Embed Size (px)
Transcript of Horne Lowe
JOURNAL OF THE EXPERIMENTAL ANALYSIS OF BEHAVIOR 1997, 68, 271296 NUMBER 2 (SEPTEMBER)
TOWARD A THEORY OF VERBAL BEHAVIOR
PAULINE J. HORNE AND C. FERGUS LOWE
UNIVERSITY OF WALES
This paper is a reply to an accompanying set of six commentaries by Sidman; Hayes and Barnes;Schusterman, Kastak, and Reichmuth; Tonneau and Sokolowski; Lowenkron; and Moerk. Thosecommentaries were prompted by our article On the Origins of Naming and Other Symbolic Be-havior (1996), which was, in turn, followed by 26 commentaries and a reply. In the course of thepresent reply, we further develop the naming account to embrace more complex verbal relationssuch as same, different, more, and less. We also examine what we see as the lack of conceptual coherencein equivalence theories, including relational frame theory, and the disparities between these accountsand the findings from empirical research.
Key words: naming, verbal behavior, language, equivalence, relational frame theory, mediation,match to sample
Our previous reply (Lowe & Horne, 1996)addressed issues raised in 26 initial commen-taries on our (Horne & Lowe, 1996) accountof early language learning and its relevanceto recent research on stimulus classes, in par-ticular those identified with the term equiva-lence. In this reply we tackle the main issuesraised in six further commentaries, clarifyingevident misunderstandings about our analysisand resharpening the focus upon what we seeas the central concerns for future theoreticaland experimental work. We begin by review-ing the objectives of the naming account andexamine the criticism that it is mediational. Wethen consider some of the main points raisedby Sidman, particularly concerning the rela-tionship between equivalence theory and lan-guage and, more generally, the theoreticaland empirical status of the concept of equiv-alence itself. In response to Hayes andBarnes, we attempt a more detailed evalua-tion of relational frame theory than we pro-vided in the initial essay, and a direct com-parison of the ways in which it and thenaming account, respectively, explain derived
Our thanks to A. Charles Catania for taking the timeto read and comment upon a draft of the manuscriptand for being so positive in his response. We are indebtedto Carl Hughes for his help with literature searches, toGareth Horne for producing Figure 1, and to Pat Barron,Judith Brooke, and Sue Peet for their assistance in thepreparation of the manuscript. Again we are particularlyindebted to Pat Lowe for her invaluable editorial contri-bution which has enhanced every aspect of the paper.
Correspondence and request for reprints should be ad-dressed to Pauline Horne or Fergus Lowe, School of Psy-chology, University of Wales, Bangor, Gwynedd, LL572DG, United Kingdom.
stimulus relations and the findings of rela-tional frame research. We reexamine the ev-idence of Schusterman and colleagues as tothe purported success of a sea lion in equiv-alence tests, and the arguments provided byTonneau and Sokolowski in favor of molartheories of equivalence. Finally, we considerpoints raised by Lowenkron concerning thebehavioral processes by which naming is es-tablished and more general theoretical issuesaired by Moerk. We are grateful for the con-tributions of all the commentators, in bothrounds of this discussion; without such shar-ing of ideas the advancement of theory andrelated empirical research would be slow in-deed.
Equivalence, along with related issues, hasfeatured prominently in the commentaries.This is hardly surprising, given that it hasbeen the focus of considerable conceptual in-terest and innovative research within behav-ior analysis for some time. We are in no doubtthat the legacy of much of this work will bepositive. Murray Sidman, in particular, hasmade a central and lasting contribution to-wards ensuring that stimulus classification, orcategorization, and the role it plays in lan-guage are put firmly at the top of the re-search agenda and, whatever the researchprocedures finally adopted in this endeavor,his methodological rigor and inventivenesswill continue to serve as a model for us all.Thanks to Steven Hayes and colleagues, thetargets of research have been broadened toinclude a range of derived relations. Howev-er, these advances notwithstanding, our ap-praisal of the concept of equivalence and its
272 PAULINE J. HORNE and C. FERGUS LOWE
role in behavior analysis remains highly crit-ical.
In any case, as we outlined in our targetarticle, our prime concern is not with equiv-alence and its related issues, but rather withthe development of an overall theory of be-havior. Such a theory should embrace notonly the contingency-shaped behavior of non-human animals but also complex human be-havior, particularly the behavior that Skinnertermed rule governed or that we term verballycontrolled (Horne & Lowe, 1996, pp. 212213). As Skinner recognized, an understand-ing of the distinction and interaction betweenthese two basic types of behavioral relation iscrucial in the analysis of human behavior. Butsuch an analysis cannot proceed without anaccount of verbal behavior itself and of howit is incorporated into the broader behavioralsystem and so generates a range of phenom-ena that include verbal rules. This was thetask Skinner set himself in Verbal Behavior(1957) and numerous subsequent writingsand that we also set ourselves in our paperon naming.
For many years the need to make specialaccommodations to account for distinctly hu-man functioning went largely unrecognizedwithin behavior analysis. Indeed, it was widelyassumed that the functional determinants ofoperant behavior in humans were no differ-ent from those that applied in other animalspecies (see Lowe, 1979, 1983). Human sub-jects routinely participated in experimentalinvestigations of the effects of schedules ofreinforcement, and researchers discussed theresults in terms of the latter contingencies,just as they did similar procedures with ani-mals. Unfortunately, however, there were sub-stantial differences between the animal andhuman sets of data that unless they were ex-plained seemed to rule out straightforwardcontingency analyses. Progress began to bemade only when investigators acknowledgedthat the verbal repertoire that verbally ablehuman subjects bring with them to an exper-iment inevitably transforms the experimentalenvironment into one that is also substantiallyverbal. Whether or not these subjects are ex-plicitly provided with experimental instruc-tions, they instruct themselves about theirown behavior and its outcomes. For them,many features of that environment, includingthe experimental procedure itself, occasion
naming and complex forms of self-instruc-tion. Such verbalizations give rise to a rangeof behavior patterns that are never seen inthe animal literature (Bentall, Lowe, & Beas-ty, 1985; Catania, Horne, & Lowe, 1989; Ca-tania, Matthews, & Shimoff, 1982; Horne &Lowe, 1993; Lowe, 1979; Lowe, Beasty, & Ben-tall, 1983; Lowe & Horne, 1985). Quite sim-ply, much of human operant behavior, notleast that studied under laboratory condi-tions, is rule governed or verbally controlledrather than contingency shaped (Lowe,1979). This language hypothesis is nowwidely accepted (see Hayes & Hayes, 1992),and there seem to us no good grounds forassuming that human behavior on condition-al discrimination tasks should be any differ-ent. It is implausible that human subjects inthese conditions would respond only in ac-cordance with the experimenter-defined con-tingencies of reinforcement and not also totheir own verbal behavior about these contin-gencies (see also Remington, 1996; K. Saun-ders & Spradlin, 1996). Indeed, the need toconsider the effects of verbal control is par-ticularly compelling in this domain given that(a) the differences between humans and oth-er animal species are even more clear-cut inmatch-to-sample than in schedule perfor-mance; (b) the involvement of verbal behav-ior in the forming of arbitrary stimulus class-es in match-to-sample procedures is so welldocumented (Horne & Lowe, 1996, pp. 215227); (c) the operation of verbal effects onmatch-to-sample performance would be con-sistent with what we know about the rest ofhuman operant performance. We suggest,therefore, that until it addresses the role ofverbal control, conditional discrimination re-search with humans will make little signifi-cant progress.
In our target article we addressed languagedevelopment up to the point when childrenbegin to combine names, usually during the2nd year of life. We hoped, in the course ofdescribing what takes place during this phaseof development, to specify namingin ourview, the higher order behavioral relationthat is the basic behavioral unit of language.As a central feature of that specification, weaimed to show how naming classifies or cat-egorizes objects and events and is the basisfor rule-governed or verbally controlled be-havior, but we did not attempt to go beyond
273A THEORY OF VERBAL BEHAVIOR
this to deal with autoclitic verbal behavior(Skinner, 1957). However, in order to re-spond to some of the commentaries in thepresent round (see Hayes & Barnes), we havedeveloped the account to embrace some ofthese more complex verbal relations.
NAMING AND MEDIATION
Because it involves naming, ours is thoughtby some to be a mediational account of equiv-alence (see commentaries by Sidman and byTonneau & Sokolowski; see also Hayes, 1994;Hayes & Hayes, 1992; Stromer & Mackay,1996; Stromer, Mackay, & Remington, 1996).Perhaps because many authors identify media-tional with the kind of mental constructs crit-icized by Skinner (1950), the term is not of-ten used without pejorative overtones. Butthere is an important distinction to be made:Unlike an unobservable hypothetical con-struct, verbal behavior is behavior that can oc-cur equally well in overt or covert form. Toinclude it in an explanatory system is thus notto invoke some shadowy construct that existsat another level of explanation in which thesubject matter is measured in different di-mensions (Skinner, 1950). And clearly, itwould be an odd account of human behaviorthat excluded explanation of language and itseffects on other behavior. Indeed, were prej-udices against this type of mediational ac-count to hold sway, then much of Verbal Be-havior (1957) and Skinners other writings inthis area would have to be discarded. Out, forexample, would go all of his work on the au-toclitic, rule-governed behavior, and verbalthinking. For Skinner, whether one appealsto mediating events, overt or covert, is not anideological matter but depends upon wheth-er the functional analysis demands it. VerbalBehavior is replete with examples for which itis necessary to infer such mediating behavior;not to make such inferences, he argues,would leave embarrassing gaps in the ac-count. For example, he writes, When some-one solves a problem in mental arithmetic,the initial statement of the problem and thefinal overt answer can often be related onlyby inferring covert events. We have to ac-count for verbal behavior which is under thecontrol of covert speechwhich reports it orqualifies it with autoclitics (1957, p. 434).
More specifically, Tonneau and Sokolowski
suggest in their commentary that our posi-tion here is comparable to Pavlovian media-tional theories that are based on unobserva-ble inferred stimulusresponse chaining.This indicates a misunderstanding not only ofour account of verbal behavior but also ofSkinners. The naming relations we describe,incorporating as they do stimulus classes andbidirectionality within an operant account,are far removed from anything described insuch mediational theories. Tonneau and So-kolowski, however, propose that in the nam-ing framework each covert verbal response(a) can be traced back to some prior overtbehavior (A), much as in Pavlovian media-tional accounts of the behavior of nonverbalanimals (e.g., Holland, 1981, 1990), in whichenvironmental events (A, B, and C) are sup-posed to evoke covert representations (a, b,and c) in the subjects. If this be the case, theyargue, why not establish what were the rele-vant overt behaviors and confine the accountto the overt domain? But the naming andother verbal behavior that feature in our ac-count occur both overtly and covertly; in-deed, in young children they exist initially atthe overt level alone and only later occur cov-ertly as well (Horne & Lowe, 1996). And likeSkinner (1957) and Ryle (1949a), we main-tain that verbal behavior, regardless of wheth-er it is overt or covert, is not a ghostly echoof environmental events. It is free-operant be-havior. The verbal responses it incorporatesbear no formal resemblance to, and occur ina different modality from, the environmentalstimuli that evoke them. In addition, verbalresponses to stimuli can occur many timesand in any order and are entirely free of thetemporal ordering that is critical to standardmediational accounts. Verbal behavior is, inshort, certainly not an unobservable hypo-thetical construct such as features in the me-diational theories of Holland and others. Andas far as its effects on an individuals otherbehaviors are concerned, it is largely irrele-vant whether it occurs covertly or overtly.
This is not to say, however, that covertchaining of sorts is never involved in match-to-sample experiments but that, if it occurs,it is likely to be intraverbal chaining (Skin-ner, 1957, p. 423; and see Horne & Lowe,1996). In addition, we maintain that it is pre-cisely because intraverbal behavior and com-plex autoclitic behavior are sometimes in-
274 PAULINE J. HORNE and C. FERGUS LOWE
volved, and such behavior takes time to emit,that nodal distance and other similar effectsoccur in studies of stimulus equivalence andrelational frames (Horne & Lowe, 1996;Lowe & Horne, 1996; and see RelationalFrame Theory and Naming, below). On theother hand, Horne and Lowe (1996) providenumerous examples for which intraverbal be-havior is not necessary but common namingmay suffice. A single common name can bea very quick and highly effective means ofclassifying physically different stimuli in con-ditional discrimination tasks. We stress, how-ever, that the main focus of interest for us isnot how subjects succeed on match-to-sampletasks, but the classifying and categorizing be-havior inherent in naming itself. There is nomediation involved here. Indeed, we have ar-gued that every time individuals name an ob-ject they directly classify it (Horne & Lowe,1996; and see Skinner, 1957, pp. 107129).
There is a real sense, on the other hand,in which Sidmans concept of equivalenceand the various frames and abstract conceptsof relational frame theory can be said to me-diate language, symbolic functioning, or ver-bal behavior. In our view the conceptualterms that Sidman and Hayes employ do notrefer to behavior but are redescriptions of be-havioral relations then used to explain the be-havior from which they were derived. And be-cause, in both their accounts, equivalence isregarded as a prerequisite for language, it isimplied that we cannot look directly at thefunctional relations between the environ-ment and different forms of verbal behaviorbut must do so instead through the mediat-ing concept of equivalence. Explanations inrelational frame theory rely on elaborateframing mediations that intervene betweenthe environment and behavior. To take justone of many possible examples, Hayes, Gif-ford, and Wilson (1996) write,
For example, imagine a situation in which agroup of boxes are waiting on the lawn to beplaced into a moving van. On what basis areproperties of the boxes selected, in order todetermine the order in which to move them?According to weight, to size, to fragility of con-tents? The dimension(s) controlling behaviormay be verbally describedmay acquire func-tions by participating in relational framesand may then in turn constitute the basis foradditional responding. (p. 294)
What does the mediation of acquire func-tions by participating in relational framesbetween the objects and behavior possiblyadd to our understanding of the relations in-volved here? A virtue of the naming accountis that it dispenses entirely with such media-tion.
CAN EQUIVALENCETHEORIES ACCOUNT
Mouse is a syllable. Now a mouse eats cheese;therefore, a syllable eats cheese. Suppose nowthat I cannot solve this problem. . . . Withoutdoubt I must beware, or some day I shall becatching syllables in a mousetrap, or, if I growcareless, a book may devour my cheese! (Sen-eca, Epistulae Morales)
In his contribution, Sidman is sympatheticto Dugdale and Lowes (1990) suggestionthat naming involves a symmetric relation be-tween the name and the thing named. In-deed Sidman goes further and proposes therelation between names and the stimuli thatoccasion them to be not only symmetric butreflexive and transitive as well (p. 262).Thus Sidmans view, shared by Hayes, is thata name is equivalent to the stimulus that oc-casions it, and the latter is the meaning of thename (Sidman, 1994, pp. 343, 346, 365; andsee Wulfert & Hayes, 1988). Much of our tar-get article, however, was based upon a rec-ognition that an account of naming in termsof stimulusresponse symmetry or equiva-lence simply did not fit the facts. Accordingto Sidmans account, behavioral symmetry ex-ists when, for example, subjects trained in aconditional discrimination procedure to emita selection response to Stimulus B upon pre-sentation of Stimulus A, proceed without fur-ther training to emit the same response to Awhen presented with B; that is, any pair ofsymmetrically related stimuli are functionallysubstitutable or interchangeable, and evokethe same response form. In his commentaryhe cites as an example of object-name sym-metry, a child naming a boy. But symmetry inthis context would yield the following: Thechild, having learned upon hearing /wheresthe boy?/ to look at and point to a boy, shouldthen upon seeing a boy look at and point tothe auditory stimulus /boy/; it is clear thatwithin the symmetry relation there is no be-
275A THEORY OF VERBAL BEHAVIOR
havioral basis for the child to emit the vocalresponse boy. And similarly with the re-verse condition: Having learned to emit thevocal response boy when she sees a boy,there is no behavioral basis for the childsemitting a response of an entirely differentform (i.e., looking at or pointing to a boy,when she hears /boy/). The problem encoun-tered by such a symmetry account is thatnaming has a variety of behavioral compo-nents; for example, objects and events evokespeaker behavior (e.g., vocalizing or signing),which produces stimulation that evokes lis-tener behavior such as orienting, pointing,reaching, and so forth. Accordingly, the boyand the auditory stimulus /boy/ cannot befunctionally substitutable because eachevokes different behavior from the other; theboy evokes a vocal response, whereas the au-ditory stimulus /boy/ evokes behavior of avery different form such as orienting andpointing.
That there is a complete absence of anyform of symmetry between the speaker andlistener components of the name relation be-comes even clearer when one considers thatwhen a child sees a stimulus (e.g., a boy), thismay evoke a particular vocal response (e.g.,boy), but the stimulation (/boy/) therebyproduced can evoke a range of disparate re-sponses relating not just to that particularstimulus (i.e., the boy just encountered) butto the entire class of events (e.g., the accu-mulated characteristics of boys in general)encompassed by the name relation con-cerned (and see Skinner, 1957, p. 117).
This absence of symmetry is, in fact, at theheart of the fundamental differences that ex-ist between names and the objects named.Naming is, after all, categorizing behavior re-lating to classes of objects and events. But anobject such as, for example, a tree, does notclassify or categorize. In addition, because aname evokes orienting behavior not only tothe particular stimulus that occasions it butto an indefinite class of objects or events, aparticular referent can never be the meaningof a name (and see Hunter, 1974; Ryle,1949b; Scruton, 1994; Skinner, 1957, pp. 110,117). Whenever naming is evoked, there is abehavioral transition from the particular tothe general; naming is inherently generaliz-ing, categorizing behavior. Herein, we sug-gest, lies the solution to the problem that Sid-
man ponders in his commentary when heasks, how does language help us to abstract,to generalize, to analyze, and to synthesize,and how does it come to do so? (p. 263; andsee Horne & Lowe, 1996).
The fact that naming relations cannot bedescribed in terms of symmetry or equiva-lence casts considerable doubt on Sidmansimportant claim that the formation of anequivalence class permits us to say that se-mantic correspondences have been estab-lished and that each of the class membershave the same meaning or that each is themeaning of the other (Sidman, 1994, p. 343;see also Hayes & Hayes, 1992; Wulfert &Hayes, 1988). A hypothetical example servesto illustrate the extent of the difficulty. Sup-pose that English-speaking subjects are pre-sented with novel graphic stimuli. On someoccasions when shown Stimulus A, their se-lection of Stimulus B from a range of alter-natives is reinforced; on other occasionswhen shown Stimulus A, their selection ofStimulus C is reinforced. A large number ofsuch conditional discriminations are estab-lished. The novel stimuli are in fact Chinesecharacters; those stimuli presented as samplesare questions, and the reinforced selectionsare the corresponding correct answers. Thus,presented with Stimulus A, which is furni-ture? (in Chinese), the subjects respond cor-rectly when they select Stimulus B, chair,or Stimulus C, bed (both also in Chinese).What then would they have learned? Al-though they may indeed respond correctly tothe questions with answers as well asmight be expected of native Chinese speak-ers, as Searle (1980) has argued, there are nogrounds to suppose that they have in factlearned the semantic relations represented bythese questions and answers, or any Chinese.Thus far, this conclusion appears to be con-sonant with the equivalence theories of Sid-man and Hayes: Because all of the relationshave been directly trained, there can be noevidence of symmetry or transitivity and,hence, the linguistic prerequisite (Sidman,1994) is missing. But assume the subjects goon to pass the necessary tests of symmetryand transitivity. When presented with bedthey respond either with which is furni-ture? or with chair. According to equiva-lence theory the subjects success here shouldprovide the evidence that semantic relations
276 PAULINE J. HORNE and C. FERGUS LOWE
have been established. In fact, of course, nowhaving succeeded on the equivalence tests,the subjects selections are even less consis-tent with the semantics of Chinese than theywere during initial training. This exampleclearly demonstrates that the existence ofsymmetrical or equivalence relations betweenstimuli cannot tell us which stimulus is thename of which, or indeed whether any par-ticular linguistic relations are involved. Thiswould be equally true if the subjects of suchan experiment were sea lions or computers.
Further evidence of the impossibility of ac-counting for language in terms of eitherequivalence or relational frames attendsclaims that the use of words such as is oris called establish symmetry and a frame ofcoordination (i.e., equivalence) betweenwords and objects (Hayes et al., 1996; Hayes& Hayes, 1989, p. 169). Thus, Hayes et al.(1996) write, For example, suppose a childis told this is called a ball. If this is called aball, then a ball is this. That is, the words iscalled quite reliably predict reinforcementfor symmetry in this context (p. 288). Butwhoever says a ball is this? And if there re-ally were symmetry, what should emergewould be the observation a ball is calledthis which, of course, would not make senseat all. In short, if the word is really wereable to establish equivalence or a frame ofcoordination between terms so related, thenSenecas quandary, outlined in the openingquotation of this section, would be real.
Conclusion. We believe that the view thatthere is a symmetrical relation betweennames and objects has arisen at least in partfrom the absence of a clear specification ofwhat a name is. For example, the word is var-iously used as a term for the vocal responseto a stimulus, the auditory stimulus producedby the vocal response, or a combination ofboth. When the behavioral components ofnaming are clearly specified, however, it isnot difficult to see that there can never besymmetry between names and objects (seealso Catania, in press).
THE CONCEPT OFEQUIVALENCE
Much of Sidmans commentary is devotedto a discussion of definitional aspects ofequivalence, but just as we have questioned
whether the concepts of symmetry and equiv-alence are of value in the analysis of linguisticrelations, it is similarly pertinent to ask wheth-er these concepts are of value in accountingfor performance on the conditional discrim-ination tasks from which they were originallyderived. We also question whether equivalenceis any longer a coherent concept.
Definitional problems. R. Saunders andGreens (1992) paper on the Nonequival-ence of Behavioral and Mathematical Equiv-alence dealt a considerable blow to the no-tion that one could simply translatemathematical into behavioral equivalence.They showed, for example, that generalizedidentity matching cannot be used to test forreflexivity (cf. Steele & Hayes, 1991). The im-plication of this is that among all the manystudies purporting to show stimulus equiva-lence there are none, in fact, in which allthree criteria for equivalence enjoined by Sid-man are satisfied. According to the operation-al definition, we have therefore never had anestablished instance of stimulus equivalence!Saunders and Green showed, in addition,that even where the tests prove to be nega-tive, subjects may nevertheless have classi-fied the stimuli in ways that meet the equiv-alence criteria of symmetry and transitivity.Their paper concluded that tests for equiva-lence are thus neither definitive nor exhaus-tive and that stimulus equivalence specifi-cally, and stimulusstimulus relations ingeneral, are far more complex behavioralphenomena than the invocation of the math-ematical analogy implies (p. 238).
What is even more problematic for equiv-alence theories is that recent research has re-ported behavioral relations that should notexist were behavior to follow the orderly reg-ularities of the mathematical relations towhich Sidman alludes in his commentary.Studies by Pilgrim and Galizio (1995) and byPilgrim, Chambers, and Galizio (1995), forexample, have shown a dissociation betweenbaseline and symmetry versus transitivity-equivalence patterns of responding on amatch-to-sample procedure. According toequivalence theory, the relation betweenstimulus pairs, including which particularpairs are symmetrically related, should bespecified by the test for equivalence, but inthese studies equivalence and symmetry go indifferent directions and relate to different
277A THEORY OF VERBAL BEHAVIOR
stimulus pairings. Pilgrim and Galizio (1995)conclude that these findings cast into doubtthe nature of equivalence as an underlying,fundamental, and integrated process (p.237). Other studies (for reviews, see Horne& Lowe, 1996; Pilgrim & Galizio, 1995, 1996;R. Saunders & Green, 1992) have shown that(a) extensive training on conditional discrim-inations may be entirely unsuccessful in es-tablishing success on any of the tests; (b) suc-cess on some or all of the tests may dependupon the particular subjects being studied;and (c) some or all of the tests can be passedin the absence of any conditional discrimi-nation training.
In his commentary Sidman indicates howhe has radically revised his original theory inorder to account for some of these findingsbut, as we have argued (Horne & Lowe,1996), most of the serious problems remain,not least the core problem of how to providea coherent behavioral definition or specifi-cation of equivalence. Indeed, we have pro-posed that in trying to put right the deficien-cies in the theory, additional major problemshave been introduced. This is a view that isnow also shared by Hayes and Barnes who, intheir commentary, argue that the concept ofequivalence has broken down, leaving onlythe concept of partition or class standing . . .and that was a concept we had before equiv-alence arrived (p. 238). How the abandon-ment of the concept of equivalence, uponwhich relational frame theory was largelybased, affects that latter theory is a matter towhich we shall return.
Problems with the data. Sidman proposes thathis account of equivalence relations providesa useful framework within which to organizethe existing data and principles of behavioranalysis. Here, however, we briefly summarizewhat we see as the main empirical deficien-cies of his theory, most of which are commonto relational frame theory. It does not ac-count for (a) major differences between hu-mans and nonhumans in their success ontests of equivalence, or (b) the finding thathuman subjects success on equivalence testsis related to their linguistic skills. Contrary towhat is suggested by Sidman in his commen-tary, we maintain that there is no good evi-dence that subjects without naming skillshave ever passed equivalence tests. It hasbeen acknowledged that to furnish as evi-
dence postexperimental naming tests that, al-though they purportedly show absence ofcommon naming, at the same time ignore in-traverbal naming and autoclitic behavior, isnot convincing (K. Saunders & Spradlin,1996, p. 30; see also de Rose, 1996; Galizio,1996; Pilgrim, 1996; Remington, 1996). Thetheory also does not account for (c) rule-gov-erned or verbally controlled success on equiv-alence tasks and a possible distinction be-tween this and contingency-shapedperformance (and see Relational Frame The-ory and Naming, below); (d) why it is oftennecessary to provide prolonged testing ofequivalence before subjects show correctperformances (Pilgrim et al., 1995; Pilgrim &Galizio, 1995; and see Relational Frame The-ory and Naming, below); (e) why perfor-mances on equivalence tests are an inversefunction of nodal distance (Fields, 1996) orrelational frame complexity (Steele & Hayes,1991; and see Relational Frame Theory andNaming, below); and (f) how context deter-mines success on equivalence tests. Neitherdo equivalence theories contribute much toprediction and control. According to Sidman,the fact that stimulus classes are ever ob-served in the laboratory is itself a mystery un-less one takes into account context. He main-tains that if we do get the experimenter-definedclasses it is because the context, and perhapsalso the response and the reinforcer, dropsout of the equivalence relations. And if wedo not, and instead all the stimuli in the ex-periment form one large overarching class,then this will be because the context has notdropped out. The question of how this drop-ping out occurs and the conditions underwhich it might or might not happen, Sidmandoes not address. Indeed, he goes so far as toassert, The mathematics does not pretend toprovide a basis for predicting whether or nota particular set of baseline conditions willgenerate equivalence relations. It only tells ushow to find out whether particular eventpairs belong to that relation (1994, p. 540).Context, both past and present, is certainlyresponsible for subjects behavior in equiva-lence experiments, relational frame studies,and indeed just about all experiments in psy-chology. The problem is that if context is ev-erything it is also nothing when it comes toprediction and control; its role should bespecified in ways that serve a functional anal-
278 PAULINE J. HORNE and C. FERGUS LOWE
ysis. Finally, (g) we have argued that equiva-lence theories have failed to account for lan-guage itself, or advance in any way Skinners(1957) exposition.
These are seven key issues that any theoryin this domain must address, and in our nam-ing account (Horne & Lowe, 1996; Lowe &Horne, 1996) we have attempted to do so.They expose what we believe to be anotherfatal flaw in equivalence theories, namely,that they fail to account for the data.
RELATIONAL FRAMETHEORY AND NAMING
In the previous section we listed seven phe-nomena that we believe pose major difficul-ties for equivalence-based theories, includingrelational frame theory. We have asked twofurther questions of the latter: What is thehistory that gives rise to a frame of coordi-nation? And how does the history work? Inour view, neither of Hayes commentaries se-riously address these issues. Hayes has, how-ever, raised questions for the naming accountto answer. What we aim to do here is to re-spond to the points raised in the Hayes andBarnes commentary and, in so doing, to takethe opportunity to compare naming and re-lational frame theory explanations of howclassifications such as same and different andmore and less are acquired. We also comparenaming with framing accounts of relationalframe theory research.
Stimulus Classes and Stimulus Relations
We first need to clarify a conceptual pointraised by Hayes and Barnes. They draw a dis-tinction between stimulus classes and stimu-lus relations and criticize Sidman and our-selves for focusing upon the former at theexpense of the latter. Their disenchantmentwith an approach that is primarily class basedseems motivated by a desire to distance them-selves from the concept of equivalence, whichthey consider to have collapsed. We find thisironic, given the centrality of that same con-cept to relational frame theory as it has beendescribed in the past. For example, Hayes(1991) has written of equivalence that it isthe most common and fundamental type ofverbal construction (p. 32) and, accordingto Barnes (1994, pp. 102103),
Hayes views equivalence as the most importantrelational frame because it is central to the oc-currence of referential relations in naturallanguage . . . and is therefore necessary beforederived relations can be learned. It would bequite difficult, for example, to teach a childthat hot is the opposite of cold if thewords hot and cold did not participate inequivalence relations with the physical eventsof hot and cold respectively (see Hayes &Hayes, 1989, pp. 174175).
How then, if the crumbling concept of equiv-alence is the very foundation on which therelational framing edifice has been erected,can the latter theory itself survive?
We think that this attempt to disavow theconcept of stimulus classes is mistaken be-cause, like Sidman, we consider that under-standing how organisms come to form cate-gories is central to understanding complexhuman behavior in general and language inparticular. To neglect the study of stimulusclasses would be to remove the foundationsfor any coherent theory of human behavior.In any case, stimulus classes do not form ofthemselves; on the contrary, in studying themwe are studying behavioral relations. And be-sides, as Barnes (1994) has noted (seeabove), without class-based concepts one can-not study derived relations.
How Do We Come to Classify Objects orEvents as the Same or Different?
The naming account differs radically fromrelational frame theory in its approach to thisquestion. But before we look in detail at thesedifferences we should outline how each ac-count approaches the question of how a ver-bal response comes to be related to objectsand events. In the naming relation, as wehave described it, each exemplar of a class ofobjects or events evokes a common verbal re-sponse (i.e., speaker behavior) that in turnevokes listener behavior (e.g., orienting to ex-emplars of the stimulus class). This bidirec-tional sequence of behavior can be initiatedin a variety of ways, including, for example,by the subjects seeing an object or event, orby her hearing the common verbal stimulus,spoken either by herself or by another. Oncesuch behavior (a naming relation) is estab-lished as a free-operant class, it occurs in var-ied contexts, its functions evolving throughinteractions with the environment. Through
279A THEORY OF VERBAL BEHAVIOR
naming, verbally able humans categorize fea-tures of their environment, thereby pro-foundly altering the ways in which they re-spond to it (Horne & Lowe, 1996).
Contrast this with the problems faced by achild who has to learn a frame of coordination,as posited by relational frame theory. In or-der to learn an arbitrarily applicable relation-al frame of coordination the child must bepresented with an object (e.g., a dog) and averbal stimulus (e.g., /dog/) in the presenceof a contextual cue (e.g., is called) thatspecifies that these two stimuli should be in-terrelated or treated as if they were the sameas each other. As we understand it, central tothe relational frame account is the notionthat it is the contextual stimulus exclusivelythat exerts control over the relational response;the latter is not occasioned by the formalproperties (or physical characteristics) of thestimuli to be related (see Hayes & Wilson,1996). Thus, in the above example of thedog, saming (or coordination) would beunder the exclusive control of the contextualstimulus is or is called. But, in that case,how would the child name objects in her en-vironment if there were no one there to pro-vide the contextual cue that controls this be-havior (e.g., that is called a dog)? And inthe absence of such a cue how would she,upon seeing a dog, ever say dog, and if shedoes not, where is the verbal stimulus to forma relation of coordination with the dog itself?Even if she were somehow to utter dog, re-lational frame theory would have it that inthe absence of the contextual cue the rela-tion of coordination would nonetheless fail tooccur. That is, although a child could learnto emit a vocalization upon seeing an object,the two events could not be further relatedwithout a contextual cue being present to de-termine which of the many possible frames(e.g., saming, moring, lessing, op-positing) might apply. In brief, without suchcontextual cues she would be unable to nameanything in her environment or, by defini-tion, emit any verbal behavior, but insteadwould respond at the level of a nonverbal or-ganism. This seems to us to be a serious flawthat runs right through the relational frameaccount; relational framing is dependentupon contextual cues that have to be provid-ed in order for the relating to occur.
There is another possibility: If the verbal
community had taught the child to say dogupon seeing a dog and to orient to dogs uponhearing the verbal stimulus /dog/, then thedog in and of itself would be sufficient to oc-casion the childs naming of it. The childwould thus be free, within the behavioralbounds set by the verbal community, toname, and this naming would influence oth-er aspects of her reacting to any stimulus inher environment. But this is, of course, notas relational frame theory would have it butrather is Horne and Lowes naming accountof the matter. The distinction between nam-ing and framing is nicely captured in a pas-sage by Mead:
The dog only stands on its hind legs and walkswhen we use a particular word, but the dogcannot give to himself that stimulus whichsomebody else gives him. He can respond toit but he cannot himself take a hand so tospeak, in conditioning his own reflexes. . . .Now it is characteristic of significant speechthat just this process of self-conditioning is go-ing on all the time. (1934, p. 108)
We would argue that to be bound by relation-al frames would leave one in no better posi-tion with respect to language than the dog inMeads example.
Naming, categorizing, and contextual control. Akey problem for all accounts in this area ishow categorizations or, to use Sidmans term,partitions, of arbitrary stimuli are established.For relational frame theory there is the par-ticular problem of how to account for con-textual control of such behavior. Within thenaming account, names are key determinantsof contextual control. For example, if grapes,apples, bananas, and oranges are the itemsencompassed by a childs name relationfruit, when she hears /pass me the fruit/ thisevokes her selecting, from an array of objectsthat includes many nonfruit items, only thoseobjects that feature in that name relation(i.e., grapes, apples, bananas, and oranges).Names thus directly specify object classifying(see Horne & Lowe, 1996, pp. 199208). Butit is not clear how relational frame theorycould account for such contextual control. Inorder for these items to be selected, the au-ditory stimulus /fruit/ would have to beplaced in a frame of coordination with eachof the fruit items. But which component ofthe instruction /pass me the fruit/ would serveto establish the frame? Given that the contex-
280 PAULINE J. HORNE and C. FERGUS LOWE
tual cue for coordination must be extrinsic tothe items to be coordinated, it could not bethe auditory stimulus /fruit/ itself. This leavesonly /pass/, /me/, and /the/, none of whichcould serve this function. We maintain thatin such cases, yet again, the contextual cuethat is vital to enable framing to occur cannotbe specified by the theory.
But Children learn not only to name objects andevents but also their properties. As Skinner(1957) described, the naming of a propertysuch as redness is established by members ofthe verbal community teaching a child to sayred in the presence of a variety of red ob-jects for which the only reliable accompani-ment is that property, and, similarly, to re-spond appropriately to /which is red?/questions. Having learned to name a numberof stimuli within the red wavelength range asred, the child is able to extend this cate-gorizing name to all red objects, irrespectiveof their other characteristics. In like fashionshe learns to name other colors, shapes, sizes,numbers, and so on. Such a repertoire ofproperty names then provides her with anenormous variety of contextual possibilities;the same set of objects may be categorized(or partitioned) in many different ways, de-pending on the name she (or someone else)applies to them: red, sharp, smooth,big, noisy. In this sense, any object mayparticipate in not one but many name rela-tions and thus may be contextually related toothers in a variety of ways through the oper-ation of any one, or some combination, ofthose name relations. When a new functionis acquired for any member, it may transferto other members of the class in accordancewith the particular name relation concerned(see Horne & Lowe, 1996, pp. 204207). Thisstraightforward account of how childrenlearn to categorize objects in terms of theirproperties contrasts with that of relationalframe theory, which, we have argued, hasconsiderable difficulty in explaining suchcontextual control.
Naming same and different. At aroundthe time when they have acquired about 50name relations, children are spontaneouslyable to sort on the basis of identity; they showthis before they begin to show intraverbalnaming or learn to produce the namesame (Gopnik & Meltzoff, 1992). With re-spect to oddity, several studies of infants ori-
enting behavior have demonstrated their abil-ity to attend selectively to a novel comparisonobject that differs from a previously present-ed sample (e.g., Cohen & Strauss, 1979; butsee Thompson, 1995). These early identityand oddity discriminations may serve as thebasis for childrens learning to name objectsor events as same or different. But theirlearning of those names is essentially no dif-ferent from their learning to name chairor red (Skinner, 1957).
Naming our naming. Skinner (1957) madean important advance in proposing that,Once verbal behavior has occurred and be-come one of the objects of the physical world,it can be described like any other object (p.319). This is the basis of his account of au-toclitic behavior. It follows from our accountof naming also that the objects that humansname include the naming responses them-selves as well as the objects that first occasionthose responses. Thus, for example, the care-giver may say to the child, Youre called Joand she [pointing to another child] is alsocalled Joyou both have the same name.Following repetitions of such episodes, whichmay also often occur in relation to items thatare physically similar, the childs verbal re-sponse same may at times come to be oc-casioned whenever she emits the same namefor two or more objects. Just how differentthe name relation same is from supposedlyanalogous behavior established in nonverbalorganisms is seen in work reviewed byThompson (1995). He argues that the test ofwhether responding is truly relational andcompletely free from control by absolute cuesis that subjects should be able to match stim-uli in a match-to-sample task when the onlybasis for matching is the relation between twoelements of a sample stimulus compound(e.g., AA) and the relation between anothertwo elements within each of the comparisonstimuli (e.g., BB vs. CD). If subjects can learnto respond to the sameness of the relationthat exists in Sample AA and Comparison BB,they should then be able to learn to respondcorrectly with novel samples (e.g., FF) andcomparisons (e.g., GG as correct and KN asincorrect). Similarly, if presented with KM assample, the subject should select the correctcomparison PQ and not the incorrect com-parison, SS. Oden, Thompson, and Premack(1990) found that even chimpanzees that
281A THEORY OF VERBAL BEHAVIOR
showed evidence of generalized identity andoddity matching failed on such a task de-spite heroic training efforts on the part ofthe experimenters (p. 211). Whatever the ba-sis of the chimps discriminative respondingin generalized identity tasks, these animalswere apparently unable to respond to that re-sponding as a way of performing correctly onthis relational task.
There are, however, a variety of ways inwhich linguistically competent humans couldpass Thompsons acid test of truly relation-al responding. One of the ways may be il-lustrated as follows: In an identity matchingtask, the sample AA may evoke in a 5-year-oldchild the basic level naming responsesame, as may also the comparison BB. Thatis, when she looks at the sample and correctcomparison, a child may say same andsame in turn, or that is the same andthat is the same. But because the child canrespond to her names as objects, she may nowrespond to these two consecutive same re-sponses with the second-order naming re-sponse same, listening to which she maynow orient again to the sample and correctcomparison (see Horne & Lowe, 1996). Re-inforcement for selecting BB would alsostrengthen the naming of AA and BB sameresponses as same and ensure that this lat-ter second-order name response would be-come discriminative for selecting the correctcomparison in the future. In similar fashion,after looking at the sample AA and sayingsame and then the incorrect comparisonCD and saying different the child, respond-ing to these consecutive basic level namessame and different as objects, may thenemit the second-order name different.Novel configurations (e.g., FF as sample, GGand HJ as comparisons) could, in subsequenttests, easily be encompassed by the childsnaming same and different at both basicand second-order response levels. And in anidentity matching task in which the stimuluscomponents of the sample are different (e.g.,where KM is the sample and PQ is the cor-rect, and SS is the incorrect, comparison),the child may respond to herself saying dif-ferent different to the sample and correctcomparison, respectively, with the second-or-der name same, and thus achieve successon the task. A similar account could also be
applicable to her performance on compara-ble oddity matching tests.
A human subjects naming of name rela-tions is thus one of the ways that could ensurehis or her successful performance on abstractidentity tasks. Other verbal strategies (e.g.,intraverbal naming) may be equally success-ful. However, that it is certainly verbal behav-ior of some sophistication that is necessaryfor such success (and see Thompson, 1995)is indicated by the fact that children do notgenerally succeed on tasks of this kind untilthey are about 5 years of age (House, Brown,& Scott, 1974; Lowenkron & Colvin, 1992).
Naming more and less. As is the case fora childs naming of same and different,learning to name more and less occursover a long period. More, although it isone of the first utterances of the young child,initially functions simply as a mand (e.g., asin more milk, more toys) (Bloom, 1970;Brown, 1973; Weiner, 1974). Gathercole(1979, and see 1985) has reviewed 24 studiesconcerned with the development of appro-priate responding to more and less. Ac-cording to the findings of these studies, it ap-pears that young children show a generalnonlinguistic response bias in favor of addingitems to a stimulus array, or of choosing thegreater of two or more stimulus arrays. Thisearly preference, of course, is likely to pro-vide good opportunities for teaching a youngchild the name more. In simultaneous dis-crimination tasks in which they are presentedwith two unequal stimuli and are askedwhich one has more? young children ini-tially learn to respond to more correctlyonly within one particular stimulus dimen-sion at a time. For example, the childs learn-ing to respond to the longer of any pair ofstimuli does not entail that she will also beable to respond to more correctly in rela-tion to stimuli differing in terms of, for ex-ample, number, volume, or area. However,Gathercoles review shows that by the timethey are 5 years old, children have generallylearned to respond to more correctlyacross a range of dimensions. Their respond-ing with comparable accuracy to less, onthe other hand, lags a long way behind, oftenby more than 2 years (see Palermo, 1973,1974). This basic asymmetry in respondingcorrectly to more and less argues strong-ly against any account of such behavior in
282 PAULINE J. HORNE and C. FERGUS LOWE
terms of mutual entailment; to the young child,being able to correctly select a stimulus thathas more entails nothing about selection ofthe alternative stimulus as less. These de-velopmental data show that even for nonar-bitrary stimulus dimensions, Hayes accountof the relational behavior of moring andlessing is unconvincing. The evidence rath-er is that children first learn a unipolar namerelation, more, in which a great variety ofstimulus dimensions (e.g., length, number,volume, area, brightness, loudness) evoke thename more. Once they have learned thisrelation, it is possible for the verbal commu-nity to instruct them to emit less in re-sponse to the stimulus that is the alternativeto that named more. This then enableschildren to respond appropriately to lessacross a range of novel contexts and stimulusdimensions for which they had previouslylearned to respond correctly only to more.
Nonarbitrary transitive responding. Once chil-dren have learned to name more andless appropriately, do they go on to maketransitive inferences about relations among a se-ries of objects that vary in their properties ordimensions? For example, according to rela-tional frame theory, a child who has seenpairwise combinations of different coloredsticks that vary in length (e.g., Stick A . B, B. C, C . D, D . E) should, on the basis ofcombinatorial mutual entailment and withoutthe opportunity for direct visual comparison,be able to infer the relation between novelpairwise combinations (e.g., BD) correctlyand so say which one is longer than the other.However, this prediction is not confirmed. Instudies conducted by Chapman and Linden-berger (1988), for example, it was found thatonly 5% of 7- to 9-year-old children respond-ed correctly on such tasks. Clearly, even withreal-world objects and late into childhood,children show little evidence of any facilitywith relational responding that has the prop-erty of combinatorial mutual entailment.
Making Sense of Relational Frame Research
Hayes and Barnes have chided us for notresponding adequately to Hayes (1996) chal-lenge that we explain the Steele and Hayesdata. Because it provides yet another oppor-tunity to contrast the respective explanatorymerits of the relational frame theory andnaming accounts, we shall here consider
some of the core features of relational frameexperiments and, in particular, the study bySteele and Hayes (1991).
Establishment of contextual cues. A centralcharacteristic of the procedures used in theserelational frame experiments (see also Roche& Barnes, 1996) is that they aim to establishnonlinguistic stimuli as contextual cues forresponding in accordance with such arbitrar-ily applicable relations as larger andsmaller. This is done by reinforcing sub-jects selection of, for example, the larger oftwo or more stimuli, the greater number, thelonger, and so forth, in the presence of pu-tatively nonlinguistic contextual cues. Asimilar procedure is employed for establish-ing cues relating to same, different, oropposite. But after such training can it beassumed that these cues are nonlinguistic? Allthe subjects in these relational frame studiesare linguistically competent and, like R. Saun-ders (1996), we consider it far more likelythat the cues come to be incorporated withinthe subjects existing name relations ofmore, less, same, different, andopposite as occasioned by particular stim-ulus configurations in the experiment. Thisraises the further question of why the re-searchers do not explicitly name the cues(e.g., as more, less, etc.) in the first in-stance (see R. Saunders, 1996). This has, in-deed, been recognized as a possible strategyby Steele and Hayes (and see Barnes &Roche, 1996). But, we maintain, if nameswere indeed provided it would make explicitwhat is already implicit in these proceduresas presently conducted; that is, that the con-textual cues derive their functional controlfrom preexisting naming repertoires. Giventhat with verbally sophisticated subjects onecannot be sure that one has taught any newbehavioral relations by means of such cue-es-tablishing procedures, for the relationalframe theory to be at all convincing theseprocedures need to be conducted with non-verbal subjects including young children. Insuch circumstances, however, we would con-fidently predict failure on tests of derived re-lations.
Testing-induced framing. Another feature ofthese procedures is that testing for derivedrelations is continued until the subjects suc-ceed; that is, until they respond in accor-dance with the experimenter-defined stimu-
283A THEORY OF VERBAL BEHAVIOR
lus classes. This means that many test sessionsare often required to establish each derivedrelationup to 35 sessions in the case of theSteele and Hayes (1991) study. We have al-ready argued, as have R. Saunders and Green(1996), that such testing-until-success strate-gies, when used with verbally sophisticatedsubjects, cannot but provide implicit instruc-tions and reinforcement for responding thatconforms to the experimenter-defined class-es. On these grounds also, therefore, it seemsunwise not to take account of verbal controlfactors when interpreting the findings of re-lational frame studies.
Difficulty in learning relational framing. Ac-cording to Hayes and Barnes, the relevant ar-bitrarily applicable modes of responding re-lationally (moring, lessing, oppositing,and differencing) are established early inchildhood. This should indeed be the case if,as relational frame theory suggests, they arebasic verbal processes. One would thereforeexpect well-educated adolescents and adultsto have little difficulty performing appropri-ately on relational frame procedures. The re-ality, however, is that it takes a very long timeand much effort to establish such perfor-mances and, even once established, correctresponding is fragile. For example, in Dy-mond and Barnes (1995, Experiment 2) the2 undergraduate subjects needed up to 800trials to reach the first criterion for baselinerelations. Both initially failed the tests for thederived relations. Only after an extended os-cillation between being retrained in baselinerelations followed by failing tests of derivedrelations did both eventually succeed; one ofthem required 1,280 baseline training trialsbefore reaching the derived relation criteri-on. Relational frame theory cannot accounteither for these failures or for why, if they aresupposed to be verbal processes, these rela-tions should be so extremely difficult to es-tablish. (Even 2-year-olds can learn entirelynew name relations within a few presenta-tions of a novel arbitrary stimulus and a novelobject; see Baldwin, 1991; Nelson & Bonvilli-an, 1973.) We agree with Barnes and Roches(1996) comment about such studies: If therelational pretraining did not readily producearbitrarily applicable relational responding inverbally sophisticated adults, RFT would be invery serious trouble as an account of humanverbal behavior (p. 492).
Why are stimuli within a frame of coordinationnot equally substitutable? Although the data arenot discussed, Steele and Hayes (1991) re-corded subjects reaction times on test trials.There was a strong positive correlation be-tween duration of reaction time and the com-plexity of the conditional stimulus relationstested in the probes (see also Wulfert &Hayes, 1988). But given that the stimuli with-in a frame of coordination (i.e., equivalence)are equally substitutable, then why shouldsubjects take longer to respond on symmetrythan on combined symmetry and transitivitytrials (see also Spencer & Chase, 1996)? In-deed, if measures of response accuracy andlatency do not covary on tests of frames ofcoordination or equivalence, then which, ifany, is the true measure of these concepts?These are central questions which, to date,have not been answered by either Hayes orSidman, nor have they accounted for thefinding of Steele and Hayes and other studiesthat responding to supposedly equivalentstimuli differs systematically as a function ofnodal distance or complexity. Such resultsare, however, readily explained within a nam-ing account (see Horne & Lowe, 1996, pp.235, 237; Lowe & Horne, 1996, p. 333).
Rule-governed framing. The performance of1 subject in the Steele and Hayes (1991)study failed to satisfy the relational framingcriteria. This subjects verbal reports clearlyindicated that he had formulated rules for re-sponding that ran counter to those requiredfor successful entailment of relations estab-lished in baseline. Although he was the onlysubject for whom Steele and Hayes provideverbal reports, his account also supports ouroriginal suggestion (Lowe & Horne, 1996, p.333) that much of the behavior observed inthis study, including that which conformed torelational framing tests, was controlled bysubjects verbal formulations and rules. Butthis raises the question of whether there aretwo kinds of equivalence performanceonethat is rule governed and one that is contin-gency shapedand, if so, how these differentkinds of equivalence come about. Neither re-lational frame theory nor Sidmans theoryhas addressed this issue, which is central tobehavior analysis.
ConclusionWe find the objectives of all these experi-
ments on relational framing, including the
284 PAULINE J. HORNE and C. FERGUS LOWE
new framing procedures outlined by Hayesand Barnes, to be obscure. As R. Saunders(1996) has observed, That newly trainedsymbolic substitutes for the words same, oppo-site, different control responding to pairs ofstimuli . . . is not unexpected; elementaryschool children perform this feat daily in theclassroom (p. 486). And it is perplexing thatin devising their studies the relational frametheory researchers should go to such lengthsto avoid the inclusion of explicit verbal cuesas to render it very difficult indeed for sub-jects to perform the feats required of them.What such language-avoidance research withhumans could possibly be expected to tell usabout language, logic, or learning in generalis most puzzling of all.
With regard to relational frame theory it-self, we can see little to recommend it. Itseems to have many of the flaws of the equiv-alence account upon which it is based, but itextends and amplifies some of the most prob-lematic features of the original. Instead ofjust one central abstractionequivalencethere are several different kinds of framesthat intervene between behavior and the en-vironment, described in only the most ab-stract of terms. The considerable gulf that ex-ists between this theory and the empiricalevidence on relational learning in childrenonly further undermines its plausibility.
DO NONVERBAL ANIMALSPASS EQUIVALENCE
Schusterman, Kastak, and Reichmuthscommentary indicates that they have misun-derstood our views on the relation betweenthought and language. We do not assumethat thought is dependent solely on words orthat the word is the sole sign of thought(p. 252). As we indicated in our original pa-per, our views on this issue are closely alignedwith those of Skinner, Vygotsky, and Mead,none of whom provided what could be mis-taken for Cartesian or non-Darwinian ac-counts. We recognize that there must be aclear continuity of those behavioral processesthat might be termed thinking from nonhu-man to human animals, but we also believethat there are discontinuities, particularly in-sofar as language affects human behavior(Lowe, 1983; Lowe & Horne, 1985; Lowe,
Horne, & Higson, 1987). Like Skinner, Mead,and Vygotsky, we propose that the form ofbehavior that we term verbal thinking (Horne& Lowe, 1996) is unique to the human spe-cies, and unsurprisingly, perhaps, we considerlanguage to be a sine qua non for that be-havior. We have consistently maintained thatto ignore this qualitative difference is to missa central aspect of human behavior, as well assome of the central tenets of radical behav-iorism.
Also contrary to Schusterman et al.s un-derstanding of our position, we have not triedto account for equivalence; given that theconcept seems to us to lack coherence itmakes little sense that we should try to ex-plain it. Our interests in this domain focusinstead mainly upon the symmetry and com-bined symmetry and transitivity behavioral re-lations that are commonly observed in hu-man performance in match-to-sampleprocedures. And our concern here is withwhether language is necessary to generatesuch performances. Our position is that wecan see how verbally competent humanscould pass these tests, but we accept that an-imals might also succeed. If they did, howev-er, we would need to know how, because, as-suming that the test procedures were free ofartifacts, we would maintain that such behav-ior could not be predicted from animal learn-ing principles as presently known. Whateverfindings finally emerge, our main concern isthat success on these tasks should not be con-fused with language itself.
The Schusterman and Kastak (1993) study. Asa possible demonstration of stimulus equiva-lence in animals, many would argue that theSchusterman and Kastak (1993) study of a sealion, Rio, has been the only real contender(see Lowe & Horne, 1996, p. 331). Clearly,Schusterman et al. are in little doubt aboutits validity. But, particularly in the light of newexperimental evidence, we are highly skepti-cal of this conclusion. It is important to re-consider in some detail the procedure thatwas used, not least because many of its fea-tures are repeated in the new studies of sup-posedly derived relations outlined in theircommentary. In Figure 1 we have attemptedto illustrate the stimulus configurations seenby Rio on training and test trials. For purpos-es of illustration we consider only one of the30 stimulus sets, Set 16, but the same analysis
285A THEORY OF VERBAL BEHAVIOR
Fig. 1. Examples of the stimulus configurations seen by the sea lion in the Schusterman and Kastak (1993) study. Thisshows stimuli from one of the 30 stimulus sets (Set 16, labeled here A, B, and C) and a selection of the negative comparisonstimuli, which are not labeled. The top left panel presents four of the 29 three-stimulus arrays used during AB training;comparable arrays for BC training are shown in the top right panel. The center panels show the symmetric compoundstimuli, consisting of AB/BA pairs (left) and BC/CB pairs (right) that may have been established during training. Thelower left panel reorders these compound stimuli to show how transitive stimulus compounds may also have beenestablished. Examples of the stimulus arrays presented in CA testing are shown in the lower right panel.
286 PAULINE J. HORNE and C. FERGUS LOWE
applies to the remaining 29 sets. Figure 1shows Set 16 stimuli (labeled A, B, and C)and a selection of the negative comparisonstimuli (not labeled) that were presented dur-ing training and testing. The top left panelshows four of the 29 three-stimulus arraysused during AB training, in each of which thecorrect comparison, B, consistently appearedeither on the left or on the right of SampleA. Note that the incorrect comparison stimuliwere never the same from trial to trial. Dur-ing AB training, Rio was not required to ap-proach or otherwise respond to the sample,so each trial essentially consisted of the pre-sentation of a three-element array, two ele-ments of which were consistently paired whilethe third was consistently varied. Thus, whenRios responding to B was reinforced this al-ways occurred in the presence of the adjacentA and of no other reliable accompaniment.The same was true for BC training, as shownin the top right panel.
Schusterman et al. might wish to considerhow similar their procedure is to that em-ployed in a recent series of studies by Sie-mann, Delius, Dombrowski, and Daniel(1996). In the first of these experiments, pi-geons responding on four successively pre-sented target stimuli was reinforced withgraded amounts of reward and penalty(A11, B1, C2, D22). These target stimuliwere accompanied by four different neutralstimuli (Na, Nb, Nc, Nd); thus, Na accom-panied A11, Nb accompanied B1, Nc ac-companied C2, and Nd accompanied D22.Responses to the neutral stimuli had no con-sequences. After discrimination of the targetstimuli had been established, tests of the neu-tral stimuli revealed a graded preference in-creasing from Nd to Na. A second experi-ment used two target stimuli, A1 and B2,accompanied by two neutral stimuli, Na andNb, respectively. The pigeons developed avery strong preference for Na over Nb, rang-ing from 92% to 100% in unreinforced test-ing. To ensure that these effects were not dueto adventitious reinforcement of operant re-sponding on the neutral stimuli, a third ex-periment showed similar strong effects evenwhen the opportunity to respond on the neu-tral stimuli during training was removed. Ifone considers the Schusterman and Kastakstudy in the light of this evidence (see alsoWynne, 1995, 1997), one observes that on all
Rios AB training trials the centrally posi-tioned Stimulus A was, effectively, a neutralstimulus accompanying the outer situatedStimulus B on which responding was rein-forced. Even though there were never anyscheduled consequences for responding to A,following Siemann et al., we would expectthat if Rio were provided with the opportu-nity to respond on A, that is, if A were situ-ated in either of the outer positions (re-sponding to the central location was neverreinforced), then the animal would do so. Be-cause on other trials A was presented withother stimulus pairs as an incorrect compar-ison, responding would occur on A only inthe presence of B. Thus, by the end of train-ing, the stimulus configurations AB (i.e.,where B is to the right of A) or BA (i.e.,where B is to the left of A) would be discrim-inative for responding on either B or A, de-pending on which occurred in the outer lo-cation. The same analysis applies to controlby the BC/CB stimuli. It follows that if Riowere tested for symmetry with these stimuliand B were presented in the center with A asthe correct outer stimulus, then, again fol-lowing Siemann et al., we would expect re-sponding to be directed at A. This wouldyield success on any symmetry test BA. Thesame would be true for any CB symmetry testfollowing BC training.
The argument that the AB pair also func-tioned as a compound Pavlovian conditionedstimulus (CS1; see center panels of Figure 1)is based on the observation that AB was reli-ably followed by delivery of food and soshould have become a strong appetitive CS1.The same was true of BC. Because B was com-mon to both the AB/BA and BC/CB com-pound stimuli, it is reasonable to assume thatby the end of the AB and BC baseline train-ing, any combination of A, B, or C (see lowerleft panels of Figure 1) would have func-tioned as a compound appetitive CS1. Giventhat A shared both the discriminative andCS1 functions of B and that B shared thediscriminative and CS1 functions of C, it fol-lows that whichever two of these three stimuliwere present, the animal should respond tothe outer element. Consequently, in equiva-lence testing (see Figure 1, bottom right pan-el), when presented with the AC/CA com-pound Rio should respond to its outerelement, A, and not to the other stimulus,
287A THEORY OF VERBAL BEHAVIOR
which would never previously have been pre-sented with AC or CA. Thus, our conclusionis that simpler conditioning principles thanequivalence can account for Rios perfor-mance (and see Horne & Lowe, 1996, pp.223224; Lowe & Horne, 1996, pp. 330331).
Other sea lion studies. We here briefly con-sider the previously unpublished experimentsto which Schusterman et al. refer, althoughin the case of these, even more than for theearlier study, we lack important proceduralinformation. The first of these experimentswas designed to investigate transfer of dis-criminative function within the three-elementstimulus sets employed by Schusterman andKastak (1993). We believe that our previousinterpretation of that study, showing as itdoes how stimulus compounding and transferof discriminative control across elementswithin the compound were likely to have oc-curred, readily accounts also for the outcomeof this transfer experiment, thus obviatingany need to appeal to equivalence as an ex-planation.
In their second experiment, which had assubjects 2 sea lions, Rio and Rocky, the ex-perimenters aim was to establish functionalclasses using contingency-reversal procedures(Sidman, 1994, pp. 451453). Rio was thentested for the emergence of equivalence be-tween functional class members. Although itis difficult to determine whether they applyto this particular study, Sidman (1994) hasmade some interesting observations aboutacross-trial discrimination learning duringsuch procedures to the effect that when si-multaneous discrimination performance ap-proaches criterion levels, then from one trialto the next, Set 1 stimuli will invariably beselected and reinforced, and thus conditionalsequences between Set 1 members may be re-inforced. Moreover, because the order of pre-sentation of Set 1 stimuli is randomly varied,all possible conditional relations between Set1 members can be learned in this way. Whenthe contingency is reversed, and selection ofSet 2 stimuli in the simultaneous discrimina-tion task is reinforced, all possible relationsbetween Set 2 stimuli can be learned in sim-ilar fashion. With each reversal, such adven-titious conditional discrimination learningbetween set members would be strengthened.In this experiment conducted by Schuster-man there were, for each set, 90 possible con-
ditional relations to be learned. Each sessionconsisted of 40 randomized trials, so Rio hadample opportunity to learn these across-trialrelations for both stimulus sets. If correct,Sidmans interpretation would account bothfor the establishment of the functional classesthat occurred in this study, and, because allof the stimulus relations had already beentrained, for Rios above-chance match-to-sam-ple performance.
The final phase of this experiment againconcerned only Rio. Bidirectional relationswere trained between a new stimulus, K, andone of the Set 1 stimuli, J, and between a newstimulus, 11, and one of the Set 2 stimuli, 10.When Rio was tested for bidirectional condi-tional relations between K and all other Set1 members, and between 11 and all other Set2 members, she passed all of the tests. How-ever, given, as has already been suggested,that all of the bidirectional relations betweenstimuli in each of the two sets had alreadybeen trained in the previous two phases,then, in training the bidirectional link be-tween K and J, the K stimulus should becometransitively related to all of the Set 1 stimuli.The JK compound trained during this phasewould have an element, J, in common withthe stimulus compounds trained in the pre-vious phase (i.e., AJ, BJ, CJ, DJ, and so on).These are just the conditions that we havealready argued are sufficient to establish re-lations between A and K, and indeed betweenK and all other Set 1 stimuli. The same anal-ysis applies to the relation of Stimulus 11 toSet 2 stimuli.
As Sidman has ably demonstrated on manyoccasions (see Sidman, 1994), knowledge ofthe procedural details of match-to-sample ex-periments, particularly those conducted withnonhuman animals, is crucial for accurate in-terpretation of data. In the absence of suchinformation about these experiments, any in-terpretations of the findings, including ourown, must remain provisional. Furthermore,in all but one of them Rio alone participatedas subject; for some reason neither Rocky norany other sea lion featured in any of the cru-cial tests. We would also need to know moreabout the role of the set-specific reinforcers(a different type of fish was used for each set)that were provided in three of the four ex-perimental procedures; one would need tobe certain that the possibility of inadvertent
288 PAULINE J. HORNE and C. FERGUS LOWE
cues (e.g., olfactory) was excluded. If therewere such cues, of course, there could be aneven simpler interpretation of the findingsthan the one we have offered here. Such is-sues, however, can be straightforwardly re-solved by the provision of more detailed in-formation, and what is most important, byreplication of the effects using other subjectsand appropriate control procedures.
Equivalence in other species. Whatever thecase with Schustermans own experimentalwork, there is clearly a problem of overinter-pretation of data in his discussion of otheranimal literature. In the general quest forderived relations that has taken place overrecent years, one of the distinctive merits ofSidmans approach has been the care he hastaken to ensure that the stimulus relationsunder investigation could not have been es-tablished by preexperimental histories, onto-genetic or phylogenetic. Given the least rigorin this regard and a basic application of LloydMorgans (1894) canon, it seems that noneof Schustermans examples of equivalence inthe behavior of animals in natural social set-tings could pass muster. First, the example ofcross-modal equivalence in pinnipeds can beexplained more simply in terms of behaviorbeing occasioned by multisensory stimuluscompounds; there is, accordingly, no emer-gent behavior to account for. Second, the no-tion that when vervet monkeys hear an absentjuveniles scream and then look at its mother,there exists an equivalence relation betweenthe scream, the juvenile, and its mother,might have some credibility if the juvenilewere never seen in the company of its motherand she never responded to its screams. Butthis seems unlikely. Similarly, there are nogrounds for invoking equivalence to explaingroup aggression in vervet monkeys. Also,vervet monkeys responding to acousticallydifferent calls undoubtedly has a major phy-logenetic component. Once again, there isno basis for invoking derived relations suchas equivalence in this case. Indeed, in his re-view of the literature, Thompson (1995), cit-ed by Schusterman, says of overinterpreta-tions of these alarm call studies, that Theprocesses need be no more complex thanthose involved in associative conditioning(p. 207). Finally, appealing to equivalence asa factor in monkeys social relationships (Das-ser, 1988) is equally unfounded, because, as
Thompson (1995) has observed, such behav-ior is almost certainly mediated by commonfunctional associations such as temporal/spa-tial proximity and interactive outcomes (p.206).
Do We Need a Molar Account of Behavior?
Tonneau and Sokolowski consider ouranalysis of stimulus equivalence to be largelycorrect. However, in their own contributionthey reveal some fundamental misconcep-tions about our account as well as, it wouldseem, about some of the basic phenomenathat we deal with. To take a minor issue first,we do not attribute the failure of equivalenceexplanations to the use of match-to-sampleprocedures. Our point is simply that if themain goal is to understand how language islearned and has its effects on other behavior,then one should not be bound by a rigid ad-herence to match-to-sample procedures. It isalso not the case that the core of the namingaccount lies in intraverbal behavior. On thecontrary, our basic unit of verbal behavior isthe name relation. Our account accordinglypays particular attention to the powerful andpervasive role of common naming in the clas-sifying of objects and events, both in the con-text of match-to-sample and other categori-zation studies and in the world at large. Butwe have also been concerned to show thatnames evoke names (as in intraverbal behav-ior) and that names themselves can benamed (as in autoclitic behavior) and that,between them, intraverbal and autoclitic be-havior can give rise to an almost infinite va-riety of rules for responding in match-to-sam-ple tasks. For example, the reversibilityinherent in intraverbal name sequences (seeHorne & Lowe, 1996) may become the oc-casion for autoclitic responses such as greenmeans up (see above and Skinner, 1957). Allthese variants of verbal behavior are, however,basically founded upon the name relationfunctioning either as a common name or asindividual names for the stimuli concerned.
Of more importance for Tonneau and So-kolowskis own thesis is their mistaken beliefthat we maintain that the behavior of non-verbal animals should . . . conform to stan-dard molecular Pavlovian and operant laws(p. 267). The problematic words here arestandard and molecular. We do not havea rigid commitment to well-established, as op-
289A THEORY OF VERBAL BEHAVIOR
posed to novel or nonstandard, accounts;nor have we any particular prejudice in favorof molecular over molar accounts of behav-ior. Indeed, in our treatment of the animalliterature in this area (see above) we havetried to incorporate some of the recent re-search and thinking about new behavioralphenomena, including the complexities thatresult from interactions between operant andPavlovian conditioning. We have tried to re-flect these complexities in our account of thename relation itself (see Horne & Lowe,1996).
We have already dealt with some of the mis-understandings concerning whether ours is amediational account (see Naming and Medi-ation, above). We also need to point out, how-ever, that Tonneau and Sokolowskis enthu-siasm for a languageless molar account,which they illustrate with Hollands (1990)mediational theory, appears to be foundedupon a mistake as to what constitute the be-havioral criteria commonly taken to indicateequivalence. They assume that having trainedAB and AC to meet the criteria for stimulusequivalence, it is sufficient to show that BChas emerged. But they thus miss what is per-haps the central imperative in this researcharea, namely the necessity of demonstratingsymmetry. To demonstrate that the relationsare symmetric rather than merely transitive,it is essential to show not merely the emer-gence of BC but also of CB, BA, and CA. Itis these symmetrical relations that are themost elusive in the animal literature. And giv-en that existing learning principles, even bol-stered by Hollands (1981, 1990) mediationaltheory, cannot account for them, this shouldcome as no surprise. It should be noted, how-ever, that a molar explanatory system basedon Hollands model would be unable to ac-count for the phenomena that seem to be ofcentral concern, that is, emergent relationson tests of stimulus equivalence. From ABand AC trained relations, Hollands modelyields BC but not CB, BA, or CA. (To be fairto Holland, he does not claim that his modelaccounts for all of these emergent relations.)Invoking other studies that have appeared toshow backward conditioning effects (albeitsmall and fragile), usually with food or someother unconditioned stimulus as one of thestimulus elements, does not counter the evi-dence that, as yet, there has been no reliable
demonstration of even these simple symmet-ric relations in animals other than humans.We propose, of course, that it is the bidirec-tionality inherent in verbal relations such asnaming that enables humans to readily pro-duce bidirectional performances in match-to-sample tests.
To conclude, (a) it is not possible to extendthe naming account to the behavior of otheranimal species because, so far, it has not beenshown that they can learn to name objectsand events, and (b) if by a molar account,Tonneau and Sokolowski mean a return to asimple methodological behaviorism in whichthe complexities of verbal behavior are ig-nored, and even simple phenomena like sym-metry cannot be explained, then we are cer-tainly less than enthusiastic about itsadvantages.
How is naming established? Lowenkron aloneamong the present commentators does notaddress equivalence. Instead he focuses uponwhat we ourselves regard as a much more im-portant issue, that is, how the basic unit ofverbal behavior that we term naming is estab-lished. Although in all other respects heseems to be in accord with our approach, hedoes perceive problems in our description ofthe name relation. He thus proposes an ad-ditional factorjoint controlthat is a behav-ioral process supplementary to those we de-scribe; the inclusion of this process hebelieves would deal with the apparent diffi-culties in our account and would thus en-hance its plausibility and generality. We aregrateful to Lowenkron for his careful scrutinyof the detail of our explanations; part of thepurpose of this project was certainly to iden-tify problems and solutions where they exist-ed. However, in this particular case, we do notsee that there is a problem to be accountedfor, nor do we see the need for insertion ofanother behavioral process. Indeed, in ourview, Lowenkrons joint control solutionwould introduce insoluble problems.
As Lowenkron understands it (see Lowen-kron, 1996), the difficulty lies in that part ofour account where, the basic name relationhaving been established, a new object is pre-sented to the child and is named by the care-giver; the child not only names this object but
290 PAULINE J. HORNE and C. FERGUS LOWE
also emits appropriate listener behavior in re-sponse to the utterance (i.e., looks at the ob-ject or reaches for it). Not being able to iden-tify any source of differential reinforcementfor such listener behavior in this context,Lowenkron is puzzled as to how it comes tobe established when only the tact relation(i.e., objectutterance) has been trained. Inother words, if one trains only speaker be-havior, how does listener behavior emerge?Interestingly, the converse process, where lis-tener behavior to a name is trained but speak-er behavior (i.e., the tact component) emerg-es, is not viewed by Lowenkron as a problem(see also Michael, 1996).
In view of Lowenkrons return to this topic,it seems that we did not devote sufficientspace in our last reply to dealing with it. It is,we believe, an entirely straightforward matter.In our account we describe how, prior totraining any verbal productions, caregivers es-tablish an extensive repertoire of listener be-havior in the young child and then an echoicrepertoire. Naming is established by the care-giver (a) pointing to (or showing) and look-ing at the object to be named, (b) saying thename of the object, (c) asking the child toecho the name, and (d) reinforcing appro-priate listener behavior and verbal produc-tions (e.g., saying the word while looking atthe object). The vital feature here is that boththe childs speaker behavior (i.e., her utter-ance of the word) and her listener behavior(i.e., her looking at or reaching for the ob-ject) are differentially reinforced; saying,without looking at the object, is not normallyreinforced, and neither is looking at the ob-ject without saying. Many names are estab-lished in this manner, so that the child is pro-vided with an extensive history ofreinforcement for appropriately conjoinedsaying and looking. Most initial names arebased upon existing listener behavior reper-toires, with speaker behavior lagging well be-hind comprehension up to and beyond 2years of age (Baldwin, 1991; Benedict, 1979).
Now we come to the seat of Lowenkronsproblem, which he locates in what he termsobjectname, or tact, training. Let us lookclosely at what actually occurs in the courseof such training, for example, when a newobject (a bell, say) is introduced to the child.The caregiver says, This is a bell. Can yousay bell? What is it? The childs previous
name training (see above) will have estab-lished that she behave along the followinglines: (a) Upon hearing the novel verbal stim-ulus /bell/, she orients first to the caregiverand then to the bell that the caregiver is look-ing at. This listener responding, because ofthe extensive listener repertoire she has al-ready acquired, occurs very reliably, ofteneven in the absence of explicit reinforcement(Baldwin, 1991). (b) As she looks at the bell,she says bell. This part of the sequence es-tablishes the discriminative control of the bellover her saying bell where previously it wasonly an echoic. The caregiver then saysGood girl! Yes, its a bell! and thus rein-forces the entire behavioral sequence, fromher hearing the name stimulus, to seeing theobject, to echoing while looking at the object,and once more hearing the name. The prob-lem Lowenkron perceives (namely, how oneaccounts for listener behavior when the care-giver apparently only trains the tact or objectname relation) thus in reality simply does notexist. In reinforcing the would-be tact, care-givers at one and the same time reinforceboth echoic behavior and appropriate listen-er behavior. As we have previously argued(Horne & Lowe, 1996, p. 200), tacts are arare feature in child behavior precisely be-cause, in the normal course of events, asspeaker behavior is trained, listener behavioris simultaneously acquired, and both functiontogether as the name relation.
There is thus, in our view, no hiatus in theaccount that joint control need fill. In anycase, as we understand it, the notion wouldrequire that, for each name relation, caregiv-ers should train first echoic responding andthen the corresponding tact relation, but notany listener behavior. This is highly unlikelyin the real world of infants language learn-ing. Furthermore, it is clearly a redundantconcept when applied to what is the primaryroute to naming, that is, where listener be-havior is already established and all that is re-quired is to reinforce echoing in the pres-ence of the object. One has to say, also, thatif language were dependent solely upon jointcontrol as described by Lowenkron, it wouldbe a poor thing indeed, for most of the stim-ulus categorization and semantic relations de-scribed in our account would have to be elim-inated. As Lowenkron acknowledges, it wouldmean a return to the exclusively speaker di-
291A THEORY OF VERBAL BEHAVIOR
mension of Skinners Verbal Behavior (1957)and an abandonment of any aspirations to-wards understanding speaker-listener effects.If joint control were the principal determi-nant of verbal behavior, how would one re-spond to the utterance The stars are brighttonight? The speaker here has not instruct-ed the listener to find bright stars, so no par-ticular listener behavior (e.g., looking up atthe stars) should be occasioned. But even ifthis utterance were somehow to occasionsearch and find behavior in the mannerdescribed by Lowenkron, it would then benecessary for the listener to continually echobright stars while scanning the environ-ment until stars were encountered andevoked the tact response stars; the con-junction of the tact and the echoic responseswould then in turn evoke the listener behav-ior of looking at and, perhaps, pointing tothe stars. But we maintain that all of thesesteps are unnecessary: An auditory stimulussuch as /stars/ alone is sufficient to occasionthe conven