mnemotechnics: some limitations of the mnemonic keyword ......for the keyword-generated condition it...

Journal of I'Mucational Psychology1981, Vol. 73, No. 3, 345-357

Copyright 1981 by the American Psychological Association, Inc.0022-0663/81/7303-0345$00.75

Mnemotechnics: Some Limitations of the MnemonicKeyword Method for the Study of Foreign

Language Vocabulary

James W. Hall, Kim P. Wilson, and Richard J. PattersonNorthwestern University

The effectiveness of the mnemonic keyword method was examined in four ex-periments with college students learning lists of 24-32 pairs of Spanish nounsand their English equivalents. The first three experiments, in which the listswere presented for periods of free study, yielding the following results: (a)The keyword condition, with keywords supplied by the experimenter, wassimilar or somewhat inferior to the control conditions both in the Spanish-English (forward) and in the English-Spanish (backward) direction; (b) thekeyword condition was clearly inferior to controls when students were re-quired to generate keywords; (c) keyword and control conditions were similarin retention over 1 week. In Experiment 4 the keyword method was superiorto the control condition with successive, experimenter-paced presentation butinferior with free-study presentation; free study was markedly superior topaced presentation for both keyword and control conditions. Implications forthe further study and application of the keyword method, and mnemotechnicsmore generally, are discussed.

The keyword method was proposed byAtkinson and his colleagues (Atkinson, 1975;Atkinson & Raugh, 1975; Raugh & Atkinson,1975) as a promising instructional methodfor the acquisition of foreign language vo-cabulary and as an example of more generaleducational application of mnemonic tech-niques, or mnemotechnics. It is a three-stage method: (a) A common English word,the keyword, is derived from the foreignword based on acoustic and/or orthographicsimilarity; (b) a stable association is builtbetween the foreign word and the keyword;and (c) a visual image containing the ref-erents of the keyword and the Englishtranslation is produced by the learner. Forexample, consider the Spanish word cand-ada, which means padlock. Step 1 is to

Portions of this paper were presented at the annualmeetings of the American Education Research Associ-ation, San Francisco, April 1979.

We are grateful to Kay Alderman, Jean Carlson, andNancy Green of the Northeastern University faculty forpermitting their students to participate in Experiment2, to Walter Secada for assistance with the data of Ex-periment 4, and to Kathy Meyers for typing the man-uscript.

Requests for reprints should be sent to James W.Hall, School of Education, Northwestern University,Evanston, Illinois 60201.

derive from candada a keyword, whichmight be candy or can. It also could be dad,but Raugh and Atkinson (1975) suggest that,other things being equal, keywords derivedfrom the first syllable are likely to be moreeffective than those derived from later syl-lables. It also could be candid, but onewould expect more difficulty with less con-crete keywords in generating linking images.Having determined a keyword, candy in thiscase, it will be important to rehearse thecandada-candy pair enough so that candybecomes a reliable mediating link in thechain between candada and padlock. Thethird step is to establish a strong associationbetween the keyword and the Englishtranslation. In the version of the keywordmethod systematically studied by the At-kinson group, keywords were provided bythe experimenter, and the keyword-Englishword association was formed through thesubject's production of a compound image.In the present case one would generate aninternal visual image of a padlock andcandy—perhaps a padlocked box of candy.Again, presumably through rehearsal of thisimage, a strong link between the keywordand the English translation would be es-tablished.

In several studies involving both Spanish

345

346 J. HALL, K. WILSON, AND R. PATTERSON

and Russian vocabulary, impressive differ-ences favoring the keyword method werereported by Atkinson and his collaborators.Further, the work was an unusually clearexample of the synthesis of earlier more basicresearch and the application of the methodsand findings to an important educationalproblem, foreign language acquisition. Theevidence reported in these experiments wasaccompanied by a discussion that illumi-nated a number of issues relating to the useof the method and raised other questions forfurther empirical study. Among the mostobvious applied questions were those con-cerning the range of conditions under whichthe method is effective. The studies re-ported in this article address several suchquestions.

The first experiment examined recall ofEnglish equivalents of Spanish nouns bothimmediately and after 1 week for keyword-provided (by the experimenter), keyword-generated (by the subject), and controlconditions, using a group administrationprocedure. In Experiment 2 the keyword-provided and control conditions were com-pared both for forward (Spanish-English)and backward (English-Spanish) perfor-mance. Experiment 3 also examined pos-sible retention effects, focusing especially onthe keyword-generated method and makingadjustments in the study phase in an effortto equate that condition with the controlcondition on immediate recall. These ex-periments differed from those reported byAtkinson and others (e.g., Pressley & Levin,1978) also in permitting the free study of avocabulary list, which more nearly repro-duces the usual conditions of study than doexperimenter-paced procedures. Experi-ment 4 directly compared the effectivenessof the keyword method under free-studyversus experimenter-paced presentation.

Experiment 1

Although the keyword-generated condi-tion has been described as less effective thanthe keyword-provided condition in the At-kinson reports, no data were given regardingkeyword-generated performance. This firstexperiment directly compared these twoconditions and used a control condition thatseems more appropriate than those generally

employed in the Atkinson studies. For ex-ample, in the first two experiments reportedby Raugh and Atkinson (1975), the keywordsubjects were given an initial period of key-word learning. Instead of allowing thecontrol students to use that period for study,as they presumably would do in an appliedsetting, the control subjects were required tolearn keywords. Moreover, the controlstudents were instructed to study by a sim-ple repetition method, which one might ex-pect to be inferior to a free-choice controlcondition such as was used in our experi-ments. In the third of Raugh and Atkin-son's studies, subjects served as their owncontrols. The subjects were instructed touse the keyword method on some items andanother method of their choice on the re-maining items. Whether the Raugh andAtkinson results (superiority of the keywordcondition) would hold for a between-subjectsdesign in which control subjects were en-couraged to devise their own strategies wasa concern in the first experiment. A secon-dary concern was the question of possibleretention differences. Raugh and Atkinson(1975, Experiment 4) reported similar lossesover a week for control and keyword condi-tions, but again no keyword-generated con-dition was included. If there is any reten-tion advantage to the keyword method, onemight expect it to occur when the subjectsgenerate their own keywords.

To summarize, Experiment 1 included twokeyword conditions—keyword generatedand keyword provided—and a control con-dition in which the subjects were encouragedto use a strategy of their choice. All thesubjects were given paired-associate recalltests immediately following a 10-minutestudy period and again 1 week later.

Method

Subjects and design. The subjects were 60 North-western University undergraduates whose participationearned them course credit. In this and the other ex-periments reported here, in order to exclude studentswith prior knowledge of items, students were notifiedthat participation was restricted to students with noformal course work in Spanish and who were not fromSpanish speaking homes. At the end of each experi-mental session, compliance was checked by asking thestudents to indicate their level of proficiency in Spanish.Those who violated our regulations were then elimi-nated from the study. As the subjects reported for the

LIMITATIONS OF THE KEYWORD METHOD 347

group experimental sessions, they were assigned ran-domly (except to ensure equal ras) to one of three studyconditions. In the keyword-provided condition, theexperimenter explained and demonstrated the keywordmethod and also provided a keyword for each Spanishword. The keyword-generated condition differed fromthe keyword-provided condition only in requiring thesubjects to produce their own keywords. In the controlcondition there was no mention of the keyword method,and the subjects were free to use whatever study methodthey wished. Five random orders of the 30-item studylist were used. All of the subjects were tested imme-diately following the study period and 1 week later; forall subjects a single different order of items was used oneach occasion. Subsequent to testing, the data of twosubjects in each of the keyword-generated and controlconditions and three in the keyword-provided conditionwere discarded because of the subjects' previous fa-miliarity with Spanish vocabulary. An additionalprotocol was randomly eliminated from the keyword-provided and control conditions so that each conditioncontained 17 subjects.

Materials and procedure. The list of 30 Spanishwords shown in the top section of Table I was selectedfrom a longer list of Spanish equivalents of 280 nounsdrawn randomly from the Paivio, Yuille, and Madigan(1968) imagery norms. Pilot data and the Paivio et al.(1968) norms were used to select the 30 items and key-words for the keyword-provided condition such that thekeywords were relatively easily produced based on or-thographic similarity and that low imagery items wereavoided. So that all three conditions could be admin-istered within a given session, instructions, study lists,and test lists were presented in booklet form. The in-structions for the keyword-provided and keyword-generated subjects described and illustrated each stepin the use of the keyword method and differed only inthe emphasis given to discovering appropriate keywordsin the keyword-generated instructions. The controlinstructions were slightly briefer and encouraged thesubject to examine each word carefully and to "practicewhatever techniques you want to use." The study listfor the keyword-provided conditions is shown in Table1. For the keyword-generated condition it was identicalexcept that the parentheses were unfilled, and for thecontrol condition the keyword column was omittedentirely. All subjects were given 10 minutes for studyand were told that they could write anything theywished in their booklets to help them perform better.In addition, the keyword-generated subjects were in-structed to write in the keywords that they generated.At the end of the study period, the subjects were told toturn to the test page containing the Spanish words onlyand were given 4 minutes in which to fill in the Englishequivalents. When the subjects returned a week later,ostensibly for a different task, they first were given a2-minute free recall test on the English response termsbefore they were given the 4-minute paired associatetest. Then the keyword groups were given an addi-tional 2 minutes in which to record as many of thekeywords as they could remember.

Results and Discussion

The paired associate test data are sum-

marized in Table 2. The main effects ofstudy condition and of retention intervalboth were significant, F(2, 48) = 4.08, MSB= 54.03, p < .05, and, F(2, 48) = 416.54, MSe= 9.55, p < .001, respectively, as was the in-teraction between them, F(2, 48) = 4.45,MSe = 9.55, p < .05. In post hoc compari-sons (Tukey HSD) on the immediate recalldata, the keyword-generated condition waslower than both the keyword-provided andcontrol conditions (p < .05); the latter twoconditions did not differ. The inferiority ofthe key word-generated to the keyword-provided condition was not surprising, giventhe assertions of Raugh and Atkinson (1975).The absence of a difference between thekeyword-provided and the control condi-tions was unexpected, given the consistentsuperiority of the keyword method to thevarious control conditions in the Atkinsonstudies. Possible explanations for this ap-parent discrepancy are discussed in theGeneral Discussion section.

The interaction between study conditionand retention interval, reflecting the loss offewer items over 1 week by the keyword-generated condition relative to the others,suggests the possibility of slower forgettingby that group. However, interpretation iscomplicated by the generating group's lowerimmediate recall, so that when one examinesthe percentage of items lost, one finds onlya very slight difference among the groups.Removal of the ambiguity requires that thegroups be similar in original learning andthat a between-subjects design be used.Steps were taken to assure these conditionsin Experiment 3.

A closer examination of performance andself-reports in the keyword-generated con-dition may be useful in understanding theinferior performance of that group. It waspossible that some subjects did not use themethod, given the keyword-generation de-mands, and that their lower performancepulled the group mean downward. In fact,3 of the 17 keyword-generated subjects ap-parently made no use of the keyword meth-od. However, their mean recall scores werehigher than for those who did. Of the re-maining subjects, keywords were generatedfor only 80% of the items. Self-reports in-dicated that some of the keyword omissionswere due to the difficulty in generating an

348 MALI,, K. WILSON, AND R. PATTERSON

appropriate keyword and that in other casesthe keyword was omitted because a moreappealing way of remembering the item wasdiscovered. Finally, the effectiveness ofkeywords recalled after a week in mediatingrecall of the target words was compared bycalculating the probability of recalling a

target word when its keyword had been re-called. The resulting difference (M = .48 forkeyword provided and .44 for keyword gen-erated) did not approach significance,suggesting that the deficiency of the key-word-generated group was in keyword gen-eration and learning and not in the forma -

Table 1Lists Used in Experiments 1, 2, and 3

Spanish Keyword English Spanish Keyword English

Experiment 1

LardonTubeGrepusculoCocinaPedrejonMancharCulebraAbuelaCiudadManzanaDoronicoChiquilloFrenteFingidoPecho

(Ladle)(Tuba)(Crepes)(Coke)(Pedestrian)(Man)(Coal)(Bull)(Dad)(Banana)(Door)(Chinchilla)(Friend)(Finger)(Peach)

RobberPipeTwlightCuisineBoulderStainSnakeGrandmotherCityAppleLeopardChildForeheadDummyBreast

BanderaPinzasAnochecidaG arraPeritoPanueloMiembroTablonMuerteCaballeroTonelRopaNinoForasteroMitin

(Bandit)(Pizza)(Anchovy)(Garage)(Parrot)(Pan)(Embryo)(Table)(Mother)(Ball)(Tunnel)(Rope)(Nine)(Forest)(Mitten)

BannerTweezersNighti'allClawConnoisseurKerchiefLimbPlankDeathGentlemanKegMaterialBoyOutsiderMeeting

Experiment 2

BatataSopleteTonclPenaCandadaBarcaColaBaldeRopaTaconSaltonCampana

(Bat)(Soap)(Toe)(Pen)(Can)(Bark)(Coal)(Ball)(Rope)(Tack)(Salt)(Camp)

Sweet PotatoBlowtorchBarrelRockPadlockBoatTailBucketClothingHeelGrasshopperBell

BandejaPavoMantaTarimaPatinBomberoTestuzSonrisaMatonGarnetDadosPandero

(Band)(Paw)(Man)(Tar)(Pat)(Bomb)(Test)(Sun)(Mat)(Car)(Dad)(Pan)

TrayTurkeyBucketPlatformSkateFiremanForeheadSmileBullyNotebookDiceTambourine

Experiment 3

AcallarMendigoOrgulloVueltaGolpeForasteroTenedorTabanoTristezaPasmoGalloZorraRocioJuezCocina

SilenceBeggarPrideCircuitDelugeOutsiderForkGadflySadnessAmazementFowlFoxSprayJudgeCuisine

LodoApurarSenaMiembroLurteCascoMuchachaRegaloTonelClavoLapizCautivoPapoTerronSoltera

SlushExhaustBosomLimbAvalancheHoofGirlPresentKegNailPencilCaptiveThistledownLumpMaiden


Table 2: Means and Standard Deviations forRecall, of English Equivalents Given SpanishWords in Experiment 1

Keyword KeywordTime of test provided generated Control

ImmediateMSD

1-weekMSD

24.414.11

10.294.27

17.949.05

8.064.52

24.125.34

10.714.88

lion of a functional association betweenkeywords and targets.

The difference among conditions in freerecall of response terms, obtained at thebeginning of the second session, was signif-icant, F(2, 28) = 6.64, MSe = 6.84, p < .01,with considerably higher recall for the con-trol than for the keyword conditions. Themeans for the keyword-provided, keyword-generated, and control conditions were 4.18,4.41, and 7.12, respectively. In other words,response learning was best for the controlcondition, consistent with the assumptionthat the strength of the keyword method isin the associative range of this paired-asso-ciate learning task.

Experiment 2

Experiment 2 compared the control andkeyword-provided conditions again, thistime with items used as a study by Pressley,Levin, Hall, Miller, and Berry (1980) inwhich sixth graders performed better in akeyword-provided than in a control condi-tion. To provide greater opportunity tocompare the conditions at different stages oflearning, three study-recall trials were used,rather than a single study period. Finally,Experiment 2 also examined backward aswell as forward recall. Although Atkinson(1975) has reported an advantage for thekeyword method in the backward direction,the relevant evidence was not described insufficient detail to permit its evaluation.The subjects in Experiment 2 were collegestudents presumed to be lower in academicabilities than those who served in Experi-ment 1. Thus, one would expect less so-phistication with respect to spontaneousstudy strategies by control subjects, with a

corresponding increase in the probability ofenhanced performance for those using thekeyword method.

Method

Subjects. The participants were 40 students un-schooled in Spanish, enrolled in educational psychologycourses in a vocationally oriented public commutercollege in Chicago. Several additional students par-ticipated in the experiment hut were not included he-cause of a Spanish language background; another wasexcluded so that each condition would include 20 stu-dents.

Design, materials, and procedure. Through the useof test booklets, the students were administered the 24pairs shown in the middle section of Table 1 for 3 studytrials of 2.5 minutes each followed by test periods of 2.5minutes. The layout for the keyword-provided stu-dents was as in Table 1, with the keywords in paren-theses. The control students' sheets were identicalexcept for omission of the keyword. A familiarizationsheet preceded the series of study and test sheets: Forthe key word-provided students this consisted of theSpanish words (in alphabetical order) paired with theirrespective keywords; for the control students this wassimply the list of the Spanish-English pairs to belearned. For each study trial the words were type-written in 2 columns, with the Spanish words on the leftof each pair and with the order of the pairs randomlydetermined.

For each test trial, half of the 24 pairs were tested byproviding the Spanish item and testing for the English(as in Experiment 1), and half by the reverse procedure.Over the three test trials, each item was tested at leastonce in each direction. On each test sheet the Span-ish-English items were placed along the left side of thesheet, and the English-Spanish items along the rightside. Except for these restrictions, placement of itemsand direction of testing were determined randomly foreach trial.

The experimental manipulation (keyword providedor control) was accomplished by the use of differentinstruction sheets in the test booklets and by differentsupplementary oral instructions. The 40 participantswere administered the task in three group sessions.Within each session, students were assigned randomlyto the two conditions with approximately equal num-bers in each. The students in the control condition thenwere taken into an adjoining room by an assistant whilethe keyword provided students read their instructionsheets. These consisted of a description and illustrationof the steps in the use of the keyword method, as wellas two practice pairs. After the written instructions hadbeen read, they were briefly summarized by the exper-imenter, who also elaborated on the role and charac-teristics of "good keywords" and "good images" andanswered questions raised by students. During thattime, test booklets were given to the control studentsand they were asked to read them. Their instructions,briefly reiterated by an assistant, simply described thetask, provided the same practice items as for the key-word-provided condition, and encouraged the studentto think of ways to help remember the words.


Finally, the total group was reassembled. The con-trol students were told that the keyword-providedstudents had been given a "special method" and wereurged to "beat them." The keyword-provided studentswere urged to use their special method to make sure thatthey learned more. The students then turned to thefirst study page of the booklet for a 1-minute "famili-arization period," followed immediately by the threestudy -test periods as described above, with the exper-imenter in control of the timing. The main purpose ofthe familiarization trial was to force the students in thekeyword-provided condition to focus first on the es-tablishment of a reasonably strong association betweenthe Spanish items and their keywords. This is consis-tent with the procedure used by Raugh and Atkinson(1975) in their Experiments 1 and 2, except that in thepresent experiment an equivalent amount of time tostudy the Spanish-English pairs was given the controlgroup. Thus, each condition received the same totalamount of study time for study activities appropriateto the condition.

The 24 pairs and the keywords, shown in the middlesection of Table 1, were taken from a set that had beenused by 1'ressley et al. (f980) in a study in which thekeyword method enhanced performance of sixth grad-ers. Note that in each case the keyword was identicalin spelling to the first syllable of the Spanish word.Thus, one would expect very rapid acquisition on thepart of the keyword-provided students.


Each student received a forward and abackward recall score (each ranging from 0to 12) for each test that depended on anitem's being correctly recalled and correctlyplaced. A Spanish word was scored as notcorrect if it was misspelled by more than asingle letter. (Requiring perfect spelling ledto essentially the same results.)

The major results are summarized inTable 3. A Study Condition X Direction(Spanish-English vs. English-Spanish, a

Table 3Recall Means and Standard Deviations inExperiment 2

Spanish-English English-Spanish

Condition

KeywordMsn

ControlMSD

1

4.203.74

6.052.96

Trial

2

6.953.83

7.952.58

3

8.402.96

9.552.80

1

1.201.28

2.852.50

Trial

2

3.953.39

5.703.77

3

5.503.91

7.253.97

within-subject variable) X Trial (also withinsubjects) analysis of variance (ANOVA) wasapplied to the learning data. This analysissnowed significantly better performance inrecall of English versus Spanish items, F(l,38) = 94.10, MSe = 4.91, p < .01, significantimprovement with trials, F(2, 76) = 129.54,MSe = 2.65, p < .01, and a marginally sig-nificant superiority of the control over thekeyword conditions, F(l, 38) = 2.93, MSe =47.70, .05 < p < .10. None of the interac-tions approached significance, ps > .10.The most striking finding was failure of thekeyword method to reach, let alone exceed,the level of performance of the control stu-dents. Several plausible explanations of thisfinding can be ruled out. One is that thekeyword group was slower in the early stagesof learning because of the unfamiliar de-mands of the method and especially byhaving to devote some time to ensuring ad-equate associative strengths between key-words and their corresponding Spanishwords. This possibility suggests a StudyMethod X Trial interaction, but no such in-teraction was found. A second possibility isthat learning in the English-Spanish direc-tion was much more difficult for the keywordthan the control group, so that keywordsubjects spent relatively more of their timeon backward study at the expense of forwardlearning. If that were true, one would expectthe maximum difference in English-Spanishperformance to occur on the first trial, witha decrease over trials as the keyword stu-dents diverted study time to backwardlearning. Again, the absence of significantinteractions involving trial argues againstthat possibility. In addition, the results ofa series of studies with school children byPressley et al. (1980) indicate that no specialdifficulty in backward learning is producedby the use of the keyword method. Thus, atthis point, we are left with no clear accountfor the ineffectiveness of the keywordmethod in this experiment relative to thosecited earlier (e.g., Raugh & Atkinson,1975).

Experiment 3

The performance of the keyword groupsrelative to control subjects in Experiments1 and 2 was not impressive. The possible


exception to that generalization was thesuggestion in the data of Experiment 1 thatthe rate of long-term forgetting may beslower when the keyword method is usedwith the students having to generate theirown keywords. The equivocal nature of thisevidence for differential forgetting waspointed out earlier; Experiment 1 simply wasnot designed to answer this question. Ex-periment 3 was designed for that purpose byvarying study time to approximately equateoriginal learning between control and key-word-generated conditions and by using theproper between-subjects design with respectto the two retention intervals.

Method

Subjects and design. The subjects were 48 North-western University undergraduates, who either receivedcourse credit or a small fee. The Spanish-English taskwas administered to small groups of subjects, who wererandomly assigned to one of four conditions. Thesefour conditions resulted from the orthogonal combi-nation of two study conditions (keyword generated andcontrol) and two retention intervals (2 minutes and 1week).

Materials and procedure. The target words, shownin the bottom section of Table 1, were 30 nouns selectedfrom a larger pool of 280 nouns from the Paivio et al.(1968) norms. The Spanish equivalents were obtainedfrom a Spanish-English dictionary. Selection of theitems was random except that several cognates wereexcluded. The 30 Spanish-English pairs were typedon a sheet of paper in two columns, Spanish on the leftand English on the right. The recall test consisted ofthe Spanish words (in an order different from the studyorder) in a column on the left with a blank line on theright of each.

The subjects were administered the task singly or insmall group sessions, with sessions randomly assignedto conditions except to ensure equal res. To begin thesessions, the subjects were given booklets containinginstructions and study and test materials. The controlsubjects were instructed that their task was to learn theEnglish equivalents of a list of Spanish words and thatthey should decide on a method to accomplish this.The keyword group was given detailed instructionssimilar to those for the keyword-provided group inExperiment 1 concerning the use of the keywordmethod. Four nontarget practice items were given, andthe subjects were instructed to write their keywords sothat we could ensure their understanding and ability toproduce appropriate keywords.

The data from Experiment 1 were used to estimatethe amount of study time for each group that wouldresult in optimal learning. Optimal learning was re-garded as a level similar for the keyword-generated andcontrol conditions and relatively high but not at ceiling.Study periods of 10 minutes for the keyword-generatedgroup and 7 minutes for the control group were selected.

For both conditions a 3-minute recall test immediatelyfollowed the study period, followed in turn by a 2-minute filler task. That task, for the 2-minute reten-tion group, was followed by a second recall test. For the1 week group, the 2-minute filler task ended the session.Those subjects returned a week later and were given therecall test. Two steps were taken to prevent subjectsfrom expecting the 1-week retention test and thus re-duce the probability of rehearsing items during thatinterval. First, they already had been tested—that wasthe main function of the immediate recall test. Second,they were told that a new task would be administeredin the next week's session and were given the assignmentof learning the Spanish words for the numbers 1-12during the interim.


The first question is whether the condi-tions have been met that permit a satisfac-tory test of possible forgetting differencesbetween the keyword-generated and controlgroups. An affirmative answer is indicatedby the following three results. First, thekeyword and control groups were nearlyequal in immediate recall (means of 20.88and 20.00, respectively, F < 1) and in 2-minute recall (means of 20.42 and 19.67, re-spectively, F < 1). Second, the level of recallafter 2 minutes was nearly ideal, being suf-ficiently high to permit considerable for-getting to be displayed but low enough toavoid a ceiling effect. Third, the amount offorgetting was sufficient to be sensitive toany substantial retentive difference betweenthe groups but not so great as to produce afloor effect on the 1-week test.

The conditions just described permit anyobserved interaction between study condi-tions and retention interval to be interpretedin terms of a differential rate of forgetting.It is clear that not even the slightest sug-gestion of such an interaction was found.The mean losses from 2 minutes to 1 weekwere 14 items (69%) for the keyword-gener-ated group and 12.5 items (64%) for thecontrol group. Note too that the equivalentlevels of learning found with different studytimes also serves to replicate the superiorityfound for the control condition over thekeyword-generated condition in Experi-ment 1.

Experiment 4

Before describing Experiment 4, it will beinstructive to recapitulate and discuss some


implications of the results of the previousthree experiments. Those experimentsyielded four main results. First, the key-word-provided condition was no better than,and perhaps slightly inferior to, a controlgroup urged to generate their own studystrategies. Second, the requirement ofgenerating one's keyword resulted in per-formance markedly inferior to either thekeyword-provided or the control condition.Third, learning in the backward (English-Spanish) direction followed about the samepattern as in the forward direction. Fourth,insofar as long-term retention is concerned(given equivalent original learning), there isniether an advantage nor a disadvantage tothe use of the keyword method.

The various studies cited earlier had ledus to expect better performance for thekeyword-provided group than for the controlgroup. This clearly was not the case in ourresearch, as is shown in both Experiments 1and 2. In fact, the advantage, if any exists,belongs to the control condition. The reasonfor the apparent discrepancy between ourresults and those of others is unclear, butseveral possibilities may be considered. Thefirst of these concerns the nature of thecontrol conditions used. We regard ourbetween-subjects design in which the groupsare given equal functional study time and thecontrol subjects are encouraged to generatean effective study method of their choice asa more reasonable approach than experi-menter-paced presentation to the evaluationof the effectiveness of the keywordmethod.

Of the remaining differences among thestudies, the one we suspect to be the mostimportant determinant of the differing re-sults is in the method by which items werepresented for study. In the experiments wehave reported here, all items were presentedsimultaneously for free study, whereas theAtkinson studies employed experimenter-paced presentation. There is evidencesuggesting that simultaneous presentationis substantially superior to successive pre-sentation for free recall (e.g., Hall, Grossman,& Elwood, 1976) and for paired-associatelearning (Moursund & Chape, 1966) whenslower rates of presentation are used. Thiseffect seems likely to occur because the free

study condition is more facilitative of effec-tive subject-initiated study strategies thanis single-item presentation. Our data, whencompared with those of Raugh and Atkinson(1975) in which subjects and materials seemreasonably comparable, are consistent withthis reasoning. The control condition per-formance in their four experiments wasconsistently below 35% with study times (peritem) of 10 to 15 sec. In our Experiment 1,with a mean of 20 sec per item, recall of thecontrol group was over 80%. For Trial 1 inExperiment 2, with just under 9 sec per item,mean recall was nearly 60%. In Experiment3, with 14 sec per item, mean recall wasnearly 67%. Although comparisons acrossexperiments are problematic, these perfor-mance differences are so massive that seriousconsideration seemed warranted. What issuggested is that the keyword method isminimally effective when control conditionsare more favorable to spontaneous strategyuse. In this case favorable is defined as theopportunity to allocate study time howeverthe student wishes with all materials present.Unfavorable conditions, then, are seen asthose which restrict that allocation by se-quential presentation controlled by the ex-perimenter.

Experiment 4 was designed to test theanalysis just offered by directly comparingkeyword and control conditions under thepaced presentation method used by Atkin-son and others and under the free studymethod used in our three experiments.Since Experiment 3 of Raugh and Atkinsonwas used as a model in designing this study,some detail regarding their procedure will begiven. Their experiment was conducted infour sessions. Session 1 was devoted tofamiliarization and practice with Spanish-English pairs and with the keyword method.In each of the remaining three sessions, thesubjects were given three study-test trialswith 40 different Spanish-English pairs.For each item the English word was pre-sented for 10 sec on a slide, with the keywordpresent for those items to be studied by thekeyword method. As the slide was shown,the Spanish word was pronounced twice, sothat the slide remained visible for some timefollowing oral presentation of the Spanishword.

LIMITATIONS OF THK KEYWORD METHOD 353

In our experiment two conditions closelyparalleled the Raugh and Atkinson proce-dure, except that the keyword-control ma-nipulation was done between subjects for thereasons outlined earlier. Also, only a singlesession was used in which the familiarizationand practice was followed by a single list of32 Spanish-English pairs, taken from Raughand Atkinson, for two study-recall trials.The shift to a single session should notjeopardize comparison of the two studies,since the superiority of the keyword condi-tion to the control condition remainedroughly constant over the three sessions inthe Raugh and Atkinson study. With theseexceptions, the procedure in one of ourconditions, the paced auditory condition,was identical to that used by Raugh andAtkinson. Based on Raugh and Atkinson's(1975) results, superior performance by thekeyword method, compared with its control,would be expected for the paced auditorycondition.

The second critical condition is one thatdiffered from the paced auditory conditiononly in that the items for study and test werepresented in columns on a single sheet ofpaper and the subjects were permitted tostudy the items freely for a period equal tothe total presentation time in the paced au-ditory condition. Keywords were insertedbetween Spanish-English pairs in this freestudy condition for the keyword subjects andnot for the control group.

If the results are as hypothesized, it mightbe argued that presentation mode (auditoryvs. visual) rather than the paced-free studydifference was critical. Therefore, a thirdcondition was included that was identical inevery respect to the paced auditory groupsexcept that each Spanish word was pre-sented visually rather than orally. If theresults under this paced visual conditionparalleled those of the paced auditorygroups, then any differences between thepaced condition and the free study condi-tions could not be attributed to differencesin presentation mode.

College students served as subjects andwere run in small groups, a procedure usedby Raugh and Atkinson (1975) in some ex-periments, although not in their Experi-ment 3.

MethodSubjects and design. The subjects were 100

Northwestern University undergraduates with minimalfamiliarity with Spanish. They received credit forparticipation toward their grade in the introductorypsychology course in which they were enrolled. Twosuhjects were dropped from the analyses due to theirfailure to follow instructions, and eight more weredropped to equalize ns at 15 in each of the six conditionsformed by the orthogonal combination of three pre-sentation and two study methods. In addition, twostudy-test trials were used, so that the design was a 3X 2 X 2 factorial with trial as a within-subject variable.The subjects were run in small groups (generally fewerthan 5 per group), and groups were assigned randomlyto conditions.

Materials and procedure. Fifty-four Spanish-English pairs and their keywords were selected ran-domly from the set of 120 items used in Experiment 3of Raugh and Atkinson (1975). These words were as-signed randomly to make up a 32-item main study list,a 10-item practice list, and a 12-item familiarization list,paralleling the kinds of lists used by Raugh and At-kinson.

The experimental session, which required about 50minutes, was divided into three main phases. In thefirst, or familiarization, phase the subjects were pre-sented the familiarization items, consisting of 12Spanish words (with their keywords for all keywordconditions). For the keyword subjects instructionswere to learn the keyword for each Spanish word, andthey were given two study-test trials at the rate of 10 secper item for that purpose. The concept of a keywordwas explained and the general description of the key-word method was given at this time. The controlsubjects received a single presentation of the Spanishitems simply to provide some familiarization with suchitems. The method of presentation varied with pre-sentation condition. The paced auditory keywordcondition viewed each keyword for 10 sec, and at thebeginning of each 10-sec period heard the Spanish wordpronounced three times within the first 4 sec of keywordpresentation; the paced visual keyword condition wasidentical except that the Spanish items were seen(rather than heard) during the first 4 sec of keywordpresentation; the free study keyword subjects simplyexamined a typewritten list of the Spanish -keywordpairs for a total of 120 sec. The control subjects weretreated similarly except that only one trial was givenand keywords were never presented.

The second, or practice, phase involved the presen-tation of 10 Spanish-English pairs, supplemented bykeywords for each keyword condition. The instructionsused for keyword subjects closely paralleled those usedby Raugh and Atkinson (1975), and following keywordinstructions the subjects were permitted to ask ques-tions regarding the method. The control subjects weresimply told to study the items in the most effective waythey could think of. For the paced conditions the En-glish equivalents (plus keywords, for the keyword con-ditions) were presented on a series of slides shown at a10-sec rate, with the Spanish items presented eitheraurally or visually during the first 4-5 sec of each ] 0-secinterval. The free study conditions were presented the

354 J. HAM,, K. WILSON, AND R. PATTERSON

practice list in typewritten form on a single sheet ofpaper for 100 see of study. Two study-test trials wereused. The main study list phase was identical to thepractice phase except that 32 different items were usedand the free study groups were given 320 sec to study thelist. Before beginning this final study list, a short breakwas given, after which the subjects were instructed tostudy this final list as they had the practice list. Thekeyword subjects were urged particularly to use thekeyword method.

Testing in all three phases followed presented con-ditions. That is, the paced auditory subjects were in-structed to record the English equivalents in a testbooklet as the Spanish words were pronounced once ata presentation rate of 10 sec. The same rate and pro-cedure was used for the paced visual subjects exceptthat the Spanish items were shown on slides at a 4-5-secrate. Subjects in the free study condition wrote theirresponses next to the Spanish words listed vertically intwo columns on a single sheet. In each phase the ordersof items were varied randomly across the various studyand test trials.

Following these main phases, supplementary sets ofdata were collected. First, immediately after the finalmain list test, all subjects were given a typewritten listof all 32 English equivalents and were asked to writedown all the Spanish words, correctly paired if possible,that they could. Second, the keyword subjects weregiven a list of both the Spanish and the English wordsand were asked to reproduce the keywords and to in-dicate those items on which they used the keywordmethod. Finally, the control subjects were asked toprovide a rather detailed description of the variousstudy strategies that they used.


Recall of English equivalents. TheSpanish-English recall data are summarizedin Table 4. The results for the paced audi-tory conditions parallel those of Raugh andAtkinson (1975) reasonably closely. Theirapproximate recall percentages (taken fromtheir graph in Figure 3) for Trials 1 and 2 fortheir first study list were 34 and 75 for the

Table 4Recall Means and Standard Deviations inExperiment 4

Presentationcondition

Keyword trial Control trial

1 2 1

11.27 20.53 7.93 17.535.34 7.05 3.86 6.60

Paced-auditoryMsn

Paced-visualM 16.13 24.47 11.53 23.93SI) 6.80 6.92 5.78 6.30

Free study-visualM 18.40 27.53 23.67 29.20SI) 7.24 5.38 7.32 6.42

keyword condition and 22 and 51 for thecontrol condition. Our corresponding per-centages were 35 and 64 (keyword) and 25and 55 (control). The similarity betweentheir data and ours from comparable con-ditions suggests that the essential elementsof the Raugh and Atkinson experiment werereproduced in our paced auditory conditions,making contrasts between that and our re-maining conditions more clearly interpret-able.

Two ANOVAs were applied, the first ofthese examining Trial 1 performance as afunction of study (keyword vs. control) andpresentation (paced auditory, paced visual,and free study visual). Trial 2 was excludedfrom this analysis because of an obviousceiling effect in the free study visual condi-tion. In this analysis the main effect ofpresentation method was significant, F(Z,84) = 26.26, MSe = 38.16, p < .001, as wasthe interaction between presentation andstudy conditions, F(2, 84) = 5.66, MSe =38.16, p < .01. The main effect of studymethod did not approach significance (F <1.0). These results are in accord with thereasoning outlined earlier concerning theinfluence of presentation condition (pacedvs. free study) on the relative effectivenessof the keyword method. Performance wasuniformly higher with free study, and therelative effectiveness of the keyword methodwas greater under paced presentation thanunder free study. In fact, under free studythe control condition was significantly higherthan the keyword condition, £(28) = 2.01, p^.05.

The second ANOVA was restricted to thepaced conditions and included both trials.The main effect of presentation conditionwas reliable, F(l, 56) = 9.87, MSe = 67.17, p< .01, indicating that the visual presentationof the Spanish words was superior to oralpresentation. The absence of an interactionbetween study and presentation method (F< 1) indicates that this superiority of visualstimulus presentation held across both study(keyword and control) conditions. Althoughmain effect of study condition was onlymarginally reliable, F(\, 56) = 3.67, MSe =67.17, p ^ .06, it seems safe enough to in-terpret this as indicating keyword superi-ority, given the earlier results of Raugh andAtkinson (1975). The main effect of trialswas reliable, F(l, 56) = 333.49, MSe = 8.82,


p < .001, as was the interaction between trialand study method, F(l, 56) = 4.12, MSe =8.82, p < .05. That interaction reflects thefact that keyword superiority was greater onTrial 1 than on Trial 2. However, the in-terpretation of that interaction is cloudedsomewhat by the marginally reliable three-way interaction among study method, pre-sentation method, and trial, .F(l, 56) = 2.92,p < .10, resulting from the virtual absence ofany keyword superiority on Trial 2 for thepaced visual condition.

Recall of Spanish items. Since the pacedauditory subjects were writing down Spanishwords that they had never seen, with little orno knowledge of sound-spelling correspon-dence in Spanish, only the visual conditionswere compared in backward learning. Thedata first were scored as in Experiment 2,with items scored as correct if they werecorrectly paired and misspelled by no morethan one letter, and incorrect otherwise.The mean numbers of items correct underthis scoring were 15.27 for the paced visualkeyword condition, 19.13 for the paced visualcontrol condition, 16.93 for the free studyvisual keyword condition, and 18.33 for thefree study visual control condition. Al-though the control condition exceeded thekeyword condition under both presentationmethods, no effects approached significance,F(l, 56) = 1.80, p > .10. The data also werescored by a native Spanish speaker, blind toexperimental conditions, who simply judgedwhether each item was close enough to cor-rect to communicate effectively. The pat-tern of group differences found by using thismethod was identical to those reportedabove, and the correlation between score setswas extremely high (r - .98).

Recall of keywords and reports of studystrategies. Mean keyword recall was 23.6for the free study visual keyword condition,19.00 for the paced visual keyword group,and 20.53 for the paced auditory keywordgroup. The differences did not approachsignificance (F < I). Mean numbers ofitems on which the keyword subjects re-ported using the keyword method were23.73, 16.0, and 22.4 for the three groups,respectively. The difference among thesemeans was significant, F(2, 42) = 3.64, MSe= 70.36, p < .05. Evidently, with pacedpresentation, the exclusively visual presen-tation of the stimulus item is not conducive

to the adoption of the keyword method. Aninspection of the Spanish-English recalldata for keyword subjects reporting heavyversus little use of the keywords indicatedonly minor differences. Had the subjectswho reported little keyword use or who re-called few keywords been omitted, the effecton the group means would have been slightand would have produced an even largerPresentation X Study Method interaction.

The study strategy self-report data fromcontrol subjects were so variable in theirspecificity and completeness that no attemptwas made to analyze and report them in aformal fashion. However, a number of thesubjects did provide very detailed reports ofhow they studied many of the items, and allgave some information.

Nearly every control subject reported theuse of more than a single strategy. Thosemost frequently reported included (a) simplerepetition, (b) attention to formal similari-ties between pair members, such as the factthat desarollo and development begin withthe same letter, (c) noting similarities be-tween Spanish words and French, Italian, orLatin words known already by the subject,and (d) variations of the keyword method.Nearly every subject reported having usedEnglish words similar in sound or spelling tothe Spanish words to provide links betweenthe stimulus and response terms. Severalversions of this self-generated keywordmethod were described. In a number of in-stances, the particular keyword selected wasbased partly on semantic qualities of theEnglish equivalent as well as perceptualfeatures of the Spanish items. Two exam-ples will clarify what we mean here. Fromthe Spanish word pestana, several controlsubjects in the paced auditory conditionderived the keyword paste on instead ofpest, which was furnished as a keyword tothe keyword subjects. The English equiv-alent was eyelash, and the preexperimentalassociation with eyelashes (pasting on falseeyelashes) almost certainly contributed tothe selection of paste on as a keyword. Asecond example is exito, meaning success,for which a subject generated excited (abouta success) as a keyword, in contrast to exit,which was provided to the keyword subjects.In both cases, and in many more reported bysubjects, the semantic features of the Englishresponse terms played an important role in


the generation of keywords. What is im-portant here is that such keywords have anobvious advantage over those based solely onstimulus features, in that there is a preex-isting link between the keyword and the re-sponse term, instead of the rather arbitraryimagery link that characterizes the keywordmethod as originally conceived by At-kinson.

A final point is that it appears that verbalelaboration was used more frequently thanimagery elaboration by control subjects tolink keywords with English responses.Some use of imagery links was reported,however, and in many cases the reports werenot clear as to whether imagery rather thanverbal elaboration was used.

General Discussion

To summarize our findings, in Experiment1 the control and keyword-provided condi-tions performed similarly, and both weresuperior to the condition in which subjectsgenerated their own keywords. In Experi-ment 2 the control subjects were somewhatbetter than the keyword-provided subjectson each of the three study-test trials givenboth in the Spanish-English and in the En-glish-Spanish direction. In Experiment 3,similar levels of forgetting after 1 week wereobserved for keyword-generated and controlsubjects. Finally, in Experiment 4, thekeyword-provided condition was superior tocontrols when presentation was experi-menter-paced, but inferior when free studywas permitted. Free study produced betterperformance than paced presentation forboth control and keyword conditions.

The major implications of our results de-rive from our finding that the superiority ofthe keyword method to an appropriate con-trol condition does not hold for college stu-dents allowed to study the target itemsfreely. This finding was somewhat seren-dipitous, in that it was an unexpected out-come of the initial decision to display allpairs simultaneously for study rather thanusing experimenter-paced presentation.That decision was based on the convictionthat the simultaneous presentation proce-dure would better simulate ordinary studyconditions and thus would provide a moreappropriate context in which to evaluate thekeyword method in terms of retention,

backward leaning, and so forth. It was notuntil an absence of facilitation for the key-word condition was found in Experiments 1,2, and 3 that the possibility was noted thatthis presentation variable might be critical;that is, that the keyword method might be apowerful enough study strategy to provesuperior when self-generated strategies arerestricted by rigidly paced presentation butnot so powerful as strategies generated whenbright students are given greater control overtheir study activities. The results of Ex-periment 4 are consistent with that rea-soning. The keyword condition was inferiorto the control condition when students wereallowed to study freely, whereas the Raughand Atkinson (1975, Experiment 3) resultswere replicated with our paced presentationcondition.

It is not clear to what extent the simulta-neous presence of the items in the free studyconditions is critical. It may be that it is thesubjects' control over the amount of timeallocated to each item and the frequencywith which items are restudied that arecritical across presentation conditions.Another possibility is that a critical differ-ence between the paced and free study con-ditions is in the amount of time that theSpanish items were available for study.Those items of course were availablethroughout the study period in the free studyconditions, whereas in the paced conditionsthey were presented briefly at the beginningof each 10-sec study interval. This wouldseem to be particularly disadvantageous tocontrol subjects who had less assistance inachieving a stable code for the stimulus item.These various possibilities remain to besorted out in more analytical experiments.

Our data, and the reasoning describedabove, lead us to propose a generalizationthat may be useful both in considering ap-plications of such mnemonic devices as thekeyword method and in planning researchrelative to that applicability: A particularimposed study strategy will be maximallyuseful when ordinary study conditions areminimally facilitative of learning; such im-posed strategies will be minimally usefulwhen ordinary conditions are maximallyfacilitative of learning. The term conditionis used broadly here to refer to any charac-teristics of the task or the learner that affectordinary performance. For example, highly


expert learners such as university studentswould be expected to profit less from mne-motechnics, other things being equal, thanless effective learners. In the latter categorywould be both the less bright subject and theyounger subject. Thus Pressley (1977) andPressley and Levin (1978) have obtainedimpressive levels of facilitation when thekeyword method was used by sixthgraders.

The very high level of performance ofuniversity students with free study, and theconsequent absence of keyword superiority,may not hold for younger subjects, since theywould not be expected to take full advantageof the study opportunities afforded by si-multaneous presentation. In fact, exactlythat is suggested by data recently obtainedby Levin, Pressley, McCormick, Miller, andShriberg (1979), who found equivalent key-word superiority to a control condition using10th grade students under both paced pre-sentation and free study. However, no adultgroup was included in that study. What isneeded to clarify this issue is a develop-mental study in which college students andyounger students are compared directly inkeyword effectiveness with paced versus si-multaneous presentation.

Other variables that warrant more sys-tematic study are particular item charac-teristics (e.g., concreteness, meaningfulness).In the studies reported to date, the foreignwords used have been selected because theyappeared to be appropriate to the keywordmethod. Put another way, we are unawareof any demonstration of facilitation by key-word use with unselected foreign words, norhas any "tried- and true" method of selectingitems been demonstrated, to our knowledge.Also, our self-report data from Experiment4 suggest the advisability of using keywordsin which some preexisting link other thanformal similarity exists between that wordand the response term. The possibility ap-pears to warrant an experimental test.

Finally, a comprehensive evaluation of theutility of such techniques as the keywordmethod will require more than a series ofcomparisons between such methods usedalone and a single nonstrategy control con-dition. In fact, the keyword method was notoriginally proposed as a replacement forother study procedures, but as a supplement

to such procedures (Raugh & Atkinson1975). Thus, the keyword method may turnout to be useful even with college studentsunder free study conditions, provided thatit is used selectively along with other meth-ods of study. For example, given a long vo-cabulary list, a student might profit by firstapplying ordinary self-generated strategiesfor study, then applying the keywordmethod to a subset of items that seem par-ticularly difficult, perhaps using a combi-nation of imagery and verbal elaboration.More complex experiments than we have yetconducted will be necessary to move fromour present state of knowledge to the fullapplication level with respect to the keywordmethod, and that generalization seems likelyto be equally true for most mnemotech-nics.

References

Atkinson, R. C. Mnemotechnics in second-languagelearning. American Psychologist, 1975, 30,821-828.

Atkinson, R. C., & Raugh, M. R. An application of themnemonic keyword method to the acquisition of aRussian vocabulary. Journal of ExperimentalPsychology: Human Learning and Memory, ] 975,104,126-133.

Hall, J. W., Grossman, L. R., & Elwood, K. D. Differ-ences in encoding for free recall vs. recognition.Memory & Cognition, 1976, 4, 507-513.

Levin, J. R., Pressley, M., McCormick, C. B., Miller, G.E., & Shriberg, L. K. Assessing the classroom po-tential of the keyword method. Journal of Educa-tional Psychology, 1979, 71, 583-594.

Moursund, J. P., & Chape, C. M. Comparison of threepresentation methods in paired-associate learning.Psychonomic Science, 1966, 6, 167-168.

Paivio, A., Yuille, J. C., & Madigan, S. Concreteness,imagery, and meaningfulness values for 925 nouns.Journal of Experimental Psychology, 1968, 76,10-25.

Pressley, M. Children's use of the keyword method tolearn simple Spanish vocabulary words. Journal ofEducational Psychology, 1977, 69, 465-472.

Pressley, M., & Levin, J. R. Developmental constraintsassociated with children's use of the keyword methodof foreign language vocabulary learning. Journal ofExperimental Child Psychology, 1978, 26, 359-372.

Pressley, M., Levin, J. R., Hall, J. W., Miller, G. R., &Berry, J. K. The keyword method for foreign wordacquisition. Journal of Experimental Psychology:Human Learning & Memory, 1980,2,163-173.

Raugh, M. R., & Atkinson, R. C. A mnemonic methodfor learning a second-language vocabulary. Journalof Educational Psychology, 1975, 67, 1-16.

Received May 2, 1980 •

mnemotechnics: some limitations of the mnemonic keyword ......for the keyword-generated condition it...

Documents