Categorization, Concept Learning, and Behavior Analysis: an Introduction

JOURNAL OF THE EXPERIMENTAL ANALYSIS OF BEHAVIOR 2002, 78, 237–248 NUMBER 3(NOVEMBER)

CATEGORIZATION, CONCEPT LEARNING, AND BEHAVIOR ANALYSIS: AN INTRODUCTION

THOMAS R. ZENTALL,MARK GALIZIO, AND THOMAS S. CRITCHFIELD

UNIVERSITY OF KENTUCKY, UNIVERSITY OF NORTH CAROLINA AT WILMINGTON, AND ILLINOIS STATE UNIVERSITY

Categorization and concept learning encompass some of the most important aspects of behavior, but historically they have not been central topics in the experimental analysis of behavior. To intro- duce this special issue of the Journal of the Experimental Analysis of Behavior (JEAB), we deﬁne key terms; distinguish between the study of concepts and the study of concept learning; describe three types of concept learning characterized by the stimulus classes they yield; and brieﬂy identify several other themes (e.g., quantitative modeling and ties to language) that appear in the literature. As the special issue demonstrates, a surprising amount and diversity of work is being conducted that either represents a behavior-analytic perspective or can inform or constructively challenge this perspective. Key words: categorization, concept learning, stimulus class, function transfer

Categorization is not a matter to be taken taken place. Cognitive psychologists Laur- lightly. There is nothing more basic than cat- ence and Margolis (1999), in a book reviewed egorization to our thought, perception, ac- in the present issue, minced no words about tion, and speech. (Lakoff, 1987, p. 5) the association: ‘‘Concepts are the most fun- Concepts give our world stability. They capture damental constructs in theories of the mind’’ the notion that many objects or events are (p. 3). Yet scholars from many research com- alike in some important respects, and hence munities have struggled to come to grips with can be thought about and responded to in the complex repertoires that the topic en- ways already mastered. Concepts also allow us compasses. to go beyond the information given; for once The heterogeneity of this research area is we have assigned an entity to a class . . . we evident in the absence of a consensus defi- can infer some of its . . . attributes. (Smith & nition of the term concept (see Palmer, this Medin, p. 1) issue; Wasserman & Bhatt, 1992). Writers There is, perhaps, no larger or more di- tend to stress the importance of concepts verse literature within experimental psychol- rather than specifying their defining fea- ogy than that focused on categorization and tures—perhaps, as Palmer speculates in his concept learning. This topic is, to the casual review of Margolis and Laurence (1999), ‘‘re- observer, most directly associated with human garding the term as too familiar to need def- cognitive psychology, within which the largest inition’’ (p. 598). Nevertheless, an introduc- volume of research and theory building has tion to this special issue demands at least an attempt to define its subject matter. Order of authorship for this article was determined by Typically in cognitive psychology, categori- random drawing. We are grateful to the numerous re- zation is regarded as a process of determining viewers, who met the dual challenge of evaluating man- what things ‘‘belong together,’’ and a category uscripts according to JEAB’s usual expectations while em- is a group or class of stimuli or events that so bracing the conceptual and methodological diversity inherent in the topic, and to the authors of the articles cohere. A concept is thought to be knowledge contained herein, many of whom invested extra effort in that facilitates the categorization process tailoring their work to an unfamiliar audience. (e.g., Barsalou, 1991, 1992). Consistent with Address correspondence to Tom Zentall at the De- the representational style of much cognitive partment of Psychology, University of Kentucky, Lexing- ton, Kentucky 40506 (e-mail: [email protected]); theorizing, conceptual knowledge is often Mark Galizio at the Department of Psychology, University portrayed as existing independently of any of North Carolina at Wilmington, Wilmington, North particular behavior–environment relation. Carolina 28403 (e-mail: [email protected]); or Tom This is assumed partly because, once a cate- Critchfield at the Department of Psychology, Illinois State University, Normal, Illinois 61704 (e-mail: tscritc@ilstu. gorization repertoire is in place, an individual edu). may be able to categorize both previously en-

237 238 THOMAS R. ZENTALL et al. countered stimuli and novel events, suggest- departure by suggesting that ‘‘when a group ing to some observers that the latter are rec- of objects gets the same response, when they ognized via comparison to general form a class the members of which are react- information represented in memory. Thus, ed to similarly, we speak of a concept’’ (p. the goal of many studies in cognitive psy- 154). Thus, categorization may be said to in- chology is to map the knowledge that humans corporate a pattern of systematic differential presumably apply in already established pat- responding to classes of nonidentical, though terns of categorization. For example, struc- potentially discriminable, stimuli (see Fields, tured interviewing and other techniques may Reeve, et al., this issue; Wasserman & Bhatt, be used to determine what entities people in- 1992). A category is a class of stimuli that oc- clude in a category like birds; which of these casion common responses in a given context. entities are considered to be more or less typ- Such classes include stimuli involved in an ex- ical of the category; and whether hierarchical plicit learning history plus, potentially, novel relations apply to the category or instances stimuli to which the fruits of this history may within it (e.g., Rosch, 1978). transfer. Many writers use the terms category Behavior analysts are likely to regard as for- and stimulus class more or less interchange- eign this practice of describing terminal per- ably; we will follow that practice here. formance without examining the necessary When the stimuli within and between cat- and sufficient conditions for its emergence. egories vary along relatively simple dimen- Reservations about the approach are war- sions (e.g., wavelength, size, brightness), cat- ranted. Although cognitive psychologists ex- egorization is readily conceived in the same pend much energy debating the structure terms as stimulus discrimination and gener- and contents of the knowledge that is as- alization. For example, ‘‘Generalization with- sumed to underpin categorization and the in classes and discrimination between classes— means by which it is compared to new per- this is the essence of concepts’’ (Keller & ceptual experiences (Laurence & Margolis, Schoenfeld, 1950, p. 155). The analytical 1999; see Palmer, this issue, for a brief syn- challenge becomes more daunting, of course, opsis of some relevant theories), their ac- as category membership is determined more counts can be difficult to distinguish empiri- complexly (e.g., Herrnstein, 1990). Consider, cally. Barsalou (1992) has noted a tendency as an instructive case, the balan category of for competing cognitive theories to make similar predictions and to account equally well the Australian aboriginal language Dyirbal, for data obtained from human subjects. Per- which ‘‘includes women, fire, and dangerous haps more important for present purposes, things. It also includes birds that are not dan- this focus on knowledge may discourage at- gerous, as well as exceptional animals such as tention to the role of experience in creating the platypus, bandicoot, and echidna’’ (Lak- and maintaining conceptual behavior (Astley off, 1987, p. 5). Any plausible account must & Wasserman, 1996). also explain how categories add and lose members, merge and fracture, share members that may belong to different categories AN OPERATIONAL APPROACH under different circumstances, support the TO CONCEPT LEARNING spontaneous transfer of function from one From a behavior-analytic perspective, the member to another, and so forth. present topic provides an opportunity to ap- Some writers have gone so far as to label ply the operational analysis of psychological the capacity to glean abstract relations, such terms that Skinner (e.g., 1945) frequently es- as those that apparently unite many catego- poused. Rather than speculating about the ries, as the essence of what it means to be status of hypothetical knowledge structures, it human (e.g., see Deacon, 1997). Regardless is possible to examine the circumstances un- of whether conceptual repertoires are der which we speak of conceptualization— uniquely human (many articles in the present that is, what individuals are doing when they issue suggest that they are not), they are are said to behave conceptually, and how they clearly among the most interesting behavioral came to behave in that way. Keller and phenomena available for study. Because of Schoenfeld (1950) specified this very point of their richness, generativity, and adaptability, CATEGORIZATION AND CONCEPT LEARNING 239 they invite a thorough experimental and the- examples and nonexamples of category oretical analysis. members (Barsalou, 1991). The goal is ‘‘to Unfortunately, categories and concepts establish high degrees of control over cate- have been addressed only sporadically within gory knowledge’’ (Barsalou, 1992, p. 31), behavior analysis. This neglect may be under- with knowledge operationalized more or less stood partly as a rejection of the cognitive as above. Such paradigms, which are com- theoretical worldview that has helped to de- fortingly reminiscent of discrimination learn- fine the topic (in the present issue, Palmer ing procedures, form the basis of most of the catalogues some key points of contention be- research described in the present issue and tween behavior-analytic and cognitive views). constitute an important point of contact be- But taking issue with the theoretical perspec- tween diverse communities of researchers. tive that has dominated the concepts and cat- Although the scientific community with egories literature does not render the rele- which JEAB is most associated has not made vant behavioral phenomena any less categorization and concept learning the cen- provocative. Moreover, objection through si- terpiece of its empirical and theoretical con- lence persuades no one. Although Keller and tributions, a remarkable amount of work is Schoenfeld (1950) provided the outlines of a being conducted that expressly represents a behavioral analysis of conceptual behavior behavior-analytic perspective, can inform this more than half a century ago, scholars out- perspective, or can constructively challenge side behavior analysis have scarcely noticed, this perspective. The overarching purpose of and they are unlikely to do so in the absence this special issue, therefore, is to bring to- of persuasive empirical evidence (Schwartz, gether the best current work of behavioral re- Wasserman, & Robbins, 2002). searchers who seek to determine which spe- The needed research requires a clearly de- cies are capable of which conceptual fined subject matter, and we have already sug- repertoires, and what experiences and cir- gested that concepts, defined largely in terms cumstances make these repertoires possible. of abstract knowledge, provide a slippery We anticipate that some JEAB readers will foundation for an experimental analysis. For- have only passing acquaintance with this re- tunately, ‘‘knowledge’’ can be regarded, not search area. For these readers, we seek to pro- as an entity, but as a linguistic surrogate for vide a cross section of contemporary research learning histories that can be operationalized efforts and illustrate something of the and, in many cases, studied experimentally breadth and value of this area of study. To (e.g., see Gagne´, this issue; Maddox, this is- place this work into a broad context, below sue; Skinner, 1977; Wixted & Gaitan, in we provide some further orienting remarks press). In search of more secure theoretical about the area. Because the literature on cat- footing, the authors of most articles in the egorization and concept learning is so vast, present issue have, in some fashion, recast the even those who work in this area may be un- notion of knowledge in terms of necessary familiar with the efforts of other investigators. and sufficient conditions for conceptual be- A key goal of this special issue is to promote havior. This subtle shift in emphasis, from a cross propagation of empirical and theoret- concepts to concept learning, diverts attention ical insights, and to demonstrate that JEAB at least partly away from taxonomizing knowl- can be the locus for a productive exchange edge and toward identifying the functional on a rich and compelling topic. relations between behavior and environment that provide the basis for conceptual behavior. TY PES OF CONCEPTS The preceding insights are not, of course, The work described in the present issue unique to behavior analysis. Many research draws from multiple scholarly communities. traditions (including human cognitive psy- One community of investigators examines chology; see Maddox, this issue) include at- categorization and concept learning in non- tempts to create artificial categories uniting humans, both to plumb the abilities of spe- experimenter-selected rather than everyday cific species and to assess interspecies similar- stimuli. Almost universally, these efforts re- ities and differences. A second community of quire subjects to respond, with feedback, to researchers has been concerned with the for- 240 THOMAS R. ZENTALL et al. mation of stimulus classes (including equiva- imental contingencies, by what the experi- lence classes) primarily in humans. Much of menter recognizes as an image of a person). their work bears on questions about catego- Following this kind of training, subjects may rization and concept learning, and some of it show a high degree of class-consistent re- addresses these topics explicitly. Finally, a sponding to novel exemplars of the stimulus large community of cognitive psychologists sets (Herrnstein, Loveland, & Cable, 1976). seeks to understand categories and concepts In the present issue, Vonk and MacDonald in humans. To avoid unnecessary entangle- extend this type of procedure to gorillas, ment in theoretical debates between and whose conceptual abilities have been studied within communities, below we emphasize infrequently. three broad types of relations that appear to In studies like that of Herrnstein and Love- unite events within a category. In perceptual land (1964), the pictures in the positive set concepts, stimuli are grouped primarily on the (to which responding is reinforced) and in basis of shared physical features. In relational the negative set (to which responding is ex- concepts, it is not the physical features of stim- tinguished) vary in terms of perspective, uli per se but the relations among these fea- background, color, number of relevant items tures that are grouped. Finally, in associative portrayed, and so forth. In such cases, as concepts, stimuli are grouped on the basis of Vonk and MacDonald (this issue) note, it is shared function (e.g., a common response more difficult to think in terms of simple that they engender, or a common conse- stimulus control processes. One noteworthy quence with which they are correlated). feature of the Vonk and MacDonald study is Perceptual Concepts an extensive, and apparently unsuccessful, attempt to identify simple stimulus features that The classes of stimuli that are united in guide categorization. In such cases, stimulus perceptual concepts may be said, from a sub- variation within a picture set may be as great ject’s perspective, to bear physical similarity as the variation between the sets and, when to one another. Because stimuli often are said many pictures are used (Herrnstein & Love- to be similar on the circular basis that sub- land, 1964, used several hundred), it seems jects respond similarly to them, similarity is unlikely that each separate picture becomes construed here as relatively little separation an independent discriminative stimulus. along a well-defined physical dimension. Straightforward examples can be found in Even when complex stimuli are employed, the present issue in articles by Maddox (line however, concept learning mirrors discrimi- length) and by Fields and colleagues (grada- nation learning in tantalizing ways. For ex- tions between two photographs as altered by ample, in the acquisition of simple discrimi- commercial morphing software). Although nations, experience with both positive and most laboratory concept tasks involve static negative stimuli appears to be necessary; es- stimuli, Herbranson, Fremouw, and Shimp tablishing a response in the presence of one (this issue) show that pigeons can accurately stimulus does not ensure differential re- categorize dynamic properties such as stimu- sponding when other stimuli are introduced lus movement and direction. (e.g., Newman & Baron, 1965). Perhaps anal- In the interest of modeling categorization ogously, training animals to respond to a sin- as it occurs in the natural environment, many gle set of stimuli may be insufficient to ensure studies employ stimuli that are not defined differential responding among multiple sets. by a few simple features. For example, re- Sutton and Roberts, in the present issue, research with nonhumans involves the discrim- port that discrimination training between cat- ination of photographs or drawings that con- egories is needed for pigeons to respond dif- tain a particular type of object, such as a ferentially to untrained members of the person, from those that do not contain the training set versus untrained members of a object. Herrnstein and Loveland (1964) different training set (i.e., category distinc- taught pigeons to discriminate photographs tions may not be made until members of at containing a person from photographs that least two categories are compared). Thus, ex- did not (in such a case, category membership perience with both ‘‘examples’’ and ‘‘nonex- may be said to be determined, in the exper- amples’’ appears to be essential. CATEGORIZATION AND CONCEPT LEARNING 241

Relational Concepts reinforced choice would be BB. Thus, it is not the match between stimuli but the match be- Compared to perceptual concept learning, tween the stimulus–stimulus relations that in which absolute properties of the stimuli must be learned. Chimpanzees acquire such may guide responding, relational concept an abstract conditional discrimination and learning makes use of more abstract proper- can transfer it to novel exemplars (Thomp- ties of the stimuli. One of the simplest and son, Oden, & Boysen, 1997). In the present most studied relational concepts is same ver- issue, possibly analogous performances are sus different. Same–different concept learn- shown in human ‘‘equivalence–equivalence ing can be studied using conditional discrim- responding’’ (Stewart, Barnes-Holmes, Roche, inations in which subjects must respond to & Smeets). the comparison that matches the sample. For example, after pigeons have been trained to Associative Concepts match two shapes, they show facilitated ac- In associative concept learning, the stimuli quisition, relative to an appropriate control, within classes bear no obvious physical simi- when trained with two novel hues (Zentall & larity to one another, but rather cohere be- Hogan, 1978). Even better transfer can be cause of shared functional properties. Sid- found when pigeons are trained on an iden- man’s (e.g., 1994) seminal work on stimulus tity task with a large number of stimuli equivalence has generated enormous inter- (Wright, Cook, Rivera, Sands, & Delius, est, in the behavior-analytic community, in 1988). the study of non-similarity-based stimulus Perhaps the strongest evidence of same– classes. This interest arose, in part, because different learning by pigeons comes from re- arbitrary relations were thought to provide a search in which the simultaneous display of tool for studying symbolic processes relevant similar objects is discriminated from displays to language and cognition (e.g., Sidman, in which all of the objects are clearly different 1971, 2000). As research on stimulus equiva- from each other. Pigeons trained in this way lence has evolved into a major focus in the learn to discriminate the training exemplars experimental analysis of human behavior, re- and also to discriminate novel stimuli involv- searchers of animal learning and cognition ing the same relations as in training (e.g., have been investigating similar phenomena Cook, Katz, & Cavoto, 1997). In the present in nonhumans. issue, Cook reports that the resulting same– A brief review of the basic methods used to different categories are general and can in- study stimulus equivalence, as deﬁned by Sid- clude stimuli that vary along different per- man (e.g., 1994), will illustrate one way that ceptual dimensions such as texture, feature, associative concept learning can be studied in geometric shape, naturalistic drawing, or an the laboratory. Typically, arbitrary match-to- object depicted in a photograph. Also in the sample training is used to establish at least present issue, Wasserman, Young, and Peissig two conditional discriminations involving report that detecting same and different ar- physically unrelated stimuli. On each trial, rays occurs very rapidly and thus does not ap- the correct comparison is conditional on the pear to require the sequential comparison of particular sample presented. For example, the stimuli in the array. given Sample Stimulus A1, a choice of Com- There is also evidence that some species parison Stimulus B1 is reinforced, and given are capable of second-order same–different Sample A2, choosing B2 is reinforced. Anoth- learning, in which the relation between two er set of relations could then reinforce a objects must be matched rather than the ob- choice of C1 given Sample B1 and C2 given jects themselves. With this procedure, the Sample B2. After such training, three kinds sample consists of a pair of objects that are of untrained performances often emerge. either identical or different, and the correct Sidman designated these reﬂexivity (e.g., giv- comparison consists of two stimuli that are en A1 as the sample, the subject chooses different from the sample but that bear the Comparison A1), symmetry (e.g., given B1 as same relation to each other as those in the the sample, the subject chooses Comparison sample. For example, if shown Sample AA A1), and transitivity (e.g., given A1 as the and given a choice between BB and CD, the sample, the subject chooses Comparison C1). 242 THOMAS R. ZENTALL et al.

Humans have shown the emergence of re- of class members, leaving the distinctions be- flexivity, symmetry, and transitivity without ex- tween equivalence and functional classes plicit reinforcement of these relations, and, blurred at best (see Sidman, 2000). To illus- historically, the demonstration of all three re- trate, we consider some of the procedures lations was considered to be the definition of through which such functional classes have an equivalence class (e.g., Sidman, 1994; Sid- been examined. man & Tailby, 1982). Over the past 25 years Common-response training. Stimuli can be- or so, many studies have been directed to- come united into a class by virtue of associa- ward evaluating the mechanics of equivalence tion with a common response. Many-to-one class formation (e.g., types of training and conditional discrimination procedures, for testing formats, preconditions of class merg- example, arrange the requisite common re- er, expansion, and fracture; and the stimulus sponse by reinforcing the matching of two or control processes that promote or preclude more samples (e.g., a red field and vertical class formation; see Arntzen & Holth, 1997; lines) to the same comparison selection (e.g., Carrigan & Sidman, 1992; Lane & Critch- a large circle), and different samples (e.g., a field, 1998; Pilgrim & Galizio, 1995; Saunders green field and horizontal lines) to another & Green, 1999; Sidman, Kirk, & Willson-Mor- comparison selection (e.g., a small circle). ris, 1985). Several lines of evidence indicate that in non- Some commentators have wondered humans a relation develops between the sam- whether stimulus equivalence is a uniquely ples associated with a common comparison human phenomenon (e.g., see Hayes, 1989), (e.g., Zentall, Sherburne, & Urcuioli, 1993; because research has suggested that not all of Zentall, Steirn, Sherburne, & Urcuioli, 1991), the formal properties of equivalence classes just as is the case for humans (e.g., Arntzen (as defined by Sidman & Tailby, 1982) are & Holth, 1997; Fields, Reeve, et al., this issue; readily demonstrated in nonhumans. For ex- see also McDaniel, Nuefeld, & Damico-Nettle- ample, symmetrical relations may fail to ton, 2001). emerge in nonhumans after training involv- A variety of means exist to assess the asso- ing arbitrary stimuli (Sidman et al., 1982). ciative relations that emerge in nonhumans One possible explanation of this failure em- following many-to-one training. For example, phasizes not a generic deficit but rather com- if Samples A and B share a common compar- peting control by factors such as stimulus lo- ison selection, C, and further training relates cation. In the present issue, Lionello-DeNolf Sample A to new Comparison D, Sample B and Urcuioli describe an experiment in will be matched to Comparison D without ad- which pigeons were tested for symmetry after ditional training (Urcuioli, Zentall, Jackson- multiple sample-location training designed to Smith, & Steirn, 1989). Another approach in- reduce control by a particular location. Sym- volves comparing the discriminability of metrical responding was directly reinforced samples associated with a common compari- with some stimulus sets before testing for son selection with that of samples associated symmetry with new sets. Although baseline with different comparison selections. If rela- matching transferred to novel stimulus loca- tions have formed between samples associat- tions, no evidence of the emergence of sym- ed with the same comparison selection, then metry was obtained in any of their experi- it should be harder to discriminate between ments. those samples than between samples associ- As the preceding case illustrates, recent ated with different comparison selections. Re- years have seen a convergence of interests be- cent research suggests that this is indeed the tween equivalence researchers and research- case (Kaiser, Sherburne, Steirn, & Zentall, ers who study associative concepts in animal 1997). cognition (Zentall & Smeets, 1996). Some Common-sample training. It is worth noting studies influenced by this convergence now that, in humans, associative concepts develop suggest the capacity of certain nonhuman fairly readily through one-to-many training, species to acquire equivalence relations (see in which a common sample is associated with Kastak & Schusterman, this issue; Kastak, two or more comparison selections (e.g., Schusterman, & Kastak, 2001). Many more Arntzen & Holth, 1997). Surprisingly, similar studies have shown functional substitutability results have not been found with pigeons, CATEGORIZATION AND CONCEPT LEARNING 243 even when conditions are otherwise similar to issue, Jitsumori, Siemann, Lehr, and Delius those under which associative concepts have report that similar results can be obtained us- been shown to develop during many-to-one ing simultaneous discriminations. The use of training (Urcuioli & Zentall, 1993). The simultaneous discriminations permits the asymmetry of these effects hints at differenc- presentation of novel pairings of the training es in the way humans and other animals form stimuli. For example, the positive stimulus these associative concepts, although interspe- from one training pair could be presented cies similarities may yet exist. In the present with the negative stimulus from a different issue, Fields, Reeve, et al. report that the like- training pair. lihood of humans acquiring a generalized ca- Symmetry training. Recent research suggests tergorization repertoire (one that applies that symmetry training, in which A → B and across several different mixed perceptual-as- B → A relations are taught, can serve to form sociative categories) is positively related to a stimulus class consisting of those stimuli. the number of sample stimuli used during Zentall, Clement, and Weaver (in press) training and is negatively related to the num- found that, following this training, if one ber of comparison stimuli. Their findings sug- stimulus was also associated with a new stim- gest that one-to-many training may interfere ulus (e.g., B → C) in a conditional discrimi- with generalized categorization in humans. nation, the untrained relation A → C Common-outcome training. Stimuli can co- emerged as well. here into classes by way of their association Choice by exclusion. When animals learn to with a common outcome. For example, Ed- match, it is assumed that they learn to select wards, Jagielo, Zentall, and Hogan (1982) the comparison based on its acquired associ- trained pigeons on two identity-matching ation with the sample. Articles in the present tasks, one with lines and the other with hues. issue indicate that if chimpanzees or sea lions In each task, correct choices of one of the are presented with a novel sample and one comparisons (e.g., red and vertical) were re- novel and one familiar comparison (i.e., a inforced with peas, whereas correct choices comparison already associated with a differ- of the other comparisons (e.g., green and ent sample), the animals will tend to choose horizontal) were reinforced with wheat. Fol- by exclusion (Beran & Washburn; Kastak & lowing training, pigeons showed positive Schusterman). That is, they will avoid the fa- transfer to trials in which hues and lines as- miliar comparison (that with an experimen- sociated with a common outcome could be tally defined function) and choose the novel matched. For analogous findings with hu- (undefined) one. Control procedures reveal mans, see Dube, McIlvane, Maguire, Mackay, that this is not simply a preference for novel and Stoddard (1989). comparisons but rather a general capacity to Serial reversals. Another procedure for de- match undefined stimuli. Such choices there- veloping stimulus classes is to treat a set of fore suggest an indirect measure of concept stimuli the same through a number of trans- learning. formations (Vaughan, 1988). Vaughan ran- Transfer of function. In many of the preced- domly assigned photographs of trees to two ing examples, concept learning was evaluated arbitrary sets, A and B, and then trained pi- through selection-based repertoires involving geons to respond to those in Set A but not to the matching of one class member to anoth- respond to those in Set B. Following training, er. Such repertoires have been described as the valence associated with each set was re- part of a more general transfer of function versed, and then reversed again, repeatedly. (also called transformation of function or in- Across reversals of stimulus sets, the number heritance of meaning), in which the stimuli of trials needed for responding to follow suit within a class spontaneously share whatever decreased to only a few trials, suggesting that functions each has acquired separately these arbitrarily assigned stimuli had become (Dougher & Markham, 1996; Dymond & two functional stimulus classes, in spite of the Rehfeldt, 2000). Manabe, Kawashima, and fact that stimuli differed as much within sets Staddon (1995) described a procedure in as between sets in terms of physical similarity which, to produce comparison stimuli during and reinforcement histories. In the present match-to-sample training, budgerigars were 244 THOMAS R. ZENTALL et al. required to make distinct vocal responses to OTHER ISSUES each of two samples. Later, new samples were Theoretical Views associated with the same comparisons (creating many-to-one training), but without dif- Much research on categorization and con- ferential vocalization requirements. Differen- cept learning is guided by cognitive theories tial vocalizations spontaneously transferred to that run counter to traditions in behavior the new stimuli, apparently on the basis of analysis (Palmer, this issue), but the cognitive association with common comparison selec- literature merits attention for at least two rea- tions. In the present issue, Urcuioli et al. de- sons. First, regardless of its theoretical under- scribe a successful attempt to replicate this pinnings, this literature describes many em- effect in pigeons, substituting high-rate versus pirical phenomena that any noncognitive low-rate pecking patterns for differential vo- theory must explain (e.g., see Concepts and calization. Language, below). Second, cognitive psychology is not monolithic, and behavior analysts Transfer of function also has been a focus may be surprised to find that they share many in many studies of stimulus equivalence. For assumptions with some of their cognitive example, Dougher and colleagues first cre- counterparts. In the present issue, Maddox ated classes of arbitrary stimuli using stimulus and Gagne´ provide cases in point. Maddox, equivalence procedures, then paired shock for instance, stresses the importance of pro- with one member of a class, making it a con- viding individuals with extended exposure to ditioned stimulus for skin conductance re- experimental manipulations and of evaluat- sponses indicating fear elicitation. The elici- ing individual-subject response functions. His tation function appeared spontaneously in quantitative models of perceptual concept other members of the same class (Dougher, learning draw heavily on signal-detection Augustson, Markham, Greenway, & Wulfert, principles and the effects of consequences on 1994). As Gagne´ (this issue) suggests, such behavior. transfer-of-function outcomes are reminis- Readers of research on stimulus relations cent of the shared meaning evident in hybrid (such as equivalence) in humans will discover lexical concepts. Gagne´ proposes an experi- the influence of three main theories. Sid- ential basis for this transfer of function to man’s (2000) stance is that equivalence rela- which readers may readily apply familiar be- tions represent a fundamental outcome of a havior principles (e.g., conditional relations reinforcement contingency such that all ele- and the matching law). ments of the contingency (conditional and In general, the circumstances that facilitate discriminative stimuli, responses, and rein- transfer of function are poorly understood forcers) are posited to become equivalence (e.g., see Barnett & Ceci, 2002; Barsalou, class members. Several papers in the present 1992; Dymond & Rehfeldt, 2000). In the pres- issue incorporate Sidman’s framework (e.g., ent issue, Urcuioli et al. offer an extended Fields et al.; Griffee & Dougher). Two com- analysis of conditions under which such ef- peting views—naming theory (Horne & fects might arise, and address the possibility Lowe, 1996) and relational frame theory that adventitious reinforcement might create (Hayes, Barnes-Holmes, & Roche, 2001)— both propose that the emergence of equiva- spurious cases of function transfer. Also in lence relations reflects previously established this issue, Markham and Markham explore higher order operant behavior. some of the necessary and sufficient condi- In the case of naming theory, the prereq- tions for transfer of function in humans. They uisite operant is a bidirectional relation be- describe a case in which stimulus class for- tween objects and the speaker-listener behav- mation (on the basis of common respondent ior (names) they occasion (note that naming, functions) apparently failed, but an operant construed somewhat differently, also plays a response function, trained to one stimulus, role in some cognitive accounts; see Mark- propagated through the putative class never- man, 1991). An example of research con- theless. The results provide fodder for inter- ducted within the framework of naming the- esting speculation about what exactly creates ory is provided in present issue by Lowe, and defines stimulus classes. Horne, Harris, and Randle, who describe a CATEGORIZATION AND CONCEPT LEARNING 245 variation of the common-response procedure in the style typical of many investigations pub- in which children were trained to make one lished in JEAB, has not played a large role in common naming response to a set of physi- the study of conceptual behavior, but cases cally unrelated stimuli and another common worthy of examination can be identified in naming response to a different set of stimuli. human cognitive psychology. Gluck and Bow- The common names were sufficient to estab- er (1988), for example, presented a model of lish stimulus classes determined by sorting concept learning with roots in the Rescorla- the stimuli in a category match-to-sample test. Wagner (1972) account of respondent con- Relational frame theory posits that much ditioning. In the present issue, Maddox de- complex behavior, including equivalence, is scribes a quantitative model of learning better understood as the result of ‘‘arbitrarily perceptual categories, and Gagne´ describes a applicable relational responding’’ called a re- model of conceptual combination. lational frame (relational frames also are in- voked in some cognitive accounts; see Barsal- Concepts and Language ou, 1993, and Gagne´, this issue). From this Conceptual behavior often is thought to be vantage point, equivalence is one of many linked to complex human capabilities such as possible relational operants that Hayes et al. language, and much research has examined (2001) call a frame of coordination. They use arbitrary stimulus relations in humans as a the terms mutual entailment and combinatorial theoretical basis of language and cognition entailment, rather than symmetry and transitiv- (see Hayes et al., 2001; Horne & Lowe, 1996; ity, because in some relational frames (e.g., Sidman, 1971, 2000). This work is in its in- ‘‘greater than’’) entailed relations are differ- fancy, however, and it remains to be seen how ent from those that emerge from equivalence adequately laboratory-generated stimulus re- relations. Thus, relational frame theory is de- lations can simulate, and support accounts of, signed to account for a broad range of rela- everyday language and cognition. It is en- tions including equivalence phenomena. couraging that several phenomena typically Some aspects of the theory are illustrated in observed in lexical classes have been mod- the attempt by Stewart et al. (this issue) to eled with equivalence relations. For example, craft a behavior-analytically based model of demonstrations of category clustering in free analogy. recall (Galizio, Stewart, & Pilgrim, 2001), fast It is worth noting that both relational lexical mapping (Wilkinson, Dube, & Mc- frame theory and naming theory have been Ilvane, 1996), and semantic priming (Hayes criticized for vagueness about the origins of & Bisset, 1998) have all been accomplished higher order operants that supposedly un- with equivalence class procedures. derpin stimulus class formation. Although Several papers in the present issue extend not targeting these theories per se, Fields, such analyses to additional aspects of lan- Reeve, et al. (this issue) describe some of the guage-based categories. For example, lexical training conditions under which a general- classes often include both members that are ized repertoire for categorizing perceptually related perceptually and members that are defined stimuli is likely to arise. Their exper- not (e.g., Lane, Clow, Innis, & Critchfield, iment illustrates the feasibility of conducting 1998). The category furniture includes exem- studies that bear on the plausibility of ac- plars such as chairs that are perceptually re- counts based on higher order operants. lated to one another, but might not be perceptually related to other category members Formal Modeling like cabinets or beds. Fields et al., in two pa- Some investigations of concept learning pers in the present issue, explore the merger test, or support the development of, formal of perceptually based classes with equivalence models. In human cognitive psychology, this classes in generalized equivalence classes. is most evident in connectionist paradigms Language-based categories often take on a that focus on computer simulations of neural hierarchical structure involving nonequiva- networks (e.g., Barsalou, 1993). Computer lence relations between members. For in- simulations also play a role in some areas of stance, all birds have wings but not all wings concept learning research with nonhumans are found on birds; all primates are mam- (e.g., Wynne, 1995). Quantitative modeling, mals, but not all mammals are primates; and 246 THOMAS R. ZENTALL et al. so forth. In the present issue, Griffee and ulus class research provide a possible means Dougher use differential reinforcement con- of modeling speaker effects. tingencies to establish hierarchical category relations that provide an elegant model of su- perordinate, subordinate, and basic level cat- CONCLUSION egorization. The Stewart et al. model of anal- We have presented a brief introduction to ogy in this issue provides another relevant some of the problems that have concerned example. investigators of categorization and concept The preceding cases all involve attempts to learning. We did not attempt a comprehen- generalize from concept learning, as ar- sive survey of this research area because the ranged in the laboratory, to naturally occur- relevant literature is too extensive. Consider, ring language phenomena outside the labo- for example, that the Margolis and Laurence ratory. In the present issue, Gagne´ illustrates (1999) volume reviewed in this issue, which the opposite approach, in which naturally oc- includes only selected seminal papers from curring abilities are brought into the labora- only one research community, consists of over tory and experimentally analyzed. Gagne´’s re- 600 densely packed pages. The present dis- search promotes testable predictions about cussion is intended to place the articles of the the learning histories that may underlie con- special issue into a context of interest to jour- ceptual combination involving the merger of nal readers, but it barely scratches the surface lexical categories through noun–noun com- of a complex topic. One goal of the special binations (e.g., exam headache). issue, therefore, is to promote further read- As Gagne´’s article indicates, members of ing in this area. lexical classes often exchange multiple func- For a review of the literature on categori- tions, and transfer of function may not always zation and concept learning in nonhumans, reciprocate among stimuli within a class (see we recommend Roberts (1998, chap. 11) and also Rosch, 1978). Unfortunately, data rele- Honig and Fetterman (1992). For an intro- vant to these complex effects are scarce in the duction to the research on stimulus equiva- literature on stimulus class formation, mak- lence in humans, we recommend Sidman ing an example from Clow (2000) worthy of (1994). For discussion of the common re- mention. After humans had acquired equiv- search agendas that have united animal learn- alence classes, separate training procedures ing and human equivalence communities, we established additional discriminative func- suggest a volume edited by Zentall and tions for two class members. These additional Smeets (1996; see also a 1993 special issue of functions were not mutually exclusive, poten- The Psychological Record, Vol. 43, No. 4). Final- tially allowing the blending of functions ly, no consideration of categorization and through transfer. Class members to which no concept learning will be complete without discriminative function had been explicitly some attention to the voluminous literature trained almost always inherited one such in human cognitive psychology. Barsalou function, but conjunctive transfer, reflecting (1992, chap. 2 and 7) provides a brief, acces- the blending of discriminative repertoires, oc- sible introduction to mainstream cognitive curred in only about half of these cases. Stim- work. Extensive surveys can be found in uli to which one discriminative function had Smith and Medin (1981) and the Margolis been explicitly trained almost never acquired and Laurence (1999) volume reviewed in the a function that had been trained to another present issue. stimulus in the same class. These outcomes We hope that, ultimately, this special issue broadly mimic patterns seen in lexical classes, will encourage new applications of the exper- and they highlight an important possible con- imental analysis of behavior to the phenom- tribution of stimulus class research. Studies of ena and processes involved in categorization conceptual combination almost always focus and concept learning. Systematic programs of on listener behavior (i.e., comprehension; research that complement, challenge, and ex- see Gagne´, this issue) or do not distinguish tend those sampled in this special issue are clearly between speaker behavior (produc- called for, and the reports collected here il- tion) and listener behavior (e.g., Barsalou, luminate some of the many questions worth 1991). Function-transfer procedures in stim- addressing. If nothing else, the present issue CATEGORIZATION AND CONCEPT LEARNING 247 should demonstrate that, no matter how vast inforcer-specific expectancies. Journal of Experimental the community of scholars who investigate Psychology: Animal Behavior Processes, 8, 244-259. Galizio, M., Stewart, K. L., & Pilgrim, C. (2001). Clus- categorization and concept learning, there is tering in artificial categories: An equivalence analysis. work enough for all who are interested. Psychonomic Bulletin & Review, 8, 609–614. Gluck, M. A., & Bower, G. H. (1988). From conditioning to category learning: An adaptive network model. REFERENCES Journal of Experimental Psychology: General, 8, 37–50. Hayes, S. C. (1989). Nonhumans have not yet shown stimulus equivalence. Journal of the Experimental Anal- Arntzen, E., & Holth, P. (1997). Probability of stimulus ysis of Behavior, 51, 385–392. equivalence as a function of training design. The Psy- Hayes, S. C., Barnes-Holmes, D., & Roche, B. (2001). chological Record, 47, 309–320. Relational frame theory: A post-Skinnerian account of hu- Astley, S. L., & Wasserman, E. A. (1996). Mediating as- man language and cognition. New York: Plenum. sociations, essentialism, and non-similarity-based cat- Hayes, S. C., & Bisset, R. (1998). Derived stimulus rela- egorization. In T. R. Zentall & P. M. Smeets (Eds.), tions produce mediated and episodic priming. The Stimulus class formation in humans and animals (pp. Psychological Record, 48, 617–630. 111–133). Amsterdam: Elsevier. Herrnstein, R. J. (1990). Levels of stimulus control: A Barnett, S. M., & Ceci, S. J. (2002). When and where do functional approach. Cognition, 37, 133–166. we apply what we learn? A taxonomy for far transfer. Herrnstein, R. J., & Loveland, D. H. (1964). Complex Psychological Bulletin, 128, 612–637. visual concepts in the pigeon. Science, 146, 549–551. Barsalou, L. W. (1991). Deriving categories to achieve Herrnstein, R. J., Loveland, D. H., & Cable, C. (1976). goals. In G. H. Bower (Ed.), The psychology of learning Natural concepts in pigeons. Journal of Experimental and motivation (Vol. 27, pp. 1–64). San Diego: Aca- Psychology: Animal Behavior Processes, 2, 285–302. demic Press. Honig, W. K., & Fetterman, J. G. (1992). Cognitive aspects Barsalou, L. W. (1992). Cognitive psychology: An overview of stimulus control. Hillsdale, NJ: Erlbaum. for cognitive scientists. Hillsdale, NJ: Erlbaum. Horne, P. J., & Lowe, C. F. (1996). On the origins of Barsalou, L. W. (1993). Flexibility, structure, and lin- naming and other symbolic behavior. Journal of the Ex- guistic vagary in concepts: Manifestations of a com- perimental Analysis of Behavior, 65, 185–241. positional system of perceptual symbols. In A. F. Col- Kaiser, D. H., Sherburne, L. M., Steirn, J. N., & Zentall, lins, S. E. Gathercole, M. A. Conway, & P. E. Mathis T. R. (1997). Perceptual learning in pigeons: De- (Eds.), Theories of memory (pp. 29–89). Hillsdale, NJ: creased ability to discriminate samples mapped onto Erlbaum. the same comparison in many-to-one matching. Psy- Carrigan, P. F., Jr., & Sidman, M. (1992). Conditional chonomic Bulletin & Review, 4, 378–381. discrimination and equivalence relations: A theoreti- Kastak, C. R., Schusterman, R. J., & Kastak, D. (2001). cal analysis of control by negative stimuli. Journal of Equivalence classification by California sea lions using the Experimental Analysis of Behavior, 58, 183–204. class-specific reinforcers. Journal of the Experimental Clow, J. K. (2000). The combination of response func- Analysis of Behavior, 76, 131–158. tions in merged equivalence classes: A behavioral par- Keller, F. S., & Schoenfeld, W. N. (1950.) Principles of adigm for conceptual combination. Dissertation Ab- psychology. New York: Appleton-Century-Crofts. stracts International, 61(3B), 1619. Lakoff, G. (1987). Women, fire, and dangerous things: What Cook, R. G., Katz, J. S., & Cavoto, B. R. (1997). Pigeon categories reveal about the mind. Chicago: University of same-different concept learning with multiple stimu- Chicago Press. lus classes. Journal of Experimental Psychology: Animal Be- Lane, S. D., Clow, J. K., Innis, A., & Critchfield, T. S. havior Processes, 23, 417–433. (1998). Generalization of crossmodal stimulus equiv- Deacon, T. W. (1997). The symbolic species: The co-evolution alence classes: Operant processes as components in of language and the brain. New York: Norton. human category formation. Journal of the Experimental Dougher, M. J., Augustson, E. M., Markham, M. R., Analysis of Behavior, 70, 267–279. Greenway, D., & Wulfert, E. (1994). The transfer of Lane, S. D., & Critchfield, T. S. (1998). Classification of respondent eliciting and extinction functions through vowels and consonants by individuals with moderate stimulus equivalence classes. Journal of the Experimental mental retardation: Development of arbitrary rela- Analysis of Behavior, 50, 125–144. tions via match-to-sample training with compound Dougher, M. J., & Markham, M. R. (1996). Stimulus classes stimuli. Journal of Applied Behavior Analysis, 31, 21–41. and the untrained acquisition of function. In T. R. Zen- Laurence, S., & Margolis, E. (1999). Concepts and cog- tall & P. M. Smeets (Eds.), Stimulus class formation in hu- nitive science. In E. Margolis & S. Laurence (Eds.), mans and animals (pp. 137–152). Amsterdam: Elsevier. Concepts: Core readings (pp. 3–81). Cambridge, MA: Dube, W. V., McIlvane, W. J., Maguire, R. W., Mackay, H. MIT Press. A., & Stoddard, L. T. (1989). Stimulus class forma- Manabe, K., Kawashima, T., & Staddon, J. E. R. (1995). tion and stimulus-reinforcer relations. Journal of the Ex- Differential vocalization in budgerigars: Towards an perimental Analysis of Behavior, 51, 65–76. experimental analysis of naming. Journal of the Experi- Dymond, S., & Rehfeldt, R. A. (2000). Understanding mental Analysis of Behavior, 63, 111–126. complex behavior: The transformation of stimulus Margolis, E., & Laurence, S. (Eds.). (1999). Concepts: Core functions. The Behavior Analyst, 23, 239–254. readings. Cambridge, MA: MIT Press. Edwards, C. A., Jagielo, J. A., Zentall, T. R., & Hogan, D. Markman, E. M. (1991). The whole-object, taxonomic, E. (1982). Acquired equivalence and distinctiveness and mutual exclusivity assumptions as initial con- in matching-to-sample by pigeons: Mediation by re- straints on word meanings. In S. A. Gelman & J. P. 248 THOMAS R. ZENTALL et al.

Byrnes (Eds.), Perspectives on language and thought (pp. relations between relations in a conceptual matching- 72–106). New York: Cambridge University Press. to-sample task. Journal of Experimental Psychology: Ani- McDaniel, M. A., Nuefeld, K. H., & Damico-Nettleton, S. mal Behavior Processes, 23, 31-43. (2001). Many-to-one and one-to-many associative Urcuioli, P. J., & Zentall, T. R. (1993). A test of comparison- learning in a naturalistic task. Journal of Experimental stimulus substitutability following one-to-many matching Psychology: Applied, 7(3), 182–194. by pigions. The Psychological Record, 43, 725–744. Newman, F. L., & Baron, M. R. (1965). Stimulus gener- Urcuioli, P. J., Zentall, T. R., Jackson-Smith, P., & Steirn, alization along the dimension of angularity. Journal of J. N. (1989). Evidence for common coding in many- Comparative and Physiological Psychology, 60, 59–63. to-one matching: Retention, intertrial interference, Pilgrim, C., & Galizio, M. (1995). Reversal of baseline and transfer. Journal of Experimental Psychology: Animal relations and stimulus equivalence: I. Adults. Journal Behavior Processes, 15, 264–273. of the Experimental Analysis of Behavior, 63, 225–238. Vaughan, W., Jr. (1988). Formation of equivalence sets Rescorla, R. A., & Wagner, A. R. (1972). A theory of in pigeons. Journal of Experimental Psychology: Animal Pavlovian conditioning: Variations in the effectiveness Behavior Processes, 14, 36–42. of reinforcement and nonreinforcement. In A. H. Wasserman, E. A., & Bhatt, R. S. (1992). Conceptuali- Black & W. F. Prokasy (Eds.), Classical conditioning: Vol. zation of natural and artiﬁcial stimuli by pigeons. In 2. Current research and theory (pp. 64–99). New York: W. K. Honig & J. G. Fetterman (Eds.), Cognitive aspects Appleton-Century-Crofts. of stimulus control (pp. 203–223). Hillsdale, NJ: Erl- Roberts, W. A. (1998). Principles of animal cognition. Bos- baum. ton: McGraw-Hill. Wilkinson, K. M., Dube, W. V., & McIlvane, W. J. (1996). Rosch, E. (1978). Principles of categorization. In E. A crossdisciplinary perspective on studies of rapid Rosch & B. B. Lloyd (Eds.), Cognition and categorization word mapping in psycholinguistics and behavior anal- (pp. 27–48). Hillsdale, NJ: Erlbaum. ysis. Developmental Review, 16, 125–148. Saunders, R. R., & Green, G. (1999). A discrimination Wixted, J. T., & Gaitan, S. C. (in press). Cognitive the- analysis of training-structure effects on stimulus equiv- ories of reinforcement history surrogates: The case of alence outcomes. Journal of the Experimental Analysis of likelihood ratio models of human recognition mem- Behavior, 72, 117–137. ory. Animal Learning & Behavior. Schwartz, B., Wasserman, E. A., & Robbins, S. J. (2002). Wright, A. A., Cook, R. G., Rivera, J. J., Sands, S. F., & Psychology of learning and behavior. New York: Norton. Delius, J. D. (1988). Concept learning by pigeons: Sidman, M. (1971). Reading and auditory-visual equiva- Matching-to-sample with trial-unique video picture lences. Journal of Speech and Hearing Research, 14, 5–13. stimuli. Animal Learning & Behavior, 16, 436–444. Sidman, M. (1994). Equivalence relations and behavior: A Wynne, C. D. L. (1995). Reinforcement accounts for research story. Boston: Authors Cooperative. transitive inference performance. Animal Learning & Sidman, M. (2000). Equivalence relations and the rein- Behavior, 23, 207–217. forcement contingency. Journal of the Experimental Zentall, T. R., Clement, T. S., & Weaver, J. E. (in press). Analysis of Behavior, 74, 127–146. Symmetry training in pigeons can produce functional Sidman, M., Kirk, B., & Willson-Morris, M. (1985). Six- equivalences. Psychonomic Bulletin & Review. member stimulus classes generated by conditional-dis- Zentall, T. R., & Hogan, D. E. (1978). Same/different crimination procedures. Journal of the Experimental concept learning in the pigeon: The effect of negative Analysis of Behavior, 43, 21–42. instances and prior adaptation to the transfer stimuli. Sidman, M., Rauzin, R., Lazar, R., Cunningham, S., Tail- Journal of the Experimental Analysis of Behavior, 30, 177– by, W., & Carrigan, P. (1982). A search for symmetry 186. in the conditional discrimination of rhesus monkeys, Zentall, T. R., Sherburne, L. M., & Urcuioli, P. J. (1993). baboons, and children. Journal of the Experimental Anal- Common coding in a many-to-one delayed matching ysis of Behavior, 37, 23–44. task as evidenced by facilitation and interference ef- Sidman, M., & Tailby, W. (1982). Conditional discrimi- fects. Animal Learning & Behavior, 21, 233–237. nation vs. matching to sample: An expansion of the Zentall, T. R., & Smeets, P. M. (Eds.). (1996). Stimulus testing paradigm. Journal of the Experimental Analysis of class formation in humans and animals. Amsterdam: El- Behavior, 37, 5–22. sevier. Skinner, B. F. (1945). The operational analysis of psy- Zentall, T. R., Steirn, J. N., Sherburne, L. M., & Urcuioli, P. chological terms. Psychological Review, 52, 270–277. J. (1991). Common coding in pigeons assessed through Skinner, B. F. (1977). Why I am not a cognitive psy- partial versus total reversals of many-to-one conditional chologist. Behaviorism, 5, 1–10. and simple discriminations. Journal of Experimental Psy- Smith, E. E., & Medin, D. L. (1981). Categories and con- chology: Animal Behavior Processes, 17, 194–201. cepts. Cambridge, MA: Harvard University Press. Thompson, R. K. R., Oden, D. L., & Boysen, S. T. (1997). Received June 19, 2002 Language-naive chimpanzees (Pan troglodytes) judge Final acceptance July 1, 2002