physical space where items are stored and later on searched for and retrieved. This metaphor has been very powerful and even dominant in the history of psychology. It uses some well-known source domains such as libraries, storehouses, and computers and thus helps us to transfer many inferences about memory. That is why the storehouse metaphor is so widespread. Even our terminology is influenced by it, so that we speak about storage and retrieval from memory.
On the other hand, starting with Sir Frederick Bartlett (1932), the spatial metaphor has been under continuous fire and a new dynamic and constructive view on human memory has emerged. One particularly notable new metaphor is due to Ulric Neisser (1967). He likens human memory to the constructive work of a paleontologist who uses a small set of bone fragments as well as general knowledge about dinosaurs and other similar animals in order to reconstruct and piece together the skeleton: "out of a few bone chips, we remember the dinosaur" (p. 285).
According to the spatial metaphor, memory traces are "stable objects" or "information structures" placed in a store. The "retrieval" process then attempts to locate and select the appropriate ones given a probe. Once a particular memory trace has been retrieved, all the information stored in it is accessible. In other words, memory consists of static structures and active processes. The former simply lie there, possibly indexed and organized in some useful way, while the latter operate on them when necessary. The constructive view (Bartlett, 1932; Neisser, 1981; Barclay, 1986; Brewer, 1988; Metcalfe, 1990; Schacter, 1995; McClelland, 1995; Whittlesea, 1997) takes a different perspective. It does not separate structures from processes and considers memory as a constructive process. Memory traces are conceptualized as temporary states constructed on the spot rather than as "fortune cookies" cracked open to reveal the message contained in them.
There are no true and false metaphors, and each metaphor could be useful in certain contexts. The question is which metaphor would be more useful in the context of analogy-making and problem solving. The two schools of thought have been conducting experiments in different ways. The proponents of the first metaphor have experimented mostly with simple artificial material--lists of words, lists of numbers, sets of pictures, etc. The dependent measure of main interest has been the success/failure ratio (or d' in more recent studies). In contrast, the protagonists of the second school have been studying memory in more natural settings. They have been interested in autobiographical memory, in memory for complex events or stories (like a party or a witnessed burglary or car accident). Under these circumstances what is really interesting is not whether people remember the event or not, but rather which details they do remember and what types of errors they make. Focusing on the errors people make in recalling from memory became an important source of insights. Thus the main message sent across by the storehouse metaphor is that one may have trouble finding the book in the library or perhaps that the book might have been spoiled. However, one cannot find a book that does not exist in the library, one cannot find a modified (rewritten) book, etc. In contrast, the second metaphor easily passes over the message that since the paleontologist reconstructs the skeleton (even though constrained by the given fossils) the result might be quite different from the reality. It might even be the case that the reconstructed skeleton has not existed or even that it cannot exist. The reconstruction might also be a skeleton of a centaur--an nonexistent mixture of two or more kinds of animals. The paleontologist might make a second re-construction which could be different from the first one since something new was learned in between, or some fossils have disappeared, or new ones were found.
The empirical question is whether such phenomena happen with human memory, and the answer is yes. During the long history of the second school much evidence has been gathered for false and illusory memories, memory distortions, and so on (see Schacter, 1995, for a recent review). These constructive-memory effects are especially likely when the episode that is to be recalled is complex and agrees with common-sense knowledge. These are the exact characteristics of the sources for many analogies--past problem-solving episodes, familiar events, and real-world situations rather than lists of words. Therefore we argue that the constructivist view of memory is highly relevant to analogy research and can bring important behavioral constraints for the modeling endeavor. The next section reviews some of the evidence supporting this position.
Brown and Kulik (1977) suggested the existence of a special type of memory for important events in our life which they called flashbulb memory. They claimed that "it is very like a photograph that indiscriminately preserves the scene in which each of us found himself when the flashbulb was fired" (p. 74). They presented the results of a study which demonstrated that most Americans had a very vivid memory about the assassination of John F. Kennedy including details about the place they were, the informant, the ongoing event, etc. So, they supported Livington's idea for a special neurobiological mechanism called Now print! which is triggered when we evaluate an event as very important for us. The flashbulb memory theory has inspired a whole line of research and many controversial results have been obtained (Neisser & Harsch, 1992; Conway, 1995). What is clear nowadays is that there are differences in the degree of vividness and the details that we retain about different events. It is also clear that even "flashbulb memories" are partial and probably also distorted. For the sake of accuracy, we must point out that Brown and Kulik wrote in the same article that "a flashbulb memory is only somewhat indiscriminate and is very far from complete" (p. 75).
Now, if even flashbulb memories are not complete, what about our ordinary memories? Bartlett (1932) showed that people ignore many important details of a story. Nickerson and Adams (1979) tested the memory Americans have for a commonly used object such as a penny. It turned out that on average each element was omitted by 61% of the participants. Some elements, such as the text Liberty, were omitted by 90% of the participants. Others, such as United States of America, E Pluribus Unum, and even one cent, were omitted by about 50% of them. And, of course, each of us has had personal experiences when we could recall an episode, but we could not recall some important aspects, such as the name of the person, the color of his or her eyes, or where we met.
Our inability to recall the details might mean that we have simply not attended and encoded them; in this case memory would not be responsible for the omissions. However, on a particular occasion in a specific context one might be able to recall these specific details. This means that the details are encoded but one cannot always reproduce them. There is a huge number of studies of the effect context plays on our ability to recall or recognize objects and events (see Davies & Thomson, 1988, for a review). These studies show that although some details can be recalled on one occasion they may not be recalled on another. Thus Salaman (1982) and Spence (1988), in their reports of involuntary reminding, also claim that people are reminded about the same episode on different occasions at different level of detail, omitting various aspects of the event. Godden and Baddeley (1975) had divers study the material either on the shore or 20 feet under the sea. The divers were then asked to recall the material in either the same or a different environment. Participants clearly showed superior memory when they were asked to recall in the same context in which they studied. Similar environmental context effects on recall have been found in numerous experiments (for an overview see Smith, 1988). Human memory turned out to be mood-dependent as well (for a review see Guenther, 1988). Thus when in an elated mood participants tend to produce more "happy" memories, while when in a depressed mood they tend to produce more unhappy memories. Just having some cookies in the waiting room may influence them to produce more "positively colored life experiences" (Isen, Shalker, Clark, & Karp, 1978).
Many experiments have also demonstrated robust context effects on recognition. For example, Craik and Kirsner (1974) and Kolers and Ostry (1974) have shown that the same voice (vs. different) and same typography (vs. different) facilitate performance in a memory recognition test for words. Davies (1988) provides an exhaustive review of the experimental studies of memory for faces and places. The review shows that recognizing a face in an familiar context is much easier than recognizing it in an unusual one. Thus, for example, Thomson, Robertson, and Vogt (1982) manipulated systematically the setting in which a given person was observed, the activity this person was performing, and the clothing of the person. They found that all three factors had significant effects on a later face-recognition test.
Implicit memory has also been shown to be context-specific. Thus priming effects are decreasing with every single difference between study and test conditions (Tulving & Schacter, 1990; Roediger & Srinivas, 1993).
To summarize, people make many omissions and describe objects and events only partially, but they do so in a context-sensitive manner: different omissions on different occasions. There is an apparent hyperspecificity of human memory which leads us to think that all aspects of an episode are encoded and all of them facilitate our memory for that episode, but on any occasion only a very small part of them can be reproduced. The conclusion we draw is that memory representations are very flexible and context-dependent. This challenges the classic view of memory as consisting of stable representations of past episodes and objects. Spence (1988) also concluded that memories for episodes have "no clear boundaries"-- neither in the details they describe, nor in the timing of the episode (when it starts and when it ends). He suggested that the "enabling context" which triggered the involuntary memory for the episode sets an "acceptance level", which is then used to filter out some aspects of the episode.
Barsalou has demonstrated that concepts also change their structure in different contexts. He suggested a context-sensitive representation of concepts--they are constructed on the spot rather than retrieved from memory (Barsalou, 1982; Barsalou & Medin, 1986; Barsalou, 1987; Barsalou, 1993). He studied the variability of the graded structure of concepts and demonstrated that it is highly context-sensitive. It varies substantially with changes in linguistic context and with changes in point of view. High variability occurs both within and between individuals (Barsalou, 1987). Moreover, people can dynamically change their judgments of typicality when the context changes. In a related study Barsalou (1993) demonstrated context effects on the characterization of concepts. He came to the conclusion that "Invariant representations of categories do not exist in human cognitive systems. Instead, invariant representations of categories are analytic fictions created by those who study them" (Barsalou, 1987, p. 114). Furthermore, he claimed that "... people have the ability to construct a wide range of concepts in working memory for the same category. Depending on the context, people incorporate different information from long-term memory into the current concept that they construct for a category" (Barsalou, 1987, p. 118).
The conclusion is that explaining the context-sensitive character of human memory for both episodes and concepts probably requires much more dynamic and flexible representations, which can be constructed on the spot rather than retrieved pre-packed from some static memory store.
The extensive literature on this topic shows clearly that there is much evidence for false memories, i.e. "memories" for aspects of events that did not occur. Moreover, in many cases people strongly believe in these false memories. False memories arise by two major means: either by blending two or more episodes, or by intrusions from some generic knowledge or schema. We will briefly review both aspects.
The study of this phenomenon starts probably with the wave of research surrounding the interference theory of forgetting. Although the theory itself has long been forgotten, the experimental facts that were established remain important. Basically, these studies showed the interference between the traces of two learning events. The participants studied two different lists of items. Later on, at the test session, they mixed up items from the two lists. Just to mention one particular example out of many: Crowder (1976) has demonstrated an interference effect between pair-associations learned on two different occasions. A similar effect was observed by Deese (1959), who demonstrated false memories for non-studied but strongly-associated items.
Loftus and her colleagues (Loftus, 1977, 1979; Loftus & Palmer, 1974; Loftus, Miller & Burns, 1978; Loftus, Feldman & Dashiel, 1995) developed a new paradigm for studying memory for complex real-world events such as crimes and accidents. These studies typically involve two sessions. On the first session the participants watch a slide show or a movie about some event and on the second session they answer questions or listen to narratives describing the same event. The second session provides some misinformation about the event. It has been demonstrated that even though the context of learning and the sources were very different in the two sessions, there was blending between the two episodes in participants' memory. In a recent review, Loftus, Feldman and Dashiel (1995) report: "In some studies, the deficits in memory performance following exposure to misinformation have been dramatic, with performance difference exceeding 30%. With a little help from misinformation, subjects have recalled seeing stop signs when they were actually yield signs, hammers when they were actually screwdrivers, and curly-haired culprits when they actually had straight hair" (p.48). Moreover, the same authors have shown that in many cases people do believe they have really seen the mistaken element.
Neisser and Harsch (1992) have also demonstrated that people can have vivid memories and believe strongly in them though in fact they are false. They interviewed people immediately after the Challenger accident and asked them to write down a report of how they learned about the accident, what they were doing, where they were, etc. One year later the experimenters asked the same subjects whether they still remember the accident and how they learned about it. People claimed they had very vivid ("flash-bulb") memories about every single detail. However, the stories they told on the second interview were often very different from the ones they had written on the previous one. Many participants were shocked when confronted with their original versions. Moreover, even in the face of this indisputable evidence (and what could be more convincing than an archive report written in one's own handwriting) some people still maintained that their second versions reflected better their memory of the accident. The exaggerated form of this memory distortion is called confabulation (Schacter, 1995; Moscovitch, 1995). Neuropsychological patients with this symptom report their own biography in a very creative way. The misinformation effects of Loftus, the distorted Challenger reports told to Neisser and Harsch, and the confabulation of patients were attributed by Schacter (1995) to the same possible cause: people's failure to distinguish between various sources of information about an event; that is to say from episode blending or source confusion. Since the pieces that are used in the memory-reconstruction process come from real (although different) episodes, the (false) memories constructed in this way can be very vivid and people can strongly believe they are real.
Blending of objects (as opposed to episodes) seems possible as well. Several experiments are particularly informative in this respect. McClelland and Mozer (1986) have shown that people can mix two items (words in this case) and produce an nonexistent item which is composed of phonetic elements from the original items (e.g. producing land out of lane and sand). Reinitz, Lammers, and Cochran (1992) presented people with human faces and asked them to learn them. Later on, on the test session, the participants were shown some novel faces that had not been presented before but were constructed out of elements of faces presented previously. This manipulation produced an illusion of memory for the novel faces (i.e., many participants "recognized" them as seen during the learning session). Finally, Nystrom and McClelland (1992) produced a blending of sentences which they called synthesis errors. About 10% of all errors were false recognitions of sentences in which one word came from one old sentence and another from a second one. The participants were asked to rate the confidence of their judgments and 40% of the synthesis errors received the highest possible ranking. One particularly important observation that McClelland (1995) makes based on a simulation of these data is that "Intrusions from the other sentence rush in when the most active trace provides no information" (p.78).
Another type of false memories come from intrusions from generic knowledge. Thus Bartlett (1932) showed that episodes are remembered in terms of generic schemata and their representations are systematically shifted or changed in order to fit these schemata. He demonstrated, for example, the intrusions of expectations and rationalizations which were part of participant's schematic knowledge, but were not part of the real event (in this case a folk tale). Research on autobiographical memory has also provided evidence that people use generic knowledge to fill in missing elements as well as to change existing elements in order to fit them into a schema (Barclay, 1986). It has also been shown that people systematically reconstruct their past in order to fit into their current self-image schema (Neisser, 1998; Neisser & Jopling, 1997).
Sulin and Dooling (1974) had their subjects read a brief paragraph about a wild and unruly girl. Then in one of the conditions they mentioned that the name of the girl was Helen Keller, whereas in the other condition they called her Carol Harris. Later on, they tested the rote memory of the participants for the sentences of the story. The test demonstrated robust false recognition of a completely novel sentence--"She was deaf, dumb, and blind"--in the first condition but not in the second. This intrusion obviously came from the generic knowledge the participants had about Helen Keller.
Loftus and Palmer (1974) demonstrated that subjects may claim they have seen broken glass in a car accident, whereas there was no broken glass in the slide show they had observed. Moreover, the percentage of subjects making this wrong reconstruction depended on the wording of the question ( smashed into versus hit). In other words, the reconstructed episode contained intrusions from generic knowledge about car crashes. Similar results have been obtained in numerous other experiments summarized by Loftus, Feldman, and Dashiel (1995) as follows: "Subjects have also recalled non-existing items such as broken glass, tape recorders, and even something as large and conspicuous as a barn in a scene that contained no buildings at all" (p. 48).
Williams and Hollan (1981) used the think-aloud technique to study how people recollect the names of their classmates. They found that the participants in the experiment typically first looked for a specific context (e.g., a swimming pool or a specific trip), then searched this context to find the corresponding classmate(s) who were part of that context, and finally verified the information. Williams and Hollan described memory retrieval as a reconstructive and recursive process of problem solving. Partial information about a target item is used to construct a partial description of the item and this description is then used to recover new fragments. A new description is constructed and the process continues recursively. Obviously the result will depend on the starting point and in particular on the specific context in which the memory reconstruction takes place. Kolodner (1984) also found that people tend to construct details that they do not remember. The reconstruction is based on general schemata for similar events. Thus, for example, a person would say "I must have gone to a hotel" and then possibly remember the specific hotel they were accommodated in.
Tulving (1983) also endorses the constructivist idea that memory traces result from a synthesis between stored information and current retrieval information. Schacter (1995) provides additional data from brain studies and argues that the fact that many cortical areas are jointly involved in the recollection process suggests that information from various sources is being collected in order to reconstruct the episode.
Summarizing the results from this section, we may conclude that there are no clear-cut boundaries between episodes, or between episodes and generic knowledge. Episodes may become blended and elements of generic knowledge may be instantiated and implanted into an episode as if they had been part of the event. Which particular elements from other episodes or from generic knowledge will intrude depends on the context of recall.
Recollecting an episode is not an instantaneous process. It takes time, which according to Anderson and Conway (1997) may run up to 15 seconds in a laboratory experiment. Sometimes reminding is spontaneous, but recalling an episode may also be an effortful process. Even spontaneous memories come into our minds in portions.
As remembering is a slow and gradual process, we may be interested in the order in which various aspects of the event are being recalled. It turns out that this order may differ on different occasions (Salaman, 1982; Spence, 1988). The order in which the elements of the episode are recalled must have an effect on the mapping in analogy-making. We call these effects memory order effects (to contrast them with the order effects due to the timing of perceiving-- see the end of Section 2.4.3).
Ross and Sofka (1986), in an unpublished work, describe a protocol analysis they performed on remindings of old episodes. They presented subjects with problems and asked them which old problems they were reminded of. They found that reminding was slow and gradual rather than an instantaneous process, and that it runs in parallel and interacts with mapping. In particular, Ross and Sofka found that the subjects relied on the established mapping to recall details about the old episode. In other words, this study suggests that the mapping process (and, more broadly, reasoning) influences and guides the memory process.
Here is how Ross (1989) summarized these results: "... other work (Ross & Sofka, 1986) suggests the possibility that the retrieval may be greatly affected by the use. In particular, we found that subjects, whose task was to recall the details of an earlier example that the current test problem reminded them of, used the test problem not only as an initial reminder but throughout the recall. For instance, the test problem was used to probe for similar objects, and relations and to prompt recall of particular numbers from the earlier example. The retrieval of the earlier example appeared to be interleaved with its use because subjects were setting up correspondences between the earlier example and the test problem during the retrieval" (p. 465).
This study was, however, performed in the context of a pure memory task. Subjects were not asked to solve the problems; they were rather asked to recall the problems they were reminded of. The next section looks at the complex interactions between memory, reasoning and perception in the context of problem solving.
| Next page | > > > | Interaction Effects |
| Previous page | > > > | Introduction |
| [ Kokinov's Home Page ][ Kokinov's Publications ] | |
| [ Petrov's Home Page ][ Petrov's Publications ] | |
| [ DUAL Main Page ] | [ Get pdf file ] |
| [ AMBR Main Page ] | [ Back to top ] |
Page maintained by
Alex Petrov