An implicit technology of generalization stokes and baer
JOURNAL OF APPLIED BEHAVIOR ANALYSIS 1977, 10, 349-367 NUMBER 2 (SUMMER) 1977 AN IMPLICIT TECHNOLOGY OF GENERALIZATION TREVOR F. STOKES AND DONALD M. BAER THE UNIVERSITY OF MANITOBA AND THE UNIVERSITY OF KANSAS Traditionally, discrimination has been understood as an active process, and a technology of its procedures has been developed and practiced extensively. Generalization, by con- trast, has been considered the natural result of failing to practice a discrimination technology adequately, and thus has remained a passive concept almost devoid of a technology. But, generalization is equally deserving of an active conceptualization and technology. This review summarizes the structure of the generalization literature and its implicit embryonic technology, categorizing studies designed to assess or program generalization according to nine general headings: Train and Hope; Sequential Modifi- cation; Introduce to Natural Maintaining Contingencies; Train Sufficient Exemplars; Train Loosely; Use Indiscriminable Contingencies; Program Common Stimuli; Mediate Generalization; and Train "To Generalize". DESCRIPTORS: generalization, treatment-gain durability, followup measures, main- tenance, postcheck methodology Traditionally, many theorists have considered Schoenfeld, 1950, p. 168ff.). Thus, generaliza-generalization to be a passive phenomenon. Gen- tion was something that happened, not some-eralization was not seen as an operant response thing produced by procedures specific to it.that could be programmed, but as a description If generalization seemed absent or insignifi-of a "natural" outcome of any behavior-change cant, it was simply to be assumed that the teach-process. That is, a teaching operation repeated ing process had managed to maintain unusuallyover time and trials inevitably involves varying tight control of the stimuli and responses in-samples of stimuli, rather than the same set volved, allowing little sampling of their varie-every time; in the same way, it inevitably evokes ties. This assumption was strongly supported byand reinforces varying samples of behavior, the well-known techniques of discrimination: byrather than the same set every time. As a conse- differential reinforcement (in general, by anyquence, it is predictable that newly taught re- differential teaching) of certain stimuli relativesponses would be controlled not only by the to others, and/or certain responses relative tostimuli of the teaching program, but by others others, generalization could be programmaticallysomewhat resembling those stimuli (Skinner, restricted and diminished to a very small range.1953, p. 107ff.). Similarly, responses resembling Thus, it was discrimination that was understoodthose established directly, yet not themselves ac- as an active process, and a technology of its pro-tually touched by the teaching procedures, would cedures was developed and practiced extensively.appear as a result of the teaching (Keller and But generalization was considered the natural result of failing to practice discriminations tech- Preparation of this paper was supported in part by nology adequately, and thus remained a passivePHS Training Grant 00183, Program Project GrantHD 00870, and Research Grant MH 11739. Reprints concept almost devoid of a technology. Never-may be obtained either from T. F. Stokes, Department theless, in educational practice, and in the devel-of Psychology, University of Manitoba, Winnipeg, opment of theories aimed at serving both practiceManitoba, Canada R3T 2N2, or D. M. Baer, Depart-ment of Human Development, University of Kansas, and a better understanding of human function-Lawrence, Kansas 66045. ing, generalization is equally as important as dis- 349
350 TREVOR F. STOKES and DONALD M. BAERcrimination, and equally deserving of an active scheduling of the same events in those conditionsconceptualization. as had been scheduled in the training conditions. Generalization has been and doubtless will Thus, generalization may be claimed when noremain a fundamental concern of applied behav- extratraining manipulations are needed for extra-ior analysis. A therapeutic behavioral change, training changes; or may be claimed when someto be effective, often (not always) must occur extra manipulations are necessary, but their costover time, persons, and settings, and the effects or extent is clearly less than that of the directof the change sometimes should spread to a intervention. Generalization will not be claimedvariety of related behaviors. Even though the when similar events are necessary for similar ef-literature shows many instances of generaliza- fects across conditions.tion, it is still frequently observed that when a A technology of generalization programmingchange in behavior has been accomplished is almost a reality, despite the fact that until re-through experimental contingencies, then that cently, it had hardly been recognized as a prob-change is manifest where and when those contin- lem in its own right. Within common teachinggencies operate, and is often seen in only transi- practice, there is an informal germ of a technol-tory forms in other places and at other times. ogy for generalization. Furthermore, within the The frequent need for generalization of thera- practice of applied behavior analysis (especiallypeutic behavior change is widely accepted, but it within the past 5 yr or so), there has appearedis not always realized that generalization does a budding area of "generalization-promotion"not automatically occur simply because a behav- techniques. The purpose of this review is to sum-ior change is accomplished. Thus, the need ac- marize the structure of that generalization litera-tively to program generalization, rather than ture and its implicit embryonic technology. Somepassively to expect it as an outcome of certain 270 applied behavior analysis studies relevant totraining procedures, is a point requiring both generalization in that discipline were reviewed.2emphasis and effective techniques (Baer, Wolf, A central core of that literature, consisting ofand Risley, 1968). That such exhortations have some 120 studies, contributes directly to a tech-often been made has not always ensured that nology of generalization. In general, techniquesresearchers in the field have taken serious note designed to assess or to program generalizationof and, therefore, proceeded to analyze ade- can be loosely categorized according to ninequately the generalization issues of vital concern general headings:to their programs. The emphasis, refinement, andelaboration of the principles and procedures that 1. Train and Hopeare meant to explain and produce generalization 2. Sequential Modificationwhen it does not occur "naturally" is an impor- 3. Introduce to Natural Maintaining Contin-tant area of unfinished business for applied be- gencieshavior analysis. 4. Train Sufficient Exemplars The notion of generalization developed here 5. Train Looselyis an essentially pragmatic one; it does not 6. Use Indiscriminable Contingenciesclosely follow the traditional conceptualizations 7. Program Common Stimuli (Keller and Schoenfeld, 1950; Skinner, 1953).In many ways, this discussion will sidestep much 2Ninety per cent of the literature reviewed wasof the controversy concerning terminology. Gen- from five journals: Behaviour Research and Therapy; eralization will be considered to be the occur- Behavior Therapy; Journal of Applied Behavior Anal- rence of relevant behavior under different, non- ysis; Journal of Behavior Therapy and Experimental Psychiatry; and Journal of Experimental Child Psy- training conditions (i.e., across subjects, settings, chology. Seventy-seven per cent of the literature re- people, behaviors, and/or time) without the viewed has been published since 1970.
AN IMPLICIT TECHNOLOGY OF GENERALIZATION 35 1 8. Mediate Generalization examples of the programming of generalization, 9. Train "To Generalize". they are a sound first step in any serious analysis of generalization. When generalization is desired, This review characterizes each category, and but is shown to be absent or deficient, program-describes some examples of research that illus- ming procedures can then be instituted.trate the generalization analyses or program- For example, useful generalization across set-ming involved in each category. Obviously, all tings was documented by Kifer, Lewis, Green,the relevant references cannot be discussed in this and Phillips (1974). In an experimental class-review.3 The nine categories listed above were room setting, parent-child pairs were taught toinduced from the literature; they are not a priori negotiate in conflict situations. During simulatedcategories. Consequently, studies do not always role-playing, instructions, practice, and feedbackfit neatly into these categories. It should also be were used to teach the negotiation behaviors ofnoted that not all studies reviewed were thorough fully stating ones position, identifying the issuesexperimental analyses of generalization. Often of conflict, and suggesting options to resolve theinferences were necessary to categorize the re- conflict. The data showed increased use of nego-search. However, the following discussion still tiation behaviors and the reaching of agreementsmay provide a useful organization and concep- in actual parent-child conflict situations at home.tualization of generalization and its program- An assessment of generalization across experi-ming. menters was described by Redd and Birnbrauer1. Train and Hope (1969), who demonstrated that control over the cooperative play of retarded children did not In applied behavior analysis research, the most generalize from an adult who dispensed contin-frequent method of examining generalization, so gent edible reinforcement to five other adultsfar, may be labelled Train and Hope. After a who had not participated in training.behavior change is effected through manipula- Studies that are examples of Train and Hopetion of some response consequences, any existent across time are those in which there was a changegeneralization across responses, settings, experi- from the intervention procedures, either to a lessmenters, and time, is concurrently and/or sub- intensive but procedurally different program, orsequently documented or noted, but not actively to no program or no specifically defined pro-pursued. It is usually hoped that some generali- gram. Data or anecdotal observations were re-zation may occur, which will be welcomed yet ported concerning the maintenance of the origi-not explicitly programmed. These hopeful probes nal behavior change over the specified timefor stimulus and response generalization char- intervening between the termination of theacterize almost half of the applied literature on formal program and the postchecks. An examplegeneralization. The studies have considerable of a followup evaluation was the study by Azrin,importance, for they begin to document the ex- Sneed, and Foxx (1973). An intensive trainingtent and limits of generalization of particular program involving reinforcement of correct toi-operant intervention techniques. While not being leting and positive practice procedures promptly 3Complete reference lists and detailed tables de- decreased bedwetting by 12 retarded persons.scribing subjects, procedures, and generalization of The reduced rate of accidents was maintainedall studies reviewed are deposited with the National during a three-month followup assessment.Auxiliary Publications Service (NAPS). For copies,order NAPS Document #02873. Order from ASIS/ Perhaps there are many more studies in theNAPS Co., C/O Microfiche Publications, 305 East Train and Hope category than would have been46th Street, New York, New York 10017. Remit expected (about 135, of which 65 % are acrosswith order for each copy $3.00 for microfiche or$19.50 for photocopies. Make checks payable to Time). However, despite its obvious value, thisMicrofiche Publications. research is frequently characterized by a lack of
352 TREVOR F. STOKES and DONALD M. BAERcomprehensiveness and depth of the generaliza- lines as being essentially independent of the mod-tion analysis. Even though generalized behavior ified baseline; thus, to report nongeneralizationchange was frequently reported, extensive, wide- would serve no useful purpose, for its nonoccur-ranging, and practical generalization was not rence would be expected. Again, any such docu-often noted or even sought. The continued de- mentation contributes to our understanding ofvelopment of behavior analysis almost surely the extent and limits of generalization, as well aswill demand more extensive collection of gener- serving as an indication of the frequent necessityalization data than is presently the fashion. The of generalization-programming techniques.extent and limits of applied behavioral interven- There is another reason for the predominancetions may be well documented and understood if of positive results in this section: if nongeneral-measurement is extended over longer periods of ization was clearly evident, and the modificationtime, over more than one circumscribed part of of this state was important, then a form of lim-the day, with more than one related response, ited programming was frequently instituted. Ex-and with more than a restricted part of the social amples of this research will be discussed in theand physical environment. It is as important for next category, "Sequential Modification".the field to formalize the conditions of the non-occurrence of generalization as it is to document 2. Sequential Modificationthe conditions associated with the display of un- These studies exemplify a more systematicprogrammed generalization. approach to generalization than the Train-and- Most of the Train-and-Hope research described Hope research. Again, a particular behaviorsuccessful generalization-approximately 90 % change is effected, and generalization is assessed.of Train-and-Hope studies. By definition, there But then, if generalization is absent or deficient,was no further need to program generalization procedures are initiated to accomplish the desiredin those studies where generalization had been changes by systematic sequential modification inexhibited within the Train-and-Hope para- every nongeneralized condition, i.e., across re-digm-presuming, of course, that the generali- sponses, subjects, settings, or experimenters. Thezation exhibited was considered sufficient to meet possibility of unprogrammed generalization typi-the therapeutic goals of the various modification cally was not examined in these sequential modi-programs (not necessarily a valid presumption fication studies, because after the initial demon-in the Train-and-Hope research). This prepon- stration of nongeneralization, all other baselinesderance of positive data may simply reflect the were exhausted. That is, after changes had beentendency of some researchers not to report their produced directly in all baselines, generalizationgeneralization data if measurement procedures to nonrecorded responses, subjects, settings, andwere instituted to probe for any generalized be- experimenters may have occurred, but could nothavior changes, but generalization was shown to be examined.be absent. Some researchers may view nongen- For example, Meichenbaum, Bowers, and Rosseralization as reflecting a deficiency or ineffective- (1968) reported an absence of generalization ofness of their procedures to develop a desirable behavior changes from an afternoon interventiongeneralized performance. Behavior analysts, period to the morning period in a classroom fornevertheless, should encourage their fellow re- institutionalized female adolescent offenders.searchers to document and to analyze experimen- Money dispensed contingent on on-task behav-tally their apparent failures, rather than allowing iors effected desired behavior changes during thethem to slide into oblivion. A detailed and sys- afternoon, but generalization to the morningtematic understanding of generalization and its period required that the same manipulations beprogramming could result. Alternatively, re- applied there as well (sequential modificationsearchers might view their generalization base- across settings). Similarly, generalization across
AN IMPLICIT TECHNOLOGY OF GENERALIZATION 353settings of the disruptive and oppositional behav- ping", where a preschool child was taught anior of two children was investigated by Wahler entry response that exposed the child to the(1969). He demonstrated control of these behav- natural contingencies of peers in the preschooliors in the home by using differential attention environment. Preschool teachers modified theand timeout operations. When generalization to low rate of skillful interaction of the child bythe childrens school behavior was not evidenced, priming others to interact with the subject andsimilar contingency operations were employed to reinforcing appropriate interactions. The dataaccomplish changes in that setting as well. showed that over time the teachers lost control The category of Sequential Modification char- of the interaction behavior, which remainedacterizes much of the actual practice of many high; it was assumed that the groups naturalbehavior analysts. Sequential modification is consequences for interaction had taken controlmerely a systematized experimental procedure of the subjects behavior. Thus, to program gen-that formalizes and allows evaluation of these eralization, the child perhaps needed only to betypical therapeutic endeavors. The tactic of introduced adequately to the natural reinforcersscheduling behavior-change programs in every inherent in active preschool play and interaction.condition to which generalization is desired is Some early analyses of preschool childrens be-frequently employed. The rationale for these havior have stressed that if the child can be soprocedures is as follows. If a desired generaliza- introduced (through the operation of differentialtion is not likely to be exhibited after changing a attention from teachers) to a reinforcing pre-behavior in a particular condition, or a number school natural environment, then the behaviorsof conditions, e.g., settings, then the researcher or eventually do not need to be maintained by con-practitioner works to effect changes across con- tinued contrived modification of the environ-ditions as a matter of course, rather than as an ment. For example, Hall and Broden (1967)outcome of the display or nondisplay of general- modified the manipulative play, climbing, andization. Thus, a behavior analyst is likely to ad- social interaction of three subjects through socialvise the scheduling of consequences in every reinforcement operations. Behavior changes wererelevant condition in preference to the dispens- shown to be durable and successful followuping of consequences in only one or a few condi- data at three months were described.tions, while hoping for generalization, but likely Buell, Stoddard, Harris, and Baer (1968) dem-not seeing it. onstrated the collateral development of appro- priate social behavior (e.g., touching, verbalizing,3. Introduce to Natural Maintaining and playing with other children) accompanying Contingencies the reinforcement of increased use of outdoor Perhaps the most dependable of all general- play equipment by a 3-yr-old girl. This entry re-ization programming mechanisms is one that sponse to the natural reinforcement communityhardly deserves the name: the transfer of behav- was tactically sound because the childs motorioral control from the teacher-experimenter to behavior was modified in a setting where thestable, natural contingencies that can be trusted resulting behavior would tend automatically toto operate in the environment to which the sub- increase social contact with other children, andject will return, or already occupies. To a con- this natural social environment could maintainsiderable extent, this goal is accomplished by the childs new skills, but indeed may also bechoosing behaviors to teach that normally will expected to sharpen and refine them, and addmeet maintaining reinforcement after the teach- entirely new ones as well.ing (Ayllon and Azrin, 1968). Most of the research concerning natural main- Baer and Wolf (1970) reported a study by taining contingencies has involved children, per-Ingram that illustrated the mechanism of "trap- haps because such techniques seem particularly
354 TREVOR F. STOKES and DONALD M. BAERsuitable, especially to their social behavior. Re- by the girls, perhaps on the presumption that thesearch would profit by determining what natural girls were "bad" and not reinforcible in any case.reinforcement communities exist for various be- However, the experimenters were able to teachhaviors and subjects, and what economical means the girls that when their work was objectivelymay be employed to ensure entry to these behav- good, and when staff persons were nearby, aioral traps. simple skill of calling these adults attention to Unfortunately, in some instances there may be their good work would result in fairly consistentno natural reinforcement operating to develop reinforcement. Thus, this was a case in whichand maintain skills. For example, in the case of experimental reinforcement was used to developretarded and institutionalized persons whose de- a response in the subjects that would tap andpendency has become a stable fact in the lives of cultivate the available but dormant natural com-their caretakers, some re-arrangement of the munity. In theory, this new skill should havenatural environment may be necessary. A few obviated the need for further experimental re-studies have introduced subjects to semicontrived inforcement, for the praise evoked should haveor redesigned "natural" reinforcement communi- functioned to maintain both the girls work andties. A simple but meaningful example was pro- cueing, and the cueing, in turn, should have func-vided by Horner (1971), who taught a 5-yr-old tioned to maintain staff praise. The Seymour andinstitutionalized retarded boy to walk on crutches Stokes study could not be continued long enoughin an experimental setting. The child was then to establish whether this would happen, and soprompted to generalize the new walking skill it remains a logically appealing but still unex-to other settings and activities to which he previ- plored method of enhancing generalization:ously had been taken in a wheelchair by solici- teaching the subject a means of recruiting a nat-tous caretakers, by enlisting those caretakers to ural community of reinforcement to maintainrefrain from offering this help. Within 15 days that generalization. Perhaps an even greater ad-after treatment was concluded, the child walked vantage of such procedures is a change in theon crutches to all those activities and settings, locus of control: the subjects can become moreeventually extending his ambulation skills to any prominent agents of their own behavior change,part of his world. Stolz and Wolf (1969) trained rather than being hapless pawns of more-or-lessa 16-yr-old, "blind" retarded male to discriminate random environmental contingencies.visual stimuli. Then, the environment was so Restructuring the environment thus becomes astructured that assistance was not given in situa- target of research aimed at extending the gener-tions where it had previously been given as a alization of newly taught skills; even though, atmatter of course. When the boy was required to a technical level, this operation may not be con-use visual cues to help himself in a cafeteria line, sidered generalization, but rather transfer ofhe soon emitted the necessary behaviors. How- control from one reinforcement contingency toever, these studies did not establish the function- another. In any event, it is a much neglected ality of their procedures in the maintenance of topic of experimental research, although widely behavior changes. recognized as a desirable, and even essential Another significant example was provided by characteristic of any rehabilitative effort. Seymour and Stokes (1976). In their study, insti- Some natural contingencies are inevitably at tutionalized delinquent girls were taught to so- work contributing to the maintenance of inap- licit reinforcement (cf. Graubard, Rosenberg, propriate behavior. For example, peer-group and Miller, 1971) from their natural community, control of inappropriate behavior has often been the staff of their residential institution. In their suspected and sometimes documented (Buehler, case, the staff had rarely displayed any systematic Patterson, and Furniss, 1966; Gelfand, Gelfand, attempts at reinforcing desirable behavior shown and Dobson, 1967; Solomon and Wahler, 1973).
AN IMPLICIT TECHNOLOGY OF GENERALIZATION 355It would seem reasonable, then, that if the pat- alization to only a few (and often only one)tern of reinforcement of inappropriate behavior extraexperimental responses, subjects, settings,is modified, the observed outcome may errone- experimenters, or times. When the absence ofously, but happily be attributed to generalization. generalization was noted, sometimes it was ac-For example, Bolstad and Johnson (1972) pre- complished by further direct intervention insented data that showed that both experimental every nongeneralized condition (i.e., Sequentialand control subjects in the same classroom were Modification). Having completed such modifica-all affected (although not to the same extent) by tions, the possibility of more extensive general-experimental manipulation of the reinforcement ized effects (i.e., beyond the two or three modifiedcontingency for the experimental subjects, baselines) was not examined. In the training ofwhereas control subjects in a different classroom sufficient exemplars, generalization to untrainedwere not so affected. The authors presented data stimulus conditions and to untrained responsesthat may account for these differences. The con- is programmed by the training of sufficient ex-trol subjects in the experimental classroom, who emplars (rather than all) of these stimulus con-were also disruptive students, had fewer disrup- ditions or responses.tive interactions with the experimental subjects A systematic demonstration of programmedduring the treatment phases than during base- generalization and measurement of generalizedline. This possible generalization effect may be effects beyond intervention conditions was re-due to the disruption of the natural contingencies ported by Stokes, Baer, and Jackson (1974). Theyoperating in that environment. That is, other established that training and maintenance of re-disruptive students previously supported some of tarded childrens greeting responses by one ex-the disruptive behavior of the control subjects, perimenter was not usually sufficient for thebut during treatment these experimental subjects generalization of the response across experi-did not support the disruptive behavior of their menters. However, high levels of generalizationpeers and, thus, a "generalized" decrease in dis- to over 20 members of the institution staff (andruptive behavior by the control subjects resulted. newcomers as well) who had not participated in the training of the response were recorded, 4. Train Sufficient Exemplars after a second experimenter trained and main- If the result of teaching one exemplar of a tained the response in conjunction with the firstgeneralizable lesson is merely the mastery of the experimenter. Thus, when generalization did notexemplar taught, with no generalization beyond prevail after the training of one stimulus exem-it, then the obvious route to generalization is to plar, it was programmed by training a greaterteach another exemplar of the same generaliza- diversity of stimulus (trainer) conditions. Simi-tion lesson, and then another, and then another, larly, Garcia (1974) taught a conversationaland so on until the induction is formed (i.e., until speech form to two retarded children, and, upongeneralization occurs sufficiently to satisfy the discovering a lack of stable generalization acrossproblem posed). Examples of such programming experimenters after one training input, pro-techniques will be described in this category of grammed generalization across experimenters bytraining sufficient exemplars, perhaps one of the having a second experimenter teach the samemost valuable areas of programming. Certainly responses.it is the generalization-programming area most A sufficient-stimulus-exemplars demonstrationprominent and extensive in the present literature. of programmed generalization across settings has In the research discussed previously under been described by Allen (1973). Allen modifiedthe categories of Train and Hope and Sequen- the bizarre verbalizations of an 8-yr-old boy bytial Modification, the typical analysis of gener- differential attention procedures. Ignoring bi-alization concerned the measurement of gener- zarre verbalizations and praise for appropriate
356 TREVOR F. STOKES and DONALD M. BAERinteraction reduced bizarre verbalizations during Although very little research has been re-evening camp activities. However, there was no ported, the analysis of generalization program-generalization to three other camp settings. After ming by training in a number of settings is aadditional training in a second setting, some virtually untapped area of far-reaching value.generalization to the unmanipulated settings was However, consistent optimism should follow ex-noted. This generalization was further enhanced amination of the studies showing generalizationby intervention in the third setting. Unfortu- after training in only a few settings. Unfortu-nately, the experimental procedures did not allow nately, behavior analysts seem too often satisfiedsufficient time to document the full extent of with the modification of a single, well-definedgeneralization after training in two settings, but behavior in one setting, e.g., a laboratory pre-generalization after training in two settings was school. Discriminated programs are often accept-clearly evident. Griffiths and Craighead (1972) able, and sometimes even desirable. When gener-similarly programmed generalization across set- alization is a valid concern, but researchers andtings. A 30-yr-old retarded woman received practitioners do not act as if this were so, thepraise and tokens for correct articulation in discriminated behavior of researchers is mostspeech therapy. Generalization to a residential probably inhibitory to the development of ancottage was not observed until the same proce- effective generalization technology.dures were instituted there. Following training Over the past 10 yr, there has developed anin these two stimulus exemplars, generalization extensive literature discussing the programmedto a third nontraining setting (a classroom) was generalization of responses through the trainingobserved. of sufficient response exemplars. A response class Very little research concerned with generaliza- has been operationally defined to describe thetion programming has dealt with the training of fact that some responses are organized such thatsufficient stimulus exemplars. The infrequent operations applied to a subset of responses in theresearch that has been published is characterized class affect the other members of that class in thelargely by programming across experimenters. same manner. For example, Baer, Peterson, andThis work has been promising, for after a modest Sherman (1967) reinforced various motor imita-number of training inputs, generalization appar- tions by retarded children. They found that asently will occur with persons not involved in long as reinforcement followed some imitativetraining-unquestionably a valuable and inex- responses, other imitations continued to be per-pensive outcome. However, the present implica- formed without training or reinforcement.tion of these studies is limited because of the A topographical analysis of generalized imita-restricted nature of the type of subjects and tion has been made by Garcia, Baer, and Fire-responses analyzed. Further work is also needed stone ( 1971). Four retarded children were trained to give direction to the optimal conditions to imitate three different topographical types ofwhereby the most extensive generalization will response: small motor, large motor, and short be achieved with a minimal training expendi- vocal. These subjects were also probed for their ture. Nevertheless, it is optimistic to note how imitation of other unreinforced responses: short frequently a sufficient number of exemplars is motor, long motor, short vocal, and long vocal. a small number of exemplars. Frequently, it is Generalized imitation was observed with each no more than two. In particular, there may well subject, but this generalization reflected the par- be reason to suspect that the use of two trainers ticular dimensions of the topographical response will yield excellent results in terms of generaliza- currently being trained or having previously re- tion. This possibility, obviously an economical ceived training. Thus, generalization may occur one, certainly merits systematic study of its po- within well-defined classes and may not gener- tential and limits. alize to other classes unless some special training
AN IMPLICIT TECHNOLOGY OF GENERALIZATION 357(generalization programming) occurs within that exemplars should undergo training. That is, toclass as well. These data depict one possible limi- program the generalized performance of certaintation of the generality of generalized imitation, responses across various setting conditions oras well as pointing to the need to train response persons, training should occur across a (suffi-exemplars that will adequately reflect the di- cient) number of setting conditions and/or withversity of the generalization being programmed. various persons. In a similar manner, generaliza- Childrens grammatical development has been tion across responses can be programmed reliablyanother prominent area of research dealing with by the training of a number of responses. Diver-generalized behavior. The concept of response sity of exemplars seems to be the rule to followclass is again pivotal in these studies, which in pursuit of the maximum generalization. Suffi-conceptualize the rules of morphological gram- cient diversity to reflect the dimensions of themar as equivalent to response class phenomena. desired generalization is a useful tactic. However,For example, Guess, Sailor, Rutherford, and diversity may also be our greatest enemy: tooBaer (1968) developed the generative correct much diversity of exemplars and not enoughuse of plurals by a retarded girl. After teaching (sufficient) exemplars of similar responses maya number of exemplars of the correct plural make potential gains disproportional to the in-response, the girl appropriately labelled new vestment of training effort. The optimal combi-objects in the singular or plural without further nation of sufficient exemplars and sufficient di-direct training relevant to those objects. Plural versity to yield the most valuable generalizationusage had become a generalized response class; is critically in need of analysis. Is the best pro-the morphological rule had been established. cedure to train many exemplars with little diver-Schumaker and Sherman (1970) rewarded three sity at the outset, and then expand the diversityretarded children for the correct production of to include dimensions of the desired generaliza-past- and present-tense forms of verbs. As past- tion? Or is it a more productive endeavor toand present-tense forms of verbs within an in- train fewer exemplars that represent a greaterflectional class were modified, there occurred a diversity, and persist in the training until gen-generalized usage of untrained verbs to similar eralization emerges?tense forms. There has been considerable research to estab-lish the importance of the training of sufficient 5. Train Looselyresponse exemplars. A survey of these (approxi- One relatively simple technique can be con-mately 60) studies shows that the number of ceptualized as merely the negation of discrimi-exemplars found to be "sufficient" for a desirable nation technique. That is, teaching is conductedlevel and durability of generalization varies with relatively little control over the stimuliwidely, probably determined primarily by the presented and the correct responses allowed, sonature of the task and the subjects prior skills as to maximize sampling of relevant dimensionsrelevant to it. Most of this research was con- for transfer to other situations and other formscerned with the development of motor and vocal of the behavior. A formal example of this mostimitations, and the beginning development of often informal technique was provided bygrammar and syntax. The development of ques- Schroeder and Baer (1972), who taught vocaltion-asking and instruction-following is also well imitation skills to retarded children in both ofrepresented. two ways, one emphasizing tight restriction of In conclusion, examination of the sufficient the vocal skills being learned at the momentexemplar research points to a significant (and (serial training of vocal imitations), and thelong-familiar) generalization-programming pro- other allowing much greater range of stimulicedure: a number of stimulus and/or response within the current problem (concurrent training
358 TREVOR F. STOKES and DONALD M. BAERof imitations). The latter method was charac- cannot discriminate in which settings a responseterized repeatedly by greater generalization to will be reinforced or not reinforced. A potentialas-yet-untaught vocal imitation problems, thus approximation to such a condition was presentedaffirming "loose" teaching techniques as a con- in a study by Schwarz and Hawkins (1970).tributor to wider generalization. In that experiment, the behavior of a sixth-grade It will be appreciated that the literature of the child was videotaped during math and spellingfield contains very few examples of this type. classes. Later, after each school day had ended,Researchers always have attempted to maintain the child was shown the tape of the math classthorough control and careful restriction and and awarded reinforcers according to how oftenstandardization of their teaching procedures, good posture, absence of face-touching, and ap-primarily to allow easy subsequent interpretation propriate voice-loudness were evident on thatof the nature of their (successful) teaching tech- tape. Although reinforcers were awarded onlyniques. Yet the import of this technique is that on the basis of behaviors displayed during thecareful management of teaching techniques to a math class, desirable improvements were ob-precisely repetitive handful of stimuli or formats served during the spelling class as well. In thatmay, in fact, correspondingly restrict generaliza- reinforcement was delayed, this technique musttion of the lessons being learned. The ultimate have made it difficult for the child to discriminateforce of this recommendation remains to be seen. in which class the behaviors were critical forWhat seems required is programmatic research earning reinforcement. In other words, the gen-aimed at assessing the generalization character- eralized success of the study may well be at-istics of lessons taught under careful, restricted tributable to the partly indiscriminable natureconditions, relative to similar lessons taught of the reinforcement contingency.under looser, more variable conditions. In general, it may be suspected that delayed reinforcement often will have the advantage of making the times and places in which the con-6. Use Indiscriminable Contingencies tingency actually operates indiscriminable to the Intermittent schedules of reinforcement have subject. However, this advantage is an advantage,been shown repeatedly to be particularly resistant by hypothesis, primarily for the goal of general-to extinction, relative to continuous schedules ization. Otherwise, delayed reinforcement would(Ferster and Skinner, 1957). Resistance to extinc- often be considered an inefficient technique, mosttion may be regarded as a form of generaliza- especially so for the initial development of a newtion-generalization across time subsequent to skill. Indeed, it may be exactly in the realm oflearning. The essential feature of intermittent disadvantaged persons such as retarded childrenschedules may be their unpredictability-the that the usual inefficiency of delayed reinforce-impossibility of discriminating reinforcement oc- ment may seem the most severe handicap to itscasions from nonreinforcement occasions until use. However, its potential for fostering general-after the fact. Thus, if contingencies of reinforce- ization suggests strongly that further research bement or punishment, or the setting events that invested in this procedure (and any others thatmark the presence or absence of those contingen- make reinforcement contingencies properly in-cies, are made indiscriminable, then generaliza- discriminable), to develop methods of applyingtion may well be observed. it perhaps only after the initial development of a In generalization, behavior occurs in settings in new skill, in the interests of promoting gen-which it will not be reinforced, just as it does in eralization.settings in which it will be reinforced. Then, the Less than a dozen studies of generalizationanalogue to an intermittent schedule, extended interpretable as cases of indiscriminable rein-to settings, is a condition in which the subject forcement contingencies can be found in the
AN IMPLICIT TECHNOLOGY OF GENERALIZATION 359 literature. Kazdin (1973), for example, showed frequently with the subject with whom general-that teacher attention to one retarded child was ization was evidenced. Thus, with this subject,responded to by another child as if it were rein- punishment of the generalization response oc- forcement for on-task behavior. Indeed, the curred more frequently when destructive behav-onlooker reacted with increased on-task behavior, ior was punished. Unfortunately, it was noteven when the teacher attended to the target determined how often the self-biting occurredchilds off-task behavior. Possibly, prior experi- at times not simultaneous with the destructiveence with reinforcement contingent on the peers behavior. Therefore, the schedule of punishmenton-task behavior was sufficient to make all future for self-biting was not established, i.e., whetherpraise (contingent or not) discriminative for on- biting occurred only when destructive behavior task behavior. In other words, with sufficient occurred and, therefore, always met the timeoutprior experience, the onlooker may have stopped contingency. In this example (which was notobserving the contingency in which the rein- intended to be a careful analysis of the indis-forcement operated and responded only to the criminable reinforcement concept), not only wasreinforcing stimulus presence, making the con- the reinforcement contingency somewhat diffi-tingency functionally indiscriminable. cult to discriminate, but the two behaviors (de- Generalization across subjects has similarly structive and self-destructive responses) also maybeen reported by Broden, Bruce, Mitchell, Carter, well have been only somewhat differentiated byand Hall (1970) in a classroom of culturally the subject.disadvantaged children. When positive teacher Thus, preventing the ready discrimination ofattention was given for one childs attention to contingencies is a generalization-programmingacademic work, the attending of a peer also in- technique worthy of application and research.creased. This generalization was also a probable Perhaps a random or haphazard delivery of re-function of the cueing properties of teacher rein- inforcement will (if luck or good judgementforcement. However, the generalization observed prevails) function to modify targetted behaviormay also have been due to the manipulation of as well as behavior occurring in proximal timenatural social consequences received by the non- or space. Even noncontingent reinforcement,target child through peer attention, or may have delivered at the outset of an intervention pro-been caused by a slight increase in the amount gram, may retard initial effects, but may workof teacher attention to the nontarget child. These to later advantage in generalization outcomes.effects deserve further systematic evaluation be- Finally, Kazdin and Polster (1973) showedcause of their relevance to the classroom prac- once again the usefulness of intermittent sched-tices of many teachers who strive to instruct ules to delay subsequent extinction, relative toeffectively but are unable to devote extensive continuous schedules of reinforcement. Socialtime to individual children. interaction by two retardates was reinforced with Pendergrass (1972) showed that timeout could tokens. After establishing social interaction, onebe employed to decrease the destructive behavior subject received continuous reinforcement andof two retarded children. With one subject, de- the other, intermittent reinforcement. Duringcreased rates were also observed with another extinction, only the subject who received inter-response (self-biting) which was sometimes mittent reinforcement continued to interact so-chained to the destructive behavior, but not cially with peers. However, these results mayitself subjected to contingent timeout. However, simply reflect different extinction rates by twowith the second subject, generalization to a sec- subjects. The research was essentially a groupond response (autistic jerking movement) was study where N 1. Adequate single-subject ex-not observed. Analysis of the data revealed that perimental control was lacking. Therefore, repli-the two behaviors occurred simultaneously more cation of these procedures would be desirable.
360 TREVOR F. STOKES and DONALD M. BAER eralization setting. Autistic children were re-7. Program Common Stimuli warded for imitation and instruction-following The passive approach to generalization de- in a training setting. Four of their 10 subjectsscribed earlier need not be a completely imprac- then did not exhibit generalization to a differenttical one. If it is supposed that generalization setting. Therefore, to program for this general-will occur, if only there are sufficient stimulus ization, various aspects of the training procedurescomponents occurring in common in both the (e.g., hand movement by therapist) or physicaltraining and generalization settings, then a rea- training environment (e.g., table and chairs)sonably practical technique is to guarantee that were systematically introduced to the generaliza-common and salient stimuli will be present in tion setting to control generalization. Makingboth. One predictor of the salience of a stimulus the experimental setting more closely resembleto be chosen for this role is its already established the regular classroom (generalization setting)function for other important behaviors of the was the programming procedure employed bysubject. Koegel and Rincover (1974). They decreased Childrens peers may represent peculiarly the teacher-to-student ratio in the experimentalsuitable candidates for a stimulus common to setting from 1-to-i to 1-to-8. After these specialboth training and generalization settings. An programming conditions were instituted, thereexample has been provided by Stokes and Baer was increased performance on previously learned(1976). In their study, two children exhibiting and new behaviors learned in the classroom.serious learning disabilities were recruited to Walker and Buckley (1972) programmed gener-learn several word-recognition skills. One child alization of the effects of remedial training ofwas taught these skills and concurrently shown social and academic classroom behavior by estab-how to teach them to the other child, thus acting lishing common stimuli between the experimen-as a peer-tutor. It was found that both children tal remedial classroom and the childrens regularreliably learned the skills, but that neither gen- classroom by using the same academic materialseralized them reliably or stably to somewhat dif- in both classrooms.ferent settings in which the other child usually The literature of this field shows only a hand-was absent. However, when the peer-tutor was ful of studies deliberately making use of a com-brought into those settings, then each child simi- mon stimulus in both training and generalizationlarly showed greatly increased and stabilized settings. Obviously, this is a technological dimen-generalization, even though there were never sion urgently in need of thorough development.any consequences for generalization. Similar The use of peers as the common stimulus hasdemonstrations have been provided by Johnston much to recommend it as a practical and naturaland Johnston (1972) for the skill of speech technique. To what extent peers need to partici-articulation. In that study, peers were rewarded pate in the training setting has not yet beenfor correct monitoring of the subjects articula- determined, although the absence of generaliza-tion. Generalization of correct articulation oc- tion sometimes shown when peers are presentcurred only when the "monitoring" peer was in nontraining settings, suggests that peers notpresent. Unfortunately, it was not determined involved in a training setting will not likelyclearly whether generalization was evidenced be- acquire sufficient discriminative function to con-cause of the discriminative properties of the trol generalized responding. The use of commonpeers presence in both settings, or whether the physical stimuli is in even greater need of sys-peers actively continued their monitoring in the tematic research. A common stimulus approachgeneralization setting. to generalization would encourage the incorpora- Rincover and Koegel (1975) have also incor- tion into training settings of (naturally occur-porated functional training stimuli into the gen- ring) physical stimuli that are frequently promi-
AN IMPLICIT TECHNOLOGY OF GENERALIZATION 361nent or functional in nontraining environments. was restricted to true reports, the reports thenIf these stimuli are well chosen, and can be made became mediators of play behavior. The lessonfunctional and salient in the training procedures, generalized, such that after several sequentialthen generalization may thereby be programmed. experiences with these procedures, the children then used reports about play as mediators, even without reinforcement being restricted to only8. Mediate Generalization true reports. Israel and OLeary (1973) used Mediated generalization is well known as a essentially the same paradigm to compare thetheoretical mechanism explaining generalization effects of having children report first what theyof highly symbolic learnings (Cofer and Foley, would play with later, in contrast to having them 1942). In essence, it requires establishing a re- report after play what they had done (the Risley sponse as part of the new learning that is likely and Hart method); they found that reinforcing to be utilized in other problems as well, and will postreports (when they were true) produced moreconstitute sufficient commonality between the actual behavior (the next day) than reinforcingoriginal learning and the new problem to result the actual behavior when it agreed with the in generalization. The most commonly used earlier promise to perform it. This technique hasmediator is language, apparently. However, the been extended subsequently to the case of socialdeliberate application of language to accomplish skills, specifically sharing and praising betweengeneralization is rare in the literature reviewed, young children (Rogers-Warren and Baer, 1976).and correspondingly little is known about what In that case, modelling was added, such that theaspects of a language response make for best young children would have a thorough chancemediation. to learn the nature of the relatively complex A sophisticated analysis of mediated general- responses at issue.ization was conducted by Risley and Hart ( 1968), Obviously, verbal mediation can easily fail,who taught preschool children to report at the most especially in those situations in which theend of play on their play-material choices. Men- verbal mediators have little meaning (i.e., tightlytion of a given choice was reinforced with snacks, restricted discriminative value) for the subjects.which produced increased mentioning of that It is commonplace to find children agreeing to achoice, but no change in the childrens actual query (e.g., about whether they praised or shared)use of that play-material. When reinforcement without any knowledge of what that must entailwas restricted to true reports of play-material in actual behavior. In the case of retarded chil-choices, however, the children then changed their dren, it might be particularly true that the abilityplay behavior (the next day) so that when to use verbal responses as mediators would lagqueried about that play, they could truthfully behind that of normal children using the samereport on their use of the specified play material language responses. It may be reasonable toand earn reinforcement. Control over any choice suggest that in the development of language-of play materials proved possible with this tech- training programs, systematic attention be givennique, which placed teaching contingencies not to the training of language skills sufficiently wellon the play, but on a potential mediator (verbal elaborated to function as mediators of nonverbalreport) of that play behavior. That the reports behavior. Language is a response, of course; it iswere only potential mediators was apparent in also, equally obviously, a stimulus to the speakerthe early stages of the study, when the children as well as to the listener. Thus, it meets perfectlyreadily reported (untruly) their use of play ma- the logic of a salient common stimulus, to beterials with no corresponding actual behavior carried from any training setting to any general-with those materials; at that stage, they earned ization setting that the child may ever enter.reinforcement even so. When the reinforcement It also perfectly exemplifies the essence of the
362 TREVOR F. STOKES and DONALD M. BAERactive generalization approach recommended and maintenance, especially when formal inter-earlier. vention manipulations have ceased to operate. The mediation of generalization is also exem- The effects of accompanying procedures shouldplified in the behavior analysis research of self- be experimentally separated from self-controlcontrol and self-management procedures. That effects, and the role of each of the various self-is, self-control procedures such as self-recording, control tactics (Glynn et al., 1973) should betaught as part of an intervention program, may individually analyzed. The potential of self-function to promote generalization: such tech- mediated generalization is apparent, but its im-niques are easy to transport and may be em- plications and practical utility still remain to beployed readily to facilitate responding under assessed.generalization conditions. Some research that hasemployed any or all of the various tactics of self- 9. Train "To Generalize"assessment, self-recording, self-determination of If generalization is considered as a responsereinforcement, and/or self-administration of re- itself, then a reinforcement contingency may beinforcement (Glynn, Thomas, and Shee, 1973), placed on it, the same as with any other operant.has also displayed maintenance and generaliza- Informally, teachers often do this when they urgetion of behavior change; however, the correla- a student who has been taught one example oftion is not perfect. a general principle to "see" another example Broden, Hall, and Mitts (1971) reported that as "the same thing". (In principle, they are alsoafter an eighth-grade girl experienced self- attempting to make use of language as a medi-recording of study behavior and teacher praise ator of generalization, relying on the supposedfor improved study, her study behavior main- characteristics of words like "same" to accom-tained at a high level for a recorded three weeks. plish the generalization.) Common observationAlthough the individual effects of the self-record- suggests that the method often fails, and thating and praise were not determined, it is possible when it does succeed, little extrinsic reinforce-that the self-recording procedures contributed ment is offered as a consequence. A more formalsignificantly to this generalization. example of the technique was seen in a study Drabman, Spitalnik, and OLeary (1973) by Goetz and Baer (1973), in which three pre-taught disruptive children to match their teach- school children were taught to generalize theers evaluations of their appropriate classroom response of making block forms (in blockbuild-behavior. Tokens were dispensed for appropriate ing play). Descriptive social reinforcement wasclassroom behavior and accurate matching. Dis- offered only for every different form the childruptive classroom behavior decreased and was made, i.e., contingent on every first appearancemaintained at low levels during a 12-day phase of any blockbuilding form within a session, butwhen tokens were not dispensed for self-record- not for any subsequent appearances of that form.ing accuracy. Generalized behavior improvement Thus, the child was rewarded for moving alongwas also evident during a 15-min no-token the generalization gradient underlying block-period within the experimental hour. These form inventions, and never for staying at anychanges were possibly a function of the close one point. In general, the technique succeeded, intemporal proximity of the token periods, which that the children steadily invented new block frequently immediately preceded or followed forms while this contingency was in use. Thus,the generalization period. there exists the possibility of programming rein- The role of self-control procedures in medi- forcement specifically, perhaps only, for move- ating generalization has often been proposed. ment along the generalization gradient desired. Research would do well to examine the contri- In largely unspecified ways, perhaps two other bution of self-control tactics in generalization studies exemplify this logic. Herbert and Baer
AN IMPLICIT TECHNOLOGY OF GENERALIZATION 363 (1972), for example, taught two mothers of technique that is being practiced when a clientdeviant children to give social reinforcement is taught to relax in a somewhat anxiety-arousingonly to their childrens appropriate behaviors, situation, and reinforced (socially) for doing so; but taught the mothers from the outset to judge and then is instructed to relax in a somewhat all behavior according to criteria they helped more powerful anxiety-arousing situation, etc.to develop, rather than attack only a few speci- That is, systematic desensitization to a heirarchyfied child responses. These mothers learned a of stimuli may be analyzed as reinforcing not justgeneralized skill because they applied correct relaxation, but also generalization along an al-social contingencies to categories that included ready constructed generalization gradient (cf.virtually all appropriate child behavior likely to Yates, 1970, p. 64ff.).occur. Behavior changes were maintained at 20and 24 weeks after completion of formal train-ing. Similarly, Parsonson, Baer, and Baer (1974) CONCLUSIONtaught two teachers of retarded children to apply The structure of the generalization literaturegeneralized correct social contingencies to all and its implicit embryonic technology has beenlikely appropriate and inappropriate behaviors summarized. The most frequent treatments ofof preschool retarded children. These effects were generalization are also the least analytical-thosealso durable over several months. Apparently described as Train and Hope and Sequentialgeneralized changes were produced in these stud- Modification. Included in the category of Trainies by Herbert and Baer and Parsonson et al., and Hope were those studies where the potentialbut the extent and quality of that generalization for generalization had been recognized, its pres-was not quantified as such. ence or absence noted, but no particular effort Very few studies of this type are found in the was expended to accomplish generalization. Byliterature of applied behavior analysis, probably contrast, some limited programming was imple-because of the preference of behaviorists to con- mented in the Sequential Modification research.sider generalization as an outcome of behavioral In these studies, given an absence of reliablechange, rather than as a behavior itself. Ulti- generalization, procedures to effect changes weremately, this behavioristic stance may well prove instituted directly in every nongeneralized condi-durable and consistent. Meanwhile, it is worth tion. Although contributing significantly to ourhypothesizing that "to generalize" may be treated understanding of the generalization of behavior-as if it were an operant response, and reinforced change programs, these studies are not examplesas such, simply to see what useful results occur. of the programming of generalization. Consequently, one other technique deserves Seven categories were discussed that directlydiscussion: the systematic use of instructions to relate to a technology of generalization. First,facilitate generalization. Thus, if a behavior is the potential role of Natural Maintaining Con-taught and generalization is not displayed, the tingencies was discussed. According to this tactic,least expensive of all techniques is to tell the generalization may be programmed by suitablesubject about the possibility of generalization trapping manipulations, where responses are in-and then ask for it. If that generalization then troduced to natural reinforcement communitiesoccurs, it may well be referred to as "instructed that refine and maintain those skills withoutgeneralization". If the effects of that instruction further therapeutic intervention. The Trainingare themselves to become generalized (yielding of Sufficient Exemplars is numerically the mosta "generalized generalizer"?), then reinforcement extensive area of programming: generalizationof the generalized behavior, on a suitable sched- to untrained stimulus conditions and to un-ule, might well be prudent, at least at first. Per- trained responses is programmed by the traininghaps it is simply a very elaborate version of this of sufficient exemplars of those stimulus condi-
364 TREVOR F. STOKES and DONALD M. BAERtions or responses. Train Loosely is a program- 4. Make unclear the limits of training contin-ming technique in which training is conducted gencies; in particular, conceal, when pos-with relatively little control over the stimuli and sible, the point at which those contingen-responses involved, and generalization is thereby cies stop operating, possibly by delayedenhanced. To invoke the tactic of Indiscrimi- reinforcement.nable Contingencies, the contingencies of re- 5. Use stimuli that are likely to be found ininforcement or punishment, or the setting events generalization settings in training settingsmarking the presence or absence of those con- as well; in particular, use peers as tutors.tingencies, are deliberately made less predictable, 6. Reinforce accurate self-reports of desirableso that it becomes difficult to discriminate rein- behavior; apply self-recording and self-forcement occasions from nonreinforcement oc- reinforcement techniques whenever possi-casions. Common Stimuli may be employed in ble.generalization programming by incorporating 7. When generalizations occur, reinforce atinto training settings social and physical stimuli least some of them at least sometimes, asthat are salient in generalization settings, and if "to generalize" were an operant responsethat can be made to assume functional or obvious class.roles in the training setting. Mediated General-ization requires establishing a response as part There are many examples of generalizationof new learning that is likely to be utilized in and nongeneralization of behavior changes. Theother problems as well, and thus result in gener- fact that apparently unprogrammed generaliza-alization. The final technique, Train "To Gener- tion has been demonstrated (particularly acrossalize", involves reinforcing generalization itself time) is valuable. It heralds a practicality de-as if it were an explicit behavior. These program- sirable in any technology of behavior: that everyming techniques should be researched further one of a subjects responses, in every setting,and usefully applied in programs in which gen- with every experimenter, and at every conceiv-eralization is relevant. able time does not need to meet specific treat- This list of generalized tactics conceals within ment consequences for that program to accom-itself a much smaller list of specific tactics. These plish and maintain important behavior changes.specific tactics can be presented as a small pic- Alternatively, the fact that generalization is notture of the generalization technology in its pres- always observed and durability is not inevitableent most pragmatic form, not only to offer a set means that there is hope for behavior modifica-of what-to-do possibilities, but also to emphasize tion: behavior can always be modified andhow very small the current technology is and changes are not necessarily irreversible. That is,how much development it requires: once behavior has been modified, there is still the possibility of reconditioning if changes are 1. Look for a response that enters a natural undesirable or inappropriate, or if new inappro- community; in particular, teach subjects to priate behaviors develop. If both appropriate cue their potential natural communities to and inappropriate behavior changes were to per- reinforce their desirable behaviors. sist and prove irreversible, it would presage the 2. Keep training more exemplars; in particu- demise of any technology of behavioral inter- lar, diversify them. vention. This occurrence of nongeneralization 3. Loosen experimental control over the stim- also underlines the need to develop a technology uli and responses involved in training; in of generalization, so that programming will be particular, train different examples concur- a fundamental component of any procedures rently, and vary instructions, SDs, social when durability and generalization of behavior reinforcers, and backup reinforcers. changes are desirable.
AN IMPLICIT TECHNOLOGY OF GENERALIZATION 365 A most important question is prompted by an be descriptive only of the active regard of re-examination of the previous research: does gen- searchers and practitioners.eralization ever occur without programming?In the above research, generalization was not al- REFERENCESways evident. In fact, the highly discriminated Allen, G. J. Case study: Implementation of behav-effects of some operant programs were some- ior modification techniques in summer camp set-times documented. We have seen that the behav- tings. Behavior Therapy, 1973, 4, 570-575.ior analysis literature describes various programs Ayllon, T. and Azrin, N. H. The token economy. New York: Appleton-Century-Crofts, 1968.that have shown that generalization may be pro- Azrin, N. H., Sneed, T. J., and Fox, R. M. Dry-bed:moted or programmed by particular intervention A rapid method of eliminating bedwetting (enu-techniques. It seems reasonable to suggest, then, resis) of the retarded. Behaviour Research and Therapy, 1973, 11, 427-434.that many of the successful Train-and-Hope Baer, D. M., Peterson, R. F., and Sherman, J. A. Theexamples cited above may be undiagnosed in- development of imitation by reinforcing behav-stances of informal or inadvertent programming ioral similarity to a model. journal of the Experi-techniques, rather than an absence of program- mental Analysis of Behavior, 1967, 10, 405-416. Baer, D. M. and Wolf, M. M. The entry into natu-ming techniques. It cannot be discounted, and is ral communities of reinforcement. In R. Ulrich,indeed possible, that these generalization exam- T. Stachnik, and J. Mabry (Eds), Control of hu-ples may simply depict successful programmed man behavior: Volume II. Glenview, Illinois: Scott, Foresman, 1970. Pp. 319-324.generalization, and neither the authors of those Baer, D. M., Wolf, M. M., and Risley, T. R. Somepapers, nor the present authors have recognized current dimensions of applied behavior analysis.or hypothesized the programming technique. Journal of Applied Behavior Analysis, 1968, 1, 91-97. Perhaps the most pragmatic orientation for Bolstad, 0. D. and Johnson, S. M. Self-regulation inbehavior analysts is to assume that generalization the modification of disruptive behavior. Journaldoes not occur except through some form of pro- of Applied Behavior Analysis, 1972, 5, 443-454.gramming. Thus, the best course of action seems Broden, M., Hall, R. V., and Mitts, B. The effect of self-recording on the classroom behavior of twoto be that of systematic measurement and analy- eighth-grade students. Journal of Applied Behav-sis of variables that may have been functional in ior Analysis, 1971, 4, 191-199.any apparently unprogrammed generalization. Broden, M., Bruce, C., Mitchell, M. A., Carter, U., and Hall, R. V. Effects of teacher attention onThese analyses should be included as part of all attending behavior of two boys at adjacent desks.research where "unprogrammed" generalized be- Journal of Applied Behavior Analysis, 1970, 3,havior changes are evidenced, for discriminated 199-203. Buehler, R. E., Patterson, G. R., and Furniss, J. M.behavior changes may well be the rule if gen- The reinforcement of behavior in institutional set-eralization is not specifically programmed. Such tings. Behaviour Research and Therapy, 1966, 4,analyses, if successful, will contribute to a tech- 157-167.nology of generalization by further developing Buell, J., Stoddard, P., Harris, F. R., and Baer, D. M. Collateral social development accompanying rein-the understanding of critical variables that func- forcement of outdoor play in a preschool child.tion to produce generalization, and would further Journal of Applied Behavior Analysis, 1968, 1,emphasize the need always to be concerned not 167-173. Cofer, C. N. and Foley, J. P. Mediated generaliza-only with generalization issues, but with the vari- tion and the interpretation of verbal behavior: I.ous techniques that accomplish generalization. Prolegomena. Psychological Review, 1942, 49, In other words, behavioral research and prac- 513-540.tice should act as if there were no such animal Drabman, R. S., Spitalnik, R., and OLeary, K. D. Teaching self control to disruptive children.as "free" generalization-as if generalization Journal of Abnormal Psychology, 1973, 82, 10-never occurs "naturally", but always requires 16.programming. Then, "programmed generaliza- Ferster, C. B. and Skinner, B. F. Schedules of rein- forcement. New York: Appleton-Century-Crofts,tion" is essentially a redundant term, and snould 1957.
366 TREVOR F. STOKES and DONALD M. BAERGarcia, E. The training and generalization of a inforcement and response maintenance in extinc- conversational speech form in nonverbal retardates. tion. Behavior Therapy, 1973, 4, 386-391. Journal of Applied Behavior Analysis, 1974, 7, Keller, F. S. and Schoenfeld, W. N. Principles of 137-149. psychology. New York: Appleton-Century-Crofts,Garcia, E., Baer, D. M., and Firestone, I. The devel- 1950. opment of generalized imitation within topo- Kifer, R. E., Lewis, M. A., Green, D. R., and Phillips, graphically determined boundaries. Journal of E. L. Training predelinquent youths and their Applied Behavior Analysis, 1971, 4, 101-112. parents to negotiate conflict situations. Journal ofGelfand, D. M., Gelfand, S., and Dobson, W. R. Applied Behavior Analysis, 1974, 7, 357-364. Unprogrammed reinforcement of patients behav- Koegel, R. L. and Rincover, A. Treatment of psy- ior in a mental hospital. Behaviour Research and chotic children in a classroom environment: I. Therapy, 1967, 5, 201-207. Learning in a large group. Journal of AppliedGlynn, E. L., Thomas, J. D., and Shee, S. M. Behav- Behavior Analysis, 1974, 7, 45-59. ioral self-control of on-task behavior in an ele- Meichenbaum, D. H., Bowers, K. S., and Ross, R. R. mentary classroom. Journal of Applied Behavior Modification of classroom behavior of institu- Analysis, 1973, 6, 105-113. tionalized female adolescent offenders. BehaviourGoetz, E. M. and Baer, D. M. Social control of form Research and Therapy, 1968, 6, 343-353. diversity and the emergence of new forms in chil- Parsonson, B. S., Baer, A. M., and Baer, D. M. The drens blockbuilding. Journal of Applied Behav- application of generalized correct social contin- ior Analysis, 1973, 6, 105-113. gencies by institutional staff: an evaluation ofGraubard, P. S., Rosenberg, H., and Miller, M. B. the effectiveness and durability of a training pro- Student applications of behavior modification to gram. Journal of Applied Behavior Analysis, teachers and environments or ecological ap- 1974, 7, 427-437. proaches to social deviancy. In E. A. Ramp and Pendergrass, V. E. Timeout from positive reinforce- B. L. Hopkins (Eds), A new direction for educa- ment following persistent high-rate behavior in tion: behavior analysis. Lawrence, Kansas: Sup- retardates. Journal of Applied Behavior Analysis, port and Development Center for Follow 1972, 5, 85-91. Through, 1971. Pp. 80-101. Redd, W. H. and Birnbrauer, J. S. Adults as dis-Griffiths, H. and Craighead, W. E. Generalization criminative stimuli for different reinforcement in operant speech therapy for misarticulation. contingencies with retarded children. Journal of Journal of Speech and Hearing Disorders, 1972, Experimental Child Psychology, 1969, 7, 440- 37, 485-494. 447.Guess, D., Sailor, W., Rutherford, G., and Baer, D. M. Rincover, A. and Koegel, R. L. Setting generality An experimental analysis of linguistic develop- and stimulus control in autistic children. Journal ment: the productive use of the plural morpheme. of Applied Behavior Analysis, 1975, 8, 235-246. Journal of Applied Behavior Analysis, 1968, 1, Risley, T. R. and Hart, B. M. Developing corre- 297-306. spondence between the nonverbal and verbal be-Hall, R. V. and Broden, M. Behavior changes in havior of preschool children. Journal of Applied brain-injured children through social reinforce- Behavior Analysis, 1968, 1, 267-281. ment. Journal of Experimental Child Psychology, Rogers-Warren, A. and Baer, D. M. Correspondence 1967, 5, 463-479. between saying and doing: teaching children toHerbert, E. W. and Baer, D. M. Training parents share and praise. Journal of Applied Behavior as behavior modifiers: self-recording of contin- Analysis, 1976, 9, 335-354. gent attention. Journal of Applied Behavior Anal- Schroeder, G. L. and Baer, D. M. Effects of concur- ysis, 1972, 5, 139-149. rent and serial training on generalized vocal imi-Horner, R. D. Establishing use of crutches by a tation in retarded children. Developmental Psy- mentally retarded spina bifida child. Journal of chology, 1972, 6, 293-301. Applied Behavior Analysis, 1971, 4, 183-189. Schumaker, J. and Sherman, J. A. Training genera-Israel, A. C. and OLeary, K. D. Developing corre- tive verb usage by imitation and reinforcement spondence between childrens words and deeds. procedures. Journal of Applied Behavior Analysis, Child Development, 1973, 44, 575-581. 1970, 3, 273-287.Johnston, J. M. and Johnston, G. T. Modification of Schwarz, M. L. and Hawkins, R. P. Application of consonant speech-sound articulation in young delayed reinforcement procedures to the behavior children. Journal of Applied Behavior Analysis, of an elementary school child. Journal of Applied 1972, 5, 233-246. Behavior Analysis, 1970, 3, 85-96.Kazdin, A. E. The effect of vicarious reinforcement Seymour, F. W. and Stokes, T. F. Self-recording in on attentive behavior in the classroom. Journal of training girls to increase work and evoke staff Applied Behavior Analysis, 1973, 6, 71-78. praise in an institution for offenders. Journal ofKazdin, A. E. and Polster, R. Intermittent token re- Applied Behavior Analysis, 1976, 9, 41-54.
AN IMPLICIT TECHNOLOGY OF GENERALIZATION 367Skinner, B. F. Science and human behavior. New behavior in a "blind" adolescent retardate. Journal York: Macmillan, 1953. of Applied Behavior Analysis, 1969, 2, 65-77.Solomon, R. W. and Wahler, R. G. Peer reinforce- Wahler, R. G. Setting generality: some specific and ment control of classroom problem behavior. general effects of child behavior therapy. Journal Journal of Applied Behavior Analysis, 1973, 6, of Applied Behavior Analysis, 1969, 2, 239-246. 49-56. Walker, H. M. and Buckley, N. K. The use of posi-Stokes, T. F. and Baer, D. M. Preschool peers as tive reinforcement in conditioning attending be- mutual generalization-facilitating agents. Behavior havior. Journal of Applied Behavior Analysis, Therapy, 1976, 7, 549-556. 1968, 1, 245-250.Stokes, T. F., Baer, D. M., and Jackson, R. L. Pro- Yates, A. J. Behavior therapy. New York: John gramming the generalization of a greeting re- Wiley and Sons, 1970. sponse in four retarded children. Journal of Ap- plied Behavior Analysis, 1974, 7, 599-610. Received 22 December 1975.Stolz, S. B. and Wolf, M. M. Visually discriminated (Final acceptance 3 June 1976.)