Memory

In the study of memory there have been many metaphors adopted in the search for an explanation of the memory process. The fourth century B.C.E. Greek philosopher Aristotle compared memorizing to making impressions in wax, and the idea that memories are copies of reality that a person stores and later retrieves has been widespread. This is sometimes called the storehouse metaphor, and many of the ways in which people talk about memory (searching for memories, bringing them back from the recesses of one's mind) assume such a metaphor. The computer metaphor that has been popular with psychologists researching memory is a version of the storehouse view. It conceptualizes the stages involved in remembering in terms of encoding, storage, and retrieval in which information is entered into memory, retained, and then found again at a later time. Thinking about remembering in this way can be valuable, but it can lead to the incorrect assumption that what is remembered is a simple copy of what was originally experienced. In reality, much that is remembered captures the gist rather than the details of the original experience, and remembering is often a process of reconstruction. Examples of constructive remembering can be found in research on false memories. Elaborate and detailed false memories of events from an individual's past can be easily created. More mundanely, hearing a list of close associates to a particular word leads to recall of the word itself even though it was not presented. One alternative to the storehouse metaphor is the correspondence metaphor that emphasizes the deviation between the memory and the original experience.

Memory Structure

Researchers who study memory use a number of terms to subdivide the enormous field. One major distinction is that between explicit and implicit memory. Explicit memory refers to the conscious recall of information. Conscious awareness of past experiences involves explicit memories. Often, however, people are influenced by experiences that are not consciously recallable. For example, the ease and speed with which a person solves the anagram rbocoilc depends upon how recently the person has encountered the word broccoli. This facilitation reflects implicit memory. Processing of new information is primed by past experiences without conscious awareness. The distinction between explicit and implicit memory may reflect different underlying memory systems. Quite different timescales and sensitivities have been demonstrated for some explicit and implicit memory tasks. The differences may arise, however, from the processing requirements of the tasks rather than from different memory systems.

A distinction that overlaps with explicit and implicit memory is that between episodic and semantic memory. This distinction, associated with Endel Tulving, is between memory for events and memory for facts. Episodic memory is for events that people can remember happening, whereas semantic memory is for facts that people know about the world without necessarily retaining any recollection of the situation in which they learned the information. One's memory for eating breakfast on a particular morning is an episodic one, whereas one's memory that Coca-Cola is a drink is a semantic one. One area of episodic memory is autobiographical memory–memory for personal events in one's own life. Autobiographical memories from the first two years of life are very rare, while memories from the late teens and early twenties are more frequently held than the average. Certain autobiographical memories seem to be so distinct and full of the apparently irrelevant details from the original event that they have been called flashbulb memories because the nature of the memory is similar to a photograph of the moment. Archetypal examples of flashbulb memories are associated with hearing or seeing particularly dramatic events such as the assassination of a famous person or a major accident.

Submemories. One approach to understanding the structure of memory has been to seek separate submemories that are responsible for retaining information over differing time periods. In 1968 Richard Atkinson and Richard Shiffrin proposed a model with three types of memory: a sensory store, a short-term store, and a long-term memory. Visual information, for example, is believed to be retained for about one second in a sensory store while perceptual processing takes place. Similar sensory memories aid in the processing of acoustic and other inputs. Beyond the perceptually based sensory memories is the short-term memory, which retains information for a few seconds before selected elements of that information are transferred to a long-term memory. Atkinson and Shiffrin recognized that there were control processes in short-term memory that influence what is attended to and processed. The Atkinson and Shiffrin model has been elaborated into the working memory system, which has been particularly investigated by Alan Baddeley and his colleagues. Baddeley has subdivided the working memory into several subcomponents, the most heavily researched of which are the phonological loop, the visuospatial sketchpad, and the central executive. The phonological loop holds a couple of seconds of speech sounds and plays a role in reading. The visuospatial sketchpad is used in the creation of mental images and in the solution of visual and spatial problems. The central executive is a controlling attentional system that supervises and coordinates current cognitive processing.

Formal models of memory. A number of formal models of memory that can be run as computer simulations have been developed. Among the most influential of these are Jerome Raaijmaker's and Richard Shiffrin's 1981 SAM model, James McClelland, David Rumelhart, and Geoffrey Hinton's 1986 PDP model, and John Anderson's 1993 ACT model.

SAM (Search of Associative Memory) is a mathematical model based upon items and the strength of associations between them. It is particularly appropriate to the learning of lists of words. Each word has a memory strength as a result of it being studied, and each word has an associate strength with the other words in the studied list. The memory strength is combined with the association between the word and the context in which it was learned to produce a strength that is the basis of recognition or retrieval. The model can account for many of the memory phenomena associated with the learning of lists, but it shares with the other two formal models described here the difficulty that many of its assumptions are not based on observations and are difficult to test.

The PDP (Parallel Distributed Processing) model is a neural network model inspired by the analogy of neural circuits in the brain. The network consists of units that are connected to form a network. The strengths of the connections (weights) are adjusted as the network is trained to produce correct responses. Activation spreads through the network and the weightings direct that spread. A response is selected when it achieves a sufficient level of activation. One feature of neural network models is that memory is not located in one place but is captured by particular patterns of activation over many units and links. The neural network models are attractive in apparently simulating the structure of the brain. The choice of the particular structure of units and their interconnections, however, turns out to be important for each simulation of human memory. A general representation that is applicable to many types of remembering has yet to be developed.

The ACT framework is a production system theory for both memory of facts and skills. Anderson has developed several versions of ACT including ACT-R (Adaptive Control of Thought-Rational). Production rules are condition-action rules of the form: If this is the condition, then execute that action. Within the system, units of information are linked by associations, with the association strength being increased through use. The ACT models were developed to account for problem solving and skill acquisition as well as memory. As with the other formal models discussed here, there are many assumptions that make a model difficult to evaluate.

Memory Functions

What is remembered of a particular event depends upon the way in which it is processed. Elaborate processing that emphasizes meaning and associations that are familiar leads to good recall. So, for example, the word albatross would be remembered poorly if only the font in which it was printed was noticed and little thought was given to its meaning. It is much more likely to be remembered, however, if at the time the word is read the reader thinks about how albatrosses are white seabirds living in southern oceans. On the other hand, if what is encountered is difficult to understand, then not only will it be poorly remembered but what is remembered may be distorted by an effort to comprehend the meaning.

The processing of new information draws very heavily upon memory of past experience. Schemas have been developed for often-encountered familiar situations such as going to a supermarket or eating at a restaurant. These schemas guide understanding and memory of the new events but may also lead to memory errors by adding expected events that did not actually occur. Information that is organized on the basis of one's existing knowledge is much easier to learn and remember than is disorganized information. So, for example, a list of the names of animals is much easier to memorize if it is categorized according to type of animals (domestic, farm, wild) and if the categories are laid out in a structured way. Experts in an area memorize new information within their area of expertise much more quickly than do novices. So, soccer fans easily learn new soccer scores and chess masters memorize real board configurations easily.

When material is restudied to strengthen the memory of it, the shorter the interval between the first and second study periods, the less the improvement in recall. This spacing effect is large, so that studying in two spaced sessions can produce twice as much recall as a single session of equal length. The rereading of factual material makes only a small contribution to the further learning of it. Testing oneself by retrieving studied material, however, is a particularly effective technique for improving memory.

What is remembered depends upon the information that is available to cue recall when it is retrieved. In 1983 Tulving summarized much research in the encoding specificity principle. This principle asserts that retrieval is successful to the extent that the cues available at retrieval match those that were processed by the learner at the study phase. The retrieval cues may be aspects of the material that was studied, but they also include environmental cues and the mood and mental state of the learner.

The learning of information that is similar creates a problem for retrieval. There is interference from similar material learned earlier (proactive interference) and from material encountered since the original learning (retroactive interference), and these reduce recall. More insidious are misinformation effects. These occur when misleading information is presented, for example, to eyewitnesses during questioning. The misleading information is then frequently recalled, and the original information becomes very difficult to retrieve.

When tested across time, forgetting follows a logarithmic curve–information loss is rapid initially but then information is lost more slowly. Nevertheless, the fate of information that has been initially very well learned is rather different. Where facts, names, or foreign-language vocabulary have been used repeatedly but are no longer regularly recalled, the pattern of their forgetting is an initial loss over a three-year period, after which recall may be equally good with delays of one or twenty-five years.