vol. psychological review - semantic scholar · 242 allan paivio call may be entirely verbal, but...

VOL. 76, No. 3 MAY 1969

PSYCHOLOGICAL REVIEW

MENTAL IMAGERY IN ASSOCIATIVE LEARNINGAND MEMORY1

ALLAN PAIVIO2

University of Western Ontario

Nonverbal imagery and verbal symbolic processes are considered in relationto associative learning and memory. These two hypothesized processes areoperationally distinguished in terms of stimulus attributes and experimentalprocedures designed to make them differentially available as associativemediators or memory codes. The availability of imagery is assumed to varydirectly with item concreteness or image-evoking (/) value, whereas verbalprocesses are presumably independent of concreteness but functionally linkedto meaningfulness (m) and codability. Stimulus characteristics are hypothe-sized to interact with mediation instructions, presentation rates, and type ofmemory task. Performance and subjective-report data resulting from experi-mental tests of the model indicated that imagery-concreteness is the mostpotent stimulus attribute yet identified among meaningful items, while m andother relevant attributes are relatively ineffective; that both processes canbe effectively manipulated by mediation instructions, but imagery is a "pre-ferred" mediator when at least one member of the pair is relatively concrete;and that the two mechanisms are differentially effective in sequential andnonsequential memory tasks. The findings substantiate the explanatory andheuristic value of the imagery concept.

This paper is concerned with the functional sists of sensory images awakened [1890, p.significance of nonverbal imagery and verbal 265]." As manifested in the "wax tablet"processes in associative meaning, mediation, model of memory, imagery was the prototypeand memory. As every psychologist knows, of stimulus trace theories (Gomulicki, 1953)imagery once played a prominent role in the and, as associative imagery, it was assumedinterpretation of such phenomena. It was to play a mediational role in mnemonic tech-widely regarded as the mental representative niques which originated long ago as a prac-of meaning—or of concrete meaning at least, tical art (see Yates, 1966). Despite theWilliam James, for example, suggested that criticisms that have been repeatedly directedthe static meaning of concrete words "con- at such views, common-sense experience con-

tinues to make their acceptance compelling.iThe research program on which this paper is Thus, in a recent discussion of the possible

based has been supported by grants from the rQje Q{ nonverbal schemata in verbal learning,National Research Council of Canada (Grant _, , ,1fv-,-, . e

APA-87) and the University of Western Ontario shepard (1966) writes,Research Fund. The author is grateful to R. C. jf j am now asked about the number of windowsGardner and P. C. Smythe of the University of ;n my house, I find that I must picture the house,Western Ontario, D. O. Hebb of McGill Univer- as viewed from different sides or from within dif-sity, and J. C. Yuille of the University of British ferent roOms, and then count the windows presentedColumbia for their comments on an earlier draft ;n these various mental images. No amount ofof this paper. purely verbal machinations would seem to suffice

2 Requests for reprints should be sent to Allan [p. 203].Paivio, Department of Psychology, Middlesex Col-lege, University of Western Ontario, London, On- The crucial point here is that the elicitingtario, Canada. question and the behavioral expression of re-

241

© 1969 by the American Psychological Association, Inc.

242 ALLAN PAIVIO

call may be entirely verbal, but the mediatingmechanism apparently is not.

It is precisely such views that were re-jected by Watson (1913). Partly on philo-sophical grounds and partly on the basis ofthe experimental evidence then available, heconcluded that mental images are mereghosts, without any functional significancewhatever. The mediating functions that hadbeen attributed to images in thought andmemory became the burden of implicit verbalresponses or their gestural substitutes (e.g.,see Goss, 1961; Watson, 1930, pp. 265-268)and until very recently this emphasis con-tinued to dominate research on problems ofmeaning and mediation processes in verballearning, memory, and language. The mean-ingfulness of verbal units has been dennedmost commonly in terms of such verbal asso-ciative measures as association value (Glaze,1928) and meaningfulness (m) (Noble,1952). Consistent with this definition, ef-fects of meaningfulness on associative learn-ing and memory have been interpreted interms of the availability of implicit associatesas mediators of stimulus-response (S-R)associations, as in the associative probability(Underwood & Schulz, 1960, p. 296) or"grapnel" theory (Glanzer, 1962). A simi-lar verbal emphasis has been generally ap-parent in research on mediated transfer andgeneralization, clustering in free recall, asso-ciation in language, and natural languagemediators (e.g., Bousfield, 1961; Cofer,1965; Deese, 1965; Jenkins, 1963; Martin,Boersma, & Cox, 1965; Montague, Adams,& Kiess, 1966; Underwood, 1965). Thepossible role of imagery in such phenomenahas been largely ignored.

Deese (1965) aptly summarizes the cur-rent view in the context of association theoryas follows:

The modern experimental psychologist works al-most exclusively with linguistic associations for thegood reason that these provide controllable materialfor his laboratory studies; he ignores the extraexistence of perceptual imagery. . . . Without neces-sarily denying either their reality or their impor-tance, [he] finds images difficult to manage inempirical study. Partly for this reason and partlyfor others, association theory in modern psychologyhas become a theory of the succession of elementsin verbal behavior [p. 4].

This is the classical behavioristic argument—imagery is subjective and inferential, wordsare objective and manageable. It constitutesa valid case against imagery if the interestsof contemporary psychologists remained atthe empirical level, but they do not. AsDeese goes on to say,

We study associations in order to make inferencesabout the nature of human thought. . . . The wholeof the current concern is with the associative prop-erties of explicit verbal behavior as a model for theimplicit verbal processes of thought [p. 4, italicsadded].

Here ends the empirical superiority of words,for implicit verbal processes are no less infer-ential than images. It may seem more directand parsimonious to infer a verbal mediatorwhen the response is verbal but this followsonly if one assumes a one-to-one relationbetween an associative reaction and the men-tal process that caused it. Such an assump-tion would be unwarranted for one can re-spond verbally to pictures as well as to wordsand so, by analogy, one's verbal responsecould just as logically be mediated by a"mental picture" as by "mental words." Bothare pure inferences and which one it is morelogical to infer in research concerned withsuch problems depends on the total set ofconditions in a given study, including suchvariables as stimulus attributes, instructionalsets, and introspective reports, not merely onthe nature of the overt response. That is nota problem, however, if one chooses to ignorethe extra existence of imagery.

The situation as portrayed is changing.On the one hand there is evidence of grow-ing dissatisfaction with traditional behavior-istic approaches to verbal learning and be-havior (Dixon & Horton, 1968). On theother, the negative attitude that Watsonfostered in regard to imagery has softenedand the concept has received considerableattention from theorists with diverse inter-ests (e.g., Beritoff, 1965; Hebb, 1966; Holt,1964; Mowrer, I960; Singer, 1966; Skinner,1953; Tomkins, 1962). More specifically inregard to verbal behavior, Osgood (e.g.,1961) has long argued in favor of nonverbalmediating processes and Staats (1961) hasindicated how imagery can be incorporatedinto such a model. Finally, active research

IMAGERY IN LEARNING AND MEMORY 243

on the possible role of mediating imagery inverbal learning and memory is proceedingapace (e.g., Bower, 1967; Bugelski, Kidd, &Segman, 1968; Paivio, 1965; Reese, 1965;Smith & Noble, 1965; Rohwer, 1966; Wood,1967), suggesting that systematic considera-tion of some of the relevant issues may betimely. The following is a review of suchproblems, organized around a particular em-pirical-theoretical approach. The main fea-tures of the approach are first outlinedbriefly, followed by a more elaborate treat-ment in the context of relevant research.

An Overview of the Approach

The possible role of nonverbal imagery willbe emphasized without denying the obviousrelevance of verbal mechanisms. Images areregarded as symbolic processes which arelinked developmentally to associative experi-ences involving concrete objects and events.In relation to language, they could be re-garded as conditioned sensations for whichappropriate words function as conditionedstimuli (cf. Mowrer, 1960; Sheffield, 1961;Skinner, 1957; Staats, 1961), or as con-structions that are actively generated by theindividual (e.g., Neisser, 1967; Piaget &Inhelder, 1966). Acceptance of a particularhypothesis is not essential here. Both imagesand verbal processes are operationally de-fined and the concern is with their functionalsignificance rather than their origin. Thetwo processes are assumed to function asalternative coding systems, or modes of sym-bolic representation (cf. Bruner, 1964).These may be relatively directly aroused inthe sense that a concrete object or eventevokes a perceptual trace, and a word, animplicit verbal response. Or, they may beassociatively aroused in the sense that anobject elicits its verbal label, or memoryimages of objects of the same class; and aword, if it is concrete, arouses an object im-age. In addition, chains of associative trans-formations involving either words or imagesor both (cf. Berlyne, 1965) presumably canoccur and serve a mediational function inverbal learning.

The mediational function and arousal ofimagery are theoretically coordinated to anabstract-concrete dimension of stimulus

meaning, which is defined in terms of direct-ness of sensory reference. For present pur-poses, it can be understood as extending fromabstract nouns to concrete nouns to picturesand objects, in increasing order of concrete-ness. The higher the concreteness of stimu-lus items, the more likely are they to evokesensory images that can function as media-tors of associative learning and memory.Verbal mediators, on the other hand, are notassumed to be functionally linked to con-creteness, but may be correlated with verbalassociative meaning as indexed by suchmeasures as association value and m. Inaddition, the arousal and mediational poten-tial of both symbolic modes presumably canbe influenced by instructional sets and otherexperimental procedures. Stimulus attributesand instructional sets thus may be regardedas different operations that define the postu-lated symbolic processes.

The research to be discussed in relationto the theory has involved both classes ofdefinitions, and inferences are also based onthe kinds of mediators subjects (Ss) reportusing in a learning situation. One series ofstudies is concerned with the relative effectsof the two kinds of stimulus meaning onassociative learning. Other studies involveboth stimulus attributes and instructionalsets, inferences concerning images and ver-bal processes being based on the interactiveeffects of these variables on performancescores and reported mediators. A final sec-tion of this paper extends the theoreticalviews and variables to memory tasks otherthan paired associates. The central questionthroughout is whether or not it is theoreti-cally necessary, or at least useful to postulateboth kinds of symbolic processes, imaginal aswell as verbal, to account for the observedeffects.

STIMULUS ATTRIBUTES AND MEDIATINGIMAGERY IN ASSOCIATE LEARNING

Memory techniques based on mediatingimagery form the point of departure. Nu-merous versions have been devised duringtheir long history (see, e.g., Hunter 1964;Yates, 1966), but the one that is analyticallymost relevant here involves the use of anumerical rhyme such as one-bun, two-shoe,

244 ALLAN PAIVIO

three-tree, four-door, and so on, togetherwith bizarre images of the objects named(see, e.g., Bugelski et al., 1968; Miller, Ga-lanter, & Pribram, 1960, pp. 134-138).Thus, to memorize a series of new objects,one imagines the first object, say a chair,inside a huge hamburger bun. The objectscan be recalled sequentially or in any ordergiven the number as a cue, which in turnreinstates the image; for example, "one"evokes the image of the bun and chair. Thistechnique changes what is essentially a free-recall situation into paired-associate (PA)learning in which the stimulus items arehighly familiar and highly differentiatedunits, and ostensibly makes use of perceptualimagery to mediate response retrieval. Thetechnique is apparently simple in practice.It is psychologically complex, however, in-volving multiple stages of symbolic trans-formation or coding, from words to imagesand back to words, together with the implicitassumption that memory for concrete objectsis somehow superior to memory for words(otherwise, why bother to make the trans-formation from words to concrete images atall?). The research described below hasbeen in part an exploration of the effectivevariables, assumptions, and complexities im-plicit in such a technique, but extended tomore conventional associative learning prob-lems. In particular, the use of word "pegs"and the imagery instructions appear to bethe crucial features of the technique, and bothhave been systematically investigated, first inthe context of standard PA learning andmore recently in relation to the mnemonictechnique in its pristine form.

The Conceptual Peg Hypothesis

The implication in regard to the wordpegs, one-bun, etc., is that their image-arousing potential may be an important vari-able independent of the instruction to makeuse of imagery, and that such word imagerymay be effective as a stimulus variable in PAlearning. This hypothesis was tested in aseries of experiments, first with adjective-noun pairs (Paivio, 1963) and later withnoun-noun pairs. The latter situation ismost directly relevant to the central issuesto be considered here and the hypothesis will

be described in that context. The argumentis that the stimulus member of a pair servesas a "conceptual peg" (a term first intro-duced, without reference to imagery, by Lam-bert & Paivio, 1956) to which its associateis hooked during learning trials when stimu-lus and response members are presentedtogether, and from which the response mem-ber can be retrieved on recall trials when thestimulus member is presented alone. On theassumption that imagery can serve a mediat-ing function, as the imagery memory tech-nique suggests, it follows that the ease oflearning the stimulus-response associationwill depend partly on the image-arousingcapacity of the individual nouns and of thestimulus member in particular. The imageryvalue of both stimulus and response wouldcontribute to the formation of a compoundimage, consisting of images evoked by theindividual items when the two are presentedtogether, thereby affecting the formation ofthe mediated association. On recall trials,however, when the stimulus is presentedalone, its image-arousing value would beparticularly important, for the stimulus mem-ber must serve as the cue that reinstitutesthe compound image from which the responsecomponent can be retrieved and receded asa word. The hypothesis leads to the predic-tion that a positive effect of noun imagerywill be greater on the stimulus side than onthe response side of pairs. Assuming, fur-ther, that concrete nouns such as "house"elicit images more readily than abstract nounssuch as "truth," it would be expected thatPA learning of nouns will be particularlyfacilitated when the stimulus noun is con-crete. To extend the metaphor, concretenouns should be more "solid" conceptual pegsthan abstract nouns.

The hypothesis was tested (Paivio, 1965)using 16 pairs of nouns, four of each possiblecombination of stimulus and response ab-stractness-concreteness. The specific predic-tion was that the four types of pairs wouldrank in the order, concrete-concrete, con-crete-abstract, abstract-concrete, and ab-stract-abstract, in increasing order of diffi-culty, thus reflecting the greater potency ofconcreteness on the stimulus side. Theresults were exactly as predicted, with ab-


stractness-concreteness of stimulus membersaccounting for eight times the variance at-tributable to response concreteness. Inter-pretations aside for the moment, it may benoted that this finding contrasts sharply withthe common empirical generalization (e.g.,Goss & Nodine, 1965, pp. 219-220; Noble& McNeely, 1957) that response factors arecrucial in PA learning whereas stimulus fac-tors are relatively unimportant. In the pres-ent instance, the stimulus attribute turnedout to have much more weight than theresponse attribute, indicating the heuristicvalue of the hypothesis that generated thenovel finding.

The assumption that concrete nouns ex-ceed abstract nouns in their image-evokingvalue was also confirmed by ratings obtainedfrom a different group of -Ss, and stimulusimagery scores correlated more highly thanresponse imagery scores with PA learning.These findings have been confirmed and ex-tended in subsequent experiments that weredesigned to clarify problems of interpretationarising from the initial study.

The Problem of the Effective WordAttribute

The identification of the effective meaningdimension is a major problem. Is nounimagery, as indexed by the ratings, "really"the effective empirical attribute or is it somecorrelated attribute such as m? If theobserved relations involving word imagerycan be attributed to some other empiricalvariable, the theoretical argument for stimu-lus-evoked imagery as the effective under-lying process is obviously weakened.

Imagery versus m. The most importantalternative to imagery is m, which has beenempirically related to learning in a varietyof situations and theoretically linked to ver-bal mediating processes. Concrete nouns gen-erally surpass abstract nouns in m as wellas rated imagery (Paivio, Yuille, & Madigan,1968), and m and imagery correlated equallywith learning in the Paivio (1965) experi-ment. It seemed unlikely that m could bethe effective variable because its effect isusually greater on the response rather thanthe stimulus side in PA learning (see Goss& Nodine, 1965; Underwood & Schulz,

1960), whereas the covarying noun attributesin the concreteness study were much moreeffective on the stimulus side. The superioreffect of response m has invariably been ob-tained with nonsense words as low-w items,however, and the possibility remained thatm is more potent on the stimulus side whenvaried entirely among familiar words.

The problem has been clarified by a seriesof experiments. Paivio, Yuille, and Smythe(1966) varied image evoking (/) and mindependently in different lists and foundthe usual stimulus effect for / despite a re-stricted range of variation, while m waseffective on both stimulus and response sideswhen varied within low-/, abstract, noun pairsbut not when varied within concrete, high-/pairs. Other studies (Paivio, 1967; Paivio& Olver, 1964) have shown that partialingout m had little effect on the positive corre-lation between learning scores and / but,with / similarly controlled, any effect of mwas reduced to zero. More rigorous controlover the range of variability in the contrast-ing variables has been achieved recentlyusing words from a sample of 925 nouns onwhich normative data are available on ratedconcreteness, rated /, and production m(Paivio, Yuille, & Madigan, 1968).3 Experi-ments involving these words (Paivio &Yuille, 1967; Smythe & Paivio, 1968) havedemonstrated conclusively that / is moreeffective than m when the two variables areindependently varied over an equivalentrange in terms of standard score units basedon the normative sample. In one of thesestudies (Smythe & Paivio, 1968), the onlyeffect of m was a slight negative one occur-

3 The size of the word pool makes it feasible tohold m constant while varying I, or vice versa, de-spite a positive correlation of .72 between thesevariables. Imagery and concreteness, which corre-late .83 in the sample, have generally been allowedto covary in the studies, although item selection hasusually been on the basis of the / scale because itappears to be superior to rated concreteness as apredictor of learning (Paivio, 1968a). Except forcertain unusual items (see Paivio, Yuille, & Madi-gan, 1968; Yuille, 1968), however, the two scalesappear largely to be alternative measures of thesame variable and the terms, noun imagery andconcreteness, are used here interchangeably exceptwhere an attempt is made to be particularly preciseabout the variable involved,

246 ALLAN PAIVIO

s2 6

List I LIstM

M-H H-t U-H L-L

ListIM

H-H H-L U-H L-L

FIG. 1. Paired-associate learning as a functionof high (H) and low (L) levels of stimulus andresponse / (List I) ; m (List M), and I-m co-varied (List IM). (From Paivio, Smythe, &Yuille, 1968.)

ring on the response side of pairs when thepairs consisted of low-/ items.

An established variable cannot be dis-missed lightly, however, so the question wasasked in one more way: Does m contributeat all to the observed effects when allowedto covary with I, as it did in the Paivio(1965) experiment? To answer this, Pai-vio, Smythe, and Yuille (1968) constructedthree mixed-attribute PA lists, one with /constant and stimulus and response m variedin a factorial design, another with m constantand 7 similarly varied, and a third in whichI and m covaried. The range of variationof the independent variable in the three listswas again roughly equivalent in standardscore units. The critical question waswhether the covarying attributes would bemore effective than / alone, thereby indicat-ing that m added something in the formercase. The results are shown in Figure 1.Separate analyses of the data for each listindicated again that m had a negative effect,recall being better for low-low pairs thanfor other pair types. The effects of 7 wereagain highly significant on both stimulus andresponse sides, with the former havinggreater weight. The covarying attributesalso had positive effects attributable pri-marily to the high recall scores for high-highpairs. In contrast with the 7 varied list, thelist with covarying attributes showed a sig-nificant interaction of stimulus and responseattributes, reflecting the relatively high recallof pairs in which both members are low 7and m. Thus, even when allowed to covarywith I, m apparently contributed a negativeeffect,

The effects of 7 have been demonstratedwith varied procedures, whereas the m effectsjust described always involved a mixed-listdesign. A final experiment was accordinglydone using homogeneous lists of the fourpossible S-R combinations of high and lowm, with 7 controlled. The results revealedno significant effects, although the trendagain was for somewhat better recall of low-low m pairs than of the other combinations.

The conclusion from these studies is clear.The relations between noun imagery and PAlearning not only can not be interpreted interms of m, but imagery is the more potentof the two empirical variables. The effectsof m have, in fact, varied from slightly posi-tive, to zero, to slightly negative in the dif-ferent experiments, which may be relevantto the "interference paradox" of associativeprobability theory (cf. Barnes & Schulz,1966), but not the interpretation of 7 effects.The significance of these findings extends be-yond the present context. A number ofstudies ostensibly concerned with effects ofm have involved nouns that obviously varyin concreteness (e.g., Mueller & Travers,1965; Saltz, 1967) and the observed effectscould be reinterpreted in terms of imageryrather than m. Furthermore, the conclusionthat m is relatively ineffective in PA learninginvolving real words extends to word classesother than nouns (Kanungo, 1968).

Other confounding semantic attributes.Although ruling out m, the above findingsdo not eliminate the possibility that someother unidentified correlate of 7 might beresponsible for the effects. This issue wasclarified by a factor-analytic study (Paivio,1968a) of 96 nouns for which mean scoreswere obtained on ease of learning and vari-ous semantic and associative characteristics.The learning variables were recall scores forthe nouns when they served as stimulusterms and when they served as responseterms in PA learning, as well as scores froma free-recall experiment. The other nounattributes, 27 in all, included several mea-sures of concreteness and imagery, and otherpotentially effective variables such as m,familiarity, distinctiveness, associative reac-tion time, semantic differential meaning di-mensions, and so on. The results showed


that only nine of the noun attributes corre-lated significantly with one or more of thelearning scores and, of these, rated / was thebest predictor of learning (the highest corre-lation was between stimulus-term I and PAscores), closely followed by the other indexesof imagery and concreteness. In addition,the imagery and concreteness variables de-fined a strong factor on which PA learningscores with the items as stimulus membersalso showed a substantial loading. Such animagery-concreteness factor has been re-cently confirmed by Frincke (1968), whofound it to be the best of a number of pre-dictors of free recall. The conclusion thatrated / defines the effective attribute betterthan rated concreteness has also been sub-stantiated (Yuille, 1968).

The caveat that causal relations cannot beinferred from correlational data obviouslyapplies fully to these findings. Nevertheless,it is apparent that concreteness-imagery is amajor dimension of word meaning—or, atleast noun meaning—which is noteworthybecause such a factor has not emerged in themultitude of studies that have been conductedon semantic dimensions (e.g., Osgood, Suci,& Tannenbaum, 1957). Moreover, the evi-dence indicates that this dimension is themost potent one yet identified among familiarwords.

The Imagery Hypothesis and ResponseAttributes

The plausibility of the imagery hypothesisgiven certain characteristics of the responsemembers of pairs can also be questioned. Ananalysis in terms of stimulus-elicited medi-ating imagery is relatively straightforwardand intuitively plausible when both membersof the pair can be assumed to evoke sensoryimages readily. It is less straightforwardwhen a concrete noun is paired with an ab-stract noun, but here too it can be seen how!a concrete stimulus might "prime" imaginalencoding of an abstract response. For exam-ple, "religion" as a response term might beencoded as a church, given a concrete stimu-lus as the cue. It is more difficult similarlyto interpret the finding (Yarmey & Paivio,1965) that noun concreteness is relatively

more effective on the stimulus side even whenthe other member of the pair is a nonsensesyllable. A mediational interpretation of theeffect apparently requires a theory of multi-ple stages of coding in which it is assumedthat a high-imagery stimulus term somehowincreases the probability that a nonsense-syllable response will be encoded into a mean-ingful word and then further translated intoconcrete imagery. It must be assumed fur-ther that decoding from the mediator back tothe nonsense syllable response is not diffi-cult, otherwise decoding errors would occur(Underwood & Erlebacher, 1965). Thisanalysis suggests that the effectiveness of anymediating imagery in such a case should be afunction of both the I value of the nounstimulus and the ease with which the non-sense syllable response can be encoded as aword.

Paivio and Madigan (1968) tested thehypothesis using high-/ and low-/ nouns aseither stimulus or response members fornonsense syllable associates, which in turnwere either high or low in association value.The hypothesis would be supported if thedifferential superiority of high-/ nouns on thestimulus side of pairs is greater with high-association-value than with low-association-value associates, since the former items morereadily suggest meaningful words that couldenter into image-mediated linkage with thehigh-/ stimuli. No such interaction shouldemerge with nouns as response members.The results were clearly as predicted for thefirst two of four study-test trials, althoughstimulus / was effective by Trial 4 even withlow-association-value (CCC) trigrams asresponses.

These results together with those previ-ously described for noun-noun pairs supportthe following conclusion: To the extent thata differential effect of noun concreteness or/ on the stimulus side is crucial to theimagery mediation theory, such mediation ismost probable when the response term is ameaningful word, although it may also beinvolved to some extent when the responseis a nonsense syllable provided that it readilysuggests a word. In addition, noun / seemsto be generally effective on both sides of pairsregardless of the nature of the associate, sug-

248 ALLAN PAIVIO

gesting that some nonmediational process,such as differentiation based on semantic dis-similarity or distinctiveness (cf. Goss, 1963;Saltz, 1963), may be involved. The latterhypothesis has been tested using associativeoverlap (Paivio, 1965) and rating scale data(Paivio, 1968a) as measures of distinctive-ness, with negative results, but the problemdeserves more systematic research for it is anattractive theoretical possibility that wordimagery facilitates item differentiation as wellas mediation.

A TWO-PROCESS THEORY OF MEANINGAND MEDIATION

Further important problems arising fromthe research concern the positive contribu-tion of verbal processes, and the provision ofmore direct evidence that mediation of anykind is involved in the observed effects.While it was never assumed that imageryalone determines the associative learning ofnouns, the relation between imaginal and ver-bal processes was not explicitly stated in theconceptual peg hypothesis. Verbal mecha-nisms are explicity postulated in the follow-ing analysis, which relates availability ofverbal associative processes as well as im-agery to the abstract-concrete dimension ofstimulus meaning. Mediation processes havebeen directly investigated in the context ofthe hypothesis, which, it should be noted, isin no way intended as a complete theory ofmeaning, but rather as a heuristic approachto what has been termed associative meaning(e.g., Deese, 1965). The hypothesis is thatconcrete terms such as "house" derive theirmeaning through association with concreteobjects and events as well as through associa-tion with other words, and thereby acquirethe capacity to evoke both nonverbal imagesand verbal processes as associative (mean-ing) reactions, which could function as alter-native coding systems affecting mediationand memory. Abstract terms such as"truth," on the other hand, derive theirmeaning largely through intraverbal experi-ences and more effectively arouse verbalassociative than imaginal processes. Such aview of associative meaning is not new (cf.James, 1890, p. 264; Staats, 1961), but itsimplications have not been stated or explored

in relation to association and learning. Itmay be noted that the hypothesis subsumesthe conceptual peg hypothesis in that anydifferential effects associated with abstractand concrete stimuli are still attributed to thegreater / capacity of the latter.

The implications of the hypothesis havebeen tested using 5V reports of associativestrategies in PA learning as evidence of thepostulated mediation processes, and by theuse of mediation instructions to encouragetheir arousal.

Evidence from Subjective Reports

The assessment of learning strategies usingsubjective reports has yielded evidence ofmediating imagery as well as verbal mecha-nisms (e.g., Bugelski, 1962; Reed, 1918;Underwood & Schulz, 1960). However, themost systematic recent approaches havestressed "natural language mediators" (e.g.,Martin et al., 1965; Montague et al., 1966)without reference to imagery, although it islikely that 5"s would have classified some oftheir mediators as nonverbal images if givena choice, as was done in the followingstudies.

The two-process theory of associativemeaning implies that S's will report frequentuse of both images and verbal mediators tolearn noun pairs in which at least the stimu-lus member is concrete, whereas verbal medi-ators should predominate in the case of ab-stract pairs. The results of a number ofexperiments have partly supported these ex-pectations. Paivio et al. (1966) used a post-learning questionnaire which described men-tal images and verbal mediators as possibleassociative strategies along with a no-media-tion category. The 5"s indicated for eachnoun pair which technique they had used,followed by a description of the specificmediator. A striking pattern of relationsemerged in which images were predomi-nantly reported as mediators for pairs inwhich both stimulus and response mem-bers were concrete, whereas verbal medi-ators predominated in the case of pairs inwhich both members were abstract. Further-more, learning scores were highest for pairsfor which imaginal mediators were reported,next highest for verbally mediated, and low-


est for unmediated pairs, suggesting thatimagery was the most effective mediator.

The associative strategies specified in thequestionnaire have been extended in subse-quent research to include rote repetition anda general "other" category. Of greatest in-terest theoretically, only the image, verbal,and repetition strategies are considered here.The data are generally consistent with thetwo-process theory of associative meaning,with the modification that images are eithermore available than words, or that 5"s havea preference for using (or at least reporting)imagery in the case of pairs in which at leastthe stimulus member is concrete and high 7.Verbal mediation reports have generallyshown smaller relations to item attributesthan in the initial study (Paivio et al., 1966).These points are illustrated by the Paivio,Smythe, and Yuille (1968) experiment, de-scribed earlier, in which noun / was variedin one list, m was varied in a second, and thetwo were covaried in a third. Figure 2 showsthe mean frequencies of imagery, verbal, andrepetition strategies for the four types ofpairs in the /-varied and w-varied lists.Reported imagery was strongly related tonoun I, particularly of stimulus members,whereas verbal mediators and repetitionshowed no such pattern. The findings weresimilar for the list (not shown in the figure)with I and m covaried. In the case of thelist in which only m was varied, however, itcan be seen that none of the reported strate-gies was significantly related to m. A corre-lational analysis involving item attributes,learning scores, and frequency of reportedmediators for individual items over all threelists confirmed the inference that the strong-est relations were those between I, reporteduse of imagery, and learning scores when theitems served as stimulus members of pairs.Furthermore, reported imagery correlatedsignificantly with learning even with stimu-lus I partialed out, increasing the plausibilityof the argument that associations were in factmediated by imagery. Although the causalsequence obviously remains equivocal in suchdata (alternative interpretations are dis-cussed in detail by Paivio et al., 1966; Paivio,Smythe, & Yuille, 1968), the observed rela-tions are remarkable, especially if one takes

3- Listl en Imagery^ Verbal•I Repetition

H-H H-L L-HPAIR TYPE

L-L

FIG. 2. Mean number of pairs for which imagery,verbal mediation, and repetition strategies werereported as a function of stimulus and response /(List I) and m (List M). (From Paivio, Smythe,& Yuille, 1968.)

the traditional behavioristic view that themediation process involved must be entirelyverbal. If this were so, why should verbalmediators fail to show strong relations toitem attributes and learning when 5s haveas much opportunity to report such mediatorsas they do imagery? Introspective reportsalone are unlikely to settle the issues, how-ever, and we turn to another approach.

Evidence from Research InvolvingInstructional Sets

Instructional sets have been used in aseries of studies to test deductions from thetwo-process theory. One prediction is thatthe latency of an associated image to a stimu-lus word will be longer when the stimulusword is abstract and low / than when it isconcrete and high / whereas verbal associa-tive latency will be less affected by variationin concreteness. To test this, Paivio (1966)had 5"s press a key either when a mentalimage or an implicit verbal associate occurredto a stimulus word. The results showed thatreaction times were longer to abstract than

250 ALLAN PAIVIO

concrete words under both the verbal andimaginal associative instructions but, as pre-dicted, the difference was much greater underthe imagery set. Yuille and Paivio (1967)obtained analogous results for noun pairs.The latency of discovery of imaginal medi-ators was slower for pairs with abstract,low-/ stimuli than for those with concrete,high-/ stimuli, but this difference did notoccur under a verbal mediation set.

According to a verbal mediation approachsuch as Glanzer and Clark's (1962) verballoop hypothesis, it might be argued that ab-stract nouns simply arouse a longer covertverbal response chain under imagery instruc-tions. Such an interpretation fails to accountfor the interactions between stimulus con-creteness and instructional set for, if imageryinstructions simply increase covert verbaliza-tion, why would they not increase imaginalreaction time to concrete words as much asto abstract? Direct evidence on the role ofverbalization in such a situation was inciden-tally obtained in a recent study (Simpson& Paivio, 1968), an aspect of which required5" to press a key when an image occurred toa stimulus word, with or without a subse-quent description of the image. The key-press latency increased significantly whenverbalization was required, but it did soequally for concrete and abstract stimuluswords. This finding indicates that differen-tial (covert) verbalization alone cannot ac-count for the interactions obtained in thereaction time studies involving both imageryand verbal sets.

The hypothesis was extended to PA learn-ing studies in which 5s were given instruc-tional sets to use verbal or imaginal medi-ators to learn lists of noun pairs varying inconcreteness. The predictions were that aset to use imagery would facilitate the learn-ing of pairs in which at least the stimulusmember is concrete but not pairs with ab-stract stimuli, whereas a verbal mediation setshould be equally effective with concrete orabstract items. Paivio and Yuille (1967)compared the two mediation set conditionswith a control condition in which 5s wereinstructed to use rote repetition and foundthat the imaginal and verbal sets producedmuch better learning than did the repetition

set, but the predicted interaction with con-creteness did not occur. That is, the imageryset was no less effective than the verbal evenwith abstract pairs. Yuille and Paivio(1968) attempted to induce stronger media-tion sets by having 5s first practice associat-ing pairs of nouns using images or verballinks, after which they learned another PAlist using the technique they had practiced.Both imaginal and verbal sets again facili-tated learning relative to a repetition controlcondition (more so when at least one memberof the pair was concrete), but the differentialeffects predicted for imaginal and verbal setsagain failed to appear. Moreover, the effectsof the set variable had disappeared by Trial 3whereas the effects of noun concreteness per-sisted throughout learning. A possible ex-planation suggested by mediation report data(Paivio & Yuille, 1967) was that associativestrategies are only partly controlled by theexperimental sets and that, over trials, 5sincreasingly revert to associative habitsaroused by the semantic characteristics of theto-be-learned items. They may abandon theuse of rote repetition generally, resorting toimagery in the case of concrete pairs and ver-bal mediators in the case of abstract pairs,and so on.

Paivio and Yuille (1969) tested the hy-pothesis using a trial-by-trial probe of asso-ciative strategies as a function of instructionalset and noun concreteness. Different groupswere given one of four instructional sets(Imagery, Verbal, Repetition, and No Set)and subgroups within each were questionedafter 1, 2, or 3 trials concerning the type ofstrategy they had used to learn each pair.The results confirmed and extended the pre-vious findings in that the verbal and imaginalmediation sets facilitated learning in com-parison with the No Set as well as theRepetition condition, and the effects of sethad essentially disappeared by Trial 3whereas concreteness remained highly effec-tive throughout. Some of the expectedchanges in the pattern of reported mediatorsover trials also occurred. The 5s given theimagery and verbal mediation sets most oftenreported having used the corresponding typeof mediator to learn the pairs, and this wastrue for all trials. In the case of the Repeti-


tion set, on the other hand, rote repetitionwas frequently reported after one learningtrial and decreased thereafter despite the factthat the two- and three-trial groups had beenreminded before each trial to use repetition.The pattern of reports suggested that verbalmediators to some extent replaced repetitionon the second trial whereas imagery wasclearly favored at Trial 3.

The effects were also qualified by the itemattribute. Repetition and verbal mediationreports were relatively little affected by con-creteness, although verbal mediators tendedto show a greater increase over trials forpairs in which both members are abstractthan for other pair types. By contrast,imagery reports were strongly affected byconcreteness in a manner closely resemblingits effect on learning. The parallel effects onreported imagery and recall scores as a func-tion of concreteness and trials are shown inFigure 3. In addition to the identical rank-ing of the four types of pairs on both re-sponse measures, it can be seen that concrete-abstract pairs show the most striking in-

Cease over trials on each dependent variable.The findings consistently point to imagery

,™ a preferred and an effective mediationj strategy in the case of pairs in which at least

;' the stimulus is concrete and high /, whereas[ verbal mediation is less affected by concrete-\ ness. In these respects the data accord with

the two-process theory of associative mean-ing and mediation, but the results provideonly weak support for the suggestion thatimagery set ,Ss resort to verbal mediatorsin the case of abstract pairs. Such changescould actually have occurred but were notdetected by the mediation questionnaire. Al-ternatively, it may be that some 5"s can makeeffective use of imagery even in the case ofabstract pairs. Support for such an interpre-tation was obtained by MacDonald (1967)in a study of the retention of imaginal andverbal mediators and recall of PAs. The 5sinstructed to use images frequently reportedthe use of a concrete image to symbolize oneor both members of abstract pairs (e.g., "boyscout" for the pair, chance-deed) and themnemonic effectiveness of such mediatorswas evidenced by high recall of the pair onlywhen the mediator was correctly recalled

003eci•5

Z 1

zUJ

• IREPORTEO IMA6ERYI 6

1 2TRIAL

2TRIAL

FIG. 3. Mean number of pairs for which imaginalmediators were reported and mean number recalledas a function of stimulus-response concreteness andtrials. (Adapted from Paivio & Yuille, 1969.)

(cf. Montague et al., 1966). Still, it shouldbe relatively more difficult to generate appro-priate mediating images for abstract than forconcrete pairs, as indicated by the mediationlatency data, whereas verbal mediators shouldnot be similarly affected. A further alterna-tive is that verbal mediators, although easilydiscovered, are relatively ineffective unlessaccompanied by imagery (Yuille & Paivio,1968). This view is not likely to find manysupporters, but it is a logical alternative thatshould not be lightly dismissed, particularlysince verbal mechanisms were also found tobe deficient as the basis of effective m.

Other investigators have contributed rele-vant data in the area but have not clarifiedthe theoretical issues. Imagery mnemonicinstructions have produced effects of varyingpotency (e.g., Bower, 1967; Bugelski et al.,1968; Seibel, Lockhart, & Taschman, 1967;Smith & Noble, 1965), possibly becausewidely different procedures have been em-ployed in regard to presentation rates, char-acteristics of the items, and so on. Hulickaand Grossman (1967) found both imaginaland verbal mediation instructions to be effec-tive with concrete noun pairs, but imageryhad the stronger effect. Wood (1967), whoalso found both types of sets to be facilitating,tested the prediction of an interaction be-tween type of set and noun concreteness, withnegative results, although the differencestended to be in the expected direction. Yar-mey and Csapo (1968) found support for thetwo-process hypothesis in that 5s given animagery set did more poorly than those givena verbal set when the pairs were abstract, but

252 ALLAN PAIVIO

not when they were concrete. They ranintact groups under the different experimen-tal conditions, however, and the possibilityof group sampling error cannot be ruled outeven though the results are fully consistentwith predictions from the theory.

Concreteness, Imagery, and the RhymeMnemonic System

Coming around full circle to generalizefrom the data thus far presented to the "one-bun" technique which provided the frame-work for the analysis, it would seem reason-able to conclude that the / value of therhyming peg words is indeed a most impor-tant feature of the mnemonic system. Theusefulness of the instructions to use imageryhas also been supported, although this factorappears to account for less of the variance inrecall scores than does stimulus concretenesswhen both are varied. Neither conclusion isfully warranted by the evidence presented.The effect of the concreteness of the pegwords was not tested by any of the PA learn-ing studies involving noun pairs inasmuch asthe peg words are not explicitly presented asstimuli in the mnemonic system as they arein PA learning. Instead, the recall cues inthe technique are numerals, and the pegs,together with any images they may arouse,must be implicitly supplied by 5.

The effects of the imagery instructions alsohave been generally investigated using nounPAs rather than the one-bun technique, asin the pioneering study by Wallace, Turner,and Perkins (1957) and the more recentresearch described above. Wood (1967)recognized that the explicit presentation ofpeg words during learning and recall trialsdeparted from the mnemonic system, andthat generalizations from his own results maytherefore be limited. He nevertheless rea-soned that the procedure is justified inas-much as it should maximize the effectivenessof the mnemonic system. It will be seen,however, that precisely this conclusion isunwarranted.

Where the mnemonic system has beendirectly investigated, the adequacy of thecontrol for the mnemonic instructions is inquestion. Bugelski et al. (1968) found muchhigher recall after 5s had been taught the

imagery mnemonic technique than underprior control conditions. A standard no-mnemonic control group and a rhyme controlgroup that learned the rhyme but was notinformed of its mnemonic function showedno such facilitation. Unfortunately, the ef-fectiveness of the imagery component of themnemonic instructions was left indeterminatesince the study did not include a rhyme con-trol group that had been given mnemonicinstructions without being told to use images.

Such a control was included in a study(Paivio, 1968b) which directly tested theeffect of the imagery instructions and theconcreteness of the mnemonic rhyme in theone-bun technique. The 5s were presentedone study and test trial with each of two10-item lists of concrete nouns. The condi-tions for the first list excluded prior mne-monic instructions and were identical for all5s. The items were presented auditoriallyat a 4-second rate, the to-be-recalled nounsbeing preceded by the numerals 1-10. Onthe test trial, the numerals alone were readin random order and 5" attempted to recall thecorresponding items. The second list waspreceded by mnemonic instructions, with orwithout reference to the use of imagery, andtraining on either a concrete or an abstractrhyme. The concrete rhyme consisted of thefamiliar one-bun, two-shoe device, in whichthe "peg" words bun, shoe, etc., are concreteand relatively high /. The abstract rhymeincluded peg words such as fun, true, free,bore, etc., all of which have lower I valuesthan do the corresponding concrete pegs.The 5s given the imagery set were instructedto use mental images along with the rhymein order to recall the items. The 5s notgiven the imagery instructions were in-structed to recall the list by saying to them-selves the rhyming words along with theto-be-remembered item, for example, one-bun-pencil. Following the instructions, all5s learned the appropriate rhyme, afterwhich they were presented one study-testtrial with the second list of nouns.

The results were quite unequivocal. Sig-nificant main effects obtained for controlversus mnemonic conditions and for no-imagery versus imagery instructions werecompletely qualified by a highly significant


interaction of the two variables, presentedin Figure 4. It can be seen that recall wascomparable for all groups on the first no-mnemonic list, and that recall increased dra-matically on the second list for Ss given theimagery set regardless of whether the mne-monic rhyme was concrete or abstract. Themnemonic instructions without imagery hadno beneficial effect—in fact, recall tended tobe lower under the mnemonic than underthe control condition.4

The unexpected outcome was that con-creteness showed no effects approaching sig-nificance. The only suggestion of an effectwas that the reduction in recall for the secondlist was significant on a post hoc test forthe no-imagery abstract-rhyme group, buttheir performance on the second list did notdiffer significantly from that of the concrete-rhyme group. These findings contrast withthose obtained using the standard PA pro-cedure, where imagery-concreteness has con-sistently been highly effective with or withoutthe addition of imagery instructions. In-deed, as already mentioned, item / accountsfor more of the variance in recall scores thandoes the instructional set when both areincluded in the design. The relative effectsof the two variables are completely reversed,therefore, in the context of the mnemonicsystem.

How are the contrasting effects to be ex-plained? They can be interpreted in termsof the relative weights of the cues that deter-mine the arousal of mediating imagery. Inthe case of the PA studies, the stimulus pegsare explicitly presented on both study andrecall trials, and their image-evoking capacityis more potent than instructional sets indetermining mediation strategies and re-sponse recall, as indicated by the mediationreports and performance data in the studiesdescribed above (e.g., Paivio & Yuille,1969). In the mnemonic system, however,the pegs are implicit and even ones that areconcrete cannot function as effective cues forthe arousal of mediating images. The con-tribution of concreteness as a variable is ac-

4 A further control group that was merely pre-sented two successive lists without mnemonic in-structions showed no practice effect, replicatingBugelski et al. (1968).

§

CONTROL MNEMONICLEARNING CONDITIONS

FIG. 4. Mean number of items recalled in ordinalposition under control conditions and under mne-monic conditions involving image (I) and no image(NI) instructions and concrete (C) and abstract(A) peg words. (From Paivio, 1968b.)

cordingly depressed while that of the im-agery instructions is relatively enhanced.

The analysis suggests that the discoveryand utilization of mediating images shouldbe more efficient under the standard PA pro-cedure, when both the concrete stimuli andimagery instructions are explicitly presented.A comparison of the effects of presentationrates in different experiments provides in-direct support for the deduction. Bugelskiet al. (1968) found the one-bun procedureto be facilitative at exposure rates of 4 ormore seconds, but not at 2 seconds, whereasGruber, Kulkin, and Schwartz (1965) foundthat imagery instructions facilitated recall ofword pairs at rates as fast as 1 second. Thepresence of the pairs together with the in-structional set apparently permitted fastermnemonic processing than did the rhymetechnique. The problem deserves furtherinvestigation, particularly since it appears tobe related to the more general issue of thenature of retrieval cues and the conditionsunder which they are effective (cf. Tulving& Osier, 1968).

It can be concluded from the Paivio(1968b) results that the imagery instructionsare indeed an effective component of themnemonic techniques. It might still be ar-gued that the instructions simply arouseeffective verbal mediators, although the con-trol instruction simply to associate the rhym-ing pegs with the to-be-remembered words

254 ALLAN PAIVIO

could be regarded as a kind of verbal media-tion set, and it proved quite ineffective. Bu-gelski et al. (1968) also reported informallythat 5s given sentence mediators did not doat all well, and BowerB has found suchmediators to be inferior to imagery. Never-theless, the problem is complex and requiresfurther research to determine the conditionsthat affect the potency of each type of (opera-tionally denned) mediator. A further alter-native to the imagery hypothesis is that themnemonic instructions simply increase moti-vation or interest (e.g., Seibel et al., 1967).To the extent that this is so, it could beinterpreted to mean that motivated 5"s makemore use of effective mediators (cf. McNulty,1966), or that mnemonic processing is effec-tive because it works (Gruber et al., 1965).In other instances, such as the experimenton the one-bun mnemonic described above,it is difficult to understand how the mereaddition of brief instructions to use imagerycould enhance the motivational effect of al-ready elaborate instructions and mnemonictraining sufficiently to produce the dramaticincrement in recall. This conclusion appliesat least as strongly to results obtained byBower (1967), who has used a variety oftechniques to manipulate imagery with dra-matic effects in memory tasks.

The upshot of the argument thus far isthat the alternative hypotheses are strainedto account for the effects attributed to theoperations that define imagery. This is notto deny the effectiveness of verbal mediatorsand other variables, but rather to affirm whathas long been denied, that imagery is aviable explanatory concept in associativelearning and memory.

SOME EMPIRICAL AND THEORETICALEXTENSIONS

The variables and theoretical views dis-cussed in relation to PA learning of nounshave been extended to include picture-wordcomparisons, learning situations other thanPAs, and further theoretical distinctions be-tween images and verbal symbolic processes.These are briefly considered in the remainderof the paper.

B G. H. Bower, personal communication, October,1967.

Picture-Word Comparisons in PA LearningThe abstract-concrete dimension of stimu-

lus meaning can be extended to includeobjects or pictures of objects at the concreteend of the scale, providing a further meansof investigating the conceptual peg hypothe-sis and the general contribution of nonverbalfactors in associative learning. Objects ortheir pictorial representations arouse non-verbal perceptual images directly and, to theextent that such images facilitate the forma-tion of associative connections with responsemembers, pictures should be even more ef-fective than concrete nouns, at least asstimulus members of pairs.

A number of experiments (e.g., Epstein,Rock, & Zuckerman, 1960; Rohwer, 1966)have shown that object-pairs or picture-pairsare easier to learn than pairs comprised oftheir concrete noun labels, and Iscoe andSemler (1964) found object-pairs to beeasier than pictures. Wimer and Lambert(1959) reported, further, that PAs with non-sense syllable responses were better learnedwhen the stimulus members were objectsrather than the names of the objects. Thesestudies did not compare objects and wordson stimulus and response sides, however,and such a comparison is necessary to testthe conceptual peg hypothesis. Paivio andYarmey (1966) obtained support for thehypothesis in a study in which pictures andtheir noun labels were factorially varied onstimulus and response sides of pairs. Similarfindings were obtained with young childrenas 5"s (Dilley & Paivio, 1968), at least withrespect to the stimulus side of pairs. As astimulus variable in PA learning, therefore,concreteness appears to function as a singleeffective dimension that extends from ab-stract nouns, to concrete nouns, to objects orpictures, in increasing order of effectiveness.Such an ordering of three levels of concrete-ness has been demonstrated within a singleexperiment by Csapo (1968) and Dominow-ski and Gadlin (1968). Since the stimuluseffect has been obtained without words beinginferior in m (Dominowski & Gadlin, 1968;Wimer & Lambert, 1959), it is difficult toexplain the superiority of objects and pic-tures over words in terms of verbal mecha-nisms alone. Differentiation has been pro-


posed as an explanation by Dominowski andGadlin (1968) and Wimer and Lambert(1959), whereas mediating imagery is im-plied by the conceptual peg hypothesis. Assuggested earlier in the case of noun I, per-haps "picturability" contributes to both dif-ferentiation and mediation.

The linear effect of the concreteness dimen-sion does not obtain on the response sideof pairs. Whereas learning has generallybeen easier when response nouns are con-crete rather than abstract, pictures as re-sponse members are not facilitative in com-parison with concrete nouns, possibly becausethe former involve a symbolic labeling ordecoding problem. That this may be thecase is indicated by the negative main effect

* of pictures as response terms in the case ofi young children (Dilley & Paivio, 1968),\ which was predicted from the hypothesis that\ they may experience greater difficulty than

adults in transforming the concrete memoryimage of a pictorial response item into anovert verbal response.

A number of investigators (e.g., Epsteinet al, 1960; Milgram, 1967; Reese, 1965;Rohwer, Lynch, Levin, & Suzuki, 1967)have compared pictorial and verbal (e.g.,sentence context) depictions of pair relations,with interesting results. For example, pairsare easier to learn when the context providedby either mode of representation makes agood "conceptual unit" of the pair than whenit does not (Epstein et al., 1960) ; and, in thecase of children, learning was facilitated bymoving as compared to stationary objects,and by verb phrases more than by preposi-tions or conjunctions (Rohwer, 1966). Al-though here, too, it has been difficult to teaseout the separate contributions of visualimagery and of verbal symbolic processes,and the investigators are generally cautiousin their interpretations (and sometimes nega-tive concerning the contribution of imagery,e.g., Milgram, 1967), the findings point tothe operation of both processes.

Extensions to Memory Tasks Otherthan PA Learning

The discussion thus far has centered onimagery as a stimulus variable and as amediator of associative connections in PA

learning, but imagery variables have beenfound effective in other memory-tasks as well.Following a brief review of the relevant find-ings, an attempt will be made to explainthem in terms of a theoretical model thatincorporates and extends the concepts al-ready discussed.

Recognition memory. Results from therelatively few studies that have been con-ducted on the problem suggest that recogni-tion memory increases directly as a functionof concreteness. Recognition scores arehigher for concrete than abstract nouns(Gorman, 1961; Jampolsky, 1950) evenwhen m is held constant (Olver, 1965),and pictures in turn are easier to recognizethan are their names or words generally(Fozard & Lapine, 1968; Jenkins, Neale, &Deno, 1967; Shepard, 1967). Taken to-gether, these findings indicate that recogni-tion memory increases from abstract words,to concrete words, to pictures." The differ-entiation hypothesis is~ obviously applicablehere in the sense that image-arousing valueor picturability may contribute to an item'sdistinctiveness, but direct supporting evi-dence is lacking.

Free recall. Several investigators (e.g.,Dukes & Bastian, 1966; Olver, 1965; Win-nick & Kressel, 1965) have found that freerecall also is higher for concrete nouns thanfor abstract nouns, while others (Bowers,1931; Paivio, 1967; Tulving, McNulty, &Ozier, 1965) have reported better recall fornouns rated high on measures of imagerythan for nouns rated low in imagery. Theseeffects cannot be interpreted in terms of m,which was controlled in several of thesestudies. Paivio, Yuille, and Rogers (1969)demonstrated further that noun / is signifi-cantly more effective than m in free recall, asit was shown to be in PA learning, whenthe two attributes are varied over an equiva-lent range.

The effective dimension again extends toconcrete objects. As early as 1894, Kirk-patrick found that free verbal recall washigher for familiar objects than for the namesof the objects, and this has been repeatedlyconfirmed in studies comparing either objectsor pictures with verbal stimuli (Calkins,1898; Ducharme & Fraisse, 1965; Lieberman

256 ALLAN PAIVIO

& Culpepper, 1965; Paivio, Rogers, &Smythe, 1968). The finding is of uncom-mon theoretical interest because the requiredresponse is verbal in both cases and the to-be-remembered name therefore should be morereadily available when the stimulus is a wordrather than a picture. Since it is not, verbalprocesses alone seem insufficient to accountfor the findings.

These results are open to alternative inter-pretations that have been suggested as ex-planations of free-recall learning (for adiscussion of these, see Tulving, 1968).According to one hypothesis, free recall isa function of the response strength or availa-bility of the items as independent units.According to the other, some kind of asso-ciative principle operates to reduce the mem-ory load involved by facilitating interitemorganization or clustering. With respect tonoun imagery, Tulving et al. (1965) foundthat the better recalled, vivid words werealso organized to a greater extent than lessvivid words, and accordingly suggested thehypothesis that "vividness or picturability isan important component of meaning of wordsthat affects the ease with which words canbe grouped into higher-order memory units[p. 250]." _

Such an interpretation of the superiorityof pictures is also supported by Scott's(1967) finding of higher clustering scoresfor objects than for words. On the otherhand, Paivio, Yuille, and Rogers (1969) didnot find greater clustering for the betterrecalled, high-/ words, and Frincke (1968)found noun imagery to be significantly re-lated to free recall with ease of clusteringheld constant. The higher recall for picturesobtained by Paivio, Rogers, and Smythe(1968) also could not be explained in termsof subjective organization. Thus, while theorganization hypothesis of the imagery-con-creteness effect in free recall has some sup-port, the independence hypothesis cannotbe rejected.

Serial learning and memory span. Animplication of the organization interpretationof the effect of imagery-concreteness in freerecall is that the effect should diminish if thelearning situation does not permit 6" to re-organize the input series into associated clus-

ters, as is the case in serial memory tasks.The finding that memory span does not differfor concrete and abstract nouns (Brener,1940) is consistent with such an interpreta-tion. The same might be expected in seriallearning, except that imagery could facilitateassociative chaining over trials just as itfacilitates PA learning. Recently, however,the validity of the chaining hypothesis hasbeen seriously questioned (e.g., Jensen &Rohwer, 1965) and, while it cannot be en-tirely rejected (cf. Postman & Stark, 1967;Saufley, 1967; Shuell & Keppel, 1967), itappears likely that serial learning is not theprototype of stimulus-response associationthat it was once thought to be. From thatviewpoint, therefore, as well as from theorganization hypothesis, imagery as an asso-ciative mechanism should be less potent inserial than in free-recall learning. With thisargument in mind, Paivio, Yuille, andRogers (1969) investigated serial learning asa function of / and m using the same lists thathad been used in the free-recall study re-ferred to above. Again, / was positivelyeffective, the only suggestion that it might berelatively less so than in free recall beingthat its effect did not significantly exceedthat of m. This could be taken as supportfor the view that imagery indeed facilitatesinteritem associations in serial learning, orit may mean that imagery somehow makesthe nouns more available as response units,or both.

Analogous information on picture-wordcomparisons in serial learning and memoryspan is generally lacking, but some recentdata will be presented in the context of thefollowing theoretical model, which deals withthe relation between the concreteness dimen-sion and performance in the various memorytasks that have been considered.

THE AVAILABILITY AND FUNCTIONSOF CONCRETE-IMAGE AND VERBAL

CODING SYSTEMS IN MEMORY

The theoretical analysis involves an as-sumption concerning the relative availabilityof concrete (visual) images and verbal mem-ory codes as a function of stimulus concrete-ness, and a hypothesis concerning the func-tional distinction between the two coding


systems. The availability assumption can beregarded either as a special case or an exten-sion of the two-process approach to associa-tive meaning described earlier. It differsfrom it in that the verbal code refers specifi-cally to an implicit labeling or naming re-sponse (like the representational responseof Bousfield, 1961) rather than to verbalassociates. Concrete imagery is used in thesame sense as previously to refer to associa-tive images, aroused in this case by picturesas well as words. In the case of pictures,it is important to note that the reference isnot to a short-duration perceptual trace of thestimulus (cf. Sperling, 1960), which shouldin no way differ whether the stimulus is aprinted word or a picture. Instead, thereference is to long-term memory represen-tations of objects, the evocation or generationof which is no less an associative processwhen the stimulus is a picture than when itis a word.

Given these definitions, the assumptionsconcerning availability are straightforward.Pictures, if familiar and unambiguously la-beled, readily evoke both concrete imageryand verbal coding, but the availability of thelatter is relatively lower because an extratransformation is involved, and would belower still if the pictures are unfamiliar. Theverbal code is directly and equally availablein the case of concrete and abstract wordsbut the former are more likely to evoke im-ages. Empirical support for and a roughquantification of the assumptions is pro-vided by reaction time data. Words canbe read faster than objects can be named(Fraisse, 1964), indicating the higher avail-ability of the verbal code in the formercase. As previously mentioned, the latencyof image arousal as indicated by a key pressis faster for concrete than abstract words(Paivio, 1966). A comparison of the reac-tion time means reported in the differentexperiments indicates further that imagearousal to words is slower than verbal codingof words or familiar pictures. No compara-ble data are available on image arousal topicture stimuli but it can be assumed thatthe latency would be no greater than thatinvolved in the implicit reading of a word.A ranking of the availability, or arousal

probability, of each code for the differentstimuli can be inferred from these data.Image arousal in the case of pictures and'verbal coding in the case of words have thehighest availability, the verbal code to pic-tures second, imagery to concrete words ^third, and images to abstract words fourth. >Thus, the summative availability of bothcodes is highest in the case of pictures, inter-mediate for concrete nouns, and lowest forabstract nouns.

The effect of concreteness on memory ishypothesized to be a direct function of theavailability of each code in that the appro-priate verbal responses can be retrieved fromeither. Their availability could contribute tounit retrieval in accordance with the inde-pendence hypothesis, or both processes mayfacilitate associative connections and theformation of higher order memory units.The picture, concrete word, abstract wordordering of free recall and recognition mem-ory in the above studies is thus interpretedto be a function of the differential availabilityof concrete imagery as a supplementarycoding system, which enhances the probabil-ity that items will be correctly retrieved onthe test trial.

The hypothesis concerning the functionaldistinction is related to assumed modalitydifferences in the nature of the representa-tional mechanism. Visual images are as-sumed to be functionally related to visualperception and are specialized for spatial rep-resentation—visual imagery is primarily aparallel processing system. It is not special-ized for serial processing unless linked to anintegrated (symbolic) response sequence,such as might be involved in imagining one-self moving along a familiar route, or sequen-tially "tracing" the outline of a familiarfigure, or unless the imagery is linked to asequentially organized verbal system, as inthe one-bun mnemonic technique. The ver-bal symbolic system, on the other hand, isassumed to be specialized for sequentialprocessing by virtue of the temporal natureof the auditory-motor speech system, al-though it undoubtedly functions also as anoperationally parallel system in that verbalunits can be processed independently of eachother (see Neisser, 1967). While the ra-

258 ALLAN PAIVIO

tionale just described has not always beenmade explicit, a spatial-serial distinction be-tween visual and verbal memory systems hasbeen previously suggested by others (for arelevant discussion, see Bryden, 1967). Oneimplication of the analysis is that the additionof imagery should have no effect in a taskwhich involves only verbal sequential proc-essing. Brener's (1940) finding of equiva-lent memory spans for concrete and abstractwords is consistent with this interpretation.

The theory was explicitly tested by Paivioand Csapo (1969), who attempted to varythe availability of the two memory codesalong with the necessity of sequential proc-essing in the memory task. Availability wasmanipulated by varying stimulus concrete-ness and the rate of presentation. Sincewords can be read faster than objects can benamed, the arousal of the verbal code pre-sumably can be prevented during input inthe case of pictorial (P) stimuli withouteliminating it in the case of concrete (CW)or abstract (AW) words by using a suffi-ciently fast presentation rate. That is, at a

MEMORY SPAN

O.

8

<Ul

SERIAL LEARNING

,-OA

FREE RECALL RECOGNITION

SLOW

PRESENTATION RATE

FIG. 5. Memory scores for pictures (P), con-crete words (C), and abstract words (A) in fourmemory tasks as a function of presentation rate.(From Paivio & Csapo, 1969.)

fast rate 5 could implicitly read a word with-out having time to label a picture.

Given the hypothesized distinction betweenparallel and sequential processing functionsand a discrete stimulus series without anyintrinsic serial order, it follows that memoryshould be poorest at a very fast rate for Pstimuli in tasks such as immediate memoryspan and serial learning, which require se-quential processing. Pictures should notsuffer even at a fast rate in free recall andrecognition memory tasks, however, sinceserial order information need not be retainedand the appropriate verbal labels presumablycan be retrieved from concrete memory im-ages after the stimulus series has been pre-sented. At a slower rate, where 5" can sup-ply a verbal label, P stimuli should not beinferior to words even in the serial memorytasks, and they would be expected to surpasswords in the free recall and recognition taskson the basis of previous findings (e.g., Pai-vio, Yuille, & Rogers, 1968; Shepard, 1966),presumably because the high availability onboth codes enhances the probability of theirretrieval. No difference would be expectedbetween CW and AW on any task at thefast rate, since only the verbal code can beutilized. At the slower rate, however, CWshould exceed AW in free recall, recognition,and serial learning because of the arousal ofassociative imagery which can enhance unitretrieval or the formation of higher orderunits, as suggested above, but not in imme-diate memory span which involves only ver-bal memory for serial order. A corollaryprediction is that variation in rate shouldhave the greatest effect generally on memoryfor P stimuli and the least effect on AW.

The experimental test of the predictionsinvolved the presentation of a nine-itemlist of P, CW, or AW stimuli over a seriesof trials at either a fast (5.3 items per sec-ond) or a slow rate (2 items per second),in one of the four memory tasks. The resultswere remarkably consistent with the pre-dictions. The rate effect was significant overall and a double interaction of Rate X Stimu-lus Attribute indicated, as predicted, that theeffect was clearly greatest for pictures andleast for abstract words. The double inter-action was further qualified by a triple inter-


action of Rate X Stimulus Attribute X Task,which is presented in Figure 5. A seriesof comparisons revealed the following effects.Consistent with predictions, memory spanand serial learning scores were significantlypoorest for pictures at the fast rate. At theslow rate, pictures remained inferior in mem-ory span, although none of the differencesbetween stimulus types was significant. Bycontrast, pictures, along with concrete words,were superior to abstract words in seriallearning, a complete crossover from the fastrate. The effects for free recall and recogni-tion memory were similar to each other anddiffered from the patterns for the serial mem-ory tasks. At the fast rate, none of thedifferences between stimulus types was sig-nificant within either task. At the slow rate,pictures were remembered best, abstractwords most poorly, and concrete words wereintermediate, as in previous free recall andrecognition experiments. Finally, the effectof presentation rate was significant for pic-tures on all tasks; for concrete words inmemory span, serial learning, and margin-ally in free recall; and, for abstract words,only in serial learning.

In summary, memory for pictures tendedgenerally to suffer at the fast rate but theeffect was significant only in the sequentialmemory tasks, presumably because the verbalcode essential to performance was leastavailable. The slow rate benefited picturememory most, presumably because bothmemory codes could be utilized. Memoryfor abstract words was least affected by rate,presumably because only (or mainly) theverbal code was involved.

The theoretical model obviously requiresfurther testing before its strengths and weak-nesses can be adequately assessed. However,it has already demonstrated considerableheuristic value and integrative potential asan extension of the two-process theory ofmediation discussed in relation to PA learn-ing. In addition, it makes contact with re-cent proposals by others concerning multiplemodes of symbolic representation and func-tional distinctions between them (e.g., Koen,1966; Taylor & Posner, 1968). Taylor andPosner, for example, concluded from theirexperimental results that the visual memory

code is organized to represent the environ-ment spatially, whereas the verbal memorycode is organized so that at any point intime items are retrieved in a fixed order.This proposal is obviously consistent with themodel presented here. Others (e.g., Brooks,1967; Dornbush, 1967; Glucksberg, Fisher,& Monty, 1967; Wallach & Averbach, 1955)have suggested functional distinctions basedon the sensory modalities to which the sym-bolic systems are presumably linked. Theircontributions must be taken into account inany future tests and modifications of themodel.

Further Empirical-Theoretical Extensionsof the Approach

The hypotheses and research reviewed inthis paper involved operational definitionsof imaginal and verbal coding systems interms of the semantic attributes of stimuli,and instructional sets. The approach can beextended to take into account individual dif-ferences in relevant symbolic habits, such asimagery ability and verbal associative skills.Successful prediction of interactions involv-ing the three classes of independent variables—stimulus attributes, experimentally ma-nipulated mediators, and individual differ-ences—would yield convergent evidence thattheir effects-are mediated by common inter-vening processes. The third approach islikely to prove difficult considering the rela-tive failure of individual differences to bepredictive of memory in the research on im-agery in the early part of the century, butencouraging evidence has been providedrecently by Kuhlman (1960), and Stewart(1965), who found that imagery ability canpredict performance in situates involvingpictorial and verbal material. On the de-pendent variable side, the approach can takeadvantage of physiological measures as pos-sible indicators of the processes aroused byimaginal and verbal associative tasks (e.g.,Paivio & Simpson, 1966; Simpson & Paivio,1968). It can be extended also to naturallanguage by investigating the effects of suchvariables as concreteness in the productionand understanding of language (Reynolds &Paivio, 1968), and imagery as a possible"surrogate structure" (Pompi & Lachman,

260 ALLAN PAIVIO

1967) that might mediate memory for con-nected discourse. Whatever the outcome ofsuch extensions, it is clear that imagery hasreestablished itself as a scientifically usefulconcept, even in areas long dominated by anemphasis on verbal mechanisms.

REFERENCESBARNES, L. D., & SCHULZ, R. W. Meaningfulness

and the associative phase of paired-associatelearning. Paper presented at the meeting of theMidwestern Psychological Association, Chicago,May 1966.

BERITOFF, J. S. Neural mechanisms of higher verte-brate behavior. (Trans. & ed. by W. T. Liber-son) Boston: Little, Brown, 196S.

BERLYNE, D. E. Structure and direction in think-ing. New York: Wiley, 196S.

BOUSFIELD, W. A. The problem of meaning inverbal behavior. In C. N. Cofer (Ed.), Verballearning and verbal behavior. New York: Mc-Graw-Hill, 1961.

BOWER, G. H. Mental imagery and memory. Col-loquium address, University of Western Ontario,May 1967.

BOWERS, H. Memory and mental imagery. Anexperimental study. British Journal of Psychol-ogy, 1931, 21, 271-282.

BRENER, L. R. An experimental investigation ofmemory span. Journal of Experimental Psychol-ogy, 1940, 26, 467-482.

BROOKS, L. R. The suppression of visualization inreading. Quarterly Journal of ExperimentalPsychology, 1967, 19, 289-299.

BRUNER, J. S. The course of cognitive growth.American Psychologist, 1964, 19, 1-15.

BRYDEN, M. P. A model for the sequential organi-zation of behavior. Canadian Journal of Psychol-ogy, 1967, 21, 37-S6.

BUGELSKI, B. R. Presentation time, total time, andmediation in paired-associate learning. Journalof Experimental Psychology, 1962, 63, 409-412.

BUGELSKI, B. R., KIDD, E., & SEGMEN, J. Imageas a mediator in one-trial paired-associate learn-ing. Journal of Experimental Psychology, 1968,76, 69-73.

CALKINS, M. W. Short studies in memory and inassociation from the Wellesley College psycho-logical laboratory. Psychological Review, 1898,S, 451^62.

COFER, C. N. On some factors in the organizationalcharacteristics of free recall. American Psy-chologist, 1965, 20, 261-272.

CSAPO, K. The effects of presentation rate and itemattributes in paired-associate learning. Unpub-lished master's thesis, University of WesternOntario, 1968.

DEESE, J. The structure of associations in languageand thought. Baltimore: Johns Hopkins Press,1965.

DILLEY, M. G., & PAIVIO, A. Pictures and wordsas stimulus and response items in paired-associate

learning of young children. Journal of Experi-mental Child Psychology, 1968, 6, 231-240.

DIXON, T. R., & HORTON, D. L. (Eds.). Verbalbehavior and general behavior theory. Engle-wood Cliffs, N. J.: Prentice-Hall, 1968.

DOMINOWSKI, R. L., & GADLIN, H. Imagery andpaired-associate learning. Canadian Journal ofPsychology, 1968, 22, 336-348.

DORNBUSH, R. L. Shadowing in bisensory memory.Proceedings of the 75th Annual Convention of theAmerican Psychological Association, 1967, 2, 61-62.

DUCHARME, R., & FRAISSE, P. fitude genetiquede la memorisation de mots et d'images. Cana-dian Journal of Psychology, 1965, 19, 253-261.

DUKES, W. F., & BASTIAN, J. Recall of abstractand concrete words equated on meaningfulness.Journal of Verbal Learning and Verbal Behavior,1966, 5, 455-458.

EPSTEIN, W., ROCK, I., & ZUCKERMAN, C. B.Meaning and familiarity in associative learning.Psychological Monographs, 1960, 74(4, WholeNo. 491).

FOZARD, J. L., & LAPINE, R. Comparison of dis-crimination of recency of pictures and names ofcommon objects. Paper presented at the meetingof the Eastern Psychological Association, Wash-ington, D. C., April 1968.

FRAISSE, P. Le temps de reaction verbale. AnneePsychologique, 1964, 64, 21-46.

FRINCKE, G. Word characteristics, associative-relatedness, and the free-recall of nouns. Journalof Verbal Learning and Verbal Behavior, 1968,7, 366-372.

GLANZER, M. Grammatical category: A rote learn-ing and word association analysis. Journal ofVerbal Learning and Verbal Behavior, 1962, 1,31-41.

GLANZER, M., & CLARK, W. H. Accuracy of per-ceptual recall: An analysis of organization. Jour-nal of Verbal Learning and Verbal Behavior,1962, 1, 289-299.

GLAZE, J. A. The association value of nonsensesyllables. Journal of Genetic Psychology, 1928,35, 255-267.

GLUCKSBERG, S., FISHER, D. F., & MONTY, R. A.Brief visual memory as a function of visual andacoustic confusability. Proceedings of the 75thAnnual Convention of the American PsychologicalAssociation, 1967, 2, 55-56.

GOMULICKI, B. R. The development and the pres-ent status of the trace theory of memory. BritishJournal of Psychology, 1953, No. 29 (Monogr.Suppl.).

GORMAN, A. M. Recognition memory for nounsas a function of abstractness and frequency.Journal of Experimental Psychology, 1961, 61,23-29.

Goss, A. E. Early behaviorism and verbal mediat-ing responses. American Psychologist, 1961, 16,285-298.

Goss, A. E. Comments on Professor Noble's paper.In C. N. Cofer & B. S. Musgrave (Eds.), Verbal


behavior and learning: Problems and processes.New York: McGraw-Hill, 1963.

Goss, A. E., & NODINE, C. F. Meaningfulness,familiarisation, and similarity in paired-associateslearning. New York: Academic Press, 1965.

GEUBEE, H. E., KULKIN, A., & SCHWARTZ, P. Theeffect of exposure time on mnemonic processing.Paper presented at the meeting of the EasternPsychological Association, Atlantic City, April196S.

HEBB, D. O. A textbook of psychology. Phila-delphia: Saunders, 1966.

HOLT, R. R. Imagery: The return of the ostra-cized. American Psychologist, 1964, 19, 254-264.

HULICKA, I. M., & GROSSMAN, J. L. Age-groupcomparisons for the use of mediators in paired-associate learning. Journal of Gerontology, 1967,21, 46-51.

HUNTER, I. M. L. Memory. Harmondsworth,England: Penguin, 1964.

ISCOE, I., & SEMLER, I. J. Paired-associate learn-ing in normal and mentally retarded children asa function of four experimental conditions. Jour-nal of Comparative and Physiological Psychology,1964, 57, 387-392.

JAMES, W. Principles of psychology. Vol. 1.New York: Holt, 1890.

JAMPOLSKY, M. Etudes de quelques 6preuves dereconnaissance. Annee Psychologique, 1950, 49,63-97.

JENKINS, J. J. Mediated associations: Paradigmsand situations. In C. N. Cofer & B. S. Musgrave(Eds.), Verbal behavior and learning: Problemsand processes. New York: McGraw-Hill, 1963.

JENKINS, J. R., NEALE, D. C, & DENO, S. L.Differential memory for picture and word stimuli.Journal of Educational Psychology, 1967, 58, 303-307.

JENSEN, A. R., & ROHWER, W. D., JR. What islearned in serial learning? Journal of VerbalLearning and Verbal Behavior, 1965, 4, 62-72.

KANUNGO, R. Paired associate learning of func-tion words. Psychonomic Science, 1968, 10, 47-48.

KIRKPATRICK, E. A. An experimental study ofmemory. Psychological Review, 1894, 1, 602-609.

KOEN, F. Codability of complex stimuli: Threemodes of representation. Journal of Personalityand Social Psychology, 1966, 3, 435-441.

KUHLMAN, C. K. Visual imagery in children.Unpublished doctoral dissertation, Radcliffe Col-lege, 1960.

LAMBERT, W. E., & PAIVIO, A. The influence ofnoun-adjective order on learning. Canadian Jour-nal of Psychology, 1956, 10, 9-12.

LlEBERMAN, L. R., & CuLPEPPER, J. T. Words

versus objects: Comparison of free verbal recall.Psychological Reports, 1965, 17, 983-988.

MACDONALD, V. N. Effects of word concreteness,mediation set, and retention interval on reportedmediators and recall of noun paired associates.Unpublished master's thesis, University of West-ern Ontario, 1967.

MARTIN, C. J., BOERSMA, F. J., & Cox, D. L. Aclassification of associative strategies in paired-associate learning. Psychonomic Science, 1965,3, 455-456.

McNuLTY, J. A. The effects of "instructions tomediate" upon paired-associate learning. Psy-chonomic Science, 1966, 4, 61-62.

MILGRAM, N. A. Verbal context versus visualcompound in paired-associate learning by chil-dren. Journal of Experimental Child Psychology,1967, 5, 597-603.

MILLER, G. A., GALANTER, E., & PRIBRAM, K. H.Plans and the structure of behavior. New York:Holt, 1960.

MONTAGUE, W. E., ADAMS, J. A., & KIESS, H. O.Forgetting and natural language mediation. Jour-nal of Experimental Psychology, 1966, 72, 829-833.

MOWRER, O. H. Learning theory and the symbolicprocesses. New York: Wiley, 1960.

MUELLER, D. J., & TRAVERS, R. M. W. Temporalrelations and meaningfulness in paired-associatelearning. Psychological Reports, 1965, 17, 491-497.

NEISSER, U. Cognitive psychology. New York:Appleton-Century-Crofts, 1967.

NOBLE, C. E. An analysis of meaning. Psycho-logical Review, 1952, 59, 421-430.

NOBLE, C. E., & MCNEELY, D. A. The role ofmeaningfulness (m) in paired-associate verballearning. Journal of Experimental Psychology,1957, 53, 16-22.

OLVER, M. A. Abstractness, imagery, and mean-ingfulness in recognition and free recall. Unpub-lished master's thesis, University of WesternOntario, 1965.

OSGOOD, C. E. Comments on Professor Bousfield'spaper. In C. N. Cofer (Ed.), Verbal learningand verbal behavior. New York: McGraw-Hill, 1961.

OSGOOD, C. E., Suci, G. J., & TANNENBAUM, P. H.The measurement of meaning. Urbana: Univer-sity of Illinois Press, 1957.

PAIVIO, A. Learning of adjective-noun paired-associates as a function of adjective-noun wordorder and noun abstractness. Canadian Journalof Psychology, 1963, 17, 370-379.

PAIVIO, A. Abstractness, imagery, and meaning-fulness in paired-associate learning. Journal ofVerbal Learning and Verbal Behavior, 1965, 4,32-38.

PAIVIO, A. Latency of verbal associations andimagery to noun stimuli as a function of ab-stractness and generality. Canadian Journal ofPsychology, 1966, 20, 378-387.

PAIVIO, A. Paired-associate learning and freerecall of nouns as a function of concreteness,specificity, imagery, and meaningfulness. Psycho-logical Reports, 1967, 20, 239-245.

PAIVIO, A. A factor-analytic study of word attri-butes and verbal learning. Journal of VerbalLearning and Verbal Behavior, 1968, 7, 41-49,(a)

262 ALLAN PAIVIO

PAIVIO, A. Effects of imagery instructions andconcreteness of memory pegs in a mnemonic sys-tem. Proceedings of the 76th Annual Conventionof the American Psychological Association, 1968,3, 77-78. (b)

PAIVIO, A., & CSAPO, K. Concrete image andverbal memory codes. Journal of ExperimentalPsychology, 1969, 80, 279-285.

PAIVIO, A., & MADIGAN, S. A. Imagery and asso-ciation value in paired-associate learning. Jour-nal of Experimental Psychology, 1968, 76, 35-39.

PAIVIO, A., & OLVER, M. Denotative-generality,imagery, and meaningfulness in paired-associatelearning of nouns. Psychonomic Science, 1964,1, 183-184.

PAIVIO, A., ROGERS, T. B., & SMYTHE, P. C. Whyare pictures easier to recall than words? Psy-chonomic Science, 1968, 11, 137-138.

PAIVIO, A., & SIMPSON, H. M. The effect of wordabstractness and pleasantness on pupil size duringan imagery task. Psychonomic Science, 1966, S,55-56.

PAIVIO, A., SMYTHE, P. C., & YUILLE, J. C. Im-agery versus meaningfulness of nouns in paired-associate learning. Canadian Journal of Psy-chology, 1968, 22, 427-441.

PAIVIO, A., & YAEMEY, A. D. Pictures versuswords as stimuli and responses in paired-associ-ate learning. Psychonomic Science, 1966, 5, 235-236.

PAIVIO, A., & YUILLE, J. C. Mediation instructionsand word attributes in paired-associate learning,Psychonomic Science, 1967, 8, 65-66.

PAIVIO, A., & YUILLE, J. C. Changes in associativestrategies and paired-associate learning overtrials as a function of word imagery and type oflearning set. Journal of Experimental Psychol-ogy, 1969, 79, 458-463.

PAIVIO, A., YUILLE, J. C., & MADIGAN, S. Con-creteness, imagery, and meaningfulness valuesfor 925 nouns. Journal of Experimental Psy-chology, 1968, 76(1, Pt. 2).

PAIVIO, A., YUILLE, J. C., & ROGERS, T. B. Nounimagery and meaningfulness in free and serialrecall. Journal of Experimental Psychology,1969, 79, 509-514.

PAIVIO, A., YUILLE, J. C., & SMYTHE, P. C.Stimulus and response abstractness, imagery, andmeaningfulness, and reported mediators in paired-associate learning. Canadian Journal of Psy-chology, 1966, 20, 362-377.

PIAGET, J., & INHELDER, B. L'image mentale cheeI'enfant. Paris: Presses Universitaires deFrance, 1966.

POMPI, K. F., & LACHMAN, R. Surrogate processesin the short-term retention of connected discourse.Journal of Experimental Psychology, 1967, 75,143-150.

POSTMAN, L., & STARK, K. Studies of learning tolearn, IV. Transfer from serial to paired-associ-ate learning. Journal of Verbal Learning andVerbal Behavior, 1967, 6, 339-353,

REED, H. B. Associative aids: I. Their relation tolearning, retention and other associations. Psy-chological Review, 1918, 25, 128-155.

REESE, H. W. Imagery in paired-associate learn-ing in children. Journal of Experimental ChildPsychology, 1965, 2, 290-296.

REYNOLDS, A., & PAIVIO, A. Cognitive and emo-tional determinants of speech. Canadian Journalof Psychology, 1968, 22, 164-175.

ROHWER, W. D. Verbal and visual elaboration inpaired associate learning. Project Literacy Re-ports (Cornell University), 1966, No. 7, 18-28.

ROHWER, W. D., LYNCH, S., LEVIN, J. R., &SUZUKI, N. Pictorial and verbal factors in theefficient learning of paired associates. Journal ofEducational Psychology, 1967, 58, 278-284.

SALTZ, E. Compound stimuli in verbal learning:Cognitive and sensory differentiation versusstimulus selection. Journal of Experimental Psy-chology, 1963, 66, 1-5.

SALTZ, E. Thorndike-Lorge frequency and m ofstimuli as separate factors in paired-associatelearning. Journal of Experimental Psychology,1967, 73, 473-478.

SAUFLEY, W. H., JR. An analysis of cues inserial learning. Journal of Experimental Psy-chology, 1967, 74, 414-419.

SCOTT, K. G. Clustering with perceptual and sym-bolic stimuli in free recall. Journal of VerbalLearning and Verbal Behavior, 1967, 6, 864-866.

SEIBEL, R., LOCKHART, R., & TASCHMAN, C. S.The learning of 100 paired associates in a single,self-paced trial. Paper presented at the meetingof the Psychonomic Society, Chicago, October1967.

SHEFFIELD, F. D. Theoretical considerations inthe learning of complex sequential tasks fromdemonstration and practice. In A. A. Lumsdaine(Ed.), Student response m programmed instruc-tion. (NAS-NRC Publ. No. 943) Washington,D. C.: National Academy of Sciences-NationalResearch Council, 1961.

SHEPARD, R. N. Learning and recall as organiza-tion and search. Journal of Verbal Learning andVerbal Behavior, 1966, 5, 201-204.

SHEPARD, R. N. Recognition memory for words,sentences, and pictures. Journal of Verbal Learn-ing and Verbal Behavior, 1967, 6, 156-163.

SHUELL, T. J., & KEPPEL, G. A further test ofthe chaining hypothesis of serial learning. Jour-nal of Verbal Learning and Verbal Behavior,1967, 6, 439-445.

SIMPSON, H. M., & PAIVIO, A. Effects on pupilsize of manual and verbal indicators of cognitivetask fulfillment. Perception and Psychophysics,1968, 3, 185-190.

SINGER, J. L. Daydreaming: An introduction to theexperimental study of inner experience. NewYork: Random House, 1966.

SKINNER, B. F. Science and human behavior. NewYork: Macmillan, 1953.

SKINNER, B. F. Verbal behavior. New York:Appleton-Century-Crofts, 1957,


SMITH, R. K., & NOBLE, C. E. Effects of a mne-monic technique applied to verbal learning andmemory. Perceptual and Motor Skills, 1965, 21,123-134.

SMYTHE, P. C., & PAIVIO, A. A comparison ofword imagery and meaningfulness in paired-associate learning. Psychonomic Science, 1968,10, 49-50.

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

STAATS, A. W. Verbal habit families, concepts,and the operant conditioning of word classes.Psychological Review, 1961, 68, 190-204.

STEWART, J. C. An experimental investigation ofimagery. Unpublished doctor dissertation,University of Toronto, 1965.

TAYLOR, R. L,, & POSNER, M. I. Retrieval fromvisual and verbal memory codes. Paper pre-sented at the meeting of the Midwestern Psycho-logical Association, Chicago, May 1968.

TOMKINS, S. S. Affect, imagery, consciousness.Vol. 1. New York: Springer, 1962.

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.

TULVING, E., MCNULTY, J. A., & OZIER, M. Vivid-ness of words and learning to learn in free-recalllearning. Canadian Journal of Psychology, 1965,19, 242-252.

TULVING, E., & OSLER, S. Effectiveness of re-trieval cues in memory for words. Journal ofExperimental Psychology, 1968, 77, 593-601.

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

UNDERWOOD, B. J., & ERLEBACHER, A. Studies ofcoding in verbal learning. Psychological Mono-graphs, 1965, 79(13, Whole No. 606).

UNDERWOOD, B. J., & SCHULZ, R. W. Meaningful-ness and verbal learning. Chicago: Lippincott,1960.

WALLACE, W. H., TURNER, S. H., & PERKINS,C. C. Preliminary studies of human information

storage. Signal Corps Project No. 1320, Insti-tute for Cooperative Research, University ofPennsylvania, 1957.

WALLACH, H., & AVERBACH, E. On memorymodalities. American Journal of Psychology,1955, 68, 249-257.

WATSON, J. B. Psychology as the behavioristviews it. Psychological Review, 1913, 20, 158-177.

WATSON, J. B. Behaviorism. Chicago: Universityof Chicago Press, 1930.

WIMER, C. C., & LAMBERT, W. E. The differentialeffects of word and object stimuli on the learningof paired associates. Journal of ExperimentalPsychology, 1959, 57, 31-36.

WINNICK, W. A., & KRESSEL, K. Tachistoscopicrecognition thresholds, paired-associate learning,and immediate recall as a function of abstract-ness-concreteness and word frequency. Journalof Experimental Psychology, 1965, 70, 163-168.

WOOD, G. Mnemonic systems in recall. Journal ofEducational Psychology, 1967, 58(6, Pt 2).

YARMEY, A. D., & CSAPO, K. G. Imaginal andverbal mediation instructions and stimulus attri-butes in paired-associate learning. PsychologicalRecord, 1968, 18, 191-199.

YARMEY, A. D., & PAIVIO, A. Further evidence onthe effects of word abstractness and meaningful-ness in paired-associate learning. PsychonomicScience, 1965, 2, 307-308.

YATES, F. A. The art of memory. London: Rout-ledge & Kegan Paul, 1966.

YUILLE, J. C. Concreteness without imagery inPA learning. Psychonomic Science, 1968, H,55-56.

YUILLE, J". C., & PAIVIO, A. Latency of imaginaland verbal mediators as a function of stimulusand response concreteness-imagery. Journal ofExperimental Psychology, 1967, 75, 540-544.

YUILLE, J. C., & PAIVIO, A. Imagery and verbalmediation instructions in paired-associate learning.Journal of Experimental Psychology, 1968, 78,436-441.

(Received June 21,1968)

vol. psychological review - semantic scholar · 242 allan paivio call may be entirely verbal, but...

Documents