DOCUMENT RESUME

ED 024 099 24

By-Ramsay. James G.Concept Learning as a Function of the Type of Material and Type of Classification.Wisconsin Univ.. Madison. Research and Development Center for Cognitive Learning.Spons Agency-Office of Education (DHEW). Washington, DC. Bureau of Research.Bureau No- BR-5-0216Pub Date Jun 68Contract OEC- 5-10-154Note- 33p.EDRS Price MF-S0.25 HC-S1.75Descriptors-*Cognitive Processes, College Students, *Concept Formation, *Learning Processes, *VerbalLearning. Visual Learning, Visual Stimuli

This reports the effects of the number of relevant stimulus dimensions andfigural versus verbal stimuli on the concept learning ability of college students.Results force a consideration of mediational variables in explaining this form ofcognitive learning. A set of verbal materials analogous to a set of dimensionalizedfigural materials was constructed. The figural stimuli were 16 H-patterns, thecombinations of values of the four binary dimensions of color (red or green), size(large or small). number (one or two), and orientation (upright or tilted): the verbalstimuli were 16 nouns. four sets of four, which had been shown to be associated withfour adjectival categories: hard-white, soft-white, hard-brown, and soft-brown.Tested were two hypotheses: (1) that the difficulty of three classifications would bean inverse function of the number of relevant dimensions composing theclassifications. and (2) that figural instances would be more difficult to categorizecorrectly than verbal instances. The hypotheses were supported. but needed to bequalified because of significant interaction. Alternative interpretations of results arediscussed. (Author)

CG 003 339

Technical Report No. 58

CONCEPT LEARNING AS A FUNCTION OF THE TYPE OF MATERIAL

AND TYPE OF CLASSIFICATION

By James G. Ramsay

Report from the Project on Situational Variables and Efficiency of Concept LearningHerbert J. Klausmeier, Principal Investigator

Wisconsin Research and DevelopmentCenter for Cognitive LearningThe University of Wisconsin

Madison Wisconsin

June 1968

U.S. DEPARTMENT OF HEALTH, EDUCATION & WELFARE

OFFICE OF EDUCATION

THIS DOCUME9T HAS BEEN REPRODUCED EXACTLY AS RECEIVED FROM THE

PERSON OR ORGANIZATION ORIGINATING IT. POINTS OF WEW OR OPINIONS

STATED DO NOT NECESSARILY REPRESENT OFFICIAL OFFICE Of EDUCATION

POSITION OR POLICY.

The research reported herein was performed pursuant to a contract with the United States Office ofEducation, Department of Health, Education, and Welfare, under the provisions of the Cooperative

Research Program.

003 33'9

Center No. C-03 / Contract OE 5-10-154

,1,4rarafiost

PREFACE

This Technical Report is based upon the dissertation of James GordonRamsay. The examining committee consisted of Professors Herbert J. Klausmeier(chairman), Nathan S. Blount, Gary A. Davis, Harold J. Fletcher, and ThomasA. Ringness.

One major program of the Wisconsin R and D Center for Cognitive Learningis Program 1 which is concerned with fundamental conditions and processes oflearning. This Program consists of laboratory-type research projects , eachindependently concentrating on certain basic organismic or situational determi-nants of cognitive learning, but all attempting to provide knowledge which canbe effectively utilized in the construction of instructional systems for tomorrow'sschools.

Of critical importance to the field of human learning is the area of conceptlearning, an area in which most experimentation is designed primarily to revealtask or situational determinants of performance. Mr. Ramsay, continuing theseempirical investigations, reports the effects of the number of relevant stimulusdimensions and figural versus verbal stimuli on the concept learning ability ofcollege students. His results force a consideration of mediational variables inexplaining this form of cognitive learning.

Harold J. FletcherDirector of Program 1

CONTENTS

List of Tables and Figures

page

vii

Abstract ix

I Introduction 1

Purposes and Hypothesis of the Study 2

Method 2

Significance of the Study 2

II Review of Related Literature 3Type of Classification 3

Descriptions of Classifications 3Classifications and Similarity 5

Type of Material 5

Figural Material 5

Verbal Material 6Mediational Considerations 7

Summary 8

III Identification of Associational Norms 9

IV Method 12Experimental Material 12Subj ects 13Experimental Procedure 13Experimental Design 13

V Results 15Analysis of Variance on Total Correct Responses 15Mean Time per Instance 16Subjects' Comments 17

VI Discussion 19

Appendix: Instructions 21

References 25

LIST OF TABLES AND FIGURES

Table page

1 A Comparison of Shepard, Hovland and Jenkins' Type 1 andType VI Classifications with the Classifications in the PresentStudy Termed Two, One, and Zero Relevant Dimensions 4

2 Shepard, Hovland, and Jenkins' (1961) Types of Classificationsin Symbolic Notation 5

3 Percent of Ss that used a Given Adjective to Describe GivenNoun: Data from Ramsay (R), Underwood and Richardson(U & R), and Mayzner and Tresselt (M & 10

4 Correlation Between Association Values found by Ramsay andby Mayzner and Tresselt for the Concept White 11

5 Pearson Product-Moment Correlation Coefficients for Asso-ciational Values found by Underwood and Richardson (U &Mayzner and Tresselt (U & T), and Ramsay 11

6 Association Values for Instances Hypothesized to Fit theCategories Hard-Brown and Soft-Brown 11

The 16 Verbal Instances 11

8 The 16 Figural Instances 12

9 Summary Table for Analysis of Variance on Total NumberCorrect Responses 15

10 Mean Number Correct for Treatment Groups 15

11 Summary Table for the Analysis of Variance on Mean Timeper Instance 17

12 Labels given to the Four Categories by Ss in the R-1 VerbalTreatment Group 18

Figure

1 Six Types of Classifications (from Shepard, Hovland, &Jenkins, 1961) 4

vii

Figure

2 Performance of Treatment Groups as a Function of Type ofMaterial and Type of Classification

3 Percentage of Correct Responses as a Function of TreatmentGroup and Number of Blocks of Trials Completed

viii

page

16

17

ABSTRACT

A set of verbal materials analogous to a set of dimensionalized figuralmaterials was constructed. The figural stimuli were 16 H-patterns, the com-binations of values of the four binary dimensions of color (red or green), size(large or small), number (one or two), and orientation (upright or tilted). Theverbal stimuli were 16 nouns, four sets of four, which had been shown to beassociated with four adjectival categories; hard-white, soft-white, hard-brown, and soft-brown. Two hypotheses were tested; (1) that the difficultyof three classifications would be an inverse function of the number of relevantdimensions composing the classifications , and (2) that figural instances wouldbe more difficult to categorize correctly than verbal instances.

The Ss, 36 male college students, were run individually in one of sixtreatments, with six Ss randomly assigned to each group. The experiment wasa 3 x 2 factorial design with three types of classification (0, 1, or 2 relevantdimensions) and two types of material (figural or verbal instances). The taskwas to learn to categorize the instances into four groups of four. Instanceswere presented sequentially and Ss responded by pushing one of four responsebuttons. Correct responses were reinforced by a green feedback light over thebutton pushed; for a wrong response, a red light came on over the button pushedand a green light over the correct button. Presentation of instances and feed-back was automated. Criterion performance was correct categorization of ablock of 16 instances or 15 such blocks.

An analysis of correct responses revealed that type of classification, typeof material, and the interaction of these effects were significant sources ofvariation (p < .001). Subsequent tests showed the order of difficulty of theclassifications (from most to least difficult) was 0 = 1 > 2 for the figural mate-rial and 0 > 1 = 2 for the verbal material. In the 0 and 1 conditions, figuralmaterial was more difficult than verbal material, while in the 2 condition therewas no difference between the types of material. The Ss spent longer meantimes per instance on the figural material than on the verbal material (p < .01).The hypotheses were thus supported, but needed to be qualified because of thesignificant interaction. Alternative interpretations of the results were discussed.

ix

-

INTRODUCTION

The 'study of concept learning in the UnitedStates since 1950 has typically been approachedthrough the use of three types of stimulus ma-terial. These types might be described as con-crete, figural, and verbal materials. Concretematerials are frequently used where subjects(as) are young children and where proceduresare based on the theoretical formulations ofJean Piaget. The use of figural or verbal ma-terials would also seem to be determined, tosome extent, by the theoretical orientation ofthe experimenter. Thus, experimenters inter-ested in describing quantitatively and in manipu-lating the amount of information transmitted byvarious instances tend to use figural stimuligenerated by combinations of values of binarydimensions (e.g., Bulgarella & Archer, 1962).The use of the term "information" in this con-text is derived from the information theory ofShannon and Weaver (1949).

Underwood and Richardson (1956a), on theother hand, developed a set of verbal materialsto study concept learning as a form of verballearning. It is the comparison of dimensional-ized figural material and the verbal materialsgenerated by Underwood and Richardson towhich the present study is addressed.

Arnstine (1967) has suggested that the gath-ering of evidence for theories of learning hasproceeded in the following manner. Supposeyou are asked to judge which of two theoriesabout the taste of water is correct. One theoristwho says it is salty gives as evidence samplesof water which he has taken from the ocean foryou to taste. The other theorist gives samplesof lake water as his evidence. You, as onenaive about the taste of water, ask for moreevidence and the first theorist continued totake samples from the ocean, the second fromthe lakes. Clearly, as long as evidence isgathered in this manner, your choice is an arbi-trary one. Similarly, Arnstine argues, "...the

'facts' of human behavior are drawn from dif-ferent wells and are quite reasonably fittingfor the different theoretical buckets which con-tain them [p. 511."

The description just cited seems to be par-ticularly relevant to the study of concept learn-ing. It may be that evidence drawn from studieswhere dimensionalized figural material has beenused has little relationship to evidence fromstudies with verbal material. Where such kindsof evidence are used only to facilitate thedesign of further experiments, the matter maynot be a crucial one. But when the evidenceis used to change educational practice, thetype of material used becomes very relevant.This point can be clarified by a further example.Bruner, Goodnow and Austin (1956) conducteda series of experiments using dimensionalizedfigural material. Later, Bruner (1960) produceda highly influential set of statements about howthe education of children should be properlyundertaken. He emphasized discovery learning.It is not coincidental that this approach closelyparallels the selection paradigm developed byBruner et al. in research on concept learning.

The mOve from evidence based on dimen-sionalizeiii figural material to revision of cur-ricula may or may not be a good one. It seemsreasonable to assume that most concept learn-ing in the classroom is closely linked withverbal media. Whether or not Brunner's evi-dence applies to such media can be questioned.Ausubel (1961), for one, has been critical ofthe approach advocated by Bruner. What seemsto be needed is research in which a verbalanalogue to dimensionalized figural materialsis developed so that variables shown to beeffective with figural material can be testedon verbal material. Until such research isinitiated, the study of concept learning willproduce evidence limited to the type of materialused in the particular experiments.

1

7

PURPOSES AND HYPOTHESIS OF THE STUDY

The purposes of the study were (1) to con-struct from the Underwood and Richardson wordlist a set of verbal materials analogous todimensionalized figural materials; (2) to studyconcept learning as a function of type of mate-rial and type of classification.

Two hypotheses were tested. The first wasthat the rank order of difficulty in learning toclassify 16 instances into four categories wouldbe inversely related to the number of relevantdimensions used to determine the type of clas-sification. Thus, it was predicted that zerorelevant dimensions would be a more difficultclassification to learn than one relevant dimen-sion, and the latter would be more difficult thantwo relevant dimensions. The second hypotheseswas that learning to classify figural instanceswould be more difficult than learning to classifyverbal instances. Difficulty was defined as thenumber of errors made to criterion.

METHOD

Thirty-six volunteer Ss, all males, partici-pated in the experiment. The figural materialconsisted of 16-H-patterns which varied onfour binary dimensions. The Underwood andRichardson (1956a) list and the Mayzner and

2

Tresselt (1961) judgmental procedure were usedto obtain an analogous set of 16 verbal.instances.

An S's task waS, on observing an instance,to identify its membership in one of four cate-gories by pushing one of four reponse buttons.instances and feedback information were pre-salted by automated apparatus through a pro-cedure described by Archer, Pourne, and Brown(1955). The experiment was conducted in thelaboratory facilities of the Wisconsin R & DCenter.

Type of material, figural or verbal, was oneof the independent variables used in the study..The second was type of classification. Thethree levels of this variable were identified interms of number of dimensions relevant to aparticular category of instances: 2, 1, or 0.

SIGNIFICANCE OF THE STUDY

To the author's knowledge, the comparisonof dimensionalized materials and analogousverbal materials identified from associationaldimensions had not been made before this studywas conducted. It was thus the first steptoward more extensive research where variablesshown to be effective with figural materials canbe applied to verbal materials. Since mostconcept learning in the schools takes placewith verbal media, the step is an important one.

(

II

REVIEW OF RELATED LITERATURE

Two search procedures were used for thisreview. First, listings from PsychologicalAbstracts (Jan., 1950 - Jan., 1967), ConceptLearning and Problem Solving: A Bibliography,1950-1964 (Klausmeier, Davis, Ramsay, Fred-rick, & Davies, 1965) and Tasks Employed inConcept Identification and Problem SolvingStudies (Davies, Cooper, Davis, & Stewart,1965) were consulted. Secondly, bibliographiesof articles and books related to concept learningwere scanned.

One area not covered by this review is theapproach to concept learning developed by jeanPiaget. As Bourne (1966) has pointed out,Piaget's influence is appreciable, but hisunique methods, his broad domain of study,and his divergence from conventional Americantheory make it difficult to compare his theoryand datk to that of American psychologists.Similarly, there is a vast literature on verballearning, but much of it is not relevant to theproblems posed in the present study. Thus,only those.studies of verbal learning (1) whichare specifically concerned with concept learning,(2) which present experimental paradigms com-parable to the one used in the present study, or(3) which have theoretical relevance for thepresent study have been included.

TYPE OF CLASSIFICATION

One dictionary definition of "classification"is "...arrangement according to some systematicdivision into classes or groups [Websters NewWorld Dictionary, 1966]," This definition isacceptable for this study. The term will beused synonomously with "categorization." Theterms used to describe the result of a classifi-cation will be "sets of instances," "groups,""categories," and "concepts."Descriptions of 'Classifications

Shepard, Hovland, and Jenkins (1961) gavea thorough description of the ways in which

instances could be classified, and tested thedifficulty of different classifications. Theydefined a classification as a grouping of agiven set of stimuli into two or more mutuallyexclusive and exhaustive classes.

The stimuli which Shepard, Hovland andJenkins used for explanatory purposes wereeight unique instances, the combinations ofvalues of the three binary dimensions of size(large or small), color (black or white), andshape (triangle or square). Their descriptioncan be applied to any set of eight instancesbased on three binary dimensions, but is limitedto types of classifications which resulx in thepartitioning of the eight instances into two setsof four instances each. The authors identifiedsix types of classifications which are illustratedas CI through CVI in Figure 1.

Though the present study used sixteen in-stances (four sets of four), some of the Shepardet al, classifications are comparable to theclassifications used in the present study, inwhich either two relevant dimensions (R-2),one relevant dimension (R-1), or zero relevantdimensions (R-0) were used to group instances.Table 1 is an illustration, in binary notation,of CI and CVI of the Shepard, Hovland aridJenkins study and groups R-2, R-1, and R-0of the present study. Each row of digits withina set represents an instance; each column ofdigits within a set represents a dimension whichcan take one of two values for a given instance.

If the first binary digit is dropped from eachinstance in R-2, sets W and X and sets Y andZ of the R-2 display become identical to setsA and B of CI. Similarity also exists betweenCVI and R-0. An inspection of the columnsof binary digits in each of the sets of instancesof CVI and R-0 reveals that each value ofevery dimension is represented equally oftenin a given set of instances. In both CVI andR-0 there are thus no consistent cues to aidin categorizing the instances into their appro-priate sets. There is no obvious comparison

3

zbEL AEA ,Lo A,A LA,a

6.0CIO 0E3 60

Figure 1.

EAO 6A

Six Types of Classifications (from Shepard, Hovland, & Jenkins, 1951)

TABLE 1. A comparison of Shepard, Hovland and Jenkins' Type I and Type VI Classifications withthe Classifications in the Present Study Termed Two, One and Zero Relevant Dimensions

C-1 CVI

A B A

000 100 000 100001 101 110 010010 110 101 001011 111 011 111

R-2 R-1 R-0

X W X X

0000 0100 1000 1100 0000 0100 0001 1000 0000 0001 0010 00110001 0101 1001 1101 0011 0111 0010 1011 1001 1011 1000 10100010 0110 1010 1110 1011 1101 0101 1100 0110 0100 0111 01010011 0111 1011 1111 1010 1110 0110 1111 1111 1110 1101 1100

of R-1 to any of the classifications describedby Shepard, Hovland and Jenkins. It can benoted that dropping the first digit from instancesin sets W and X and the last digit from setsY and Z in R-1 makes W and X, and Y andZ, identical to A and B of C-1. But, the shiftfrom dropping the first to dropping the last digitin accomplishing this equivalence indicates thatthe relationship between R-1 and CI is morecomplex than the relationship between R-2 andCI.

Shepard, Hovland and Jenkins found, undera variety of different conditions, a consistentrank-order of difficulty of learning the classifi-cations. From easy to difficult, the order ofclassifications was: CI; CII; CIII = CIV= CV; CVI. Shepard and Chang (1963) founda nearly identical rank order of difficulty usingcolors obtained from a Munsell color chart.Davis and Bourne (1965) found CI was easiestto learn, CVI hardest, with CII, CIV and CV of intermediate difficulty. Thesefindings suggest that the difficulty of the learn-

4

ing of classifications used in the present studywould be R-2 (easiest), R-0 (most difficult),and R-1 (intermediate difficulty).

Several authors (e.g. , Hunt & Hovland,1960; Hunt, 1962; Bruner et al. , 1956) havedescribed classifications of instances accord-ing to conjunctive and disjunctive rules. Con-junction corresponds to the intersection ofsets, the verbal connective being "and" as inthe concept "red and circle." Disjunctioncorresponds to the union of sets , the verbalconnective being "or" as in "red or circle."Disjunction can be further described as inclu-sive or exclusive. An example of the former is"red or circle or both;" an example of the latteris "red or circle but not both."

It has been shown that concepts based onconjunctive rules are easier to learn than thosebased on disjunctive rules (Neisser & Weene,1962; Conant & Trabasso, 1964; Haygood &Bourne, 1965). In the Haygood and Bournestudy it was also shown that difficulty was adirect function of length of rule. The increase

in difficulty due to disjunction and length ofrule may account for Shepard, Hovland, andJenkins' results. To reconsider the classifi-cations illustrated in Figure 1, let v stand fordisjunction; & stand for conjunction; and theletters W, B, L, Sm, T, and Sq stand for white,black, large, small, triangle, and square, re-spectively. The six classifications CI throughCVI have been expressed in symbolic form inTable 2. CI is based on shorter rules thanCVI and there is a regular increase in thenumber of disjunctions from CI to CVI.Although the rules needed to describe the clas-sifications in the present study are longer dueto the introduction of a fourth dimension, itcan be demonstrated thac a similar increase inlength of rule and number of disjunctions occursfrom R-2 to R-1 to R-0. A corresponding in-crease in difficulty for these latter classifica-tions could thuS be expected.

TABLE 2. Shepard, Hovland and Jenkins' (1961)Types of Classifications in SymbolicNotation

Clas-sifica-tion

SET A

(B)II (B&T)v(W&Sq)

III (B&L)v(T&Sm)IV (B&L)vT&[(L&W)v(Sm&B)]V (B&T)vSq&[(B&L)v(W&Sm)]

VI T& [(B&L)v(W&Sm)]vSq& [(W&L)v(B&Sm)]

SET B

(W)II (W&T)v(B&Sq)

III (W&L)v(Sq&Sm)IV (W&Sm)vSq& [(Sm&B)v(L&W)]V (W&T)vSq&[(W&L)v(B&Sm)]

VI T& [(W&L)v(B&Sm)]vSq& [(B&L)v(W&Sm)

Classifications and Similarity

Shepard, Hovland and Jenkins (1961) sug-gested that it would surely be easier to learnto make one response to four horses and anotherto four dogs than to make the first response totwo horses and two dogs and the second to theremaining two horses and two dogs. It hasbeen shown for concrete materials (Baum, 1954;Buss , 1950), for figural materials (Oseas &Underwood, 1952) and for verbal materials(Underwood, 1957; Neuman, 1957) that asinstances of different concepts being learned

simultaneously become more similar, the learn-ing of the correct placement of instances be-comes more difficult. Underwood (1957) con-structed three lists of nouns which varied onintralist similarity. Each list was a set of 16nouns, four instances each for the conceptsround, small, white and soft. In List 1, in-stances of each concept had no associationswith other concepts. In List 2, an instanceof a given concept also had an associationwith one other concept (e.g. , "round" was theappropriate response for the noun "bean," but"bean" also had an association with the con-cept "small"). Instances on List 3 had anaverage of 1.9 associations with conceptsother than the appropriate one. Underwoodfound concepts for List 1 to be significantlyeasier to learn than those for Lists 2 or 3.

In the present study, intralist similaritycannot be said to vary since the same instancesare used in each of three classifications.However, similarity between instances of dif-ferent concepts does vary inversely with num-ber of relevant dimensions, and based uponthe research cited above, the prediction thatincreasing the number of relevant dimensionswill facilitate performance is a tenable one.

TYPE OF MATERIAL

In this section, research on concept learningconducted with figural and verbal instances willbe discussed and implications of the theoreticalconstruct of mediation will be considered.

Figural Material

Hull (1920) defined a concept as a commonresponse to dissimilar stimuli having commonelements. A problem with this definition mayoccur if it is applied in its narrowest context.The definition suggests that the proper studyof concept learning is limited to the construc-tion of instances of concepts which have im-mediately observable commonalities. Hull's(1920) initiation of the study of concept learningwas a classic experiment and certainly a con-tribution, but may have set a precedent for theuse of figural materials. Smoke's (1933) studyof the relative contributions of positive andnegative. instan'ces to concept learning alsoemployed figural material. Gibson in 1940proposed a theory of verbal learning whichstressed the importance of stimulus generali-zation, a construct which came to play an im-portant role in considerations of concept learn-ing (e.g., Buss, 1950; Baum, 1954; Shepard,1957; Shepard, 195; Shepard & Chang, 1963).It is somewhat ironic that Gibson's own testsof her theory were conducted with figural

5

material (Underwood, 1961). Until as late as1956, when Underwood and Richardson pub-lished verbal materials for use in studyingconcept learning, the main materials used werefigural.

The use of figural material does have ad-vantages. One is that such material providesthe same kind of control over extra-experimentalfactors that nonsense syllables do in the studyof verbal learning. Dimensionalizing the ma-terials adds a further element of control since,with this operation, a finite population of in-stances is generated which can be dividedunambiguously into positive and negative in-stances. Outside of the laboratory, such con-trol over instances is unattainable. Shipstone(1960) has suggested that this is the very rea-son why dimensionalized figural material haslimitations, since ambiguity is characteristicof most situations in the real world.

In two studies (Ramsay, 1965; Fredrick,1965), the effects of figural and verbal instanceson the identification of concepts were tested.The task was what Bourne (1966) has describedas a selection paradigm. The figural instanceswere the 64 combinations of six binary dimen-sions. The verbal instances had the values ofdimensions typed in words on the cards ratherthan represented directly. Thus, a figural in-stance might have shown two small red spottedcircles with one border surrounding them, whilethe comparable verbal instance gave this sameinformation in words. Ramsay found that figuralinstances led to significantly shorter times tocriterion. Fredrick, who used the figural in-stances as a practice task, found no differencesbetween these instances and the verbal ones.

These studies, however, do not appear tocapitalize on the inherent differences betweenthe two types of material. Had the verbal in-stances been written in Swahili rather thanEnglish, non-Swahili speaking Ss could stillhave learned the appropriate classificationinto positive and negative instances since thewords themselves could be regarded as config-urations giving rise to observable cues con-sistently associated with positive and negativeinstances. Another approach to the construc-tion of verbal material is to relate the instancesand concepts by association value rather thanby direct symbolization of figural material.

Verbal Material

Underwood arid Richardson (1956a) devel-oped verbal concept learning materials consist-ing of a list of nouns and a set of adjectivalcategories associated with the nouns. Theadjectival categories were based on sensoryim pre s sions (e . g . , round , smelly,, yellow,,

6

rough) and the nouns stood tor concrete objects(e.g., brick, pill, cabin, whale). Each nounwas presented to a sample of 153 college stu-dents who were given four seconds to respondwith the first sensory impression that theythought of to describe the noun.

Data were tabulated in terms of the percentof Ss who gave the same adjectival responsefor a particular noun. For example, for thestimulus word "apple," 67% of the Ss respondedwith the adjective "red," 19% with "round,"and 5% with "sweet." The published table wasa list of the 213 nouns with entries after eachnoun for the adjectives which 5% or more ofthe Ss gave as a description of the noun. Therewere 40 adjectives which met this 5% criterion.The adjectives were considered to be conceptsand the subsets of nouns associated with theadjectives, instances of those concepts. Inthe typical experiment, S was presented witha list of several instances of different conceptsand learned to respond with the correct conceptfor each instance.

With the material, sets of instances withdifferent mean association values for the sameconcept could be constructed. These meanshave been termed levels of a variable called"dominance level" (Underwood & Richardson,1956b). It has been shown that ease of conceptlearning is a direct function of dominance level(Underwood & Richardson, 1956b; Coleman,1964). Other factors which have been shownto improve performance have been increasingthe associative rank of instances (Mednick &Halpern, 1962), increasing the variance of thedominance values of a set of instances (Freed-man & Mednick, 1958), increasing the speci-ficity with which Ss are instructed (Underwood& Richardson, 1956b), and decreasing the de-gree of overlap of associations among instancesof different concepts (Underwood, 1957).

Runquist and Hutt (1961) selected from theUnderwood and Richardson materials four verbalinstances each for the concepts round, soft,sharp, and slimy and made comparable sets ofpictorial instances. For example, the corres-ponding pictorial instances for the verbal in-stance "bed" was a line drawing of a bed.Both the word "bed" and the picture bed wereinstances of the concept "soft." Two formsof each pictorial instance were constructed,one which emphasized the characteristic thatmade the instance an exemplar of the concept,and one in which the characteristic was notemphasized. The emphasized form showed a"soft and billowy" bed while the other formshowed a bed which looked "like an army cot."

Subjects were run individually and in onlyone condition, seeing each of the 16 instances15 times, the set of 16 instances being shuffled

after each block. The Ss were instructed torespond verbally with a descriptive word withinthree seconds of the presentation of each in-stance. Correct responses received a "right"from the experimenter (E). Sixty Ss were runin a 3 x 4 factorial design with three levels oftype of material (Verbal, Picture Dominant, orPicture Nondominant) and four grade levels(Ss were freshmen, sophomores, juniors orseniors in high school). Both main effectswere statistically significant while the inter-action was not. Verbal instances resulted ina significantly higher mean number correct re-sponses' than the Picture Dominant instances(p < .01) and Picture Dominant instances re-sulted in a significantly higher mean numbercorrect responses than the Picture Nondominantinstances (p < .02).

Runquist and Hutt offered two interpretationsfor the results. The first was that verbal in-stances resulted in higher performance becausethe stimuli and responses in the verbal condi-tion were given in the same medium. Thisstimulus-response compatibility allowed thehighly likely associations between instancesand concepts to be used without Ss necessarilyforming "an image" of the word as an objectwhen he responded. The second interpretation,and a more plausible one according to theauthors, was that three of the concepts usedin the study (sharp, soft, slimy) were tactualrather than visual, that pictorial instances ofthese concepts would tend to be responded towith visual rather than tactual terms, and thatthe visual descriptive responses would tend tointerfere with the responses designated as thecorrect ones.

Both interpretations are subject to criticism.It could be argued that the visual presentationsof words and pictures were the same mediumrather than different media. Also, there wasthe omission of a simple test which might havegiven support for the second interpretation.The test would have been to compare numberof correct responses for "round" (a visual con-cept) with number of correct responses for eachof the tactual concepts. According to the sec-ond interpretation it could be predicted that thenumber of correct responses for the visual con-cept would be greater than for the other con-cepts. Although this result could not have beenconstrued as conclusive evidence, it wouldhave supported the interpretation.

In the Underwood and Richardson (1956a)scaling technique, the dominance value for aconcept and an instance of that concept wasthe percentage of Ss who responded to the in-stance with the concept. With this approachthere seems to be the implicit assumption thatdifferent dominance values of two concepts

which share the same instance reflect a hier-archy of responses for a particular S. Forexample, because more Ss responded with"red" to the stimulus word "apple" than with"sweet," "red" is inferred to be the more highlydominant response for the average S. Thisapproach obscures individual differences.Clearly, for those Ss who responded with"sweet" to "apple," "sweet" is in Underwoodand Richardson's terms the dominant responsefor these Ss.

The judgmental technique of Mayzner andTresselt (1961) has a certain advantage toUnderwood and Richardson's associationalmethod. Mayzner and Tresselt's Ss were pre-sented with a list of 300 nouns and judged theinclusion of the nouns in none, one, or morethan one of six adjectival categories. Whilethis approach limited the number of conceptsfor use in latter experiments, it did allow Ssto respond with more than one concept to aparticular instance, rather than being limitedto one concept per instance as in the Underwoodand Richardson study.

Medi ational Considerations

Kendler (1961) and her associates have sug-gested that S's covert response may serve amediating function. This hypothesis was usedto clarify data on the effects of reversal andnonreversal shifts. In the task typically usedby Kendler, the stimulus material was figuraland varied on binary dimensions. Initiallyinstances were placed into a correct categorydepending on one value of one dimension. Inthe nonreversal shift, correct categorizationbecame based on one value of another dimen-sion, while in the reversal shift situation cor-rect categorization became based on the othervalue of the relevant dimension of the initialtask. If the relevant dimension in addition tothe relevant value of that dimension was rein-forced, the reinforced dimension could serveas a mediator, in which case it couid be pre-dicted that the reversal shift would be easierto learn than the nonreversal shift. It wasshown that rats and young children performedbetter on the nonreversal shift (Kelleher, 1956;Kendler, Kendler, & Wells, 1960), while olderhuman Ss performed better on the reversal shift(Buss, 1953; Kendler & D'Arnato, 1955; Harrow& Friedman, 1958). These data were consideredto support the hypothesis that the relevant di-mension served as a mediator and that theavailability of the mediator facilitated perform-ance.

If relevant dimensions serve a mediatingfunction for more mature Ss, the college stu-dents used in the present study might perform

7

better on classifications with more relevant di-mensions. In the R-0 condition, no such medi-ators could be used, while the R-2 conditionwould allow two dimensions to serve as mediators.

If Ss in the verbal condition were able toarrange the verbal instances into syntacticalunits, these units might also serve a mediatingfunction. For example, the disparate verbalinstances "cork" and "enamel" have no commondescriptive adjective in the Underwood andRichardson tabulations. However, an S whosees that both instances are categorized underresponse button 4 might form a mediating sen-tence like "Enamel the four corks" so that upoilater presentations of the instances , the medi-,ator would help him categorize both instancescorrectly. No such easily formed mediationalunits wouid seem to be available for the dis-parate figural instances "one large green up-right 14" and "two small red tilted H's." Ver-bal mediators like "Enamel the four corks" arerelated to a consideration of memory load.Miller (1956) has suggested that a considerableamount of information can be stored as long asit is recoded into a relatively small number ofsymbols. The implementation of verbal medi-ators would allow information to be recodedand to be recalled in what Miller has termed"chunks." The subsequent reduction of thenumber of units to be recalled would tend tofacilitate performance. If it is assumed thatthe verbal instances lend themselves more eas-ily to such recoding it could be predicted thatSs in the verbal condition would have a betterchance of correctly categorizing instances.

8

SUMMARY

It has been suggested: (1) that the threetypes of classifications used in the presentstudy are in some ways comparable to thosedescribed by Shepard, Hovland, and Jenkins(1961); (2) that based on the results of Shep-ard, Hovland and Jenkins and others who haveused the saute classificationS, the rank orderof difficulty of the three types of classificationsused in the present study could be expected tobe R-0 (most difficult), R-1 (intermediate dif-ficulty), R-2 (easiest); (3) that the increasingcomplexity and length of rules needed to des-cribe the classifications symbolically are basesfor explaining why this rank order might beobtained; (4) that decreasing the number ofrelevant dimensions increases the similaritybetween instances of different concepts andincreases the difficulty; (5) that increasingthe number of relevant dimensions may allowthese dimensions to serve as mediators andthus to facilitate performance. These pointsallow the firm prediction that the rank order ofdifficulty for the three classifications will beR-0, R-1,. R-2. Further, it was suggestedthat Ss in the verbal condition would have anadvantage in using verbal mediators as mne-monic devices. Such mediators would allowthese Ss to reduce memory load by "chunking"the instances into units which were more easilyrecallable. It is thus predicted that the figuralcondition would be more difficult than theverbal condition.

III

IDENTIFICATION OF ASSOCIATIONAL NORMS

Underwood (1957) characterized the manipu-lation of similarity in figural materials as cen-tering on commonalities of physical character-istics, while in his study of verbal conceptlearning he focused on similarity as a functionof commonalities of descriptive characteristicsof nouns. Underwood's comparison was usedas the basis for creating analogous figural andverbal materials in the present study. In itssimplest form the analogy can be made betweenfigures which all have the color green in com-mon and nouns which all have the association"white" in common. Further, just as figuralstimuli can vary along a physical dimensionlike color, so can verbal stimuli vary along anassociational dimension of color. Finally,where sets of figural stimuli can be categorizedwith respect to relevant physical dimensions,so can sets of verbal stimuli be categorizedwith respect to relevant associational dimen-sions.

The Underwood and Richardson (1956a) listprovided the initial source for verbal instances.Many of the forty adjectives on the list couldbe considered to be values or dimensions.Some of these dimensions and values are size(small, big), hardness (hard, soft), texture(slimy, smooth, fuzzy, sticky, rough), weight(light, heavy), and taste (sweet, sour-bitter).Ideally, the set of verbal instances used in thepresent study would have been associated withvalues on four such dimensions. However, itwas impossible to identify 16 verbal instances,each of which was associated with a uniquecombination of values of four binary dimensionsas was possible with the figural material. Sucha procedure was prohibitive because of incom-patible joint associations. For example, anoun described as red, sweet, and juicy wasnot square. In a more general sense, it wouldappear that associational dimensions in theEnglish language are not orthogonal. Given anoun for which some values of associational

dimensions are present, the presence of othervalues of other dimensions can, to some ex-tent, be predicted.

The next step was an attempt to identifyverbal instances of the four categories basedon the conjunction of values of two binary di-mensions. The four most likely categorieswere: hard-white, soft-white, hard-brownand soft-brown. Even at this reduced level ofcomplexity, however, the Underwood andRichardson (1956a) list of 213 nouns was notsufficient to obtain four instances for each ofthe four categories so recourse was made tothe Mayzner and Tresselt (1961) procedure.

Underwood and Richardson gave 213 nounsto 153 Ss and identified 40 adjectives used todescribe the nouns. Mayzner and Tresselt tooksix of these adjectives and obtained adjectivaldescriptions from 100 Ss for each of 300 nouns(the 213 used by Underwood and Richardsonplus 67 more). In the present study three ofMayzner and Tresselt's adjectives (white,hard, small) and three other adjectives fromUnderwood and Richardson's material whichhad not been used by Mayzner and Tresselt(brown, soft, big) were listed as headings ofsix columns of a response sheet, and 24 nounswere listed as headings of rows on the responsesheet. Sixteen of these nouns were common tothe Underwood and Richardson and the Mayznerand Tresselt lists; the remaining eight wereidentified by the present author as possibleinstances for the categories hard-brown andsoft-brown, the two categories where therewere missing instances. The twenty-four wordswere:

acornbeer bottlebread*bronze

closet** ivory*cork* linen*enamel* loamgarlic** mahogany

potteryrabbit*salt*sheep*

chamois* gavel mink skull*chestnut* grass** moccasin* snail**

The starred nouns were taken from the Under-

9

wood and Richardson Mayzner and Tresseltlist. Those with two stars were selected ran-domly from the Underwood and Richardson listas buffer items which had no associations withhard, soft, brown or white. The unstarred nounswere those identified by the present author.

Twenty volunteer Ss, 13 males and 7 femalesfrom the staff of the Wisconsin R & D Center,responded to the nouns by checking those adjec-tives which could be used to describe a givennoun. These data and the data from the Under-wood and Richardson and the Mayzner andTresselt lists all took the same form. In eachcase, the percent of Ss who responded to agiven noun with a given descriptive adjectivewas tabulated. These percent response fre-quencies are an estimate of the degree to whichthe adjective would be elicited upon presenta-tion of the noun. In a more general sense, thepercentages are an indication of the strengthof association between an instance (noun) anda concept or category response (adjective).In Table 3 is presented the percent of Ss whoresponded to a given noun with white, small,or hard, the three adjectives used in each ofthe three stUdies. The entries in this tablewere used to calculate Pearson Product-Momentcorrelation coefficients between associationvalues found in the present study and thosefound by Underwood and Richardson andMayzner and Tresselt. For example, in Table4 are the words and values used to calculatethe degree of relationship between the results

of Mayzner and Tresselt and those of the pres-ent study for the concept "white."

In order to avoid spurious inflation of thecorrelation coefficient due to associationvalues of zero, only those words which hadvalues of 5 or more were used. Instances like"grass" were not included in Table 4 sinceneither in the present study nor in Mayznerand Tresselt's data was grass associated withthe category "white." Adding words like grasswith zero association values would haveincreased the value of the correlation since itwould have added a perfectly correlated pairof 'scores to the table, but would have obscuredthe actual relationship. Tables similar to Table4 were constructed for nouns associated withsmall and hard. All of these correlation coef-ficients are given in Table 5.

Association values had been obtained forthe nouns identified by the present author.These values are presented in Table 6. Sincethese words did not appear on either the Under-wood and Richardson or the Mayzner andTresselt list, no correlation could be run usingthem. From the size of the correlations whichwere run, however, it was assumed that thevalues obtained for the nouns in Table 6 werereasonable estimates of the values which wouldhave been obtained if the nouns had, in fact,been used in the other two studies. From thenouns in Table 6, acorn, bronze and mahoganywere selected as instances of the concept hard-brown, and mink was selected as an instance

TABLE 3. Percent of Ss that used a Given Adjective to Describe a Given Noun: Data from Ramsay(R), Underwood and Richardson (U & R), and Mayzner and Tresselt (M & T)

White Small Hard

R U&R M&T U&R M&T R U&R M&T

bread 90 35 55 5 0 7 10 0 7chamois 5 0 9 5 0 11 0 0 3

chestnut 0 0 0 70 9 56 80 18 57closet 0 0 5 40 24 33 0 0 5cork 0 0 1 50 0 40 5 0 10enamel 40 28 40 0 0 1 80 20 65garlic 40 0 10 35 0 20 5 0 4grass 0 0 1 0 0 7 0 0 0ivory 95 65 95 0 0 1 80 14 50linen 85 59 86 0 0 1 0 0 0moccasin 5 0 0 30 0 18 0 0 1

rabbit 45 25 41 35 6 54 0 0 1

salt 100 53 66 20 0 33 35 7 11sheep 85 23 75 5 0 6 0 0 0skull 95 25 22 10 0 5 95 36 71snail 0 0 2 85 42 72 10 0 10

10

TABLE 4. Correlation Between Association Val-ues found by Ramsay and by Mayznerand Tresselt for the Concept White

Noun

breadchamoisenamelgarlicivorylinenrabbitsaltsheepskull

Percent of Ss Who Respondedwith "White" to a Given Noun

Ramsay

Mayznerand

Tresselt

90 555 9

40 4040 1095 6585 5945 41

100 6685 7595 22

TABLE 5. Pearson Product-Moment CorrelationCoefficients for Associational Valuefound by Underwood and Richardson(U & R), Mayzner and Tresselt (M& T), and Ramsay

Adjective

white

small

hard

Source forValues Correlated Pearson

U & R - RamsayM & T - Ramsay

U & R - RamsayM & T - Ramsay

U & R - RamsayM & T - Ramsay

. 49

. 72

. 60

.90

. 81

.97

of the concept soft-brown. These four in-stances, plus 12 more which had been selectedfrom the Underwood and Richardson list, arepresented in Table 7.

TABLE 6. Association Values for Instances Hypothesized to Fit the Categories Hard-Brownand Soft-Brown

noun white brown small big hard soft

acorn 0 85 85 0 95 0

beer bottle 0 85 5 5 50 0

bronze 0 50 0 5 65 15gavel 0 40 25 0 85 0

loam 0 45 0 5 0 50mahogany 0 85 0 15 85 15mink 5 60 30 0 0 80pottery 5 20 15 5 85 0

TABLE 7. The 16 Verbal Instances

Type ofClassi-fication 1 2

Category

3 4

R-2

(white-hard)bonesaltenamelskull

(white- soft)rabbitlinenbreadsheep

(brown-hard)gavelchestnutacornbronze

R-1

(white)boneskulllinenbread

(brown)gavelbronzeminkchamois

(hard)saltenamelchestnutacorn

R-0

)

bonebreadbronzechamois

skullsheepacornmoccasin

saltlinengavelmink

(brown-soft)corkminkchamoismoccasin

(soft)rabbitsheepcorkmoccasin

enamelrabbitchestnutcork

11

IV

METHOD

In the summary section of Chapter II, it wassuggested that two predictions might be madewith regard to type of classification and typeof material: (1) that the rank order of difficultyof learning the three types of classificationswould be: R-0, the most difficult classifica-tion followed by R-1, with R-2 the easiestclassification, and (2) that learning with thefigural material would be more difficult thanlearning with the verbal material. The methodby which these predictions were put to empiri-cal test is the content of this chapter. Thechapter consists of four sections which are:Experimental Material, Subjects, ExperimentalProcedure, and Experimental Design.

EXPERIMENTAL MATERIAL

The 16 figural instances used in the stuwere H-patterns which varied as to number(one or two), size (large or small), color (ior green), and orientation (upright or tiltedEach instance was constructed, photograpion color slides, and mounted in 2 in. x 2 islide frames. These slides were projecteda 12 in. x 15 in. rear-projection screen.large upright patterns were 2 1/2 irk. tall IAthe small, upright patterns were 1 1/4 in.Variations in values of each dimension wereasily discriminable. The instances are resented in Table 8. The 16 verbal instance:

TABLE 8. The 16 Figural InstancesType of CategoryClassification 1 2 3

R-2

(large-green) (large-red) (small-green)

HH HH HH$

4(small-red)

R-1

(green)

HH

(red)

zH H

(large)

zHH

HH

H H

(small)

R-0HH Fi

12 I H H

)

HH

zz

( )

HH

were presented in Table 7 of the last chapter.Each was typed on mimeograph stencil, cut out,and mounted on 2 x 2 in. slides. The lettersin the projected words were white, 3/4 high,in lower case type.

The presentation of the slides and of feed-back information was fully automated. Theapparatus, which has been described in detailelsewhere (Davis, 1968), consisted of threeunits: a four channel response unit, a tapereader, and a Kodak Carousel slide projector.The response unit housed the electronic cir-cuitry, four response buttons , eight feedbacklights (a red and a green light over each but-ton), and the projection screen. A continuousloop of tape was punched with the correct re-sponses and fed through the tape reader. Thisunit, in conjunction with the response unit,controlled the feedback lights, while the re-sponse unit controlled the slide advance.

The function of the apparatus can be madeclear by the following sequence of events. (1)An instance was presented. (2) The S. who wasself-paced, pushed the response button corres-ponding to his choice of a category. (3) Theinstance was removed and if S had correctlycategorized the instance the green feedbacklight over the button he pushed came on; if hewas incorrect, the red feedback light over thebutton he pushed came on and the green feed-back light over the correct button came on.The feedback lights remained on for four sec-onds. (4) The next instance appeared.

SUBJECTS

The Ss who participated in the experimentwere 36 volunteers, all males, residing in theRegent Apartments of the University of Wiscon-sin. The mean age of the Ss was 19.4 years.One S failed to understand the instructions andwas replaced.

EXPERIMENTAL PROCEDURE

Subjects were scheduled.at their conven-ience and were assigned randomly to treatmentgroups with the restriction that six Ss were runin each treatment group. On arrival at thelaboratory, S sat in front of the response unitand E read the instructions appropriate for thetreatment group to which S was assigned. Fullsets of instructions can be found in the Appen-dix. The Ss were instructed as to the type ofcategorization which they would be required tolearn, because the omission of such instructionsmight have resulted in a continuing search bySs in the zero relevant dimension condition forsome systematic categorization of the instances.Since there was no such system, this search

would have been fruitless and might have causedpoorer performance not due to the condition perse, but because Ss were looking for a relation-ship which did not exist.

After the instructions were given, the taskbegan and E recorded each response. A dualcriterion,for the termination of the task was set;the task ended after S had correctly categorizedone block of the 16 instances or after 240 trials.Time to criterion was kept with a stopwatch.Upon finishing, Ss in the R-1 and R-2 treat-ments were asked if they could label the fourresponse buttons and all Ss were interviewedinformally about their reaction to the task.

The order of correct responses was randomwith the restriction that each of the four re-sponse buttons was correct twice in a block ofeight trials. Slides of the 16 instances wereordered in blocks of 16. For a given trial,one of four slides could have been presentedand paired with the correct response. Whichof the four possible slides was in fact pairedwith the correct response was randomly deter-mined with the restriction that the slide couldnot again appear until the block of 16 slideswas completed. The capacity of the slidemagazine was 80 slides, so five differentblocks of 16 slides were presented before thefirst one repeated. It was assumed that thelearning of response sequences due to the repe-tition of blocks could be ignored.

The criterion of 16 correct responses in ablock of 16 instances was chosen rather thanthe more commonly used one of 16 correct re-sponses in a row so that Ss would have to cate-gorize correctly each of the 16 slides. If thecriterion of 16 correct responses in a row hadbeen used it was conceivable that S could haveachieved this criterion without having respondedcorrectly to all 16 slides.

EXPERIMENTAL DESIGN

The two independent vatiables were type ofclassification (2, 1, or 0 relevant dimensions)and type of material (figural or verbal instances).In the R-2 figural 'condition the dimensions of,color and size were relevant to each categoryso that button 1 represented large-green pat-terns, button 2 large-red, button 3 small-green,and button 4 small-red. In the R-1 figural con-dition one dimension, color or size, was rele-vant for each category so that button 1 repre-sented green, button 2 red, button 3 large, andbutton 4 small. In the R-0 figural conditionno dimensions were relevant to a given cate-gory. Thus no value of any dimension was heldconstant for any of the categories. In the figuralcondition for R-2, instances appropriate forbutton 1 varied on number and orientation but

13

all were large and green. In R-1, the instancesfor button 1 varied on number, orientation, andsize but were all green. In R-0, the instancesfor button 1 varied on number, orientation,size, and color. In the verbal condition, therelevant associational dimensions for R-2were color and hardness; for R-1, color orhardness; and for R-0 neither color nor hard-

14

ness were consistently associated with anyof the categories.

A 2 x 3 factorial design was used with sixreplications in each of the six treatment groups.A two-way fixed effects analysis of variancemodel was assumed with the mean square errorterm as the denominator in the F ratio both formain effects and for the interaction.

L.

V

RESULTS

It was assumed that the Ss who achievedthe criterion of the correct classification ofall 16 instances in a block of 16 trials wouldhave performed without error 'had they continuedto the end of the 15 block trials. Consequently,from the trial at which S achieved this criterion,a number of correct responses was added to his'score such that this number plus the total num-ber of trials he had undergone to achieve cri-terion summed to 240 trials. Two hundred fortytrials was the alternative criterion, so theadjustment just cited allowed the comparisonof scores Of Ss which had reached either onecriterion or the other.

The results of the analysis of variance ontotal correct responses will be presented asthe initial section of this chapter. Subsequent.sections will deal with the analysis of varianceon mean time per instance and the results of aninform& interview with Ss after the experiment.

ANALYSIS oF VARIANCE ONTOTAL CORRECT RESPONSES

The summary table for'the analysiS of vari-ance is given in Table 9, and the means fortreatment groups are given in Table 10. TheF ratios for both of the main effects and for theinteraction were significant (p < .001). Therelationship among the means and the level ofattainment of treatment groups are shown inFigure 2. The nature of the interaction becomesobvious from an inspection of this figure.

A trend analysis (Myers, 1966) was includedin the analysis of variance in order to describein detail the curves for the figural and verbalireatments over the three levels of type of clas-sification. The statements derived from thisanalysis were: (1) The linear trend in the datawas significant (F1,30 = 219.35; p < .001) andaccounted for 9 2.1% of the variation due to typeof classification; (2) the quadratic trend in thedata was also significant (F1,30 = 18.76;p < .001) and accounted for 7.9% of the variation

TABLE 9. Summary Table for Analysis ofVariance on Total Number Cor-rect Responses

Source df MS

A. Type ofMaterial 1 38,155.11 153.75**

B. Type ofClassi-fication 2 29,546.25 119.06**Lin (B) 1 54,435.38 219.35**Quad (B) 1 4,657.12 18.76**

A x B 2 23,118,39 46.58**Lin (B) xA 1 15,657.05 63.09**Quad (B)x A 1 7,461.34 30.06**

Error 30 248.17Total 35

**p < .001

TABLE 10. Mean Number Correct forTreatment Groups

Type ofClas si-fication

R-0R-1R-2

Type of Material

Figural Verbal

M = 80.67M = 99.17M = 227.00

M = 176.50M = 205.00M = 220.67

due to type of classification. The significantlinear and quadratic components of the inter-action shoWed that both the slopes and shapesof the curves differed significantly.

Because of the significant interaction be-tween the two main effects, a meaningful des-cription of the effects depended upon a subse-quent, two-step analysis. First, separate

15

one-way analyses of variance on type of clas-sification for figural and verbal materials wereperformed. The second step was to use I testson differences between pairs of means in theevent that a one-way analysis resulted in asignificant L ratio. In each one-way analysisof variance the mean square error from the two-way analysis was retained for the denominatorof the E ratio. This error term was also usedas the estimate of error variance in thet tests.

The one-way analysis of varinace on datafrom the figural condition resulted in a signifi-cant F (F2,30 = 153.51; p < .001). Subse-quent two tailedt tests showed the differencein mean number correct between the R-2 andR-0 groups was statistically significant (t30= 16.09; p < .001). The difference betweenmeans of R-2 and R-1 groups was also signifi-cant (t30 = 14.05; p < .001). The differencebetween means of the R-1 and R-0 groups wasnot a significant one.

The one-way analysis of variance for Ss in theverbal condition also resulted in a statisticallysignificant E (F2,30 = 12.12; p < .001). The

tests showed a significant difference betweenthe R-2 and R-0 means 430 = 4.86; p < .001),and between the R-1 and R-0 means (t30

=

100

90

60

70

60

50

40

30

20

10

0

3.13; p < .05). The difference between meansof the R-2 and R-1 groups was not significant.

The learning curves of the six treatmentgroups shown in Figure 3 also reflect the sig-nificant interaction. While Ss in the figuralR-0 treatment were categorizing an average of6.5 instances correctly in the 15th block of 16instances , and while Ss in the R-1 conditionwere categorizing an average of 9 out of 16instances correctly in the 15th block, all Ss inthe R-2 group had categorized all 16 instancescorrectly by the end of the fifth block. Thelearning curves for Ss in the verbal treatmentgroups show a much more consistent increasefrom R-0 to R-1 to R-2 and look more liketypical learning curves.

MEAN TIME PER INSTANCE

Total time to criterion in seconds for eachS was divided by the total number of responsesmade to criterion. The resulting number indi-cated the mean time in seconds S had spent oneach instance. From .a two-way analysis ofvariance on the scores it was determined thatSs who had observed the figural instances tooka significantly longer mean time per instancethan Ss who had observed the verbal instances(F1,30. = 11.22; P < .01). Neither Type of

0a

1 2

NUMBER OF RELEVANT DIMENSIONS

Figure 2. Performance of Treatment Groups as a Function ofType of Material and Type of.Classification

16OPO soo-as4-4

100

90

BO

70

60

50

40

30

20

10

0

A°i

ii

tt

A0t :

et

"Ilk. ...so a III%%.

A.es.....110# .

., °.jr% a. S.air' '' .

. ... , .9 . ....... ,. .... ..e e.S.

he . .5,4,0 . , ...-.*

. .. . . ... '. Jr* S.

II° - 1I ' '

A. IIS

A A

#

a S

1 2 3 4 5 6 7 9 9 10 11 12 13 14 15

BLOCKS

Figure 3. Percentage of Correct Responses as a Function of TreatmentGroup and Number of Blocks of Trials Completed

Classification nor the interaction of Type ofMaterial and Type of Classification were sig-nificant effects. The mean time per instancefor Ss in the figural condition was 9.43 sec-onds; for Ss in the verbal condition, 6.57seconds. The summary of the analysis of vari-ance is presented n Table 11.

TABLE 11. Summary Table for the Analysisof Variance on Mean Time perInstance

Source ofVariance df MS

A. Type ofMaterial 1

B. Type ofClassi-fication 2

A x B 2Error 30Total 32

73.59

10.494.146.56

**p < .01

SUBJECTS' COMMENTS

During the experiment several Ss made spon-taneous comments which were recorded by E.

verbalRa 1

Ago op A R' IA 90figural

01 R I

R 0

After an S had completed the experimental taskhe was informally interviewed as to what hethought about the task, and any questions hehad about the nature of the experiment wereanswered. The Ss said that once they had begunthe task, they found it very interesting and,in some cases, frustrating. One S said after32 trials, "It made me so angry to see thatlittle red light come on." On the 34th trialthis same S said as he observed the instance,"I'm going to make this mistake again," andin making a res.lonse did indeed make the sameerror on that instance that he had made in theprevious block of trials.

An S in the R-2 verbal treatment group saidaloud in the course of the experiment after in-correctly responding to the instance "rabbit,""Oh, you're using white rabbits." A similarcomment was made about the same instanceby an S in the R-1 verbal group. Both Ss latersaid that they would have described rabbits anbrown rather than white. These comments indi-cate that with the verbal material some Ss re-lied on idiosyncratic associations rather thanthose which are experimenter-defined.

A clear use of idiosyncratic categorizationwas illustrated in Ss' responses to the ques-tion of what labels they had attached to thefour response buttons. This question wasasked of those Ss in the R-2 and R-1 condi-tions. All of the Ss in the R-2 figural group

17

and four out of the six in the R-2 verbal groupgave the experimenter-defined labels, thevalues of the two relevant dimensions, whenasked to label the response buttons. The othertwo Ss in the R-2 verbal group gave "animal"as one of their categories. In the R-1 situationhowever, only one S in the figural group wasable to give the experimenter defined labels.The other Ss in that group said that they hadnot used labels but had simply tried to memo-rize which slide went with which butthn, Themost interesting labeling occurred in the R-1verbal group. Their responses are listed inTable 12. Not one of these Ss completely rep-licated the experimenter-defined categoriesalthough S No. 28 gave three out of the fourlabels. All of the Ss identified the instancesassociated with button 1 as white, but noneidentified the instances of button 4 as soft.For the latter category, all but one of the Ssindicated that they had used the property ofliving thing or animal to categorize the instances

One easily definable strategy for gatheringinformation was used by two of the Ss (a No.26 in the R-1 verbal group and S No. 3 in theR-0 figural group). In the first block of 16trials S No. 26 pushed button 1 thirteen times.In the second block he pushed button 1 fifteentimes. In these W7st two blocks he averagedfour correct responses. On the third block he

made 13 out of 16 correct responses, andreached criterion of 16 out of 16 on the sixthblock. This S reported that .s.ince he would beguessing the first time through the instances,he had simply pushed button 1 and then hadpaid attention to the feedback lights whichinformed him of the correct response.

A variation of this strategy was used by SNo. 3. In this case, however, the strategywas not employed until the fifth block ofinstances. Up to that block there was nosystematic pattern of responses and his suc-cesses were at chance level. Then on thefifth block he pushed button 1 fifteen times.On the sixth block he pushed button 2 twelvetimes and also got three out of the four instancesassociated with button 1 correct. On the sev-enth block of trials he was able to categorizeseven out of the eight instances associatedwith buttons 1 and 2 correctly. On the eighthblock he shifted his attention to button 3. On

. blocks 5 through 14, or 160 trials, he onlypushed button 4 six times. On the 15th blockhe switched to button 4 and in this final blockof trials made 13 out of the 16 correct responsesas compared to the 5.4 out of 16 averaged bythe rest of the Ss in his treatment group. ThisS said that he had deliberately set about focus-ing his attention on each of the buttons in turn,

TABLE 12. Labels given to the Four Categories by Ss in the R-1 Verbal Treatment Group

ExperimenterDefined white brown hard softCategory

Instances bone gavel salt rabbitskull bronze enamel sheeplinen mink chestnut corkbread chamois acorn moccasin

Ss' Code Numbersand Labels

25 white brown miscellaneous living property26 white porosity brown27 white brown white-hard brown-animal28 white brown hard animal29 white catch-all minerals and nuts animals and leather goods30 white brown element,

hard,white,smooth

not brown,animal oranimalproduct

18

VI

DISCUSSION

Two hypotheses were tested in the study.The first was that the rank order of difficultyin learning to classify instances would be aninverse function of the number of relevant di-mensions determining the type of classification,where the number of relevant dimensions waszero, one, or two. Secondly, it was hypothe-sized that figural material would lead to moredifficulty in correct categorization of instancesthan would verbal material.

While the data generally support the firsthypothesis, the exceptions must be noted,viz., that there was no difference betweenR-0 and R-1 in the figural condition nor be-tween R-1 and R-2 in the verbal condition.A similar statement and qualification can bemade with regard to the second hypothesis con-cerning type of material. While verbal instanceswere clearly easier to categorize in the R-0 andR-1 conditions, in the R-2 condition, perform-ance was very nearly the same for both types ofmaterial. In this last case, the small differencewhich did occur was opposite to the predicteddirection.

From the reports of labeling in the R-1 andR-2 conditions, one possible interpretation ofthe results can be presented. In the R-1 con-dition labeling of the buttons might have facili-tated performance. In the R-2 condition, thenon-experimenter-defined labels used by twoof the Ss in the verbal condition might haveactually inhibited performance since the animalcategory used by these Ss was not mutuallyexclusive of instances in other concepts,whereas the labels for the buttons used by Ssin the figural R-2 condition were mutually ex-clusive. The suggestion is that labeling isbetter than no labeling, but mutually exclusivelabels are better than labels which allow in-clusion of instance from other categories.

The interpretation Just posed is one whichrelates to the construct of mediation. The labelsmight function as mediators. It was evident thatSs in the verbal R-1 condition had no difficulty

in labeling the buttons while Ss in the figuralR-1 condition did have difficulty. The use ofverbal labels as mediators may have facilitatedperformance. The mutually exclusive experi-menter-defined labels used in the R-2 figuraltreatment could thus be postulated to be themost efficient mediators.

Another interpretation of the results restson the assumption of interference occuring asa Joint function of type of classification andtype of material. According to this interpreta-tion, in the R-0 condition there was interfer-ence occurring with both types of material butgreater interference with the figural material.This interference was minimized in the R-1verbal group. At R-2, interference was mini-mized in the figural group. For this interpre-tation to be tenable, however, the source ofthe interference must be specified.

The R-0 condition will be considered first.in this condition where there were no valuesof dimensions which could be consistentlyassociated with each response button, figuralinstances were more difficuh to categorizecorrectly than verbal instances. Interferencemay have been reduced in the latter conditionif Ss responded to each instance as a wholerather than as a complex of values on dimen-sions. For example, learning to push button1 on the presentation of the instance "salt"may have been easier if salt was rememberedas a unitary item rather than as a set of itemsdescribed as white, hard, grainy, etc. In con-trast to this, the figural instances may havebeen remembered on the basis of their entireset of figural values so that an S needed toremember that button 1 was, for example, thecorrect response for one large green uprightpattern. The finding that longer time per in-stance was spent on figural material than onthe verbal material could be attributed to thisanalytical vs. holistic perception of the twotypes of materials. If this difference in theperception of the two types of material occurred,

19

then interference between figural instancesmay have occurred among any of the valuesmaking up the instances while interferencebetween the verbal instances may have occurredon the basis of the unitary item which the ver-bal instances represented, rather than on thebasis of the fractionation of the instances intoseveral associative values.

In the R-1 condition it could be postulatedthat the same kind of differential perception ofthe materials occurred resulting in continuedinterference for the figural group. Assumingthat the verbal instances were responded to assingle items, the presence of one consistentassociative value for each response buttonwould contribute to the similarity of theeset offour items associated with each button thusdecreasing interference.

In the R-2 condition, the attention given tothe specific values of the figural instancescould have been highly facilitative to perform-ance since this would be a way to identify justwhich values were relevant to a particular but-ton and which were not. The perception of theverbal instances as single items in the R-2condition may have prevented the identificationof relevant associative values.

An analysis of the experimental events wouldalso seem to contribute to the understanding ofthe results. An S had to remember the instancehe had just observed in order to associate itwith the correct response, since the slides andfeedback lights were presented sequentiallyrather than contiguously. Furthermore, he hadonly four seconds to concentrate on this asso-ciation until the next instance appeared. Sinceone instance only appeared once in a block of16 instances, an average of 15 instances camebetween the presentation of an instance and thenext presentation of the same instance.

This sequence of events may have contributedto the relatively poor performance in the figuralR-0 and R-1 treatment groups. The Ss had tostore in memory an instance composed of fourvalues, associate it with one of four responsebuttons, and remember this association while

20

responding to, and trying to remember the cor-rect response for an average of 15 more four-valued instances. Since five out of six Ss inthe figural R-1 group were not able to labelany of the buttons, it might be assumed thatthey were unable to identify the relevant valuesassociated with each button and were thereforefaced with the same memory load as Ss in thefigural R-0 condition. Assuming that Ss in theverbal treatments stored the nouns as unitsrather than as a set of discrete values, it wouldseem likely that they had much less of a memoryload imposed on them than Ss in the figuraltreatments.

It is likely that each of the considerationsmentioned abovelabeling, mediation, inter-ference, instances stored as units vs. sets ofvalues, time limits in the experimental sequenceof events all contributed to the results of theexperiment. This study was not designed toprovide evidence as to the efficacy of any ofthese alternatives, but simply to determine theeffects of the independent variables. In thislatter respect, the study would seem to havebeen successful. Some of the differences be-tween treatment group means were so large asto have made statistical tests a validation ofthe obvious. Now that type of classificationand type of material have been shown to bepowerful variables, further experimentationcan be designed to deal with the alternativeexplanations of their effects.

For both types of materials, performancewas shown to be an increasing function of num-ber of relevant dimensions, but the interactionof type of material with type of classificationindicated that the effects of increasing the num-ber of relevant dimensions was not consistentfor the two types of material. This inconsist-ency would suggest that caution be used ingeneralizing from results of experiments onconcept learning based on dimensionalizedfigural materials to situations where conceptlearning takes place in a primarily verbalmedium.

APPENDIX,

INSTRUCTIONS

FIGURAL, TWO RELEVANT DIMENSIONS

What you are about to begin is a learningtask. You will be shown a series of 16 patternslike these (show some patterns). After seeingeach pattern you will push one of these fourbuttons. What you are to learn is, which pat-tern goes with which button. For a.given pat-tern, if the button you push is the correct one,a green light will come on over that button.If your choice is incorrect, a red light willcome on over the button you pushed and a greenlight will come on over the button you shouldhave pushed.

The patterns and buttons have a systematicrelationship. This relationship can best bedescribed by using as an example, the sortingof playing cards. A deck of playing cardscould be sorted as to suit (clubs, diamonds,hearts, spades), as to color (red or black), asto whether the numbers are even or odd, as toface vs. non-face, etc. Also, the deck couldbe sorted on the basis of some combination oftwo of these characteristics. We could sorton the basis of color and face vs. non-face.If we did this we would wind up with four cate-gories of cards: red face cards, black facecards, red non-face cards, and black non-facecards. The 16 patterns you will see have anumber of characteristics upon which theycould be classified. These characteristics arecolor (red or green), number (one or two), size(large or small), and orientation (upright or onthe side). Each button represents one of thefour categories obtained from the combinationof two of these characteristics. For example,if the two characteristics were number andcolor, button one might represent two red,button two - one red, button three - two green,and button four - one green. Using the cate-gories just mentioned a slide showing a patternwith one large green upright figure would becategorized under button four which stood forone green.

To review, your task is to learn which buttonto push on seeing each of 16 patterns. Eachbutton represents a category and the four cate-gories are obtained from the combination oftwo characteristics. What you must learn isthe category membership of each pattern. Onseeing each pattern you will push one of thebuttons. If you are correct, a green light willcome on over that button. If you are wrong, ared light will come on over the button and agreen light will come on over the button youshould have pushed. To begin with you willnot know what the four categories are whichthe buttons represent. You will have to guess.The pattern will change only when you havepushed a button, so that you may work at yourown pace. I promise you that there is nothingtricky about this task. It is a straightforwardlearning task. Do you have any questions?

FIGURAL, ONE RELEVANT DIMENSION

What you are about to begin is a learningtask. You will be shown a series of 16 patternslike these (show some patterns). After seeingeach pattern you will push one of these fourbuttons. What you are to learn is, which pat-tern goes with which button. For a given pat-tern, if the button you push is the correct one,a green light will come on over that button.If your choice is incorrect, a red light willcome on over the button you pushed and agreen light will come on over the button youshould have pushed.

The patterns and buttons have a partiallysystematic relationship. This relationship canbest be described by using as an example thesorting of playing cards. A deck of playingcards could be sorted as to suit (clubs, dia-monds, hearts, spades) as to color (red orblack), as to odd vs. even, as to face vs.non-face, etc. Also the deck could be sortedon the basis of two of these characteristics.We could sort on the basis of color and face

21

vs. non-face. If we did this, we would windup with four categories: red cards, black cards,face cards, and non-face cards. As you cansee, this kind of classification system is notcompletely precise. The five of spades fitstwo categories, black cards and non-face cards.All the other cards also would fit two categories.Once you learned the categories, however, youwould have limited your choice to two ratherthan four categories. You would know for ex-ample that the Jack of diamonds would belongto either the category red cards or the categoryface cards.

The 16 patterns you will see also have anumber of characteristics upon which they couldbe classified. These characteristics are color(red or green), number (one or two), size (largeor small), and orientation (upright or on theside). Each button represents one of the fourcategories obtained from two of these charac-teristics. For example, if the two character-istics used to classify the cards were numberand color, button one might represent two pat-terns, button two -1 one pattern, button three -red patterns, and button four - green patterns.As in the playing card example, the basis forclassification is not precise. Using the cate-gories just mentioned, a slide showing twogreen patterns might be categorized under buttonone which stood for two patterns or under buttonfour which stood for green patterns. What youmust learn are the categories and also the spe-cific category for each pattern.

To review, your task is to learn which buttonto push on seeing each of 16 patterns. Eachbutton represents a category and the four cate-gories are obtained from two characteristics.What you must learn is the category membershipof each pattern. On seeing each pattern youwill push one of the buttons. If you are correcta green light will come on over that button. Ifyou are wrong, a red light will come on overthe button you pushed and a green light willcome on over the button you should have pushed.To begin with you will not know what the fourcategories are which the buttons represent orwhich pattern goes with each specific category.You will have to guess. The pattern will changeonly when you have pushed a button, so thatyou may work at your own pace. I promise youthat there is nothing tricky about this task. Itis a straight-forward learning task. Do youhave any questions?

FIGURAL, ZERO RELEVANT DIMENSIONS

What you are about to begin is a learningtask. You will be shown a series of 16 patternslike these (show some patterns). After seeingeach pattern you will push one of these four

22

buttons. What you are to learn is, which pat-tern goes with which button. For a given pat-tern, if the button you push is the correct one,a green light will come on over that button.If your choice is incorrect, a red light willcome on over the button you pushed and a greenlight will come on over the button you shouldhave pushed.

The patterns you will see have a number ofcharacteristics upon which they vary. Thesecharacteristics are color (red or green), number(one or two), size (large or small), and orienta-tion (upright or on the side). There is however,no systematic relationship between the char-acteristics of the patterns, and the buttons.

To review, your task is to learn which but-ton to push on seeing each of 16 patterns. Onseeing each pattern you will push one of thebuttons. If you are correct, a green light willcome on over that button. If you are wrong ,a red light will come on over the button and agreen light will come on over the button youshould have pushed. To begin with you willnot know which pattern goes with which button.You will have to guess. The pattern will changeonly when you have pushed a button, so youmay work at your own pace. I promise you thatthere is nothing tricky about this task. It is astraight-forward learning task. Do you haveany questions?

VERBAL, TWO RELEVANT DIMENSIONS

What you are about to begin is a learningtask. You will be shown a series of 16 wordslike these (show some words). After seeingeach word you will push one of the four buttons.What you are to learn is, which word goes withwhich button.. For a given word, if the buttonyou push is the correct one, a green light willcome on over that button. If your choice isincorrect, a red light will come on over thebutton you pushed and a green light will comeon over the button you should have pushed.

The words and buttons have a systematicrelationship. This relationship can best bedescribed by using as an example, the sortingof playing cards. A deck of playing cards couldbe sorted as to suit (clubs, diamonds, hearts,spades), as to color (red or black), as towhether the numbers are even or odd, as toface vs. non-face, etc. Also the deck couldbe sorted on the basis of some combinationsof two of these characteristics. We could sorton the basis of color and face vs. non-face.If we did this we would wind up With four cate-gories of cards: red face cards, black facecards, red non-face cards and black non-facecards. The 16 words you will see have a num-ber of associational characteristics upon which

they could be classified. Some of these asso-ciational characteristics are texture, smoothvs. rough, color, white vs. brown, shape,round vs. flat, and hardness, hard vs. soft.Each button represents one of the four categoriesobtained from the combination of two of thesekinds of associational characteristics. Forexample, if the characteristics were size andcolor, button one might represent large brownthings, button 2 - small brown things, button3 - large white things,,and button 4 - smallwhite things. Using the categories just men-tioned, a slide showing the word "pill" wouldbe categorized under button four which stoodfor small white things.

To review, your task is to learn which buttonsto push on seeing each of 16 words. Each buttonrepresents a category based on two characteris-tics, so what you must learn is the categorymembership'of each word. On seeing each wordyou will push one of the buttons. If you arecorrect, a green light will come on over thatbutton. If you are wrong, a red light will comeon over the button and a green light will comeon over the button you should have pushed. Tobegin with you will not know what the fourcategories are which the buttons represent:You will have to guess. The word will changeonly when you have pushed a button, so thatyou may work at your own pace. I promise youthat there is nothing tricky about this task. Itis a straight-forward learning task. Do youhave any questions?

VERBAL, ONE RELEVANT DIMENSION

What you are about to begin is a learningtask. You will be shown a series of 16 wordslike these (show some words). After seeingeach word you will push one of these four but-tons. What you are to learn is, which wordgoes with which button. For a given word, ifthe button you push is the correct one, a greenlight will come on over that button. If yourchoice is incorrect, a red light will come onover the button you pushed and a green lightwill come on over the button you should havepushed.

The words and buttons have a partially sys-tematic relationship. This relationship canbest be described by using as an example thesorting of playing cards. A deck of playingcards can be sorted as to suit (clubs, diamonds,hearts, spades) as to color (red or black) as toodd vs. even, as to face vs. non-face, etc.Also the deck could be sorted on the basis oftwo of these characteristics. We could sorton the basis of color and face vs. non-face.If we did this, we would wind up with fourcategories: red cards, black cards , face cards,

and non-face cards. As you can see, this kindof classification system is not completely pre-cise. The five of spades fits two categories,black cards and non-face cards. All the othercards also would fit two categories. Once youlearned the categories, however, you wouldhave limited your choice to twb rather than tofour categories. You would know for examplethat the jack of diamonds would belong to eitherthe category red cards or to the category facecards.

The 16 words you will see have a number ofassociational characteristics upon which theycould be classified. Some of these associa-tional characteristics are texture (smooth thingsvs. rough things), color (white things vs. brownthings), shape (rough things vs. flat things),and hardness (hard things vs. soft things).Each button represents one of the four cate-gories.obtained from two of these kinds ofcharacteristics. For example, if the charac-teristics were size and color, button one mightrepresent large things, button two - smallthings, button three - brown things and buttonfour -.white things. As in the playing cardexample, the basis of classification is notprecise. Using the categories just mentioneda slide showing the word "pill" might be cate-gorized under button two which stood for smallthings or under button four which stood forwhite things. What you must learn are thecategories and also the specific category foreach word.

To review, your task is to learn which buttonto push on seeing each of 16 words. Each but-ton represents a category and the four categoriesare obtained from two as sociational character-istics. What you must learn is the categorymembership of each word. On seeing eachword you will push one of the buttons. If youare correct, a green light will come on overthat button. If you are wrong, a red light willcome on over the button you pushed and a greenlight will come on over the button you shouldhave pushed. To begin with you will not knowwhat the four categories are which the buttonsrepresent, or which word goes with each spe-cific category. You will have to guess. Theword will change only when you have pusheda button, so that you may work at your ownpace. I promise you that there is nothing trickyabout this task. It is a straight-forward learningtask. Do you have any questions?

VERBAL, ZERO RELEVANT DIMENSIONS

What you are about to begin is a learningtask. You will be shown a series of 16 wordslike these (show some words). After seeingeach word, you will push one of these four

23

buttons. What you are to learn is, which wordgoes with which button. For a given word, ifthe button you push is the correct one, a greenlight will come on over that button. If yourchoice is incorrect, a red light will come onover the button you pushed and a green lightwill come on over the button you should havepushed.

The words you will see have a number ofassociational characteristics upon which theyvary. Some of these characteristics are texture(smooth vs. rough), color (white vs. brown),shape (round vs. flats, and hardness (hard vs.soft). There is, however, no systematic rela-tionship between these associational charac-teristics of the words and the buttons.

24

To review, your task is to learn which but-ton to push on seeing each of 16 words. Onseeing each word you will push one of thebuttons. If you are correct, a green lightwill come on over that button. If you arewrong, a red light will come on over the buttonand a green light will come on over the buttonyou should have pushed. To begin with youwill not know which word goes with whichbutton. You will have to guess. The wordwill change only when you have pushed a but-ton, so you may work at your own pace. Ipromise you that there is nothing tricky aboutthis task. It is a straight-forward learningtask. Do you have any questions?

GPO 11011.411441

REFERENCES

Archer, E. J., Bourne, L. E., & Brown, F. G.Concept identification as a function ofirrelevant information and instruction.Journal of Experimental Psychology, 1955,49, 153-164.,

Arnstine, D. Philosophy of education: Learn-ing and schooling. New York: Harper &Row, 1967.

Ausubel, D. P. In defense of verbal learning.Educational Theory, 1961, 11, 15-25.

Baum, M. H. Simple concept learning as afunction of intralist generalization. Journalof Experimental Psychology, 1954, 47, 8994.

Bourne, L. E. Human conceptual behavior.Boston: Allyn and Bacon, 1966.

Bruner, J. S. The process of education. Cam-bridge, Mass.: Harvard University Press,1960.

Bruner, J. S., Goodnow, J. J., & Austin, G.A. A study of thinking. New York: JohnWiley & Sons, 1956.

Bulgarella, R., & Archer, E. J. Concept identi-fication of auditory stimuli as a function ofamount of relevant and irrelevant information.Journal of Experimental Psychology, 1962,63, 254-257.

Buss, A. H. A study of concept formation asa function of reinforcement and stimulusgeneralization. Journal of ExperimentalPsychology, 1950, 40, 494-503.

Buss, A. H. Rigidity as a function of reversaland nonreversal shifts in the learning ofsuccessive discriminations. Journal of Ex-perimental Psychology, 1953, 45, 75-81.

Coleman, E. B. Verbal concept learning as afunction of instruction and dominance level.Journal of Experimental Psychology, 1964,68, 213-214.

Conant, M. B., & Trabasso, T. Conjunctiveand disjunctive concept formation underequal-information conditions. Journal ofExperimental Psychology, 1964, 67, 250255.

Davies, M. H., Cooper, C., Davis, J. K.,& Stewart, D. Tasks employed in conceptidentification and problem solving studies.Unpublished working paper, Research andDevelopment Center for Learning and Re-education, University of Wisconsin,Madison, Wisconsin, 1965.

Davis, G. A., & Bourne, L. E. Effects ofresponse type and problem complexity uponclassification learning. Journal of GeneralPsychology, 1965, 73, 151-159.

Davis, J. K. Concept identification as a func-tion of cognitive style, complexity, andtraining procedures. Technical Report fromthe Wisconsin R & D Center for CognitiveLearning, University of Wisconsin, 1968,No. 32.

Fredrick, W. C. The effects of instructions,concept complexity, method of presentation,and order of concepts upon a concept attain-ment task. Technical Report from the Wis-consin R & D Center for Cognitive Learning,University of Wisconsin, 1965, No. 3.

Freedman, J. L., & Mednick, S. A. Ease ofattainment of concepts as a function of re-sponse dominance variance. Journal ofExperimental Psychology, 1958, 55, 463466.

Gibson, E. J. A systematic application of theconcepts of generalization and differentia-tion to verbal learning. Psychological Re-view, 1940, 47, 196-229.

Harrow, M., & Friedman, G. B. Comparingreversal and nonreversal shifts in conceptformation with partial reinforcement con-trolled. Journal of Experimental Psychology,1958, 55, 592-598.

Haygood, R. C. , & Bourne, L. E. Attributeand rule learning aspects of conceptualb,thavior. Psychological Review, 1965,72, 175-195.

Hull, C. L. Quantitative aspects of the evo-lution of concepts. Psychological Mono-graphs, 1920, 28 (1, Whole No. 123).

25

Hunt, E. B. Conce tual learnin : An informa-tion processing problem. New York: JohnWiley & Sons, 1962.

Hunt, E. B,., & Hovland, C. I. Order of con-sideration of different types of concepts.Journal of Experimental Psychology, 1960,59, 220-225.

Kelleher, R. T. Discrimination learning as afunction of reversal and nonreversal shifts.Journal of Experimental Psychology, 1956,51, 379-384.

Kendler, H. H., & D'Amato, M. F. A compari-son of reversal shifts and non-reversal shiftsin human concept formation behavior. Journalof Experimental Psychology, 1955, 49, 165174.

Kendler, T S. Concept formation. AnnualReview of Psychology, 1961, 12, 447-472.

Kendler, 'T. S., Kendler, H. H., & Wells, D,Reversal and non-reversal shifts in nurseryschool children. Journal of Comparativeand Physiological Psychology, 1960, 53,83-88.

Klausmeier, H. J., Davis, J. K., Ramsay, J.G:, Fredrick, W. C., & Davies, M. H.Concept learning and problem solving: Abibliography, 1950-1964. Technical Reportfrom the Wisconsin R & D Center for Cog-nitive Learning, University of Wisconsin,1965, No. 1.

Mayzner, M. S., & Tresselt, M. E. A com-parison of judgmental and associationaltechniques in developing verbal conceptformation materials. Journal of Psychology,1961, 51, 331-342.

Mednick, S. A., & Halpern, S. Ease of con-cept attainment as a function of associativerank. Journal of Experimental Psychology,1962, 64, 628-630.

Miller, C. A. The magic number seven, plusor minus two: Some limits on our capacityfor processing information. Psychological-Review, 1956, 63, 81-97.

Myers, J. L. Fundamentals of experimentaldesign. Boston: Allyn and Bacon, 1966.

Neisser, U. & Weene, P. Hierarchies inconcept attainment. Tournal of ExperimentalPsychology, 1962, 64, 644-645.

Neuman, S. E. Effects of contiguity and simi-larity on the learning of concepts. Journalof Experimental PsychologY, 1956, 62,349-353.

Oseas, L., & Underwood, B. J. Studies ofdistributed practice: V. Learning and reten-tion of concepts. Journal of ExperimentalPsychology, 1952, 43, 143-148.

26

Ramsay, J. G. The attainment of conceptsfrom Vgural and verbal instances, by indi-viduals and pairs. Technical Report fromthe Wisconsin R & D Center for CognitiveLearning, University of Wisconsin, 1965,No. 4.

Runquist, W. N., & Hutt, V. H. Verbal con-cept learning in high school students withpictorial and verbal representation of stim-uli. Journal of Educational Psychology,1961, 52, 108-111.

Shannon, C., & Weaver, W. The mathemati-cal theory of communication. Urbana:University of Illinois Press, 1949.

Shepard, R. N. Stimulus and response gen-eralization: A stochastic model relatinggeneralization to distance in psychologicalspace. Psychometrika, 1957, 22, 325-345.

Shepard, R. N. Stimulus and response gen-eralization: Tests of a model relating gen-eralization to distance in psychologicalspace. Journal of Experimental Psychology,1958, 55, 509-523.

Shepard, R. N., & Chang, J. Stimulus gen-eralization in the learning of classifications.Journal of Experimental Psychology, 1963,65, 94-102.

Shepard, R. N., Hovland, C. I., & Jenkins,H. N. Learning and memorization of clas-sifications Psychological Monographs,1961, 75, (13, Whole No, 517).

Shipstone, E. I. Some variables affecting pat-tern conception. Psychological Monographs,1960, 74 (Whole No. 504).

Smoke, K. L. Negative instances in conceptlearning. Journal of Experimental Psychol-ogy, 1933, 16, 583-588.

Underwood, B. J. Studies of distributed prac-tice: XV. Verbal concept learning as a func-tion of intralist interference. Journal ofExperimental Psychology, 1957, 54, 33-40.

Underwood, B. J. An evaluation of the Gibsontheory of verbal learning. In C. N. Cofer(Ed.), Verbal learning 'and verbal behavior.New York: McGraw-Hill, 1961, 197-217.

Underwood, B. J., & Richardson, J. Someverbal materials for the study of conceptformation. Psychological Bulletin, 1956,53, 84-95. (a)

Underwood, B. J., & Richardson, J. Verbalconcept learning as a function of instructionsand dominance level. Journal of ExperimentalPsychology, 1956, 61, 229-238. (b)

Webster's new world dictionary of the Americanlanguaqe. (College ed.) Cleveland: WorldPublishing Co., 1966.

GPO 110/11-244-.2*

(TOP)

001

100101

102

103

200

300310

320330

340350400

00501

600601602

603604605606

607

800801802

803804805806807808809810811

812813814815816817818819820821

822

OE UM licv. 9:46) DL.PAN IMEN 1 01 NI. Al III, C:DUC Al ION, ANt JI.L F ANL: ,?*P--77e-I.ITI( ACCEINIDN NC),

ACCr5SION NUPAIIL.R

g ex 3 33RI:St.J".1) DATE

l().-78-0 1

ERIC&e.7/(C..""J ....'e' ciiir":65REPORT RESUME

is DOCUMENT COPYRIGHTED) YES

ERIC REPRODUC rtoN RELEASE' vES M NO 0P.A. 1.b.,

1, L.F.

Concept learning as a function of the type of material and type ofclassification

.

.PERSONAL AUTHORIS)

Ramsay, James C.

----------is.,7-6

..

I N 5 TITD1ION IsouRCE,universi tyWisconsin Research and

of Wisconsin,DevelQpmantReport No53

.

Madison, WlsconsinGeni-or.. for____Cogni-tive-Learning-

REPORT/SERIES NO. TechnicalsouRcE coot:.

.----- --i

OTHER SOURCE.

OlHER'REPORI NO.OTHER SOURCE SOURCE COE

..---- .-...--------...-0-THER REPOR T NO. --.PUWL.OATE June - 68 I CON 7 R AC T/GOiANT NUMBER Center No. C-03Blantmaci- OF 5-10615A,PAGINATION, ETC.

24 pp.RETRIEVAL 'I ERMS

concept formation learning processesconcept developmentconceptual distinctions

:verbal learningvisual learningvisual stimuli,cognitive processes

IDENTIFIERS

AE3S TR AC I

This reports the effects of the number of relevant stimulus dimensionsand figural versus verbal stimuli on the concept learning ability of collegestudents. Results force a consideration of mediational variables in explain-ing this form of cognitive learning.

A set of verbal materials analogous to a set of dimensionalized figuralmateiials was constructed. The figural stimuli were 16 H-patterns, the com-binations of values of the four binary dimensions of color (red or green), size(large or small), number (one or tuo), and orientation (upright or tilted). Theverbal stimuli were 16 nouns, four sets of four, which had been shown to beassociated with four adjectival categories: hard-white, soft-white, hard-brown,and soft-brown. Two hypotheses were tested: (1) that the difficulty of threeclassifications would be an inverse function of the number of relevant dimensionscomposing the classifications, and (2) that figural instances would be moredifficult to categorize correctly than verbal instances.

The hypotheses were supported, but needed to be qualified because ofsignificant interaction. Alternative interpretotions of results arediscussed.

ATC{-71.,