ENCODING SPECIFICITY AND RETRIEVAL PROCESSES IN EPISODIC...

Psychological Review1973, Vol. 80, No. 5, 352-373

ENCODING SPECIFICITY AND RETRIEVALPROCESSES IN EPISODIC MEMORY1

ENDEL TULVING 2

Yule University and University of Toronto,Toronto, Canada

ANT) DONALD M. THOMSON

Monask University, Clayton, Victoria, slttstraliu

Recent changes in prctheorclical orientation toward problems of humanmemory have brought with them a concern with retrieval processes, and anumber of early versions of theories of retrieval have been constructed. Thispaper describes and evaluates explanations offered by these theories to ac-count for the effect of extralist cuing, facilitation of recall of list items by non-list items. Experiments designed to test the currently most popular theory ofretrieval, the generation-recognition theory, yielded results incompatible notonly with generation-recognition models, but most other theories as well:under certain conditions subjects consistently failed to recognize manyrecallable list words. Several tentative explanations of this phenomenon ofrecognition failure were subsumed under the encoding specificity principleaccording to which the memory trace of an event and hence the propertiesof effective retrieval cue are determined by the specific encoding operationsperformed by the system on the input stimuli.

The current transition from traditionalassociationism to information processing andorganizational points of view about humanmemory manifests itself in many ways. Oneof the clearest signs of change has to do withthe experimental and theoretical separationbetween storage and retrieval processes. Inan important early paper, Melton (1963),for instance, pointed out that "the principalissues in theory of memory . . . are abouteither the storage or the retrieval of traces[p. 4]." Only 10 years before Melton madethe statement, it would have puzzled moststudents of verbal learning. At that timememory was still a matter of acquisition, re-tention, transfer, and interference of asso-ciations betwen stimuli and responses.While everyone was aware of the logicaldistinction between acquisition and reten-tion on the one hand and retention and recallon the other hand, these distinctions shaped

1 This research was supported by the NationalScience Foundation Grant 24171X.

The paper was written during the first author'sresidence as a Fellow at the Center for AdvancedStudy in the Behavioral Sciences in Stanford,California.

2 Requests for reprints should be sent to EndelTulving, either at Department of Psychology, YaleUniversity, New Haven, Connecticut 06510 or atthe Department of Psychology, University ofToronto, Toronto M5S 1A1, Canada.

neither experiment nor theory. At the levelof conceptual analysis, the mechanism of re-call was included in the concept of associa-tion ; at the level of experimental operations,recall was observable behavior whose mea-surable aspects simply served to provide evi-dence about strength of associations. More-over, the act of recall was empirically neutralin that it did not affect the state of the sys-tem ; it was theoretically uninteresting be-cause it could not be studied independentlyof acquisition.

The last 10 or 15 years have changed theideational framework for studying memory.Today the orienting attitudes clearly includethe notion that both recall and recognitionare more or less complex retrieval operationsor processes that can be studied and analyzedin some sense separately of storage opera-tions or processes. Retrieval operationscomplete the act of remembering that beginswith encoding of information about an eventinto the memory store. Thus, rememberingis regarded as a joint product of informationstored in the past and information presentin the immediate cognitive environment ofthe rememberer. It is also becoming in-creasingly clear that remembering does notinvolve a mere activation of the learned asso-ciation or arousal of the stored trace by astimulus. Some sort of a more complex in-

352

ENCODING SPECIFICITY AND RETRIEVAL 353

teraction between stored information andcertain features of the retrieval environmentseems to be involved in converting a potentialmemory into conscious awareness of theoriginal event and corresponding behavior.

Relations between the effects of the pastand present inputs and the interaction of thememory trace with the retrieval environmentconstitute the domain of theories of retrieval.Although the important issues are not yetentirely clear, many questions do seem to becentral to the understanding of retrieval pro-cesses and are likely to come more directlyunder experimental and theoretical scrutiny.What, for instance, determines the high de-gree of selectivity of retrieval, the fact thatat any given moment a person can only re-member one discrete event? Is the outputmode of the memory system different fromthe input mode? Can information be re-trieved at the same time that some otherinformation is stored, or does storage al-ways involve retrieval, and retrieval storage ?How do we conceptualize the nature of theeffect of a retrieval cue on stored informa-tion? Does it activate the trace, does itelicit it directly or indirectly, does it provideaccess to it, does it restrict the size of thesearch set, does it somehow complement theinformation contained in the trace, or what?Is there a fundamental difference betweenrecall and recognition, or do these two re-trieval operations differ only in terms of thenature of retrieval cues present at output?What controls retrieval in a situation inwhich no specific cues seem to be present,such as in free recall? What makes somestimuli effective cues for retrieval of a givenevent and others not?

Since the separation of the total act ofremembering into storage and retrieval pro-cesses has only recently been translated intoexperimental observations and theoreticalspeculations, there are as yet no theories ofretrieval that provide satisfactory answersto these and many other possible questionsone might want to ask. The development ofthese theories is one of the important ob-jectives of research on memory. But amodest beginning has been made, and it isthe purpose of the present paper to examineand describe the early state of the art.

In the first part of the paper, we describea familiar phenomenon that has recentlybeen brought into the laboratory and that hasbecome the source of a minor theoreticalcontroversy. This is the so-called extralistcuing effect, facilitation of recall of a listitem by a retrieval cue that was not explicitlya part of the input list. We summarizeand briefly evaluate seven theories that havebeen advanced to explain the effect. In thesecond major section of the paper, we pre-sent some new data from experiments thatwere initially designed to test the most popu-lar and widely accepted explanation of theextralist cuing effect, the generation-recogni-tion theory. Somewhat unexpectedly, theresults of the experiments showed largesuperiority of recall over recognition, a stateof affairs that cannot occur according to thegeneration-recognition theory as well as mostother extant theories of retrieval. In thethird and final section, we consider severalpossible interpretations of the new data andsubsume them, together with other data al-ready in the literature, under the encodingspecificity principle: What is stored is de-termined by what is perceived and how it isencoded, and what is stored determines whatretrieval cues are effective in providing accessto what is stored.

The present analysis applies to memoryfor simple events of the kind frequently usedin laboratory studies. A familiar word orsome other object is presented as a memberof an unfamiliar collection or list, and thesubject's task is to remember that he sawthat particular word or object in that par-ticular list in that particular situation. Insuch a list-item task, the subject may be in-structed to retrieve the whole set of presenteditems, some particular subset thereof, or asingle item. At the time of retrieval, he isgiven general instructions as to the set orsubset of items to be retrieved, and some-times the instructions are supplemented withspecific cues. The cues may be units ofmaterial that were presented as a part ofthe input list (intralist cues). They mayalso be items not explicitly present at thetime of the study (extralist cues). An in-tralist cue to aid retrieval of List Item Bmay be another List Item A, or it may be

354 ENDEL TULVING AND DONALD M. THOMSON

a literal copy of the Target Item B. In theformer case, it is customary to refer to thesubject's task as that of cued recall; in thelatter we say that the task is one of recogni-tion. (Despite the apparent differences inthe nature of these two kinds of cues, weassume that the processes involved are es-sentially the same in both cases, and we makeno theoretical distinction between recall andrecognition. The subject's task is the samein both cases, namely to utilize the infor-mation provided in the retrieval environmentto select some specific stored information.)

The paper is about retrieval processes inepisodic rather than semantic memory. Thedistinction between these two memory sys-tems has been described in detail elsewhere(Tulving, 1972). It forms part of the gen-eral background of the present analysis.Briefly, episodic memory is concerned withstorage and retrieval of temporally dated,spatially located, and personally experiencedevents or episodes, and temporal-spatial re-lations among such events. Appearance ofa word in a to-be-remembered list in an ex-perimental task is such an event. (In typi-cal laboratory experiments the spatial co-ordinates of events to be remembered areheld constant, hence the focus on their tem-poral dates and relations only.) Semanticmemory is the system concerned with storageand utilization of knowledge about wordsand concepts, their properties, and interre-lations. Thus, episodic information about aword refers to information about the eventof which the word is the focal element, orone of the focal elements, while semantic in-formation about a word is entirely indepen-dent of the word's occurrence in a particularsituation or its temporal co-occurrence withsome other words.

We agree with the widely held pretheo-retical assumption that the central repre-sentation of the to-be-remembered event,the memory trace, is a multidimensional col-lection of elements, features, or attributes(e.g., Bower, 1967; Norman & Rumelhart,1970; Posner & Warren, 1972; Underwood,1969; Wickens, 1970). We also assume,along with many other contemporary stu-dents of memory (e.g., Bower, 1967, 1972;Martin, 1968, 1972; Melton & Martin, 1972;

Underwood, 1969, 1972; Wickens, 1970)that an encoding process intervenes betweenthe perception of an event and the creationof the corresponding trace, a process of asyet unknown nature that converts the stimu-lus energy into mnemonic information. Theconcern with retrieval processes in the pre-sent paper necessarily means that we musttake an interest in the nature of stored in-formation and in conditions determining theformat of this information, that is, in encod-ing processes. It makes little sense to talkabout retrieval without knowing something,or at least making some assumptions, aboutwhat it is that is being retrieved.

One term that we will use rather fre-quently in the paper is "effectiveness of re-trieval cues." By this we mean the proba-bility of recall of the target item in thepresence of a discretely identifiable retrievalcue. Effectiveness, as thus defined, can al-ways be expressed in absolute terms, but forcertain reasons it is more convenient to de-scribe the effectiveness of any specific cue inrelation to the basic reference level of re-trieval observed in absence of any specificcues. An effective cue, in this sense, is onewhose presence facilitates recall in compari-son with free or nominally noncued recall.3

EFFECTIVENESS OF EXTRALIST RETRIEVALCUES : SEVEN THEORIES

Effectiveness of extralist cues as an em-pirical phenomenon is well documented. Anearly experiment by Postman, Adams, andPhillips (1955), experiments by Bilodeauand his associates (e.g., Bilodeau, 1967;Bilodeau & Blick, 1965; Fox, Blick, &Bilodeau, 1964), as well as an experiment

3 Subjects may retrieve items through schemesother than those suggested by the experimenter,and match the item to the cue after retrieval hasoccurred. Such cases could distort and inflate theabsolute measures of effectiveness, and this iswhy measures relative to the noncued base lineare preferred. One can then talk about inhibitingeffects of specific cues, although it only meansthat cues assumed to be present in the nominallynoncued situation are more effective in absoluteterms than those the subject attempts to use atthe experimenter's request (e.g., Earhard, 1969;Postman, Adams, & Phillips, 1955; Slamecka,1969).


of Bahrick's (1969) have all shown thatstrong extraexperimental associates of listitems, when presented to the subject as re-call aids, increase the probability of correctresponses. Names of conceptual categoriesof list' words as extralist cues have been usedby Hudson and Austin (1970), Tulving andPsotka (1971), and Wood (1967). Light(1972) has shown that homonyms and syno-nyms facilitate recall of list words. Greatereffectiveness of two than of one associativeextralist cue has been reported by McLeod,Williams, and Broadbent (1971). In a con-tinuous study-test paradigm, Bregman(1968) studied the effectiveness of semantic,phonetic, and graphemic cues at differentretention intervals. Finally, extralist asso-ciates have been shown to be effective cueswhen the subject expects a free-recall testbut not when he expects to be tested for listwords with other cues (Thomson & Tulving,1970).

We can identify at least seven theories ofretrieval that have been or could be proposedto account for the effectiveness of extralistcues, provided that we use the term "theory"in the sense of its most modest dictionarymeaning, as "an idea or a set of ideas aboutsomething."

To make the discussion a bit simpler, weoccasionally use a concrete example in lieuof more general and abstract terminology.In this example, CHAIR is one of the wordevents that is presented for study, and table,a word not shown in the input list, serves asan effective retrieval cue. The question tobe answered then is this: Why does tablefacilitate the recall of CHAIR as the targetword?

Guessing from Semantic Memory

Even if the subject does not have the fog-giest notion that CHAIR occurred in the list—the stored information has been completelywiped out, or he never saw the word in thelist—he can, if he chooses to guess, or is soinstructed by the experimenter (e.g.,Bahrick, 1969; Fox, Blick, & Bilodeau,1964; Freund & Underwood, 1970), boostthe correspondence between the input listand his output protocol by thinking of and

producing words semantically related to thecue word table and thus create the appear-ance of recalling the specific word event fromthe list.

Although the possibility of semantic guess-ing is sometimes mentioned in connectionwith the interpretation of extralist cuingeffects (e.g., Freund & Underwood, 1970),and although it is difficult to deny that somesubjects in some episodic memory experi-ments may utilize information from seman-tic memory only, the theory cannot be takenseriously for at least two reasons. First,subjects on their own seldom resort to thestrategy of semantic guessing in an episodicmemory task, as evidenced by the fact thatin most extralist cuing experiments a largemajority of recall errors consists of plainomissions. Second, the magnitude of cuingeffects is usually (a) considerably greaterthan one could expect on the basis of guess-ing from semantic memory, regardless ofwhat strategy the subject might follow insuch guessing, and (b) independent of, oreven inversely related to, the probability ofchoosing the correct response on the basisof semantic information alone (e.g., Bahrick,1969; McLeod et al., 1971).

Convergence of Episodic and SemanticAssociations

Some theorists are not interested in sub-jects' memories for particular events butrather in changes in probabilities of particularresponses. One might say they are not in-terested in memory but in learning. In theirpretheoretical scheme of things, Word A isidentical with Word A, regardless of thesituations in which they occur. For in-stance, "chair" as a response indicating mem-ory for a particular word event in an experi-mental task and "chair" as a response to theword table in the free-association test arelumped together into the same response class.The changes in the probability of emissionof responses of this class then constitute be-havioral happenings of experimental andtheoretical interest.

Given this type of orienting framework,the following simple explanation of extralistcuing effects can be and has been advocated.


A semantic association exists between tableand CHAIR prior to the experiment. An-other association is created between CHAIR,or the corresponding response, and the gen-eral contextual stimuli present in the experi-ment. In the noncued recall test, only thecontextual association is reactivated, whilein the cued test both the contextual associa-tion derived from the experimental input andthe specific association with table originatingoutside the experiment converge upon theresponse CHAIR, producing a higher proba-bility of the correct response in the cuedsituation.

Such a theory of convergence of experi-mental and extraexperimentally acquired as-sociations has been proposed by Bilodeau andhis associates (e.g., Bilodeau, 1967; Bilodeau& Blick, 1965) to account for their ownobservations in extralist cuing experiments.It tacitly assumes that the activation of theextraexperimental association between thecue and the target word at the time of recallis independent of what happened at the timeof presentation of the target word for study.The demonstration that encoding operationsperformed on a target word affect the effec-tiveness of extralist cues would be incom-patible with the theory. Such experimentalfindings are described later in this paper.

Increments in Trace Strength

A theory closely related to the idea ofconvergent associations holds that presenta-tion of the cue at recall increases thetrace strength of the target word, perhapsthrough the mechanism of elicitation of im-plicit associative responses. Many theoristsassume that the trace of a word can be"strengthened" through repeated presenta-tion of the word for study. They may alsofind it not too difficult to imagine that thetarget word could occur as an implicit asso-ciate to the cue at recall and that such acovert repetition of the target word wouldalso "strengthen" the trace. Such cue-pro-duced strengthening of originally subliminaltraces might thus be responsible for theheightened recallability of target words inthe cued test.

The theory of increased trace strength

differs from that of convergent associationsprincipally in that it does not require theassumption that retrieval of CHAIR dependson associative connections between it andfable, or between it and the experimentalcontext. The theory can, therefore, be ex-tended to recognition tests of trace"strength," in which the role of associativeconnections among items is thought to beminimal.

This theory has been considered by Broad-bent (1973, p. 90) in a discussion of effec-tiveness of extralist retrieval cues and re-jected on the basis of data from an unpub-lished study by Wingrove and Giddingsshowing lack of facilitative effect of extralistcues in recognition tests of target items.Similarly, data from earlier experiments ofours, designed to study the effect of contexton recognition of list words (Thomson,1972; Tulving & Thomson, 1971), showedthat the presence of the cue word table in-hibits recognition of the target word CHAIRif CHAIR occurred in the input list by itself, afinding quite contrary to the theory of in-creased trace strength.

Restricted Search Set

William James's rich legacy includes acolorful metaphor of retrieval as a searchprocess:

We make search in our memory for a forgottenidea, just as we rummage our house for a lostobject. In both cases we visit what seems to usthe probable neighborhood of that which we miss.We turn over the things under which, or withinwhich, or alongside of which, it may possibly be;and if it lies near them, it soon comes to view[James, 1890, p. 654].

Current conceptualizations of retrieval assearch (e.g., Atkinson & Shiffrin, 1968;Norman, 1968; Shiffrin, 1970; Shiffrin &Atkinson, 1969; Yntema & Trask, 1963)have been as much influenced by the develop-ments in computer technology as they havebeen by William James. These searchtheories reject the possibility that retrievalof information about the occurrence of anyspecific event entails the examination of theentire contents of memory. Rather, theyassume that search for desired informationalways takes place in a more or less narrowly


circumscribed region of the memory store,the search set. In any given situation thesize and nature of the search set is specifiedby the information available to the retrievalsystem at the beginning of the search, in-cluding specific retrieval cues, and it may bemodified as search proceeds. In this scheme,an extralist retrieval cue, for reasons thatwe have already met in the previous theories,creates a search set more restricted thanwould otherwise be the case, and since thesuccess of search is, other things being equal,an inverse function of the size of the searchset, cued recall is higher than noncued recall.

Search-set theories assume that search setsare smaller in recognition than in recall tests.Probability that a learned item would berecognized, therefore, cannot be smaller thanthe probability of the item's recall. Experi-mental data inconsistent with this predictionof search theories are described below.

Generation-Recognition Models

At the present time the most widely ac-cepted theories of retrieval are various ver-sions of the generate-test model of informa-tion processing. The generation-recogni-tion models assume that retrieval of storedinformation consists of two successive oroverlapping stages: (a) implicit generationof possible response alternatives and (b)recognition of one of the generated alterna-tives as meeting certain criteria of accept-ability. The generation phase is frequentlyguided by semantic information the systempossesses about the cue word: given tableas cue, the implicitly generated responsesconsist of words semantically related to it,including CHAIR. The operations in therecognition phase, on the other hand, can besuccessful only to the extent that relevantepisodic information is available. The gen-erated response alternative CHAIR would beidentified as the desired word if its internalrepresentation carries an appropriate "occur-rence tag" (Handler, 1972) or "list marker"(Anderson & Bower, 1972), informationabout the membership of the word in a par-ticular list in a particular situation.

Effective retrieval cues of all sorts, in-cluding extralist cues, facilitate recall (to

follow the reasoning of generation-recogni-tion models) because they reduce the proba-bility that the desired information, althoughavailable in the memory store, cannot befound. Cued recall, in this view, producesa higher level of retrieval than does noncuedrecall for the same reason that recognition ishigher than recall. Bahrick (1969), in oneof the most explicit accounts of the workingsof the generation-recognition mechanism,said that a cue or a prompt

is likely to produce a hierarchy of responses as aresult of past learning. . . . One of these re-sponses is likely to be the training response. Sis thus unburdened of the search strategy involvedin unaided recall tasks. He continues to produceresponses associated with the prompt until he canidentify one of them as the response presentedduring training. This portion of the prompted re-call task functionally approximates a recognitiontask [Bahrick, p. 217].

In Bahrick's formal model (1970), the prob-ability of recall of the target item in responseto the extralist cue is the product of theprobability of its implicit generation and theprobability of its recognition.

Many other contemporary thinkers advo-cate, or at least approvingly mention, thegeneration-recognition model as an appro-priate explanation of cuing effects (e.g.,Bower, 1970; Bower, Clark, Lesgold, &Winzenz, 1969; Fox & Dahl, 1971; Kintsch,1970; Murdock, in press; Norman, 1968;Shiffrin & Atkinson, 1969; Slamecka, 1972;Underwood, 1972). The terminology usedby different theorists is not always the same,but the basic ideas are identical. For in-stance, Shiffrin and Atkinson (1969, p. 187)referred to the generation part of the processas search and recovery, and to the recogni-tion phase as response generation. Under-wood (1972) talked about response produc-tion and evaluation of the response for cor-rectness "by the attributes used in makingsuch recognition decisions [p. 11]" andBower (1970), in explaining the effective-ness of rhyming cues, assumed that the sub-ject implicitly generates "the plausible candi-date-responses" among which "he needs onlyto recognize the one that has recently oc-curred in the list context [p. 22]."


Anderson and Bower's (1972) theory ofrecognition and retrieval processes deals withfree recall, but it could be extended to coverextralist cuing effects without introducingany additional mechanisms. In this theorythe generation of potential response itemstakes the form of activation of previouslymarked word nodes in an associative net-work, through pathways also appropriatelymarked during the presentation of the list.The marking of the pathway between theextralist cue node and the target node—anecessary condition for recall in the model-—could be assumed to result from the attemptto link the target node to other list-itemnodes in the input phase of the trial.

All versions of the generation-recognitiontheories subscribe to two closely relatedassumptions. One is the assumption oftranssituational identity of words: everyword has but a single representation in mem-ory, or at most a few representations, cor-responding to its dictionary meaning ormeanings. Perception, encoding, or use ofthe word in any one of a wide variety ofsituations and contexts, including list-itemexperiments, leads to the kind of activationand modification of its internal representa-tion that is thought of as the addition of alist marker or an occurrence tag.

The other assumption is that of "auto-matic access": long-term memory is a self-addressing storage system in which the loca-tion of each unit of information is determinedby the nature and contents of the information(Kintsch, 1970; Norman, 1968; Shiffrin &

Atkinson, 1969). When a to-be-rememberedword is presented for study in a list-itemtask, its meaning and other properties de-termine into what part of the memory storeit is entered. When the same word occursagain—as a repeated study item, as an im-plicit response generated by an associated cueword, as an "old" test item in the recogni-tion test, or in some other way—automaticaccess to the same memory location takesplace, permitting the system, among otherthings, to examine the location for the pres-ence of previously stored episodic informa-tion.

The generation-recognition models pro-

vide plausible accounts of effects of extralistcues. They are also consistent with a largenumber of experimental results showing ab-sence of differences in recognition of learnedmaterial between experimental conditionsthat do produce differences in recall (e.g.,Kintsch, 1970; McCormack, 1972). Butthese models cannot handle situations inwhich an otherwise effective retrieval cue,such as table, fails to facilitate recall of atarget item with which it is closely relatedsemantically, such as CHAIR. They are alsoincompatible with findings showing recall tobe higher than recognition. These kinds ofdata are described presently.

Mediation by Input-Generated ImplicitAssociative Responses

According to a widely accepted assump-tion, when a word is presented for study, itactivates many existing associations (e.g.,Underwood, 1965). For instance, when theto-be-remembered word CHAIR is presentedfor study, the implicit associative responsesmade to it include the word table. When atrecall the word table is presented as a cue,it elicits the word CHAIR, and the backwardversion of the implicit associative responsethus revived adds its strength to whateverother associations are elicited by generalcontextual stimuli responsible for the recallof CHAIR under noncued conditions. Thetheory is rather similar to the theory of con-vergent experimental and preexperimentalassociations, except that it makes the criticalassumption of the necessity of the experi-mental updating of the preexperimental asso-ciation.

This theory was mentioned as a possibleexplanation of extralist cuing results byFreund and Underwood (1970). Althoughthese authors rejected it, since they thoughttheir own data did not demonstrate a "true"extralist cuing effect, the theory is quiteplausible within the framework of currentassociative views of memory. Indeed, Post-man et al. (1955) interpreted their extralistcuing data along the lines of this theory, andit has been used to account for data fromother kinds of experiments that are relatedto cuing studies (e.g., Cramer, 1970; Puff,


1966). The theory is also highly compatiblewith a well-known principle according towhich probability of recall of an item is adirect function of the similarity between therecall situation and the original learning en-vironment (e.g., Hollingworth, 1928; Mel-ton, 1963).

Encoding Specificity Principle

The encoding specificity principle is thefinal idea about retrieval and extralist cuingeffects we discuss. In its broadest form theprinciple asserts that only that can be re-trieved that has been stored, and that how itcan be retrieved depends on how it wasstored. In its more restricted senses, theprinciple becomes less truistic and hencetheoretically more interesting. For instance,we assume that what is stored about the oc-currence of a word in an experimental list isinformation about the specific encoding ofthat word in that context in that situation.This information may or may not includethe relation that the target word has withsome other word in the semantic system.If it does, that other word may be an effec-tive retrieval cue. If it does not, the otherword cannot provide access to the storedinformation because its relation to the targetword is not stored.

Thus, the effectiveness of retrieval cuesdepends on the properties of the trace of theword event in the episodic system. It isindependent of the semantic properties ofthe word except insofar as these propertieswere encoded as a part of the trace of theevent. The distinction between semanticcharacteristics of words as lexical units andwords as to-be-remembered events can bereadily demonstrated with homographs—forinstance, if VIOLET is encoded and stored asa color name, it normally cannot be re-trieved as an instance of the category offlowers, or girls' names—but the same prin-ciple presumably holds for all verbal items.The cue table facilitates recall of the targetword CHAIR if the original encoding of CHAIRas a to-be-remembered word included seman-tic information of the kind that defines therelation between two objects in the sameconceptual category. Most intelligent sub-jects in episodic memory experiments rou-

tinely encode to-be-remembered words se-mantically, and hence words meaningfullyrelated to target items will serve as effectiveretrieval cues.

A recent application of the encoding speci-ficity principle to the interpretation of effec-tiveness of retrieval cues appeared in a studyby Tulving and Osier (1968) ; one of itsmore interesting implications was explicitlytested in three experiments by Thomson andTulving (1970) ; and its bearing on resultsfrom intralist cuing experiments has beendiscussed by Postman (1972). Since theprinciple asserts that it is the encoded traceof the target word rather than the character-istics of the target word in semantic memorythat determines the effectiveness of extralistretrieval cues, as well as all other cues, itcan be experimentally contrasted with theo-ries that attribute the effectiveness of extra-list cues to their preexperimental relationswith target words. Such contrasts, how-ever, are possible only under specialconditions.

Logic of Experimental Comparison betweenTheories

The main difference between the genera-tion-recognition models of retrieval and theencoding specificity principle that is subjectto test lies in the encoding stage of an item'sprocessing as the locus of the effect of cues.According to the generation-recognitionmodels the encoding stage is not important,as long as it does not disturb the capacityof the extralist cue to produce the target itemas an implicit response. According to theencoding specificity principle, the target itemmust be encoded in some sort of referenceto the cue for the cue to be effective.

Both theories can account equally well forthe finding that a given cue in fact is effec-tive. Thus, for instance, if table does facili-tate the recall of the target word CHAIR, it ispossible that an implicit response "chair" wasmade to the cue at retrieval and subsequentlyrecognized. It is also possible that the targetCHAIR was semantically encoded at the timeof presentation in a specific way that ren-dered the cue word table effective. Experi-ments in which specific encoding conditions


are unknown cannot provide critical data forevaluation of theories that differ in claimsthat they make about the importance of theseconditions. Specific encoding operationsperformed on an input usually are not identi-fiable, but they can be experimentally manip-ulated through instructions and other means.Attempts to determine the nature of specificencoding operations in a particular situationthrough subjects' introspections (e.g., Light,1972) are not likely to be any more success-ful than have been most other attempts togain theoretical knowledge about mental pro-cesses from observations of this sort.

To contrast the encoding specificity prin-ciple with other theories, some experimentalcontrol over encoding of target items mustbe exercised. Thus, for instance, the targetword CHAIR could be presented, and the sub-ject induced to encode it in the specific con-text of another word such as glue that issemantically unrelated or only vaguley re-lated to the otherwise effective cue wordtable. If CHAIR is encoded in relation toglue, it is less likely to be encoded at thesame time in relation to table, and hence,according to the encoding specificity princi-ple, the effectiveness of the extralist cuetable should now be reduced. According tothe generation-recognition models, on theother hand, the effectiveness of table as anextralist cue should not be impaired, as longas the semantic relation between table andCHAIR is intact.

The comparison of effectiveness of extra-list cues under conditions in which the sub-jects were free to encode the target item inany way they wished and in which they wereinduced to encode the target item in the con-text of a specific list cue was undertaken byThomson and Tulving (1970). Targetwords such as CHAIR were presented eitheralone or in the context of list cues such asglue, and the effectiveness of extralist cuessuch as table was observed under the twotypes of encoding condition. The resultsshowed that extralist cues did facilitate recallof target words, but only if the target wordsappeared in the list as single items. In thiscase (Experiment II, Groups 1 and 3) cuedrecall was. on tb,e grder of 70% as compared

with the values in the neighborhood of 45 %in the nominally noncued test. Under theconditions in which the target word occurredat input with, and was presumably stored insome relation to another word (ExperimentII, Groups 5 and 6), cuing with extralistcues resulted in a much lower recall levelof some 23%. Noncued recall of targetwords encoded in presence of specific listcues (Experiment II, Group 4) was approxi-mately 30%. These data seem to be morecompatible with the encoding specificityprinciple than with the generation-recogni-tion theory. It is not known, however, towhat extent the pairing of target words withspecific list cues at input may have reducedthe capacity of extralist cues to produce tar-gets as implicit responses, and therefore theresults of the Thomson and Tulving experi-ment are not entirely unequivocal.

We next report data from three experi-ments in which the two phases of retrievalas envisioned by generation-recognition theo-ries, generation and recognition, could bedirectly observed. The experiments werepatterned after the Thomson and Tulvingstudy, except that we did not test the effec-tiveness of extralist retrieval cues under con-ditions where target words were presentedin absence of any specific intralist context.Previous experiments have made it quiteclear that extralist associates of target wordswould be quite effective retrieval cues underthese conditions. This kind of result, as wehave seen, would not distinguish amongtheories.

WHEN RETRIEVAL CUES FAIL :THREE EXPERIMENTS

In the experiments to be described, sub-jects studied a list of target words, such asCHAIR, each presented in the company of aspecific input cue, such as glue. Since thesubjects expected to be tested with thesecues, they presumably encoded target wordsin an appropriate relation to the input cue.After studying the list, the subjects wereasked to produce free association responsesto strong extraexperimental associates oftarget words such as table. The probabilityof generation of target words (CHAIR) in


TABLE 1

MATERIALS USED IN THE CONSTRUCTION OFCRITICAL LISTS IN EXPERIMENTS 1, 2, AND 3

List A

Weakcue

groundheadhathcheesestomachsunprettycavewhistlenoisegluecommandfrui thomegraspbutterdrinkbeatclothswiftladybladeplantwish

Strongcue

hotdarkwantgrasssmallnightskydrytennisblowtablewomanbloombitterinfantroughtobaccoachelambstopkingscissorsinsectsoap

Targetword

COLDLIGHTNEEDGREENLARGEDAYBLUEWETBALLWINDCHAIRMANFLOWERSWEETBABYSMOOTHSMOKEPAINSHEEPGOQUEENCUTBUGWASH

List B

Weakcue

hopestemwhiskeymothcabbageglasscountrytoolmemorycoveringbarnspidercrustdeeptrainmountaincottageartadultbravedoorrollthinkexist

Strongcue

lowlonglakeeatsquaresoftclosedfingerfastliningcleaneaglebakebedwhiteleafhateboylaborstrongcolorcarpetdumbhuman

Targetword

HIGHSHORTWATERFOODROUNDHARDOPENHANDSLOWCOATDIRTYBIRDCAKESLEEPBLACKTREELOVEGIRLWORKWEAKREDRUGSTUPIDBEING

response to the extralist cues was one ob-servation of interest. Next, subjects wereasked to identify those generated words thatthey remembered having seen in the inputlist, their success in doing so being anotherobservation of interest. Finally, a cued re-call test involving input cues (glue) wasgiven to the subjects in an attempt to esti-mate the extent to which information abouttarget words (CHAIR) was available in thememory store.

General Method

In all three experiments the same procedure wasused up to a critical point in time when differenttreatments and tests were administered to subjects.The procedure common to all experiments follows.

Every subject was shown and tested on threesuccessive lists. The sole purpose of the first twolists was to induce subjects to encode each targetword with respect to, or in the context of, anotherword. The word pairs in these two set-establishinglists were comparable to weak-cue input lists usedby Thomson and Tulving (1970). The targetwords, each paired with its cue word, were shownvisually, one at a time, at the rate of three seconds/pair. Immediately at the end of the presentationof the list the subjects were provided with 24haphazardly ordered input cue words on a recallsheet and instructed to write down the target words.Three minutes were given for the recall of the list.The mean number of words recalled for these twolists were 14,3 and 17,6 in Experiment 1; 15.7 and

18.3 in Experiment 2; and 14.4 and 17.7 in Experi-ment 3.

The third list in each experiment was the criticallist, providing the data of interest. This list, too,consisted of 24 cue-target pairs, with the materialpresented exactly as in the first two lists.

The target words in the third list in each experi-ment were those designated as such in Table 1.The two sets, A and B, were used equally frequentlywith two subgroups of subjects in each of the threeexperiments; otherwise, all subjects in a givenexperiment were treated identically. In Table 1,each target word is accompanied by two cue words,one "weak" and the other one "strong." Thesetriplets were selected from free-association norms(Bilodeau & Howell, 1965; Riegel, 1965) to con-form to the following criteria: (a) the target wordis a low-frequency (mean of 1% for the whole set)associate to its weak cue; (b) the target word isa high-frequency associate to its strong cue (meanof 52%), and (c) weak and strong cues of a giventarget word are not associatively related to eachother in the norms.

The 24 pairs of words in critical lists consistedof weak cues and their corresponding target words.The strong cue words were not shown at the timeof the presentation and were used only as extralistretrieval cues in the subsequent test phase. Forinstance, subjects tested with List A saw the pairs,(iround COLD, head LIGHT, bath NEED, and so on, onepair at a time, for all 24 pairs, the cue word appear-ing in lower-case letters above the capitalized tar-get word. The same instructions that had beengiven to the subjects in the first two lists wereroutinely repeated: their task was to remember thecapitalized words, but paying attention to cuesmight help them at the time of the subsequent test.No mention was made of any change in the pro-cedure between Lists 2 and 3. Each pair was againshown visually for three seconds. After all 24pairs had been presented once, subjects receiveddifferent treatments in different experiments, asdescribed below.

Experiment 1

After the presentation of the third, critical list—henceforth referred to as "the list"—all 40 subjects,undergraduates at Yale University, were given thesame four successive tasks.

First, each subject received a sheet of paperlisting 12 extralist cues corresponding to one halfof the to-be-remembered words from the list.For instance, subjects tested with List A were giventhe words hot, dark, want, and so on. They weretold that each of the listed cue words was relatedto one of the capitalized words in the list that theyhad just studied, and that their task was to writedown as many of the capitalized words as theyremembered, each one beside its related cue word.Three minutes were allowed for this task. Themean number of target words recalled in thisextralist-cue test was 1.8 (15%).

362 ENDEL TULVXNG AND DONALD M. THOMSON

Second, subjects were given the remaining 12extralist cues, briefly told about the free associa-tion procedure, instructed to look carefully at eachcue word, produce free associates to each mentally,and then, "if one of the words you generate as afree association is a word from the list that you havejust studied," to write it down beside its stimulusword. Again, three minutes were given for thistask. The mean number of target words recalledin this test was 3.6 (30%). This figure was sig-nificantly higher than the 15% recall level in thefirst phase.

Third, all 24 extralist cue words, that is, the cuesthat the subjects had just seen in the first andsecond phase of testing, were presented to thesubjects once more, listed on a sheet of paper,together with instructions to "write down all thewords that you can generate as free associations"to these stimulus words. Beside each stimulusword were six blank spaces for up to six responses.Twelve minutes were allowed for this task. Eachsubject generated, on the average, 104 free asso-ciation responses to the 24 stimulus words, a meanof 4.4 response words. Among the 104 responsesthus generated there were, on the average, 17.7(74% of the 24) words matching the target wordsof the list. Of these words, 70% were given bysubjects as the primary response to the stimuluswords. Thus, the proportion of target words gen-erated as primary responses to the high-frequencystimuli was 52% (70% of the 74%), matching thenormative data exactly.

Fourth, the subjects were instructed to look overall their generated responses and to circle all wordsthat they recognized as target words from the listthey had learned last. They were given as muchtime as they needed for this task. The meannumber of words circled was 4.2, out of the maxi-mum of 17.7, producing a hit rate of 24%. Thepercentage of false positives, circled words thatwere not target words from the list, was 4.5%. .

A fifth task in Experiment 1 was an afterthoughtwhose relevance became clear after we had testedand seen the results from 30 subjects. It wasadministered to the final 10 subjects. (Their per-formance on the first four tasks was not dis-tinguishable from that of the first 30 subjects.)These 10 subjects were provided, on two sheetsof paper, with the 24 input cues from the list andwere instructed to recall the capitalized words fromthe list they had seen last. The mean number oftarget words recalled on this cued recall test was15.2, for a hit rate of 63%. Further analysis ofthe data from these 10 subjects revealed that intheir fourth and fifth tasks, they both recognizedand recalled a total of 43 words; recognized butdid not recall 4 words; and recalled but did notrecognize 69 words.

These data can be summarized as follows: (a)Pairing of to-be-remembered words at input withcue words associatively unrelated to subsequentlypresented extralist cues has no adverse effect onsubjects' ability to utilize these extralist cues in

generating target words from semantic memory.(6) Regardless of whether subjects are or are notinstructed to use the generate-recognize strategy,under conditions where cues are switched from in-put to output, the level of their recall performance inpresence of extralist cues does not materially exceedthat expected under noncued conditions (cf. Thom-son & Tulving, 1970). (c) Under the experimentalconditions as described, subjects cannot recognizemany generated copies of target words, althoughthey can produce these words in presence of whatappear to be more effective cues, context itemsfrom the input list. The recognition hit rate of24% contrasted with the hit rate of 63% in the cuedrecall test.

Experiment 2

One purpose of Experiment 2 was to replicateExperiment 1; another was to find out to whatextent the low hit rate in the recognition of gen-erated free association responses was attributableto the source of the recognition test items. InExperiment 1, each subject generated the "old"items and distractors for the recognition test. InExperiment 2 this procedure was replicated withone half of the list items, while for the other halfof the list the source of "old" test items and dis-tractors for a given subject was the free associationprotocol of a yoked subject.

Twenty-two subjects, undergraduate students ofboth sexes at Yale University, participated in thisexperiment. Again, they were divided into twogroups for the sole purpose of using both Lists Aand B, as shown in Table 1, in the critical, thirdlist position. Otherwise, all subjects were admin-istered an identical sequence of tasks.

Each subject was again first tested with twosuccessive set-establishing lists, and then the thirdlist was presented, as in Experiment 1. Followingthe presentation of the list, subjects were given fivesuccessive tasks.

1. They were asked to generate and write downsix free association responses to each of the 12extralist cue words corresponding to the targetwords. The mean number of words generated was54 per subject, of which 9.6 (80% of the 12)matched target words in the list. Primary re-sponses coincided with target words 53% of thetime.

2. Subjects were asked to examine all the wordsthey had generated in the free association test andto write down, on a separate piece of paper, allthose words they recognized as target words fromthe list. On the average, 1.8 target words werethus recognized, for a hit rate of 18%. The falsepositive rate was 2.8%.

3. Each subject was given the free associatesgenerated by another subject to the remaining 12extralist cues, and instructed to perform the samekind of recognition test on these words that theyhad performed on their own. This time the 22subjects correctly identified 51 out of the total of


204 copies of target words, for a hit rate of 25%,a score not significantly different from the 18%hit rate of subjects' own generated responses,( = 1.18, # = 21.

4. Subjects were asked to write down, besideeach of the recognized target words from the secondset (yoked subjects' free association responses),the corresponding input cues they remembered fromthe list. The 22 subjects could provide 36 (71%)such cues for the 51 targets.

5. Finally, all subjects were given all 24 originalinput cues from the list and were asked to recallas many target words as they remembered. Themean number of words recalled on this test was14.1, for a hit rate of 59%. Again, as in Experi-ment 1, there were very few words that wererecognized in the second task but not recalled inthe fifth task (a total of 5 for all 22 subjects), andthere were numerous words recalled in the fifthtask but not recognized, although generated, inthe second task (total of 73 for 22 subjects).

The results of Experiment 2 thus confirmed theresult's of Experiment 1: when provided with strongextraexperimental associates of target words asstimuli, subjects generated many responses matchingthe target words but they did rather poorly inidentifying these as target words from the list.In Experiment 2, the hit rate in the recognitiontest of generated target words was, on the average,22%, while the false positive rate was 2.8%. Thesource of the "old" test words and distractor itemsin the recognition test—whether the subject's ownsemantic memory or that of another subject—didnot seem to be an important determinant of therecognition performance under these conditions, al-though the confounding of order of tests withexperimental treatment may have contributed tothe absence of a significant difference between thetwo conditions. Subjects also remembered a sizableproportion of input cues for the target words theycorrectly identified from among those generated bytheir yoked partners. Finally, the data from Ex-periment 2 confirmed those from Experiment 1 inshowing that a low hit rate in the recognition teston generated target words did not prevent subjectsfrom doing reasonably well in recalling these wordsin presence of the original input stimuli. The meanrecognition hit rate of 22% in Experiment 2 againcontrasted starkly with the cued-recall hit rate of59%.

While the data in these experiments were aver-aged over all target words in the two critical lists,A and B, as shown in Table 1, it may be of someinterest that large differences in recognition ofindividual words occurred. The words in bothlists in Table 1 are ordered from least to mostrecognizable, on the basis of data in Experiments1 and 2. Target words at the top of each list(COLD, LIGHT, HIGH, SHORT, WATER) were neverrecognized, even though each was generated byanywhere from 13 to 21 subjects; words in themiddle (BALL, WIND, CHAIR, MAN, FLOWER, SLOW,COAT, DIRTY) showed individual hit rates of ,14 to

.16; words at the bottom (QUEEN, CUT, BUG, WASH,STUPID, BEING) were correctly recognized over50% of the time.

Experiment 3

A possible interpretation of the very low recogni-tion hit rates of generated to-be-remembered wordsin Experiments 1 and 2 might be provided by in-voking the concept of "high criterion": for reasonsunknown, the subjects adopt a very cautious atti-tude in the recognition test and check off onlythose words that they are extremely confidentabout as being identical with target words fromthe list. While the observed false positive rates,in relation to the observed hit rates, are sufficientlyhigh to weaken the force of this argument, wemade a more direct attempt to evaluate this in-terpretation in Experiment 3.

Fourteen subjects, from the same source as thosein Experiments 1 and 2, served in this experiment.Again, they were divided into two groups for thesole purpose of counterbalancing materials in thecritical test conditions. The materials and otherconditions of the experiment were identical withthose used in Experiments 1 and 2.

Subjects were again given two set-establishinglists first. The third list was then presented underthe same instructions and conditions as Lists 1 and2, and followed by three tasks. First, subj ects wereprovided with two sheets of paper, listing the 24extralist cues corresponding to the target wordsfrom the list. They were instructed to write downfour free association responses to each of the 24cue words, words that the cue words "made themthink of."

Second, a forced-choice recognition test wasgiven. Subjects were instructed to look at thefour words they had generated for each of thestimulus words and circle the word that appearedto be the most likely member of the set of targetwords in the last list they had learned, guessingwhenever necessary. In addition, the subjectswere asked to indicate their confidence in the cor-rectness of the response, on a scale on which thethree values, 1, 2, and 3, were labeled as "guessing,""reasonably sure," and "absolutely sure."

The 14 subjects generated a total of 221 targetwords in the free association test, or an averageof 15.8 (66%) of the possible 24. The proportionof target items among primary responses was 46%,only a little less than the normative value of 52%.In what was effectively a four-alternative forced-choice recognition test, the subjects correctly circled118 out of the 221 copies of target words and failedto recognize the other 103 words, for a hit rateof 53%. Of the 118 to-be-remembered words cor-rectly circled in the recognition test, 47 were labeledas guesses, while the remaining 71 were given con-fidence ratings of 2 or 3. The standard guessingcorrection yields a corrected recognition score of38%. Alternatively, if we consider only those to-be-remembered words as recognized for which the


subjects gave confidence ratings of 2 or 3, we ob-tain a recognition score of 71/221, or 32%.

In the third and final task, subjects were testedfor recall of target words in presence of theiroriginal input cues, as in Experiments 1 and 2.The mean number of target words recalled was 14.2,for a hit rate of 61%, considerably higher than therecognition scores of 32% or 38%.

The major finding in Experiment 3 was that thefailure to recognize recallable words also occurs,perhaps in a somewhat attenuated form, under theconditions of the forced-choice recognition test. Itrules out the response bias as the sole explanationof the similar results obtained under free-choicerecognition procedure in Experiments 1 and 2.

General Results

A summary of the data from the three ex-periments, together with those from Experi-ment II in the Thomson and Tulving (1970)paper, is presented in Table 2. Table 2shows probability of recall of target words inpresence of three kinds of specific cues: (a)input cues, ( b ) copies of target words gen-erated by subjects in free association tests,and ( c ) strong preexperimental associates oftarget words used as extralist retrieval cues.

The importance of the data pertaining torecall in presence of input cues is twofold:the observed levels of recall in various ex-periments indicate the extent of effectivenessof input cues in relation to other cues, andthey provide a lower-bound estimate of theavailability of information about target wordsin memory. The data thus constitute a criti-cal link in the argument that takes us fromthe results to the general conclusion of theexperiments: conditions can be created inwhich information about a word event isavailable in the memory store in a form suffi-cient for the production of the appropriateresponse and yet a literal copy of the wordis not recognized. This phenomenon of rec-ognition failure was a striking one. Ignoringwords that could be both recalled and recog-nized and those that could be neither recog-nized nor recalled—response categories de-fining theoretically uninteresting outcomes—-we found, in Experiments 1 and 2, that thenumber of words that were recalled but notrecognized exceeded the number of wordsthat could be recognized but not recalled bya ratio of approximately 15:1.

The phenomenon of recognition failure of

TABLE 2PROBABILITY OF RECALL OF THE TARGET WORD

IN THE PRESENCE OF THREE KINDSOF RETRIEVAL CUES

Cue

Input cueExtraexperimental

associateCopy of target

Tulving &Thomson

(1970)"Exp. II

.65

.23--•

Exp. 1

.63

.15, .30.24

Kxp. 2

.59

—.22

Exp. 3

.61

—.32, .38

a Data are from Group 6, Experiment II, by Thomson andTulving (1970).

recallable items is not a novel one. Otherexperiments have been reported in whichlearned materials were recalled at higherlevels than they were recognized (Bahrick &Bahrick, 1964; Lachman & Field, 1965;Lachman, Laughery, & Field, 1966). Butin these experiments the subjects were givenonly a limited amount of time to make therecognition judgments, and it is not knownto what extent the outcome was a conse-quence of time pressure. Tt is quite likelythat if the subjects in the Lachman and Field(1965) experiment, for instance, in whichthe learned material was a meaningful prosepassage, would have been given more time,they might have been able to recognize allwords that they could recall, simply by re-producing each word as an element of thelearned passage and matching it with thetest word. This strategy may not have beenfeasible under the conditions of Lachmanand Field's recognition task in which sub-jects only had 1.5 seconds to make the de-cision about each word.

FAILURE OF RECOGNITION OFGENERATED WORDS

The recognition failure of recallable wordsis an empirical phenomenon that cannotoccur according to the two-process theoryof recall and recognition (Kintsch, 1970)and other versions of the generation-recogni-tion model of retrieval (e.g., Bower et al.,1969; Murdock, in press; Norman, 1968;Shiffrin & Atkinson, 1969; Slamecka, 1972 ;Underwood, 1972). Recovery of informa-tion through two bottlenecks (generation andrecognition) in a recall situation cannot be


more effective than that through only one ofthe two (recognition). Since the experi-mental data show that under certain con-ditions generation and recognition produce ahigher level of retrieval than recognitionalone, existing generation-recognition modelsrequire revision.

The phenomenon of recognition failure ofrecallable words violates the two criticalassumptions on which generation-recogni-tion models are based, the assumption oftranssituational identity of words and theassumption of automatic access to the inter-nal representation of a word. If a word hadonly one representation in memory, and ifthat representation were modified by theappearance of the word in an experimentallist, then recognition could fail only becauseof loss or deterioration of the relevant listor occurrence information. If recall alsodepended upon an intact list marker, its levelcould never exceed that of recognition. Inour experiments, the copy of the recallabletarget word that was not recognized mayhave provided automatic access to some in-ternal representation of the word, or to somespecific location in the memory store, but theinformation about the membership of theword in a particular list was apparently notstored in that location. The question thusis not whether the occurrence of a stimulusword, produced in whatever fashion, pro-vides automatic access but rather automaticaccess to what ?

One might perhaps argue that the con-ditions of our experiments lie outside theboundaries of the domain of the generation-recognition models, and that in "normal"experimental situations the models do ratherfaithfully represent the retrieval process. Inthat case we need to be told what the domainof the generation-recognition models is, andwhat kind of a model would account for thedata that show strong associates of targetwords to be ineffective cues. In this con-nection it is helpful to remember that theprocedure we used rendered many recallablewords not recognizable, but it did not seemto affect the ability of subjects to success-fully generate copies of target words in re-sponse to extralist cues. The failure ofretrieval as envisaged by the two-process

theory had its source in the recognition phaseand not the generation phase.

A plausible modification of the generation-recognition models that would bring theminto line with the phenomenon of recognitionfailure of recallable words would involvemaking the assumptions that (a) the pairingof the target word with another word at in-put creates a specific semantic meaning forthe target word, and (b) this specific mean-ing is stored with the "core" representationof the word and marked with a list tag. Theinput cue may provide access to the appro-priate location in the memory store, becauseof its list-specific association with the markedsense of the target word. But a copy of thetarget word may fail to do so, because itwould be interpreted in a different sense inabsence of the input cue, and thus provideaccess to an incorrect storage location.

A generation-recognition model thus modi-fied would become in many ways quite simi-lar to the encoding specificity principle inits general orientation. In the modified ver-sion the requirement would be relinquishedthat a given lexical item be represented inonly one location. It remains to be seenwhether it would be possible to retain someother basic features of these models, such asthe assumption that episodic information isstored in the same system as semantic infor-mation, that occurrence of the to-be-remem-bered word in a list results in modification ofthe existing representation of the word inthe memory store, and that recall, but notrecognition, involves generation of alterna-tives other than the desired target words.

The phenomenon of recognition failurecreates difficulties not only for extant ver-sions of generation-recognition models butfor most other theories of retrieval as well.Apart from the encoding specificity principle,to which we will return in the final sectionof the paper, the theory that seems to bemost readily reconcilable with the data isthat of mediation by implicit associative re-sponses. The theory represents a specialcase of the principle of reinstatement ofstimulus conditions and we will briefly con-sider it as such. This principle states thatthe success of retrieval depends on the com-pleteness with which stimulating conditions


present at input are reinstated at the time ofattempted retrieval (Hollingworth, 1928;Melton, 1963). Since the stimulating con-ditions present at input could be thought toinclude unobservable processes of the kindthat have been labeled implicit associative re-sponses, presentation at the time of the recallof extralist retrieval cues that match implicitassociative responses would produce agreater degree of similarity between the testand study conditions than that prevailing inthe noncued test situation. While the princi-ple of reinstatement of stimulus conditionsdoes provide a plausible interpretation of theeffectiveness of extralist retrieval cues, itcannot account for the phenomenon of recog-nition failure observed in our experiments.Given that the input consists of a compoundstimulus such as glue-CHAix, and given thatthe subject clearly knows that his memorywill be tested for the target word CHAIR, whyshould the stimulating conditions be "morecompletely" reinstated when the glue part ofthe compound rather than the CHAIR part ispresented at test ? Thus the theory of medi-ation by implicit associative responses andthe principle of reinstatement of stimulusconditions do not explain the phenomenonof recognition failure any more readily thando generation-recognition models.

ENCODING AND RETRIEVAL

In this last section of the paper, we con-sider some possible tentative explanationsof the phenomenon of recognition failure,relate the phenomenon to other relevant datareported in the literature, and suggest anoverall pretheoretical framework withinwhich this phenomenon as well as otherproblems of retrieval could be experimentallyand theoretically analyzed. The demonstra-tive nature of our own experiments precludesan enlightened interpretation of the failureof extralist associates and copies of targetitems as retrieval cues, but some possibilitiesfor interpretation should be mentioned.

Recognition Failure: TentativeInterpretations

One may wish to approach the phenome-non of recognition failure of recallable words

in terms of the concept of asymmetry of asso-ciations between the input cue and the targetitem. The data show and imply that the"forward" association between the cue andthe target in our experiments was strongerthan the "backward" association between thetarget and the cue: given the cue, the targetword could be recalled without difficulty inmany cases; given the target word, the inputcue could be recalled only infrequently.Thus, it would be possible to argue that therecognition failure of target words was aresult of the weak association between thetarget and the cue. If that association hadbeen stronger, the input cue could have beenretrieved given the target word, and the tar-get word itself recognized in the presence ofthe cue.

If one adopted this approach, the problemwould become that of explaining the asym-metry of the associations between the inputcue and the target word. One would alsohave to keep in mind an equally plausiblealternative line of reasoning: backward asso-ciation appears to be weaker than the forwardassociation whenever the target word is notas readily recognized as the cue word. Rec-ognition of the stimulus item, which dependson the consistent encoding of the item atinput and at test, can be thought of as anecessary condition for associative recall(Martin, 1968, 1972). The difficulty herelies in the uncertainty of whether to accountfor failure of recognition in terms of a weakassociation or to interpret the apparent weakassociation in terms of the recognition failure.

Another approach to the problem of inter-pretation of the phenomenon of recognitionfailure could begin with the distinction be-tween nominal and functional memory units.This distinction has often been made in dis-cussions of associative and organizationalprocesses in memory (e.g., Asch, 1969;Kohler, 1947; Tulving, 1968). A verbalunit may be designated as an independentto-be-remembered item by the experimenter,but the processing system, that is the subject,may treat it as a part of a larger functionalunit. For instance, a subject shown a listof words under instructions to rememberthem could do reasonably well if he wereasked to recall the letter bigrams he saw in


the list. He would "recall" the bigrams byretrieving information about the words andconstruct the units designated by the experi-menter in a postretrieval operation; he wouldnot store bigrams in the same sense as hestores words. Even under intentional learn-ing conditions, subjects are known to havegreat difficulty in recognizing bigrams theycan readily recall as parts of integrated words(Watkins, 1973). Recognition failure ofwords that were recalled quite well inresponse to the specific input cues in ourexperiments may also have come about be-cause the "old" test item in the recognitiontest did not match the functional memoryunit, the complex consisting of the inputcue and the target word. If the whole unithad been presented for recognition, the sub-jects probably would not have had any greatdifficulty.

While the explanation of the phenomenonof recognition failure in terms of the distinc-tion between nominal and functional memoryunits is incomplete as long as we do notknow exactly what are the functional unitsin any given situation, the notion might pro-vide some guidelines for identification ofmore appropriate units of analysis than thosedefined by the experimenter. For instance,it may be reasonable to claim that a func-tional memory unit is one which, amongother things, must be recognized if it is tobe recalled. Thus, if a nominal unit couldbe recalled as a part of a larger complex butcould not be recognized outside the complex,it would not correspond to a functional unit.The recall-cum-recognition criterion couldbe combined with other rules for identifica-tion of functional units, for instance, thatthey are activated and emitted in an all-or-none fashion (Handler, 1967b).

Another closely related interpretation ofrecognition failure could be based on theGestalt concept of "embeddedness." It iseasy to demonstrate failure of perception ofa figure embedded in a larger complex, andsimilar demonstrations have been reportedin memory situations. Koffka (1935, p.622), for instance, discussed an experimentby Harrower (1933) in which punch linesof jokes could not be recognized as readilywhen they appeared embedded in a meaning-

ful passage than when they were presentedas a part of a collection of unrelated state-ments. Winograd, Karchmer, and Russell(1971) interpreted their data demonstratingeffectiveness of retrieval cues in recognitionmemory by assuming that under certain con-ditions nominal elements are compoundedinto a unitary representation. The principleof embedding would also apply to the resultsof an earlier demonstration of recognitionfailure of recallable words (Tulving, 1968b)in which the to-be-remembered words couldbe encoded by subjects as second halves ofcompound words, for example, air-pORT;house-HOLD; sand-HOG. In our present ex-periments the encoding operation performedon the input pairs also may have producedtightly knit compounds. The well-knownGestalt principle that properties of elementschange when these elements become partsof larger wholes would then explain whycopies of target words were not particularlyeffective in providing access to the tracecreated by the compound of the input cue andthe target word. Thus, CHAIR may havebeen as difficult to retrieve from the "gluc-CHAIR" complex stored in memory as itwould be difficult for a person to recognizeHIT as an "old" item when the stored unitis "architecture" or POT when the storedcomplex is "hippopotamus." The embeddedpart can be discerned easily enough in a per-ceptual display of the whole, both before thewhole has been stored in and after it has beenretrieved from memory, but its identificationmay be impossible after storage and beforeretrieval.

The major problem with the explanationof the recognition failure in terms of theconcept of embedding has to do with theassumed difference in the recognition ofinput cues and target words: Why is the in-put cue not embedded in the cue-target com-plex in the same way as the target wordis? Data reported elsewhere (Thomson,1972) show the same pattern we are assum-ing here and create the same problem ofinterpretation: right-hand members ofweakly associated input pairs were not aswell recognized as left-hand members whenthe test word was presented alone.

Given the fact that the target word can


be more readily recalled in presence of theinput cue than in presence of its own literalcopy, it is possible to think of the input cueas a "control element" or "code" that governsthe access to the complex of stored informa-tion about the target word. The concept of"control element" has been used by Estes(1972) and the closely related concept of"code" by Johnson (1970) in theoreticalaccounts of processing of serial information.In the present context, we think of anythingthat controls retrieval of a particular memoryunit at a particular time as a control elementof that unit. A retrieval cue that is not onlyeffective but also necessary for the recoveryof certain stored information (an access cue)would be such a control element.

Recent experiments by Martin (1971) andSlamecka (1972) have produced ratherstriking demonstrations of the role that con-trol elements play in determining access tostored information. While the two experi-ments were different in both conception andmethod, they did make one and the samepoint: certain stored items can be retrievedonly through the activation of certain otheritems. The latter can be regarded as con-trol elements for the former. In Martin'sexperiment, subjects learned groups of wordsconsisting of a three-word compound stimu-lus and a single response word. A subse-quent recall test showed that one stimulusword as a cue provided access to another oneonly through the response word. InSlamecka's study, subjects learned lists ofwords belonging to conceptual categories.In the recall test, one item in a given categoryprovided access to other items in the samecategory only through the control element ofthe category label. These experiments, aswell as those of Johnson's (1970) on chunk-ing, do not only illustrate the part that con-trol elements and codes play in retrieval, butthey also point to the important distinctionbetween experimenter-designated retrievalcues and control elements. The two are notalways the same.

Although the data from our own experi-ments can be interpreted to reflect the roleof input cues as control elements, this in-terpretation fails to tell us exactly what de-termined the identity of control elements in

our situation. Why were the input cues, andnot the target items, control elements; whynot both?

Still another tentative interpretation of ourfindings could be advanced under the labelof the hypothesis of complementary informa-tion. It can be briefly stated as follows.The address of an event stored in episodicmemory is essentially temporal, defined interms of the event's temporal relations toother experienced events. The specificationof the address is frequently sufficient for thehomunculus to read out the contents of thataddress. Failure of retrieval of desired in-formation in a situation in which only thetemporal address is given may sometimescome about because the information storedat that address is fuzzy or incomplete. Aneffective retrieval cue then is one that pro-vides the missing information. Thus, theretrieval information complements the in-formation available in the trace and permitsthe homunculus to read the trace withoutdifficulty. According to this formulation, acopy of the list cue presented in the finalcued tests in our experiments did containinformation that was lacking in the storedtraces, while extralist cues and copies of tar-get items did not. The complementary in-formation hypothesis does not say why thelist cue, and not the target item, carried thenecessary additional information, and there-fore it is as incomplete as all others we haveconsidered.

Encoding Specificity

These tentative interpretations of the rec-ognition failure we have briefly consideredhave not explained the phenomenon, but theyhave suggested questions that may providesome guidance and direction to future re-search. At the present time, it is possibleonly to point once more to the encodingspecificity principle as a general answer tothe questions. What produced the asym-metry of associations between list cues andtarget items in our experiments ? What de-termined the functional memory units andtheir relations to the nominal units? Whywas one of the two nominal units embeddedin the whole complex more readily recog-nized when it appeared alone than was the


other ? Why did the list cue serve the func-tion of a control element or code of the storedcue-target trace while the target item didnot? Why was the information containedin the input cue sufficient to complementthe stored information while that in theextralist cue and the copy of the target itemwas not?

A general answer to all these questions isprovided by the encoding specificity princi-ple : Specific encoding operations performedon what is perceived determine what isstored, and what is stored determines whatretrieval cues are effective in providing accessto what is stored.*

In our experiments, encoding of targetwords was influenced by the list cues presentat input and by the subjects' expectationsthat they would be tested with those cues.But it was influenced by other things as well,factors that cannot as yet be adequatelyidentified or labeled. We have referred tothe totality of conditions determining theencoding of a perceived item as its cognitiveenvironment (Tulving & Thomson, 1971),and we think of encoding operations as somesort of an interaction between the perceptualinput and its cognitive environment. Theterms are ill defined, and the concepts do notexplain too much at this time. Yet theyserve to remind us that something else be-sides the properties of a presented item de-termines how well the item is rememberedand that an important research problem is tofind out what this something else is and howit works. Specific encoding of input mate-rials in our studies was responsible for asym-metry of associations, embeddedness, controlelements, and the nature of effective retrievalinformation. The same material in othersituations, however, in a different cognitive

* Although we have emphasized the importanceof encoding conditions at input, we do not wish toimply that once an event has been encoded its tracedoes not undergo further changes. Among manypossible modifications of stored information, thoseproduced through active recoding of the traceare most relevant to determining its subsequentretrievability in different retrieval environments.For our present purposes, encoding processes areconsidered always to subsume recoding processesthat occur prior to the act of explicit retrieval ofinterest.

environment, could have been encoded inways in which access routes to the storedinformation would have been different.

It has been known for a long time, ofcourse, that how well a thing is remembereddoes not depend only on what it is, but alsoon how it is stored in memory. Ancientorators knew it when they used the methodof loci to ensure that they would know theright thing to say at the right time (Yates,1966), Gestalt psychologists knew it whenthey kept telling others that the propertiesof an element depended on the company theelement kept in a larger whole, Bartlett(1932, p. 188) emphasized the importance ofthe "conditions of the prior perception" indetermining recall and recognition, and along tradition of research on incidental andintentional learning has converged on theconclusions that what is learned depends onwhat happens in the learning situation (Post-man, 1964).

In recent research the effects of encodingoperations on what is stored, the memorytrace, have been studied in a variety of set-tings and using a number of different tech-niques and paradigms (Craik & Lockhart,1972). One method has been to change thecontext of an item. It has been found thatrepetition of a to-be-remembered item mayor may not facilitate its recall, dependingupon its intralist context (e.g., Asch, 1969;Murdock & Babick, 1961), on its semanticinterpretation as biased by context (e.g.,Bobrow, 1970), and on its membership inthe same or different higher order unit (e.g.,Bower et al., 1969). Similarly, recognitionof a previously seen list word has been shownto be influenced by its presentation and testcontexts (e.g., DaPolito, Barker, & Wiant,1972; Light & Carter-Sobell, 1970; Thom-son, 1972; Tulving & Thomson, 1971;Winograd & Conn, 1971).

In another type of experiment, encodinghas been manipulated by asking the subjectsto do different things with the material whenit is presented. Thus, for instance, Handler(1967a) had one group of subjects sortwords into conceptual categories while an-other group was exposed to the same wordsunder instructions to study and rememberthem. In a subsequent free-recall test, both


groups did equally well. In Hyde andJenkins' (1969) experiment, subjects in onegroup made judgments about semantic prop-erties of words, while in another group theystudied the same set of words in expectationof a recall test. Both groups recalled thesame number of words and did considerablybetter than a third group that had madejudgments about graphemic properties ofwords prior to the test. A particularly ef-fective manipulation of input in terms of en-coding operations performed on the presenteditems has been described by Craik (1973).Words were presented tachistoscopically, al-ways for the same short duration, and thesubjects were instructed to answer differentquestions about each presented word, suchas, "Is it printed in capital or lower-caseletters?" or "Does it belong to the categoryof fruits?" Large differences in subsequentrecall of these words were observed, depend-ing upon the encoding operation performedat input. These and other similar studies(e.g., Gardiner, 1972; Johnston & Jenkins,1971; Paivio & Csapo, 1972; Schulman,1971) thus provide convincing evidenceabout the important role that encoding opera-tions play in determining subsequent rc-trievability of perceived items.

Differences in recall and recognition ofidentical material stored under different en-coding conditions do not simply reflect es-tablished differences in "strength" of traces.The concept of trace strength makes littlesense and has little value if: (a) strengthmust be estimated from observed levels ofrecall or recognition and ( b ) these observedlevels vary with specific conditions of re-trieval. A number of recent experimentssupplement common sense in demonstratingthat recall and recognition of items storedunder identical encoding conditions are in-fluenced, sometimes greatly influenced, bythe nature of information present in the re-trieval environment. These experiments(e.g., Bahrick, 1969; Light, 1972; McLeoclet al., 1971; Tulving, 1968b; Tulving &Pearlstone, 1966; Tulving & Psotka, 1971;Winograd & Conn, 1971) clearly suggestthat a stored trace of an item is more ac-cessible through certain cues than others.Memory traces may be said to vary in

strength, or quality, or durability, but moreimportantly they vary in the specificity ofcode they carry as to the effectiveness ofvarious kinds of retrieval information thatgovern the recovery of the stored informa-tion (Earhard, 1969).

Thus, research has shown that it is pos-sible to hold constant the to-be-remembereditem and observe large differences in its re-call and recognition depending upon its en-coding conditions, and that it is possible tohold constant the encoding conditions of theitem and observe large differences in its recalland recognition depending upon retrievalconditions. These two basic sources of vari-ability of recall, encoding and retrieval con-ditions, usually interact in the sense that acue effective in one situation may or may notbe effective in another. The new data wedescribed in this paper strongly imply suchan interaction, and other data already in theliterature in fact demonstrate it (e.g., Ear-hard, 1969; Frost, 1972; Ghatala & Hurlbut,1973; Lauer & Battig, 1972; Mikula, 1971;Thomson, 1972; Thomson & Tulving, 1970;Tulving & Osier, 1968; Tulving & Thomson,1971; Tversky, in press; Winograd et al.,1971).

All these data suggest that the effective-ness of a particular cue depends on how theto-be-retrieved item was encoded at input.The recognition failure of recallable wordsrepresents an extreme case of the generalprinciple that encoding determines the trace,and the trace determines the effectiveness ofretrieval cues. The trace itself is simplythe link between encoding conditions and theretrieval environment and, from the pointof view of psychological analyses of memory,need have no more reality than is impliedby the relation between encoding and re-trieval.

The formulation of the encoding specificityprinciple is so general that it covers allknown phenomena of episodic memory andretrieval and in a weak sense provides anunderstanding of them. More detailed con-ceptual analyses of the relation between en-coding and retrieval processes may also befruitfully undertaken within the same generalparadigm.


REFERENCES

ANDERSON, J. R., & BOWER, G. H. Recognitionand retrieval processes in free recall. Psy-chological Review, 1972, 79, 97-123.

ASCH, S. E. Reformulation of the problem ofassociation. American Psychologist, 1969, 24,92-102.

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. Vol. 2. New York: Academic Press, 1968.

BAHRICK, H. P. Measurement of memory byprompted recall. Journal of Experimental Psy-chology, 1969, 79, 213-219.

BAHRICK, H. P. A two-phase model for promptedrecall. Psychological Review, 1970, 77, 215-222.

BAHRICK, H. P., & BAHRICK, P. O. A re-examina-tion of the interrelations among measures ofretention. Quarterly Journal of ExperimentalPsychology, 1964, 16, 318-324.

BARTLF.TT, F. C. Remembering: A study in experi-mental and social psychology. Cambridge, Eng-land : University Press, 1932.

BILODEAU, E. A. Experimental interference withprimary associates and their subsequent recoverywith rest. Journal of Experimental Psychology,1967, 73, 328-332.

BILODEAU, E. A., & BUCK, K. A. Courses of mis-recall over long-term retention intervals as re-lated to strength of preexperimental habits ofword association. Psychological Reports, 1965,16(Monogr. Suppl. 6).

BILODEAU, K. A., & HOWELL, D. C. Free associa-tion norms. (Catalog No. D210.2:F87) Wash-ington, D. C.: U.S. Government Printing Office,1965.

ROBROW, S. A. Memory for words in sentences.Journal of Verbal Learning and Verbal Be-havior, 1970, 9, 363-372.

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

BOWER, G. H. Organizational factors in memory.Cognitive Psychology, 1970, 1, 18-46.

BOWER, G. H. Stimulus-sampling theory of en-coding variability. In A. W. Melton & E.Martin (Eds.), Coding processes in humanmemory. Washington, D. C.: Winston, 1972.

BOWER, G. H., CLARK, M. C., LESGOLD, A. M., &WINZENZ, D. Hierarchical retrieval schemes inrecall of categorized word lists. Journal ofVerbal Learning and Verbal Behavior, 1969, 8,323-343.

BREGMAN, A. S. Forgetting curves with semantic,phonetic, graphic, and contiguity cues. Journalof Experimental Psychology, 1968, 78, 539-546.

BROADBENT, D. E. In defence of empirical psy-chology. London: Methuen, 1973.

CRAIK, F. I. M. A "levels of analysis" view ofmemory. In P. Pliner, L. Krames, & T. M.

Alloway (Eds.), Communication and affect:Language and thought. New York: AcademicPress, 1973.

CRAIK, F. I. M., & LOCKHART, R. S. Levels ofprocessing: A framework for memory research.Journal of Verbal Learning and Verbal Be-havior, 1972, 11, 671-684.

CRAMER, P. Magnitude and selectivity as in-dependent factors in semantic generalization.Journal of Verbal Learning and Verbal Be-havior, 1970, 9, 509-524.

DAPOLITO, F., BARKER, D., & WIANT, J. Theeffects of contextual changes on the componentrecognition. American Journal of Psychology,1972, 85, 431-440.

EARHARD, M. Storage and retrieval of wordsencoded in memory. Journal of ExperimentalPsychology, 1969, 80, 412-418.

KSTES, W. K. An associative basis for codingand organization. In A. W. Melton & E. Martin(Eds.), Coding processes in human memory.Washington, D. C.: Winston, 1972.

Fox, P. W., BLICK, K. A., & BILODEAU, E. A.Stimulation and prediction of verbal recall andmisrecall. Journal of Experimental Psychologv,1964, 68, 321-322.

Fox, P. W., & DAHL, P. R. Aided retrieval ofpreviously unrecalled information. Journal ofExperimental Psychology, 1971, 88, 349-353.

FREUND, J. S., & UNDERWOOD, B. J. Restrictedassociates as cues in free recall. Journal ofVerbal Learning ami Verbal Behavior, 1970, 9,136-141.

FROST, N. Encoding and retrieval in visualmemory tasks. Journal of Experimental Psy-chology, 1972, 95, 317-326.

GARDINER, J. M. Studies of word retrieval inhuman memory. Unpublished doctoral disserta-tion, University of London, 1972.

GHATALA, E. S., & HURLBUT, N. L. Effectivenessof acoustic and conceptual retrieval cues inmemory for words at two grade levels. Journalof Educational Psychology, 1973, 64, 347-352.

HARROWER, M. R. Organization in higher mentalprocesses. Psychologische Forschung, 1933, 17,56-120.

HOLLINGWORTH, H. L. Psychology: Its facts andprinciples. New York: Appleton, 1928.

HUDSON, R. L., & AUSTIN, J. B. Effect of con-text and category name on the recall of cate-gorized word lists. Journal of ExperimentalPsychology, 1970, 86, 43-47.

HYDE, T. S., & JF.NKINS, J. J. Differential effectsof incidental tasks on the organization of recallof a list of highly associated words. Journal ofExperimental Psychology, 1969, 82, 472-481.

JAMES, W. Principles of psychology. New York:Holt, 1890.

JOHNSON, N. F. The role of chunking andorganization in the process of recall. In G. H.Bower (Ed.), The psychology of learning andmotivation. Vol. 4. New York: AcademicPress, 1970.


JOHNSTON, C. D., & JENKINS, J. J. Two moreincidental tasks that differentially affect as-sociative clustering in recall. Journal of Experi-mental Psychology, 1971, 89, 92-95.

KINTSCH, W. Models for free recall and recog-nition. In D. A. Norman (Ed.), Models ofhuman memory. New York: Academic Press,1970.

KOFFKA, K. Principles of Gestalt psychology.New York: Harcourt, Brace & World, 193S.

KOHLER, W. Gestalt psychology. New York:Liveright, 1947.

LACHMAN, R., & FIELD, W. H. Recognition andrecall of verbal material as a function of degreeof training. Psychonomic Science, 1965, 2, 225-226.

LACHMAN, R., LAUGHERY, K. R., & FIELD, W. H.Recognition and recall of high frequency wordsfollowing serial learning. Psychonomic Science,1966, 4, 225-226.

LAUER, P. A., & BATTIG, W. F. Free recall oftaxonomically and alphabetically organized wordlists as a function of storage and retrieval cues.Journal of Verbal Learning and Verbal Be-havior, 1972, 11, 333-342.

LIGHT, L. L. Homonyms and synonyms as re-trieval cues. Journal of Experimental Psy-chology, 1972, 96, 255-262.

LIGHT, L. L., & CARTER-SOBELL, L. Effects ofchanged semantic context on recognition memory.Journal of Verbal Learning and Verbal Be-havior, 1970, 9, 1-11.

MANDLER, G. Organization and memory. In K.W. Spence & J. T. Spence (Eds.), The psy-chology of learning and motivation. Vol. 1. NewYork: Academic Press, 1967. (a)

HANDLER, G. Verbal learning. In, New directionsof psychology, III. New York: Holt, Rinehart& Winston, 1967. (b)

MANDLER, G. Organization and recognition. InE. Tulving & W. Donaldson (Eds.), Organiza-tion of memory. New York: Academic Press,1972.

MARTIN, E. Stimulus meaningfulness and paired-associate transfer: An encoding variability hypo-thesis. Psychological Review, 1968, 75, 421-441.

MARTIN, E. Stimulus component independence.Journal of Verbal Learning and Verbal Behavior,1971, 10, 715-721.

MARTIN, E. Stimulus encoding in learning andtransfer. In A. W. Melton & E. Martin (Eds.),Coding processes in human memory. Washing-ton, D. C.: Winston, 1972.

McCoRMACK, P. D. Recognition memory :• Howcomplex a retrieval system? Canadian Journalof Psychology, 1972, 26, 19-41.

McLEOD, P. D., WILLIAMS, C. E., & BROADBENT,D. E. Free recall with assistance from oneand from two retrieval cues. British Journalof Psychology, 1971, 62, 59-65.

MELTON, A. W. Implications of short-termmemory for a general theory of memory.

Journal of Verbal Learning and Verbal Be-havior, 1963, 2, 1-21.

MELTON, A. W., & MARTIN, E. (Eds.). Codingprocesses in human memory. Washington, D. C.:Winston, 1972.

MIKULA, G. Der Einfluss mnemotechnischer Hilfenauf das Erlernen und Behalten verbalen Mate-rials. Psychologische Forschung, 1971, 34, 312-324.

MURDOCK, B. B. Human short-term memory.Potomac, Md.: Lawrence Earlbaum Associates,in press.

MURDOCH, B. B., JR., & BABICK, A. J. The effectof repetition on the retention of individualwords. American Journal of Psychology, 1961,74, 596-601.

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

NORMAN, D. A., & RUMELHART, D. E. A systemfor perception and memory. In D. A. Norman(Ed.), Models of human memory. New York:Academic Press, 1970.

PAIVIO, A., & CSAPO, K. Picture superiority infree recall: Imagery or dual coding? (ResearchBulletin No. 243) London: University ofWestern Ontario, Department of Psychology,1972.

POSNER, M. I., & WARREN, R. E. Traces, concepts,and conscious constructions. In A. W. Melton& E. Martin (Eds.), Coding processes in humanmemory. Washington, D. C.: Winston, 1972.

POSTMAN, L. Short-term memory and incidentallearning. In A. W. Melton (Ed.), Categoriesof human learning. New York: Academic Press,1964.

POSTMAN, L. A pragmatic view of organizationtheory. In E. Tulving & W. Donaldson (Eds.),Organization of memory. New York: AcademicPress, 1972.

POSTMAN, L., ADAMS, A., & PHILLIPS, L. W.Studies in incidental learning: II. The effects ofassociation value and of the method of testing.Journal of Experimental Psychology, 1955, 49,1-10.

PUFF, C. R. Clustering as a function of the se-quential organization of stimulus word lists.Journal of Verbal Learning and Verbal Be-havior, 1966, 5, 503-506.

RIEGEL, K. F. Free associative responses to the200 stimuli of the Michigan restricted associa-tion norms. (USPHS Tech. Rep. No. 8,Grant MH 07619) Ann Arbor: University ofMichigan, 1965.

SCHULMAN, A. J. Recognition memory fortargets from a scanned word list. British Jour-nal of Psychology, 1971, 62, 335-346.

SHIFFRIN, R. M. Memory search. In D. A.Norman (Ed.), Models of human memory. NewYork: Academic Press, 1970.

SHIFFRIN, R. M., & ATKINSON, R. C. Storage andretrieval processes in long-term memory. Psy-chological Review, 1969, 76, 179-193.


SLAMECKA, N. J. Testing for associative storagein multitrial free recall. Journal of Experi-mental Psychology, 1969, 81, 557-560.

SLAMECKA, N. J. The question of associativegrowth in the learning of categorized material.Journal of Verbal Learning and Verbal Be-havior, 1972, 11, 324-332.

THOMSON, D. M. Context effects in recognitionmemory. Journal of Verbal Learning andVerbal Behavior, 1972, 11, 497-511.

THOMSON, D. M., & TULVING, E. Associativeencoding and retrieval: Weak and strong cues.Journal of Experimental Psychology, 1970, 86,2SS-262.

TULVING, E. Theoretical issues in free recall. InT. R. Dixon & D. L. Horton (Eds.), Verbalbehavior and general behavior theory. Engle-wood Cliffs, N. J.: Prentice-Hall, 1968. (a)

TULVING, E. When is recall higher than recogni-tion? Psychonomic Science, 1968, 10, 53-54. (b)

TULVING, E. Episodic and semantic memory. InE. Tuiving & W. Donaldson (Eds.), Organisa-tion of memory. New York: Academic Press,1972.

TULVING, E., & OSLER, S. Effectiveness of retrievalcues in memory for words. Journal of Experi-mental Psychology, 1968, 77, 593-601.

TULVING, E., & PEARLSTONE, Z, Availabilityversus accessibility of information in memoryfor words. Journal of Verbal Learning andVerbal Behavior, 1966, S, 381-391.

TULVING, E., & PSOTKA, J. Retroactive inhibitionin free recall: Inaccessibility of ihformationavailable in the memory store. Journal ofExperimental Psychology, 1971, 87, 1-8.

TULVING, E., & THOMSON, D. M. Retrieval pro-cesses in recognition memory: Effects of as-

sociative context. Journal of Experimental Psy-chology, 1971, 87, 116-124.

TVERSKY, B. Encoding processes in recognitionand recall. Cognitive Psychology, in press.

UNDERWOOD, B. J. False recognition produced byimplicit verbal response. Journal of Experi-mental Psychology, 1965, 70, 122-129.

UNDERWOOD, B. J. Attributes of memory. Psy-chological Review, 1969, 76, 559-573.

UNDERWOOD, B. J. Are we overloading memory?In A. W. Melton & E. Martin, (Eds.), Codingprocesses in human memory. Washington, D. C.:Winston, 1972.

WATKINS, M. J. When is recall spectacularlyhigher than recognition? Journal of Experi-mental Psychology, 1973, in press.

WICKENS, D. D. Encoding categories of words:An empirical approach to meaning. Psychologi-cal Review, 1970, 77, 1-15.

WINOGRAD, E., & CONN, C. P. Evidence fromrecognition memory for specific encoding of un-modified homographs. Journal of Verbal Learn-ing and Verbal Behavior, 1971, 10, 702-706.

WINOGRAD, E., KARCHMER, M. A., & RUSSELL, I. S.Role of encoding unitization in cued recogni-tion memory. Journal of Verbal Learning andVerbal Behavior, 1971, 10, 199-206.

WOOD, G. Category names as cues for the recallof category instances. Psychonomic Science,1967, 9, 323-324.

YATES, F. A. The art of memory. Chicago: Uni-versity of Chicago Press, 1966.

YNTEMA, D. B., & TRASK, F. P. Recall as asearch process. Journal of Verbal Learningand Verbal Behavior, 1963, 2, 65-74.

(Received January 10, 1973)

ENCODING SPECIFICITY AND RETRIEVAL PROCESSES IN EPISODIC...

Documents

Transcript of ENCODING SPECIFICITY AND RETRIEVAL PROCESSES IN EPISODIC...