Haskins Laboratories Status Report on Speech Research1991, SR-105/106, 69-82

Modularity and Dissociations in Memory Systems*

Robert G. Crowdert

A cynical attitude toward progress in psychologyis that we simply move back and forth betweenwell-defined polarities, on pendulum swings,without really getting anywhere. Personally, Imuch prefer the model of a helix, in which we canrecognize steady progress in one direction whilenot denying oscillations of perspective on certainother issues. One unmistakable trend that hasbeen sweeping across the behavioral and cognitivesciences is the advancement of genetic explana­tions over environmental explanations. This iseasy to find in such diverse fields as linguistics,intelligence, personality, and mental health, tosay nothing of medicine. My grandparents werestaunch believers in genetic causation, too, but Ilike to think we now have better reasons for ourattitudes than they did.

A different trend back to earlier ideas isbecoming evident in the reappearance of facultypsychology in cognition, generally, and in theanalysis of memory, in particular. Publication ofFodor's Modularity of Mind (Fodor, 1983) onlycelebrates this newest cycle. The histories of suchtopics as localization of function in the brain andthe interpretation of intelligence put the trend inperspective. The impulse to subdivide memoryinto isolated subsystems should be appreciated asa manifestation of Fodor's views, with modularityitself manifesting a historical rhythm that governsour approach to many of the great issues.

TWO SORTS OF MODULARITY INMEMORY

In this first section of the paper, I will identifytwo very different interpretations of whatmodularity could mean in memory, one that maywell be generally accepted as conventional wisdom

I greatly appreciate thecommenta of my colleagues DonaldBroadbent, Fergus Craik, and Michael Watkins on an earlierversion of this paper, whose preparation was also supported byGrant BNS 8608344 from the National Science Foundation.

69

and the other controversial. These correspond, Ithink, to Fodor's "horizontal" and "vertical" modu­larity, and to Jackendoffs (1987) "representation­based modules" and "fundamental principles," butI will call them, less abstractly, coding modularityand process modularity. In later sections, I takethe dissociation of short-term storage (STS) andlong-term storage (LTS) as a case in point, raisingcaution about concluding in favor of separatememory systems. I will show that belief inHebbian consolidation is quite general in theanimal field, just as the STSILTS distinction hasbeen among students of human memory. Iconclude that the evidence is equally fragile inboth cases. Finally, the status of empirical disso­ciations will be discussed in general, including thecases of recognition and recall and of declarativeand procedural memory.Coding modularity

Any attitude toward information processingmust acknowledge that different kinds of informa­tion use different parts of the brain. Trivially, theauditory and visual systems engage distinct inputpathways and cortical projections, as do the so­called minor senses. The specialization of the lefthemisphere for language processing has beenmore tantalizing for the cognitive psychologist,perhaps only because it gives tangible reality tothe concept oflanguage.

As students of memory, many ofus have blandlyrecognized this left-right specialization and yet, atthe same time, we have held to a sort of Lashleyposition about learning and memory-that theydepend on having something between the ears, butit is all-purpose machinery (some say oatmeal) tobe found there. Such reliance on all-purposeequipment causes no one to lose sleep, providedlearning and memory are themselves consideredspecialized functions of the mind. This assumptionis dubious, however.

The growing identification of perception andinformation processing as being synonymous with

70 Crowder

learning, or memorization (Craik & Lockhart,1972; Crowder & Morton, 1969; Kolers, 1975) ledmany of us to change our attitudes in animportant way: Once memory is regarded as a by­product of information processing, the implicitconcept of all-purpose memory storage dissolves.To me, a comfortable language to describe thisattitude is to say that memory is not a storageprocess as such; it is simply the property ofinformation processing that extends in timeafterwards.

In much the same way, the neutrinos detectedat several international observatories starting onFebruary 24, 1987, were not a sort of time capsule,laid down by Supernova 1987A for our benefit.These neutrinos represent part of that originalevent itself, in a galaxy called the LargeMagellenic Cloud-an event that occurred 163,000years ago and a billion billion miles away fromhere. These neutrinos are the event, observed at aremote context. And so it is with at least somekinds of memories. The retention is just an aspectof the original episode itself, manifest at sometemporal remove. To me this attitude is theessence of proceduralism.

This attitude makes notions like memory stores,specialized to hold information over time, super­fluous. If the memory is in some sense, an aspectof the original event, then the memory resideswherever in the nervous system the event did. TheProustian mechanism of redintegration based onolfactory/gustatory cues must then reside in thebrain structures specialized for olfactory/gustatoryprocessing, and of course the connections of thesestructures with others.· Recognition memory for akaleidoscope slide, for a snowflake, for a musicaltimbre, or perhaps for a spatial layout, wouldactivate the participating units from their ownrespective information-processing episodes.Presumably, in these cases the processing, andtherefore the memory residue, would have beenscarcely verbalized. And when the informationprocessing of interest consist, all along, oflinguistic manipulations, the success or failure ofretrieval efforts will depend on whether theoriginally active systems get reactivated by thecues at hand. I consider this view of episodicmemory to be faithful to Tulving's (see Tulving,1983) encoding specificity principle. It is alsoprofoundly compatible with Hebb's (1949) ideas oncell assemblies and phase sequences as theagencies of memory. Related ideas in the animal­physiological approach to memory are found inSquire (1987, chapters 5 & 8).

Process modularitySo, coding modularity in memory is the

inevitable consequence of (a) brain specializationin cognition and (b) proceduralism in memory.What Fodor calls horizontal modularity, which Icall processing modularity, is quite another story.In this case, the separate subsystems cut acrossinformation formats (coding, or verticalmodularity) in favor of common processes. Anexample that now seems rather crude, and whichwas cited derisively by Fodor (1983, pages 13-14),would be to propose distinct short- and long-termmemory subsystems in which all sorts ofmemories would belong to one system up untilsome number of seconds had elapsed, and then toanother system afterwards. Another case ofprocess modularity is the distinction betweenprocedural and declarative memory, increasinglypopular in the 1980s. The same originalexperience, according to this last distinction, getsentered into distinct memory systems, oneprocedural, accessible to implicit memory tests,and intact in the amnesic, the other declarative,accessible to explicit probes of memory andseriously compromised in the amnesic.

What does it mean to propose distinct memorysystems? Perhaps an analogy with visualinformation processing is instructive: It is nowconventional to separate visual processingpathways, and their associated cortical relay sites,into two broad classes, some concerned with theidentification of objects and others with theirlocation in space (see, for example, Squire, 1987,p. 69). In addition, the same sort of distinctionmight cover sound localization and pitchperception, where the same stimuli might undergoprocessing in the two systems, for differentpurposes.

In the remainder of this paper, I shall try toexamine the evidence and arguments that oncemade us embrace the distinction between short­and long-term storage. The case was grounded insolid empirical dissociations. I conclude, as I havebefore (Crowder, 1982) that the case for twomemory systems is weak, even when thedistinction is broadened to cover the Hebbianconsolidation theory of memory.

On the basis of empirical dissociations betweenimplicit and explicit memory tests, some are nowurging us to make a distinction between anotherprocessing dichotomy, procedural and declarativememory (Squire, 1987, chapter 11; Tulving, 1987).These systems of memory would constituteprocessing modules of just the sort Fodor and

Modularity and Dissociations in Memory Systems 71

others have warned against. We can see in boththe dual-trace approach to memory storage(STS/LTS), and also in the more recentprocedural/declarative distinction, that empiricaldemonstrations of dissociation are central to thearguments. Why have the dissociations of the1960s come on hard times? What does this tell usabout the usefulness of empirical dissociations inproposing memory subsystems? I will argue thatTulving (1987) is too eager to make thisinferential leap from processing dissociations toseparate subsystems. Before going on, however, Ishould affirm that I attach the highest importancefor theory to dissociations between and amongmemory variables. I take that importance to bebeyond argument. The only question is whetherthey are sufficient to differentiate memorysystems.

THE CASE OF SHORT-TERM STORAGEI have reviewed elsewhere (Crowder, 1982) my

revisionist interpretations of the concept of short­term storage (primary memory), and so I need notgo into detail here. In a few words I will try topoint out how the evidence relied on empiricaldissociations. I pause to do that because the termdissociation was not popular in mainstreamcognition until the exciting work on amnesicsenriched our language with this particularlymedical connotation.

Review of evidenceRecency. The traditional association of the

recency effect in free recall with some transientmemory has now been discredited by the work ofBjork and Whitten (1974; Whitten & Bjork, 1972,see also Baddeley & Hitch, 1977; Greene, 1986;Tzeng, 1973; Watkins & Peynircioglu, 1983). Theyshowed that a post-list period of distraction,sufficient by itself to disrupt normal recency, doesnot eliminate recency if the items were spacedsufficiently from one another during presentation.If the distractor eliminates short-term memorythen this long-term recency effect must not becaused by short-term storage. Maybe the "normal"recency effect is not caused by it either.

I fully appreciate that this inference is notuniversal: For example Schneider and Detweiler(1987) preferred to associate the short-term andlong-term recency effects with differentunderlying mechanisms. Although I agree thatparsimony is not a serious virtue in the theory ofcognition, still, I think the burden of evidencemust lie with those who wish to impose acomplicated theoretical interpretation on a simple

data pattern. If two creatures both look likegoldfish and both act like goldfish, we should beprepared to accept that they are, in fact, not bothgoldfish, but the job of proving the distinctionbelongs to whomever believes in it.

Brown-Peterson forgetting. The second signatureof the short-term storage system was the slope ofthe Brown-Peterson forgetting curve. Again, Ihave detailed elsewhere (Crowder, 1976) how thisevidence was handled by the two-store theory andthe argument is quite standard. Whether one'stheory of loss in the Brown-Peterson task is anappeal to temporal decay, to displacement, or toperturbation of ordered rehearsal cycles, itoverpredicts forgetting, specifically on the firsttrial of the experiment. That is, these mechanismssay nothing about proactive inhibition. There isnothing in these theories about why decay,displacement, and perturbation, respectively donot occur or have only negligible effects on thefirst trial of an experimental session. Far less dothey anticipate that a change in the meaningcategory of the stimulus items would make decay,displacement, or perturbation suddenly becomeineffective; nor why allowing a long intertrialinterval would nullify these three mechanisms.

Broadly, then, a theory of Brown-Petersonforgetting has to be a theory of how proactiveinhibition works in this task (see Crowder &Greene, 1987a). Theories of proactive inhibitiontend to fall into two main categories: First, someassume that the essence of proactive inhibition isreally negative transfer-subsequent items neverget learned as well as the items presented on thevery first trial. Besides the studies reviewed inCrowder (1982), a recent application of this secondidea can be found in Schneider and Detweiler(1987) Second, some theories assume thatproactive inhibition represents a problem ofrecency discrimination among traces that arewidely available: We must always rememberwhich stored items were those that appeared ononly the most recent trial. This latter hypothesisis particularly well equipped to handle thefindings listed above, as Gorfein (1987) hasexplained and as I shall assert again below. Noticethat neither hypothesis for proactive inhibitionhas anything to do with special-purposemechanisms of short-term storage. Both are well­understood principles of memory, plain andsimple.

But recency and the Brown-Peterson slope wereoriginally considered evidence for STS mainlybecause of 4issociations. Recency recall andprerecency recall were sensitive to different

72 Crowder

variables. Recency depended on distraction butrecall from early and middle positions did not. Theasymptote depended on such factors as wordfrequency, rate of presentation, and age of thesubjects, but the recency effect did not

The analysis of Brown-Peterson forgettingdepended on the theoretical dissociation firstadvanced by Waugh and Norman (1965) betweenthe asymptote and slope of forgetting functions.But this assumption never had much in the way ofexperimental test because experimenters neverhad the patience to collect enough data points toallow separate estimates of these two performanceparameters. This effort would have requiredcurve-fitting based on tests of many differentretention intervals within the same experiment.

Alternative account using temporal codingBoth the recency effect and the Brown-Peterson

slope were dissociated from other performancemeasures in the same task (prerecency recall andthe Brown-Peterson asymptote, respectively).There is a good case that both were originallymisinterpreted. Furthermore, both results cannow be explained in essentially the same way.They may depend on retrieval by means of thetemporal context.

Recency has periodically been described interms of "temporal" coding (e.g., Murdock, 1960).Recently, a temporal-discriminability approach torecency has been suggested by Glenberg (1987;Glenberg & Swanson, 1986).1 The reasoning isthat events that just happened are moredistinctive with regard to their time of occurrencethan are events from the more distant past. Iftime of occurrence were to be a retrieval cue, therecency advantage would result. Bjork andWhitten (1974) evoked a principle of "temporalperspective" in which an evenly spaced list ofitems was approached, for purposes of recall, fromthe recency direction, much as a line of evenlyspaced telephone poles could be seen in spatialperspective from the near end. They assumed thatthe temporal discriminability of the recency itemswould be better than that of earlier items to theextent that the items were themselves spacedwidely in time. Recall might depend on a sort ofWeber-fraction, expressing (a) distance of theobserver from the most recent event, as aproportion of (b) the spacing between events. Theratio would be increased either by recallingpromptly after occurrence of the most recent item,or alternatively, by. increasing the interitemspacing. The rule seems to predict recall well(Glenberg, Bradley, Kraus, & Benzaglia, 1983;

Glenberg, Bradley, Stevenson, Kraus, Tkachuk,Gretz, Fish, & Turpin, 1980; Hitch, 1985). In ageneral sense, then, recency and temporal codingseem to have been joined in a coherent and quitegeneral principle of human memory. Such atheory is able to address with one voice the long­and short-term recency effects, unlike the dual­trace position. As such, it must be taken seriouslyas a possible explanation of why people rememberrecent experiences better than distant ones.

It is not hard to imagine that factors such asword frequency, intelligence, presentation rate,and age would have minimal effects on temporalcoding, and hence be ineffective in controllingrecency. These variables would, however, continueto exert their familiar positive effects on degree oflearning, and so affect performance wheretemporal coding has little impact, on the pre­recency segments of the serial position curve.Thus, the empirical dissociations on which thecase for two stores was based are equally plausiblefrom the alternative account.

One approach to the Brown-Peterson task, onethat dispenses with talk of separate short-termmemory systems, also stresses temporal coding.Bennett (1975) and Gorfein (1987) have advancedspecific versions of this approach. They assumethat subjects always face the problem ofdistinguishing the most recent memory items fromthose that occurred on earlier trials. On the firsttrial, these most recent memories are the onlyones eligible and so the "discrimination problem"disappears and there is no forgetting. After thefirst trial, the difficulty in making the recencydiscrimination depends on several factors. One isthe retention interval: Even if there are manycompeting items from earlier trials, if the mostrecent one has just been presented, with nointerpolated material, it is still in the foregroundof the temporal context. Lengthening theretention interval destroys the special foregroundprivilege of the most recent item, perhaps inconformity with Glenberg's (1987) "ratio rule" forrecency in general. This accounts qualitatively forwhy there is no forgetting on the first trial of aBrown-Peterson experiment and for whyforgetting increases with retention interval afterthe first trial. Gorfein's position also easilyaccounts for data on shifts in taxonomic categoryof the memory items and data showing sensitivityof Brown-Peterson forgetting to the intervalelapsing between trials.

I cannot guarantee that this sort of approachwill carry the day, in the end, for the Brown­Peterson task. But the account is very good now,

Modularity and Dissociations in Memory Systems 73

and so the burden of evidence must now fall onthose who believe in some mechanism of short­term storage, such as a limited-capacity buffer,decay, or the perturbation principle. The detailedfacts of forgetting in Brown-Peterson experimentsdo not demand any such extra principle, so theseadvocates must find other sorts of evidence tosupport them. Furthermore, these principles allmake one crucial prediction for which evidence isweak: They expect that there should be memoryloss on the very first trial of a Brown-Petersonexperiment. Nothing in perturbation, limitedcapacity, or buffer principles says that "all betsare oft" during the first experimental trial. Indeed,our lives are full of "first trial" situations, with thevarieties of experiences outside, as opposed toinside, the laboratory. So, the first trial is not anexceptional circumstance, to be set aside as afluke; it is ecologically central to our task astheorists. Let us examine first-trial forgetting fora moment:

Forgetting without PI? A survey of theliterature

A crucial question for theory is thereforewhether, after all, people forget on trial one of aBrown-Peterson experiment. I have been amazedthat this question is not stressed more given thequality and quantity of evidence available andgiven the importance for theory of the question.Baddeley and Scott (1971) deliberately soughtgood evidence on the point, explicitly appreciatingits importance. (Part of the problem withinvestigating this issue is that a subject is wastedafter his or her first few trials in an experiment, amatter of a few minutes, and must then be senthome.) Baddeley and Scott combined individualstudies by Baddeley and his associates in which asingle trial had been given, in order to amasssufficient observations to trace first-trialforgetting curves in detail. The total n was animpressive 922. The combined result showed areliable decline over the first 5 seconds, but thefunction was not monotonic, showingimprovement in performance between 6 and 9seconds. So this ambitious effort leaves somereaders unsatisfied and in need of additionalevidence.

Most of the many remaining experiments pub­lished on proactive inhibition set a fixed retentioninterval, often 20 seconds, and examined build-upand release as a function of other variables. AsBaddeley and Scott (1971) said, of the data per­mitting estimation of trial-one performance, someare furthermore uninterpretable owing to ceiling

effects on the first-trial-notably the famous andfrequently reprinted data of Keppel andUnderwood (1962) themselves.

Among the few adequate studies, Loess (1964)showed no loss on the first trial in twoindependent experiments. In the first,performance was near the ceiling at someintervals (scores of .958, .875, and .958,respectively, for intervals of 9, 18, and 27 seconds.In Loess's second experiment the first-trial scoreswere.79, .75, and .79 for intervals of 3, 9, and 18seconds. In both experiments there were 24observations per data point. Noyd (1965; seeFuchs & Melton, 1974) tested independent groupsof 27 subjects on either two-, three-, or five-worditems following delays of either 4, 8, or 24 s ofdigit reading. The two-word items were at ceiling.Performance on the three-word items was .927 ateach interval, perhaps also uncomfortably close tothe performance limit. Corresponding results forthe five-word items were .676, .630, and .638,respectively. The decline between 4 and 8 secondswith the five-word items corresponds to 23% of aword and was not reliable, and so in thisexperiment there was no loss over time in theabsence of proactive inhibition. Cofer andDavidson (1968) tested 18 subjects each at 3 or 18seconds of counting backwards on three-consonantsyllables and obtained perfect recall rates of .78and .83, respectively. This gain as a function ofretention interval was nonsignificant. Wright

. (1967) tested 240 people, 80 each at 3, 9, and 18seconds of counting backwards and obtainedcorrect recall proportions of .93, .90, and .96,respectively. These proportions are close to theperformance ceiling, but the number ofobservations per data point is commanding andwhat trend might be evident in the data is not adeclining one. Turvey, Brick, and Osborn (1970)tested subgroups of 40 subjects on three­consonant items at 5, 10, 15, 20, or 25sec,obtaining recall probabilities of .87, .85, .93, .93,and .95. Finally, an experiment by Gorfein andViviani (1980, data given in Gorfein, 1987) showedoff-ceiling performance that did not deterioratewith retention interval.

So, what should we conclude about forgetting onthe first trial of a Brown-Peterson experiment?Perhaps the most conservative conclusions is thatthere are some inconsistencies in the literaturebut that first-trial forgetting is the exceptionrather than the rule. At best, the role of anyadditional short-term storage or primary memorymechanism, over and above the principlesresponsible for proactive inhibition, is empirically

74 Crowder

slight in relation to observed performance lossesin the Brown-Peterson task. This task cantherefore no longer be exhibited as the showcasefor short-term storage. Only with faith andenormous effort can evidence for short-termstorage be coaxed from this task.New sources of evidence?

So much for the techniques once thought tobring short-term storage into the laboratory,recency and Brown-Peterson forgetting slopes.What new kinds of task are cited for the conceptnowadays? In a recent review of the evidence forbuffer storage, Schneider and Detweiler (1987)cited a third phenomenon, the span effect. Thereference is to the fact that immediate orderedrecall is limited to about a handful or two of items,depending in well-understood ways on what thememory items are. But the early hypothesis thatthis performance represents a fundamentalconstant in cognitive capacity (e.g., Miller, 1956)has not aged well. Watkins (1977) hasdemonstrated different forms of coding for theearly and late serial positions of an eight-itemmemory-span list. Others have dissociated ageeffects on the primacy and recency segments(Cohen & Sandberg, 1977; Huttenlocher & Burle,1976; Samuel, 1978). A more plausible candidatefor memory span may be that achieved withrunning-memory-span tests (Crowder, 1969;Pollack, Johnson, & Knaff, 1959) but its size ismore on the order of 2 or 3 items than 7 ± 2(Glanzer, 1972). Such a miniscule memory span isunattractive to those who would like to give short­term storage a role in other cognitive tasks. As weshall see, the working memory system of Baddeleyand Hitch (1974; Baddeley, 1983) does haveinformative things to say about memory span, butmore that it results from a highly specializedtrick-the articulatory loop-than that it is afundamental manifestation of an all-purposecapacity to buffer information.

To my knowledge, new candidates forepitomizing short-term memory within an explicittesting format have not received widespreadacceptance, the way the recency and distractortechniques were accepted in the past. Rather, thenewer orientation is to see short-term retention asa participant in integrated cognitive functioning,as working memory, to which attitude we nowtum.Working memory

One commendable development in the academicstudy of memory has been its integration withongoing cognitive tasks that are not, inthemselves, memory tests. This began with the

early computer models of associative memory(Anderson's FRAN, etc.) and has continued, forexample, with models of reading (Daneman &Carpenter, 1983) and reasoning (Case, Kurland, &Goldberg, 1982). The introduction of the termworking memory by Baddeley & Hitch (1974; seeBaddeley, 1986) has underlined this increasedecological perspective on memory. From relativelysimple beginnings the model of working memoryhas been elaborated steadily, and now stands asour most comprehensive theory of short-termmemory (Baddeley, 1983, 1986). The evidencefrom experiments that functionally distinguishamong components such as the articulatory loopand visuo-spatial scratchpad is convincing. Noticethat these distinctions are largely based on codingdifferences rather than storage-time differences;they exemplify coding modularity and not processmodularity.2

Working memory does not, for me, merely reori­ent the older concept of STS in a new dressing. Inits most fully articulated components, it is morelike a bag of tricks, each a modular coding formatin the sense discussed earlier in this paper. Thetight connection between the articulatory loop andspecific motor codes ·(Baddeley, Thomson, &Buchanan, 1975; Ellis & Hennelly, 1980) illus­trates this point admirably. So does the experi­ment of Reisberg, Rappaport, and O'Shaughnessy(1984) in which people were taught to use a sim­ple, thoughtless, finger-tapping "loop" to hold anadditional item or two for memory span, beyondwhat they could otherwise handle. This fonn of"buffer storage" is fundamentally procedural,highly code specific, and not at all representativeof a "memory system" to be distinguished fromLTS.

Necessity of buffer memory for cognition?The utility of talking about working memory is

that it stresses that people require-without anypossible argument-to remember small packagesof information briefly in order to succeed in anycomplex thinking task. But is this an argumentfor the necessity of memory per se in humancognition, or is it rather an argument for aseparate memory subsystem that is different instructure or function from "regular" memory,whatever that is? Here, I think we drift oft' fromissues on which evidence has been brought tobear. Evidence for (a distinct subsystem of) short­term storage is not at all the same as evidence thatpeople need to store things over the short term. Noquarrel is possible with the assertion that peopleneed memory for the recent past in order to

Modularity and Dissociations in Memory Systems 75

engage in language comprehension, written orspoken, in problem solving, in musical cognition,in playing bridge, or in almost any reasoning orlanguage activity. Perhaps the truest sense inwhich we need a working memory mechanism isthat we need a memory system that works.Distinguishing theoretically the operatingprinciples of (a) memory for recent events from (b)memory for remote events is a much stifferassignment, and one that I find relativelyneglected these days.

Why, then, does it seem so intuitively correctthat we attribute working or immediate memoryphenomena to a different system than passive,long-term memories? Part of the answer is thatsome proposed subsystems, such as thearticulatory loop, do indeed have integrity asseparate forms of information processing. Becausethese codes are especially useful in the short term,we are fooled into thinking they are distinctivebecause of their short-term properties, rather thanbecause of their coding format.

The assumption of process modularity­horizontally distinct memory subsystems-mustmake a much stronger claim, namely that short­and long-term storage are different within acommon coding format. Otherwise, if codingformat and short- versus long-term status areconfounded, there may be no need for the latterdistinction. What evidence is there that a phoneticmemory code used in digit span is different inkind than a phonetic memory code used in wordencoding experiments like those of Fisher andCraik (1977), where words were cued by phonetichints some 5 minutes after learning? I believethere is none.

Historically, the theoretical confounding ofcoding and process modularity probably arose inthe following way: The early workers stressedmainly the coding distinctions in the belief thatthey were subdividing short-term storage. In otherwords, the lesson of coding diversity was firstappreciated with short-term storage. As evidenceaccumulated that traditional memory-long-termstorage-was comparably subdivided into codingmodules, the implicit responsibility arose fordistinguishing which type of modularity-codingor process-was the more important. That is thecentral issue faced in this essay, and the answer Igive is obviously that coding modularity is wellsupported but process modularity is not.

In answer to our question of why primarymemory seems so distinctive, I can only appeal tolanguage so often quoted from William James(1890) on the nature of primary memory. With

virtually unchanged internal and external context,information from the immediate past seems still tobelong to the psychological present, in the sense ofTulving's (1983) "recollective experience." Memoryand forgetting do not demonstrably obey differentprinciples provided we equate for their codingformat. That is, what marks semantic coding inmemory is the same whether testing occurswithout appreciable delay (which may be rare) orquite a bit later. What marks phonetic coding islikewise continuous between short and longtesting delays, the same for olfactory coding andfor visual-imagery. These different forms of codingmay be more or less durable, perhaps because ofthe density of interference that occurs afterlearning, but different laws do not suddenly comeinto play with long retention intervals. Thedifference between long and short intervals is, inall cases, that the immediate test occurs with littlecontextual change and therefore the system oftime perception registers almost no change,whereas at longer intervals the change has beenconsiderable.

Thus, the same dissociations that led us todistinguish STS and LTS need not imply twoseparate memory stores, as we once thought.Coding (vertical) modularity proves to be the morevaluable principle than process (horizontal)modularity, in memory theory as Fodor (1983)claimed for cognition in general. In the nextsection, I examine a closely related aspect ofmemory theory, consolidation, in order to traceparallels with the dual-store ideas.

Where did STS come from? The case ofconsolidation

I recently had occasion to review some work inthe neuropsychogical theory of memory. I wasreminded that we cognitive psychologists shouldnot be possessive of STS, as if we had invented it,via such workers as Broadbent (1958), Brown(1958), and Peterson and Peterson (1959).Neuropsychologists believe that the concept islegitimately theirs, and indeed their case is a goodone: The history of thinking on consolidationtheory over recent decades is instructive when weconsider dissociations and the evidence for aseparable state of short-term storage. To mysurprise, I found the same bankruptcy in theoriginal concept of consolidation as I have in theconcept of STS. As we shall see, a new conceptionof consolidation has emerged.

Hebb and GerardHebb's (1949; see also Gerard, 1949)

neuropsychological theory was the landmark in

76 Crowder

the modern history of consolidation theory. Hisstatement had a wide influence in what we nowcall the neurosciences, as well as its influence inpsychology, anticipating as it did the populardual-trace (STS/LTS) distinction. According toHebb's version of the dual-trace hypothesis,experience is first recorded in the form of labile,reverberating, organized patterns of firing amongneural units, which, if allowed to remain activelong enough in concert, lead to the formation ofstructural changes in the nervous system, thebasis oflong-term memory. Notice how congenialHebb's formulation is with the famous lines ofWilliam James that characterized primarymemory as the persistence of (active) attentionand secondary memory as memory proper. We seeHebb's ideas honored even more in the moderntwo-process theory of Estes (1972; Lee & Estes,1981), about reverberatory cycles of orderedinformation giving way, with rehearsal, tostructural representations of serial order.

The continued debate on the idea of memoryconsolidation has centered on several researchareas. Of these, the two most prominent are (a)animal memory and (b) human clinical amnesia.As Weingartner (1984) said, the concept ofmemory consolidation, central as it was to manydevelopments in the modern psychobiology oflearning and memory "oo.was either ignored orrejected by investigators of cognitive processes inunimpaired human subjects" (p. 204).

Exceptions to Weingartner's observation arefew: Interest continues sporadically in sleep as afactor in retention (Ekstrand, 1972). No doubtseems to exist that retention is better following aninterval containing sleep than following one thatdoes not (see also Hockey, Davies, & Gray, 1972).The theoretical weight of this fact is not easilymeasured, though. Subjects in conditions callingfor sleep almost immediately after learning mustbe irresistibly drawn to rehearsal while they are"drifting off" to sleep. But Ekstrand (1972) citedevidence that memory is a reliable function ofwhether or not, during sleep, there has been rapideye movement (REM) activity. REM activityindicates the presence of dreaming during deepsleep. One outcome reported by Ekstrand is thatmemory performance is better following periodswithout REM activity than following REMepisodes. But other reports (Empson & Clarke,1970) have selectively deprived people of REMsleep with resulting damage to recall performance(see Jones, 1979, for discussion and more citationson this point.) Comparing sleep with and withoutREM activity is obviously better than comparing

sleep versus wakefulness, but the content of REMdreaming itself might provide interference (orsources of reinstatement of the learning activity).

Another use of the consolidation idea in researchon normal human subjects is due to Landauer(1974; 1977). Essentially, Landauer's experimentsshow that a given amount of high-similarityinterference, in short-term paired-associatelearning, has a larger effect if it comes right afteracquisition than if it comes after a delay.Landauer's (1974, 1977) results are consistentwith consolidation theory in that perseveratoryactivity specifically promoting consolidation of anitem is more likely to be broken apart by highlysimilar items than by an items sharing little withthe original learning. Relatively speaking only,low-similarity interference may be said to corre­spond to sleep. If so the earlier perseveratory ac­tivity is disrupted, the more its consolidation forthe long run should be compromised. But that re­sult is also consistent with other ideas aboutmemory and so it does not uniquely favor consoli­dation theory. For example, both the acid-baththeory of Posner and Konick (1966) and Estes'stimulus perturbation model (Estes, 1972) predictjust this result. At any rate, we do not need to re­sort to perseveration-consolidation to explain theespecially damaging effects of prompt, as opposedto delayed, interference. Now, the alternative ex­planations for this pattern of results may eventu­ally boil down to formal equivalence with consoli­dation theory once consolidation theory is workedout in detail. For the moment, the evidence aboutconsolidation from conventional experimentationis not nearly so powerful as the clinical evidence.

Wickelgren (1977, 1979) is the leading advocateof consolidation theory who is not primarilyidentified with research on amnesia. Wickelgren'streatment of consolidation (1977) distinguishesbetween two possibilities for what changes whenperseveration is allowed to run its course. Anhypothesis based on unitary strength wouldassume that the memory trace just gets strongerand stronger following learning. If this were true,reminiscence would be the rule rather than adelicate and elusive phenomenon. Alternatively,(a) the beneficial influence of consolidationfollowing learning and (b) the detrimentalinfluence of decay could be rationalized with someversion of the STSILTS distinction. However, itwould be preposterous to assign the STSmechanism a role lasting up to several decades,and the data on ECT do indeed provide evidencethat the consolidatory process extends overdecades (Squire, Slater, & Chace, 1975).

Modularity and Dissociations in Memory Systems 77

Wickelgren's own formulation of consolidationtheory, based on these rational considerations andon the study of amnesia, is called the decreasedfragility hypothesis. This idea is a single-trace­strength hypothesis about memory representa­tion, but it has a two-factor account of trace dy­namics over time. As traces age, they undergo de­cay. But at the same time, they "grow in resis­tance to decay;" they decrease in fragility. Decay,for Wickelgren, need not be simple disuse with thepassage of time. Resistance to decay (decreasedfragility) is assumed to grow indefinitely as thememory trace gets older and older, but with di­minishing returns, so that the first moments afterlearning are the most important ones.

Two historical "laws" of memory anticipateWickelgren's notion of trace fragility. Ribot (1881)deserves priority in this: His law of regressionstates that the vulnerability of memories todisruption lessens with their age. Ribot derivedthis generalization from a survey of amnesia casesproduced by head injury. As now, the evidence forthis important proposition then came from theclinic more than from the laboratory. Jost's secondlaw (1897; see McGeoch, 1942; Woodworth, 1938)says that if two associations are of equal strengthbut different age, the older one diminishes lesswith time. Jost had been interested in a relatedidea as manifested in experimental studies ofdistribution of practice.

Wickelgren showed (see summary inWickelgren, 1977) that the detailed analysis offorgetting curves was consistent with amathematical model including separate tracestrength and trace fragility expressions. However,the main evidence was clinical, from amnesiacases, including the Squire, Slater, & Chace

. (1975) study of ECT. This study, and data on headinjury seem to show that recent memory isdisrupted according to a temporal gradient by atraumatic event. The "lost memories" are verydefinitely "still there," however, because (a) beforean ECT session, subjects are fine at rememberingmaterial from the most recent decades (which isunavailable just after ECT), and (b) as Russelland Nathan (1946) pointed out for head injurycases, the memories spontaneously recover withthe passage of time since the injury (that is, theamnesic material comes back in reverse order toits age).

The modern version of consolidation theorytakes these facts seriously, and the rest of usshould quit ignoring them. That is one lesson ofthis section. The other lesson is that Hebbianconsolidation, the dual-trace theory, which was so

congenial to STSILTS distinctions, now has littleto recommend it as a general theory, for the sametwo reasons just examined, as well as others. Andwhat of research evidence from animals?

Evidence from animal experiments. One of themost persuasive reasons for experimentation onanimals, rather than humans, is that radical ma­nipulations comparable to the trauma of head in­jury can be produced at will with animals but not,ethically, with humans. Accordingly, the truly ex­perimental analysis of consolidation has belongedto the animal laboratory for many years. With an­imals, the investigator is free to introduce someelectrical or chemical agency that might block or­ganized perseveration in the brain and the result­ing consolidation. The time course of consolidationcan then be studied by manipulating the delay be­tween learning and the administration of such anamnesic agency.

Skipping over much history, we can focus on theexperiment in which rats are taught to step downfrom a platform to escape shock to their feet. Inexperimental conditions, electroconvulsive shock(ECS) is administered at various delays after asingle step-down trial. The question is whether itimpairs storage of the remembered footshock. Theresults of the Chorover and Schiller (1965)experiment and other subsequent ones did indeedshow amnesic effects of an ECS treatmentfollowing one-trial punishment training, but onlywhen the time between initial acquisition andECS was less than 10 seconds. A temporalgradient occurred in that ECS delays of more than10 seconds gave results like control conditionswithout ECS. Hilgard and Bower (1975) havewritten a thorough review of subsequentdevelopments in the animal ECS experiments. Fornow, the important point is that an experimentalamnesia, with a temporal gradient, can beproduced in rats. This observation supports theHebbian consolidation theory of memory but eventhis support has not been unequivocal in light offurther reports:

1. Experiments on different species (forexample, mice) or even different strains of rat, orexperiments using slightly different task detailshave turned in wildly variable time constants forconsolidation, even using the Chorover-Schillerexperimental logic (Chorover, 1976; McGaugh &Gold, 1974). At the least, consolidation timeestimates frustrate those who would have hopedfor a single "magic number" for such afundamental brain process as consolidation.

2. Memories impaired by ECS just afterlearning can be recovered spontaneously just by a

78 Crowder

lapse in time before testing, as Russell andNathan, (1946) documented long ago for headinjury cases in humans. Miller and Springer(1973) reviewed some of the evidence for recoveryin the animal studies. If the temporal gradient oflost memories shrinks with the passage of timebefore testing, we cannot say for sure thatmaterials "forgotten" in a test are necessarilyunavailable. It is always possible that waiting alittle longer would show recovery.

3. The effects of ECS can be reversed, or largelyreversed, if a "reminder" is given during theinterval before testing. In one situation, rats weregiven a simple footshock outside the training andtest apparatus. This apparently (R. Miller &Springer, 1973) reinstateds the trainingcontingency between stepping off the platform andfootshock. If the memory can be reinstated bysuch a reminder, it must have been laid downafter all, and not obliterated by disruption of theconsolidation process.

For these and other reasons, some workers havechosen to regard the induction of retrogradeamnesia in animals, by ECS, as compromisingprimarily the retrieval process and not theconsolidation process as originally thought (Gold& McGaugh, 1984; McGaugh & Gold, 1974; Miller& Marlin, 1984; Zeckmeister & Nyberg, 1982). R.Miller and Marlin (1984) are especially emphaticin rejecting the consolidation interpretation ofamnesia as it was originally intended by Mullerand Pilzecker, suggesting that: "To define allretrograde disruption of acquired information asconsolidation failure and then cite retrogradedisruption of acquired information as evidence ofconsolidation failure is circular and does not addto knowledge..." (pp. 86-87).Hebbian theory as an article of faith

However, these same authors (R. Miller &Marlin, 1984) endorse Hebbian memoryconsolidation almost on logical grounds: The veryfirst moments following an experience must, theysay, carry that experience in some form of activitytrace. Nobody would maintain that a structuralbrain change occurs instantaneously. But astructural brain change must happen sometime,for it is unreasonable to assume all memory is anactivity pattern even years later. Therefore, theysaid, there must be a transition between the twoforms of storage, as Hebb proposed. Miller andMarlin report some analyses that suggest suchconsolidation can take place as rapidly as within500 milliseconds following an experience.3 Oncethis reasoning is accepted, then so is theconsolidation hypothesis of memory. Whether

consolidation is a concept that explains anyobservable behavior is altogether anotherquestion, however. If the reversibility of ETCamnesia and the recovery of memories lost toretrograde amnesia argue against disruption ofconsolidation in humans, now we find that little orno evidence can be cited for this hypothesis inanimals, either.Consolidation reinterpreted

Renewed interest in consolidation, nowreinterpreted in terms of Ribot's Law, candetermine whether this is just a different way oftalking about familiar mechanisms (rehearsal,test events, and so on) or whether we have beenmissing out on important discoveries. More to thepoint for present purposes, the status ofconsolidation, as that idea was originallyunderstood, turns out to be quite like that ofshort-term storage: Workers seem disposed totrust the idea, almost on sheer faith, without clearevidence that can be cited as uniquely favoring it.

Taking stock of the arguments presented here,we note that the discussion of consolidation theoryhas pertained most directly to the bankruptcy ofthe concept of short-term storage. Beyond this,one may well wonder whether I am not proposingto substitute one form of processing modularity foranother, by revising what interpretation ofconsolidation is tenable in light of the evidence.This question, in turn, depends on whetherconsolidatory changes could one day beunderstood in terms of coding modularity. So faras I know, the issue has not been addressed. Withagencies such as ECT and memories of humanbeings for information from the recent and remotepast, changes in coding might well be important.With lower species, such reasoning appearsfanciful. Currently, we must restrict ourselves tothe more conservative point that the theory ofconsolidation does nothing to limit the generalityof my earlier conclusions about short-termstorage.

DISSOCIATING RECALL ANDRECOGNITION

It would have been a mistake, I believe I haveshown, to accept dissociations as grounds fordistinguishing short- and long-term storagesystems. Let us take still another example, briefly:If empirical dissociation were the criterion fordifferentiating memory systems, our field ofmemory might soon become a taxonomic scienceresembling botany. Then surely not only STS andLTS would have remained processing modules,but also recall and recognition would have. I have

Modularity and Dissociations in Memory Systems 79

organized the dissociating evidence for recall andrecognition elsewhere (Crowder, 1976), but thebest known factors are word frequency,intentional versus incidental learning, andsemantic organization. The effects of thesevariables on recall are either opposite to theireffects in recognition, a pattern called crosseddouble dissociation by Dunn and Kirsner (1987),or there are null effects in the case of one measureand positive effects in the case of the other (simpledissociation).

The recalVrecognition dissociations are impor­tant theoretically. They serve to falsify single-pro­cess theories like strength, which might claim thatrecognition is only a more sensitive measure of theunderlying trace than recall (a point first made byAnderson & Bower, 1972). By the same token, thedissociations advance the case for theories thatassume more than one underlying process in re­trieval (generate-recognition theories, for exam­ple, or Mandler's [1980] familiarity-plus-retrievaltheory). But nowadays recognition and recall areconsidered as belonging to declarative memory,tested explicitly, and requiring deliberate recollec­tion of the encoding episode. Evidence we usedformerly to distinguish them is still valid for justthat, but we would not maintain they representdifferent systems of memory, in the sense of pro­cess modularity.

Dunn and Kirsner (1987) have argued thatcrossed double dissociations can be expected evenfrom single-process models whenever there aretwo performance measures that are necessarilyreciprocal to each other. In the Anderson andBower (1972) theory of recognition and recall, twoseparate processes are postulated in order toaccount for the dissociations mentioned in the lastparagraph, pathway tagging and contextassociations. The dissociation data are indeedconsistent with this two-process theory. But theyare also consistent with the view that one of thesefactors is just what is left over after the other isused. Anything that then increases one of thefactors would necessarily have to reduce the other.If going from incidental to intentional learningincreases pathway tagging, for example, it wouldhave to reduce the importance of what remains,perhaps context associations, without affectingthis latter factor directly. Single dissociations,acording to Dunn and Kirsner, are even moreeasily dismissed. If a common single factor isincluded in two tasks, its effective range of actionmight be more suited to one than another. In thelimit, if this factor affects one task reliably, itcould fail to affect another because within the

context of this second task it is at floor or ceiling.For example, the experimental factor of retentioninterval seems to have different effective ranges ofaction on primed fragment completion than it doeson explicit recall and recognition (Sloman,Hayman, Ohta, Law, & Tulving, 1988).

In summary, then, the empirical dissociations ofthe past have not been sufficient for postulatingmultiple memory systems. I have includedinformation about short-term storage, aboutconsolidation theory, and about recalVrecognitiondifferences, to make this point. Now what of thedissociations popular today?

THEPROCEDURAUDECLARATIVEDISTINCTION

We read much evidence, these days, for multiplememory systems based on dissociations betweenthe relation of different tasks and differentindependent variables (Squire, 1987; Tulving,1987). The particularly dramatic reports are ofdissociations between explicit and implicitmemory in amnesics (Cohen & Squire, 1980; Graf& Schachter, 1985; see chapters in Squire &Butters, 1984). The same two systems have alsobeen dissociated in normal subjects as a functionof experimental operations (for example, Jacoby,1983). Nobody doubts that these dissociationshave exciting implications for theory; the onlyquestion is the legitimacy of concluding thatdissociations show the existence of differentmemory systems (Dunn & Kirsner, 1987; Jacoby,1983; Roediger, 1984). In this section I will arguethat such a conclusion is risky at best. I shall notattempt to review the growing evidence onimplicit and explicit tests of memory, however,because I think no such review is needed.

Dissociations within implicit and explicit mem­ory. Besides repeating the cautions I have alreadymentioned about interpreting empirical dissocia­tions, I shall pause here to cite a particularlysobering context for the proceduraVdeclarative (orperhaps implicit/explicit) subsystem distinction:Roediger and Blaxton (1987) have shown that weshould not be so quick to declare that adissociation of one memory measure from anotherheralds a distinction between systems of memory.They find that striking empirical dissociationsoccur when several testing procedures for implicitmemory are compared in response to the sameindependent variable. In the end, a dissociation isa particularly well-behaved and replicableinteraction (Tulving, 1987) in which one set oftasks responds to a manipulation and anotherdoes not, or vice versa. We are all used to getting

80 Crowder

theoretical mileage from interactions, of course,but additional arguments are needed to defend adistinction between memory systems in the senseofprocessing modularity.

Dunn and Kirsner (1987) maintain that dissoci­ations, and even double dissociations, are not logi­cally acceptable grounds for distinguishing mentalprocesses. They suggest the method of "reversedassociation" as a more defensible pattern on whichto distinguish process. A reversed association isessentially a nonmonotonic relationship betweentwo tasks across conditions and subjects.

At the least, the preceding review of two-processmemory theory should have made clear thatproposing distinct processing on the basis of evenorderly empirical dissociations is premature. Nowwe are seeing strong and fascinating empiricaldissociations between implicit and explicit tests ofmemory. As we seek to interpret these, perhapswe should be cautious in suggesting twounderlying memory systems, such as proceduraland declarative memory.

Two Systems or a modular component? Inrelation to procedural and declarative memory, wetalk as if two systems have been isolated, butreally there is only one element-the declarativeencoding of temporal context-that is separatefrom all the other diverse procedural formats.Each of the latter is "stored" in its processinglocus in the brain. Procedural memory is really an .umbrella term for processing residues of all sortsdepending on the mode of original informationprocessing. The structure of Schneider andDetweiler's (1987) recent "connectionist/control"model seems to recognize some of these attitudes:They propose processing modules assorted byprinciples of vertical modularity, that is, assortedby processing formats-coding modules-such asvisual, auditory, phonetic, semantic, lexical, andso on. Among these is a module representingcontext. In on-line processing, the connectionsbetween this context module and other processingcenters constitutes attention. Although I disagreewith Schneider and Detweiler on the necessity ofspecial assumptions for a system of buffer storage,the survival of contextual connections in theshort- and long-term will very plausibly commenton two of our concerns in this essay. First, theuninterrupted pace of contextual change couldprovide the constancy that gives "primarymemory" its Jamesian recollective experience ofbelonging to the conscious present. Second, thesource of classical amnesia may be understood byvirtue of a special vulnerability, or fragility in thesense of neo-consolidation theory, of the

contextual connections with other aspects of theprocessing system. The argument that amnesiacould not be a disturbance of consciousness is awell-understood proviso in the theory of amnesia-amnesics show no obvious signs of not beingaware of the world around them as they go abouttheir lives. It is the survival of traces connectingthis awareness with the processing systems usedin the past that may be disturbed.

This does not sound to me like a distinctionbetween procedural and declarative memory astwo systems, as such. It sounds like one elementof normal memory-the knowledge that thatprocessing occurred in that context-iscompromised. On the other hand, if we have amodel of memory that is vertically modular, eachof quite a few coding formats distinct from theothers, then loss of context connections, would bejust the loss of information of one among manyspecialized codes.

REFERENCESAnderson, J. R., & Bower, G. H. (1972). Recognition and retrieval

processes in free recall Psychologiall Reoiew, 79,97-123.Baddeley, A. D. (1983). Working memory. Philosophical

Tl'llnsactions of the Royal Socidy of London, H, 382, 311-324.Baddeley, A. D. (1986) Working rru:mory. New York: Oxford

University Press.Baddeley, A. Do, & Hitch, G. Jo (1974). Working memory. In G. H.

Bower (Ed.), The psychology of learning and motmuion (Vol. 8, pp.47-89). New York: Academic Pres&

Baddeley, A. D., &:: Hitch. G. J. (1977). Recency revisited. In S.Domic (Ed.), Attention and performance 6 (pp. 647-667).Hillsdale, NJ: LawrenO! Erlbaum Associates.

Baddeley, A. Do, & Scott, D. (1971). Short-term forgetting in theabsence of proactive inhibition. Quarterly Journal ofExperimentalPsychology, 23, 27~283.

Baddeley, A. D., Thomson, N., & Buchanan, M. (1975). Wordlength and the structure of short-term memory. Journal ofVerbldlaming and Ver/1Ql HehafJior, 14, 57~589.

Bennet, R. W. (1975). Proactive interference in short-termmemory: Fundamental forgetting processes. JOlll7llJl of Ver/1QlLearning and Ver/1Ql BehafJior, 14, 123-144.

Bjork, R. A., & Whitten, W. B. (1974). Recency-sensitive retrievalprocesses in long-term free recall. Cognitiw Psychology, 6,173­189.

Broadbent, D. E. (1958). Perception and commllllication. New York:Pergamon.

Brown, J. (1958). Some tests of the decay theory of immediatememory. Quarterly Journal of Experimt:lltal Psychology, 10, 12-21­

Case, R., Kurland, M. Do, & Goldberg, I. (1982). Operationalefficiency and the growth of short-term memory span. JOlITIIlJ1of Experimental Child Psychology, 33,386-404.

Chorover, S. L (1976). An experimental critique of the"consolidation studies" and an altemative "model systems"approach to the biophysiology of memory. In M. R.Rosenzweig & E. L Bennett (Eds.), Neul'lll rru:chanisms oflearningand rru:mory (pp. 561-582). Cambridge Mass. MIT Press.

Chorover, S. L, &:: Schiller, P. H. (1965). Short-term retrogradeamnesia in rats. JOllrnal of Comparatiw and PhysiologicalPsychology, 59, 73-78.

Modularity and Dissociations in Memory Systems 81

Cofer, C. N, &: Davidson, E. H. (1968). Proactive interference inSTM for consonant units of two sizes. JouT1lll1 of VtTbGl UilrningImd Verbill Be1umior, 7, 268-270.

Cohen, N. J., &: Squire, L. R. (1980). Preserved learning andretention of pattern-analysing skill in amnesia: Dissociation ofknowing how and knowing that. Science, 210, 2lY7-210.

Cohen, R. L., &: Sandberg, T. (1977). Relations betweenintelligence and short-term memory. Cognitive Psychology, 9,534-554.

Crail<, F. I. M., &: Lockhart, R. S. (1972). Levels of processing; aframework for memory research. Jounud of Vemu Uilrning lindVerilal Beh4vior, 11, 671-684.

Crowder, R. G. (1969). Behavioral strategies in immediate mem­ory. JouT1lll1 ofVerbill Uilrning lind Verblll Beh4vior, 8, 524-528.

Crowder, R. G. (1976).1'Tinciples of learning lind memory.Hillsdale,NJ: Lawrence Erlbaum Associates.

Crowder, R. G. (1982). The demise of short-term memory. ActlzPsychologiCA, SO, 291-323.

Crowder, R. G., &: Greene, R. L. (1987a). The context ofremembering. In D. S. Gorfein &: R. R. Hoffman <Eds.), Memorylind learning: The Ebbinghaus cenunnW conference. (pp. 191-199)Hillsdale NJ: Lawrence Erlbaum Associates.

Crowder, R. G. &: Greene, R. L. (1987b). On the remembrance oftimes past: The irregular technique. Jounud of ExperimentlllPsychology: GenerIIl , 116, 265-278.

Crowder, R. G., &: Morton, J. (1969). Precategorical acousticstorage (PAS). Percqtioft and Psychophysics, 5, 365-373.

Daneman, M., &: Carpenter, P. A. (1983). Individual differences inworking memory and reading. Jounud of Vemu Learning lindVmill Behavior, 19,4~.

Dunn, J. c., &: Kirsner, K. (1987). Discovering functionlillyindependent mentlll processes: The principle of reversed IISsociation.Unpublished manuscript.

Ekstrand, B. R. (1972). To sleep, perchance to dream (about whywe forget). In C. P. Duncan, L. Sechrest, &: A. W. Melton (Eels.),HUmll7l memory: Festschrift for Benton J. Underwood (pp. 59-82).New Yorlc Appleton Century Crofts.

Ellis, N. c., &: Hennelly, R. A. (1980). A bilingual word-lengtheffect: Implications for intelligence testing and the relative easeof mental calculation in Welsh and English. British JouT1lll1 ofPsychology, n, 43-52 .

Empson, J. A. c., &: Qarke, P. R. E. (1970). Rapid eye movementsand remembering. Nature, 227,287-288.

Estes, W. K., (1972). An associative basis for coding andorganization in memory. In A. W. Melton &: E. Martin (Eds.),Coding processes in huPflll1l memory. Washington, DC: Winston &:Sons.

Fisher, R. P., &: Craik, F. I. M. (1977). Interaction betweenencoding and retrieval operations in cued recall. JouT1lll1 ofExperimentlll Psychology: Humlln Uilrning and Memory, 3, 701­711.

Fodor, J. A. (1983) Modularity ofmind. Cambridge, MA: MIT Press.Fuchs, A. H., &: Melton, A. W. (1974). Effects of frequency of

presentation and stimulus length on retention in the Brown­Peterson paradigm. JouT1lll1 of Experimentlll Psychology, 103, 629­637.

Gerard, R. W. (1949). Physiology and psychiatry. AmeriCAn JounudofPsychiatry, 105, 161-173.

Glanzer, M. (1972). Storage mechanisms in recall. In G. H. Bower&: J. T. Spence (Eds.), The psychology of larning lind motiTJlltion(Volume 5, pp. 129-193). New York: Academic Press.

Glenberg, A. M. (1987). Temporal context and memory. In D. S.Gorfein &: R. R. Hoffman (Eds.) Memory and larning: TheEbbinghaus centennilll conference (pp. 173-19O).Hillsdale NJ:Lawrence Erlbaum Associates.

Glenberg, A. M., Bradley, M. M., Stevenson, J. A., Kraus, T. A.,Tkachuk, M. J., Gretz, A. L., Fish, J. F., &: Turpin, B. A. M.(1980). A two-process account of long-term serial positioneffects. Jounud of Experimentlll Psychology: HUmll7l UiJrning andMemory, 6,355-369.

Glenberg, A. M., &: Swanson, N. C. (1986). A temporaldistinctiveness theory of recency and modality effects. Journal ofExperimentlzl Psychology: Uilrning, Memory, and Cognition. 12,3-24.

Gold, P. E., &: McGaugh, J. L (1984). Endogenous processes inmemory consolidation. In H. Weingartner &: E. S. Parkers(Eds.), Memory consolidation: Psychobiology of cognition(pp. 65-84). Hillsdale NJ: Lawrence Erlbaum Associates.

Gorfein, D. S. (1987). STM &: discriminability. In D. S. Gorfein &:R. R. Hoffman (Eels.) Memory and laming: The EbbinghauscenunnW conference (pp. 153-172). Hillsdale, NJ: LawrenceErlbaum Associates.

Grai, P., &: Schacter, D. L (1985). Implicit and explicit memory fornew associations in normal and amnesic subjects. JouT1lll1 ofExperimentlzl Psychology: Learning, Memory, lind Cognition, 11,501-518.

Greene, R. L. (1986). Sources of recency effects in free recall.­PsychologiCAl Bulletin, 99, 221-228.

Hebb, D. O. (1949). Organimtion ofbeJulTJior. New York: Wiley.Hilgard, E. R., &: Bower, G. H. (1975). Theories of learning (4th ed.).

Englewood Cliffs, NJ: Prentice-Hall.Hitch, G. J. (1985) Short-term memory and information processing

in humans and animals: Towards an integrative framework. InL G. Nilsson &: T. Archer (Eds.) Perspectives on learni"g lindmemory. Hillsdale, NJ: Lawrence Erlbaum Associates..

Hockey, G. R. J., Davies, S., &: Gray, M. W. (1972). Forgetting as afunction of sleep at different times of day. QuIIrterly JOUTnllI ofExperimentlll Psychology, 24,~393.

Huttenlocher, J., &: Burke, D. (1976). Why does memory spanincrease with age? Cognitive Psychology, 8,1-31.

Jackendoff, R. (1987). The modularity of the computational mind.In R. Jackendoff, Consciousness and the computlztionlll mind(Chapter 12). Cambridge, Bradford/MIT.

Jacoby, L L (1983). Remembering the data: Analysing interactiveprocesses in reading. JouTnilI of Verblll Uilrning lind VerblllBeMvior, 22, 485-508.

James, W. (1890). Principles of psychology. New York: Holt.Jones, D. M. (1979). Stress and memory. In M. M. Gruneberg &: P.

E. Morris (Eds.), Applied problems in memory (pp. 185-214).London: Academic Press.

Jost, A. (1897). Die Associazionfestigkeit in ihrer Abhangigkeitvon der Verteilung der Weiderholungen. Zeitscllrift furPsychologie, 14,~72

Keppel, G., &: Underwood, B. J. (1962). Proactive inhibition in theshort-term retention of single items. Jounud of Verblll UilrningII7ld VerbIIl Be1umior, 1, 153-161.

Kolers, P. (1975). Specificity of operations in sentence recognition.Cognitive Psychology, 7,289-306.

Landauer, T. K. (1974). Consolidation in human memory:Retrograde amnestic effects of confusable items in paired.associate learning. JouT1III1 of VerbIIl Uilrning II7ld VerbIIl Be1Iat1ior,12,119-131.

Landauer, T. K. (1977). Remarks on the detection and analysis ofmemory deficits. In E. M. Birnbaum &: E. S. Parker (Eds.),Alcohol lind hUmll7l memory (pp. 23-42). Hillsdale, NJ: LawrenceErlbaum Associates.

Lee, Co L., &: Estes, W. K. (1981). Item and order information inshort-term memory: Evidence for multilevel perturbationprocess. JouT1III1 of Experimentlll Psychology: Human Uilrning lindMnnory, 7, 149-169.

82 Crowder

Loess, H. (1964). Proactive inhibition in short-term memory.Jounlll1 ofVeriHIllami1lg II1Id VeriHIl 8dIavior, 3, 362-368.

Mandler, G. (1980). Recognition: The judgment of previousooc:urrence. PsydwlogiCill RnJiew, 87,252-271.

McGaugh, J. L., ok Gold, P. E. (1974). Conceptual andneurobiological issues in studies of treatments affectingmemory storage. In G. H. Bower (Ed.), T1Ie psydwlogy of leamingII1Id motiwtion (Vot 8, pp. 233-264). New York: Academic Press.

McGeoch, J. A. (1942). 1M psycJwlogy of Ialrrum leami1lg.New York:Longmans Green ok Co.

Miller, G. A. (1956). The magical number seven plus or minustwo: Some limits on our capacity for processing information.PsycJwlogiaIl Rn1iew, 63,81-97.

Miller, R. R., ok Marlin, N. A. (1984). The physiology andsemantics of consolidation. In H. Weingartner ok E. S. Parkers<Eds.), Memory consolitllJtion: Psychobiology of cognition (pp.85-110). Hillsdale NJ: Lawrence Erlbaum Associates.

Miller, R. R., Springer, A. D. (1973). Amnesia, consolidation, andretrieval. Psychologiad Rmero, 80, 69-79.

Murdock, B. B., Jr. (1960). The distinctiveness of stimuli.PsychologiClll Rn1iew, 67, 16-31.

Noyd, D. E. (1965, June). ProIIctiw IUltl i1ItN stimulus inte~(% inshort-term memory for two-, three, lind fiW-7DOrtl stimuli. Paperpresented at meeting of the Western Psychological Association,Honolulu.

Peterson, L. R., &: Peterson, M. J. (1959). Short-term retention ofindividual items. JOUTIIIIl ofExpmmnttal PsycIwlogy 61, 12-21.

Pollack, 10, Johnson, I. B., ok Knaif, P. R. (1959). Running memoryspan.Jounlll1 ofExpmmnttal Psychology, 57,137-146.

Posner, M. I., ok Konick, A. W. (1966). On the role of interferencein short-term retention, Jounlll1 of Experimental Psychology, 72,221-231.

Reisberg, D., Rappaport, I., ok O'Shaughnessy, M. (1984). limits ofworking memory: The digit digit-span. Jounlll1 of ExperimentalPsychology: laming, Memory, III Cognitiml, 10,203-221.

Ribot. T. (1881). Us PrIII1Iulies de III memoirt. Paris: Germer Baillere.Roediger, H. L., III (1984). Does current evidence from

dissociation experiments favor the episodic/semanticdistinction? &1uJT1iorAlll1ld BrAm SciDras, 7, 252-254.

Roediger, H. L., III, ok Blaxton, T. A. (1987). Retrieval modesproduce dissociations in memory for surface information. In. D.S. Garfein ok R. R. Hoffman (Eds.), Memory lind ltlmting: T1reEbbmghAus cmtennw confermce (pp. 349-379). Hillsdale, NJ:Erlbaum.

Russell, W. R., ok Nathan, P. W. (1946). Traumatic amnesia. Brlli1l,69,2sa.300.

Salame, P., ok Baddeley, A. D. (1982). Disruption of short-termmemory by unattended speech: Implications for the structureof working memory. JoumIIl of Verblll Lellming lind VerlHzlBeJwmior,21,150-164.

Samuel, A. G. (1978). Organization versus retrieval factors in thedevelopment of digit span. Joumlll of Erperimentlll ChildPsyclwlogy, 26, 308-319.

Schneider, W., ok Detweiler, M. (1987). A connectionist/controlarchitecture for working memory. In G. H. Bower (Ed), T1repsyclwlogy of leIIming lind memory, Volume 21 (pp. 53-119). NewYork: Academic Press.

Sloman, S. A., Hayman, C. A. G., Ohta, N., Law, J., &r: Tulving, E.(1988). Forgetting in primed fragment completion. Joumlll ofExpmmnttal Psychology: lami1lg, Memory, lind Cognitimr.

Squire, L. R. (1987). Memory lind brllin. New York: OxfordUniversity Prerss.

Squire, L. R., ok Butters, N. (Eds.), (1984) Neuropsychology ofmemory. New York: Guilford Press.

Squire, L. R., Slater, P. c., ok Chace, P. M. (1975). Retrogradeamnesia: Temporal gradient in very long-termmemory following electroconvulsive therapy. Science, 187, 71­79.

Tulving, E. (1983). Elements ofepisodic memory. New York: OxfordUniversity Press.

Tulving, E. (1987). Introduction: Multiple memory systems andoonllCiOWl11ellll. Hurrum Neurobiology, 6, 67-80.

Turvey, M. T., Brick, P., ok Osborn., J. (1970). Proactiveinterference in short-term memory as a function of prior-itemretention interval. QuIIrterly JOIU7IIIl of ErperimmtIIl Psyclwlogy,22,142-147.

Tzeng. O. J. L., (1973). Positive recency eHect in delayed freerecall Jourrllll ~Vm.llami1lg lind Verl1IIl BeIunJior, 12, 436-439.

Watkins, M. J. (1977). The intricacy of memory span. Memory &Cogrritimr,5,529-534.

Watkins, M. J., &r: Peynirdoglu. Z. F. (1983). Three recency effectsat the same time.Joumlll of Verl1IIllami1lg II1Id VeriHIl Be1un1ior,22,375-384. .

Waugh, N. c., ok Norman, D. A. (1965), Primary memory.Psyc1wIogiCill Rmew, 72, 89-104.

Weingartner, H. (1984). Psychobiologic:a1 determinants of memoryfailures. In L. R. Squire and N. Butters (Eds.), Neurupsydwlogy ofmemory (pp. 205-212). New York: Guilford Press.

Whitten, W. B., ok Bjork, R. A. (1972). Test ewnts /IS lermIi1Ig triIIls:T1re importtmce of being impnftct. Paper presented at theMidwestern Mathematical Psychological meeting,Woomington, IN.

Wickelgren, W. A. (1977). laming lind memory. Englewood Cliffs,NJ: Prentic~Hall.

Wickelgren, W. A. (1979). Chunking and consolidation: Atheoretical synthesis of semantic networks, configuring inoonditioning, 5-R versus cognitive learning, normal forgetting,thE amnesic syndrome, and the hippocampal arousal system.PsychologicIIl Rmero, 86, 44-60.

Woodworth, R. S. (1938). ErperimerttIIl psychology. New York: Holt.Wright, J. H. (1967). Effects of formal interitem similarity and

length of retention interval on proactive inhibition inshort-term memory.JOIlnlll1 of Erperimerttal PsycJrology,75,366­395.

Zec:hmeister, E. B., ok Nyberg, S. E. (1982). Human memory: Anintrotluctimr to restllrcIi IUltl theory. Monterey, CA: Brooks/Cole.

FOOTNOTES·In H. L. Roediger, m &r: F. I. M. Craik, (Eds.), Vllrieties of memory

lind COIISCiousness (pp. 274-294). Hillsdale, NJ: Lawrence ErlbaumAssociates (1989).

tyale University, Department of Psychology.'Crowder and Greene (1987a, b) have commented on a possible

error in conceptualizing the motllllity effect in this way, but myconcern here is with the general idea that recency results fromtemporal discriminability of items at the end of a series.

21 say "largely'" with cause: Salame and Baddeley (1982) havesubdivided the articulatory loop into storage and processingcomponents, which are examples of processing modularitywithin a form of coding.

30thers (Hilgard ok Bower, 1975, pp. 508-516) have reviewedessentially the same literature and reached the conclusion thatoonsolidation is the concept.