DOC MIT 111/11611 SD Center for Cogaitive Learning. (DM ...This finding in turn suggests that meaningful-ness (as defined here) may well be a crucial ... "dory," with a normative frequency

DOC MIT 111/11611

SD 011$ 3S CS 001 501

10TOOR Ghatala, Elisabeth S.; lad OthersTITLE 1 Clarification of Pregueacy Iffects in Children's

Discriniaatioa Learning. Techaical Report No. 294.INSTITUTION isconsis Naive, Sadism*. lessarch sad Bovelopmeat

Center for Cogaitive Learning.AGEWCT National Inst. of Education (DM). Vashingtoa.

ft. C.IMPORT NO TR-294PBS DIT2 Par 74CONTNICT NE-C-00-3-0065NOTE 19p.EONS PRICE E1-00.75 SC-$1.50 PUSS POSTAGEDESCRIPTORS *Behavior Patterns; Child Developsest; Cognitive

Development; *Cognitive Processes; *Discrisinatioa.Learning; *Educational Research; SloseataryEducation; Grade 6; Learning Processes; *Memory;Retention

ILILSVRACTPoor experiments were conducted to assess the effects

of ,:-fain stimulus variables on children's discrimination learning.7a genoral, it was found that word frogmeacy was negatively relatedco discrimination learning as long as the words were seeningful tothe subjects. Noreover# the relationship between word and performascereversed in free- recall Learning,' as was expected. Equivaleatrel&tioaships between frequency sad learning were &Malted withverbal and pictorial materiels. The implications; of thqse resultswere discussed in the context oZ popular acconats of memorialrepresentation. (Author)

erc

I

A CLARIFICATION OF FREQUENCY

EFFECTS IN CHILDREN'S

DISCRIMINATION LEARNING

IS( )r\IS( R F. ARCH A Ni D4DE-VILOPMENT

. CENTER FOR

COGNITIVE LEARNING4

Technical Report No. 294

A CLARIFICATION OF FREQUENCY EFFECTS INCHILDREN'S DISCRIMINATION LEARNING

by

Elizabeth S. Ghatala, Joel R. Levin, andLois A. Makoid

Report from the Project onChildren's Learning and Development

Wisconsin Research and DevelopmentCenter for Cognitive LearningThe University of Wisconsin

Madison, Wisconsin

March 1974

Published

supported

Institute

expressed

Institute

it

11?

by the Wisconsin Research and Development Center for Cognitive Learning,

in part as a research and development center by funds from the National

of Education, Department of Health, Education, and Welfare. The opinions

herein do not necessarily reflect the position or policy of the National

of Education and no official endorsement by that agency should be inferred.

Center Contract No. NE-C-00-3-0065

Statement of Focus

Individually Guidect Education (IGE) is a new comprehensive system ofelementary education. The following components of the IGE system are invatrying stages of development and implementation: a new organization forinstruction and related administrative arrangements; a modei of instructionalprograming for the individual student; and curriculum components in prereading,reading, mathematics, motivation, and environmental education. The develop-ment of other curriculum components, of a system for managing instruction bycomputer, and of Instructional strategies is needed to complete the system.Continuing programmatic research is req4ired to provide a sound knowledgebase for the components under development and for improved second generationcomponents. Finally, systematic implementation is essential so that the prod-ucts will function properly in the IGE schools.

The Center plans and carries out the research, development, and imple-mentation components of its ICE program in this sequence: (1) identify theneeds and delimit the component problem area; (2) assess the possible con-straintsfinancial resources and availability of staff; (3) formulate general,./plans and specific procedures for solving the problems; (4) secure and allo-s-ate human and material resources to carry out the plans; (5) provide foreffective communication among personnel and efficient management of activi-ties and resources; and (6) evaluate the effectiveness of each activity andite contribution to the total program and correct any difficulties through feed-back mechanisms ald appropriate management techniques.

A self-renewing system of elementary education is 'Projected in eachparticipating elementary school, i.e., one which is less dependent on externalsources for direction and is more responsive to the needs of the children attend-ing each particular school. In the IGE schools, Center-developed and othercurriculum products compatible with the Center's instructional programing modelwill lead to higher student achievement and self-direction in learning and inconduct and also to higher morale and job satisfaction among educational per-sonnel. Each developmental product makes its unique contribution to IGE asit is implemented in the schools. The various research components add to theknowledge of Center practitioners, developers, rind theorists.

iii

Acknowiedgntents

We are grateful to Benton J. Underwood and Larry Wilder for their inputs duringthe planning of the study; to Jerri Belliston for collecting the data in Experiments I and II;to Ann E. McCabe for assistance in constructing the materials for Experiments III and IV;and to the staffs and students of Grandview and 1'. O. Smith Elementary Schools inOgden, Utah, Caswell and Rockwell Elementary Schools in Fort Atkinson, Wisconsin,and Yahara Elementary School in Stoughton, Wisconsin.

iv

Table of Contents

Page

Acknowledgments ivAbstract vii

I. Introduction 1

II. Experiment I 3

III. Experiment II 7

IV. Experiments III and IV 9

V. General Discw:sion 13

References. 15

List of Tabks and Figures

Table

1. Mean Number of Correct Responses Over Four Test Trials byExperimental Condition and Performance on the DefinitionsTest (Experiment II) 7

Figure

Mean percent correct responses by experimental conditionin discrimination learning (Experiment III) and free recall(Ex2eriment 11

Abstract

Four experiments were conducted to assess the effects of certainstimulus variables on children's discrimination learning. In general, it wasfound that word frequency was negatively related to discrimination learningas tong as the words were meaningful to as. Moreover, the relationshipkletween word frequency and performance reversed in free-recall learning,as was expected. Equivalent relationships between frequency and learningwere obtained with verbal and pictorial materials. The implications ofthese results were discussed in the context of popular accounts of memorialrepresentation.

y,/vii

IIntroduction

The purpose of the present series of experi-ments was to determine the effects of certainstimulus variables on children's discrimina-tion learning. The three dimensions on whichmaterials were varied were "backgroundfrequency" (as inferred from norms such asthose of Thorndike and Lorge, 1944), meaning-fulness (as defined by Es' semantic responsesto items), and modality (verbal versus pictorialrepresentations.

According to the tenets of frequencytheory (Ekstrand, Wallace, & Underwood,1966): (1) discrimination learning isassumed to involve subjective frequencydiscriminations between "correct" and"incorrect" pair members. It has furtherbeen assumed that: (ii) the accrual ofsubjective frequency to items in a pair maybe influenced by the background or preexperi-

mental frequency of the items, in a mannerakin to Weber's Law. Under the secondassumption, discrimination of situationalfrequency differences should be easier foritems low in background frequency than forthose high in background frequency.

In earlier studies (Ghatala & Levin,1973; Ghatala & Levin, 1974; Ghatala,Levin, & Wilder, 1973; Levin, Ghatala,& De Rose, in press), we have invoked thebackground-frequency assumption to accountfor various phenomena in children's discrimina-tion learning. The emphasis in the presentresearch is on clarifying the role of backgroundfrequency as it operates in conjunction withother stimulus variables previously demon-strated (or presumed) to have an effect ondiscrimination learning.

1 /2

IIExperimeet I

One of the assumptions just mentionedwas that discrimination learning should bebetter with materials of low backgroundfrequency than with those of high backgroundfrequency. However, tests of this predic-tion utilizing high and low-frequency wordshave been equivocal. While some studieshave found that low-frequency word pairs arelearned significantly better than high-frequencypairs (e.g., Rowe & Paivio, 1971b, Experi-ments I and IV; Underwood, Broder, & Zimmer-man, 1973), others have not (e.g., Ingison& Ekstrand. 1970; Paivio & Rowe, 1970; Rowe& Paivio, 1971b, Experiments II and III).And as Pelvic, (1971) has argued, even whensuch word-frequency effects are found theyare generally not as potent as those producedby other stimulus variables--in particular,stimulus concreteness as defined by Paivio,Yui lle, and Madigan's (1968) norms.

Recently Ghatala and Levin (1974)presented evidence to suggest that theelusive effect of word frequency in discrimina-tion learning might be due to the operation ofanother factor which may be regarded as"meaningfulness" (though not in the usual ver-bal-learning sense--cf. Underwood & Schulz,1960). In a frequency judgment task, it wasfound that subjective frequency differencesbetween high-frequency words and low-frequency words for which as knew themeanings were in accordance with predictionsstemming from Weber's Law. In contrast,the lack of difference between high-frequencywords and low-frequency words for which Esdid not know the meanings was not inaccordance with Weber's Law. This wastrue even thougn the "meaningful" and

no:mei:mime " low-frequency words werefairly comparable in terms of their averagenormative (Thorndike & Lorge, 1944)

frequencies, these being 8.2 and 6.0occurrences per million respectively.

The Ghatala and Levin resultsindicated that within the context of a fre-quency judgment task, Weber's Law holdsfor materials which are meaningful to Esbut not for materials which are not meaningful.This finding in turn suggests that meaningful-ness (as defined here) may well be a crucialvariable to control when investigating theeffects of background frequency in discrimina-tion learning. The purpose of Experiment Iwas to follow up on this suggestion. Predic-tions based on the Chagas and Levin resultswere: (a) discrimination lists consisting oflow-frequency words which are meaningful toSs will be better learned than lists consistingof high-frequency words; but (b) lists con-sisting of low-frequency words which are notmeaningful to As will be at least as difficultas high-frequency word lists.

Method

Subjects

The Es were 80 sixth-grade childrenattending an elementary school located in amiddle-class neighborhood in Ogden, Utah.The .as were randomly assigned in equal num-bers to the four conditions of the experiment.

Design and Materials

Four types of verbal-discriminationlists comprised the conditions of the experi-ment. One list (Hi-F) consisted of high-frequency words from the Alt end A range ofthe Thorndike and Lorge (1944) norms. Asecond list (to-F/111-M) consisted of low-

3

frequency words (less than 25 occurrencesper million) whose meanings were known byas in thi.c age group (e.g., "hatchet," witha normative frequency of 8), and a third list(Lo-F/Lo-M) consisted of low-frequencywords (also less than 25 occurrences per million)whose meanings were unknown by theseSs (e.g. , "dory," with a normative frequencyof 7).

The meaningfulness of the low-frequencywords was determined in a previous experiment(Ghatala & Levin, 1974) by having sixth-grade children pronounce and then define thewords. (Any definition was taken to indicatethat the word had meaning for the A.) TheLo-F/HI-M words were those which at least80% of thels could both pronounce and define.The Lo-F/Lo-M words were those which atleast 80% of the Ss could pronounce, butno more than 20% could define. The meanThorndike-Lorge frequency of the wordsselected for the Lo-F/Hi-M list was 5.93,and that for the Lo-F/Lo-M words was 5.80.

A fourth list consisted of nonsense itemswhich were obtained by transforming theLo- F/to -M words according to the followingrule: Replace each consonant with the nextconsonant in the alphabet, but retain thesame vowels. The nonsense condition wasincluded to sample the lower extremes of themeaningfulness dimension. That is, whilethe Lo -F/Lo M words have little semanticcontent for Es of this age, their possiblecloser resemblance to known English words(in terms of orthographic structure andpronunciability) might afford more meaningand/or associations for As than would non-sense words. The frequency judgmentresults for nonsense words appeared tosupport this speculation (of. Ghata la & Levin,1974) , and led to the present predictionof inferior discrimination learning with non-szLse words in comparison to 111-F materials.

All lists consisted of 15 pairs. Twoversierg of each list contained differentrandom pairings of items. For each versionof each list, one member of each pair wasselected as correct; in a second list theother members of the pairs were correct. Ofthe 20 as in each condition, five wereassigned to each list variation, All lists wereconstructed such that: (a) the 15 pairsoccurred In three random serial orders; (b)within any order, the correct members a thepairs occurred approximately equally oftenin the right and left positions; and (c) acrossorders, the correct member of a pair occurredno more than twice in the same position. Theitems in the pairs were typed side by side on

4

5 by 8 inch plain white cards and placed innotebooks. The correct members of the pairswere starred.

Procedure

The is were run individually in aprivate room in the school building. E.,:h

received one silent (no-guess) anticipationtrial followed by four anticipation responsetrials. The pairs were presented at a three-second rate timed by means of an electronicmetronome and Es Indicated their choicesby pointing.

Results and Discussion

Mean discrimination learning perfor-mance over four trials wan 51.10, 46.80, 46.70,and 41.00 in the Hi-F, Lc c/Hi-M, Lo-F/Lo-M,and Nonsense conditions respectively. Inkeeping with the Ghatala and Levin (1974)analyses, Dunnett comparisons (utilizingthe Hi-F condition as the "control group")were conducted to assess the predictedeffects. According to this procedure, itwas found that although the Nonsensecondition was significantly inferior to Hi-F,neither of the Lo-F conditions differedsignificantly from Hi-F (a is .05). Thus, whilethe prediction of inferior performance in theNonsense condition was confirmed, the predictionof superior performance in the Lo-F/Hi-Mcondition was not.

The lack of difference between the Lo-F/and Hi-F conditions was Puzzling in

light of differences obtained with these samematerials in a frequency judgment task (Ghatala& Levin, 1974). However, an inspectionof the sample variances revealed that thevariance for the Lo-F/Hi-M group was twicethat of Hi-F and three times that of Lo-F/Lo-M,suggesting the operation of factors peculiarto Lo- F/Hi -M Es, In an attempt to ascertainthe reason for the large variation amongindividuals in the Lo-F/Hi-M condition, theDA in this group were administered thedefinitions test originally used by Ghatalaand Levin.

On the definitions test, the as werepresented each of the 30 Lo- F/Hi -M wordsand required to pronounce and then define eachone. The results obtained from this procedureclearly indicated that while the item haduniformly high imeaningfulness (using thecriteria previously described), individual Esexhibited considerable variation in their abilityto pronounce and define the words. Ofparticular interest in this regard was the sig-

nificant < .01) correlation, r .68, betweennumber of wards correctly recognized( pronounced and defined) and total correcton the discriv.ALation task: a trend which pro-vides evidence in support of the originalhypothesis , in that as As' semantic knowledgeof the words increased so did their discrimina-tion learning scores.

There were two stages to the definitionstest--S first had to pronounce each word andthen define it. In cases where failed topronounce the word reasonably, E. pronounced itfor Ito define. Accordingly, three types oferrors were possible: (0,1 could fail topronounce the words correctly yet give anacceptable definition once pronounced it; (ii)a could pronounce the word correctly yet notbe able to give an adequate definition; or (iii)S could neither pronounce nor define the word.The majority of the errors (58%) tell into thefirst category, with 34% and 8% falling into thesecond and third categories respectively.Following the frequency - meaningfulness hypo-thesis, this result suggests that pronouncingthe words for Si during discrimination learningshould improve the performance of the Lo-F/Hi-Mgroup since many words which are "meaning-less" when unpronounced would become"meaningful" when pronounced by E. On theother hand, pronouncing the words for Ss inthe Hi-F and Lo-F/Lo-M groups should have

little effect. In the former case, wordsare highly familiar and are probably pronouncedcovertly by Es, and in the latter, it is doubtfulthat simple pronunciation would increase themeaningfulness of Lo-F/Lo-M words.

Another change in procedure was sug-gested by consideration of the latencies ofthe responses on the de,finitions test. The Eswere allowed up to fiv seconds to pronouncea word and then a fort er 20 seconds to defineit. Manyis, even t se who correctly definedall the words, displeVed fairly long latencies(especially in the prtuncing stage, but alsoin the definition stag ). The three-secondpresentation rate utilized in the verbal-discrimi-nation task was obviously too short for manySs in the Lo-F/Hi-M condition to pronounce(covertly) and get a meaning response for bothwords in a pair.

In this regard, the results of Ghatalaand Levin (1974) which led to the presentpredictions for verbal-discrimination learningwere obtained in an absolute frequencyjudgment task in which items were presented oneat a time for five seconds, withls explicitlyinstructed to pronounce each item to themselvesas it appeared. The present predictions mighttherefore be confirmed under conditions moreclosely resembling those in which the effectsof frequency and meaningfulness were firstdemonstrated.

mExperiment II

The purpose of Experiment II was tocompare the three word groups from Experi-ment I using a slower rate of presentatioand with E's pronunciation of the words.(Since significantly poorer performance in theNonsense condition was already demonstratedunder the procedures of Experiment Landsince this condition is not crucial to thepresent hypothesis, it was excluded here.)

Method

Subjects

The Ss were forty-eight sixth-gradechildren from an Ogden, Utah, elementaryschool demographically similar to the onein Experiment I. The Es were randomly assignedin equal numbers to thft,shree conditions ofthe experiment (Hi-F. Lo-F/Hi-M, andLo-F/Lo-M).

Materials

The lists for the three conditions werethe same as in Experiment I.

Procedure

The procedure was the same as inExperiment I with two exceptions: The Epronounced both words in a pair during theanticipation phase on all trials, and therate of presentation was slowed to five seconds.

Results and Discussion

Performance on the discriminationlearning task is summarized in the last rowof Table 1. While the three means are in thepredicted order, Dunnett tests (a in .05)revealed that relative to Hi-F, Lo-Fito-Mwas significantly inferior but LoF/141-M wasmat significantly superior.

TWILLMEAN NUMBER OF CORRECT RESPONSES OVER FOUR TEST TRIALS

SY EXPERIMENTAL CONDITION AND PERFORMANCE ON THEDEFINITIONS TEST (EXPERIMENT

Hi-F Lo-F/Hi-M Lo-F/LoM

High Definitions 47 . 75 53.22 40.33

(N -B) (Ni.9) (N -6)

Low Definitions 47.12 45 . 00 41.90

(Nise) (117) (N20140)

Weighted Mean 47 .44 49.62 41.31

6,6

Once again, in order to analyze thisresult more fully, the definitions test wasadministered to Es in the L-F/Hi-M condition.As in Experiment I, it was anticipated thatEs who would do well on the definitionstest were those who also did well on thediscrimination learning task. However, it ispassible that this could be due simply to thefact that, in general, "brighter" Ss (i.e.,Es who know mons low-frequency worddefinitions) are also better learners. In orderto choose between a general hypothesis (thatbrighter children learn faster) and a specifichypothesis (that children who know the mean-ings of low-frequency items learn thoseparticular items faster), Es in the Hi-F andLo -Fito-M conditions were also administeredthe Lo-F/Hi-M definitions test. To beconsistent with the procedure followed in thediscrimination task, E pronounced all of thewords and, after hearing each word, A wasrequired to define It

The results are presented in the bodyof Table 1. The 48 Ss were divided into twoapproximately equal-sized groups based ontheir definitions test performance. It may beseen that only in the Lo-F/141-M condition isthere a substantial effect of knowing thedefinitions an discrimination performance.Looked at another way, for those Es whoknew most of the definitions (High: 27 ormore correct out of 30), a nested comparisonrevealed that Lo-F/Hi-M Es were superior toHi-FAB tat 2.03, a 42, _p < .05); but for

8

those who knew fewer definitions (Low:between 17 and 26 correct out of 30), nosignificant difference was observed (ILI< i).

Thus, consideration of only the overalldiscrimination performance did not yieldsignificant differences between Es in theLo-F/Hi-M and Hi-F conditions, contrary topredictions. Despite the procedural changesfrom Experiment I to Experiment II (E-rronuncia-tion of the items and a slower press-zetionrate), there was still appreciable variationamong as in their knowledge of the meaningsof the Lo-F/Hi-M words, However, theanalysis of discrimination learning scoresfor the Hi-F and Lo-F/111-M Ls as a functionof their scores on the definition test clearlysupported the hypothesis that low-frequencywords are better discriminated than high-frequency words as long at Es know themeanings of the low - frequency words.

Taken together, the results of Experi-ments I and II provide 30410 support for thenotion that background frequency influencesthe difficulty of discrimination learnirxt, butthat the meaningfulness of the stimuli is avariable moderating the effects of frequency.The data of the first two experiments wereless than satisfying, however, becausecomplete control over meaningfulness was notobtained with the materials used. Consequently,in the remaining experiments new high- and low-frequency materials were selected so that bettercontrol over meaningfulness could be obtained.

IVExperiments III and IV

In Experiment III, discrimination learn-ing with revised Hi-F and Lo-F/Hi-M verbalmaterials was compared. As in the previousexperiments, it was expected that the Lo-F/Hi-M materials would produce superior learning.However, in order to demonstrate that thisfinding could not be attributable simply to theparticular idiosyncracies of the materialsselected, in Experiment IV the same materialswere compared under free-recall learning con-ditions. Based on the well-known differentialeffects of word frequency on recognition- andrecall-type measures of learning (cf.Kintsch,1970), it was expected that the presentLo-F/Hi-M materials, while superior to Hi-Fmaterials in discrimination learning would beinferior in free recall.

In addition, experimental line-drawingscorresponding to the verbal stimuli wereincluded to see if these would be similarlyresponsive to background-frequency manipu-lations., Although unforeseen at the time,the inclusion of pictorial materials providesdata relevant to current theorizing aboutpicture-word differences in learning (cf.Pelvic), 1971). Such issues will be discussedfollowing a presentation of the experimentalresults.

Method

Materials

Fifty-two concrete nouns were selected,with half designated as Hi -F and half as Lo -F`as determined from Carroll, Davies, and Rich-man's (1971) word-frequency norms (third-gradelevel). By selecting from actual materialsused by children, we sought to obtain morerealistically-based high- and low-frequency

words than those determined from more remotenorms such as those of Thorndike and Lorge(1944). In particular, the Carroll et al.norms are derived from samples of children'sreading materials, grade level by gradelevel. Overall, the Lo-F words (mean of 7.5occurrences in third-grade materials, range of1 to 19) appeared in such samples much lessfrequently than the Hi-F words (mean of 351.5,range of 232 to 785). An attempt was made tomatch Hi-F and Lo-F words with respect totheir general object class (e.g., "dog" with"ape"; "window" with "chimney") as much aswas possible. Line-drawings of each of theseitems were also created.

The final selection of words and picturesresulted from initial pilot testings with a largersample of materials. In these pilotings, therewere two major concerns: (a) that the Hi-F andLo-F items generated from the Carroll et al.(1971) norms corresponded to our as' phenomenalexperience with such items; and (b) that Sspossessed the desired labels for each of thepictures (thereby also indicating that allmaterials were "meaningful" to Ss). To dealwith the first concern, we presented mixed-frequency word pairs aurally to beginningfourth graders, with Es instructed to circlethe letter on supplied answer sheets whichcorresponded to the word in each pair thatthey had "heard, seen, or used more often. "To deal with the second concern, we showedthe pictures one at a time to additional Ss,and asked them to label each one Followingthese procedures. it was possible to select26 Hi-F and 26 Lo-F items which met at leastan 80% agreement criterion on both thephenomenal-frequency judgment and the picturelabeling tasks. From these items, 13 Hi-Fpairs were randomly formed for the discriminationlearning task (Experiment III). with one item

9

in each pair designated correct. The matched(from the pilot studies) Lo-F items were thenselected to form comparable pairs. A randomsample of 18 of the Hi-F items and their Lo-Fcounterparts were selected for the free-recalltask (Experiment IV). Both verbal andpictorial items appeared on cards insertedinto looseleaf binders.

Subjects and Design

A total of 123 fourth graders (differeqtfrom those used in the pilot studies) from anelementary school in the Midwest participatedin the two experiments. Of these, 75 Ssparticipated in Experiment III, and 48 inExperiment IV. Within each experiment, Sswere randomly assigned in approximatelyequal numbers to the four cells of the designas defined by the combination of Frequency(Hi-F vs. Lo-F) and Modes (Words vs .Fictures).

Procedure

In the discrimination task, Sip wereshown each pair for three seconds under theanticipation method (with one silent studytrial). Additionally, g pronounced the two pairmembers during the anticipation phase in the,Word _gendiums_smly (to guarantee that eachword would be recognized) . Following theinitial study trial, two response trials wereprovidecl.

In the free-recall task, stimuli werepresented one at a time for three eeoondsapiece. As in the discrimination tusk, Enamed aloud each stimulus in the Wc..rdconditions during presentation. Four alternatingpresentation trials and S-paced recall trialswere provided.

Results and Discussion

Mean performance on the two tasks ispresented in Figure 1. On the free-recalltask, E.-produced labels that were synonymousto the intended ones were scored as correctin the Picture conditions.

Since there were slightly unequalnumbers of E.,: in the four discriminationlearning conditions, the harmonic mean wascomputed ails 18.72) and unweighted meansanalysis of variance applied to the twofactors. According to this procedure,pictures were discriminated better than words(Ent 20.26, df 10 1/71, 2, < .001), a result in

10

accord with previous data for both children(e.g., Wilder & Levin, 1973) and adults(e.g., Rowe & Paivie,1971a). Moreover, asanticipated with meaningful naterials, Lo-Fwas found to be superior to Hi-F L= 7.36,ga- v71, 2 < .01). The Frequency by Modesinteraction was not significant (j < 1).

The picture-over-word effect remainedon the free-recall task (E 10.76, su- 1/44,

< .01), which is consistent with previousresults (e.g. , Cole, Frankel, & Sharp, 1971;Paivio & Csapo, 1969). However, as wasexpected, the frequency effect reversed:that is, Hi-F stimuli were better recalledthan Lo-F stimuli =15.68, ant V44, 2 <.001). Once again, the interaction was non-significant (El* 1.93, 1/44, 2> .10).

Thus, the major premise of the presentresearch (y.a. , that background frequency isnegatively related to discrimination learningperformance) was supported. Conversely, apositive relationship between backgroundfrequency and free-recall performance wasobserved, suggesting that different cognitiveprocesses were evoked by the two tasks eventhough they included the same materials.

Since no interaction between frequencyand stimulus mode was obtained in either task,it might be concluded that background frequencyinfluences the learning of pictorial as wellas verbal materials. While this seems obviousin the case of free recall (where must storeand retrieve the stimuli's labels which areidentical for both types of material), it isless so in the case of discrimination learning(where production of the stimuli's verballabels is not required).1 Consequently,additional data to those of Experiment III werecollected to corroborate the existence of theeffect with pictures. Seventeen fourth gradersfrom the same school used in Experiment IIIwere given a much longer (26-pair) list,formed by including both the previous Hi-F andLo-F picture pairs in a single mixed list. Theobtained difference of 1.65 correct responsesin favor of Lo-F was statistically significant

'However, recent evidence (cf.Tversky, 1973) suggests that verbalprocesses are involved in ostensibly non-verbal tasks (such as the pictorial discrimi-nation task of the present study).

Dis

crim

inat

ion

Lea

rnin

gFr

ee R

ecal

l

Hi-F

Lo-

FW

ords

Hi-F

Lo-

FP

ictu

res

HiF

Lo-

FW

ords

HiF

LoF

Pic

ture

s

Figu

re 1

. Mea

n pe

rcen

t cor

rect

res

pons

es b

y ex

peri

men

tal

cond

ition

indi

scri

min

atio

n le

arni

ng {

Exp

erim

ent I

II)

and

free

rec

all (

Exp

erim

ent I

V).

u..= 3.04, giLE 16, g < .01). with 11 As perform-ing better on Lo-F pairs and only 3 As perform-ing better on Hi-F pairs (and 3 as showing nodifference).

Two recent accounts of the picture-worddifferences found in a variety of learningtasks have been offered by Paivio (1971). The"concreteness" explanation asserts thatvisual imagery is a powerful determinant oflearning efficiency, and that pictures elicitsuch imagery more directly than do theirassociated verbal labels. The "dual-coding"explanation, on the other hand, assertsthat two internal codes (imaginal end verbal)are more efficient than one, and thatpictures are more likely to produce such adual coding than words--due to the presumedgreater propensity ofais to label picturesspontaneously than to generate visual imagesfor words spontaneously.

Paivio and Csapo (1973) have providedevidence in support of the "dual-coding"(rather than the "concreteness") interpreta-tion of picture-word differences in freerecall. While this also seems to be the casein discrimination learning (as determined fromsome unpublished data of our own), thepresent study suggests that neither explana-tion is completely adequate if "backgroundfrequency" is ignored. That is, neither cansatisfactorily account for discriminationlearning differences between high- and low-

12

frequency materials of comparable concrete-ness (i.e., between high- and low-frequencywords or between high- and low-frequencypictures). However, in light of the presentfinding that pictures with low-frequencylabels were better discriminated than werethose with high-frequency labels, Paivio'ssuggestion that pictures are dually encoded(i.e., verbally, along with their nonverbalimages) is eminently reasonable in its ownright, and should be considered in conjunctionwith an alternative hypothesis previouslyoffered to account for picture-word differencesin discrimination learning (Ghatala, Levin, &Wilder, 19 73.)

At the same time, even though a-"background frequency" effect was producedwith pictorial materials here, its locus isdifficult to trace. That is, are low-frequency pictures discriminated better thanhigh-frequency pictures because theirrespective verbal labels differ in backgroundfrequency gr_because their respectiveobject referents differ in background frequency,or both? Certainly the word "giraffe" isencountered less frequently by (NorthAmerican) As than is the word "cat"; yet §ffialso encounter more cats than giraffes intheir day-to-day existence. Although thepresent research does not lend itself to ateasing apart of these naturally-correlatedfactors, it would seem possible to do so.

spo 01110-747.0

VGesiend Discussion

The present series of experimentsindicates that background frequencyinfluences the diecrimination learning ofverbal materials in a manner prescribed byWeber's Law. More importantly, however, theexperiments have shown that the negativerelationship between frequency and discrimina-tion learning is evident only for verbalmaterials which are meaningful toThe analysis in Experiment II, which showedthat only those As in the Lo-F/Hi-M groupwho actually knew the definitions of theLo-F/Hi-M words performed significantlybetter than comparable is In the Hi-F group,appears to be particularly strong evidence insupport of the frequency/meaningfulnesshypothesis advanced by Ghata la and Levin(1974) .

The first two experiments revealed sub-stantial individual differences among childrenin their knowledge of word meanings,which in turn illustrates the importance ofselecting high- and low-frequency wordsthat are not contaminated by differences inmeaningfulness. While such individualdifferences in word knowledge might beexpected to be reduced with adult Is, it isnot unlikely that they still exist (particularlywith very low-frequency words) and may beresponsible for the equivocal findings

concerning frequency effects in verbal dis-crimination learning which were describedearlier. In this regard, Experiment III demon-strated that when words are chosen which areuniformly high in meaningfulness for all is,then the negative relationship betweenfrequency and discrimination learning isslearly apparent.

Experiments III and IV substantiated thefinding of opposite effects of word frequencyin free recall and discrimination learning(the latter presumably involving recognitionmemory). Such differential effects offrequency, as well as other variables, haveled some (e.g., Kintsch, 1970; Underwood,1969) to propose that different processesor memory attributes underlie recognition andrecall.

Finally, background frequency appearedto influence the discrimination learning ofpictures as well as words although, asnoted earlier, further research is needed todetermine the locus of the frequency effect.Furthermore, in the present study the objectreferents of all pictures were meaningful tothe as. It would be of interest to determineif the discrimination learning of pictures isalso moderated by meaningfulness, as hasbeen demonstrated here for verbal materials.

13

Carroll, I. B., Davies, P., & Richman, B.The American houltesaimilEagmaivjaggk. New York: Houghton Mifflin,1971.

Cole, M., Frankel, F., & Sharp, D.Development of free recall learningin children. 2exelessaiiatakingilla-IN1X, 1971, I, 109-123.

Ekstrand, B. R., Wallace, W. P., & Under-wood, B. I. A frequency theory ofverbal-discrimination learning.j'avchgjoaical Review, 1966,566-578.

Ghatala, E. S. & Levin, J. R. Developmentaldi.lcirences in frequency Judgments ofwords and pictures. Tournal ofLicnorknental Child Psvoholoov,19?3, 11, 495-507.

Ghatala, E. S., & Levin, J. R. Discrimi-nation learning as a function ofdifferences in materials: Aproposed explanation. Memoryand Cognition, 1974, a, 395-400.

Ghatala, E. S., Levin J. R., & Wilder; L.Apparent frequency of words andpictures as a function of pronunciationand imagery. ,Journal_ ofidearaina-ArliLYeriaLibliAKL2E, 1973,A, 85-90.

Ingison, L. J., & Ekstrand, B. R. Effectsof study time, method of presen-tation, word frequency, and wordabstractness on verbal discriminationlearning- lianaLsirgweEtmenIA).7svcholoav, 1970, ja, 249-254.

Kintsoh, W. Models for free recall andrecognition. In D. A. Norman (ed.) ,lifisialL2Lhuraamissaget. New York:Academic Press, 1970.

Levin, J. R. , Ghatala, E. S. & DeRoseT. M. The effect of stimulus pre-familiarization on children'sdiscrimination learning. TechnicalReport No. 285, Wisconsin Researchand Development Center for CognitiveLearning, Madison, in press.

Paivio, A.New York: Holt & Co.. 1971.

Paivio, A. , & Csapo, K. Concrete-imageand verbal memory codes. Tournal oJiiiimiumAtiLEMbilltir, 1969, a279-285.

Paivio, A., & Csapo, K. Picture superiorityin free recall: Imagery or dual coding?rdsinititlelizr11211asiz 1973. A.,176-206.

Paivio, A. , & Rowe, E. J. Noun imagery,frequencY, and meaningfulness inverbal discrimination. Tournal oflistatmeatallagraglaiu, 1970,IL, 264-269.

Paivio, A., Yuille, J. C., & Madigan, S.Concreteness, imagery, and meaning-fulness values for 925 nouns. Jpurnalaranalignentalmagi212azitaimukIiimizlementd 1969, /I (1. Part 2).

Rowe, E. J., & Paivio, A. Discriminationlearning of pictures wad words.ZeKitgainaglidence, 1971, 21,87-88. (a)

Rowe, E. J. , & ), A. WOid.frequeacyand imager", c,ffects in verbal discrimi-nation learning. kezmeLakeedamazij'sychol r. 1971, no 319-326. (b)

Thorndike, E. L. & Large, I. The teacher'sEfiribittaaLagjilanusa, NewYork: Bureau of Publicatleas, TeachersCollege, 1944.

fills

Tversky, B. Encoding processes in recog-nition and recall. Coaativebusludgoy, 1973, 275-267.

Underwood, B. J. Attributes of memory.Psvcholoaical Review, 1969, Z.559-573.

Underwood, B. J., Broder, P. K., &Zimmerman, J. Retention of verbaldiscrimination lists as a functionof number of prior lists, word fre-quency, and type of list. Ism EDW.

16

of Experimental Psych Rim,. 1973,jag., 101-105.

Underwood, B. j. , & Schulz, R. W.MeAl2/112f11/82111LACLUILIELUICIELaaaPhiladelphia: Lippincott, 1960.

Wilder, L. , & Levin, J. R. A developmentalstudy of prounounoing responses inthe discrimination learning of wordsand pictures. jouragollageamentelfehAld Psycholoav, 1973,15., 278-28 6.

GPO 111110-ve"..

National Evaluation Committee

Helen BainPast PresidentNational Education Association

Lyle E. Bourne, Jr.Institute for the Study of Intellectual BehaviorUniversity of Colorado

Sue BuelDissemination and Installation ServicesNorthwest Regional Educational Laboratory

Francis S. ChaseProfessor EmeritusUniversity of Chicago

George E. DicksonCollege of EducationUniversity of Toledo

Chester W. HarrisGraduate School of EducationUniversity of California

Hugh J. ScottConsultantNational Evaluation Committee

H. Craig SipeDepartment of InstructionState University of New York

G. Wesley SowardsDean of EducationFlorida International University

Joanna WilliamsProfessor of Psychology and EducationColumbia University

Enocutivo Committee

William R. BushDirector, Program Planning and ManagementDeputy Director, R & D Center

M. Vere De VaultProfessorSchool of Education

Herbert J. KlausmeierPrincipal InvestogatorR & D Center

Joel IL LevinPrincipal InvestigatorR & D Center

=1111===1=111111.1

Donald N. McIsaacAssociate Dean, School of EducationUniversity of Wisconsin

Richard A. Rossmiller, Committee ChairmanDirectorR & D Center

Len VanEssAssociate Vice ChancellorUniversity of WisconsinMadison

Dan WoolpertDirector, Management SystemsR & D Center

Faculty of Principal Investigators

Vernon L. AllenProfessorPsychology

B. Dean BowlesAssociate ProfessorEducational Administration

Frank H. FarleyAssociate ProfessorEducational Psychology

Marvin J. FruthAssociate ProfessorEducational Administration

John G. HarveyAssociate ProfessorMathematics

Frank H. HooperAssociate ProfessorChild Development

Herbert J. KlattarneitrV. A. C. Henmon ProfessorEducational Psychology

Gisela LalbouvieAssistant ProfessorEducational Psychology

Jowl rt. LavinAssociate ProfessorEducational Psychology

L. Joseph LinsProfessorInstitutional Studies

James LiphamProfessorEducational Administration

Wayne OttoProfessorCurriculum and Instruction

Robert PetsoldPro: armorCurriculum and Instruction

Thomas A. RombergAssociate ProfessorCurriculum and Instruction

Dennis W. SpuckAssistant ProfessorEducational Administration

Richard L. VeneskyAssociate ProfessorComputer Science

Larry M. WilderAssistant ProfessorCommunication Arts