storage and retrieval processes in long …cogs.indiana.edu/festschriftforrichshiffrin/pubs/1969...

Psychological Review1969, Vol. 76, No. 2, 179-193

STORAGE AND RETRIEVAL PROCESSESIN LONG-TERM MEMORY1

R. M. SHIFFRIN 2 AND R. C. ATKINSON

Stanford University

A theory of human memory is described in which a distinction is madebetween three memory stores: the sensory register, the short-term store, andthe long-term store. Primary emphasis is given to the processes by whichinformation is stored in and retrieved from the long-term store, a store whichis considered to be a permanent repository for information. Forgetting andrelated phenomena are attributed to a failure of the retrieval process, inwhich the search through some memory area becomes less efficient as newinformation is placed in it. Storage and retrieval in the long-term store areconceived of as parallel processes, one mirroring the other, and each isdivided into three stages for conceptual clarity. The memory trace is viewedas an ensemble of information stored in some memory location, the location ofstorage determined largely by the components of the ensemble itself. Theability of the system to cope with diverse phenomena is demonstrated by aconsideration of a number of selected experimental paradigms.

Theories of human long-term memoryhave given primary emphasis either to theorganization of that memory, in terms of thedimensions of storage and the associationsbetween the stored information (e.g., Cofer,1965; Deese, 1966; Mandler, 1968; Osgood,1963), or to the characteristics of temporaldecay from that memory as in the interfer-ence theories (e.g., Keppel, 1968; Melton,1963; Postman, 1961; Underwood, 1957).The processes by which information is storedin, and retrieved from, long-term memoryhave been relatively neglected. An exampleof this type of process is the memory searchduring retrieval; during the search, a suc-cession of memory codes is examined, eachexamination followed by a decision processin which the search is either terminated orcontinued, and in which information recov-ered is either accepted as that desired and

1 Preparation of this article was supported by agrant from the National Aeronautics and SpaceAdministration, and is an outgrowth of ideas firstdeveloped in two earlier reports (Atkinson & Shif-frin, 1968b; Shiffrin, 1968). This paper, in combi-nation with papers by Atkinson and Shiffrin (1965,1968a), represents an attempt to formulate a gen-eral schema within which to analyze memory andlearning.

2 Now at Indiana University. Requests for re-prints should be sent to Richard C. Atkinson,Institute for Mathematical Studies in the SocialSciences, Ventura Hall, Stanford University, Stan-ford, California 94305.

output, or rejected. It is the intention ofthis paper to elaborate the memory input andoutput processes. As will be indicated later,our view of storage and retrieval eliminatesthe necessity for assuming decay of informa-tion from long-term memory. It will beassumed that long-term memory is perma-nent; decrements in performance over timeare ascribed to an increasingly ineffectivesearch of the stored information.

We begin by describing the overall concep-tion of the memory system. The systemfollows that described in Atkinson and Shif-frin (1965, 1968a), and is similar to thoseproposed by 'Feigenbaum (1966) and Nor-man (1968). The major components ofthe system are diagrammed in Figure 1: thesensory register, the short-term store (STS)and the long-term store (LTS). The solidarrows in the diagram represent directionsin which information is transferred fromone part of the system to another. Note thattransfer is not meant to imply the removalof information from one store and the placingof it in the next; rather, transfer is an opera-tion in which information in one store is"copied" into the next without affecting itsstatus in the original store. It should beemphasized that our hypotheses about thevarious memory stores do not require anyassumptions regarding the physiological locusof these stores; the system is equally con-

179

180 R. M. SHIFFRIN AND R. C. ATKINSON

Stimulus analyzer programs

Activate rehearsal mechanismModify information flow from SR ICode and transfer information frorInitiate or modify search of LTS

Initiate response generator

. STSSTS li

FIG. 1. A flow chart of the memory system. (Solid lines indicate paths of information trans-fer. Dashed lines indicate connections which permit comparison of information arrays residingin different parts of the system; they also indicate paths along which control signals may besent which activate information transfer, rehearsal mechanisms, etc.)

sistent with the view that the stores areseparate physiological structures as with theview that the short-term store is simply atemporary activation of information perma-nently stored in the long-term store. Thecontrol processes listed in Figure 1 are asample of those which the subject (S) cancall into play at his discretion, depending uponsuch factors as the task and the instructions.Control processes govern informational flow,rehearsal, memory search, output of re-sponses, and so forth.

The sensory register is a very short-livedmemory store which temporarily holds in-coming sensory information while it is beinginitially processed and transferred to theshort-term store. In the visual modality,for example, information will decay from thesensory register in a period of several hun-dred milliseconds (Sperling, 1960). Infor-mation in the short-term store, if not attendedto by S, will decay and be lost in a period ofabout 30 seconds or less, but control proc-esses such as rehearsal can maintain informa-tion in STS for as long as S desires (thebuffer process in Figure 1 is one highlyorganized rehearsal scheme). While infor-

mation resides in STS, portions of it aretransferred to LTS. The long-term store isassumed to be a permanent repository ofinformation; we realize that factors such astraumatic brain damage, lesions, and de-terioration with extreme age must lead tomemory loss, but such effects should benegligible in the types of experiments con-sidered in this paper. Thus it is hypothe-sized that information, once stored in LTS,is never thereafter destroyed or eliminated.Nevertheless, the ability to retrieve informa-tion from LTS varies considerably with timeand interfering material.

The short-term store serves a number ofuseful functions. On the one hand it de-couples the memory system from the externalenvironment and relieves the system fromthe responsibility of moment-to-moment at-tention to environmental changes. On theother hand, STS provides a working memoryin which manipulations of information maytake place on a temporary basis. BecauseSTS is a memory store in which informationcan be maintained if desired, it is often usedas the primary memory device in certaintypes of tasks; in these tasks the information

PROCESSES IN LONG-TERM MEMORY 181

presented for retention is maintained in STSuntil the moment of test and then emitted.Tasks in which STS is utilized for this pur-pose, and the mechanisms and control proc-esses that may come into play, have beenexamined extensively in Atkinson and Shif-frin (1968a). In this report we are pri-marily interested in STS as a temporarystore in which information is manipulated forthe purposes of storage and retrieval fromLTS, rather than as a store in which infor-mation is maintained until test. In the re-mainder of this paper, discussion is limitedto that component of memory performancewhich involves LTS retrieval, and the com-ponents arising from STS and the sensoryregister will not be considered.

LONG-TERM STORE

In describing the structure of LTS, ananalogy with computer memories is helpful.The usual computer memory is "locationaddressable"; if the system is given a certainlocation it will return with the contents ofthat location. When given the contents ofa word (a "word" refers to a single com-puter memory location), such a system mustbe programmed to examine each location inturn in order to find the possible locationsof these contents in the memory. It seemsuntenable that an exhaustive serial searchis made of all of LTS whenever retrievalis desired. An alternative type of memorymay be termed "content-addressable"; if thesystem is given the contents of a word itwill return with the locations in memorycontaining those contents. One way in whichsuch a memory may be constructed utilizesa parallel search through all memory loca-tions; the system then returns with thelocations of all matches. If this view isadopted, however, an additional process isneeded to select the desired location fromamong the many returned by the parallelsearch. Thus, if we feed the system the word"red," it would not be useful for the systemto return with all references or locations of"red"; there are simply too many and theoriginal retrieval problem would not be sig-nificantly reduced in scale. There is, how-ever, an alternative method for forming acontent-addressable memory; in this method,

the contents to be located themselves containthe information necessary to specify the stor-age location(s). This can occur if the in-formation is originally stored in locationsspecified by some master plan dependentupon the contents of the information. Sucha system will be termed "self-addressing."A self-addressing memory may be comparedwith a library shelving system which is basedupon the contents of the books. For example,a book on "caulking methods used for 12thcentury Egyptian rivercraft" will be placedin a specific library location (in the Egyptianroom, etc.). If a user desires this book, itmay be located by following the same shelv-ing plan used to store it in the first place.We propose that LTS is to a large degreejust such a self-addressable memory. Anensemble of information presented to thememory system will define a number ofmemory areas in which that information islikely to be stored; the memory search willtherefore have certain natural starting points.The system is assumed to be only partiallyself-addressing in that the degree to whichthe storage locations are specified will varyfrom one ensemble to the next and one mo-ment to the next, in much the way as pro-posed in stimulus sampling theory (Estes,1959). Thus it may be necessary to embarkupon a memory search within the specifiedlocations, a search which may proceed seri-ally from one location to the next. Thisconception of LTS leads to a number of pre-dictions. For example, a recognition test ofmemory will not proceed via exhaustive scan-ning of all stored codes, nor will a recognitiontest eliminate in all cases the necessity for anLTS search. If information is presented andS must indicate whether this information hasbeen presented previously, then the likelystorage location (s) is queried. To the de-gree that the information has highly salientcharacteristics which precisely identify thestorage location, the extent of the LTS searchwill be reduced. Thus, for items with highlysalient characteristics, 6" should be able toidentify quickly and accurately whether theitem was presented previously, and the iden-tification might not require a memory searchwhich interrogates more than a single storagelocation. The less well-specified the storage


location, the greater the memory searchneeded to make an accurate recognition re-sponse. This view is in many respects simi-lar to that proposed by Martin (1967), inwhich he suggests that certain stimuli tend togive rise to identical encoding responses fromone instance to the next, that is, on bothstudy and test trials. From the presentviewpoint "identical encoding responses" istaken to be equivalent to identical storagelocations.

Although it is assumed that informationensembles are sorted into LTS locations ac-cording to a master plan, it is beyond thescope of this paper to attempt to outline thisorganizational structure; other workers havebeen dealing with this problem (e.g., Han-dler, 1968; Pollio, 1966). Undoubtedly thisorganization is highly complex, but the formof the organization will not be crucial to ourdiscussions of input and output mechanisms.There is one dimension of organization thatmust be mentioned here, however; this is the"temporal" dimension. We are assumingthat this dimension is like the other organi-zational dimensions in that information maybe stored along it, and that retrieval may bebased upon it. In the remaining sections ofthis paper, the temporal dimension will oftenbe singled out for special mention, not be-cause it differs in substance from otherorganizational dimensions, but because it isvirtually impossible to eliminate it as a sys-tematic variable in most memory experi-ments.

The term "location" is used in relation tothe organizational schema; an LTS locationis defined by the place in the organizationalstructure occupied by an information en-semble. The location will be defined in termsof the modality of the information (e.g.,visual versus auditory), the level of analysis,(e.g., spelling versus syntactic structure),and all other dimensions of organization thatmay be relevant. Two locations will be saidto be "close" if the information in them tendsto be retrieved together. In particular, weshall refer to a code, or an image, as anensemble of information that is closely relatedand very likely to be retrieved together. Wedo not wish to imply that there is some uni-tary atom of storage called a code or an

image. The information making up a codein one task may be considered to be severalcodes in a different task. 'For example, anentire sentence may be considered a code ifwe are comparing the meaning of that sen-tence with others; however, the same sen-tence might be considered to be made up ofa series of codes if we are comparing it withsentences of the same meaning but differentgrammatical form. Nevertheless, for mosttasks the concept of a code or image as rep-resenting a cohesive array of information ina single storage location proves useful. Thisview of the memory trace is consistent withthe analysis given by Bower (1967). In hismodel, a trace, or image, is represented by avector of attributes which serves to identifyboth the information contained in the traceand the location of the trace in LTS. Thevalues of the attributes in the vector serve toindicate the position of the trace on the vari-ous organizational dimensions; for example,one of the positions in the vector might indi-cate whether the trace is auditory or visual.It should be apparent that this quantitativeview is compatible with our description ofan image or code, and also with the descrip-tion of LTS as a self-addressable store.

STORAGE AND RETRIEVAL

Since LTS is self-addressing, storage andretrieval have many features in common,one process mirroring the other. Storage isassumed to consist of three primary mecha-nisms: transfer, placement, and image-pro-duction. The transfer mechanism includesthose control processes by which 5" decideswhat to store, when to store, and how to storeinformation in LTS. The placement mecha-nism determines the locations in which theensemble of information under considerationwill be stored. To a large degree, the com-ponents of the ensemble itself will determinethe location of storage. That is, in the actionof encoding the desired information for stor-age, 5" may supplement the information cur-rently in STS with pertinent informationretrieved from LTS; the resultant ensemblein STS determines the storage location. Theimage-production mechanism determineswhat proportion of the current ensemble ofinformation in STS will be placed in the


designated LTS location(s). The propor-tion stored should be a function of the dura-tion of the period that the ensemble is main-tained in STS. Retrieval, like storage, isassumed to consist of three primary mecha-nisms : search, recovery, and response gen-eration. The search process is a recursiveloop in which locations or images are suc-cessively selected for examination. As eachimage is examined, the recovery process de-termines how much information will berecovered from the image and placed in STS.The response generation process then exam-ines the recovered information and decideswhether to continue the search or terminateand emit a response. If the search does notterminate, the selection of the next locationor image for examination may depend uponinformation already uncovered during thesearch.

Although storage and retrieval are treatedseparately in this paper, we do not wish toimply that these processes are separated intime, one following the other. Rather, long-term storage is continually occurring for theinformation residing in short-term store. Inaddition, retrieval is continually occurringduring storage attempts by S; for example,5* may try to store a paired-associate bysearching LTS for prominent associations tothe stimulus, associations which could then beused as mediators.

The remaining portions of this report willbe directed toward a delineation of these in-put and output processes. It is beyond thescope of this paper to do more than brieflydescribe the major mechanisms of the theory.Nevertheless, an attempt will be made toindicate how these processes work, what evi-dence supports our assumptions, what pre-dictions may be derived, and how the theorymay be applied to a number of selected tasks.

Storage Processes

Transfer. The decisions involving what tostore, when to store, and how to store infor-mation are under a high degree of control byS, and therefore can result in striking per-formance changes from one task to another.Although we shall be discussing experimentalsituations in which storage takes place largelyduring designated study periods, we should

note that storage in general occurs wheneverinformation is cycled through the short-termstore; for example, in reflective thinking, indaydreaming, and in moment-to-momentconsideration of the day's happenings.

The decision concerning when to storeinformation is especially important in situa-tions where a large amount of information isbeing input rapidly to STS. Such a situa-tion taxes the capacity of the system andforces 5" to select a subset of the presentedinformation for special attention and coding.There are a number of factors that determinethe information so selected. Important oreasily stored information is likely to be givenpreference. For example, Harley (1965)has shown in a mixed-list design that paired-associate items given high monetary payoffsare selectively attended in preference to itemsgiven low payoffs. Information selection willalso be governed by the degree to which theincoming material is already known. Forexample, there is considerable evidence that5s tend to store more information on agiven item if that item's presentation is fol-lowed by other items that are well-known asopposed to being followed by items that arenot known (Atkinson, Brelsford, & Shiffrin,1967; Thompson, 1967). Finally, note thatthe decision when to store may be determinedby strategies associated with the list or taskas a whole, rather than with the individualitem. For example, in an experiment em-ploying a paired-associate list with only thetwo responses A and B, S might decide tostore only information about those stimulipaired with Response A, and always guessResponse B if the answer is not known attest.

The decisions concerning how to store in-formation will also affect performance: stor-age via visual images may be more effectivethan auditory storage (Schnorr & Atkinson,1969); overt and covert rehearsal methodsmay result in very different effects (Brels-ford & Atkinson, 1968); and mediating ver-sus nonmediating instructions may give riseto considerable performance differences(Runquist & Farley, 1964). On a somewhatdifferent level, 5" may engage in organiza-tional storage strategies such that differentitems are stored in locations determined by


some fixed organizational framework (Tul-ving, 1962).

What information is stored, given thepresentation of a particular information en-semble, will also be highly dependent uponthe control processes utilized by S. In somecases, as Underwood (1963) points out inmaking a distinction between the nominaland functional stimulus, not all the informa-tion contained in the presented item is neces-sary for correct responding (e.g., if thestimuli are all nonsense syllables differingonly in the first letter, then only the firstletters need be stored). In these cases onlythe relevant characteristics of the input needbe stored. In most cases S will selectrelevant characteristics of the presented in-formation and add to this additional codinginformation from LTS; for example, apaired-associate plus a mediator might bestored in many cases. One type of informa-tion which S often attempts to store is in-formation indicating that a particular re-sponse (usually one just given in error) isnot correct; Millward (1964) has examinedevidence for this process in simple paired-associate tasks. This process of taggingresponses as incorrect is particularly impor-tant in studies of negative transfer; thehigher the probability that the first responseassigned to a stimulus will be tagged as in-correct (when the response changes), theless will be the proactive interference effectobserved in the data (Shiffrin, 1968).

The examples of transfer mechanismsgiven above are by no means exhaustive, butthey serve to indicate the pervasiveness andimportance of these processes. Relativelysimple decisions to select one transfer schemerather than another can and do lead to largeperformance effects. These facts emphasizethe need for carefully establishing the trans-fer mechanisms used before extending analy-sis to other aspects of the data.

Placement. The location in which animage will be stored is determined by thecontents of that image; 5" therefore controlsthe storage location by manipulating theinformation complex in STS. For any givenensemble of information, however, there isa certain amount of randomness in placement.For this reason, a search for an image may

have to be undertaken at test even if theentire information ensemble originally pres-ent in STS during study is presented forconsideration. Information to be remem-bered may be stored in images in more thanone location; for example, a paired-associatemay be encoded by the use of two entirelydifferent mnemonics. This notion has beengiven quantitative form in "multiple-copy"models for the memory trace (Atkinson &Shiffrin, 1965).

The primary mechanism determining stor-age location is in all cases an organizationalframework. The self-addressing character-istic of input information depends upon theprior, already established, LTS organization;each item is sent to locations that dependupon the preexisting organizational frame-work. However, in many cases (especiallysituations involving free-recall learning andpaired-associate learning in which the samelist of items is presented over a series oftrials) it is an effective technique for 51 toform new organizational structures. In free-recall learning, for example, output on suc-cessive trials becomes increasingly organizedand similar to output on the preceding trial(Gofer, 1965; Tulving, 1962); it may beinferred that a consistent new organizationhas been imposed upon a set of words whichwere disorganized at the start of learning.This point regarding growth of organizationis especially important with regard to list-structured paired-associate tasks, that is, atask in which a list of paired-associates ispresented over and over on successive trials(possibly in a new random order from trialto trial). In such a task, organizational effectsbetween items will tend to occur over trials,and care must be taken in inferring thateffects seen by averaging data over all itemsin the list will also apply to the individualitems. For example, there is evidence indi-cating that interference effects found byaveraging over items in a list do not applyto the individual items making up the list(DaPolito, 1966; Greeno, 1967). For thisreason we shall often consider in the rest ofthis paper results from what is termed a"continuous" paired-associate task (Bjork,1966; Brelsford, Shiffrin, & Atkinson, 1968;Rumelhart, 1967). This type of task em-


ploys a long series of presentations, eachinvolving first a test and then a study on apaired-associate item. The character of theexperiment is essentially homogeneous overtime because new items are continually beingintroduced and old items deleted; a new itemmay appear at any point in the sequence,receive a fixed number of presentations dis-tributed over a subset of trials, and then bedropped and replaced by yet another item.In this way the list-structure feature of thetypical paired-associate experiment is elimi-nated, and it is extremely difficult for -5" todevelop special schemes for interitem organi-zation.

In most cases, 5"'s placement strategy in-volves choosing one of many preexisting or-ganizational dimensions for storage. Inthese cases, an organizational clue containedin the experimental design may prove useful.In categorized free-recall, for example, 5"will be induced to store (and retrieve) in thegiven categories (Bousfield & Cohen, 1956;Cofer, 1965; Cohen, 1963). Thus the loca-tion in which the word division will be storedwill be quite different if preceded by multi-plication, addition, and subtraction than ifpreceded by platoon, regiment, and battalion,

In the interests of effective memory, themost important requirement of the placementprocess is that it results in a storage locationwhich will later be searched during test. Iforganizational schemes are used for place-ment, the schemes themselves will have to bestored, recovered, and utilized in order forretrieval to be optimal.

Image-production. When an ensemble ofinformation is present in STS, some portionof it will be stored in a designated locationin LTS as a permanent image. The propor-tion of information that is stored will be afunction of the duration of time that theensemble stays in STS, or of the number oftimes that ensemble is cycled through STS.It would be most natural to look for evidenceof this mechanism in experiments varyingstudy time for particular items. However,improved performance with longer studytimes can be attributed to a higher chanceof finding a good mnemonic. Better evidenceis found in experiments in which 5"s are in-duced to utilize rote rehearsal rather than

coding, and in which longer periods of roteverbal rehearsal lead to improved perform-ance (Brelsford & Atkinson, 1968; Hellyer,1962).

To conclude the discussion of storage, weconsider the content of the image: the rangeand form of the stored information. Asingle image may contain a wide variety ofinformation, including characteristics of theitem presented for study (its sound, meaning,color, size, shape, position, etc.) and char-acteristics added by S (such as codes, mne-monics, mediators, images, associations, etc.).In addition, an image usually will containlinks to other images (other informationwhich was in the short-term store at the sametime) ; these links can be regarded as a setof directions to the locations of related im-ages in LTS.

Retrieval Processes

The retrieval mechanism forms the cruxof the present theory, since it enables a per-manent long-term store to exhibit the char-acteristics of a failing memory. The basicmechanism by which memory loss occursinvolves a partially random search throughan increasingly large set of images in somelocal set of memory locations. This localarea may be defined by one or more dimen-sions in the organizational structure of LTS;as the number of images in the local areaincreases, the search for the desired informa-tion will become increasingly ineffective.For example, consider two areas, one con-sisting of three images, the other consisting of10 images, and both containing the desiredcode. A random search through the smallerset will result in successful retrieval in ashorter period of time than a search throughthe larger set.

The retrieval process begins with the pre-sentation of an information complex whichplaces constraints on the response desiredand also provides a number of clues, ordelimiting information, concerning that de-sired output (i.e., a stimulus). On the basisof this presented information, or as the resultof an external search strategy, 5" is led tolook in some local memory area and select(possibly randomly) an image for examina-tion. The process by which information is


recovered from this image is called recovery.The recovered information will be placed inthe short-term store which may also containother information such as the search strategybeing employed, salient information recov-ered previously in the search, the LTS loca-tions that have been examined already, andsome of the links to other images that havebeen examined already, and some of the linksto other images that have been noted in thesearch but not yet examined. The short-term store thus acts as a "window" uponLTS, allowing 5" to deal sequentially with amanageable amount of information. The cur-rent contents of STS are now examined andvarious decisions made concerning whetherthe desired response has been found, whetherto emit it, whether to terminate the searchunsuccessfully, or whether to continue thesearch. These decisions and the generationof the response are called the response-gen-eration process. If a decision is made tocontinue the search, then a new location isselected either randomly, on the basis ofinformation just recovered, or in accord withan overriding external search strategy. Theprocess which continues cycling in this man-ner until termination is called the searchprocess.

Search. Each cycle in the search recursionbegins with a mechanism which locates thenext image for examination. This mecha-nism may be separated into directed and ran-dom components. The directed componentincludes strategies controlled by S and de-pends upon the input information and theself-addressing nature of the system. As aresult of the directed component, a number oflocations in some area of memory are markedfor examination. The locations and imagesso marked will be referred to as the exami-nation subset, and the directed component ofthe search may be characterized by the proba-bility that the sought-after code is in theexamination subset.

The directed component of the search canbe either under a high degree of 51 controlor relatively automatic. For example, thecontrol factor is high in a situation where5" engages in a first-letter alphabetic search,or where 5" attempts to remember lunch of2 days previously by reconstructing the

events of that day. In cases like this, thesearch assumes many of the aspects of prob-lem solving. At the other extreme is thealmost completely automatic direction ofsearch which occurs, say, in the attemptedrecognition of a word as having been pre-sented earlier in the session. In such a case,the word itself serves to direct the searchvia the self-addressing feature of the system.

In its most general form the search processcan be viewed as a series of stages in whichsearches are made successively in differentexamination subsets. That is, after choosingcodes randomly for a time from one subset,information recovered during the search maycause 5" to change the memory area currentlybeing examined for one quite different. Thistype of search could occur, for example, ina categorized free-recall task, in which thevarious categories are searched successively(Cohen, 1963). In most applications, how-ever, a restricted search model should be ade-quate, the restriction allowing only a singleexamination subset to be searched duringany retrieval period. This model should beparticularly applicable to tasks in which onlya short response period is allowed.

The random component of the searchspecifies the selection of codes in the exami-nation subset from one cycle of the search tothe next. It is assumed that the likelihood ofany particular code being selected at somepoint will depend upon the amount of infor-mation contained in it, as well as the totalnumber of codes and their temporal orderingin the examination subset. It will be seenthat the extent to which the search dependsupon the temporal ordering of the items isan important variable which, among otherthings, determines interference effects.

The division of search into directed andrandom components leads to somewhat dif-ferent memory models, depending uponwhich component is selected for elaboration.Feigenbaum (1966) and Hintzman (1968),for example, have elaborated upon the di-rected component in computer simulationmodels. In these formulations there is amechanism termed a "discrimination net,"which enables S to sort through the organi-zational structure of LTS to reach the localmemory area where desired information is


stored. On the other hand, there are modelswhich emphasize the probabilistic searchthrough an examination subset of items insome memory area (Atkinson & Shiffrin,1965). It is upon this latter type of modelthat attention will be focused in this paper.

Recovery. Once an image has been lo-cated, it is appropriate to ask what informa-tion contained in the image will be enteredinto the short-term store. This process iscalled recovery. The amount of informationrecovered from an image is assumed to beprobabilistic, depending upon the currentnoise level in the system and the amount ofinformation in the image. In particular, theamount of information recovered should bean increasing function of the amount of in-formation in the image.

Response generation. Having recoveredinformation from LTS, 5" is faced with deci-sions as to whether to terminate the searchand respond or to continue to search. Thesedecisions must first of all depend upon theconsistency of the recovered information withthat indicated by the test information. In-consistent information can be ruled out atonce. If information consistent with the teststimulus is found, and a recognition responseis desired, then the response will be givenwhen temporal or contextual information isrecovered indicating that the image wasstored recently. In a paired-associate test,three types of information may be recovered:that associated with the stimulus, that asso-ciated with the response, or associative in-formation linking the two. Recovery ofstimulus or response information will serveto lead to recognition of either; if both arerecovered in nearby locations in memory,then the response may be emitted. On theother hand, a response might be output fol-lowing recovery of associative informationlinking the response to the stimulus. Sincethe ability to output a response depends uponthe amount of response information recov-ered, the process may often be represented bya decision-theoretic model in which S is at-tempting to filter information through a noisybackground (Bernbach, 1967; Kintsch, 1967;Wickelgren & Norman, 1966). In situationswhere a long time period exists for respond-ing, a likely response may be recovered but

not emitted; in these cases the search will becontinued in the hope of finding a betterresponse. An extended search of this kindleads naturally to predictions that 5" will beable to rank responses in the order of theirprobability of being correct, with responsesranked after the first being correct at anabove-chance level (Binford & Gettys, 1965).

A decision can be made to terminate thesearch unsuccessfully if response time hasrun out, or if response time is expected torun out and .S" wishes to make a guess beforeit does, or if 5 decides that further searchwould not be useful. Termination schemesof the latter type are quite varied. Onesimple rule would terminate the search whenall images in the examination subset havebeen interrogated unsuccessfully. Otherschemes would end the search when somefixed time limit expired, or some fixed num-ber of items examined. These schemes aredescribed in more detail in Atkinson andShiffrin (1965).

The extensive decision structure that hasbeen outlined may make it seem unlikely thattime is available during the search for exami-nation of a very large number of images.This is indeed the point of view we adopt: inmost cases it is assumed that only a fewimages will be examined in the search beforetermination. For example, in a continuouspaired-associate task analyzed by Shiffrin(1968), the estimated number of codes exam-ined prior to termination was from aboutone to five.

APPLICATIONS OF THE SYSTEM

Forgetting

Decrements in performance occur in thesystem as a result of the input of additionalinformation to LTS. These decrements re-sult from three related mechanisms. Firstis a mechanical effect; information sufficientto respond correctly at one point in time mayprove inadequate after additional informationhas been added. For example, a paired-associate GAX-4 may be stored as G**-4,and this code will be sufficient for correctresponding (if recovered) when GAX istested. Suppose, however, that GEK-3 isnow presented and stored as G**-3. When


either of these stimuli is tested later, bothcodes may be retrieved from LTS and there-fore 6" will have to guess whether the correctresponse is 3 or 4.

The second cause of forgetting arises froma breakdown in the directed component ofthe search mechanism. That is, correct re-trieval requires that the same memory areabe searched at test as was used for storageduring study. This may not occur, however,if only a portion of the input information isused to direct storage during study, for adifferent portion might be utilized to locatethe storage area during retrieval. This proc-ess could be viewed within the framework ofstimulus sampling theory (Estes, 1959) ifthe stimulus elements are taken to representdimensions of organization. For clarity, letus denote the image which encodes the cor-rect response for the current test as a"c code," and denote the other codes as"i codes." Thus the i codes are irrelevantcodes which should lead to intrusion errors,whereas the c code, if examined, should leadto a correct response. Then the directedcomponent of search can be characterized bythe probability that the c code is in theexamination subset, called p0. In experimentsin which clues are available to denote theorganizational dimensions to be searched, />„may be close to 1.0. In other situations, suchas continuous tasks with randomly chosenstimuli and responses, p0 will be lower anddependent upon such factors as the amountof information in the c code and its age(where "age" denotes the position of the codeon the temporal dimension). Although the

breakdown in the directed component canprovide a reasonable degree of forgetting,we shall focus primarily upon the thirdmechanism of forgetting: the increasing sizeof the examination subset.

When searching the examination subset,there are a number of possible results. Thec code may be examined and give rise to acorrect response, one of the i codes may beexamined and produce an intrusion response,or none of the codes may give rise to a re-sponse and the search terminates. If thesearch through the examination subset is atleast partially random, then the followingconclusions may be reached. When the sizeof the subset is increased (i.e., the numberof i codes is increased), then the probabilityof giving an intrusion will increase, the aver-age time until the c code is examined will in-crease, and the probability of giving a correctresponse will decrease. When we say thatthe order of search is partially random, wemean to imply that the order in which codesin the examination subset are selected forconsideration may depend upon both theamount of information in the code and theage of the code. Clearly, as the amount ofinformation in a code tends toward zero, oras the age of a code increases, the probabilityof examining that code early in the searchshould decrease.

In order to make the sequence of events inthe search clear, an example of a search ispresented in Figure 2 for a continuouspaired-associate task. In this task, supposethat on successive trials stimulus-responsepairs are presented; on each trial the stimu-

ORDER OFSEARCH

CODES IN THEEXAMINATIONSUBSET

AMOUNT OFSTOREDINFORMATION

STIMULUSNUMBER

TRIALNUMBER

1

ILi18 33 32 31I 1 i I

70 69 68 67

130

1

66

I 1

l.l24 29 20I I I

65 64 63

//

•

1.28 27I l

62 61i i

y-* \\ \\ \

.'~ ,̂ X6

-• 1

1 -R I.I.I I26 25 24 23 22 21 20 19 18 17i I I * * '

60 69 58 57 66 55 64 53 52 61 > _ _

FIG. 2. A schematic representation of an LTS search in a continuous memory task.


lus is tested first and then the pair is studied.A new pair may be presented on any trial,and given several reinforcements and tests atvarying intervals. The bottom row in thefigure gives the trial number for a sectionof the task from Trials 51-70. The numberof the stimulus-response pairs tested andstudied is given in the second row. Thethird row shows which pairs have had codesstored in LTS, and the height of the barindicates the amount of information in thecode. The fourth row indicates those codesthat are in the examination subset on Trial70 when Stimulus 18 is presented for test.For example, the stimulus tested may havebegun with a vowel, and the items in theexamination subset could have stimulus com-ponents which also begin with a vowel.Note that in this example the code forStimulus 18 was stored on Trial 52, andhappens to be in the examination subset. Thetop row of the table gives the order of searchthrough the subset; the first four codes wereexamined and rejected but the fifth codeexamined was the c code and enabled thesearch to end with a correct response. Itmay be seen how forgetting occurs if it isimagined that Item 18 had not been testeduntil Trial 90 or 100. In this event, morei codes would be present in the examinationsubset and the probability would be greaterthat an intrusion would occur, or that thesearch would terminate before the c code wasexamined.

Interference

Various interference phenomena are read-ily predicted by the system. Although ingeneral the order of search through theexamination subset will depend upon the ageand amount of information in the codes,suppose for simplicity that the search orderis entirely random. Then both nonspecificproactive and retroactive interference effectsare predicted, and in a sense are predictedto be equal. That is, extra i codes in theexamination subset added either temporallybefore or after the c code will cause the cor-rect retrieval probability to drop; in thecase of a random search the probability de-crease will be the same whether caused byan i code preceding or following the c code.

The drop occurs because the extra codesincrease the amount of time required to findthe c code. Therefore, if the size of theexamination subset is increased, it is morelikely that either response time will run outor an intrusion response will occur. Obvi-ously the greater the degree to which thesearch is ordered temporally backwards fromthe most recent item, the less the proactive,and the greater the retroactive interferenceeffect. Thus if codes are examined strictlyin temporal order, the average amount oftime until the c code is examined will beindependent of the number of codes whichare older than the c code and hence no pro-active effect will be expected. One of thebest places to examine nonspecific proactiveand retroactive interference effects is in thestudy of free-verbal recall as a function oflist length (Murdock, 1962). In this taska list of words is read to S, who attempts torecall as many of them as possible followingtheir presentation. The data are usuallygraphed as a serial-position curve whichgives the probability of correct recall as afunction of the presentation position (andhence as a function of the number of pre-ceding and succeeding items in the list). Asthe list length is varied, the number of pre-ceding and succeeding items is systematicallyvaried and it is possible to apply the theoryto the resultant data. The application of thetheory is made particularly easy in this casebecause 6" is trying to recall all of the list,and hence the examination subset can beassumed to consist of all codes that havebeen stored. In fact, a model derived fromthe theory has been applied to free-verbalrecall data as a function of list length andhas proved remarkably successful (Atkinson& Shiffrin, 1965, 1968a). The model as-sumes that performance decreases with listlength because more codes are missed in thememory search at longer list lengths. Thismodel also predicts that some of the missedcodes should be retrieved if a second recalltest is given following the first. Just such aneffect was found by Tulving (1967), and itsmagnitude is predicted accurately by themodel. Three successive recall tests weregiven following a single list presentation andonly 50% of the items recalled were recalled


on all three tests, even though the actualnumber recalled remained constant over thethree tests.

Item-specific interference is also readilypredicted via the search mechanism. Iteminterference refers to that condition arisingwhen a stimulus originally paired with oneresponse is later paired with a second, dif-ferent response. In this case two differentcodes with the same stimulus may be placedin LTS. Thus, the amount of proactive orretroactive interference will depend upon thenumber of times the wrong code is examinedand accepted prior to the correct code. Inparticular, the degree of temporal orderingin the search will affect the relative amountsof proactive and retroactive effects. That is,the greater the degree that the search isordered temporally backwards from the mostrecent item, the less proactive and the greaterretroactive effect is predicted. The reason-ing is similar to that in the previous para-graph for nonspecific effects. This rathersimple view of interference is complicated byat least two factors. First, if S is aware thathe will eventually be tested for both re-sponses, he may link them in nearby codes,or in a single code, and thereby reduce inter-ference effects (Ballet & D'Andrea, 1965).Second, when the first response is changed,-S" may tag the first code with the informa-tion that the response is now wrong. If thefirst code is later recovered during search,then this information will enable him toinhibit an intrusion and continue the search;an effect like this was found by Shiffrin(1968).

We might ask how this view of interfer-ence phenomena compares with traditionaltheories such as the various "two-factor"interference theories (Melton & Irwin, 1940;Postman, 1961; Underwood, 1957). In anumber of respects, the present system dif-fers sharply with the traditional views; forexample, in the assumption of a permanentstore. It might be expected in the presentframework, where memory is permanent,that interference effects which appear underone form of test (say recall) would be re-duced under less stringent tests (say recog-nition) whenever the less stringent testsucceeds in making both the old and new

codes available. Evidence of this sort hasbeen found by McGovern (1964). Note thatthe present search system does not necessi-tate the introduction of such processes asproactive and retroactive inhibition and spon-taneous recovery, each with associatedchanges over time. The effects accountedfor by these processes are easy to handlewithin the search framework, at least in con-tinuous tasks of the type presented in Figure2. In list-structured tasks, however, thereis room for considerable complication, in thatlearning of lists allows for organization andretrieval schemes based on the list as a whole.Thus, when the first-list responses to stimuliare all changed in a second list, the organiza-tional strategy or the retrieval scheme, (inthe present terms, the "directed" search com-ponent) may be the mechanism which is dis-turbed, and it may be this disturbance whichis described by the traditional interferencetheories. Indeed, there is evidence from list-structured tasks that interference effectsfound over lists as a whole may not be re-lated to individual stimulus-response assign-ments within those lists (DaPolito, 1966;Greeno, 1967). In any event, when quanti-tative models derived from the present theoryhave been applied to data (Atkinson & Shif-frin, 1968a; Shiffrin, 1968), the searchscheme outlined here handled forgetting andinterference effects in a parsimonious andaccurate manner.

Intrusions

Another useful feature of the model is itsnatural prediction of intrusions, and of varia-tions in intrusion rates over differing condi-tions. In a paired-associate task, an intru-sion occurs when the response contained inan i code is recovered and emitted. Actually,the intrusion process has not yet been speci-fied clearly, since both the probability of be-ing in the examination subset and the proba-bility of accepting the recovered responsewill be smaller for an i code than a c codecontaining an equal amount of information.It may be assumed that the likelihood of ani code being in the examination subset willbe a function of its similarity to the teststimulus, since storage is carried out pri-marily on the basis of stimulus information.


The probability of accepting an t code asbeing correct will similarly depend upon thegeneralization from the test stimulus to thestimulus information encoded in the i code.Given that the i code is examined and ac-cepted, however, the probability that a re-sponse will be recovered and emitted shoulddepend directly upon the response informa-tion encoded, just as for a c code. The abovestatements allow intrusion probabilities to bepredicted for various conditions. In thesituation of Figure 2, for example, an in-crease in intrusions will be predicted over thecourse of the session (since the number ofi codes in the examination subset on testsduring the course of the session will in-crease). This increase has been found insuch a task, and a model based on the presenttheory predicts the increase accurately (Shif-frin, 1968).

Another phenomenon predicted by the the-ory is that of second-guessing, where second-guessing refers to the giving of a secondresponse after 5" has been told that his firstresponse is incorrect. A variety of assump-tions can be made about this process, thesimplest of which postulates that 5" continueshis search of the examination subset from thepoint where the intrusion occurred. Thisassumption predicts that the level of second-guessing will be above chance, an effect foundby Binford and Gettys (1965). If the searchis temporally ordered to any degree, thenstrong predictions can be made concerningthe second-guessing rate depending uponwhether the response given in error waspaired in the sequence with a stimulus occur-ring before or after the tested stimulus (as-suming that the task utilizes a set of uniqueresponses). In fact, examination of thiseffect is one method of determining the tem-poral characteristics of the search.

Latency of Responses

Another variable which may be predictedfrom the theory in a straightforward way isthe latency of responses. The basic assump-tion requires latency to be a monotonic in-creasing function of the number of imagesexamined before a response is emitted.Among the implications of this assumption

are the following. Latencies of correct re-sponses should increase with increases in thenumber of intervening items. This predic-tion holds whenever there is some temporalcomponent to the search, or whenever thenumber of items preceding the tested itemis large. If the reasonable assumption ismade that codes containing more informationare examined earlier in the search, then adecrease in correct response latency is ex-pected as the number of reinforcements in-crease, since the item will gain stored infor-mation over reinforcements and thereforetend to be examined earlier in the search.This effect has been found by Rumelhart(1967) and Shiffrin (1968) in a continuouspaired-associate task. In general, any ma-nipulation designed to vary the number ofcodes examined, whether by instructions, byorganization of the presented material, or byother means should affect the response laten-cies in a specifiable way.

Recognition and Recall

In terms of the present system the searchproceeds in a similar manner whether recog-nition or recall is the mode of test; the dif-ference lies in the size of the examinationsubset in the two cases. Once informationis recovered from LTS, however, the deci-sion process involved in response generationmay be somewhat different for recognitionand recall. In a paired-associate design, thesearch will begin with an attempted recogni-tion of the stimulus, with the decisionwhether to continue the search dependentupon a positive stimulus recognition (Mar-tin, 1967). Hypotheses which ascribe differ-ent retrieval mechanisms for recognition andrecall are not necessary. In both recognitionand recall the presented stimulus will besorted into an LTS area, and a search ini-tiated there. In the case of recognition, thissearch can be quite limited, perhaps consist-ing of an examination of a single image. Inthe case of recall, the stimulus may be recog-nized with little search needed, but the neces-sity for recovering the response may entail alarger search, although "larger" might implyonly examination of two to five additionalitems (Shiffrin, 1968).


CONCLUSIONS

The theory outlined here is descriptive;we have attempted to present a theory ofmemory in fairly general terms and to dem-onstrate for certain commonly studied varia-bles how the theory can be applied. It isbeyond the scope of this paper to presentspecific quantitative models that follow fromthe general theory and apply them to data,but such models have been set forth elsewhereand applied successfully: in continuouspaired-associate learning experiments wherethe variables examined included the numberof intervening items, rankings of responses,second-guessing, proactive interference ef-fects, intrusions, and latencies (Shiffrin,1968); in free-verbal recall where the varia-bles examined included list length andpresentation time (Atkinson & Shiffrin,1968a) ; and in paired-associate memorytasks where the variables include list length,confidence ratings, and response times (At-kinson & Shiffrin, 1965; Phillips, Shiffrin,& Atkinson, 1967). Despite these successes,we wish to emphasize that the theory is stillin an early formative stage, and awaits appli-cation to a wider range of problems. Forexample, it is not yet known whether thetheory can be extended in an elegant way toaccount quantitatively for the interferencephenomena observed in a typical list-struc-tured task. Whatever the fate of such appli-cations, the present theory serves the purposeof providing a general framework withinwhich many of the specific quantitative mod-els known to the authors may be placed,including all of our own work. In addition,we hope that this report will lead to a moredetailed consideration of memory input andoutput mechanisms, especially the memory-search process.

REFERENCES

ATKINSON, R. C., BRELSFORD, J. W., JR., & SHIF-FRIN, R. M. Multiprocess models for memorywith applications to a continuous presentationtask. Journal of Mathematical Psychology, 1967,4, 277-300.

ATKINSON, R. C., & SHIFFRIN, R. M. Mathemati-cal models for memory and learning. TechnicalReport 79, Institute for Mathematical Studies inthe Social Sciences, Stanford University, 1965.(Republished: D. P. Kimble (Ed.), Proceedingsof the third conference on learning, remember-

ing and forgetting. New York: New York Acad-emy of Science, in press.)

ATKINSON, R. C., & SHIFFRIN, R. M. Humanmemory: A proposed system and its controlprocesses. In K. W. Spence & J. T. Spence(Eds.), The psychology of learning and motiva-tion: Advances in research and theory, Vol. 2.New York: Academic Press, 1968. (a)

ATKINSON, R. C., & SHIFFRIN, R. M. Somespeculations on storage and retrieval processes inlong-term memory. Technical Report 127, Insti-tute for Mathematical Studies in the SocialSciences, Stanford University, 1968. (b)

BERNBACH, H. A. Decision processes in memory.Psychological Review, 1967, 74, 462-480.

BINFORD, J. R., & GETTYS, C. Nonstationarity inpaired-associate learning as indicated by a secondguess procedure. Journal of Mathematical Psy-chology, 196S, 2, 190-195.

BJORK, A. Learning and short-term retention ofpaired-associates in relation to specific sequencesof interpresentation intervals. Technical Report106, Institute for Mathematical Studies in theSocial Sciences, Stanford University, 1966.

BOUSFIELD, W. A., & COHEN, B. H. Clustering inrecall as a function of the number of word-categories in stimulus-word lists. Journal ofGeneral Psychology, 1956, 54, 95-106.

BOWER, G. H. A multicomponent theory of thememory trace. In K. W. Spence & J. T. Spence(Eds.), The psychology of learning and motiva-tion. Vol.1. New York: Academic Press, 1967.

BRELSFORD, J. W., JR., & ATKINSON, R. C. Short-term memory as a function of rehearsal proce-dures. Journal of Verbal Learning and VerbalBehavior, 1968, 7, 730-736.

BRELSFOHD, J. W., JR., SHIFFRIN, R. M., & ATKIN-SON, R. C. Multiple reinforcement effects inshort-term memory. The British Journal ofMathematical and Statistical Psychology, 1968,21, 1-19.

COFER, C. N. On some factors in the organiza-tional characteristics of free recall. AmericanPsychologist, 1965, 20, 261-272.

COHEN, B. H. Recall of categorized word lists.Journal of Experimental Psychology, 1963, 66,227-234.

BALLET, K. M., & D'ANDREA, L. Mediation in-structions versus unlearning instructions in theA-B, A-C paradigm. Journal of ExperimentalPsychology, 1965, 69, 460-466.

DAPOLITO, F. J. Proactive effects with independentretrieval of competing responses. Unpublisheddoctoral dissertation, Indiana University, 1966.

DEESE, J. The structure of associations in languageand thought, Baltimore: Johns Hopkins Univer-sity Press, 1966.

ESTES, W. K. The statistical approach to learningtheory. In S. Koch (Ed.), Psychology: A studyof a science. Vol. 2. New York: McGraw-Hill,1959.

FEIGENBAUM, E. A. Information processing andmemory. In Proceedings of the fifth Berkeleysymposium on mathematical statistics and proba-


bility, 1966. Vol. IV. Berkeley: University ofCalifornia Press, 1966.

GREENO, J. G. Some attempts to locate associativeinformation. Technical Report, Department ofPsychology, Indiana University, 1967.

HARLEY, W. F., JR. The effect of monetary incen-tive in paired-associate learning using an abso-lute method. Psychonomic Science, 1965, 3, 141-142.

HELLYES, S. Supplementary report: Frequency ofstimulus presentation and short-term decrementin recall. Journal of Experimental Psychology,1962, 64, 650.

HINTZMAN, D. L. Explorations with a discrimina-tion net model for paired-associate learning.Journal of Mathematical Psychology, 1968, 5,123-162.

KEPPEL, G. Retroactive and proactive inhibition.In T. R. Dixon & D. L. Horton (Eds.), Verbalbehavior and general behavior theory, Engle-wood Cliffs, N. J.: Prentice-Hall, 1968.

KINTSCH, W. Memory and decision aspects ofrecognition learning. Psychological Review,1967, 74, 496-504.

McGovERN, J. B. Extinction of associations infour transfer paradigms. Psychological Mono-graphs, 1964, 78(16, Whole No. 593).

HANDLER, G. Association and organization: Facts,fancies, and theories. In T. R. Dixon & D. L.Horton (Eds.), Verbal behavior and generalbehavior theory. Englewood Cliffs, N. J.:Prentice-Hall, 1968.

MARTIN, E. Stimulus recognition in aural paired-associate learning. Journal of Verbal Learningand Verbal Behavior, 1967, 6, 272-276.

MELTON, A. W. Implications of short-term memoryfor a general theory of memory. Journal ofVerbal Learning and Verbal Behavior, 1963, 2,1-21.

MELTON, A. W., & IRWIN, J. M. The influence ofdegree of interpolated learning on retroactiveinhibition and the overt transfer of specific re-sponses. American Journal of Psychology, 1940,53, 173-203.

MILL WARD, R. An all-or-none model for non-correction routines with elimination of incorrectresponses. Journal of Mathematical Psychology,1964, 1, 392-404.

MURDOCK, B. B., JR. The serial position effect offree recall. Journal of Experimental Psychology,1962, 64, 482-488.

NORMAN, D. A. Toward a theory of memory andattention. Psychological Review, 1968, 75, 522-536.

OSGOOD, C. E. Psycholinguistics. In S. Koch(Ed.), Psychology: A study of a science. NewYork: McGraw-Hill, 1963.

PHILLIPS, J. L., SHIFFRIN, R. M., & ATKINSON,R. C. Effects of list length on short-termmemory. Journal of Verbal Learning and Ver-bal Behavior, 1967, 6, 303-311.

POLLIO, H. R. The structural basis of word asso-ciation behavior. The Hague: Mouton, 1966.

POSTMAN, L. The present status of interferencetheory. In C. N. Cofer (Ed.), Verbal learningand verbal behavior. New York: McGraw-Hill,1961.

RUMELHART, D. E. The effects of interpresentationintervals in a continuous paired-associate task.Technical Report 116, Institute for Mathemati-cal Studies in the Social Sciences, Stanford Uni-versity, 1967.

RUNQUIST, W. N., & FARLEY, F. H. The use ofmediators in the learning of verbal paired-associates. Journal of Verbal Learning andVerbal Behavior, 1964, 3, 280-285.

SCHNORR, J., & ATKINSON, R. C. Repetition versusimagery instructions in the short- and long-termretention of paired associates. Psychonomic Sci-ence, 1969, in press.

SHIFFRIN, R. M. Search and retrieval processes inlong-term memory. Technical Report 137, Insti-tute for Mathematical Studies in the Social Sci-ences, Stanford University, 1968.

SPERLING, G. The information available in briefvisual presentations. Psychological Monographs,1960, 74(11, Whole No. 498).

THOMPSON, W. J. Recall of paired-associate itemsas a function of interpolated pairs of differenttypes. Psychonomic Science, 1967, 9, 629-630.

TULVING, E. Subjective organization in free recallof 'unrelated' words. Psychological Review,1962, 69, 344-354.

TULVING, E. The effects of presentation and re-call of material in free-recall learning. Journalof Verbal Learning and Verbal Behavior, 1967,6, 175-184.

UNDERWOOD, B. J. Interference and forgetting.Psychological Review, 1957, 64, 49-60.

UNDERWOOD, B. J. Stimulus selection in verballearning. In C. N. Cofer & B. S. Musgrave(Eds.), Verbal behavior and learning: Prob-lems and processes. New York: McGraw-Hill,1963.

WICKELGREN, W. A., & NORMAN, D. A. Strengthmodels and serial position in short-term recog-nition memory. Journal of Methematical Psy-chology, 1966, 3, 316-347.

(Received February 2, 1968)

storage and retrieval processes in long …cogs.indiana.edu/festschriftforrichshiffrin/pubs/1969...

Documents