Kinds of Memory

8/6/2019 Kinds of Memory

1/42

pashler-44076 book November 27, 2001 9:43

CHAPTER 1

Kinds of Memory

HENRY L. ROEDIGER III, ELIZABETH J. MARSH, AND STEPHANIE C. LEE

Memory is a single term, but refers to a mul-titude of human capacities. There are manydifferent kinds of memory. Philosophers have

analyzed memory for 2,000 years; psychol-ogists have studied the topic experimentallyfor 115 years; and neuroscientists have ex-amined the neural bases of memory for thepast 70 years. All these attempts have re-vealed much about phenomena of memoryour understanding has increased in leaps andboundsbut there remains no generallyagreed-upon classication of the kinds of memory that exist. Many categorizations doexist, but the difculty is that experts dis-agree on which classication is best. Analyz-

ing this problem in 1972, Tulving remarked:In a recent collection of essays on humanmemory edited by Norman (1970) one cancount references to some twenty-ve or socategories of memory, if one is willing toassume that any unique combination of anadjectival modier with the main term refersto something other than any of the refer-ents of other unique combination (Tulving,1972, p. 382). Many more adjective-nouncombinations exist in 2002 than in 1972(implicit memory, ashbulb memory, andworking memory, to name just three terms in-troduced since then), but a universallyaccepted categorization scheme does notexist. There is no periodic table for types of memory.

The foregoing considerations make itdifcult to write a chapter on varieties of memory. However, they should not be cause

for undue alarm. Yes, the eld of inquiry intohuman memory is in ux and full of healthydisagreement. But to say that there is no uni-versal agreement on a categorization schemefor types of memory is not to say that thereis no agreement. In this chapter we presenta set of categories that reect current theo-rizing for many (and maybe most) psycholo-gists who study human memory. The typologywe present is not perfect, as we will see, andsome would disagree with it; nonetheless, weseethecategorization systemwe describehere

reected in many articles, chapters, and text-books. Briey, we consider types of memoryordered generally by their persistence, begin-ning with eeting sensory memories and thenmoving on to short-term or working mem-ory (holding information in mind and work-ing with it). We nally turn to the many typesof long-term retention. This scheme impliessharp boundaries between the length of timeinformation is held by these various types of memory, but in truth the differences are moreshades of gray; one type of memory blendsinto another. Still, the framework is useful.

The chapter begins by considering groundsfor distinguishing among types of memory.The next section presents some broad distinc-tions between different classes of memory or

1


2/42


2 Kinds of Memory

tests of memory, such as explicit and implicit,which have proved useful to the eld. Nextwe consider a brief overview of the typologyof sensory, working,and long-termmemories.

The remainder of the chapter eshes out thesedifferent types of memory.

GROUNDS FOR DISTINGUISHINGTYPES OF MEMORY

We present four different bases for distin-guishing among types of memory: phenom-enology of memory, differences amongmemory tests, dissociations between forms of memory, and different neural systems under-

lying memory.

Phenomenology of Memory

Philosophers have oftenproposeddistinctionsamong forms of memory chie y on the basisof the types of experience or the languageof expression to which they give rise. Forexample, the philosopher Gilbert Ryle (1949)distinguished between declarative memory(making declarations about the past) and pro-cedural memory (carrying out actions basedon past learning) in part on the basis of differ-ent language forms. He argued that these twotypes of knowledge are accompanied by dif-ferent verbal expressions. People say I knowthat. . . when accessing declarative knowl-edge, but I know how. . . when referring toprocedural knowledge: I know that Ottawais the capital of Canada, but I know how totie my shoes.

Similarly, Tulving (1985a) argued thatdistinct forms of memory are accompaniedby different types of mental experience. For

example, episodic memory (memory for epi-sodes in speci c times and places) is accom-panied by the experience of remembering,or mentally traveling back in time and, in away, re-experiencing the events. Tulving re-ferred to remembering as re ecting autono-

etic (self-knowing) consciousness. Semanticmemory (the general knowledge of facts) is,he argued, accompanied by noetic (knowing)consciousness. (We remember salient events

of our lives; we know Napoleon was a Frenchemperor). Finally, procedural memory inTulving s system is anoetic (not knowing) initsstate of consciousexperience. That is,com-plex, practiced procedural skills (riding a bi-cycle, tying one s shoes, serving a tennis ball)are executed with little conscious mediationand control once the action has been started.

Other forms of memory, such as ashbulbmemories, have also been advanced largelyon phenomenological grounds. For example,ashbulb memory refers to the quality of vivid recollection that exists for some memo-ries of particularly highly charged emotionalevents that are often discussed later in one slife (Brown & Kulik, 1977).

All these forms of memory and others aresupported by differences that most peopleexperience and to which they canrelate. How-ever, psychologistshave repeatedly foundthatintrospection alone is a poor means for mak-ing and defending critical psychological dis-tinctions. Usually, experientialdifferencesaretaken, at best, as a starting point for mak-ing distinctions among types of memory thatneed to be veri ed by other means. For ex-ample, Tulving (1985a) proposed a techniquethat led to a program of research to vali-datehisdistinction between remembering andknowingthepast(see Gardiner& Richardson-Klavehn, 2000).

Differences among Expressionsof Memory

Psychologists have devised a large number of ways to assess a person s knowledge and totest memory. Of course, there are standard-ized tests of general knowledge or facts, suchas the Scholastic Assessment Test (SAT), andthere are other standard tests speci cally for


3/42


Grounds for Distinguishing Types of Memory 3

memory of recent experiences such as theWechsler Memory Scale. However, psycholo-gistsstudyingmemoryhave createda plethoraof memory tests for both laboratory study and

life that test many different abilities. In thelab, people may study a long series of wordsor pictures. Tests for these materials can in-clude free recall (recall the items in any orderon a blank sheet of paper), cued recall (whatpresented item was associated with lemon? ),serial recall (report the items in order), orrecognition (pick old or studied items fromnew, nonstudied items). There are many otherforms of testing for stimuli presented in a lab-oratory context, as well as tests of knowledgeacquired in life, too (such as the test of TVshows that lasted only one season in Squire &Slater, 1975).

No one has ever proposed that all thesedifferent tests represent different forms of memory. However, some distinctions pro-posed to classify types of memory re ectdifferences observed between tests, such asdifferences between general knowledge tests(semantic memory) and those forpersonalex-periences (episodic memory). The task, then,is to nd out which memory tests behave sim-ilarly when certain independent variables aremanipulated and which tests respond differ-ently. The next section presents this logic of classifying tests based on whether they pro-duce patterns of results that are similar to ordifferent from one another.

Dissociations among Memory Measures

Perhaps the primary method of distinguishingbetween different types of memory is ndinginteractions between independent or subject

variables and performance on different mem-ory measures. For example, injury to certainparts of thebrain (especially thehippocampusand surrounding areas in the medial tempo-ral lobe) renders people amnesic for certaintypes of information (Squire, 1992). When

amnesic patients are examined on tests of memory and compared to control subjectsmatched on age and education, they typi-cally showimpaired retention on certain types

of tests but not others (for a review, seeMoscovitch, Goshen-Gottstein, & Vriezen,1994). For example, measures of short-termmemory (retention of digits or words overbrief periods without distraction) are oftencomparable to those of healthy control sub-

jects, but certain measures of long-termmem-ory (e.g., free recall of a list of words after adelay of severalminutes)aregreatly impaired.This outcome leads to the conclusion that, ata minimum, these two differentmemory tasks(short-term digit memory and long-term freerecall) tap different types of memory.

The same sort of outcome can also beobtained between measures of memory innormal, healthy adults by manipulating inde-pendent variables. That is, some factor can bevaried and be shown to have a strong effect ononemeasureof memoryand eitherno effect oreven an opposite effect on a different measureof memory (see, e.g., Jacoby & Dallas, 1981;Jacoby, 1983). Once again, at a minimum wecan conclude that the measures re ect differ-ent types of memory capacity.

There are dozens of types of memory tests,andmany canbe dissociated from oneanotherby manipulatingsubject,material, or indepen-dent variables. Test differences can be usefulfor attempting to carve nature at its joints, but there are so many differences among teststhat it is hard to know how to classify them.Although test dissociations are critical in theenterprise of identifying types of memory,most classi cation systems focus rather arbi-trarily on a few test differences while ignoring

many others.

Neural Systems Underlying Memory

In the past 20 years, much research has beendirected towards understanding the neural


4/42


4 Kinds of Memory

systems underlying memory. Just as percep-tual systems can be divided based on differ-ences in function and neural circuitry (thevisual system, the auditory system, etc.), so

the hope is that memory can be fractionatedin similar ways (e.g., Schacter & Tulving,1994). Most of the data brought to bearon human memory have come either fromstudies of brain-damaged patients or from ex-periments that use neuroimaging techniques(positron emission tomography [PET] orfunctional magnetic resonance imaging[fMRI]). These techniques have been usefulfor nding differences among types of mem-ory. As already noted,patients with damage tothe hippocampus and surrounding areas showgreat losses on some types of memory tests(such as free recall) but not on others (short-term memory tests and priming on implicitor indirect memory tests). Similarly, neu-roimaging techniques often show differencesbetween tests of memory in the componentneural processes that underlie the test. Thesefacts may help specify neural differencesbetween tests of memory. Of course, when-ever a dissociation occurs between two mem-ory tests at the behavioral level as a func-tion of independent variables, there mustbe neural differences (Roediger, Buckner, &McDermott, 1999). That is, assuming that thebrain is the cause of behavior, if there aredifferences in behavior, there must be dif-ferences in the neural mechanisms causingthe behavior. This statement is as true in un-derstanding differences between measures of memory as in understanding any other be-havioral phenomena. Therefore, differencesin neural underpinnings of memory may helpsolve theclassi cationproblemin some ways,

but dissociations among behavioral measuresof memory remain fundamental.

Although there is no acid test of any typeof memory, researchers look to convergingevidence from a variety of sources. The pri-mary types of evidence are those just dis-

cussed: phenomenology, test differences, dis-sociations among measures, and differencesin neural underpinnings.

BROAD DISTINCTIONSAMONG TYPES OF MEMORY

In this section we cover important distinc-tions that have been proposed to cut acrossa variety of types of memory. Some of theserefer to types of memory tests, whereasothers are intended to refer to temporal prop-erties of memory or different types of repre-sentation.

Declarative/Procedural Memory

As noted above, Ryle (1949) proposed a fun-damental distinction between declarative andprocedural memory, or knowing that andknowing how. The modern champion of this distinction has been Squire (e.g., Squire1987; Squire, Knowlton, & Musen, 1993),who originally distinguished between declar-ative and procedural memory but more re-cently has preferred to cast the contrastbetween declarative and nondeclarative cate-gories. Declarative memory, in this typology,is composed of episodic memory and seman-tic memory. Episodic memory is de ned asmemory for events (Tulving, 1972); one mustretrieve the time and place of occurrence inorder to retrieve the event, as in answeringthe question, Where did you go on vacationlast summer? The retrieval query speci esthe time, but in order to recall the events, therememberer must retrieve the place where theevents occurred. Semantic memory refers to

relatively permanent knowledge of the world,or generic knowledge. (Generic memory hasbeen suggested as a substitute for semanticmemory because the knowledge may notalways be meaningful; Hintzman, 1978). Ourknowledge that zebras have four legs, that


5/42


Broad Distinctions among Types of Memory 5

the chemical symbol for oxygen is O, andthat Joe Dimaggio was a baseball player, aswell as thousands of other facts, constitutesour permanent knowledge or semantic mem-

ory. One idea about the relationship betweenepisodic and semantic memory is that repeat-edly experienced events may become repre-sented in a decontextualized form in seman-tic memory. That is, the rst time you heardThomas Jefferson served as president of theUnited States, you might have encoded thefact within episodic memory, but after hun-dreds of exposures you can answer Who wasthird president of the United States? withouthaving to retrieve any speci c episode (a par-ticular time or place) in which you heard thisfact.

The distinction between episodic and se-mantic memory is clear at a descriptive level,as proposed in the above paragraph, but re-mains controversial among some researchers.Even those accepting a clear division ar-gue about the relationship between these twosystems and other systems, such as proce-dural memory. Tulving (1985b; 1999) hasproposed that procedural memory is oldestin terms of evolution and is shared by allanimals. Semantic memory is thought to bea more recent adaptation during the evolu-tion of the brain and neurocognitive systems,and episodic memory is thought to be rela-tively recent in terms of evolution and per-haps unique to humans. Tulving (1999) hasargued thatthe developmentof episodic mem-ory was necessary for humans to think aboutthe future as well as the past and was criticalfor the development of civilizations. Unlessone has a concept of the future and can think beyond the here and now, Tulving suggests,

there is no reason to farm, to build cities, andso on.

According to Squire (1987), procedural(ornon-declarative, to use the later term) mem-ory encompasses a very broad range of hu-man skills and abilities: classical condition-

ing, motor skill learning, priming phenom-ena on certain tasks, complex (skill-based)problem solving, and more. All these andothers would represent subcategories of non-

declarative memory.Distinguishing experimentally among

types of memory is a challenge. At onepoint it was popular to distinguish betweentasks that were thought to re ect (ratherpurely) episodic memory, semantic memory,or procedural memory (Tulving, 1983). Thus,for example, a standard recognition memoryexperiment was thought to re ect episodicmemory, whereas asking general knowledgequestions ( What is the capital of Missouri? )was thought to re ect semantic memory.Problems abound, however, and the simpleidea that tasks and memory systems map onto one another in a simple fashion has beenabandoned. Performance on standard recog-nition tests is now thought to re ect somecombination of episodic and semantic mem-ory systems (Tulving, 1985b), and there mayeven be a dash of procedural memory tossedin (when one claims to recognize an eventbecause it is uently processed; see Jacoby& Dallas, 1981, for the idea that percep-tual uency helps drive recognition perfor-mance).

In addition, others have argued that thereis a procedural component to all acts of re-membering, even on supposed declarative orepisodic tests (Kolers & Roediger, 1984). Themental procedures instantiated during a testcan alter the pattern of data obtained, even instandard episodic tasks such as serial learning(see Slamecka, 1977) or recognition memory.Tulving(1983,pp.303 308)showedthateventhe seemingly trivial change in tasks between

subjects saying yes or no to putting checksand crosses next to each item on a recognitiontest can affect the outcome of the experiment.However, if all tasks have a procedural (ornondeclarative)component, thenthe strongestform of the distinction between declarative


6/42


6 Kinds of Memory

and procedural memory becomes dif cult todefend. Still, in the pure cases (tying one sshoes versus remembering pictures on a freerecall test), the distinction is clearly sensible.

Explicit/ Implicit Memory

Graf and Schacter (1985) rst proposed thedistinction between explicit and implicitmemory (for a history of the idea, seeSchacter, 1987). The basic distinction is be-tween tests that directly request memoriesfrom the past (explicit tests) and those thatmeasure retention indirectly, without subjectsnecessarily being aware that their responsesreect memory at all. Explicit tests are thosesuch as free recall, cued recall, and recog-nition in which subjects are required to re-trieve events from the recent past. Implicittests are transfer tests: People are exposed tosome material and then later given an osten-sibly unrelated task (such as naming picturefragmentsor completing wordfragments suchas e e h n ). Sometimes items in a prior studyphase may help, or prime, performance on thetest. Forexample, if subjects were toseeeithera picture of an elephant or the word elephant in the rst phase of theexperiment, those whosaw the picture would be better able to namethe fragmented picture of the elephant thanwould those who saw only the word. Con-versely, on thewordfragmentcompletion test,studying the word but not the picture wouldproduce a bene t (Weldon& Roediger, 1987).The measure on implicit memory tests is typ-ically priming, or the bene t of recent expo-sure to material on a later task that measurestransfer.

Explicit memory tests measure conscious

recollection, whereas implicit tests re ectretention that has been variously describedas automatic, incidental, or even unconscious.Debate swirls around how to characterizeperformance on implicit memory tests. Atleast two types of priming occur: perceptual

(data-driven) or conceptual (meaning-driven)(Blaxton, 1989; Roediger, 1990). Tests suchas completing fragmented words or picturesare perceptual in that basic perceptual manip-

ulations such as modality of study (auditoryor visual; symbolic form, i.e., words or pic-tures) have greateffects on perceptual implicitmemorytestsand little or no effect on concep-tual implicit tests (McDermott & Roediger,1996). On the other hand, conceptual implicittests (like explicit tests) are often affected bymanipulations of meaning, such as levels of processing or the difference between gener-ating material and reading it (Blaxton, 1989;Srinivas & Roediger, 1990).

Conscious/Unconscious Forms of Memory

The term unconscious memory calls to mindthe concept of repression and the writingsof Sigmund Freud, who popularized the ideain the early 1900s (e.g., Freud, 1917/1982).Brie y, the notion is that painful childhoodmemories are too threatening to the psycheand so are banished to an unconscious state,or repressed. The memories remain active,creeping out in neurotic symptoms, in slips,and in dreams. The evidence for this formof unconscious memory is scanty (althoughthere is some evidence for it; see Erdelyi,1996), but the general idea that memories canbe outside the realm of consciousness is un-contested. Try to remember thename andfaceof your sixth grade teacher and you may wellsucceed. Until you were asked the question,that knowledge was in an unconscious state.By de nition, most of our knowledge is un-conscious (in this limited sense) at any onepoint in time.

Conscious recollection in contemporarystudies of memory refers to deliberate, effort-ful remembering inTulving s (1985a) senseof traveling back in time and reliving an experi-ence. Unconscious retention typically refersto the automatic display of past experience


7/42


Broad Distinctions among Types of Memory 7

on some test such as an implicit memorytest. In the very rst experiments on mem-ory, Ebbinghaus (1885/1964) proposed arelearning/savings technique for measuring

memory because it could potentially detectunconscious knowledge (Slamecka, 1985).Modern implicit memory tests can potentiallydo the same. In addition, Jacoby (1991) pro-posed a process dissociation procedure thatseparates contributions of memory tests intoconscious and unconscious components byemploying an ingenious combination of experimental conditions.

Voluntary/Involuntary Retention

The contrast between voluntary and invol-untary retention covers much the same con-ceptual territory as do the explicit/implicitor conscious/unconscious contrasts. There isa slightly different twist, however. Voluntaryretention refers to deliberate, willful recollec-tion; the person actively tries to remember (ason explicit tests). In its purest form, involun-tary retention refers to recollection withouteffort. However, voluntary remembering mayreect automatic processes, and vice versa.For example, if I am trying to rememberan event from my childhood and a similarevent recently occurred in my experience, Imay be primed to remember more easily thechildhood event (I might think that I am re-membering the childhood event of my ownvolition, not being aware of the unconsciousinuence of the recent experience). Similarly,in a caseof involuntaryor incidental retention,one might try deliberately to use the recentpast as an aid. One complaint about implicitmemory tests such as completing fragmented

words is that subjects may use the cues de-liberately to remember the past (e.g., Jacoby,1991).

An interesting case occurswhen consciousrecollection follows automatic retrieval asituation that has been called involuntary

conscious recollection (Richardson-Klavehn,Gardiner, & Java, 1996). Brie y, a thoughtor concept might come to mind unbidden,with no apparent source for the memory.

Later, however, the source can be determined,and the rememberer realizes, after somereection, Oh, I must have thought of thatbecause of my recent experience. The studyof conscious and unconscious experience isone of the central issues in the contempo-rary psychology of memory, which uses ex-plicit and implicit memory tests, the remem-ber/know procedure, and the process dissoci-ation procedure.

Retrospective/Prospective MemoryRetrospective memory refers to memory forthe past we re ect back and recollect whathas happened to us. Thus far we have de-scribedonly situationsinvolving retrospectivememory: Remembering one s childhood, rec-ognizing a picture studied earlier in an experi-ment, and completing a word fragment with arecently seen word are all examples of mem-ory for the past. Prospective memory refersto a situation in which the focus is on thefuture: how we remember to do things in thefuture. Outside the lab, we face this task allthe time remembering to pick up milk onthe way home, remembering to take one santibiotic twice a day, and so on. The standardlaboratory paradigm for studying prospec-tive memory requires subjects to remember toperform a secondary task (e.g., press a key)when they see a target cue embedded intheir primary task (e.g., whenever a speci edword occurs in a text; Einstein & McDaniel,1990). Prospective memory researchers have

also used more naturalistic versions of thisparadigm such as having participants remem-ber to call the lab on a particular day at a par-ticular time (e.g., Maylor, 1990). Althoughprospective and retrospective memory referto clear categories descriptively, it is not yet


8/42


8 Kinds of Memory

clear whether the two types of memory testsinvolve different processes. That is, effectsof independent and subject variables may begenerally the same for prospective memory

tests and retrospective memory tests that areequated on other features.

Code-Specic Memories

A central concept in the study of memoryis recoding: Stimuli in the outside world arenot copied into the brain, but rather recoded.Psychologists refer to modalities of coding we can remember events as mental images orverbal descriptions, or perhaps remember thesmell and the feel of past events. Psycholo-

gists have tended to focus on verbal recod-ing, the conversion of experience into wordsor schemas or scripts, with the assumptionthat verbal coding is ascendant in adult hu-mans (e.g., Glanzer & Clark, 1963).Althoughlanguage is important, spatial/imaginal repre-sentations also underlie much of experience(e.g., Paivio, 1986; Kosslyn, 1995). Memoryfor odors seems to have special properties,so olfactory coding is also important (Herz& Engen, 1996). Motor coding may be crit-ical at times; performing motor actions to acommand makes simple events more memo-rable than does either hearing the commandor imagining performance of the command(see Nilsson, 2000). Therefore, all these cod-ing modalities are ways of representing expe-rience, and each probably has distinct neuralsubstrates.

THE MODAL MEMORY TYPOLOGY:SENSORY, WORKING, ANDLONG-TERM MEMORY

The distinctions discussed earlier representterms that refer to the varieties of memoryand memory experience. We now outline onecategorization scheme of memories that isbased,roughlyspeaking,on thedegreeof their

persistence. However, we must admit at theoutset that attempts to categorize memoriesby their longevity are fraught with dif cul-ties, as sharp boundaries do not exist. One

type of memory usually blends into another.Still, the three main categorizations presentedhere represent a starting place.

Sensory Memories

Sensations are not coded instantaneously inthe brain; rather, the sensations from the re-ceptors (the sense organs) linger in the ner-voussystem as information is being processedby higher cortical centers. This sensory per-sistence re ects a eeting memory of the out-side world in what is thought to be a relativelyfaithful, unrecoded form. Although sensorymemories are thought to exist for all senses,those most studied are forvisionandaudition.

Iconic Memories

The sensory memory for vision is referredto as iconic memory. Sperling (1960) devel-oped a standard technique for its study. Hepresented subjects with arrays of letters forvery brief periods (20 ms) and asked them toreport parts of the array. The entire array wastoo large to be reported from a single glimpse,but hereasoned thatif the letters wereheldin asensory store,people could rescueitems whendirected to do so. He presented a tone vary-ing in pitch (to direct attention to one line) ei-ther just as the array was nished or at variousbrief times afterward. He wanted to see if sub-

jects could report parts of the array and if thisability would rapidly decrease (as the arrayfaded or was forgotten) from the iconic store.Sperling s experiments con rmed the exis-

tence of the iconic store and estimated its du-ration at about a quarter of a second, under hisconditions. However, other techniques haveproduced a quite different estimate, at around100 ms compared with Sperling s longer es-timate (e.g., Haber & Standing, 1970). These


9/42


The Modal Memory Typology: Sensory, Working, and Long-Term Memory 9

different estimates can be reconciled by as-suming that there are really two differentbrief visual stores with different neural bases in theearly visual system (Crowder & Surprenant,

2000). One type of visual persistence seemsto correspond to the experience we all haveof visual afterimages, such as when a photo-graphic ash persists before our eyes for abrief period after the ash.

Echoic Memories

The sensory memory for hearing is calledechoic memory for the presumed brief echothat perseveres in the nervous system after in-formation is heard. The existence of echoicmemory is supported by an auditory paral-lel of the Sperling partial-report technique

just described (Darwin, Turvey, & Crowder,1972). Subjects heard brie y presentedsounds from three different locations; in thewhole-report condition they recalled as muchas possible, whereas in the partial-report con-dition they were cued to recall from just oneof the three locations. The partial-report tech-nique led to higher estimates of echoic mem-ory capacity, although this advantage disap-peared after thesignal to respond wasdelayedby about 4 s.

As with iconic memory, there seem to betwo different forms of brief auditory persis-tence, one more short-lived than the other.The shorter form lasts only hundreds of mil-liseconds, but the longer one may last be-tween 2 and10 seconds (Cowan, 1984).Theseechoic memories presumably help in compre-hending speech and other auditorysignals thatchange rapidly over time (Crowder, 1981).

Working Memory

The term working memory is used to describea temporary memory system in which infor-mation is maintained and manipulated for ashort period of time (Baddeley, 2000a). Forexample, one draws upon working memory

when solving an addition problem such as384 + 743 without a calculator or pencil andpaper. In order to accomplish this task, onemust temporarily store the addends, add the

last digits together, maintain this outcomewhile adding the next two digits together, andso forth, until the nal solution is computed.This notion of working memory is a fairly re-cent one, and it differs from prior conceptual-izations of short-term memory in that it placesan emphasis on the active manipulation anduse of information, rather than merely on itsmaintenance or storage. However, both stor-age and manipulationfunctionsare important.We consider thestorage function rstand thenturn to manipulation.

Within the information-processing tradi-tion that postulated multiple memory stores,theso-called modal model of memory(namedby Murdock, 1974) primarily emphasized thestorage functions of short-term memory. Thismodel (see Figure1.1, adapted from Atkinson& Shiffrin, 1971) postulated sensory storagesystems through which information owedto a short-term store that held informationbrie y. The information could then be trans-ferred more or less well to a long-term storedepending on how extensively it was pro-cessed while being held in the short-termsystem. Because only a limited amount of in-formation could be maintained in the short-term store at any given time, new incominginformation replaced older information in thestore. Depending on the amount of process-ing already received, this bumped informa-tion was either forgotten or transferred to thesecond, long-term store (Atkinson & Shiffrin,1968).

Several key ndings supported the exis-

tence of separate short-term and long-termstores. As already noted, some patients withbrain damage showed a loss of long-termretention while short-term processes re-mained intact. In addition, experimental evi-dence showed different forgetting rates for


10/42


10 Kinds of Memory

Control processes(e.g., coding) Long term store

Sensory stores Short term store

Retrieval

Transfer

Environmentalinput

Decay Forgetting Forgetting

ResponseRehearsal

Visual Auditory Other modalities

Figure 1.1 The modal model of memory, adapted from Atkinson and Shiffrin (1971). The modelconsists of sensory stores, a short-term store (STS), and a long-term store (LTS). Information ows fromthe sensory stores to the STS, where it is processed and maintained via rehearsal. Transfer to the LTS isdependent on the amount and quality of processing in the STS.

short- and long-term stores, as well as dif-ferent characteristics of retained information,with phonemic coding more probable forshort-term verbal memory and semantic cod-ing more likely for long-term verbal memory(Crowder, 1976). In addition, much evidenceconcerning serial position functions in single-trial free recall also was consistent with two-process theory (e.g., Glanzer & Cunitz, 1966;Glanzer, 1972).

Much of the research in the 1960s and1970s was concerned with storage character-istics of the short-term store; less emphasiswas given to other purposes of short-termretention. That focus changed with the ad-vent of Baddeley and Hitch s working mem-ory model, which broadened the concept andled to modern notions of working memory.

Baddeley and Hitch (1974) proposed theoriginal model of working memory, whichcontinues to be developed and modi ed (e.g.,Baddeley & Logie, 1999; Baddeley, 2000b).As depicted in Figure1.2, theBaddeley modelof working memory consists of twomodality-

speci c slave systems controlled by ahigher-level executive system. The idea of modality-speci c slave systems grew out of one of the ndings not handled gracefully bytheAtkinson-Shiffrinmodel,namely, the nd-ing of relatively little impairment in short-

termrecall followingcertain divided-attentiontasks (e.g., Brooks, 1968). This nding is eas-ily explained by a model that allows for sep-arate storage systems to handle information-processing demands in different modalities,but not by a model with a single modality-free short-term store.

The Phonological Loop

Baddeley and Hitch (1974) hypothesized thatworking memory is composed of three basiccomponents: the articulatory (phonological)

loop, the visuo-spatial sketchpad, and thecentral executive. According to their model,the phonological loop is responsible for

Phonological loop Subvocal rehearsal

Storage

Central executive Controls slave systems No storage capacity

Visuospatial sketchpad Imagery and spatial

rehearsal Storage

Figure 1.2 Baddeley s model of working mem-ory. The model consists of two modality-speci cslave systems that process and maintain informa-tion and are controlled by a central executive.


11/42



maintaining verbal and auditory information,while the visuo-spatial sketchpad maintainsvisual and spatial information. The centralexecutive coordinates these components and

allocates resources to them depending on thedemands of theenvironment. Thebasicmodelhas evolved over the years to include moredetailed descriptions of each component aswell as a series of subcomponents. Each of the slave systems works not only to store in-formation passively but also to maintain thatinformation actively.

The most widely studied component of Baddeley s model, the phonological loop, in-volves both the passive storage and the activerehearsal of verbal information. Four majorexperimental ndings are cited to support theexistence of this slave system; we describeeach of these in turn.

The errors people make when recallingwords suggest that working memory has aphonological rehearsal component. Baddeley(1966) found that immediate serial recall of a list of items was impaired when the itemssounded similar to one another. This phono-logical similarity effect is thought to occur be-cause the similar phonological codes used fortheitemsare dif cult to discriminate from oneanother during storage or retrieval. Phonolog-ical similarity would notmatter if theworkingmemory system did not have a phonologicalrehearsal component.

The nding that the length of words thatpeople are trying to remember affects imme-diate recall also supports the concept of ac-tive phonological rehearsal. A list of longerwords shows poorerrecall than a list of shorterwords when the number of words presented isequated (Baddeley, Thomson, & Buchanan,

1975). Surprisingly, the crucial variable inproducing the word-length effect is not thegreater number of syllables in the longerwords,but ratherthe actualspoken duration of the words. Baddeley et al. found that memoryspan is approximately equal to the number of

words that can be read aloud in 2 s (see alsoSchweickert & Boruff, 1986). This ndingsuggests that we subvocally rehearse words,and that the words can only be maintained for

about 2 s. This constraint on working mem-ory explains why digit spans vary across lan-guages; the memory span (the average num-ber of words that can be accurately recalledin correct order) is shorter in languages (e.g.,Welsh) in which words are long (relative toEnglish) and take more time to rehearse (Ellis& Hennelly, 1980). However, memory spansare longer in languages (e.g., Chinese) thathave shorter words, thus allowing for the re-hearsal of more words in the same time period(Hoosain & Salili, 1988).

Another nding that supports the ideaof the phonological loop is the unattendedspeech effect, which refers to the fact thatlistening to irrelevant auditory speech impairsrecall. That is, if people are hearing words ordigits and trying to remember them, havingother speech in the background reduces theamount that can be recalled. This effect sug-gests that the unattended speech obligatorilyclogs working memory capacity by engagingthe phonological loop and thereby prevent-ing rehearsal of the attended information thatpeople are trying to recall. Again, this effectwould not be expected unless there existed aphonological rehearsalprocesswith whichtheunattended speech interferes (Colle & Welsh,1976; Salame & Baddeley, 1987, 1989). Crit-ically, the unattended speech effect remainseven when the unattended speech is in a for-eign language or consists of nonsense sylla-bles. However, the effect does not occur withnonlinguistic materials, such as music.

A nal bit of evidence for the impor-

tance of phonological rehearsal comes fromexperiments using articulatory suppression.In this kind of experiment, subjects articu-late repeatedly some word or phrase such asthe or hiya while trying to remember in-formation presented visually. Recall is quite


12/42


12 Kinds of Memory

poor under these conditions when comparedwith conditions in which subjects are notengaged in articulatory suppression (e.g.,Richardson & Baddeley, 1975; Baddeley,

Lewis, & Vallar, 1984). This articulatorysuppression effect, as it is called, shows thatwhen conditions prevent access of studiedmaterial to the phonological loop, recall of that information suffers.

These four lines of evidence (amongothers) support the concept of the phonolog-ical loop as playing a central role in verbalworking memory. A comparable slave systemexists for nonverbal materials.

The Visuo-Spatial Sketchpad

The second slave system described byBaddeley and Hitch (1974) is the visuo-spatial sketchpad, which is hypothesized tomaintain visual and spatial information overshort durations and to permit manipulationof this information. The visuo-spatial sketch-pad has been further broken down into twosub-components: The visual cache is respon-sible for passive storage of pattern informa-tion, while the inner scribe retains sequencesof movement. However, most of the researchdescribed here is aimed more at supportingthe idea of a separate visual slave system thanat elucidating its speci c characteristics. Ingeneral, much less research has been directedat nonverbal working memory than at verbalworking memory.

Brooks (1968) was one of the rst to showthat maintenance of visuo-spatial informa-tion can be independent of storage and ma-nipulation of verbal information. He requiredpeople to remember either sentences or vi-sual patterns and then report on them either

by talking (verbal output) or pointing (spatialoutput). His study demonstrated that holdingsentences in memory leads to verbal reporttimes that are slower than spatial report times.However, when maintaining spatial patternsin memory, subjects found it quicker and

easierto reporton them verbally than bypoint-ing. That is, the outcome is not that eitherpointing or speaking is inherently more dif -cult, but that it depends on the type of in-

formation being held in memory. While peo-ple can simultaneously do activities that drawon both the phonological loop and the visuo-spatial sketchpad, they are impaired when at-tempting to do more than one activity thatdraws on the same capacity (the phonolog-ical loop or the visuo-spatial sketchpad) atthe same time. Analogous to the unattendedspeech effect, unattended visual informationdisrupts the visuo-spatial scratchpad. Logie(1986) also showed that extraneous visual in-formation selectively disrupted visual learn-ing, again supporting the idea that peoplecan rehearse verbal and visual information inseparate systems.

Neurological evidence supports the inde-pendence of visual and verbal slave sys-tems, as they draw on different brain areas.McCarthy and Warrington (1988, 1990) re-ported a single patient who suffered from anabnormality in his left temporal lobe. Thepatient demonstrated normal ability to com-prehend and maintain visual informationbut could not comprehend auditory/verbal in-formation. Using PET, Smith, Jonides, andKoeppe (1996) provided converging evidencethat verbal and spatial working memory tasksinvolve different areas of the brain. A letter-name recall task (a verbal task) activatedregions almost entirely in the left hemisphere,while remembering the positions of threedots (a spatial task) activated only right-hemisphere regions. In short, althoughtheevi-denceon operationof thevisuo-spatial sketch-pad is sparser than that on the phonological

loop, enough evidence exists to establish theindependence of the two slave systems.

The Central Executive

The central executive is the least speci edcomponent of Baddeley s working memory


13/42



model and borders on the most fundamentalproblems in all of cognition: consciousnessand control of behavior. The central executiveis hypothesized to have control over the allo-

cation of attention, the recruitment of slavesystems, and thepreparation anduseof strate-gies, as well as other features, such as acti-vating representations in long-term memory(Baddeley, 1996). However, the central exec-utive is thought to have no storagecapabilities.The central executive therefore holds most of the power in working memory but is basicallyrelegated to the role of a homunculus. If thecentral executive has control over the slavesystems, what controls the central executive?The concept of the central executive is similarto the Supervisory Attentional System, pro-posed by Norman and Shallice (1980). Thissystem oversees and controls certain behav-iors, overriding over-learned (automatic) be-haviors when necessary.

Reading Comprehension

Daneman and Carpenter (1980) argued thattypical working memory tasks (e.g., digitspan and word span) do not accurately re-ect the importance of working memory incomprehension, and that a more appropriatemeasure is reading span. In this task, sub-

jects read a series of sentences and then recallthe last word of each sentence. The num-ber of words that can be reported accuratelyis the measure of reading span. Althoughthe average number of words correctly re-called was relatively small ( just 2 to 5 wordsin the Daneman and Carpenter sample), thisspan correlated very highly with performanceon questions measuring comprehension of the prose passages. Daneman and Carpenter

argued that reading span measured workingmemory ability, which in uenced how wellmaterial could be understood. Just andCarpenter (1992) extended this argument andproposed that the important link betweenreading span and comprehension is working

memory capacity, for two main reasons. First,individuals with larger working memory ca-pacities can maintain more information aboutsyntactic constraints within a passage of text,

and can use this information to make judg-ments about the text. Second, a larger work-ing memory capacity facilitates maintenanceofmore than oneinterpretation incasesofsyn-tactic ambiguity. This relieves high-capacityindividualsof backtrackingin thetext in orderto reinterpret ambiguous sections. The criti-cal point is that a relatively simple measure of working memory (reading span) predicts animportant cognitive ability (reading compre-hension). Could measures of working mem-ory be tapping some fundamental cognitivecapacity?

Intelligence

In recent years, some researchers have pro-posed that working memory is closely re-lated to intelligence as measured by stan-dardized tests. One reason for this claim isthat performance on the verbal portion of theSAT is correlated fairly highly with Danemanand Carpenter s (1980) reading span mea-sure of working memory ( r = .59). Simi-larly, Engle, Tuholski, Laughlin, and Conway(1999) showed that performance on workingmemory tasks that measure the ability to holdand manipulate information (such as readingspan) was highly related to measures of uidintelligence (the ability to solve novel prob-lems andto adapt to new situations).The samerelationship does not show up on measuresthat simply tap storage aspects of short-termmemory. Engle et al. argued that the rela-tionship between working memory measuresand uid intelligence exists because both con-

structs tap the ability to keep a representationactive and to manipulate it, despite distrac-tions and interference.

Evidence from fMRI also shows a clearrelationship between performance on work-ing memory tasks and measures of uid


14/42


14 Kinds of Memory

intelligence. In one study (Prabhakaran,Smith, Desmond, Glover, & Gabrieli, 1997)participants solved three different kinds of uid reasoning tasks from Raven s Progres-

sive Matrices while in the fMRI scanner.Problems were classi ed as (a) visuo-spatial,which could be solved primarily by guralanalysis; (b) analytical, which required ab-stract reasoning in addition to gural analysis;and (c) perceptual-motor, which did not re-quire gural or analytic reasoning. The areasactivated by these tasks mapped onto thoseactivated by certain working memory tasks.Visuo-spatial reasoning tasks activated thesame areas that spatial working memory tasksoften activate (e.g., right middle frontalgyrus), and analytical tasks activated areasoften implicated in verbal working memoryand executive processes (e.g., left middlefrontal gyrus and left premotor cortex).

Summary

As with allpartsof this chapter, no universallyaccepted view of the working memory sys-tem exists. Baddeley s (1986) working mem-ory model has provided a fruitful theoreti-cal framework for empirical research in theeld; its parsimony and explanatory powercontinue to motivate research. Of course,several questions remain to be resolved inthe future (see, e.g., Miyake & Shah, 1999).Baddeley (2000b) has recently proposed theidea of a fourth component, the episodicbuffer. The episodic buffer is hypothesizedto integrate (bind) information from the slavesystems andlong-termmemory. Newresearchwill address this idea. Another new directionis in terms of neural instantiations of work-ing memory theory, to specify the neural sub-

strates through research with animals andhumans. For example, recent neural theoriesof the mechanisms underlying the centralexecutive have made some progress in avoid-ing the problem of its resemblance to ahomunculus by proposing speci c neural

mechanisms involved in the central executive(e.g., a dopamine-gating theory of control;Braver& Cohen, 2000). However, themecha-nisms of executive control remain largely un-

explored.

LONG-TERM MEMORY

The concept of long-term memory is some-thing of a grab-bag because many differenttypes of retention qualify as long-termmemory. There are names and faces we haveknown for years; text material studied forexams; our general knowledge or semanticmemoryfor facts; allkindsof motor skills that

would include talking, walking, driving cars,playing sports, and so on; and there are thesmells (e.g., popcorn) and touches (e.g., silk)that we can instantly recognize. And these ex-amplesare but a sample. Everything we retainthat did not occur in the last few momentscan be considered long-term memory. Herewe sample some of the primary concepts thathave been used to analyze this huge category,but there is certainly no agreed-upon taxon-omy for the various capacities that compriselong-term memory. We could have included

many more categories than the ones repre-sented here.

Episodic Memory

As mentioned earlier, episodic memory refersto memory for events (or episodes) and thecognitive and neural mechanisms involvedin remembering those events. In order to re-trieve such memories, the time and placeof occurrence of the events must be speci-ed (explicitly or implicitly) in the retrieval

query (Tulving, 1972). Examples of episodicmemory tasks include recalling the eventsexperienced last week, recalling the wordsfrom a list heard ten minutes ago, or re-calling dinner companions from the daybefore yesterday. Many of the laboratory


15/42


Long-Term Memory 15

techniques developed by psychologists overtheyears recallof stories, pictures, or wordslearned in the lab test primarily episodicmemory. As noted earlier, Tulving (1985a)

has argued that episodic memory re ects aspecial type of awareness autonoetic (self-knowing) awareness and that this abilitymay be unique to humans (Wheeler, 2000).Episodic memory is re ected by performanceon explicit memory tests (although someaspects of performance on these tests mayreect contributions from other memory sys-tems, as well). The concept of episodic mem-ory has changed over the years since Tulving(1972) rstproposed it, but it remains a centralorganizing concept in cognitive psychologyand cognitive neuroscience (for relatively re-cent treatments, see Tulving, 1993; Wheeler,2000).

One important sense in which episodicmemoryis usedis todescribe tasks such astheexamples just given. The following nine taskscan all be classi ed as episodic memory tasksbecause they require subjects to think back tothe time that the events in question occurred(Tulving, 1993). (The place is usually givenas in the lab where you are, but outside thelab the place may need to be speci ed, too.)

1. Free recall. People are exposed to a set of words or pictures and are asked to recallthem in any order after a brief delay.

2. Serial recall. People are given a series of digits, words, or pictures and are askedto recall them in the order of occurrence.Variations might include giving one itemfrom the series and asking for the item thatappeared before or after it.

3. Paired-associate recall. People learn pairs

of items that might be related (giraffe-lion)or unrelated (tightrope-pickpocket) andare later given one item (e.g., tightrope)and are asked to recall the other item. Thistask measures the formation of associa-tions.

4. Cued recall. People are given a seriesof words, pictures, or sentences and arethen given a cue (often something not pre-sented) and asked to recall a related event

from the series. If people study sentencessuch as The sh attacked the swimmer, the word shark might be given as a cue.Paired-associate learning is one type of cued recall task, but there are many varia-tions.

5. Recognition. These tests, as the nameimplies, require people to decide whetheror not they recognize an item as beingfrom thestudied set. In a typical laboratoryparadigm, subjects might study a list of

100 words (under various conditions) andthen be given a test with 200 words, half studied and half not studied. The task is toselect thestudied words. If thesubjects seethewordsone at a time, they judge whethereach one was studied and respond yes orno. This is called a free choice or yes/norecognition test. If the subjects are testedwith pairs of words, one old and one new,they have to pick the word that was stud-ied. This is called a forced choice recogni-tion test. Free and forced choice recogni-

tion tests resemble true/false and multiplechoice tests (respectively) used in educa-tional assessment. Another variation is thecontinuous recognition test, in which sub-

jects see a long stream of items and mustdecide for each item whether or not theyhave seen it earlier in the series.

6. Absolute frequency estimation tasks.Subjects study items such as words orpictures various numbers of times (say,18 times) and then later are presented theitem and have to judge how many timesthey studied it.

7. Relative recency judgments. Subjectsstudy items and then are given two andasked which one occurred earlier (or later)in the series.


16/42


16 Kinds of Memory

8. Source judgments. Information is pre-sented from a variety of sources (e.g.,spoken or written words, or by a male anda female voice if all items are spoken) and

during the later test subjects must iden-tify the source of each (spoken or written?Male or female?).

9. Metamemory judgments. Researchers canask subjects to give other kinds of rat-ings that are thought to re ect features of episodic memory. Con dence judgmentsask for ratings (on, e.g., a 7-point scale) of how con dent a person is about whetheran event occurred. People can also beasked, for items they recall or recognize,to judge whether they remember the mo-ment of occurrence of the item or rather

just know it was presented but cannot re-member the moment of actual occurrence(Tulving, 1985a). These kinds of remem-ber/knowjudgments have beenextensivelystudied (e.g., Gardiner & Richardson-Klavehn, 2000) because remember judg-ments are thought to re ect a pure mani-festation of episodic memory. Subjectscan also be asked to evaluate more speci -cally the sensory, emotional, and contex-

tual characteristics of their retrieved mem-ories (e.g., the Memory CharacteristicsQuestionnaire; Johnson, Foley, Suengas,& Raye, 1988).

All these tests (and others) tap some aspectof episodic memory by requiring subjects toretrieve information fromspeci c times inthepast. However, not all performances on theepisodic (or explicit) memory tests just listednecessarily re ect pure manifestations of episodic memory, as performance from rel-

atively automatic (Jacoby, 1991) or noetic(knowing) states of awareness (Tulving,1985a) might affect performance as well,especially on tests with strong retrieval cues.

Many variables have been shown to affectepisodic memory performance across a range

of tests. We consider just a few such variableshere variables manipulated during study orencoding. In general, meaningful processingof events produces better retention than does

processing that focuses on more super cialfeatures (Craik & Lockhart, 1972). In suchlevels-of-processing experiments, as they arecalled, subjects are exposed to words or othermaterial and are asked to make judgmentsabout them. For example, different groupsof subjects given the word RABBIT mightbe asked: Is it in upper-case letters? Does itrhyme with habit? Is it a type of animal? Theanswer to all three questions would be yes,but the rst question requires only a super-cial visual examination for an af rmativeanswer. The second question requires phone-mic (orphonologicalprocessing) to sound outthe word. The third question requires sub-

jects to think about the meaning of the word.Hundreds of experiments have shown that onlater tests of recall or recognition, meaning-ful processing produces retention superior tothat afforded by phonemic processing, whichin turn provides better recollection than pro-cessing of simple visual features such as typefont (e.g., Craik & Tulving, 1975). The exacttesting conditions for producing the effectsdo matter, as discussed later, but the levels-of-processing effect is ubiquitous in standardtests of recall and recognition (see Figure 1.3for an example). However, the interpretationof the effect is still under debate (Roediger &Gallo, in press).

Active involvement in learning, such asgenerating information rather than reading it,also promotes better retention (Jacoby, 1978;Slamecka & Graf, 1978). This generation ef-fect, as it is called, occurs even under condi-

tions in which the generation seems triviallyeasy. Jacoby (1978) had students either readword pairs (foot-shoe) or generate the sec-ond word from a word fragment (foot-s e).The fragments were easy (because the wordswere related), so the target word could almost


17/42


Long-Term Memory 17

0

.20

.40

.60

.80

1

Case Rhyme Category

Orienting task

P r o p o r

t i o n r e c o g n i z e

d

YesNo

Figure 1.3 The levels-of-processing (LOP) ef-fect. Mean proportion of words recognized as afunction of orienting task and type of response tothe question (yes or no). Adapted from Craik andTulving (1975, Experiment 9B).

always be generated easily. In the test, subjects were given the rst word and asked torespond with the paired word. When they hadgenerated the second word, they rememberedit much better than when they had read it,even though the generation process did notinvolve much effort. Slamecka and Graf pro-duced similar results in a somewhat differ-ent paradigm. Again, this generation effectcan disappear under certain conditions, butit has fairly wide generality, especially whenthe same subjects both read and generate in-formation (i.e., when the variable is manipu-latedwithin subjects; see Begg, Snider, Foley,& Goddard, 1989; McDaniel, Waddill, &Einstein, 1988; Slamecka & Katsaiti, 1987).

A third variable that reliably affects epi-sodic memory tasks is repetition. In general,

and not surprisingly, repeated items are bet-ter remembered than items presented onlyonce (therepetition effect; seeCrowder, 1976,Chapter 6). Less intuitively, however, thespacing of repetitions does matter. Massedrepetition refersto thesituationwhen an event

is studied twice in succession, whereas spacedrepetition refers to the case in which time andintervening items occur between repetitions.For tests of long-term retention, spaced repe-

tition almost always leads to better retentionthan does massed repetition; furthermore, upto some limit, the greater the lag or spacingbetween two presentations is, the better is re-tention (e.g., Melton, 1970; Dempster, 1988).This spacing or lag effect, as it is called,occurs on practically all tests and under mostconditions. Interestingly, one exception oc-curs when a test is given very quickly afterthe second of two presentations; then massedrepetition leads to better retention than doesspaced repetition (e.g., Balota, Duchek &Paullin, 1989).

Fourth, concrete materials generally pro-duce better retention on episodic memorytests than do abstract materials. For example,pictures are better recalled than words (theirnames); this is called the picture superiorityeffect (Paivio & Csapo, 1973; Paivio, Rogers,& Smythe, 1968). Words that refer to con-crete objects (umbrella, ngernail) are betterretained than are abstract words matched onsuch qualities as word length, part of speech,and frequency of occurrence in the language(Paivio, Yuille, & Rogers, 1969). The sameholds true forprose materials (Paivio & Begg,1971). To generalize, speakers and professorswho can explain an abstract theory (e.g., thekinetic theory of gases) by using a concreteanalogy or metaphor (molecules of gasbehav-ing like billiard balls on a pool table) oftencan make their subject matter not only eas-ier to understand but also more memorable.Known since the days of the ancient GreeksandRomans, mentalimageryis oneof theold-

est techniques for improving memory, and itrelies on the same principle: The mind gener-ally grasps and remembers concrete conceptsbetter than abstract ones.

Finally, distinctive items are generallybetter remembered on episodic memory tests


18/42


18 Kinds of Memory

(e.g., Hunt, 1995; Hunt & McDaniel, 1993).For example, a picture embedded in a list of words should be better remembered than thesame picture embedded in a seriesof pictures.

Distinctiveness has been used to explain su-perior memory for such items as bizarre sen-tences (McDaniel, Dunay, Lyman, & Kerwin,1988), unusual faces (Light, Kayra-Stuart, &Hollander, 1979), atypical category members(Schmidt, 1985), and words with unusualorthographies (Hunt & Elliot, 1980). Distinc-tiveness may increase attention to and pro-cessing of an item at study. Distinctive itemsalso yield excellent retrieval cues, since noother memories are associated with uniqueevents. Distinctiveness may underlie some of the effects just discussed. For example, thebetter memory associated with pictures andconcrete objects maybe due to thedistinctive-ness of their encoding. Similarly, deeper, se-manticprocessing of words leads to more dis-tinctive encoding and retrieval cues than doesmore shallow, phonological, or orthographicprocessing.

The various effects just discussed thelevels-of-processing effect, the generation ef-fect, thepicture superiority effect, the spacing(or lag)effect, and the distinctiveness effect represent just a sample of important variablesmanipulated during encoding or study thataffect episodic memory performance. How-ever, just because these variables are manipu-lated during learning does not mean that theyonly affect encoding of memories. Retrievalprocesses are critically important in the studyof episodic memory.

A common experience is forgetting somebit of information the name of an acquain-tance, where you left your keys and then

suddenly retrieving the information later.Sometimes the recovered memory seems tooccur spontaneously, but in other cases it isprompted by cues. Such recovered memoriesshow that forgetting is not necessarily dueto loss of information from memory (e.g.,

degraded memory traces) but rather that theinformation was available in memory, but notaccessible. Tulving andPearlstone (1966) rstformally distinguished between information

that is available in memory (is stored) andinformation that is accessible (is retrievableunder a particular set of conditions). Psy-chologists may wish for a perfect measureof what is stored in memory, but they willnever have one; all measures reveal the in-formation accessible under a particular setof conditions. The study of retrieval pro-cesses is therefore a key to understandingepisodic memory (Roediger & Guynn, 1996;Roediger, 2000; Tulving, 1974).

Even when we restrict our view to thestudy of episodic memory measures, we ndthat all tests do not reveal the same patternof results. For example, words that occurin the language with high frequency aretypically better recalled on a free recall testthan are words that occur with less frequency(e.g., Hall, 1954). So, we might concludethat high-frequency words simply producestronger or more durable memory traces thando low frequency words. However, this sim-ple idea is ruled out by recognition experi-ments. When high- and low-frequency wordsare presented and then retention is measuredby recognition, low-frequency words are bet-ter recognized than are high-frequency words(Kinsbourne & George, 1974; Balota &Neely, 1980). That different patterns of out-come are often obtained when different mem-ory tests are used is a fundamental fact thatmust be understood.

Two general ideas that have been for-warded to explain encoding/retrieval inter-actions are the encoding speci city princi-

ple (Tulving & Thomson, 1973) and theprinciple of transfer appropriate processing(Morris, Bransford, & Franks, 1977;Roediger, 1990). Both principles maintainthat retention is best when the conditionsof retrieval match (complement, overlap,


19/42


Long-Term Memory 19

recapitulate) the conditions of learning.Within terms of the encoding speci cityframework, the idea is that features of expe-rience are encoded and that retrieval cues are

effective to the extent that features extractedfrom thecue matchor complement thosein thememory trace (e.g., Flexser & Tulving,1978).The transfer appropriate processing principlestates that experiences during learning trans-fer to a test to the extent that the test re-quires appropriate cognitive operations topermit expression of what was learned. Testsmay be more or less appropriate to tap whatwas learned. These two principles are notmeant to be opposites; rather, they are some-what different statements of what is funda-mentallythe sameidea. Muchevidenceagreeswith these principles in episodic memory re-search (see Roediger & Guynn, 1996, for areview) and perhaps across all memory tests(Roediger & McDermott, 1993). We shall re-turn to these principles when we consider ev-idence for implicit memory tests.

The study of episodic memory is a hugetopic, and we can barely scratch the surfacein this section. Tulving s (1983) book, Ele-ments of Episodic Memory, is a good startingplace for further study of this critical topic.Much episodic memory research has beenlaboratory-based. A somewhat different tra-dition of research, but one that is also con-cerned with personal experiences, goes un-der the rubric of autobiographical memory, towhich we turn next.

Autobiographical Memory

Autobiographical memory refers to one s per-sonal history. Memories of the rst week of

college, of learning to drive, and of a friend sphone number are all autobiographical tosome extent. As these examples demonstrate,one s autobiographical knowledge consistsof many different types of knowledge: epi-sodes, procedures, and facts. Autobiograph-

ical memory is an amalgam of varieties of memories tied together by their importanceto one s sense of self and one s life history.In some sense, then, it does not represent a

distinct category of memory (like episodicand semantic memory are assumed to be), butrather a distinct research tradition within theeld. Still, consideration of autobiographicalmemory serves to bring out many of the coreconcepts of memory as considered by mostpeople who are not psychologists.

The problem of de ning autobiographi-cal memory has been discussed elsewherein depth (e.g., Conway, 1990). Brewer (1986)suggested distinguishing among personalmemories, autobiographical facts, andgenericpersonal memories. Personal memories, suchas memories of one s wedding, are describedas memories for speci c life events accom-panied by imagery. These would be episodicmemories. Autobiographical facts, such asmemories for phone numbers, are memoriesfor self-relevant facts and are unaccompaniedby imagery or spatio-temporal context (likesemantic memories, as de ned by Tulving,1972). Other knowledge, such as knowledgeof how to drive, consists of abstractions of events and is unaccompanied by speci cimages. These could be considered procedu-ral memories, but Brewer refers to them asgeneric personal memories. In this section,we focus on personal memories, with someattention to generic personal memories.

Vivid Memories of Life Events

Clearly, peoplehave accessto many vividper-sonal memories. People can retrieve detailed,emotional memories in response to a widevariety of retrieval cues, from across the life-

span and following long delays. What, then,leads to the unique, vivid personal memoriesthat characterize our sense of the past?

Historically, psychologists made surpris-ingly few attempts to capture autobiographi-cal memory. Galton (1879) conducted the


20/42


20 Kinds of Memory

rst attempt to study personal memories; heretrieved and dated personal memories in re-sponse to each of a set of 20 cue words. Otherearly research included Colegrave s (1899)

collection of people s memories of hearingthe news of Lincoln s assassination as well asFreud s clinical investigations into childhoodmemories (Freud, 1917/1982). However, ex-perimental psychologists conducted little re-search on autobiographical memory until the1970s. At that time, the pendulum swung infavor of more naturalistic research, partly inresponse to Neisser s famouscharge that If Xis an interesting or socially signi cant aspectof memory, then psychologists have hardlyever studied X (Neisser, 1978, p. 4). In ad-dition, the 1970s brought the publication of three important methods and ideas: Linton s(1975) diary study of her own memories forsix years of her life, the idea that surprisingevents imprint vivid ashbulb memories onthe brain (Brown & Kulik, 1977), and therediscovery of the Galton word-cueing tech-nique (Crovitz & Schiffman, 1974). Urged onby these results and the changing zeitgeist,experimental psychologists turned to thetricky problem of understanding how peoplecome to hold such vivid memories of theirown lives.

Results from Diary Studies

Beginning in 1972, Marigold Linton spent sixyears recording descriptions, dates, and rat-ings of 5,500 events from her own life. Shetested herself for recognition of a semirandomsample of events each month. While Lintonwas primarily interested in her ability to datethese personal events (e.g., Linton, 1975), shedid preliminary analyses of the characteris-

tics associated with remembered versus for-gotten events. She argued that rememberedevents were salient, emotional, and relativelydistinctive, and that there was some tendencyfor positive events to be better remembered(Linton, 1982).

Both White (1982) and Wagenaar (1986)followed up Linton s results, conducting diarystudies aimed more speci cally at remember-ing details of events rather than dates. Wage-

naar collected 2,400 events over a period of 6 years; he recorded the most salient eventeach day, and coded it with four cues: who,what, when, and where. He also rated thesalience (distinctiveness) of the event, as wellas its pleasantness and his emotional involve-ment. White recorded one event per day for ayear; he haphazardly selected both salient andnonsalient events. For each event, he recordeda description and chose adjective descriptors.He rated each event on a number of dimen-sions, including the degree to which he par-ticipatedin theevent,its importance tohim, itsfrequency, and its emotionality and physicalcharacteristics (e.g., sights, sounds, smells).Overall, theresults from thetwo studies corre-sponded well with Linton s observations:Recalled events were unique, and, at least inWagenaar s study, moreemotional. Bothstud-ies presented some evidence for the betterrecall of pleasant events.

Converging Results

While diary studies provide a rich sourceof autobiographical memories, such richnesscomes with methodological costs. Diary stud-ies typically involve only the experimenter assubject; the events to be remembered are notrandomly selected; and the very act of record-ing the events probably changes the way theyare encoded. Two different paradigms havebeen developed to deal with these problems.In one study, Thompson (1982) recruited 16undergraduates to participate in a diary study;the twist was that the participants recorded

events not only from their own lives but fromtheir roommates lives as well. All 32 parti-cipants later attempted to retrieve the recordedevents and used a 7-point scale to rate howwell they remembered them. Thecritical nd-ing was that memory did not differ between


21/42


Long-Term Memory 21

the recorders and their roommates, eventhough the recorders had selected and re-corded the events and had knowledge of theupcoming memory test.

In another clever study, Brewer (1988a)dealtwith the event-selection issue by recruit-ing subjects to carry pagers and record theirongoing events whenever the alarm sounded.Participants also rated their emotional state aswellas thefrequency, signi cance,and goal of the event. In the test, subjects were given oneof ve different types of retrieval cues (time,location, both time and location, thoughts, oractions) and were asked to recall the eventin question. Correctly recalled events wererated as being more associated with remem-beredsensory details,emotions,and thoughts.Consistent with the results of earlier diarystudies (Wagenaar, 1986; White, 1982), cor-rect recall was associated with exciting, in-frequent events occurring in atypical loca-tions. Similar results were also obtained inanother beeper study in which the memorytest involved recognition rather than cuedrecall (Brewer, 1988a, 1988b). As in labora-tory studies of episodic memory, distinctiveevents are well remembered.

We mention here only one of the manyother studies that support the idea that vividmemories tend to be for life events that wereunique, important, and emotional. Rubin andKozin (1984) collected data on vivid mem-ories using two paradigms. First, they askedparticipants to describe their three most vividmemories and then rate them on a number of scales (e.g.,nationalandpersonal importance,surprisingness, consequentiality, etc.). Over-whelmingly, participants provided memoriesof events such as personal injuries or romantic

episodes that were rated high in personal butnot in national importance (seealsoRobinson,1976). Second, participants retrieved autobio-graphicalmemoriesin response to 20 national(e.g., the night President Nixon resigned) andpersonal(e.g.,your 13thbirthday)cues. These

cues naturally varied in their ability to elicitvivid memories; vivid memories tended tobe associated with consequentiality, surprise,emotional change, and rehearsal (repeated

retrieval after the event).Thus, vivid personal memories tend to be

associated with exciting, emotional, unique,and even surprising life events. We turn nowto the question of what is forgotten.

Forgetting Life Events

Some researchers claim to nd less forgettingof life events than mightbe expectedgiven thehigh forgetting rates in many laboratory tasks,but of course such comparisons are fraughtwith dif culty. (No one really measures for-getting of the thousands of trivial activitiesthat occur in our lives.) Both Linton (1978)and Wagenaar (1986) reported forgetting lessthan 1% of recorded events after a year delay,but of coursetheyhad only recorded oneeventper day, and the act of recording would makeevents more memorable. Forgetting rates doincrease when events to be remembered arenot selected to be memorable; White (1982)forgot about 40% of his events, and the parti-cipants in Brewer s beeper studies failed evento recognize almost one third of events afterve months.

As in laboratory studies of words and pic-tures, forgetting ofautobiographicaleventsin-creases over time (Linton, 1978). It is criticalto note, however, that estimates of forgettingaredependenton thetypeof retrieval cueused.Emotion words are not good retrieval cues(e.g., Robinson, 1976), and temporal cues arenot as strong as content cues such as what, who, and where (Wagenaar, 1988;but seePillemer, Goldsmith, Panter, & White, 1988).

We have already noted that unique eventsare more likely to be remembered (e.g.,Wagenaar, 1986). When evaluating forgotten(non recognized) events, Linton (1982) clas-sied many as the failure to distinguish the target event from other similar events in


22/42


22 Kinds of Memory

memory. Once-salient events became lessmemorable as Linton experienced more andmoresimilarlife events,corresponding to howthe study of related material in laboratory ex-

periments increases interference effects (e.g.,Underwood, 1957) and leads to an abstractionof the gist of events or a schema for the events(e.g., Bartlett, 1932). Although participantsmay lose access to speci c event memories,they may retain more generic personal mem-ories covering a class of related life events(Brewer, 1986). Barsalou (1988) found thatstudents asked to recall the events from theirsummer vacations mostcommonly respondedwith summaries of events (e.g., I watcheda lot of TV ). Only 21% of responses wereclassi ed as corresponding to speci c events(e.g., we had a little picnic ).

As already noted, forgetting of life eventsincreases with the passage of time. Crovitzand Schiffman (1974) had college studentsrecall and date life events in response to aseries of cue words using Galton s technique.Brie y, subjects were asked to jot down afew words describing the rst memory thatcame to mind for each of 20 cue words. Theywere then asked to go back over the list of memories and date each as accurately as pos-sible. Crovitz and Schiffman counted up thenumbers of memories occurring in each of aseries of temporal categories. As expected,subjects recalled the most things from theirrecent pasts and the fewest from their dis-tant pasts (see also Rubin, 1982). However,a more complicated pattern emerges whenretention across the entire life span is exam-ined. First, thedecline is accelerated formem-ories from early childhood. Memories fromthe rst and second year of life are almost

nonexistent, and memories from the rstve years of life are infrequent (e.g., seeWetzler & Sweeney, 1986). Thisphenomenonis called childhood amnesia (Howe &Courage, 1993). Second, a different functionoccurs for older adults than for college stu-

dents.When older adultsrecall anddate mem-ories in response to word cues, they still re-port fewer memories from more distant timeperiods, and disproportionally few memories

from early childhood. However, they alsoshow what is called the reminiscence bump(see Figure 1.4); given the rest of the distribu-tion, a greater proportion of retrieved memo-ries aredated to theperiodsof late adolescenceand early adulthood than would be expected(e.g., Rubin & Schulkind, 1997). Numerousreasons have been suggested to account forthe so-called reminiscence bump, includinga preponderance of rsts occurring duringthe 20-something time period, the importanceof that time period for identity formation, and

0

100

200

300

400

110 11 20 21 30 41 503140

11011 2021304150 31 40

Age of memories in years

Approximate age of subjects at thetime of the event

SumFranklin & HoldingFitzgerald & Lawrence: NounsFitzgerald & Lawrence: AffectZola-Morgan, Cohen, & Squire

N u m

b e r o f m e m o r

i e s p e r

d e c a

d e

Figure 1.4 Distribution of autobiographicalmemories across the lifespan. In four studies, rep-resented by the lower four curves in the gure,50-year-old subjects remembered and dated life

events in response to cue words. The top curve col-lapses over studies and sums over the lower fourcurves. Subjects recalled a disproportionate num-berof eventsfrom adolescence andearly adulthood(reminiscence bump). SOURCE : Rubin et al., 1986.Reprinted with the permission of CambridgeUniversity Press.


23/42


Long-Term Memory 23

greater rehearsal frequencies for the types of events occurring during one s 20s. The exactreason for the bump remains uncertain.

Dating Autobiographical MemoriesOn what date didyou hear about theattemptedassassination of Ronald Reagan? On whatdate did you receive your acceptance letterfromthe college thatyou eventually attended?We suspect that our readers will be unlikely toanswer these questions quickly or accurately.Numerous studies have shown that peoplehave dif culty in dating their autobiograph-ical memories (for a review, see Friedman,1993), and that this dif culty increases withthe passage of time from the target event(Linton, 1975).

As introspection quickly reveals, however,it is not that autobiographical memory lacksall temporal information, which would belike a jumbled box of snapshots (Friedman,1993, p. 44). While the snapshots maylack explicit time-date stamps, we are quitecapable of relating, ordering, and organizingthe snapshots into a coherent story. Thesame subjects who cannot date a series of events within a month of their occurrence(3% correct; Brown, Rips, & Shevell, 1985)can determine the temporal ordering of theevents (rank order correlation of .88; Brownet al., 1985). There is an entire literature onhow peopleaccomplish this; dueto space con-straints, we describe only a few of the strate-gies peopleuseto reconstruct when events oc-curred. In general, people make use of whatlittle temporal information was encoded orig-inally. At least two types of temporal in-formation in memory appear relevant: thetemporal cycles that regularly occur in peo-

ple s lives, and temporal landmarks. First,natural temporal structures or cycles are en-coded that later guide memory; examples in-clude the academic calendar (Kurbat, Shevell,& Rips, 1998; Pillemer, Rhinehart, & White,1986) and the weekday-weekend cycle

(Huttenlocher, Hedges, & Prohaska, 1992).Second, peoplehavea bettersenseof thedatesof consequential landmark events, and thusboth public andprivate landmarkscan be used

to guide date reconstruction (e.g., Brown,Shevell, & Rips, 1986; Loftus & Marburger,1983; for a review, see Shum, 1998). Suchinformation about temporal and event bound-aries, combined with knowledge of some spe-cic dates, can be used to place a date on atarget event. However, people s reconstruc-ted dates tend to be too recent (Loftus &Marburger, 1983).

Other biases come into play when datingautobiographical memories; we mention justtwohere.Similar to theavailabilitybias foundin decision making (see Chapter 10), memo-ries for which people have more knowledgeare dated as more recent (the accessibility principle; Brown et al., 1985). People alsomay make rounding errors when they use in-appropriately precise standard temporal units(e.g., days, weeks, months; see Huttenlocher,Hedges, & Bradburn, 1990).

Inaccuracies in AutobiographicalMemories

Retrieval times for remembering autobio-graphical events tend to be slow and vari-able, suggesting that remembered events arereconstructed. Although diary studies havesuggested that people are good at recogniz-ing and remembering events that happened tothem, they do not show that people s mem-ories are accurate. A study by Barclay andWellman (1986) makes this point nicely. Intheir study, students took a recognition test onpreviously recorded life events that includedfour types of items: duplicates of original di-ary entries, foils thatchanged descriptive (sur-

face) details of the original events, foils thatchanged reactions to original events, and foilsthat did not correspond to recorded events.Participants were good at recognizing origi-nal diary entries (94% correct), but they alsoaccepted a large number of the foils. They


24/42


24 Kinds of Memory

incorrectly accepted50%of modi ed descrip-tions and 23% of novel events. These effectsincreased over a delay such that after a yearsubjects were accepting the majority of both

semantically related and unrelated foils.More naturalistic data also support the

idea that participants autobiographical mem-ories may not be accurate, even if they seemvivid and are con dently held. Perhaps mostfamous is the case of John Dean, a witnessin the Watergate hearings who appeared tohave an incredible memory for meetings withNixon at least until the appearance of thePresidential Transcripts, actual recordingsof Oval Of ce conversations. Neisser (1982)provided a fascinating comparison of Dean smemory with the transcripts, revealing thatsometimes Dean s memories re ected notthe truth but rather his fantasies and beliefsabout what shouldhavebeen. For example,on15 September 1972 John Dean assured Nixonthat nothing is going to come crashing downto our surprise (p. 146). When recalling thismeeting nine months later, Dean rememberedthat I also told him there was a long wayto go before this matter would end and that Icertainly could make no assurances that theday would not come when this matter wouldstart to unravel (p. 147).

Numerous laboratory experiments havesince demonstrated that people remembertheir personalhistories asconsistentwith whatthey believe should have happened, ratherthan with what did happen. Ross (1989) hasargued that people use their current statusas benchmarks and then reconstruct the pastbased on whether or not they think changesshould have occurred over time. For exam-ple, people believe that attitudes and political

beliefs remain consistent over time, so theyoften overestimate the consistency of the pastwith the present. In one study, subjects at-titudes towards toothbrushing were manipu-lated; people exposed to a pro-brushing mes-sage overestimated previous brushing reports,

whereas participants in an antibrushing con-dition underestimated their previous reports(Ross, McFarland, & Fletcher, 1981). Like-wise, people may mistakenly remember a

nonexistent change if one was expected. Par-ticipants who took a bogus study skills group(leading to no improvement) misrememberedtheir prior skills as being worse than theyactually were (Conway & Ross, 1984).

Even the most emotional, unique memo-ries are not immune from distortion. Whileit was initially argued that unexpected events(e.g., hearing of an assassination) triggered aspecial mechanism leading to capture of allevent details in a very accurate memory trace(Brown & Kulik, 1977), a spate of researchhas appeared arguing to the contrary. Theso-called ashbulb memories may be particu-larly vivid, rehearsed at high frequencies, andcon dently held, but they are prone to inac-curacies just as are memories of less emo-tional events.Early investigationsof ashbulbmemories were retrospective only, meaningthat they did not assess the consistency of participants stories over time (e.g., Yarmey& Bull, 1978). A different picture emergedfrom studies that involved the comparison of initial reports to later memories. For exam-ple, Neisser and Harsch (1993) compared ini-tial reports of having learned about the spaceshuttle Challengers explosion to those col-lected 32 to 34 months later. Even thoughtheir subjects report

Documents

Kinds of Memory