The Analysis of Verbal Behavior 1990, 8, 13-17

Do Stimulus Classes Exist Before They are Tested?William J. McIlvane and William V. Dube

E. K. Shriver CenterThis paper addresses verbal practices that are common when behavior analysts talk aboutstimulus classes. Specifically, we examine some of the conditions that may set the occasion forsaying "Stimulus classes exist prior to the tests that document their formation." We suggestthat "stimulus class" should tact behavior that is a function of training and test procedures, notentities that "form" or "exist" in any independent sense. To frame our arguments, we reviewrelevant research findings and suggest descriptive language that is more consistent withbehavior analytic traditions.

Can stimulus classes be said to existprior to test outcomes that set the occasionfor saying that classes currently exist?Phrased in this way, the question soundsodd, but it is semantically equivalent tothose that often come up when behavioranalysts talk about stimulus classes.Moreover, there have been a number ofexperiments conducted to try to gather evi-dence of class formation prior to the tests(e.g., K. Saunders, Spradlin, & R. Saunders,1989a). The provenance of these effortsappears to have been a few lines in an arti-cle by Sidman, Kirk, and Willson-Morris(1985) on six-stage equivalence class for-mation. They wrote that "Strong rationalsupport . . . exists for supposing thatclasses of equivalent stimuli form duringthe test rather than during the originalteaching" (p. 39). Subsequently, Sidmanwent on to clarify this position: He sug-gested that the equivalence tests may pro-

Preparation of this manuscript was supported byNICHD Grants HD 22218 and HD 25995, and in partby the Department of Mental Retardation of theCommonwealth of Massachusetts (Contract: 3403-8403-306). A preliminary version of this paper waspresented at the 1989 meeting of the Association forBehavior Analysis. We thank Kate Saunders andColeman Paul for helpful comments on that presenta-tion which helped us better formulate the argumentspresented here. We also thank Joe Spradlin for hisinsightful comments on an earlier version of themanuscript. Correspondence can be addressed toWilliam J. McIlvane, Department of BehaviorAnalysis, E. K. Shriver Center, 200 Trapelo Road,Waltham, MA 02254.

vide additional experience that helps sepa-rate equivalence relations from others thatare possible in the experimental situation(Sidman, in press). However, Sidman andcolleagues' (1985) discussions have beeninterpreted as asserting something more-that the tests for reflexibility, symmetry,and transitivity (Sidman & Tailby, 1982)are in fact necessary to form the classes.What followed naturally were efforts toshow that equivalence classes can formbefore the tests.The fact that experiments like this are

being conducted shows important growthwithin behavior analysis. Particularly inthe past decade, behavior analysts havesought answers to ever more complexquestions about behavior involved in cog-nition. In the excitement of pursuing newquestions, however, it is easy to lose sightof the fundamentals temporarily. We seeevidence of that when we examine some ofthe verbal behavior that is used to discusswork on stimulus equivalence and relatedphenomena. The purpose of this article isto help to encourage verbal practices thatare more consistent with behavior analytictraditions.We suggest that questions such as "Do

stimulus classes exist before they aretested?" are not good ones to ask. In ourview, such questions create conceptualproblems that behavior analysis usuallyhelps us to avoid. We also suggest alterna-

13

14 WILLIAM J. McILVANE and WILLIAM V. DUBE

tives that help to avoid the conceptualproblems.Two examples will help frame our argu-

ments. The examples involve emergentbehavior that seems to have features incommon with that emitted on tests forstimulus class formation. The first is a sen-tence spoken by Alfred North Whiteheadto B. F. Skinner while they were dining atthe Harvard Society of Fellows (Skinner,1957, p. 457). Whitehead said "Let me seeyou account for my behavior as I sit heresaying 'No black scorpion is falling uponthis table."' Whitehead's challenge was forSkinner (and behavior analysis in general)to identify the variables that determineproduction of novel sentences that hadnever been spoken before and thus had noexplicit history of reinforcement. Skinner'sattempt to do that ultimately resulted inVerbal Behavior, and much stimulus equiva-lence work has a similar purpose. For ourpurposes, we may ask about the status ofWhitehead's sentence before it was spoken.Did it exist in some sense and if so how?Our second example is that of the skilled

musician who departs from his or herprinted music and improvises entirely newvariations on old familiar themes. What isthe status of those improvisations beforethey are played for the first time? Do theyexist and if so how?

In what way is the behavior involved inthese examples comparable to that emittedon equivalence tests? Most behavior ana-lysts would agree that all three types ofbehavior are a function of variables that"lie outside the organism, in its immediateenvironment and in its environmental his-tory" (Skinner, 1953, p. 31). What separatesemergent matching-to-sample perfor-mances, novel sentences, and improvisa-tions from other forms of behavior islargely that they have not been previouslyemitted. Of course, a fully comprehensivescience of behavior must have a frame-work for accounting for emergent behav-ior. In behavior analysis, the traditionalapproach has been to trace emergent per-formances to other behavior previouslyacquired (behavioral prerequisites; cf.Epstein, 1985; Holland, Solomon, Doran, &

Frezza, 1976). As an example, considerSkinner's (1957, pp. 187-190) account of theconditions leading to the emission of amand for the first time. Some prerequisitesmay be obvious and others perhaps not so,but they must be specifiable in terms ofprevious environment-behavior interac-tions if the science of behavior is to suc-ceed. In the behavior analytic tradition, onewould never say that Whitehead's sentenceand the musician's improvisation existedbefore they were emitted. Rather, onewould say only that prior experience estab-lished the prerequisites for the emission ofthe new behavior at the appropriate time.We suggest that verbal behavior describingstimulus classes (and their "formation")should remain faithful to this tradition.When we assert that stimulus classes

exist before they are tested, we edgetoward subtly transforming "class forma-tion" into an event that is outside thedomain of behavior analysis. For example,one might view class formation as a biolog-ical event. It seems self-evident that anybehavioral change must have a biologicalcorrelate. But when we work at a purelybehavioral level, we must be cautious toavoid unintended pursuit of problems inneuroscience and neuropsychology. In ourjudgment, few behavior analysts would doso. More likely, however, is the potentialfor unintended analysis of the formation ofconnections in some conceptual nervoussystem or a cognitive analog of it. Suchanalyses, of course, are clear departuresfrom the behavior analytic enterprise, thegoal of which is to document functionalrelations between specific environmentaloperations and changes in behavior. Whenwe say that "classes exist before they aretested," we are not consistent with thenature of our science.The problem that we are addressing is

similar in kind to that which occurs whenwe try to determine what feature(s) of acomplex discriminative stimulus have con-trolled a given response during training.We present the subject with new displaysthat do or do not contain specific features,and measure the subject's response. Undersome conditions of testing, our findings

STIMULUS CLASSES 15

seem to indicate that, say, Feature A didcontrol and that Feature B did not (e.g.,Reynolds, 1961). If we use somewhat dif-ferent test procedures, however, we mayobtain very different results (cf. Wilkie &Masson, 1976). Such findings illustrate thatthe test procedures themselves exert con-trol over what the subject does. This factshould caution us against casual use of testoutcomes to support our inferences aboutthe stimuli that controlled prior to the test.All that we are truly certain about is thatthe subject, who has had exposure to cer-tain training contingencies, interacted withthe test conditions to produce the test out-come.

In this light, how can we describe find-ings taken as evidence that classes existbefore they are tested? We will considertwo examples. One very clever effort wasreported by K. Saunders, Spradlin, and R.Saunders (1989a). Figure 1 presents theessence of their experimental design.(Their design actually involved more stim-uli, but it has been simplified here to aidexposition.)Mentally retarded human subjects first

learned the conditional-discrimination per-formances represented in the left portion ofFigure 1. The procedure was matching tosample; arrows point from samples to com-parisons. The subjects learned to selectvisual comparison stimuli from Set A con-ditionally upon samples from Sets B and C.Subjects also learned to select comparisonstimuli from Set E conditionally upon sam-ples from Sets F and G. Unreinforced tests

for emergent matching performances doc-umented the formation of four three-mem-ber stimulus classes: Al-Bl-Cl, A2-B2-C2,E3-F3-G3, and E4-F4-G4.That is, on test trials subjects selected Cl

conditionally upon Bi, C2 upon B2, G3upon F3, and so forth. Note that the ABCand EFG classes were independent of oneanother.The middle portion of Figure 1 illus-

trates the next phase of the experiment. K.Saunders and colleagues instituted condi-tions sufficient for a merger of the classesvia an "unreinforced conditional selection"procedure (R. Saunders, K. Saunders,Kirby, & Spradlin, 1988). Unreinforced tri-als displayed Set F stimuli as samples andSet A stimuli as comparisons. Consistentwith previous reports of unreinforced con-ditional selection, subjects reliably selectedone comparison in the presence of onesample and the other comparison in thepresence of the other sample. For purposeof illustration, let us consider the caseshown in the center portion of Figure 1 inwhich the subject selects Al conditionallyupon F3, and A2 upon F4. Based on thefindings of R. Saunders and colleagues'(1988) study, one would predict the mergerof the ABC and EFG classes into two largerclasses, Al-Bl-Cl-E3-F3-G3 and A2-B2-C2-E4-F4-G4. Such class formation would bedocumented by test trials in which subjectsselected G3 conditionally upon Al, G4upon A2, E3 upon Bi, and so forth.However, the tests for class formation werenot conducted at this point.

Bi c Bi c Bi ClB2 /C2 B2\ 6C2 B2\ C2

\AI AlA2A2 A2

3 F3 G3 F3 G3F41.F4A F4 vb

E4 E4 E4

Fig. 1. A simplified version of the experimental designed used by Saunders, Spradlin, and Saunders (1989a).

16 WILLIAM J. McILVANE and WILLIAM V. DUBE

The right portion of Figure 1 diagramsthe final phase of the experiment. In it, theperformances that were previously dis-played via unreinforced conditional selec-tion were reversed with a differential rein-forcement procedure. Subjects were taughtexplicitly to select Al conditionally uponF4, and A2 upon F3. Given such training, areasonable prediction would be the result-ing stimulus classes A1-B1-C1-E4-F4-G4and A2-B2-C2-E3-F3-G3; the predictedemergent performances would be selectionof G4 conditionally upon Al, G3 upon A2,E4 upon Bi, and so forth. When class-for-mation tests were conducted, however, theresults were contrary to that prediction.Instead the relations that emerged were infact consistent with the subjects' originalunreinforced conditional selections (asshown in the middle portion of the figure):selection of G3 conditionally upon Al, G4upon A2, E3 upon Bi, and so forth.One interpretation of these findings is

that the classes had formed earlier-priorto the reversal phase and, of course, priorto the tests for class formation. However,an analysis in terms of specifiable environ-ment-behavior interactions tells us merelythat the behavior emitted on the class testswas consistent with the training historyprior to the reversal phase. A logical nextquestion is whether this outcome willalways occur or whether the test behaviormight be consistent with the reversalunder some circumstances. Is anythingadded by asking whether classes formedprior to the tests?One might argue that "classes form

before the tests" adds a convenient, harm-less way to tact a relationship of sufficiencybetween a given history and a given testoutcome. Our next example illustrates whysuch practice can lead to conceptual diffi-culties and why it should be avoided, espe-cially when less troublesome and equallyspare descriptions are available (K.Saunders, Spradlin, & R. Saunders, 1989b).Recent experimental work has demon-

strated transfer of discriminative (e.g., deRose, McIlvane, Dube, Galpin, & Stoddard,1988) and conditioned reinforcing (Hayes,Brownstein, Devany, Kohlenberg, &

Shelby, in press) functions among condi-tionally related stimuli. In de Rose and col-leagues' study, subjects acquired a simple,simultaneous discrimination in whichvisual stimuli Al and A2 served as S+ andS- respectively. Al and A2 also served assamples on conditional-discrimination tri-als, controlling selections of comparisonsBi and B2, respectively. When Bi and B2were displayed in the simultaneous dis-crimination format, subjects typicallyselected Bi and rejected B2; the A- and B-stimuli were thus shown to be members ofthe same functional stimulus class underthese conditions.One might go on to suggest that the

transfer of functions is a characteristic ofequivalence class members, and that suchtransfer is evidence that equivalenceclasses (Al-Bi and A2-B2) had formed pre-viously (cf. Hayes, 1990). By this logic, pos-itive transfer findings would be evidencethat equivalence classes can exist prior toreflexivity, symmetry, and transitivitytests. This interpretation, however, is notconsistent with recent research findings.For example, Sidman, Wynne, Maguire,and Barnes (1989) reported a case in whichequivalence-class test outcomes were nega-tive even though discriminative functionstransferred among the stimuli. Moreover,an accumulating body of evidence is docu-menting that functional class membership(as exemplified by de Rose and colleagues'study) and equivalence class membershipmay be independent under some circum-stances (de Rose, McIlvane, Dube, &Stoddard, 1988; Mcllvane, Dube, Kledaras,Iennaco, & Stoddard, 1989). Also, it is wellknown that oral naming can but need nottransfer among members of equivalenceclasses (e.g., Lazar, Davis-Lang, & Sanchez,1984). The implications of these variousfinding seem clear. Although emergentmatching-to-sample performances, func-tion transfers, and the like may have manyoverlapping behavioral prerequisites, theirindependence under some conditionsshows that they do not have identical pre-requisites.These findings may surprise us only

when we think or speak of emergent

STIMULUS CLASSES 17

behavior as the product of stimulus classesthat form somewhere. When we do that,we may find ourselves looking around forthe various manifestations of class forma-tion and expecting those manifestations tocovary. However, when "stimulus class"merely tacts test outcomes that are a func-tion of prior exposure to specified trainingcontingencies, then we find it unremark-able that, for example, functional classesare displayed in the absence of equivalenceclasses and vice versa. Further, to theextent that our verbal behavior sets theoccasion for other behavior-our own, ourcolleagues', or our students'-we then maybe more likely to attempt to relate suchfindings to specific experimental proce-dures and behavioral histories.

REFERENCESde Rose, J.C., Mcllvane, W.J., Dube, W.V., Galpin, V.,& Stoddard, L.T. (1988). Emergent simple discrimi-nation established by indirect relation to differentialconsequences. Journal of the Experimental Analysis ofBehavior, 50, 1-20.

de Rose, J.C., McIlvane, W.J., Dube, W.V., &Stoddard, L.T. (1988). Stimulus class formation andfunctional equivalence in moderately retarded indi-viduals' conditional discrimination. BehaviouralProcesses, 17, 167-175.

Epstein, R. (1985). The spontaneous interconnection ofthree repertoires. The Psychological Record, 35, 131-141.

Hayes, S.C. (1990). A relational control theory of stim-ulus equivalence. In L.J. Hayes & P.N. Chase (Eds.),Dialogues on verbal behavior: Proceedings of the FirstInternational Institute on Verbal Relations (pp. 19-40).Reno, NV: Context Press.

Hayes, S.C., Brownstein, A.J., Devany, J.M.,Kohlenberg, B.S., & Shelby, J. (in press). Stimulusequivalence and the symbolic control of behavior.Revista Mexicana de Analisis de la Conducta.

Holland, J.G., Solomon, C., Doran, J., & Frezza, D.A.(1976). The analysis of behavior in planning instruction.Reading, MA: Addison-Wesley.

Lazar, R., Davis-Lang, D., & Sanchez, L. (1984). Theformation of visual stimulus equivalence in chil-dren. Journal of the Experimental Analysis of Behavior,41,251-266.

McIlvane, W.J., Dube, W.V., Kledaras, J.B., Iennaco,F.M. & Stoddard, L.T. (1989). Stimulus-consequencerelations and stimulus classes in mental retardation.Experimental Analysis of Human Behavior Bulletin, 7,17-20.

Reynolds, G.S. (1961). Attention in the pigeon. Journalof the Experimental Analysis of Behavior, 4, 203-208.

Saunders, K.J., Spradlin, J.E., & Saunders, R.R.(1989a). Detecting equivalence relations that developprior to testing. Proceedings of the 22nd AnnualGatlinburg Conference on Research and Theory inMental Retardation and DevelopmentalDisabilities, 32.

Saunders, K.J., Spradlin, J.E., & Saunders, R.R.(1989b). Assessing the prerequisites for merger ofequivalence classes. Experimental Analysis of HumanBehavior Bulletin, 7, 20-21.

Saunders, R.R., Saunders, K.J., Kirby, K.C., &Spradlin, J.E. (1988). The merger and developmentof equivalence classes by unreinforced conditionalselection of comparison stimuli. Journal of theExperimental Analysis of Behavior, 50, 145-162.

Sidman, M. (in press). Equivalence relations: Wheredo they come from? In H. Lejeune & D. Blackman(Eds.), Proceedings of EMEABII. Hillsdale, NJ:Erlbaum.

Sidman, M., Kirk, B., & Willson-Morris, M. (1985). Six-member stimulus classes generated by conditionaldiscrimination procedures. Journal of the Experi-mental Analysis of Behavior, 43, 21-42.

Sidman, M., & Tailby, W. (1982). Conditional discrim-ination vs. matching-to-sample: An expansion ofthe testing paradigm. Journal of the ExperimentalAnalysis of Behavior, 37, 5-22.

Sidman, M., Wynne, C.K., Maguire, R.W., & Barnes,T. (1989). Functional classes and equivalence rela-tions. Journal of the Experimental Analysis of Behavior,52,261-274.

Skinner, B.F. (1953). Science and human behavior. NewYork: Macmillan.

Skinner, B.F. (1957). Verbal behavior. New York:Appleton-Century-Crofts.

Wilkie, B.M., & Masson, M.E. (1976). Attention inpigeon: A reevaluation. Journal of the ExperimentalAnalysis of Behavior, 26, 207-212.