EQUIVALENCE CLASSIFICATION BY CALIFORNIA SEA …people.uncw.edu/galizio/website/Articles/Kastak 2001.pdf · 131 journal of the experimental analysis of behavior 2001, 76, 131–158

131

JOURNAL OF THE EXPERIMENTAL ANALYSIS OF BEHAVIOR 2001, 76, 131–158 NUMBER 2 (SEPTEMBER)

EQUIVALENCE CLASSIFICATION BY CALIFORNIASEA LIONS USING CLASS-SPECIFIC REINFORCERS

COLLEEN REICHMUTH KASTAK, RONALD J. SCHUSTERMAN,AND DAVID KASTAK

LONG MARINE LABORATORY,UNIVERSITY OF CALIFORNIA SANTA CRUZ

The ability to group dissimilar stimuli into categories on the basis of common stimulus relations(stimulus equivalence) or common functional relations (functional equivalence) has been convinc-ingly demonstrated in verbally competent subjects. However, there are investigations with verballylimited humans and with nonhuman animals that suggest that the formation and use of classificationschemes based on equivalence does not depend on linguistic skills. The present investigation doc-umented the ability of two California sea lions to classify stimuli into functional classes using a simplediscrimination reversal procedure. Following the formation of functional classes in this context, thesecond experiment showed transfer of the relations that emerged between class members to a match-ing-to-sample procedure. The third experiment demonstrated that the functional classes could beexpanded through traditionally defined equivalence relations. In these three experiments, appro-priate within-class responding produced class-specific food reinforcers. Experiment 3 addressed therole of these reinforcers in equivalence classification and showed that the class-specific reinforcerswere sufficient to relate new stimuli to the functional classes. These findings show that sea lions canform equivalence classes in simple and conditional discrimination procedures, and that class-specificreinforcers can become equivalence class members.

Key words: stimulus equivalence, functional classes, reversal procedure, simple discrimination, con-ditional discrimination, differential outcome, California sea lions

Equivalence classification, or simply equiv-alence, occurs when perceptually dissimilarstimuli come to exert similar control over be-havior through emergent relations. Indeed,equivalence is demonstrated by a subject’ssuccessful performance on tests of emergentrelations following specific training. Since theconcept of equivalence was first applied tostudies of symbolic behavior in humans,equivalence relations have been further ex-perimentally or operationally defined. Fordifferent stimuli to be considered equivalent,the relations that emerge between them mustmeet the mathematically derived criteria of

Research was supported by ONR Grant N00014-95-1-0936 to R. J. Schusterman and a DoD AASERT Fellow-ship to C. Reichmuth Kastak. Additional support was pro-vided by the Earl and Ethel Myers Oceanographic Trust,the Friends of Long Marine Laboratory, and the UCSCOcean Sciences Board. Parts of this manuscript were pre-sented at the meeting of the Animal Behavior Society inFlagstaff, Arizona, July 1996, and at the 13th Biennial Ma-rine Mammal Conference in Wailea, Hawaii, November1999.

We thank the dedicated staff of the Pinniped Researchin Cognition and Sensory Systems laboratory at UCSC,especially Grace Ho and Shannon Spillman.

Address correspondence and reprint requests to Col-leen Reichmuth Kastak, UCSC Long Marine Laboratory,100 Shaffer Road, Santa Cruz, California 95060 (E-mail:[email protected]).

reflexivity, symmetry, and transitivity (Sidman& Tailby, 1982). Thus, the term stimulus equiv-alence describes groups of stimuli that becomeinterrelated in specific and verifiable ways.Reflexive relations are those in which a stim-ulus is conditionally related to itself (i.e., gen-eralized identity matching: A is related to A,or ArA, BrB, CrC). Symmetrical relations arethose that exhibit emergent bidirectionality(i.e., if ArB, then BrA; if BrC, then CrB).Transitive relations are those that include anemergent forward relation (i.e., if ArB andBrC, then ArC). A typical test for equivalencemight assess the emergent relation that com-bines symmetry and transitivity (i.e., if ArBand BrC, then CrA). Thus, the model of stim-ulus equivalence describes how untrained re-lations (generalized identity, symmetry, andtransitivity) arise from trained stimulus rela-tions (in this case, ArB, BrC).

Due to the explicit nature of these definingproperties, equivalence can be effectivelydemonstrated only by training subjects in thecontext of a conditional discrimination ormatching-to-sample (MTS) procedure. Thishas been done convincingly in a variety ofstudies with human subjects. Many studiesshow that mentally competent adults, chil-

132 COLLEEN REICHMUTH KASTAK et al.

dren as young as 2 years of age, and evenmany mentally disabled people readily linkphysically dissimilar stimuli into equivalenceclasses using these training procedures. Al-though verbally able subjects have successful-ly demonstrated equivalence, human subjectswho lack basic language skills have historical-ly not passed such tests. These observations,coupled with several failed attempts to dem-onstrate equivalence in nonhuman animals,have prompted some to conclude that theability to form equivalence classes is uniqueto linguistically competent humans (for re-views of relevant studies and for differingviewpoints on this issue, see Fields & Nevin,1993; Hayes, 1989; Horne & Lowe, 1996,1997; and Sidman, 1994).

However, recent investigations with nonver-bal subjects do not support this anthropocen-tric and ‘‘language-centric’’ view of stimulusequivalence. Carr, Wilkinson, Blackman, andMcIlvane (2000) demonstrated equivalencein several developmentally disabled adultswith virtually no functional spoken language.Further, although evidence for emergent re-flexivity, symmetry, and transitivity in nonhu-man animals has been difficult to obtain ex-perimentally, several recent reports documentthe requisite relations emerging throughMTS procedures in one or more nonhumanspecies (see brief review in Zentall, 1998; seealso Kastak & Schusterman, 1994; Manabe,Kawashima, & Staddon, 1995). The strongestevidence for stimulus equivalence by a non-human animal has been reported by Schus-terman and Kastak (1993), who trained a Cal-ifornia sea lion with a subset of MTSproblems that included combinations ofemergent relations (e.g., symmetry, transitiv-ity). Following this training, the sea lionshowed significant performance on a largerset of completely novel transfer problems.The authors concluded that the lack ofstrong transfer in many early attempts todemonstrate reflexivity, symmetry, and tran-sitivity (see, e.g., D’Amato, Salmon, Loukas,& Tomie, 1985; Sidman et al., 1982) could beovercome when testing procedures weremodified to provide nonverbal subjects witha large number of training exemplars orwhen the potentially disruptive effect of novelstimulus position was mitigated prior to orduring testing. They further suggested thatestablishing successful generalized identity

matching performance likely facilitated sub-sequent performance on tests of combinedsymmetry and transitivity.

Not all of the emergent relations measuredin humans or animals fall neatly into the par-adigm of stimulus equivalence. For example,Keller and Schoenfeld (1950) described stud-ies of semantic generalization conducted byRiess (1940) in which functional relations be-tween words similar in function but not inform emerged in a classical conditioning con-text. When a galvanic skin response was con-ditioned to a word such as urn, much stron-ger generalization occurred to the synonymvase than to the homonym earn. Generaliza-tion between words similar in function ratherthan structure implies an emergent equiva-lence between the words. Correspondingly,many studies have demonstrated emergentrelations in nonhuman animals outside theconstraints of the operational definition ofstimulus equivalence. The experimental pro-cedures used in these studies include classicalconditioning, MTS procedures, sequentiallearning procedures, hierarchical stimuluscategorization tasks, and simple discrimina-tion reversal procedures. The procedural andtheoretical differences introduced in theseand other tasks have led to an array of over-lapping terms used to describe emergent re-lations, including symbolic representation;acquired equivalence of cues; mediated gen-eralization; functional equivalence, categori-zation, or classification; abstract concept for-mation; and non-similarity-based classification.Because the relations between stimuli orevents examined in these studies emergedoutside the context of conditional discrimi-nation procedures or traditional testing par-adigms, it is impossible to determine in mostcases whether these instances are examples ofstimulus equivalence as currently defined.However, a review of these studies does leaveone with the distinct impression that theseare related phenomena (see reviews andcommentary in Balsam, 1988; Dube, Mc-Ilvane, Callahan, & Stoddard, 1993; Schuster-man & Kastak, 1998; Sidman, 1994; Tomo-naga, 1999; Vaughan, 1988; Wasserman &DeVolder, 1993; Zentall, 1998).

Some investigators have used broad criteriato define the emergent properties of stimulusclasses. For example, Wasserman and his col-leagues (Wasserman & DeVolder, 1993; Was-

133EQUIVALENCE CLASSIFICATION BY SEA LIONS

serman, DeVolder, & Coppage, 1992) de-scribe the formation of non-similarity-basedclasses by the emergence of untrained rela-tions arising between dissimilar stimuli. Oth-ers have defined the classification process bythe procedure used to measure it. The latterseems to be the case with the definition ofstimulus equivalence as formulated by Sid-man and Tailby (1982). This classificationscheme is also based on relations that emergebetween stimulus class members, but includesthe specific mathematically derived proper-ties of the emergent relations as criteria. Thedemonstration of these criteria is restricted toa narrow experimental context (an MTS pro-cedure) and, thus, this definition of stimulusequivalence excludes emergent abilities thatarise in other procedures. Vaughan (1988)first addressed this issue by proposing that be-haviorally and mathematically valid equiva-lence relations could arise in an alternativeprocedure without demonstrating the pres-ence of identity, symmetry, and transitivityper se. Vaughan theoretically and empiricallysought to eliminate the distinction betweenstimulus classes linked by stimulus equiva-lence and those grouped into partitions, orfunctional classes.

Functional stimulus classes are sets of dis-criminative stimuli that control the same be-havior. The members of a functional classshare a high correlation with a particular re-sponse, such that responding to all classmembers is altered when responding to anyone class member is altered (Skinner, 1935).There are limited data available on the for-mation of functional classes in nonhuman an-imals. Schusterman and Gisiner (1997) sug-gested that the grammatical sequences ofsigns or lexigrams used in animal languageresearch may lead to the formation of func-tional classes. In many of these studies, ref-erential signs of a given type (e.g., ‘‘objects,’’‘‘actions,’’ or ‘‘modifiers’’) can be inter-changed with one another without disruptingthe resulting performance of the animal. Forexample, any sign representing an object,whether a ball, cone, or cube, or even a novelobject, generates an object-oriented responsewhen placed in the correct position in an in-structional sequence. However, if the signsfor an object and an action are transposedbetween their standard positions in a se-quence, the performance of the animal de-

teriorates. Schusterman and Gisiner cited thisapparent substitutability of stimuli sharing se-quence positions as evidence of functionalclass formation, with members of each stim-ulus class controlling similar response topog-raphies that did not extend to stimuli occu-pying other sequential positions. Furthersupport for the idea that functional classescan arise from sequential procedures in non-human animals comes from studies with rhe-sus monkeys showing a transfer of functionbetween stimuli sharing the same ordinal po-sitions in different stimulus sequences (Chen,Swartz, & Terrace, 1997).

The best evidence for functional class for-mation by nonhumans comes from Vaughan(1988). He trained pigeons on a discrimina-tion reversal procedure in which the subjectswere presented with a sequence of 40 differ-ent slides of trees that were divided into twoarbitrary sets of 20 slides each. The pigeonswere conditioned to peck at any of one set ofslides, designated as positive, and to withholdpecking when presented with any of the otherset of slides, designated as negative. Followinglearning of the positive set, the reinforce-ment contingencies were reversed, and mem-bers of the formerly negative set were rein-forced as positive. After repeatedly shiftingthe reinforcement contingencies between thetwo sets of stimuli, the pigeons began chang-ing their responses to all members of a setafter experiencing the reversed contingencywith just a few. Thus, the reversed contingen-cy for slides presented at the beginning of asession predicted reversed contingencies forslides presented in the remainder of the ses-sion.

Within the context of this simple discrimi-nation reversal procedure, Vaughan (1988)showed that pigeons classified a large set ofstimuli into two functionally equivalent subsetsbased only on shared reinforcement historiesof the stimuli. He argued that the relationsthat eventually emerged between stimuli, asdemonstrated by a transfer of function be-tween each of the stimuli in a set, impliedstimulus equivalence as well as functionalequivalence. Vaughan’s viewpoint establisheda basis for investigating the specific conditionsthat give rise to equivalence. Studies with hu-man subjects show unequivocally that tradi-tionally defined equivalence classes estab-lished in an MTS procedure immediately


transfer to functional classes demonstrated ina simple discrimination procedure (Lazar,1977; Wulfert & Hayes, 1988). This transferfrom MTS to simple discriminations was alsofound following the only study to successfullydemonstrate equivalence in a nonhuman an-imal, a California sea lion (Schusterman &Kastak, 1998). However, if equivalence rela-tions and functional classes reveal the samecognitive-behavioral processes through differ-ent procedures, then functional classesshould also transfer to equivalence classes inan MTS context.

Although Vaughan (1988) did not extendhis functional classification experiment to anMTS context, Sidman, Wynne, Maguire, andBarnes (1989) followed Vaughan’s study witha multistep classification experiment to testwhether the functional classes formed by hu-man subjects were also equivalence classes.To accomplish this, these investigators (a)generated functional classes in a simple dis-crimination reversal procedure using an ap-proach similar to Vaughan’s, (b) presentedthe functional class members in the contextof an MTS procedure to determine if class-consistent conditional discriminationsemerged, and (c) tested whether the func-tional classes, following additional trainingwith new stimuli in the MTS procedure,would yield emergent equivalence relations.Two of 3 subjects passed these tests, indicat-ing that for these subjects, members of thefunctional classes were also related by equiv-alence (Sidman, 1994). These findings sup-port Vaughan’s claim that equivalence rela-tions can, in fact, emerge through simplediscrimination reversal training. A consider-ation of these procedural transfer studies,which show equivalence classes formed inMTS procedures transferring to functionalclasses in simple discriminations and vice ver-sa, has proven problematic for the traditionalview of equivalence relations (Sidman, 1994).

In the last decade, Sidman (1994, 2000)has proposed an expanded approach to de-scribing and predicting emergent behavior inthe context of equivalence classes. According-ly, potential equivalence class members in-clude responses and reinforcers, in additionto stimuli. This expanded viewpoint allowsthe demonstration of equivalence in proce-dures other than MTS, including simple dis-crimination reversal procedures. Although

the traditional definition has constrained ournotions of equivalence, this broader conceptof equivalence provides a flexible and moreuseful model of classification. It is likely thatSidman’s revised concept of equivalence en-compasses a great deal of theoretical and em-pirical work on behavioral learning and con-ditioning that, like Vaughan’s demonstrationof functional classification, has historicallybeen excluded from the equivalence litera-ture.

Vaughan’s (1988) results, showing that pi-geons formed functional classes from a largegroup of stimuli, have not been reliably rep-licated with pigeons or any other nonhumanspecies (but see Delius, Jitsumori, & Siemann,2000; Dube, Callahan, & McIlvane, 1993; To-monaga, 1999); however, von Ferson and De-lius (2000) recently reported that 2 bottle-nose dolphins trained on an auditorydiscrimination reversal task successfully trans-ferred responses established for one pair ofstimuli to a second pair of stimuli. Our ob-jective in the present experiment was to de-termine if 2 California sea lions were capableof differentiating functional classes from alarge set of stimuli. Then, following the mod-el established by Sidman et al. (1989), wesought to demonstrate the transfer of thefunctional classes to an MTS procedure andto evaluate whether class membership wouldbe extended through traditionally definedequivalence relations. The studies describedherein support an expanded view of equiva-lence by showing a nonhuman species to becapable of (a) forming functional classes in asimple discrimination reversal context, (b)transferring the relations between class mem-bers to an MTS procedure, (c) expanding theclasses through stimulus-mediated equiva-lence relations, and (d) expanding the classesthrough reinforcer-mediated equivalence re-lations.

GENERAL METHOD

Subjects

The subjects were 2 female California sealions (Zalophus californianus) named Rio andRocky. Rio was 8 years old at the start of thestudy and had previously participated in sev-eral similar experiments, including general-ized identity MTS (Kastak & Schusterman,


Fig. 1. The top photograph shows sea lion Rio per-forming a simple discrimination trial. The trial beganwhen the sea lion positioned her head at the stationingbar located in front of the center stimulus box. Followingthis stationing response, the sliding doors covering theside boxes were opened to reveal Comparison Stimuli Aand 3. The sea lion observed the stimuli from her posi-tion at the stationing bar until she was signaled by anacoustic cue to make a response. She responded by mov-ing from the stationing bar to touch Stimulus A with hernose. Her correct response was marked by an acoustictone which signaled that a fish reward would be provid-ed. The bottom photograph shows an example of a con-ditional discrimination trial. The trial was similar to a sim-ple discrimination, except that following the stationingresponse and prior to the presentation of the two Com-parison Stimuli E and 7, Sample Stimulus 10 was revealedin the center box. Rio’s correct selection of Stimulus 7as the match to the sample was rewarded.

1994) and a demonstration of stimulus equiv-alence using an MTS procedure (Schuster-man & Kastak, 1993), which later transferredto a simple discrimination procedure (Schus-terman & Kastak, 1998). Rocky was 17 yearsold at the start of the study and also had ex-tensive experience with MTS, including gen-eralized identity matching (Kastak & Schus-terman, 1994). Rocky had been trained andtested on stimulus equivalence class forma-tion using an MTS procedure, but failed todemonstrate emergent stimulus equivalencerelations. She had also received additional ex-perience with conditional discriminationlearning in the form of an artificial gesturallanguage that she had been trained with forover 10 years (for a review, see Schusterman& Gisiner, 1997). Rocky had no experiencewith simple discrimination learning prior tothis study.

Both animals were housed outdoors infree-flow seawater tanks and adjacent haulout areas at Long Marine Laboratory at theUniversity of California Santa Cruz. Each an-imal was fed between 4 and 5 kg of freshlythawed cut herring and capelin each day, onehalf of which was typically consumed duringexperimental sessions. Each animal partici-pated in experimental sessions twice each dayfor 5 days per week, generally between 9:00a.m. and 1:00 p.m. The animals were trainedusing standard operant conditioning proce-dures and fish reinforcement.

Apparatus

A two-choice visual MTS apparatus, shownin Figure 1, was used. The apparatus was athree-dimensional display constructed ofthree horizontally arranged plywood boards,each housing a window-fronted stimulus box.The boxes were 30 cm by 30 cm square and10 cm deep, and were covered by movableopaque doors. The center (sample) box waspositioned 90 cm in front of a T-bar stationat which the subject’s head rested, and thetwo side (comparison) boxes were angledsuch that each was 110 cm away from the sub-ject’s head.

The stimuli used in the experiment wereplanometric plywood squares (30 cm by 30cm) consisting of black patterns painted onwhite backgrounds. A set of 20 stimuli weredesigned and divided into two subsets of 10that were coded as ‘‘letters’’ and ‘‘numbers,’’

as depicted in Figure 2. Each pattern was con-figured to be roughly equal in area andbrightness and to be discriminable from eachof the other stimuli.

During experimental sessions, two assis-tants were seated behind the apparatus,where they were out of view of the subject.On each trial, the assistants were instructedvia headphones to place the required stimuliinto the appropriate boxes. Stimuli were al-ways placed into comparison boxes simulta-neously, so that the subject could not be cuedto the correct choice by the timing of itsplacement. Instructions were provided to the


Fig. 2. Stimulus configurations used in Experiments 1, 2, and 3. Stimuli were coded as A through J (top row)and 1 through 10 (bottom row). Stimuli K and 11 (far right) were introduced in Experiment 3. The MTS trainingsets used in Experiments 2 and 3 are shown separately for Rio and Rocky in the lower panels.

assistants by a remote experimenter who ob-served the session in real time on video. Atthe start of each session, the subject enteredthe enclosure and was signaled by an assistantto station in front of the apparatus. Sessionsconsisted of either simple discrimination tri-als or conditional discrimination trials, as de-scribed in detail in the following sections. Ineither case, a trial began when the experi-menter signaled an assistant to open one ormore of the stimulus doors to reveal the hid-den stimuli. After an observation interval of2 to 4 s, the subject was released from thestation by an acoustic cue to select one of thetwo comparison stimuli. A response was de-fined by the touch of a comparison stimulusby the subject with her nose (see Figure 1).

Correct responses were marked by a 0.5-sacoustic tone that served as a conditioned re-inforcer. The tone was followed by a piece offish tossed to the animal from behind the ap-paratus. Incorrect responses were not rein-forced, and were marked by the vocal signal‘‘no.’’ The stimulus doors were closed simul-taneously at the end of each trial. All acousticcues were triggered by the experimenter andbroadcast from a speaker mounted near theapparatus.

Analysis

Performance on experimental (novel) andbaseline (familiar) trials was measured as thenumber of correct responses out of the totalnumber of trials completed. The subjects’


rate of acquisition of novel relations was mea-sured by calculating the numbers of errorsthey made on the experimental trials prior toreaching the designated performance crite-rion. Performance in experimental condi-tions was evaluated relative to performancepredicted by chance (50% correct respond-ing) with two-tailed binomial tests. Perfor-mance between subjects or conditions wasevaluated with two-tailed Fisher’s exact tests.Changes in performance within each condi-tion were evaluated by linear regression anal-ysis, with a positive slope indicating improve-ment with time or trial number and a slopenot different from zero indicting stable per-formance. The statistical significance of thesetests was evaluated at alpha levels of .05 or.01.

EXPERIMENT 1

To evaluate the sea lions’ capability to formfunctional classes, we used a simple discrimi-nation reversal procedure in which all of themembers of a potential class shared in com-mon only a similar pattern of reinforcement.Like Vaughan (1988), we repeatedly shiftedthe reinforcement contingency from one po-tential class to the other to determine wheth-er encountering the reversed contingencywith a few class members would result in thesubjects altering their responses to the re-maining members of each class.

Procedure

The approach used in this experiment wasbased on the successive reversal proceduresused by both Vaughan (1988) with pigeonsand Dube, Callahan, and McIlvane (1993)with rats, but employed a design involving si-multaneous two-choice simple discrimina-tions rather than sequential discriminations,similar to the procedure used by Sidman etal. (1989) with humans. This procedure in-volves presenting the subject with two differ-ent visual stimuli and then signaling the sub-ject to select one of the stimuli. An exampleof such a simple discrimination trial is shownin the top panel of Figure 1. This approachwas chosen because both subjects had exten-sive prior experience with two-choice visualdiscriminations presented in an MTS proce-dure.

Each session consisted of 40 trials that in-

cluded four consecutive blocks of 10 individ-ual trials. Each trial included the presentationof one stimulus from each stimulus set (onenumber and one letter). Each block of 10 tri-als contained a single presentation of each ofthe stimuli from the two classes (i.e., Athrough J and 1 through 10). Each stimulusappeared once in each block for a total offour times per session, balanced for left andright presentation. On a given session, stimulibelonging to either the letter set or the num-ber set were designated as positive, and allresponses to members of the positive set (S1)were reinforced as correct choices. The mem-bers of the remaining set were designated asnegative (S2) and served as alternate stimulion each trial. The probability of left or rightplacement of the S1 on each trial was .5, theprobability of the S1 appearing on the sameside or alternate side as the previous trial was.5, and each session contained a unique se-quence of trials.

The general procedure consisted of a seriesof sessions in which responses to members ofthe class designated as positive were rein-forced until the subject’s performance met apreset criterion of 90% correct responses oneither one or two consecutive sessions (seebelow). Following acquisition of the positiveset, the reinforcement contingencies were re-versed so that previously positive stimuli weremade negative and previously negative stimuliwere made positive. Responses to members ofthe previously negative set were then rein-forced until the animal’s performance onceagain reached criterion, at which point thecontingencies were reversed again. This se-ries of reversals between the letter set and thenumber set continued throughout the exper-iment.

Theoretical performance following a rever-sal of reinforcement contingencies is sum-marized in the top panel of Figure 3. Priorto the first trial of a reversal, a model subjectis consistently rewarded for selecting stimulibelonging to the positive class. When the pos-itive class unexpectedly becomes negative,the subject’s performance should fall to 0%on the first trial following the reversal. Fromthis point, two scenarios describe the poten-tial performance. If each stimulus pairing istreated as an independent problem, perfor-mance levels should remain at 0% on Trials2 through 10, because the most recent feed-


Fig. 3. The top panel shows theoretical performance following a reversal with and without stimulus classification.The lower panels show the actual performance of each subject on Phases 1 through 6 of Experiment 1. Data for thelast 10 reversals of each phase are shown to represent stabilized performance levels under each condition. The trialsin Positions 1 through 10 following a reversal include a single presentation of each S1 and S2.

back the subject has about each stimulus isnow incorrect. However, if functional rela-tions have emerged between members ofeach stimulus set through reversal training,then feedback about one member of the set

should provide information about othermembers of the set. In this case, performancefollowing the first trial of a reversal shouldrapidly rise to near-perfect levels.

Experiment 1 tested these predictions by


Table 1

Each testing phase in Experiment 1 comprised a different experimental condition underwhich reversals were completed. The number of trial types was either 10 (stimuli presentedin matched pairs) or 100 (stimuli presented in shuffled pairs). The reinforcer type was eitheruncorrelated (randomized with respect to the two different stimulus sets) or correlated (spe-cific to each stimulus set). Reversals of reinforcement contingencies occurred either at thestart of session or unpredictably within a session.

Condition Trial typesReinforcer

typeReversalposition

Reversalscompleted

Rio Rocky

Initial training 10 matched pairs Uncorrelated

Phase 1:Discriminations in matched pairs 10 matched pairs Uncorrelated Start of session 20 15

Phase 2:Discriminations in shuffled pairs 100 shuffled pairs Uncorrelated Start of session 22 20

Phase 3:Assignment of class-specific reinforcers 100 shuffled pairs Correlated Start of session 22 41

Phase 4:Removal of class-specific reinforcers 100 shuffled pairs Uncorrelated Start of session 53 24

Phase 5:Reintroduction of class-specific reinforcers 100 shuffled pairs Correlated Start of session 10 10

Phase 6:Begin within-session reversals 100 shuffled pairs Correlated Within session 40 40

dividing reversals into one training and sixtesting phases. The training phase was re-quired to introduce each subject to the pro-cedure and stimuli and to establish reliableresponses to all the members of one stimulusset. Each of the six testing phases compriseda different experimental condition underwhich reversal sessions were performed. Theexperimental conditions were established foreach phase during the course of the experi-ment in a continuing effort to elicit perfor-mance consistent with functional class for-mation. The number of reversals completedby each subject in each testing phase was notstandardized. Rather, we continued to reverseand retrain the positive sets in each phase un-til performance on the first block of trials fol-lowing reversals reached an asymptotic level.When performance on at least 10 consecutivereversals showed no further improvement,the subject progressed to the subsequent test-ing phase. In all phases, regression analysisconfirmed stable performance on the last 10reversals in each phase of the experiment.Rio completed a minimum of 10 and a max-imum of 53 reversals in each test phase;Rocky completed a minimum of 10 and a

maximum of 41 reversals. Testing under eachcondition is summarized in Table 1 and wascarried out as follows:

Initial training. All 20 stimuli were intro-duced to each subject in the first session ofthe experiment. In this phase, letters andnumbers were presented on each trial inmatched pairs (i.e., A and 1, B and 2, . . ., Jand 10). This design generated a total of 10trial types that appeared four times each persession, once in each block of trials. Respons-es to stimuli belonging to the positive set(designated as letters for Rio and numbersfor Rocky) produced a 440-Hz tone thatserved as a conditioned reinforcer followedby a piece of randomly selected capelin orherring. Training proceeded until the sub-jects’ performance met a criterion of two con-secutive sessions containing at least 90% cor-rect responses.

Reversal Phase 1: Stimuli presented in matchedpairs. Following initial training, the reinforce-ment contingencies were reversed: Responsesto stimuli belonging to the previously positiveset no longer produced food, and responsesto members of the previously negative setwere now reinforced. Once performance


reached the criterion of 90% correct respons-es on two consecutive sessions, the reinforce-ment contingencies were again reversed andretrained to criterion. All correct responsesproduced the 440-Hz tone followed by eithercapelin or herring. Throughout Phase 1, thestimuli continued to be presented in matchedpairs following the same design used in theinitial training phase.

Reversal Phase 2: Stimuli presented in shuffledpairs. Following Phase 1, during which nosubstantial improvement in reversal perfor-mance occurred, several additional phaseswere incorporated into the experimental de-sign (see Table 1). In Phase 2, the stimuluspairings were shuffled so that any S1 couldbe presented with any S2 on an individualtrial (e.g., F and 2, 7 and B, J and 8). Thischange in procedure increased the numberof different trial combinations from a total of10 in Phase 1 to a total of 100 different trialcombinations in Phase 2. The presentation ofstimuli within each session remained bal-anced as outlined in the general procedure,with each S1 and S2 appearing once in eachblock for a total of four times per session. Asin Phase 1, all correct responses produced a440-Hz tone followed by either capelin orherring.

Reversal Phase 3: Assignment of class-specificfish reinforcers. Following Phase 2, reversalscontinued as before with one exception: Eachof the two different fish reinforcers that werepreviously uncorrelated with different stimu-lus sets were now assigned to one of the twostimulus sets. In this condition, correct re-sponses to stimuli from a given set produceda specific acoustic conditioned reinforcer (ei-ther a higher pitched tone or a lower pitchedtone) followed by a specific fish reward (ei-ther capelin or herring). For Rio, when mem-bers of the letter set served as S1, correctresponses produced a 587-Hz tone followedby capelin; when members of the number setserved as S1, correct responses produced a293-Hz tone followed by herring. For Rocky,the opposite outcomes were correlated witheach stimulus set.

Reversals continued throughout Phase 3,with the original criterion of two consecutivesessions with performance at or above 90%correct until each subject’s performancereached 90% on the first session following areversal. At this point (the 9th reversal of the

phase for Rio; the 15th reversal for Rocky),the defined criterion was reduced to one ses-sion with performance at or above 90%.

Reversal Phase 4: Removal of class-specific fishreinforcers. In this phase, the specific reinforc-ers that had been assigned to each class weredesegregated. As in Phase 2, correct respons-es to members of either class produced theoriginal 440-Hz tone and a mixed (either cap-elin or herring) fish reward.

Reversal Phase 5: Reintroduction of class-specificfish reinforcers. Upon completion of Phase 4,the class-specific reinforcers used in Phase 3were reinstated.

Reversal Phase 6: Within-session reversals. Inthe final phase of Experiment 1, the positionof the contingency reversal in the session wasmanipulated. Prior to this phase, reversals ofthe positive and negative stimulus sets alwaysoccurred at the start of a session. DuringPhase 6, a reversal could occur one or moretimes within a session, with the only restric-tion being at least 90% correct on the previ-ous 10 to 14 consecutive trials.

Results and Discussion

Both subjects began the training phasewith chance levels of performance. Rioreached criterion following three sessions(120 trials with 36 errors); Rocky reached cri-terion following six sessions (240 trials with75 errors). Following this training phase, thesubjects began the reversal phases of the ex-periment.

Functional classes are demonstrated whenexperience with a few members of one stim-ulus set alters responding to the remainingmembers of that set. Therefore, performanceon the first exposure of each S1 following areversal must be isolated from performanceon subsequent trials to prevent the effects oflearning from influencing assessment of func-tional class formation. Because the first 10 tri-als of each reversal included a single presen-tation of each S1 and S2, any improvementthat occurred between the trials in Positions1 through 10 can be attributed to the for-mation of relations between class members.Consequently, only the first 10 trials followingeach reversal (Block 1) were considered foranalysis in the reversal phases of the experi-ment.

The performance trends of both subjectson the first block of trials following a reversal


Table 2

Linear regressions were completed to evaluate each subject’s performance across a series ofreversals. Performance was measured as the number of correct responses made on the 10unique trials following a reversal, and that metric was evaluated across reversals. In this way,changes in performance were measured across all reversals in each testing phase, and on thelast 10 reversals within each testing phase. Stable performance evaluated at p . .05 is denotedby ns, significant improvement evaluated at p , .05 is denoted by *, and significant improve-ment evaluated at p , .01 is denoted by **. Note that in all phases, performance across thelast 10 reversals was stable.

Subject Phase

Performance across all reversals

Samplesize R2 Significance

Performance across last 10 reversals

Samplesize R2 Significance

Rio 123456

202222421040

.175

.000

.654

.006

.036

.119

nsns**nsns*

101010101010

.001

.042

.141

.288

.036

.063

nsnsnsnsnsns

Rocky 123456

152044241040

.091

.023

.358

.008

.027

.261

nsns**nsns**

101010101010

.000

.031

.054

.263

.040

.182

nsnsnsnsnsns

Fig. 4. Performance of each subject in Phases 1through 6 of Experiment 1. Each bar represents the pro-portion of correct responses made to the 10 stimuli ontheir first presentation following a reversal. The idealizedperformance maximum is 90% correct responses (per-fect performance following one error after a reversal; seeFigure 3, top panel). Data for the last 10 reversals of eachphase are shown to represent stabilized performance lev-els under each condition.

were similar for all six phases of the experi-ment. Results from regression analysis of per-formance following reversals in each phaseare shown in Table 2. For this analysis, trialsin the first block following a reversal weregrouped for each reversal. Performance wasmeasured as the total number of correct re-sponses in each block of 10 trials. When per-formances across all the reversals in a phasewere considered, both Rio and Rocky showed

stable performance (slope of regression linenot different than zero) in Phases 2, 4, and5. Both subjects exhibited substantial im-provement (slope of regression line positive)across the reversals in Phases 3 and 6.

All further comparisons of performancebetween each subject and between eachphase of the experiment were restricted tothe last 10 reversals completed in each phase.The purpose of this analysis was twofold. Per-formance on the last 10 reversals in eachphase for both animals was asymptotic, asshown by the regression analysis in Table 2.Thus, this region represents the best perfor-mance achieved by each subject under eachcondition. In addition, this measure provideda standardized way of comparing the perfor-mance of both animals, because the numberof reversals completed by each subject ineach testing phase was not the same.

The pattern of performance on trials with-in the last 10 reversals of each testing phaseis shown in Figure 4, which also documentsthe strong similarities observed between thesubjects. The performance of the 2 subjectscompared within each phase was not differ-ent for any of the six phases (Fisher’s exacttests, p . .05). Performance was lowest inPhase 1, in which the stimuli were presentedin matched pairs. In Phase 2, in which the


stimuli were shuffled, both animals showed aslight but nonsignificant improvement in per-formance (Rio improved from 21 of 95 cor-rect responses to 33 of 100, Rocky improvedfrom 24 of 95 to 29 of 100; Fisher’s exacttests, p . .05). They also showed a reductionin the average number of errors required toreach criterion following a reversal betweenPhases 1 and 2 (Rio improved from an aver-age of 126 errors per reversal on Phase 1 to55 errors on Phase 2, Rocky improved from137 to 117 errors). However, the most dra-matic change in performance occurred forboth subjects in Phase 3, when different re-inforcers were assigned to each stimulus set.Performance for both subjects under thiscondition showed dramatic improvement rel-ative to Phase 2 (Rio improved from 33 of100 to 83 of 100, Rocky improved from 29 of100 to 82 of 100; Fisher’s exact tests p , .01),and performance was also much better thanwould be expected by chance (Rio scored 83of 100, Rocky scored 82 of 100; binomialtests, p , .01). By the end of this phase, bothsubjects were responding appropriately afteronly a few trials following a reversal. Becauseboth subjects generally met criterion withinthe minimum number of sessions required,almost every session was a criterion session;thus, Rio made an average of 0 errors priorto criterion, and Rocky averaged 0.5 errors tocriterion in Phase 3.

In Phase 4, class-specific reinforcers wereremoved from the testing procedure. Thisphase replicated the testing conditions ofPhase 2. Under this condition, the perfor-mance of both subjects declined significantlyfrom Phase 3 levels (Rio’s performancedropped from 83 of 100 to 56 of 100 andRocky’s dropped from 82 of 100 to 43 of 100;Fisher’s exact tests, p , .01), but remainedelevated relative to performance in Phase 2(Rio scored 56 of 100 on Phase 4 and 33 of100 on Phase 2; Rocky scored 43 of 100 and29 of 100; Fisher’s exact tests, p # .05). InPhase 5, when the reinforcers were reas-signed to the two classes, performance recov-ered to Phase 3 levels (Rio scored 83 of 100on Phase 5 and 83 of 100 on Phase 3; Rockyscored 73 of 100 and 82 of 100; Fisher’s exacttests, p . .05).

The final measure used to assess functionalclass formation was an analysis of perfor-mance following a reversal by trial position.

This analysis can be compared to the theo-retical reversal performance predicted withand without stimulus classification as de-scribed in the top panel of Figure 3. The ac-tual data plotted by trial position for each an-imal for each phase of the experiment aredepicted in the lower panels of Figure 3. Inall phases, performance was better on trialsthat occurred in the latter half of the testblock, suggesting some degree of functionalclassification by the sea lions. Phases 3 and 5,in which test conditions were the same (shuf-fled stimulus pairings and class-specific rein-forcers), showed close to perfect perfor-mance. The only anomaly was performanceon the very first trial following a reversal. Asdepicted in the top panel of Figure 3, ex-pected performance following a reversal ofreinforcement contingencies is zero. Howev-er, the performance of both subjects on Trial1 improved from zero to near chance levelsduring Phases 1 through 5. Because reversalsoccurred frequently, and only at the start ofa session during these phases, it is likely thatthe subjects eventually began responding atrandom on the first trial of a session. To de-termine if this was the case, Phase 6 movedthe reversal contingency from the start of thesession to an unpredictable position withinthe session. In this condition, performancewas predictably zero on the first trial follow-ing a reversal, and test performance on sub-sequent trials mirrored that expected fromsuccessful functional class formation (com-pare the top and bottom panels of Figure 3).

These results document the formation offunctional classes in two California sea lionsand provide strong support for Vaughan’s(1988) finding with pigeons. The finding thatsea lions, as well as pigeons, can form classesof functionally equivalent stimuli indicatesthat this classification process may be a fun-damental learning ability. Further support forthis view comes from less definitive studiesconducted with other nonhuman subjects.Dube, Callahan, and McIlvane (1993) report-ed that rats showed some savings on sequen-tial auditory discrimination reversals usingdifferent reinforcers for correct responses todifferent classes. However, improvement didnot occur within the first exposure of eachstimulus following a reversal of reinforcementcontingencies, and only 2 of 5 subjectsshowed savings at all. Tomonaga (1999) re-


cently reported some evidence of functionalclass formation in a chimpanzee with a two-item sequential responding procedure, with-out using different reinforcers. Because thisstudy included only two two-member classesand failed to show an immediate transfer offunction following a reversal, the findings areof limited value regarding this issue. In con-trast, the strong reversal transfer perfor-mance demonstrated by 2 dolphins (von Fer-son & Delius, 2000), also using a small set ofstimuli, does provide support for the ideathat functional classification is a general pro-cess.

The measures used to assess functionalclass formation in the present study are pow-erful because they rely on Trial 1 perfor-mance, or the performance of the subjects ontheir first encounter with each stimulus fol-lowing a reversal. Despite a sample size ofonly 10 stimuli per set, both subjects eventu-ally showed that reversed reinforcement con-tingencies for a few members of a set wouldresult in reversed responses to the remainingmembers of that set on their first exposure.Although the members of each set had neverbeen directly associated and were not physi-cally similar, they became related to one an-other through a shared history of reinforce-ment and formed distinct functional classes.

The first variable manipulated in this studywas the number of exemplars (stimulus con-figurations) provided to each subject. We pre-dicted that increasing the number of exem-plars from 10 in Phase 1 to 100 in Phase 2would facilitate classification, based on otherstudies using the same subjects (Kastak &Schusterman, 1994; Schusterman & Kastak,1993). We observed, however, only a slightimprovement in reversal performance follow-ing shuffling of the stimulus pairings. Ratherthan facilitating performance immediatelyfollowing the reversal, increasing the numberof exemplars appears to have aided reversalacquisition, as shown by the reduction in theaverage number of errors made prior to cri-terion. This trend likely involved improve-ment in the memorization of specific re-sponses and in the identification of individualstimuli rather than in the strengthening ofrelations between stimuli. Thus, increasingthe number of exemplars was not directly as-sociated with stimulus classification.

The major variable assessed in this study

was the assignment of different primary andconditioned reinforcers to each stimulus set.Following the assignment of class-specific re-inforcers to each stimulus set, both subjectsrapidly improved to near-perfect perfor-mance levels within the first block of trialsfollowing a reversal. This effect culminated inthe reversal of responses to all members of astimulus set following only one or two infor-mation trials with individual stimuli.

The rationale for correlating reinforcerswith potential stimulus classes was based onseveral studies linking differential outcomesor rewards to increased performance on var-ious discrimination tasks (see Goeters, Blake-ly, & Poling, 1992, for a review). In addition,the assignment of differential reinforcers hadbeen used to facilitate class formation in bothsimple and conditional discrimination pro-cedures, ostensibly by providing an additionalsource of information about each set (see,e.g., Schenk, 1994). Most recently, Meehan(1999) used class-consistent differential rein-forcement with pigeons to demonstrate emer-gent stimulus relations, including transitiveand reflexive relations.

The assignment of class-specific reinforcersapparently catalyzed functional class forma-tion for the sea lions. Why Vaughan’s (1988)pigeons were able to demonstrate functionalclass formation without assigned reinforcersis not clear. Variability in performance be-tween our study and Vaughan’s may be relat-ed to differences in procedure (Vaughanused a sequential discrimination; we em-ployed a simultaneous procedure), responsecriteria (Vaughan used a pecking rate index;we used a discrete correct or incorrect re-sponse), or analysis (Vaughan used a likeli-hood ratio of choosing a positive stimulusover a negative stimulus; we measured thenumber of correct responses of the first ex-posure of each stimulus following a reversal).

The effect of assigning differential out-comes was not entirely predictable based onprevious work. One of our subjects, Rio, hadpreviously formed stimulus equivalence clas-ses in an MTS procedure in the absence ofspecific outcomes for responses to membersof specific classes (Schusterman & Kastak,1993). Subsequently, Rio was tested with asimple discrimination procedure similar tothe one used in the current study to deter-mine if equivalence classes formed in an MTS


procedure would transfer to functional clas-ses in a simple discrimination procedure. Rioshowed immediate transfer, indicating thatfunctional classes could emerge from equiv-alence relations without the assignment ofclass-specific reinforcers (Schusterman & Kas-tak, 1998). Whether different reinforcerswould have facilitated the formation of equiv-alence classes in the Schusterman and Kastakstudies, as reported by Schenk (1994) forchildren, is unknown. However, the resultsfrom this experiment suggest that differentialoutcomes may play a more important rolewhen stimuli are related through a commonbehavioral response, as in a simple discrimi-nation reversal procedure, rather thanthrough a common stimulus, as in an MTSprocedure.

The powerful effect of assigning reinforc-ers to stimulus classes led to the question ofwhether our subjects could retain functionalclass memberships in their absence. It waspossible that the reinforcer following eachcorrect response served as a direct discrimi-native cue for performance on the followingtrial. Essentially, a specific reinforcer follow-ing a correct response to one S1 could oc-casion responding to the S1 presented onthe following trial, without the two stimuli be-coming interrelated. Two pieces of evidencesuggest that this was not the case. Followingthe reversal of reinforcement contingencies,the responses of both subjects eventuallyshifted to members of the positive class afterresponding to only one or two members ofthe negative class, in the absence of any cuesprovided by class-specific fish or conditionedreinforcers. In addition, when reinforcer as-signments were randomized in Phase 4, per-formance declined but remained significantlyelevated with respect to the preassignmentlevels measured in Phase 2.

Despite this evidence, we could not ruleout the possibility that the emergent perfor-mance we observed was a result of condition-al control by the class-specific reinforcers. Re-moval of the class-specific reinforcers inPhase 4 was problematic. Ideally, the two re-inforcers assigned to each class should havebeen replaced with a third, neutral reinforc-er. Unfortunately, this was not possible be-cause the subjects were maintained on a dietof two fish types and could not be coaxed toaccept a third within the time constraints of

the experiment. The results for both subjectswhen food reinforcers were varied in Phase 4,however, suggest that the role of the assignedreinforcers in Phase 3 was to strengthen therelations between individual stimuli by relat-ing members of a common class to a com-mon reinforcer. When the reinforcers wereuncorrelated in Phase 4, the relations thathad formed between stimuli were sufficientto sustain elevated levels of performance rel-ative to the prior uncorrelated reinforcercondition. The decline in performance in theuncorrelated condition (Phase 4) relative tothe previous correlated condition (Phase 3)can be attributed to the confusing effects ofmixing reinforcers that had previously beenassigned to each class. When the reinforcerassignments were reestablished in Phase 5,performance immediately recovered to Phase3 levels.

EXPERIMENT 2

From an experimental standpoint, measur-ing the transferability of a stimulus class fromone procedure to another can reveal thestrength of the relations that exist betweenclass members. Based on the results of Ex-periment 1, we believed that functional rela-tions had formed between the members ofeach stimulus class. We next tested the sealions to determine if the functional classes es-tablished in a simple discrimination reversalprocedure would yield conditional discrimi-nations in an MTS procedure; given a samplestimulus from one class and comparison stim-uli from each of two classes, would the sub-jects immediately relate the two stimuli be-longing to the same functional class? Tosucceed in this experiment, the sea lionswould have to transfer functional class mem-bership to an MTS procedure, one that elim-inated the potential for discriminative controlby class-specific reinforcers.

Procedure

An MTS procedure was used for all exper-imental sessions according to the general de-sign already described. The same apparatuswas used; however, in this experiment, thecenter (or sample) stimulus box was used inaddition to the two comparison boxes. Oneach trial, the assistants were directed viaheadphones to simultaneously place the ap-


propriate stimuli into each comparison boxand then place the sample stimulus into thecenter box. To begin a trial, the door cover-ing the sample box was opened to expose thesample stimulus. The subject was given a 3-to 4-s observing interval, and then, on theexperimenter’s cue, the two comparison stim-uli were exposed. After another 3- to 4-spause during which all three stimuli were vis-ible to the subject, the subject was promptedby the acoustic release to select one of thecomparison stimuli as the match to the sam-ple. An example of such a conditional dis-crimination trial is shown in the lower panelof Figure 1.

The experiment was divided into a seriesof baseline training and transfer tests as fol-lows:

Maintenance of MTS baseline relations. Priorto testing the transfer of functional classes toconditional discriminations, a baseline of fa-miliar conditional discriminations was estab-lished for each subject. These stimuli, codedas MTS training sets, were plywood squares (30cm by 30 cm) with a variety of black shapespainted onto white backgrounds. These stim-uli, shown in the lower panels of Figure 2,were a subset of stimuli used by each animalin a previous experiment (see Schusterman& Kastak, 1993). Rio and Rocky were eachassigned 12 training stimuli divided into fourthree-member sets labeled MTS Training Sets1, 2, 3, and 4. The three stimuli in each setwere labeled A, B, and C members as shownin Figure 2; A1, B1, and C1 were all membersof Set 1, B1, B2, and C2 were all members ofSet 2, and so on. The stimuli within eachtraining set could be combined to generatesix different conditional discriminationsbased on the stimulus combinations ArB,BrC, ArC, BrA, CrB, and CrA, where the let-ter code represents the sample r S1 config-uration. The S2 presented on each trial wasthe corresponding member of one of the oth-er sets. An important characteristic of thesetrained relations was that all of the stimuli ina set appeared as samples, positive compari-sons, and negative comparisons on differenttrials, a feature that would later be incorpo-rated into the transfer tests.

When the subjects originally learned thesebaseline conditional discriminations, rein-forcement was not class specific. However,several months prior to the start of the Ex-

periment 2 transfer tests, each training setwas assigned one of the two class-specific re-inforcers used in Experiment 1. For Rio, cor-rect responses to A1, B1, C1 or A3, B3, C3produced the 587-Hz tone followed by cape-lin, and correct responses to A2, B2, C2 orA4, B4, C4 produced the 293-Hz tone fol-lowed by herring. For Rocky, the oppositeoutcomes were correlated with each MTStraining set. Both subjects were trained onconditional discriminations with these stimuliusing class-specific reinforcers until a perfor-mance criterion of 90% correct was main-tained.

Maintenance of reversal performance. Perfor-mance on simple discrimination reversalswith the letter class and the number class wasmaintained throughout Experiment 2. Thesesessions were continued in accordance withthe design used in Phase 6 of Experiment 1.A minimum of one simple discrimination re-versal session and one MTS baseline sessionwere completed at criterion prior to each ofthe transfer tests described below. The pur-poses of these sessions were threefold: (a) toensure that the subjects were properly trainedto perform the MTS procedure, (b) to estab-lish baseline performance levels to which testperformance could be compared, and (c) toensure that the integrity of the functionalclasses established in Experiment 1 was main-tained.

Conditional discrimination transfer tests. Ex-periment 2 was designed to measure thetransfer of the functional classes establishedin a simple discrimination procedure to anMTS procedure. Each of six transfer testscomprised novel pairings of functional classmembers in the MTS procedure. There were4 novel trials presented in Tests 1, 2, and 3,48 novel trials in Test 4, 24 novel trials in Test5, and 96 novel trials in Test 6, for a total of180 novel conditional discrimination trials.

The transfer tests were designed so that asubset of novel pairings would be tested,trained to criterion, and then incorporated inthe baseline of familiar conditional discrimi-nations before the subsequent test was con-ducted. This design provided an opportunityfor the subjects to become accustomed to thetesting procedure and to diminish novelty ef-fects that might disrupt test performance. Inaddition, this design provided a series of pro-cedural exemplars that could facilitate per-


formance on later transfer tests involving nov-el combinations of stimuli. The transfer-testprocedures are detailed below; an exhaustivelist of the trial configurations used in eachtest appears in Appendix A.

The test trials within each test session werepresented randomly within a baseline of fa-miliar conditional discrimination trials. Cor-rect responses on test trials were defined asselection of a letter conditionally upon thepresentation of another letter as the samplestimulus, and selection of a number condi-tionally upon the presentation of anothernumber as the sample stimulus. All correctresponses produced the same class-specific re-inforcers that had been paired with the letterand number stimulus sets in Experiment 1.

Transfer Test 1. Two members from eachfunctional class were randomly selected to betested in Transfer Test 1: E and I and 4 and8. These stimuli generated a total of fourunique conditional discrimination test trials(i.e., ErI, IrE, 4r8, and 8r4). The S2 on thesetrials was one of the stimuli being tested fromthe alternate set. Each of the four test trialsappeared three times in each session against28 familiar baseline MTS trials. Following twosessions, the number of test trials appearingin each session was increased from 12 to 24and the number of baseline trials was de-creased from 28 to 16. These sessions werecontinued until each subject reached a per-formance criterion of 90% on the test trialswithin one session; the test trials were thenincorporated into each subject’s baseline ofconditional discriminations.

Transfer Test 2. Transfer Test 2 replicatedTest 1 with two new stimuli drawn from eachfunctional class: B and H and 2 and 10. Fol-lowing testing and training to criterion, thefour test trials were incorporated into eachsubject’s baseline of conditional discrimina-tions.

Transfer Test 3. Transfer Test 3 replicatedtests 1 and 2 with two new stimuli drawn fromeach functional class: A and J and 7 and 9.Following testing and training to criterion,the four test trials were incorporated intoeach subject’s baseline of conditional discrim-inations.

Transfer Test 4. Transfer Test 4 utilized allcombinations of the stimuli tested in TransferTests 1, 2, and 3, yielding a total of 48 un-tested conditional discriminations. Testing

took place over two sessions that each con-tained 24 test trials and 16 baseline trials. Fol-lowing these two sessions, which included thefirst presentation of each novel stimulus com-bination, two additional sessions were con-ducted in which test trials appeared for thesecond time with a different S2 and in a dif-ferent configuration. After these four sessionswere completed, the 48 test trials were incor-porated into each subject’s baseline of con-ditional discriminations.

Transfer Test 5. The remaining four stimulifrom each functional class (C, D, F, G; 1, 3,5, 6) were used in Transfer Test 5. This pro-cedure generated 24 novel conditional dis-criminations. The test trials were presented inone test session with 16 baseline trials. A sec-ond session was conducted in which test trialsappeared for the second time with a differentS2 and in a different configuration. Follow-ing this session, the test trials were incorpo-rated into each subject’s baseline of condi-tional discriminations.

Transfer Test 6. The final transfer test in-cluded all new combinations of the eightstimuli used in Transfer Test 5 with the 12stimuli used in Tests 1 through 4. This pro-cedure generated 96 novel class-consistentstimulus pairings. This test was fashioned af-ter Transfer Tests 4 and 5, with testing occur-ring over four sessions with 24 test trials and16 baseline trials each. Following these foursessions, four additional sessions were con-ducted in which test trials appeared for thesecond time.


The results of these transfer tests are sum-marized in Table 3. The primary measure oftransfer performance was the subjects’ per-formance on the first exposure of each novelconditional discrimination. The use of thisTrial 1 measure precluded the possibility thatpositive results could arise from trial-and-er-ror learning of individual conditional rela-tions. Performance on the second exposureof each test trial and performance on thebaseline trials that were presented duringtesting are also reported. Finally, a measureof any trial-and-error learning that did occuris shown by the number of errors made onnew trials prior to reaching the 90% perfor-mance criterion.

Transfer Tests 1, 2, and 3 contained four


Table 3

Performance on conditional discrimination transfer tests summarized as number of correcttrials out of total number of trials. Categories in which performance is not significant relativeto chance at an alpha level of p . .05 (calculated from a two-tailed binomial) are denoted byns; categories in which performance is significant relative to chance at an alpha level of p ,.05 are denoted by *; categories in which performance is significant relative to chance at analpha level of p , .01 are denoted by **. Number of errors to criterion includes the totalnumber of errors made on test trials prior to the first session of better than 90% performance.If the subject achieved criterional performance on test trials within the first session, the num-ber of errors to criterion was zero.

SubjectTransfer

testBaseline

performance

Test performance

First exposure Second exposureErrors tocriterion

Rio 123456

27/28**27/28**24/28**28/32**15/16**59/64**

2/43/43/4

40/48**16/24 ns89/96**

2/43/42/4

47/48**21/24**96/96**

90208080

Total 180/196** 154/180** 171/180** 126Rocky 1

23456

24/28**25/28**27/28**32/32**16/16**61/64**

2/40/43/4

48/48**18/24*92/96**

4/43/43/4

48/48**21/24**90/96**

662727090

Total 185/196** 163/180** 169/180** 129

novel test trials each. The sample size foreach test was too small to evaluate Trial 1transfer of functional classes to conditionaldiscriminations. When Trial 1 performancewas pooled over the first three tests, neithersubject demonstrated significant transfer (Rioscored 66% and Rocky scored 41%; binomialtests, p . .05). Both subjects, however,showed a reduction in the number of trialsrequired to reach criterion on each successivetest.

Performance on the 48 novel trials pre-sented in Transfer Test 4 was strong for bothsubjects. Rio scored 83.3% on the first expo-sure of the test trials, and Rocky scored 100%(binomial tests, p , .01). The performanceof both subjects on test trials was not differentfrom their performance on familiar baselinetrials (Fisher’s exact tests, p . .05). An advan-tage to the design of Transfer Test 4 was thateach of the stimuli tested had previously ap-peared as both a sample and a comparison inother trial combinations. Thus, any noveltyeffects generated by the unexpected appear-ance of a letter or number in the sample po-sition were mitigated, and performance onnovel stimulus pairings could still be assessed.

Performance on the 24 novel trials in

Transfer Test 5, which included stimuli neverbefore presented in the MTS procedure butwhich did not control for the effect of novelstimulus position, was not quite as strong ason Test 4. Rio scored 66% on test trials (bi-nomial test, p . .05), and Rocky scored 75%(binomial test, p , .05). Performance on testtrials was slightly but not significantly worsethan performance on corresponding baselinetrials for both subjects (Fisher’s exact tests, p. .05). However, on the second exposure ofeach test trial, Rio’s performance rose from66% to 91%, and Rocky’s performance rosefrom 75% to 87% (binomial tests, p , .01).

Transfer Test 6 included 96 novel trial com-binations, which, like Transfer Test 4, didcontrol for the effect of novel stimulus posi-tion. Consistent with performance on Trans-fer Test 4, performance on Transfer Test 6was significantly better than expected bychance. On this test, Rio scored 92% on thefirst exposure of the test trials, and Rockyscored 95% (binomial tests, p , .01). Theirperformance on novel test trials was not dif-ferent from performance on familiar baselinetrials (Fisher’s exact tests, p . .05).

Overall, performance on Transfer Tests 1through 6 combined shows highly significant


transfer of functional classes to conditionaldiscriminations. Rio was correct on 154 of180 novel stimulus pairings (85.6% correct;binomial test, p , .01), and Rocky was correcton 163 of 180 novel pairings (90.5% correct;binomial test, p , .01). There was no differ-ence in performance on transfer trials inwhich letters served as the discriminativestimulus relative to trials in which numbersserved as the discriminative stimulus (Fisher’sexact tests, p . .05), and overall performanceon test trials was not different from perfor-mance on familiar baseline trials (Fisher’s ex-act tests, p . .05).

The weakest transfer for each subject wasobserved during tests that involved present-ing functional class members in the MTS pro-cedure for the first time. It is likely that theunexpected change in context or stimulusposition for the functional class memberscaused some disruption in performance ontransfer trials in some or all of these tests.The results of Test 5 provide evidence to sup-port this hypothesis. In this test, each subjectwas presented with 24 novel transfer trialscomposed of stimuli that had not yet beenpresented in the MTS paradigm; Rio’s per-formance on this test was marginally worsethan predicted by chance, and Rocky’s per-formance was marginally better. However, onthe second session of Test 5, when the sametest trials were presented (this time appearingwith a different S2 and in a different positiveposition), performance of both subjects roseto near-perfect levels. It is unlikely that eithersubject learned the 24 new relations present-ed in the first session after a single exposureto each. Therefore, improved performanceon the second session can be attributed to areduction or elimination of the novelty effectthat had disrupted transfer in the first ses-sion.

The two tests that involved novel combi-nations of stimuli that had already been pre-sented in the MTS paradigm provided the op-portunity to assess transfer in the absence ofa possible disruptive novelty effect. TransferTests 4 and 6 were composed of stimuli thathad been used in previous transfer tests re-shuffled into novel combinations. The per-formance of both subjects on these test trialswas not different than performance on com-pletely familiar trials; Rio was correct on 90%of the transfer trials, and Rocky was correct

on 97% of novel transfer trials on their firstexposure.

The disruptive effects of novel stimulus po-sitions are consistent with data reported forthe same subjects on tests of identity match-ing (Kastak & Schusterman, 1994). The an-ticipated disruption of test performance wasa key factor in designing the block format ofthe transfer tests incorporated into this ex-periment. The procedure allowed any disrup-tion in performance caused by novel stimulusposition to be isolated and measured by pre-senting novel combinations of individualstimuli that either had or had not been pre-viously exposed to the subjects in the MTSprocedure. In addition, the experimental de-sign incorporated a sequential componentthat provided the subjects with experience re-lating some members of functional classes toone another in the MTS procedure; this fea-ture may have facilitated transfer on subse-quent tests.

The design of the experiment was also ad-vantageous because it allowed us to resolvethe question of whether the functional classesformed in Experiment 1 were under the dis-criminative control of the reinforcer. Had thereinforcer controlled responding in the sim-ple discrimination reversal procedure, cor-rect responding would not necessarily requireemergent stimulus relations to arise betweenclass members. In the MTS procedure, how-ever, either a letter or a number could appearas the S1 on any trial, and consequently, thesequence of reinforcers alternated irregularlythroughout the session. In this context, thereinforcer given on the preceding trial servedno predictive function for correct respondingon the following trial. With reinforcers nolonger serving as potential discriminativecues, both subjects still matched functionalclass members with a great deal of accuracy.These results support the interpretation thatthe relations between members of the func-tional classes were not based solely on stim-ulus–reinforcer relations, but at least in parton stimulus–stimulus relations that emergedbetween functional class members. Thus, webelieve that the stimulus–reinforcer relationsestablished in Experiment 1 served primarilyto strengthen or facilitate the relations be-tween stimuli, rather than to replace them.


EXPERIMENT 3

In the first two experiments, the subjectsformed functional classes in a simple discrim-ination procedure and then transferred therelations between class members to condi-tional discriminations. However, as Sidman etal. (1989) noted, demonstrating the emer-gence of conditional discriminations withinfunctional classes does not suffice to dem-onstrate equivalence relations among classmembers. In this final series of experiments,we tested the sea lions to determine if func-tional class members that shared commonstimulus and reinforcer relations would gen-erate verifiable equivalence relations. Fur-ther, we attempted to determine whetherstimuli with only reinforcer relations in com-mon would also become related throughemergent equivalence relations.

Procedure

This experiment followed the same generaltesting procedures used in the previous ex-periments and consisted of two primary com-ponents. The first component tested whetheremergent equivalence relations could formfrom functional classes. For this task, the sealions were trained to relate novel stimuli toexisting class members in MTS, and werethen tested to determine whether equiva-lence relations would emerge between thenew stimuli and the remaining class mem-bers. The second component was designed toexamine the role of the different reinforcersin class formation. For this task, the sea lionswere tested to determine if a common rein-forcer would establish emergent relations be-tween functional class members and previ-ously unrelated stimuli.

The MTS apparatus and general testingprocedure used in Experiments 1 and 2 werealso used in Experiment 3. The stimuli con-sisted of the MTS training sets used for eachsubject in Experiment 2 and the two 10-mem-ber functional classes established in Experi-ment 1. In addition, one new member wasadded to each existing functional class in thecurrent experiment: K was added to the letterclass and 11 was added to the number class.All of these stimuli are shown in Figure 2.

Expansion of functional classes through stimu-lus-mediated equivalence relations. Stimuli K and11 were related to J and 10 as follows. First,

the conditional discriminations JrK and 10r11were trained with the familiar stimulus (ei-ther J or 10) serving as the sample, and thenew stimuli (K and 11) serving as compari-sons. Each training trial appeared 12 timeseach per session with 16 familiar baseline tri-als. All correct responses produced the sameclass-specific reinforcers used previously.These training sessions were continued to aperformance criterion of 90% correct ontraining trials. Following attainment of thiscriterion, the same testing procedure wasused to train the symmetrical relations KrJand 11r10. Once relations KrJ, 10r11, JrK,and 11r10 were established, the training trialswere combined into sessions that included sixtrials with each of the four newly trained dis-criminations. The subjects were required toperform one of these sessions at criterion pri-or to proceeding to the transfer test. Thetransfer test is described below; all of the trialconfigurations used in training and transfertesting are shown in Appendix B.

Transfer testing took place over two con-secutive sessions that consisted of 18 test trialsand 12 baseline trials each. Transfer trialsconsisted of presenting the new stimuli (Kand 11) with the remaining nine members ofeach functional class (A through I and 1through 9) in novel conditional discrimina-tions. On these trials, either K or 11 couldappear as the sample stimulus, paired with anS1 and S2 from each of the two functionalclasses as comparisons; in addition, K and 11could appear together as the S1 and S2 ona trial, with one of the functional class mem-bers appearing as the sample. Correct re-sponses were defined by class-consistent re-sponses (i.e., matching K with any letter andmatching 11 with any number). This test de-sign generated 36 completely novel condi-tional discrimination transfer trials.

Expansion of functional classes through rein-forcer-mediated equivalence relations. The finaltests were designed to determine whethercommon associations with specific reinforcerswould establish emergent relations betweenpreviously unrelated stimuli and the function-al classes. To accomplish this, test trials withMTS Training Sets 1 and 2 were presented inthe context of the simple discrimination re-versal procedure for the first time. Test trialsconsisted of pairing one stimulus from eachof the two training sets in simple discrimina-


tion test trials. Each of the two choice stimulipresented on a trial had previously beenpaired with a different reinforcer (stimuli inSet 1 had been correlated with capelin; stim-uli in Set 2 had been correlated with her-ring). The test trials were embedded in sixfunctional class reversal sessions that alter-nated between letters and numbers positive.When letters served as S1s, selection of thetest stimuli that had also been paired withcapelin and the 587-Hz tone was reinforced;conversely, when numbers served as S1s, se-lection of the test stimuli that had also beenpaired with herring and the 293-Hz tone wasreinforced. Thus, the functional class cur-rently designated as positive determined theS1 for each test trial. In this way, the sametest trial could appear in two different rever-sal sessions with opposite reinforcement con-tingencies. There were nine novel pairings ofarbitrary stimuli tested (e.g., 1A vs. 2A, 1B vs.2C, 1C vs. 2A). Each pairing appeared twiceduring testing with opposite reinforcementcontingencies to generate a total of 18 novelsimple discrimination test trials. Three testtrials, which included a single exposure ofeach stimulus in a set, were presented in eachof the six reversal sessions. All correct re-sponses produced the same class-specific re-inforcers that had been used throughout thestudy.

Immediately following the simple discrim-ination transfer test, a second transfer test wasconducted. This test utilized the MTS proce-dure and tested for the emergence of novelconditional discriminations between the MTStraining stimuli and the functional class mem-bers that shared common reinforcers. Priorto testing, the training stimuli had never beendirectly paired with any functional class mem-bers. During testing, the training stimuli andthe functional class members were combinedinto novel conditional discriminations. Oneach trial, the S2 presented was a stimulusassigned to one reinforcer, and the sampleand S1 presented were assigned to the alter-nate reinforcer. On test trials, a training stim-ulus could serve as the sample with two stim-uli from opposing functional classes ascomparisons, or conversely, a functional classmember could appear as the sample with twostimuli from opposing training sets as com-parisons. Correct responses were defined asselection of the comparison stimulus that had

been assigned to the same reinforcer as thesample. All correct responses produced thesame class-specific reinforcers that had beenused throughout the study. The test proce-dure generated a total of 132 novel condi-tional discrimination transfer trials. Testingoccurred over six sessions that contained 22test trials and 12 baseline trials each. A com-plete list of the trial configurations tested canbe found in Appendix C.


After trained relations were established be-tween the new stimuli K and 11 and the func-tional class members J and 10, untrained re-lations emerged between the new stimuli andthe remaining members of each functionalclass. This transfer was nearly perfect. On thefirst presentation of the 36 novel conditionaldiscriminations, Rio was correct on 100% oftrials, and Rocky was correct on 91% of trials(binomial tests, p , .01). Performance ontransfer trials was not different from perfor-mance on familiar baseline trials for eithersubject (Rio scored 36 of 36 on test and 24of 24 on baseline trials; Rocky scored 33 of36 and 24 of 24; Fisher’s exact tests, p . .05).These data indicate that the functional classesformed by the sea lions also met the requisitecriteria for stimulus equivalence classes as de-scribed by Sidman and Tailby (1982).

In the test just described, functional classeswere expanded when equivalence relationsemerged between stimuli that were relatedthrough a common stimulus as well as a com-mon reinforcer. The next set of transfer testswas conducted to determine whether a com-mon reinforcer alone was sufficient to estab-lish equivalence classes consisting of function-al class members and previously unrelatedstimuli. On the first test, MTS training stimulithat had been associated with the same rein-forcers as functional classes were presented asnovel simple discrimination trials in sessionsin which one functional class was designatedas positive. Both Rio and Rocky correctlychose the stimulus that shared a common re-inforcer with the positive functional class on16 of the 18 novel transfer trials (binomialtest, p , .01). On the second test, the trainingstimuli and functional classes were combinedinto novel conditional discriminations. Rioand Rocky both correctly matched the train-ing stimuli with the functional class member


sharing a common reinforcer on 129 of the132 transfer trials (binomial test, p , .01).Performance on novel trials was not differentfrom performance on familiar baseline trials(Rio scored 72 of 72 on baseline trials; Rockyscored 68 of 72; Fisher’s exact test, p . .05).Both of these tests provide strong evidencethat stimuli associated with a common rein-forcer can become equivalence class mem-bers. This effect was strong and immediate,with appropriate classification occurring 99%of the time on novel transfer trials.

These results show that 2 California sea li-ons formed equivalence classes following ex-perience with class-specific reinforcement.Thus, our findings support and add to pre-vious work by Schusterman and Kastak (1993,1998) that demonstrated equivalence classi-fication by a California sea lion in the ab-sence of class-specific reinforcement. Thecompletion of these studies with Californiasea lions provides the strongest available evi-dence of equivalence in nonhuman subjects.The immediate transfer of controlling stim-ulus relations between functional classes andmore traditional equivalence classes suggeststhat the two classification schemes may com-prise the same cognitive-behavioral process,that is, that functional classes are equivalenceclasses. These findings bolster Vaughan’s(1988) view of stimulus classification and areconsistent with the results of Sidman et al.(1989) and Sidman (1994) for 2 of the 3 hu-man subjects tested with a similar procedure.These findings also support the interpreta-tions of Schusterman and Kastak (1998), whofound transfer of equivalence classes to func-tional classes, by showing the bidirectionalityof this transfer.

The transfer of stimulus function solelythrough relations with common reinforcerssupports Sidman’s answer to the question,‘‘Where do equivalence relations comefrom?’’ He has proposed that equivalence re-lations arise directly from the reinforcementcontingency (Sidman, 2000). A key consid-eration in his expanded equivalence model isthe inclusion of responses and reinforcers aspotential equivalence class members. Re-sponses and reinforcers enter into equiva-lence relations directly, through the contin-gencies that connect one or more stimuli toa defined response and then a defined rein-forcer. Consequently, when responses and re-

inforcers are the same for all contingencies,the differentiation of stimuli into equivalenceclasses may be hindered. Conversely, equiva-lence relations may arise more easily betweenstimuli when responses or reinforcers arecontingency specific. Within this expandedmodel, stimuli that share common responsesor reinforcers can become equivalent inmuch the same way that stimuli related toone another become equivalent.

This theory is testable, and the currentstudy provides at least two lines of empiricalevidence in its support. In all of the experi-ments that included shared stimulus relationsas well as class-specific reinforcers, the role ofthe reinforcers in our subjects’ performancewas best explained by the idea that they func-tioned as members of an equivalence class(Dube, McIlvane, Maguire, Mackay, & Stod-dard, 1989; Sidman, 1994, 2000). As classmembers, reinforcers could serve to strength-en new equivalence relations that arise be-tween stimuli through common stimulus–re-inforcer relations. Thus, in Experiment 3,equivalence relations emerged between stim-uli that were linked through intermediatestimuli as well as through specific reinforcers.However, if reinforcers can act as fully func-tioning class members, then it follows thatclass-specific reinforcers alone should be suf-ficient to induce equivalence relations toemerge between stimuli. The results of testspairing functional class members with otherstimuli related only through shared reinforc-ers indicate that specific reinforcers did func-tion as class members in our experiments.

GENERAL DISCUSSION

The present experiments show that 2 Cal-ifornia sea lions organized perceptually dif-ferent stimuli into equivalence classes on thebasis of common functional relations and re-inforcers. The classes that emerged were ro-bust, transferring readily from one procedureto another. The equivalence relations formedwithin the simple discrimination reversal pro-cedure were maintained and even expandedacross procedures to a conditional discrimi-nation task. Most significantly, the expandedclasses formed by the 2 sea lions met the for-mal criteria of stimulus equivalence classes.Thus, this study supports Vaughan’s (1988)proposition that functional classes generated


in simple discrimination reversal proceduresare the same as equivalence classes generatedthrough MTS procedures. In addition, our re-sults are consistent with Sidman’s (1994,2000) expanded theory of equivalence rela-tions, which includes responses and reinforc-ers as potential class members and allows thedemonstration of equivalence beyond the tra-ditional, mathematically derived frameworkof stimulus equivalence (Sidman & Tailby,1982).

Based on the evidence provided by this andother studies, we agree that equivalence clas-ses should be disentangled from the restric-tive procedural definitions imposed by amathematically based characterization. Givenan expanded theoretical framework, emer-gent stimulus relations demonstrated in a va-riety of experimental contexts may also com-prise equivalence relations. This idea issupported by work with human subjects show-ing that relations that emerge between stim-uli sharing ordinal positions in sequencetraining also meet the traditional criteria ofequivalence classes (Sigurdardottir, Green, &Saunders, 1990). Sequence training conduct-ed with nonhuman species in tasks requiringcontrol by ordinal positions or discriminationof sequential sign classes likely depends onsimilar proximal mechanisms (see Chen etal., 1997; Schusterman & Gisiner, 1997). Giv-en the overlap of many stimulus control con-cepts with the expanded theory of equiva-lence relations, it is clear that concepts suchas functional classification, ordinal knowl-edge, mediated generalization, non-similarity-based classification, acquired equivalence ofcues, and symbolic representation require re-evaluation in light of the expanded model ofstimulus equivalence.

The incorporation of class-specific reinforc-ers into our experimental design appeared tobe a key element in the sea lions’ successfulclassification performance. From an etholog-ical standpoint, interactions with a variety ofenvironmental signals, including individuals,objects, and events, result in specific conse-quences. Reinforcement contingencies thatgive rise to equivalencies may enable individ-uals to behave adaptively in the presence ofdisparate signals, to recognize objects acrossthe senses, and to rapidly acquire natural cat-egories. Such categories are subject to con-textual control (Bush, Sidman, & de Rose,

1989) and as flexible constructs, they are like-ly to be useful in facilitating the conceptualorganization of predator–prey relations aswell as social relations based on variables suchas activity, age, gender, kin, friendships, andrivalries (Schusterman & Kastak, 1998; Schus-terman, Reichmuth, & Kastak, 2000).

The fact that humans with and without lan-guage skills and at least one nonhuman spe-cies have demonstrated equivalence suggeststhat processes for classifying perceptually dif-ferent stimuli are relatively fundamental. Webelieve that given the appropriate testingconditions, equivalence can be demonstratedin a variety of animals. Further investigationwith different species and procedures that es-tablish emergent relations among stimuli, re-sponses, and reinforcers will be necessary todetermine whether this is the case.

REFERENCES

Balsam, P. D. (1988). Selection, representation, andequivalence of controlling stimuli. In R. C. Atkinson,R. J. Herrnstein, G. Lindzey, & R. D. Luce (Eds.), Ste-vens’ handbook of experimental psychology: Vol. 2. Learningand cognition (2nd ed., pp. 111–166). New York: Wiley.

Bush, K. M., Sidman, M., & de Rose, T. (1989). Contex-tual control of emergent equivalence relations. Jour-nal of the Experimental Analysis of Behavior, 51, 29–45.

Carr, D., Wilkinson, K. M., Blackman, D., & McIlvane, W.J. (2000). Equivalence classes in individuals with min-imal verbal repertoires. Journal of the Experimental Anal-ysis of Behavior, 74, 101–114.

Chen, S., Swartz, K. B., & Terrace, H. S. (1997). Knowl-edge of the ordinal position of list items in rhesusmonkeys. Psychological Science, 8, 80–86.

D’Amato, M. R., Salmon, D. P., Loukas, E., & Tomie, A.(1985). Symmetry and transitivity of conditional re-lations in monkeys (Cebus apella) and pigeons (Colum-ba livia). Journal of the Experimental Analysis of Behavior,44, 35–47.

Delius, J. D., Jitsumori, M., & Siemann, M. (2000). Stim-ulus equivalencies through discrimination reversals.In C. Heyes & R. Huber (Eds.), The evolution of cogni-tion (pp. 103–122). Cambridge, MA: MIT Press.

Dube, W. V., Callahan, T. D., & McIlvane, W. J. (1993).Serial reversals of concurrent auditory discrimina-tions in rats. The Psychological Record, 43, 429–440.

Dube, W. V., McIlvane, W. J., Callahan, T. D., & Stoddard,L. T. (1993). The search for stimulus equivalence innonverbal organisms. The Psychological Record, 43, 761–778.

Dube, W. V., McIlvane, W. J., Maguire, R. W., Mackay, H.A., & Stoddard, L. T. (1989). Stimulus class forma-tion and stimulus-reinforcer relations. Journal of the Ex-perimental Analysis of Behavior, 51, 65–76.

Fields, L., & Nevin, J. A. (1993). Stimulus equivalence[special issue]. The Psychological Record, 43.

Goeters, S., Blakely, E., & Poling, A. (1992). The differ-


ential outcomes effect. The Psychological Record, 42,389–411.

Hayes, S. C. (1989). Nonhumans have not yet shownstimulus equivalence. Journal of the Experimental Anal-ysis of Behavior, 51, 385–392.

Horne, P. J., & Lowe, C. F. (1996). On the origins ofnaming and other symbolic behavior. Journal of the Ex-perimental Analysis of Behavior, 65, 185–241.

Horne, P. J., & Lowe, C. F. (1997). Toward a theory ofverbal behavior. Journal of the Experimental Analysis ofBehavior, 68, 271–296.

Kastak, D., & Schusterman, R. J. (1994). Transfer of vi-sual identity matching-to-sample in two California sealions (Zalophus californianus). Animal Learning & Be-havior, 22, 427–435.

Keller, F. S., & Schoenfeld, W. N. (1950). Principles ofpsychology. New York: Appleton-Century-Crofts.

Lazar, R. (1977). Extending sequence-class membershipwith matching to sample. Journal of the ExperimentalAnalysis of Behavior, 27, 381–392.

Manabe, K., Kawashima, T., & Staddon, J. E. R. (1995).Differential vocalization in budgerigars: Towards anexperimental analysis of naming. Journal of the Experi-mental Analysis of Behavior, 63, 111–126.

Meehan, E. F. (1999). Class-consistent differential rein-forcement and stimulus class formation in pigeons.Journal of the Experimental Analysis of Behavior, 72, 97–115.

Riess, B. F. (1940). Semantic conditioning involving thegalvanic skin reflex. Journal of Experimental Psychology,36, 143–152.

Schenk, J. J. (1994). Emergent relations of equivalencegenerated by outcomes-specific consequences in con-ditional discriminations. The Psychological Record, 44,537–558.

Schusterman, R. J., & Gisiner, R. C. (1997). Pinnipeds,porpoises, and parsimony: Animal language researchviewed from a bottom-up perspective. In R. W. Mitch-ell, N. S. Thompson, & H. L. Miles (Eds.), Anthropo-morphism, anecdotes, and animals (pp. 370–382). Alba-ny: SUNY Press.

Schusterman, R. J., & Kastak, D. (1993). A California sealion (Zalophus californianus) is capable of formingequivalence relations. The Psychological Record, 43, 823–839.

Schusterman, R. J., & Kastak, D. (1998). Functionalequivalence in a California sea lion: Relevance to an-imal social and communicative interactions. AnimalBehaviour, 55, 1087–1095.

Schusterman, R. J., Reichmuth, C. J., & Kastak, D.(2000). How animals classify friends and foes. CurrentDirections in Psychological Science, 9, 1–6.

Sidman, M. (1994). Equivalence relations and behavior: Aresearch story. Boston: Authors Cooperative.

Sidman, M. (2000). Equivalence relations and the rein-forcement contingency. Journal of the ExperimentalAnalysis of Behavior, 74, 127–146.

Sidman, M., Rauzin, R., Lazar, R., Cunningham, S., Tail-by, W., & Carrigan, P. (1982). A search for symmetryin the conditional discrimination of rhesus monkeys,baboons and children. Journal of the Experimental Anal-ysis of Behavior, 37, 23–34.

Sidman, M., & Tailby, W. (1982). Conditional discrimi-nation vs. matching to sample: An expansion of thetesting paradigm. Journal of the Experimental Analysis ofBehavior, 37, 5–22.

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

Sigurdardottir, Z. G., Green, G., & Saunders, R. R.(1990). Equivalence classes generated by sequencetraining. Journal of the Experimental Analysis of Behavior,53, 47–63.

Skinner, B. F. (1935). The generic nature of the con-cepts of stimulus and response. Journal of General Psy-chology, 12, 40–65.

Tomonaga, M. (1999). Establishing functional classes ina chimpanzee (Pan troglodytes) with a two-item sequen-tial-responding procedure. Journal of the ExperimentalAnalysis of Behavior, 72, 57–79.

Vaughan, W., Jr. (1988). Formation of equivalence setsin pigeons. Journal of Experimental Psychology: AnimalBehavior Processes, 14, 36–42.

von Ferson, L., & Delius, J. D. (2000). Acquired equiv-alences between auditory stimuli in dolphins (Tursiopstruncatus). Animal Cognition, 3, 79–83.

Wasserman, E. A., & DeVolder, C. L. (1993). Similarityand nonsimilarity-based conceptualization in childrenand pigeons. The Psychological Record, 43, 779–793.

Wasserman, E. A., DeVolder, C. L., & Coppage, D. J.(1992). Non-similarity-based conceptualization in pi-geons via secondary or mediated generalization. Psy-chological Science, 3, 374–379.

Wulfert, E., & Hayes, S. C. (1988). Transfer of a condi-tional ordering response through conditional equiv-alence classes. Journal of the Experimental Analysis of Be-havior, 50, 125–144.

Zentall, T. R. (1998). Symbolic representation in ani-mals: Emergent stimulus relations in conditional dis-crimination learning. Animal Learning & Behavior, 26,363–377.

Received September 20, 2000Final acceptance June 9, 2001


APPENDIX A

The test-trial configurations and baseline trials used in each phase of Experiment 2 are shown.Each row shows two test trials and corresponding alternate choices.

Transfertest Test stimuli

Testtrials

(Sample r S1)

Alternatechoices(S2) Baseline trials

1 E, I4, 8

ErI, IrE4r8, 8r4

4, 8E, I

MTS Training Sets 1, 2, 3, and 4

2 B, H2, 10

BrH, HrB2r10, 10r2

10, 2B, H

MTS Training Sets 1 and 2,Transfer Test 1 relations

3 A, J7, 9

ArJ, JrA7r9, 9r7

9, 7A, J

MTS Training Sets 1 and 2,Transfer Tests 1 and 2 relations

4 E, I, B, H, A, J4, 8, 2, 10, 7, 9

ErH, HrEErB, BrEErJ, JrEErA, ArEJrH, HrI

7, 99, 710, 22, 107, 9

MTS Training sets 1 and 2,Transfer Tests 1, 2, and 3 relations

IrB, BrIIrJ, JrIIrA, ArIBrA, ArBBrJ, JrB

9, 710, 22, 104, 88, 4

HrA, ArHHrJ, JrH4r2, 2r44r10, 10r44r7, 7r44r9, 9r48r2, 2r8

4, 88, 4A, JJ, AB, HH, BA, J

8r10, 10r88r7, 7r88r9, 9r82r7, 7r22r9, 9r210r7, 7r1010r9, 9r10

J, AH, BB, HE, II, EE, II, E

5 C, D, F, G1, 3, 5, 6

CrD, DrCCrF, FrCCrG, GrCDrF, FrDDrG, GrDFrG, GrF

3, 61, 55, 36, 35, 11, 6

MTS Training Sets 1 and 2,Transfer Tests 1, 2, 3, and 4 relations

1r3, 3r11r5, 5r11r6, 6r13r5, 5r33r6, 6r35r6, 6r5

G, CF, DC, FD, FG, CD, G

6 A through J1 through 10

CrA, ArCCrB, BrCCrE, ErCCrH, HrCCrI, IrCCrJ, JrCDrA, ArDDrB, BrDDrE, ErDDrH, HrDDrI, IrD

10, 41, 34, 68, 49, 310, 79, 54, 67, 101, 87, 4

MTS Training Sets 1 and 2,Transfer Tests 1, 2, 3, 4, and 5relations

DrJ, JrDFrA, ArFFrB, BrF

8, 39, 59, 3


APPENDIX A

(Continued)

Transfertest Test stimuli

Testtrials

(Sample r S1)


FrE, ErFFrH, HrF

7, 26, 2

FrI, IrFFrJ, JrF

10, 39, 1

GrA, ArGGrB, BrGGrE, ErGGrH, HrGGrI, IrGGrJ, JrG

6, 29, 37, 66, 28, 56, 8

1r2, 2r11r4, 4r11r7, 7r11r8, 8r11r9, 9r11r10, 10r1

D, FB, AI, FJ, GH, JD, J

3r2, 2r33r4, 4r33r7, 7r33r8, 8r33r9, 9r33r10, 10r3

C, DI, FC, HC, II, EE, A

5r2, 2r55r4, 4r55r7, 7r55r8, 8r55r9, 9r5

J, EF, CG, JF, HJ, C

5r10, 10r56r2, 2r66r4, 4r66r7, 7r66r8, 8r66r9, 9r66r10, 10r6

J, GD, AC, BG, BG, HB, CF, E

Total testtrials

180


APPENDIX B

The MTS test-trial configurations and baseline trials used in the first part of Experiment 3(expansion of functional classes through stimulus-mediated equivalence relations) are shown.Each row shows two test trials and corresponding alternate choices.

Training and testingphases

Train/testtrials

(Sample r S1)


Forward training JrK10r11

11K

MTS Training Sets 1 and 2A–J1–10

Symmetry training KrJ11r10

10J

Combined training JrK, KrJ10r11, 11r10

11, 10K, J

Transfer testing ArK, KrABrK, KrBCrK, KrCDrK, KrDErK, KrEFrK, KrF

11, 911, 611, 511, 211, 411, 1

MTS Training Sets 1 and 2A–J1–10K and 11 training trials

GrK, KrGHrK, KrHJrK, KrI1r11, 11r12r11, 11r23r11, 11r34r11, 11r4

11, 811, 311, 7K, HK, FK, DK, C

5r11, 11r56r11, 11r67r11, 11r78r11, 11r89r11, 11r9

K, GK, IK, AK, EK, B

Total test trials 36


APPENDIX C

The MTS test-trial configurations and baseline trials used in the final part of Experiment 3(expansion of functional classes through reinforcer-mediated equivalence relations) areshown. Each row shows two test trials and corresponding alternate choices.

Test stimuliTest trials

(Sample r S1)


A–K with MTS TrainingStimuli 1A, 1B, 1C

1–11 with MTS TrainingStimuli 2A, 2B, 2C

1ArA, Ar1A1ArB, Br1A1ArC, Cr1A1ArD, Dr1A1ArE, Er1A

4, 2A9, 2A5, 2A3, 2A1, 2A

MTS Training Sets 1 and 2A–K1–11

1ArF, Fr1A1ArG, Gr1A1ArH, Hr1A1ArI, Ir1A1ArJ, Jr1A

11, 2A8, 2A6, 2A2, 2A7, 2A

1ArK, Kr1K1BrA, Ar1B1BrB, Br1B1BrC, Cr1B1BrD, Dr1B

10, 2A2, 2B1, 2B

11, 2B7, 2B

1BrE, Er1B1BrF, Fr1B1BrG, Gr1B1BrH, Hr1B1BrI, Ir1B1BrJ, Jr1B

8, 2B9, 2B4, 2B

10, 2B3, 2B6, 2B

1BrK, Kr1B1CrA, Ar1C1CrB, Br1C1CrC, Cr1C1CrD, Dr1C1CrE, Er1C

5, 2B1, 2C

11, 2C8, 2C2, 2C7, 2C

1CrF, Fr1C1CrG, Gr1C1CrH, Hr1C1CrI, Ir1C1CrJ, Jr1C1CrK, Kr1C

10, 2C6, 2C4, 2C5, 2C9, 2C3, 2C

2Ar1, 1r2A2Ar2, 2r2A2Ar3, 3r2A2Ar4, 4r2A2Ar5, 5r2A2Ar6, 6r2A

B, 1AF, 1AK, 1AE, 1AJ, 1AA, 1A

2Ar7, 7r2A2Ar8, 8r2A2Ar9, 9r2A2Ar10, 10r2A2Ar11, 11r2A

G, 1AD, 1AI, 1AC, 1AH, 1A

2Br1, 1r2B2Br2, 2r2B2Br3, 3r2B2Br4, 4r2B2Br5, 5r2B

A, 1BJ, 1BF, 1BG, 1BC, 1B

2Br6, 6r2B2Br7, 7r2B2Br8, 8r2B2Br9, 9r2B2Br10, 10r2B2Br11, 11r2B

E, 1BD, 1BI, 1BK, 1BH, 1BB, 1B


APPENDIX C

(Continued)

Test stimuliTest trials

(Sample r S1)


2Cr1, 1r2C2Cr2, 2r2C2Cr3, 3r2C2Cr4, 4r2C2Cr5, 5r2C2Cr6, 6r2C

B, 2CI, 2CA, 2CK, 2CE, 2CD, 2C

2Cr7, 7r2C2Cr8, 8r2C2Cr9, 9r2C2Cr10, 10r2C2Cr11, 11r2C

G, 2CC, 2CJ, 2CF, 2C1A, 2C

Total test trials 132

EQUIVALENCE CLASSIFICATION BY CALIFORNIA SEA …people.uncw.edu/galizio/website/Articles/Kastak 2001.pdf · 131 journal of the experimental analysis of behavior 2001, 76, 131–158

Documents