JOURNAL OF THE EXPERIMENTAL ANALYSIS OF BEHAVIOR

CHILDREN'S IDENTITY MATCHING AND ODDITY:ASSESSING CONTROL BY SPECIFIC AND GENERAL

SAMPLE-COMPARISON RELATIONS

ROBERT STROMER AND JOAN BUTCHER STROMER

NORTHERN MICHIGAN UNIVERSITY ANDEUNICE KENNEDY SHRIVER CENTER

After children in Experiments 1 and 2 learned identity matching or oddity, control by sample-comparison relations was assessed. Tests for generalized control displayed novel samples and twocomparison stimuli, one identical to the sample. Specific relations were tested with identical or non-identical sample-comparison stimuli from one set of stimuli and substitute comparisons from eitherthe other training set or from a novel set. When tests displayed identical stimuli, patterns of comparisonselection suggested control by generalized identity and oddity. However, selection patterns varied whenstimuli were nonidentical and familiar or novel substitute comparisons were used. Therefore, controlby specific relations is not a precondition for generalized identity and oddity. One set of trainingstimuli was used in Experiment 3, and generalized performances occurred again. Moreover, controlby specific relations was shown by the oddity subjects and 2 of 6 identity subjects. Generalized andspecific control may therefore exist simultaneously. In Experiment 4, selections were irregular on testsdisplaying substitute comparisons and samples and familiar comparison stimuli; this finding supportedthe relational account of specific sample-comparison control found in Experiment 3.

Key words: stimulus control, identity matching to sample, oddity from sample, window press, children

In identity matching to sample and oddityfrom sample a subject might be given threestimuli on a given trial: a sample stimulusdisplayed on the center stimulus/responsewindow and comparison stimuli on the twoside windows. When identity-matching con-tingencies are in effect, selecting the identicalcomparison stimulus is reinforced; in oddity,selecting the differing comparison stimulus isreinforced. Identity matching and oddity areconditional discriminations because the S+ andS- functions of comparison stimuli changefrom trial to trial conditionally upon differentstimuli (Cumming & Berryman, 1965).Human identity matching and oddity per-

formance may be controlled by the general orspecific relations among samples and compar-ison stimuli. Evidence for generalized controlis inferred from performance on test trials thatdisplay arrays of novel stimuli. These arrays

This research was conducted at Northern MichiganUniversity. Final preparation of the manuscript occurredat the Shriver Center with financial support from NICHDGrant HD05515. We greatly appreciate the constructivecomments provided by Bill Mcllvane on the manuscript.Requests for reprints should be addressed to RobertStromer, Behavioral Neurology Department, E. K. ShriverCenter, 200 Trapelo Road, Waltham, Massachusetts02254.

permit selections based on identity or oddity(i.e., a sample and two comparison stimuli,one identical to the sample). The two types ofspecific sample-comparison control are thepositive relations between samples and S+comparison stimuli (sample/S+ relations) andnegative relations between samples and S-comparison stimuli (sample/S- relations)(Stromer & Osborne, 1982). Control by theserelations is inferred under test trials that dis-play novel stimuli as substitute comparisons.

Children are capable of generalized identityand oddity performance (Levin & Maurer,1969; Saunders & Sherman, 1986; Scott, 1964;Sherman, Saunders, & Brigham, 1970). Con-sistent with their training history, they selecteither the identical or odd stimulus from anovel array of three stimuli. Further, recentresearch suggests that children's identitymatching may also be controlled by specificsample/S+ and sample/S- relations (Dixon& Dixon, 1978; Dixon, Dixon, & Spradlin,1983). In Dixon and Dixon's Experiment 2,for example, preschoolers were first taught AAmatching: given Sample Al, to select Com-parison Stimulus Al and not A2; and givenA2, to select A2 and not Al. The children werealso trained with a second set of stimuli (BBmatching): if Bi, select Bi and not B2; and if

47

1989, 51, 47-64 NUMBER 1 ( JANUARY)

ROBERT STROMER and JOAN BUTCHER STROMER

B2, select B2 and not Bi. Next, tests for sam-ple/S- control displayed the following stim-uli: an A sample (e.g., Al), its correspondingS- A comparison stimulus (e.g., A2), and asubstitute B comparison stimulus (e.g., B1).Control by specific sample/S- relations wasshown when all children consistently rejectedthe S- A comparison stimuli and selected thesubstitute B comparison stimuli. Control bysample/S+ relations (e.g., if Al, select Al; ifA2, select A2) was not assessed in this study,but it was shown in a related study with men-tally retarded subjects (Dixon et al., 1983).The possibility that the subjects were also ca-pable of control by generalized identity rela-tions was suggested but not assessed.The development of control by both positive

and negative sample-comparison relations maybe an important characteristic of human con-ditional discrimination that possibly is relatedto the ability to perform generalized identitymatching and oddity (Stromer & Osborne,1982). Perhaps, control by the general rela-tions in fact depends on the ability to learnboth sample/S+ and sample/S- relations.This suggestion gains support from researchwith pigeons. Under typical identity-matchingprocedures, pigeons apparently learn whichcomparison stimulus is "correct" on a giventrial but fail to learn which is "incorrect" (Ber-ryman, Cumming, Cohen, & Johnson, 1965;Carter & Werner, 1978; Cumming & Ber-ryman, 1961; Urcuioli, 1977; Urcuioli &Nevin, 1975). With these same procedures,pigeons also fail to display generalized iden-tity. However, when pigeons are explicitlytrained to differentiate between specific sam-ple/S+ ("correct") and sample/S- ("incor-rect") relations, they show some evidence ofgeneralized control (Santi, 1982; Urcuioli,1977; Urcuioli & Nevin, 1975).

It is also possible, however, that control bygeneral and specific relations may be indepen-dent and perhaps mutually exclusive in somecases. Variables such as the number of rela-tions trained and tested are likely to influencewhether a subject who is capable of learningboth positive and negative relations in fact doesso (cf. McIlvane, Munson, & Stoddard, 1988).Generalized stimulus control implies that con-trol is not restricted to sample-comparisonrelations involving specific stimuli. Indeed,generalized control requires appropriate re-sponding to entirely new stimulus arrays.

Moreover, control by specific stimuli may beincompatible with, or discouraged, when pit-ted against generalized control. For example,Stoddard (1968) taught children to select an"odd" S+ line tilted to the right of verticalfrom an eight-stimulus array; the remainingS- stimuli were lines tilted to the left of ver-tical. During a test for generalized control, thestimulus display was reversed: A single linetilted to the left of vertical was shown withseven lines tilted to the right of vertical. Se-lecting the odd line tilt during this test wasevidence of control by the general relation ofoddity. A subsequent test assesed control bythe specific stimuli; an S+ line tilt (the formerodd stimulus) was displayed with a single S-line tilt. On this test, children whose initialtest performance suggested control by odditytypically did not select the S+ line reliably. Incontrast, reliable selections were made by chil-dren whose initial test performance did notsuggest oddity.The present study sought to clarify variables

that may contribute to the emergence of gen-eralized and specific sample-comparison con-trol. In Experiments 1 through 3, childrenwere trained with either an identity-matchingor oddity procedure. Thereafter, they receivedtests for specific sample-comparison control.Variables examined included the relative nov-elty of stimuli substituted during testing andthe number of stimuli used in baseline train-ing. Control by the general relations of identityand oddity was also assessed with selected chil-dren. Experiment 4 served as a control, ex-amining selections of comparison stimuli usedduring training in the presence of substitutesamples.

EXPERIMENT 1When one comparison stimulus involved in

a previously established conditional relation isreplaced by a novel comparison stimulus, se-lection or rejection of the substitute stimulusmay be due to its novelty alone (Dixon &Dixon, 1978; Stromer & Osborne, 1982).Dixon and Dixon in their Experiment 2 soughtto minimize novelty effects by giving the chil-dren a brief history with the substitute stimuli(e.g., Set B) prior to testing. However, thehistory was considerably shorter than that ofthe set (e.g., Set A) used to assess sample/S-relations in identity. One could argue, there-

48

IDENTITY MATCHING AND ODDITY

fore, that relative novelty could have led to thesample/S- control found in that experiment.To control for novelty in the present ex-

periment, children were trained with stimulusSets A and B from the outset (AA/BB train-ing). The reinforcement histories of the A andB stimuli were thus equated prior to tests forsample/S+ and sample/S- control. Further,control by positive and negative relations wasevaluated with children who were given odditytraining. Test arrays were the same for bothgroups of children; the critical variable wasthe identity versus oddity training histories. Ofprimary interest was whether identity trainingwould result in sample/S- control andwhether oddity training would result in sam-ple/S+ control. As will be seen in more detailbelow, both groups were given tests that dis-played samples and nonidentical comparisonstimuli from one set (e.g., Sample Al andComparison A2) and a substitute comparisonstimulus from the other (e.g., Bl). For identitysubjects, samples and nonidentical comparisonstimuli from the same stimulus set could beinvolved in sample/S- relations; for odditysubjects these stimuli could be involved in sam-ple/S+ relations. In addition, generalizedidentity or oddity was assessed with some sub-jects. A finding of generalized control duringthese tests would replicate previous studies(Levin & Maurer, 1969; Saunders & Sher-man, 1986; Scott, 1964; Sherman et al., 1970)and address the question of the coexistence ofcontrol by specific sample-comparison rela-tions and control by generalized relations.

METHODSubjectsThe subjects were 12 experimentally naive

children (6 females and 6 males). Their meanage was 5 years 9 months (range, 4 years 8months to 7 years 11 months).

ApparatusThe experimental console contained three

Grason Stadler stimulus projectors (ModelE5704B-3), a Gerbrands penny dispenser, anddoor chimes. The projectors were mounted inGrason Stadler display/response windows(Model E8670A) and produced white letter-like forms (16 mm) on dark backgrounds. Thethree windows (3.2 cm diameter) were ar-ranged horizontally on the face of the console

(24.0 cm by 54.5 cm), with the two side win-dows spaced 12.7 cm to either side of the centerwindow. Relay circuitry in an adjacent roomcontrolled stimulus presentations. Windowpresses were recorded on digital counters anda Gerbrands six-pen event recorder (ModelP2C-6).

Identity-Matching ProcedureThe children were seen three to five times

per week at about the same time each dayduring each meeting. Two to four 48-trial ses-sions, each lasting 5 to 10 min, alternated with5- to 10-min play periods. The latter wereheld in a nearby room containing age-appro-priate toys. Table 1 shows the type of trainingand order of test conditions each subject re-ceived. The number assigned to a given testcondition (Test 1, Test 2, etc.) signifies theparticular stimulus array (see Figure 1) usedduring the test.

Trials began when a sample stimulus wasdisplayed on the center window. One press onthe sample window illuminated the two sidewindows with comparison stimuli; the sampleremained until a comparison stimulus waspressed. Samples were randomized across trialswith the restriction that the same sample oc-curred no more than twice in succession. Theleft-right position of the correct and incorrectcomparison stimuli varied unsystematicallyacross trials, and each appeared in each po-sition an equal number of times. In any givensequence of 48 trials, the same position wascorrect no more than three consecutive times.When feedback contingencies were in effect,pressing the S+ comparison stimuli resultedin chimes, the delivery of a penny, a 2-s in-tertrial interval with all windows dark, andthe next trial. Pressing the S- comparisonstimuli produced only the 2-s intertrial intervaland the next trial.

Train AA/BB identity. Figure 1 shows theidentity-matching task taught to 6 children (seeTable 1): if Sample Al, select Al and not A2;if A2, select A2 and not Al; if Bi, select Biand not B2; and if B2, select B2 and not B1.Before the first session the following instruc-tions were given:When you see this window (Experimenterpointing to sample stimulus) go on, push it withyour finger; see, you made these windows(pointing to comparison stimuli) come on. Now,if you see this thing (pointing to the sample

49

ROBERT STROMER and JOAN BUTCHER STROMER

Table 1

Conditions for identity and oddity (*) subjects in Experiment 1; all were trained with AA/BBstimulus arrays. Numbers in parentheses reflect sessions required to reach training criteria.Test sessions for sample/S+ (ST/S+), .sample/S- (ST/S-), and generalized (Gen.) controlare listed in the order given. Figures 1 and 2 show stimulus arrays given during the conditions.Test numbers in table and figures correspond.

NR AS NO JB BM RJ

Trn (1-5) Trn (1-8) Trn (1-7) Trn (1-7) Trn (1-6) Trn (1-8)

Test ST/S+ Test ST/S+ Test ST/S- Test ST/S- Test Gen. Test ST/S-Control: Control: Control: Control: Identity: Control:

Test 1 Test 2 Test 1 Test 3 Test 1 Test 3Test 2 Test 1 Test 2 Test 4 Test 4

Test 3 Test 1 Test ST/S- Test 1Test ST/S- Test ST/S- Test 4 Test 2 Control: Test 2Control: Control:

Test ST/S+ Test ST/S+ Test 1 Test ST/S+Test 1 Test 3 Control: Control: Test 2 Control:Test 2 Test 4 Test 3Test 3 Test 1 Test 1 Test 2 Test 4 Test 2Test 4 Test 2 Test 2 Test 1 Test 1

Test ST/S+Control: Test Gen.

Identity:Test 1Test 2 Test 3

Test 4Test Gen. Test 1Identity: Test 2

Test 2Test 3Test 4

stimulus), push this thing (pointing to the cor-rect comparison stimulus); if you do, you'll heara bell and you'll get a penny in the cup. If youpush this thing (pointing to the incorrect com-parison stimulus), the bell won't go on and youwon't get a penny. When we're all done youget to keep all the pennies that you make dropinto the cup, so try to get as many pennies asyou can. Here, let me show you how you canmake pennies drop into the cup.

The child was then guided through four to sixtrials during which the consequences for cor-rect ("Great, you made the bell go on and yougot a penny.") and incorrect ("See, when youpush that thing there's no bell or penny.")comparison selections were demonstrated. Af-ter this, the experimenter left the room andbegan the session. At first, each correct selec-tion produced the chimes and a penny. Train-ing continued until one session of 96 to 100%accuracy occurred. After a session, the child

was allowed to keep the pennies earned or toexchange them for coins of comparable worth.Next, the child was told that only some of thetrials would produce the chimes and pennies,and "to do the best you can." At this point,50% of the trials produced chimes and pennies.After one session of 96 to 100% accuracy, theschedule was reduced so that only 25% of thetrials produced these consequences. This phasecontinued until a child responded with 96 to100% accuracy in three consecutive sessions.Under intermittent reinforcement, pennies thechild accumulated during a session were ex-change for an amount either twice (50% sched-ule) or four times (25% schedule) the face value.

Subsequent test sessions consisted of a mix-ture of 16 AA trials, 16 BB trials, and 16 testtrials. Reinforcement was discontinued for alltrials during testing, but subjects were given$0.50 at the end of each session. Subjects inboth groups each received tests for sample/S+control and for sample/S- control. Four sub-

IDENTITY MATCHING AND ODDITY

Table 1 (Continued)

JC* NC* NM* CO* MI* DF*

Trn (1-5) Trn (1-5) Trn (1-7) Trn (1-5) Trn (1-7) Trn (1-6)

Test ST/S- Test ST/S- Test ST/S+ Test ST/S+ Test ST/S+ Test Gen.Control: Control: Control: Control: Control: Oddity:

Test 1 Test 2 Test 1 Test 3 Test 1 Test 3Test 2 Test 1 Test 2 Test 4 Test 2

Test 3 Test 1 Test 3 Test ST/S+Test ST/S+ Test ST/S+ Test 4 Test 2 Test 4 Control:Control: Control:

Test ST/S- Test ST/S- Test ST/S- Test 3Test 1 Test 3 Control: Control: Control: Test 4Test 2 Test 4 Test 1Test 3 Test 1 Test 1 Test 2 Test 1 Test 2Test 4 Test 2 Test 2 Test 1 Test 2

Test ST/S-Test Gen. Control:Oddity:

Test 2Test 1 Test 1Test 2Test 3 Test Gen.Test 4 Oddity:

Test 4Test 1Test 2

jects (2 identity and 2 oddity) were added tothe experiment after initial data were gath-ered; these subjects also received tests for gen-eralized identity or oddity (see Table 1).

Testsfor sample/S+ control. Figure 1 showsthe two tests used to assess whether subjectswould continue to select comparison stimuliwhen identical samples were presented and theremaining comparison stimuli were substi-tuted from the other set of training stimuli. InTest 1, for sample/S+ control the trials in-volved samples from Set A, identical compar-ison stimuli, and substitute comparison stimulifrom Set B. In Test 2, the tests involved sam-ples and identical comparison stimuli from SetB and substitute comparison stimuli from SetA. To minimize the time devoted to the testseries, and because the primary focus of theidentity tests was on assessing control by sam-ple/S- relations, only two of the four possiblestimulus arrays for assessing sample/S+ con-trol were given to each child.

Testsfor sample/S- control. Figure 1 showsthe four tests designed to assess control by neg-ative relations (sample/S - control). Each testdisplayed a sample from one set (A or B), anS- comparison stimulus from that set, and asubstitute comparison stimulus from the otherset.

Testsfor generalized identity. Figure 1 showsthe tests used to assess generalized identitymatching with Subjects BM and RJ. The fourstimuli from Set Z were combined to producefour different test arrays, each with a sample,an identical comparison stimulus, and a non-identical comparison stimulus. These stimuliwere not used during any other tests in thisexperiment.

Oddity ProcedureFigure 2 illustrates major trial types used

during oddity training; representative tests forcontrol by sample/S+, sample/S-, and gen-eralized oddity arrays are also shown. Table

51

ROBERT STROMER and JOAN BUTCHER STROMER

Train AA/BB Identity

Al Al A2 Al A2 A2

9 +eBi B1 B2 B1 B2 B2

Tests for Sample/S- Control

1

2

3

4

B1 Al A2 Al A2 B2

___ @00B2 Al A2 Al A2 B1

Al B1 B2 B1 B2 A21

A2 B1 B2 B1 B2 Al

0@0C@0l

Tests for Sample/S+ Control

1

2

Al Al B1 B2 A2 A2

B1 B1 Al A2 B2 82

Tests for Generalized Identity

1

2

3

4

Z1 Z1 Z2 Zi Z2 Z2

Z3 Z3 Z4 Z3 Z4 Z4

Zl Z1 Z Z3 Z3

Z2 Z2 Z4 Z2 Z4 Z4

___ 9@00Fig. 1. Major trial types for identity matching in Experiment 1. Each trio of stimuli depicts a sample (center) and

two comparison stimuli (sides). During training, selecting the identical comparison was reinforced (+) and selectingthe nonidentical one was not (-). Reinforcement was not delivered on test trials. Sample/S+ tests displayed samplesand S+ comparisons from one set of stimuli (A or B) and substitute comparisons from the other (B or A). Sample/S- tests displayed samples and S- comparisons from one set of stimuli and substitute comparisons from the other.Tests for generalized identity displayed three novel stimuli (Z).

1 lists the order of conditions given to eachsubject. The basic trial procedures, criterionlevels, and methods of testing were the sameas for identity matching.

Train AA/BB oddity. Oddity contingenciesspecified reinforcer delivery only if a child se-lected comparison stimuli that were not iden-tical to the samples: if A1, select A2 and notAl; if A2, select Al and not A2; if Bi, selectB2 and not B1; and if B2, select Bi and notB2.

Testsfor sample/S+ control. Figure 2 showsthat the stimulus arrays for these tests werethe same as those used to assess sample/S-control with identity subjects: Samples andnonidentical comparison stimuli from one set

of stimuli were displayed with substitute com-parison stimuli from the other. In Test 1, forexample, if the positive relations between Asamples and nonidentical A comparison stim-uli control the behavior, subjects should con-tinue to select the A comparison stimuli whengiven the alternate B comparison stimuli. Thestimulus arrays for Tests 2 through 4 were thesame as those shown in Figure 1 (Arrays 2through 4, sample/S- control) for the identitysubjects.

Testsfor sample/S- control. Figure 2 showsthat these test arrays were the same as thoseused for the identity subjects' sample/S+ tests(Arrays 1 and 2, Figure 1). But if sample/S-relations in oddity control the behavior, sub-

IDENTITY MATCHING AND ODDITY

jects should consistently select the nonidenticalcomparison stimulus that had been substitutedfrom the second stimulus set (e.g., ifA samplesand identical A comparison stimuli, select theB comparison stimuli).

Tests for generalized oddity. Generalizedoddity was assessed with Subjects MI and DFwith the same stimulus arrays used to assess

generalized identity. Figure 2 shows the Test1 array; stimulus arrays for Tests 2 through4 were the same as those shown in Figure 1for the identity subjects.

RESULTSThe identity subjects met the AA/BB train-

ing criterion in five to eight sessions. Odditysubjects took from five to seven sessions to meetcriterion. Throughout testing, accuracy on the32 AA/BB trials never fell below 94%.

Figure 3 shows percentages of selectingcomparison stimuli on the 16 critical trials ineach consecutive test session given to the 6children trained on identity matching. Duringsample/S+ tests (open bars), subjects selectedthe identical comparison stimuli 88 to 100%of the time, performances consistent with theirreinforcement histories during training. Selec-tions of the substitute A or B comparison stim-uli during sample/S- tests (diagonal bars)ranged from 6 to 94% with the majority ofperformances at or near 50%. During tests forgeneralized identity (solid bars), BM and RJselected the identical comparison stimuli 81 to100% of the time.

Figure 4 displays test results for the odditysubjects. During sample/S- tests (open bars),oddity subjects always selected the nonidenti-cal substitute comparison stimuli, not the iden-tical comparisons. During sample/S+ tests,percentages of selecting the S+ comparisonstimuli (diagonal bars) ranged from 19 to 75%.During tests for generalized oddity (solid bars),MI and DF always selected the differing com-parisons.

DISCUSSIONBoth identity and oddity subjects showed

consistent selection patterns only when testsdisplayed comparisons that were identical tothe samples. Sample/S- control in identityand sample/S+ control in oddity could not beinferred. Thus, these findings do not supportthose previously reported (Dixon & Dixon,1978; Dixon et al., 1983). The present data

Train AA/BB OddityAl Al A2 Al A2 A2

Bi Bi B2 B2 B2 Bi

Test 1 for Sample/S+ ControlB1 Al A2 A cA2 B2

Test 1 for Sample/S- ControlAl Al Bl B2 A2 A2

Test 1 for Generalized OddityZ1 Z1 Z2 Z1 Z2 Z2

Fig. 2. Major trial types for oddity and representativetests in Experiment 1. Each trio of stimuli depicts a sample(center) and two comparison stimuli (sides). During train-ing, selecting the nonidentical comparison was reinforced(+) and selecting the identical one was not (-). Rein-forcement was not delivered on test trials. Sample/S+ testsdisplayed samples and S+ comparisons from one set ofstimuli (A or B) and substitute comparisons from the other(B or A). Sample/S- tests displayed sample and S- com-parisons from one set of stimuli and substitute comparisonsfrom the other. Tests for generalized oddity displayed threenovel stimuli (Z).

could be interpreted as evidence of sample/S+control for identity subjects and sample/S-control for oddity subjects (Berryman et al.,1965; Cumming & Berryman, 1961; Dixon etal., 1983), but control need not have been lim-ited to specific relations involving identicalsame-set stimuli. Data from 4 subjects showedcontrol by the general relations of identity (BMand RJ) and oddity (MI and DF). Control bythe general relations could account for all sub-jects' performances during tests for sample/S+ (identity) and sample/S- (oddity) rela-tions. During these tests, one of the two com-

53

ROBERT STROMER and JOAN BUTCHER STROMER

N.R. R.S. N.O.

////F-F-F-F'/-fJ.B. B.M. R.J.

IDENTITY SUBJECTS IPSWLE/S4 CONTROLSAWLE/$- CONTROLBERRLItZED IDENTITY

Fig. 3. Test results for identity subjects in Experiment 1. Bars show percentages of selecting identical comparisonsduring tests for sample/S+ control and generalized identity or substitute comparisons during tests for sample/S-control. The 16 test trials (see Figure 1) were interspersed among 16 AA and 16 BB training trials. Bars are arrangedin the order tests were given (see Table 1).

parison stimuli was always identical to thesample. Either general or specific sources ofcontrol could produce the selection patternsshown during these tests; appropriate perfor-mance would not necessitate control by specificrelations among stimuli. Had subjects failedto show generalized control, the conclusion thatthere was control among identical same-setstimuli might be justified.The irregular selection patterns during tests

for sample/S- (identity) and sample/S+(oddity) control are consistent with control bygeneral relations. During these tests, neithercomparison stimulus was identical to the sam-ple; hence, the basis for control by the generalrelations was removed from the stimulus ar-

rays. These findings are consistent with thosesuggesting the incompatibility of control bygeneralized and specific sample-comparisonrelations in identity-matching and oddity per-formance (Stoddard, 1968).

EXPERIMENT 2One difference between Experiment 1 and

previous experiments that demonstrated spe-cific sample-comparison control (Dixon &Dixon, 1978; Dixon et al., 1983) was the rel-ative novelty of the substitute stimuli used dur-ing testing. From the outset of training, sub-jects in Experiment 1 were exposed to both Aand B stimuli, and baselines of AA/BB per-

formance were maintained throughout testing.Thus, during the critical tests for sample/S-control in identity and sample/S+ control inoddity, the substitute comparison stimuli were"novel" only in the context of the sample andcomparison from the alternative set. The pos-

sibility exists, therefore, that the novelty ofsubstitute comparison stimuli is an importantvariable in demonstrating control by specificsample-comparison relations. Had the chil-dren lacked prior exposure to the substitutecomparison stimuli, patterns of specific sam-

ple-comparison control might have emerged(cf. Dixon & Dixon, 1978). In Experiment 2,sample/S- control in identity and sample/S+control in oddity were assessed following AA/BB training using substitute X and Y com-

parison stimuli, stimuli that appeared for thefirst time during testing. These unfamiliarstimuli would, presumably, clearly reveal anyinfluence of novelty.

METHODSubjects and ApparatusThe subjects were 12 children (4 females

and 8 males). Their mean age was 5 years 7months (range, 4 years 7 months to 7 years 2months). Ten children were experimentallynaive; 2 (CO and AS) had participated in Ex-periment 1. The apparatus was the same asin Experiment 1.

z

C-U

LUCf-)

:zLLJC-)OcLUCL

54

IDENTITY MATCHING AND ODDITY

CDzC-)LUI-LUC1)

CzLUC-)

LUJ0m

N.C. N.M. C.O.

ODDITY SUBJECTS I I SRWPLE/S- CONTROL

SAMPLE/Si CONTROLeGEERALIZED ODDITY

Fig. 4. Test results for oddity subjects in Experiment 1. Bars show percentages of selecting nonidentical comparisonsduring tests for sample/S- control and generalized oddity or S+ comparisons during tests for sample/S+ control.The 16 test trials (see Figure 2) were interspersed among 16 AA and 16 BB training trials. Bars are arranged in theorder tests were given (see Table 1).

ProcedureThe procedures for training and testing were

essentially the same as in Experiment 1. Table2 outlines the order of test conditions, usingthe stimuli shown in Figure 5. Six subjectswere given identity training and 6 receivedoddity training. The training stimuli were thesame as those used in Experiment 1 (Figures1 and 2). Testing began after a subject metthe AA/BB training criterion. CO and ASwere given oddity and identity training, re-

spectively, in Experiment 1. In Experiment 2,however, CO was given identity training andAS was given oddity training. Table 2 alsoshows that 4 subjects were given two additionaltests for sample/S- (WB and SN) and sam-

ple/S+ (PS and JS) control. These tests weregiven to determine whether control by specificsample-comparison relations would emergewith additional testing.

Figure 5 shows the four types of test arraysused. The tests were like those in Experiment1, except for the substitute X and Y compar-ison stimuli. For the identity subjects, stimulusarrays in Figure 5 were tests for sample/S-control and may be compared with those inFigure 1. For oddity subjects, these tests as-

sessed control by sample/S+ relations and maybe compared with those in Figure 2. Two sub-jects (SN and JS) were given somewhat dif-ferent trial mixtures during testing. They re-

ceived 16 test trials interspersed among 32

training trials from the same stimulus set (e.g.,32 AA trials plus 16 Test-1 trials). These testconditions were therefore comparable to Ex-periment 3. As in Experiment 1, 4 childrenwere also given tests for generalized identity(CM and WJ) and oddity (RM and SG) (Fig-ures 1 and 2).

RESULTSThe 12 children took from five to seven ses-

sions to meet the identity or oddity criteria.During testing, they continued to respond cor-

rectly (97 to 100%) on training trials.Figure 6 shows that during repeated tests

for sample/S- control (diagonal bars), iden-tity subjects selected the substitute X and Ycomparison stimuli 12 to 75% of the time. Se-lections of the identical comparison stimuliduring tests for generalized identity (solid bars)ranged from 75 to 100%. Figure 7 shows thatoddity subjects' selections of the A or B com-parison stimuli ranged from 19 to 75% on testsfor sample/S+ relations. Selections of non-identical comparison stimuli during tests forgeneralized oddity ranged from 81 to 100%.

DISCUSSIONFollowing AA/BB training, control by spe-

cific sample-comparison relations was assessedusing substitute X and Y comparison stimuli,stimuli that were totally unfamiliar to the chil-dren. Identity and oddity subjects gave no evi-

M.I. D.F.

155

ROBERT STROMER and JOAN BUTCHER STROMER

Table 2

Conditions for identity and oddity (*) subjects in Experiment 2; all were trained with AA/BBstimulus arrays. Numbers in parentheses reflect sessions required to reach training criteria.Test sessions for sample/S+ (ST/S+), sample/S- (ST/S-), and generalized (Gen.) controlare listed in the order given. Figures 1, 2, and 5 show stimulus arrays given during the conditions.Test numbers in table and figures correspond.

CO JW WB SN CM WJ

Trn (1-5) Trn (1-8) Trn (1-5) Trn (1-7) Trn (1-7) Trn (1-7)

Test ST/S- Test ST/S- Test ST/S- Test ST/S- Test Gen. Test ST/S-Control: Control: Control: Control: Identity: Control:

Test 1 Test 3 Test 1 Test 3 Test 1 Test 3Test 2 Test 4 Test 2 Test 4 Test 4Test 3 Test 1 Test 3 Test 1 Test ST/S- Test 1Test 4 Test 2 Test 4 Test 2 Control: Test 2

Test 1 Test 3Test 3 Test 1 Test 1 Test Gen.

Test 2 Identity:Test 3Test 4 Test 3

Test 4Test Gen. Test 1Identity: Test 2

Test 2Test 3Test 4

dence of sample/S- and sample/S+ control,respectively. These results show that specificcontrol does not result from the use of the novelsubstitute stimuli per se. Further, results thusfar show that control by specific positive andnegative relations is not prerequisite for gen-eralized identity or oddity (cf. Berryman et al.,1965; Carter & Werner, 1978; Cumming &Berryman, 1961; Urcuioli, 1977; Urcuioli &Nevin, 1975) and may, in fact, be incompatible.This apparent incompatibility might be re-lated to particulars of the training procedures.Simultaneous use of two training sets (AA/BB), as opposed to one, represents a majordeparture from previous procedures that dem-onstrated specific sample-comparison control(Dixon & Dixon, 1978; Dixon et al., 1983).The influence of this variable was assessed inthe following experiment.

EXPERIMENT 3Control by general rather than specific stim-

ulus relations might be encouraged by simul-taneous training with two stimulus sets insteadof one. Training multiple exemplars of a stim-

ulus relation has been shown to promote gen-eralized stimulus control (Stokes & Baer,1977). Control established by general relationsmay, in turn, override control by specific re-lations among stimuli (Stoddard, 1968).

In Experiment 3, we asked whether the dif-ference in results of Experiments 1 and 2 andthose of Dixon and Dixon (1978) might bedue to the more complex baselines of the pres-ent study. The tactic was to teach identity oroddity with only one set of stimuli (A or B)and then to test for sample/S- relations (iden-tity subjects) or sample/S+ relations (odditysubjects) using substitute X and Y comparisonstimuli. For the identity subjects, these con-ditions closely resembled those that led to con-trol by nonidentical sample-comparison stim-uli over unfamiliar or relatively unfamiliarsubstitute comparisons in the experiments ofDixon and Dixon. Selected subjects were alsotested for generalized identity or oddity.

METHODSubjects and ApparatusTen experimentally naive children and 2

from Experiment 1 (NC and NO) served (5

56

IDENTITY MATCHING AND ODDITY

Table 2 (Continued)

AS* RG* PS* JS* RM* SG*

Trn (1-5) Trn (1-7) Trn (1-6) Trn (1-5) Trn (1-6) Trn (1-7)

Test ST/S+ Test ST/S+ Test ST/S+ Test ST/S+ Test ST/S+ Test Gen.Control: Control: Control: Control: Control: Oddity:

Test 1 Test 3 Test 1 Test 3 Test 1 Test 3Test 2 Test 4 Test 2 Test 4 Test 2Test 3 Test 1 Test 3 Test 1 Test 3 Test ST/S+Test 4 Test 2 Test 4 Test 2 Test 4 Control:

Test 1 Test 3Test 3 Test 1 Test Gen. Test 3

Oddity: Test 4Test 1

Test 1 Test 2Test 2Test 3 Test Gen.Test 4 Oddity:

Test 4Test 1Test 2

females and 7 males). Their mean age was 5years 8 months (range, 4 years 8 months to 7years 1 month). The apparatus was the sameas in the previous experiments.

ProcedureThe procedures were the same as in the

previous experiments except for the use of asingle stimulus set during training and the pro-vision of 24 rather than 48 training trials persession; this change was made in an effort toequate the number of training trials with thosegiven in Experiments 1 and 2. Test sessions,in which no reinforcers were delivered, con-sisted of 16 test trials interspersed among 32of a subject's respective AA or BB trainingtrials (these test conditions were comparableto those experienced by SN and JS in Exper-iment 2). Table 3 shows the training sets usedand the order of tests across subjects. Thetraining procedures used with NC and NOwere identity and oddity, respectively; thistraining differed from that given in Experi-ment 1 (NC, oddity; NO, identity). The testslisted in Table 3 correspond to those shown inFigures 1, 2, and 5.

RESULTSThe subjects took from five to eight sessions

to meet training criterion. During subsequenttesting their accuracy on training trials rangedfrom 97 to 1 00%.

Figures 8 and 9 show test performances foridentity and oddity subjects, respectively. Fig-ure 8 shows that CJ and TG were the onlyidentity subjects who tended to select the sub-stitute X and Y comparison stimuli duringtests for sample/S- control (diagonal bars).Subjects who were given tests for generalizedidentity (FB and TG) typically selected theidentical Z comparison stimuli (solid bars).Figure 9 shows that all 6 oddity subjects tendedto select the S+ A or B comparison stimuli onsample/S+ tests, except on the initial ones for4 subjects (CC, BK, GT, and MC). Duringtests for generalized oddity, GT and MC typ-ically selected nonidentical comparison stim-uli.

DISCUSSIONResults for 2 of 6 subjects who learned iden-

tity and for all 6 who learned oddity with only

57

ROBERT STROMER and JOAN BUTCHER STROMER

1

2

3

4

Xi Al A2 Al A2 X2

X2 Al A2 Al A2 Xi

Y1 Bi B2 Bi B2 Y2

@@® @S®Y2 B1 B2 B1 B2 Y1

I@ I

Fig. 5. Major trial types used in Experiments 2 through4 during unreinforced tests of either sample/S - control(for identity subjects) or sample/S+ control (for odditysubjects). Each trio of stimuli depicts a sample (center)and two comparison stimuli (sides). The substitute X andY comparisons were used only during these tests.

one stimulus set provided evidence of controlby specific sample-comparison relations. Thus,the findings differed from those of Experi-ments 1 and 2. In addition to control by thespecific relations, however, results from se-lected subjects also provided clear evidence ofgeneralized control. Most convincing were theperformances of TG (identity), GT, and MC

Z!

C-)

LL

C-)

LLCL

C.O. J.W. W.B. S.N.

(oddity). Therefore, generalized control doesnot necessarily preclude control by specificstimulus relations.The reduction of the number of baseline

relations was apparently responsible for theacquisition of specific controlling relations. Theuse of one rather than two stimulus sets meantthat subjects needed to differentiate only twospecific relations during testing (e.g., sample-comparison Al -A2 and A2-Al stimuli) insteadof the four required in Experiments 1 and 2.Therefore, it may have been easier to remem-ber two as opposed to four nonidentical sam-ple-comparison relations. Another possibilityalluded to earlier is that identity or odditydimensions may become more salient whentraining is conducted with two sets of stimulias opposed to one (Stokes & Baer, 1977). Thesubjects may have been capable of demonstrat-ing control by four nonidentical relations, butan immediate, firmly established history ofgeneralized control may have discouraged suchlearning.The differences across children in the emer-

gence of specific controlling relations in thisexperiment are worth noting. Apparently, se-lecting the nonidentical (S+) stimuli in oddity,as opposed to rejecting nonidentical (S -) stim-uli in identity, is more likely to produce specificcontrol. These outcomes are pertinent to theissue of whether identity and oddity perfor-mances are governed by the same or differentprocesses (e.g., Zentall, Edwards, Moore, &

C.M. W.J.

IDENTITY SUBJECTS SSA MPLE/S- CONTROLGhENERALIZED IDENTITY

Fig. 6. Test results for identity subjects in Experiment 2. Bars show percentages of selecting substitute X or Ycomparisons during tests for sample/S- control or identical Z comparisons during tests for generalized identity. The16 test trials (see Figure 5) were interspersed among 16 AA and 16 BB training trials or among 32 AA (during Tests1 and 2 in Figure 5) and 32 BB (during Tests 3 and 4 in Figure 5) training trials (Subject SN). Bars are arrangedin the order tests were given (see Table 2).

58

IDENTITY MATCHING AND ODDITY

R.G. P.S. J.S. R.M. S.G.

ODDITY SUBJECTS R/PLPLE/SE CONTROL6ENERALIZED ODDITY

Fig. 7. Test results for oddity subjects in Experiment 2. Bars show percentages of selecting S+ comparisons duringtests for sample/S+ control or nonidentical Z comparisons during tests for generalized oddity. The 16 test trials (seeFigure 5) were interspersed among 16 AA and 16 BB training trials or among 32 AA (during Tests 1 and 2 in Figure5) and 32 BB (during Tests 3 and 4 in Figure 5) training trials (Subject JS). Bars are arranged in the order testswere given (see Table 2).

Hogan, 1981). The present data support thelatter contention.

It seems unlikely that the sample-compar-ison control demonstrated by the oddity sub-jects can be attributed solely to consistentavoidance of novel substitute stimuli. Recallthat most subjects in Experiments 1 and 2 wereindifferent to the substitute stimuli; they nei-ther selected nor rejected them consistently.Moreover, the oddity subjects in Experiment1 consistently selected the substitute stimuliunder tests for sample/S- control. Finally,previous studies have, to some extent, con-

trolled for novelty by conducting preferencetests among substitute comparison stimuli (e.g.,Al as sample, and Xl and X2 as comparison

stimuli) before and/or after critical demon-strations of sample-comparison control (Dixonet al., 1983; Stromer & Osborne, 1982).With the present procedures, children seem

to be capable of control by both general andspecific relations among samples and compar-isons. To strengthen the conclusion that thecontrol was indeed relational, a final experi-ment was conducted that involved substitutesamples as well as substitute comparison stim-uli.

EXPERIMENT 4The performances of 2 identity subjects and

all 6 oddity subjects in Experiment 3 suggestedcontrol by the specific relations among non-

identical samples and comparison stimuli. If

the selection (oddity subjects) or rejection(identity subjects) of nonidentical comparisonstimuli indicates relational control betweensamples and comparison stimuli, then the pres-ence of substitute sample stimuli should dis-rupt test performance. In Experiment 4, we

asked whether consistent patterns of selection(oddity subjects) or rejection (identity subjects)would occur given the familiar comparisonstimuli without the original samples (tests forcomparison control). Test trials displayed Aor B training stimuli, unfamiliar substitutecolor samples, and substitute form comparisonstimuli. In addition, 2 new subjects were trainedwith the oddity procedure and were then giventhe tests for comparison and sample/S+ con-

trol.

METHODSubjects and Apparatus

Immediately following Experiment 3, 1identity subject (TG) and 2 oddity subjects(GT and MC) participated in this experiment.Two naive male children (GA and Li), aged6 years 1 month and 5 years, also served. Theapparatus was the same as in all previous ex-

periments.

ProcedureFigure 10 depicts the test arrays used to

assess control by A or B comparison stimuli.Test trials displayed familiar comparison stim-uli, red (R) or green (G) samples, and novel

z

II

uJ

ul

L)CD14lLiC-

A. S.

59

ROBERT STROMER and JOAN BUTCHER STROMER

z

CI~-jC-)

LL

C-')F-

LU

LL

N.C. T.B. T.K. C.J. F.B. T.G.

IDENTITY SUBJECTS SSAHPLE/S- CONTROL

SENERWKIZE IDENTITY

Fig. 8. Test results for identity subjects in Experiment 3. Bars show percentages of selecting substitute X or Ycomparisons during tests for sample/S- control or identical Z comparisons during tests for generalized identity. (Theabsence of a bar reflects 0% selection of X or Y stimuli.) The 16 test trials (see Figure 5) were interspersed among32 AA training trials (NC, TB, and FB) or 32 BB training trials (TK, CJ, and TG). Bars are arranged in the ordertests were given (see Table 3).

substitute comparison stimuli. After odditytraining and tests for comparison control, the2 new subjects were also given tests for sam-

ple/S+ control. Table 4 shows the conditionsgiven to each subject.

RESULTSThe new oddity subjects (GA and LJ)

reached training criterion in six to seven ses-

sions; their accuracy on the training trials re-mained high during testing (97 to 100%).

Table 3

Conditions for identity and oddity (*) subjects in Experiment 3. The A or B next to theparentheses indicates the stimulus set used during training; numbers within reflect sessionsrequired to reach criteria. Test sessions for sample/S+ (ST/S+), sample/S- (ST/S-), andgeneralized (Gen.) control are listed in the order given. Figures 1, 2, and 5 show stimulus arraysgiven during the conditions. Test numbers in table and figures correspond.

NC TB TK CJ FB TG

Trn A(1-5) Trn A(1-5) Trn B(1-5) Trn B(1-5) Trn A(1-6) Trn B(1-5)

Test ST/S- Test ST/S- Test ST/S- Test ST/S- Test Gen. Test ST/S-Control: Control: Control: Control: Identity: Control:

Test 1 Test 2 Test 3 Test 4 Test 1 Test 3Test 2 Test 1 Test 4 Test 3 Test 4Test 1 Test 2 Test 3 Test 4 Test ST/S- Test 3Test 2 Test 1 Test 4 Test 3 Control: Test 4

Test 3 Test 4Test 4 Test 3 Test 1 Test Gen.

Test 2 Identity:Test 1Test 2 Test 3

Test 4Test Gen. Test 1Identity: Test 2

Test 2Test 3Test 4

60

IDENTITY MATCHING AND ODDITY

S.S. B.K. G.T. M.C.

ODDITY SUBJECTS E HSqPiWLE/S* cONTROLBEEMELIZEO ODDITY

Fig. 9. Test results for oddity subjects in Experiment 3. Bars show percentages of selecting S+ comparisons duringtests for sample/S+ control or nonidentical Z comparisons during tests for generalized oddity. The 16 test trials (seeFigure 5) were interspersed among 32 AA training trials (NO, CC, and GT) or 32 BB training trials (SS, BK, andMC). Bars are arranged in the order tests were given (see Table 3).

Figure 11 shows that during the tests forcontrol by comparison stimuli (horizontal bars),MC, an oddity subject from Experiment 3,continued to select the familiar comparisonstimuli 100% of the time. Selection patternsfor the other subjects were inconsistent duringthese tests. On tests for control by sample/S+

relations (diagonal bars), GA's selections ofS+ comparison stimuli ranged from 66 to 94%and LJ's ranged from 56 to 75%.

DISCUSSIONThe results strengthen the findings of spe-

cific relational control found in Experiment 3.

Table 3 (Continued)

NO* CC* SS* BK* GT* MC*

Trn A(1-5) Trn A(1-8) Trn B(1-6) Trn B(1-6) Trn A(1-6) Trn B(1-8)

Test ST/S+ Test ST/S+ Test ST/S+ Test ST/S+ Test ST/S+ Test Gen.Control: Control: Control: Control: Control: Oddity:

Test 1 Test 2 Test 3 Test 4 Test 2 Test 3Test 2 Test 1 Test 4 Test 3 Test 1Test 1 Test 2 Test 3 Test 4 Test 2 Test ST/S+Test 2 Test 1 Test 4 Test 3 Test 1 Control:

Test 3 Test 4Test 4 Test 3 Test Gen. Test 4

Oddity: Test 3Test 3

Test 1 Test 4Test 2Test 3 Test Gen.Test 4 Oddity:

Test 4Test 1Test 2

z

H-LU-JLUiCD)

H-1LUC-)LL0-

N.O. C.C.

61

ROBERT STROMER and JOAN BUTCHER STROMER

1

2

3

4

Bi A2 Al B2

B2 A2 Al Bi

Al B2 B1 A2

A2 B2 B1 AlIc ®DIcb®o) g

Fig. 10. Major trial types used in Experiment 4 dur-ing unreinforced tests of comparison control. Each trio ofstimuli depicts a sample (center) and two comparison stim-uli (sides). Tests displayed red (R) or green (G) samples,familiar comparisons (Set A or B), and substitute com-parisons (Set B or A) used only during these tests.

The selection patterns indicating sample/S-control in identity and sample/S+ control inoddity were apparently not based on controlby comparison stimuli alone. Oddity SubjectsGA and LJ were also tested with sample/S+

stimulus arrays. Performances generally didnot show strong control by sample/S+ rela-tions. Error patterns established during priortesting, however, may well have interfered withtheir subsequent performance (cf. Stoddard &Sidman, 1971).

GENERAL DISCUSSIONThe present study represents a first step

toward identifying the conditions under whichhuman identity and oddity performances arecontrolled by general relations, specific rela-tions, or both. The performances found in Ex-periments 1 through 3 replicate previous dem-onstrations of generalized control in identityand oddity (Levin & Maurer, 1969; Saunders& Sherman, 1986; Scott, 1964; Sherman et al.,1970). Further, Experiments 1 and 2 supportthe earlier observation of control by general,but not by specific, stimulus relations with hu-mans (Stoddard, 1968). However, control bygeneral and specific relations is not necessarilymutually exclusive. In Experiment 3, the per-formances of 2 identity subjects and all 6 odditysubjects suggested control by specific stimulusrelations. In addition, selected subjects alsoshowed control by the general relations amongstimuli. The finding of control by specific re-lations replicates, in part, earlier studies ofidentity matching (Dixon & Dixon, 1978;

ble 4

Conditions for identity and oddity (*) subjects in Experiment 4. The A or B next to theparentheses indicates the stimulus set used during new subject's training; numbers within reflectsessions to reach criteria. Test sessions for comparison (CO) and sample/S+ (ST/S+) controlare listed in the order given. Figures 5 and 10 show stimulus arrays given during these conditions.Test numbers in table and figures correspond.

TG GT* MC* GA* LJ*

Test CO Test CO Test CO Trn A(1-7) Trn B(1-6)Control: Control: Control:

Test CO Test COTest 3 Test 1 Test 3 Control: Control:Test 4 Test 2 Test 4Test 3 Test 1 Test 3 Test 2 Test 4Test 4 Test 2 Test 4 Test 1 Test 3

Test 2 Test 4Test 1 Test 3

Test ST/S+ Test ST/S+Control: Control:

Test 1 Test 3Test 2 Test 4Test 1 Test 3Test 2 Test 4

62

IDENTITY MATCHING AND ODDITY

zCD

l-)LULLCr)

zLUC)

LU0)

T

G.RA. L. J.

SUBJECTS FI IcCP RrsoN CONTROLSAMPLE/S$ CONTROL

Fig. 11. Test results for Experiment 4. Subjects TG, GT, and MC participated in Experiment 3 prior to thisexperiment. GA and Li were taught oddity prior to testing. Bars show percentages of selecting substitute comparisons(TG), or familiar A or B comparisons (all oddity subjects) during tests of comparison control and tests of sample/S+control. The 16 test trials (see Figures 5 and 10) were interspersed among 32 AA training trials (GT and GA) or 32BB training trials (TG, MC, and Li). Bars are arranged in the order tests were given (see Table 4).

Dixon et al., 1983) and extends the analysisto children's oddity performance.

If one assumes that a subject is capable ofacquiring either generalized or specific rela-tions, perhaps it is not surprising to find gen-eralized rather than specific control. Learninga single relation seems to require less effortthan does learning several specific relations.Consequently, it was interesting to find chil-dren who displayed both kinds of control si-multaneously (Experiment 3). Also interestingwas that generalized control developed withonly minimal training (a single set of two stim-uli).The lack of consistent control by sample/

S - relations in identity contrasts sharply withthe remarkably clear findings of Dixon andDixon (1978). The lack of sample/S+ controlin oddity following AA/BB training (Exper-iments 1 and 2) also requires clarification. Sev-eral procedural differences between this studyand previous ones should be examined system-atically. By themselves or collectively, factorssuch as the preliminary verbal instructions, thenumber of training trials, the kinds of conse-quences used during training, and the contin-gencies operating during testing may have in-fluenced performance.The ability to learn both positive and neg-

ative relations among samples and comparisonstimuli is considered to be critical to the for-mation of generalized identity and oddity re-

lations. Generalized control occurs when asubject learns which comparison stimulus toselect in the presence of a given sample andwhich comparison not to select. This concep-tualization has emphasized the role of controlby specific relations, either as an outcome per-formance (Dixon & Dixon, 1978) or as a train-ing variable (Urcuioli, 1977; Urcuioli & Nevin,1975). But, as shown in the present study,generalized control does not always translateinto control by specific sample-comparison re-lations. Moreover, training specific relationsis not necessarily a sufficient precondition forthe emergence of generalization (Santi, 1982;Urcuioli, 1977; Urcuioli & Nevin, 1975).

Perhaps too much emphasis has been placedon the separate functions of samples and com-parison stimuli in attempts to account for gen-eralized control. As Cumming and Berryman(1965, p. 286) suggested, identity-matching andoddity performance may be based on the entiresample-comparison stimulus array. Identityand oddity arrays may function as a "unitarydiscriminative stimulus." The stimulus con-figuration itself controls responding (cf. Carter& Werner, 1978). But with humans, such con-trol is obviously not restricted to specific stim-ulus configurations. Generalized control sug-gests that any unitary discriminative functionis unrestricted; it is shared by any stimulusarray that conforms to identity or oddity con-figurations.

I63

G. T. M. C.

64 ROBERT STROMER and JOAN BUTCHER STROMER

Additional research is needed to clarify therelationship, if any, between specific and gen-eralized sample-comparison relations. The dif-ficulties in producing specific control in thepresent study suggest that, for humans, thissource of control may be exhibited only underspecial circumstances. The ability to treat var-ious identity and oddity arrays as unitary dis-criminative stimuli may, however, be relatedin some distant way to specific relational con-trol. When children display specific control inthe laboratory, test conditions may tap an im-portant ability developed outside the experi-mental setting-the ability to discriminate bothcorrect and incorrect stimulus selections in avariety of contexts. To produce generalizedcontrol in subjects who do not show it readily(e.g., pigeons), an effective tactic might com-bine the learning set approach (Holmes, 1979)with explicit training on positive and negativesample-comparison relations (Santi, 1982; Ur-cuioli, 1977; Urcuioli & Nevin, 1975).

REFERENCESBerryman, R., Cumming, W. W., Cohen, L. R., & John-

son, D. F. (1965). Acquisition and transfer of si-multaneous oddity. Psychological Reports, 17, 767-775.

Carter, D. E., & Werner, TJ. (1978). Complex learningand information processing by pigeons: A critical anal-ysis. Journal of the Experimental Analysis of Behavior,29, 565-601.

Cumming, W. W., & Berryman, R. (1961). Some dataon matching behavior in the pigeon. Journal of the Ex-perimental Analysis of Behavior, 4, 281-284.

Cumming, W. W., & Berryman, R. (1965). The com-plex discriminated operant: Studies of matching-to-sample and related problems. In D. I. Mostofsky (Ed.),Stimulus generalization (pp. 284-330). Stanford, CA:Stanford University Press.

Dixon, M. H., & Dixon, L. S. (1978). The nature ofstandard control in children's matching-to-sample.Journal of the Experimental Analysis of Behavior, 30,205-212.

Dixon, M. H., Dixon, L. S., & Spradlin, J. E. (1983).Analysis of individual differences of stimulus controlamong developmentally disabled children. In K. D.Gadow & I. Bialer (Eds.), Advances in learning and

behavioral disabilities (Vol. 2, pp. 85-110). New York:JAI Press.

Holmes, P. W. (1979). Transfer of matching perfor-mance in pigeons. Journal of the Experimental Analysisof Behavior, 31, 103-114.

Levin, G. R., & Maurer, D. M. (1969). The solutionprocess in children's matching-to-sample. Develop-mental Psychology, 1, 679-690.

Mcllvane, W. J., Munson, L. C., & Stoddard, L. T.(1988). Some observations on control by spoken wordsin children's conditional discrimination and matchingby exclusion. Journal of Experimental Child Psychology,45, 472-495.

Santi, A. (1982). Hue matching and hue oddity in pi-geons: Is explicit training not to peck incorrect huecombinations a sufficient condition for transfer? Psy-chological Record, 32, 61-73.

Saunders, R. R., & Sherman, J. A. (1986). Analysis ofthe "discrimination-failure hypothesis" in generalizedmatching and mismatching behavior. Analysis and In-tervention in Developmental Disabilities, 6, 89-107.

Scott, K. G. (1964). A comparison of similarity andoddity. Journal ofExperimental Child Psychology, 1, 123-134.

Sherman, J. A., Saunders, R. R., & Brigham, T. A. (1970).Transfer of matching and mismatching behavior inpreschool children. Journal of Experimental Child Psy-chology, 9, 489-498.

Stoddard, L. T. (1968). An observation on stimuluscontrol in a tilt discrimination by children. Journal ofthe Experimental Analysis of Behavior, 11, 321-324.

Stoddard, L. T., & Sidman, M. (1971). The removaland restoration of stimulus control. Journal of the Ex-perimental Analysis of Behavior, 16, 143-154.

Stokes, T. F., & Baer, D. M. (1977). An implicit tech-nology of generalization. Journal of Applied BehaviorAnalysis, 10, 349-367.

Stromer, R., & Osborne, J. G. (1982). Control of ad-olescents' arbitrary matching-to-sample by positive andnegative stimulus relations. Journal of the ExperimentalAnalysis of Behavior, 37, 329-348.

Urcuioli, P. J. (1977). Transfer of oddity-from-sampleperformance in pigeons. Journal of the ExperimentalAnalysis of Behavior, 27, 195-202.

Urcuioli, P. J., & Nevin, J. A. (1975). Transfer of huematching in pigeons. Journal of the Experimental Anal-ysis of Behavior, 24, 149-155.

Zentall, T. R., Edwards, C. A., Moore, B. S., & Hogan,D. E. (1981). Identity: The basis for both matchingand oddity learning in pigeons. Journal ofExperimentalPsychology: Animal Behavior Processes, 7, 70-86.

Received February 11, 1988Final acceptance September 7, 1988