The Analysis of Verbal Behavior 1987, 5, 41-53

Teaching Mands by ManipulatingConditioned Establishing Operations

Genae Hall' and Mark L. SundbergWest Virginia University and Sundberg & Associates, Concord, CA

Skinner's (1957) analysis of verbal behavior suggests the functional independence of the verbaloperants. However, only a few empirical studies have directly examined the nature of theseoperants, and their independence. The present study evaluated whether teaching topographiesas tacts would lead to their emission as mands. The results indicated that manding only occurredreliably after direct mand training, which consisted of the use of imitative and tact prompts,and fading those prompts, to transfer stimulus control from nonverbal stimuli to conditionedestablishing operations. The results contribute to the existing data on the functional independenceof mands and tacts, as well as demonstrate the value of manipulating conditioned establishingoperations for mand training.

Skinner's (1957) book Verbal Behavior pro-vided a functional analysis of verbal eventsfrom the perspective of an individualspeaker and classified those events into the"verbal operants" (mands, tacts, echoics,intraverbals, textual behavior, taking dicta-tion, and copying a text). This analysis wasunique because it made no appeal to hypo-thetical explanatory entities such as innatelanguage acquisition devices or internalmechanisms which accept, process andstore spoken words. Skinner called theanalysis an "exercise in interpretation";behavioral principles found to be operatingunder highly controlled conditions in thelaboratory, with nonhumans, were extra-polated to complex human behavior in thenatural environment. This functionalapproach departed radically from the lin-guistic tradition of studying the grammaticalstructure and "meaning" of verbal response-products, regardless of the controlling cir-cumstances for the speaker's behavior.

'Portions of this paper are based on a thesis submittedby the senior author to Western Michigan University in1979, in partial fulfillment of the requirement of a mastersdegree.The authors wish to thank Gerald L. Shook, former

coordinator of the Kalamazoo Valley MultihandicapCenter (now Croyden Ave School), for his administrativesupport for this research. Appreciation is also extendedto Mark Stafford and Ray Miltenberger for theirassistance in conducting the study, and to Jack Michael,Wayne Fuqua and David Ray for their helpful commentson an earlier version of the manuscript. Reprintsrequests should be sent to Genae Hall, Department ofPsychology, West Virginia University, Morgantown,WV 26506.

Verbal Behavior contributed a functionalaltemativetothelinguisticmodelandrevealeda type of complexity in verbal events that hadpreviously been overlooked. Because verbalrelations included both response forms andtheir controlling variables, identical topo-graphies could participate in a number ofdifferent verbal relations and differenttopographies could participate in a singleverbal relation. The dependent variable unitnow included controlling variables as well astopographies.This functional approach suggested new

ways to predict and control verbal behavior.At the same time, its complexity revealednew practical problems for researchers in thearea. Functional units which includedobscure or temporally remote controllingvariables were more difficult to observe andmeasure than words, sentences and parts ofspeech per se. Verbal behavior's complexityalso required new research skills, such asmanipulating subjects' histories, arrangingnew types of motivative variables or"establishing operations" (Michael, 1982a)and analyzing equivalence classes (Sidman,1971).For various reasons, Verbal Behavior failed

to generate much research or theoreticalactivity in the first 20 years after its publica-tion in 1957 (Peterson, 1978; Vargas, 1986).Most behavioral language research was (andstill is) controlled by the linguistic model,and the language of the "common culture."Although not conceptualized within a verbalbehavior framework, this research has pro-

41

42 GENAE HALL AND MARK L. SUNDBERG

duced data which are relevant to Skinner'sanalysis.The early behavioral language research

was mainly designed to solve applied prob-lems (e.g., remediate language deficits withdevelopmentally delayed individuals), andresearchers initially assumed that teachingwords, sentences and parts of speech per seusing behavioral methods would establishfunctional and appropriate language reper-toires. Unfortunately, acquiring responseforms in training situations rarely led to"initiations" or "spontaneity." This problemwas pointed out by Lovaas (1977) who wrote:"Perhaps because of the highly controllednature of our language training and itsreliance on experimental rather than'natural' reinforcers, many of the childrenshowed verbal behavior that had comeunder very restricted environmental con-trol" (p. 17).A functional analysis of this problem might

compare the type of verbal relation (unit)established in the training situation with thetype of verbal relation desired in the naturalenvironment. If these verbal relations are dif-ferent, the question arises whether teachingone verbal relation should automatically pro-duce the other.

"Initiations" in the natural environmentare often mands-a specific type of verbal re-lation. Skinner(1957) definedthemandas "averbal operant inwhich the response is rein-forcedby a characteristic consequence and istherefore under the control of relevant condi-tions of deprivation or aversive stimulation"(pp.35-36). Theform ofthemand is controlledby these conditions of deprivation or aversivestimulation called "establishing operations"byMichael (1982a, 1987). An establishingop-eration (EO) is defined by Michael (1987) asIlan environmental event, operation, or stim-ulus condition which affects an organismbymomentarily altering (1) the reinforcing effec-tiveness of other events, and (2) the strengthof that part of the organism's repertoire thathas been reinforced by those other events"(p. 30). Michael goes on to distinguish bet-ween two types of EOs: unconditionedestablisfiing operations (UEO), and condi-tioned establishing operations (CEO). UEO's"are events, operations, (or) stimulus condi-tions whose value-altering effects areunlearned. They depend upon the evolu-tionary history of the particular species, and

vary from species to species" (p. 30). Fooddeprivation, for example, is an unlearnedoperation (the behavior of asking for food is,of course, learned, but the value alteringeffects are not).The second type of establishing operation,

the CEO, involves value-altering effectswhich have been learned during the indi-vidual organism's history. For example, thevalue of a certain toy may become momen-tarily strong, because of an individual child'sconditioning history, and the immediate con-tingencies. Nvichael (1982b) defines the mandas controlled by these two types of establish-ing operations, as well as the specific conse-quences linked to the relevant EOs. Accord-ing to Skinner (1957), the mand is the mostfunctional verbal operant for the speaker. Hestated that "When we come to considerother types of verbal operants, we shall findthat the behavior functions mainly for thebenefit of the listener ... The mand, however,works primarily for the benefit of thespeaker" (p. 36).Unfortunately, most structured language

sessions often contain training on only tactor echoic relations. Skinner (1957) definedthe tact as "a verbal operant in which aresponse of a given form is evoked (or at leaststrengthened) by a particular object or eventor property of an object or event" (pp. 81-82),and the echoic as "verbal behavior ...underthe control of verbal stimuli ... (where) ... theresponse generates a sound pattern similarto that of the stimulus" (p. 55). Tact trainingoften occurs as follows: The child is pre-sented with an object, asked "What is this?",then required to emit the appropriatetopography, and then given some educa-tional consequence such as social praise ora small edible. With echoic training, the childmight be asked to "Say ," then pro-vided with the same type of consequence asfor the tact.According to Skinner's analysis, teaching

tacts or echoic relations should not automati-cally lead to mands. It is the linguistic modelwhich suggests that once the child learns the"meaning" of a word (e.g., learns to say theword under tact conditions), he or she willthen "use" that word under all conditions.Skinner (1957) has instead suggested that theverbal operants are functionally indepen-dent: "In the terminology of meaning, wesay that the word 'doll' is used at one time

TEACHING MANDS 43

to 'ask for a doll' and at another to 'describeor refer to a doll.' When the response Doll!has been acquired as a mand, however, wedo not expect that the child then spon-taneously possesses a corresponding tact ofsimilar form. If we find both operants in therepertoire of the child, we must account forthem separately" (p. 187). According to thisanalysis, then manding or "spontaneity"may need to be trained directly.Without conducting this type of analysis,

a number of recent applied behaviorallanguage studies have addressed the pro-blem of restricted stimulus control. In mostcases they have trained manding, althoughthe dependent variable has generally beendescribed as "spontaneity" or "languageuse." These studies manipulated inde-pendent variables such as the presence ofreinforcing items, imitative prompts, timedelays and specific reinforcement pro-cedures, to train manding.Hart and Risley (1968, 1974, 1975, 1980)

used an "incidental teaching" procedure totrain more appropriate mands, with disad-vantaged children. This procedure was rela-tively unstructured in the sense that teach-ing interactions were child-initiated andoccurred in the natural environment.Children first made rudimentary mands forpresumed reinforcers (establishing opera-tions were not manipulated) which werevisible but out of reach. They were then imi-tatively prompted to improve or elaboratetheir manding topographies, and specificreinforcers were provided. As a result of thistraining, the children demonstrated strongermand repertoires.

Halle, Marshall and Spradlin (1979), Halle,Baer and Spradlin (1981) and Charlop,Schreibman and Thibodeau (1985) employeda "time delay" procedure with developmen-tally delayed children to establish mands orevoke previously acquired mands. This pro-cedure was more structured than incidentalteaching, in that certain times were desig-nated "opportunities" for initiations (mand-ing). At these times, certain reinforcingevents which regularly occurred were brieflydelayed. If mands for these events did notoccur within the designated time period,they were imitatively prompted. Specificreinforcers were then presented. This pro-cedure was also successful in generatingmands.

Simic and Bucher (1980) evaluated similarmand training procedures with develop-mentally delayed individuals. Two mandingresponses were successfully trained, in thepresence of the manded items. In this study,the dependent variable was described asmanding, however, establishing operationswere not manipulated.In both the incidental teaching, and the

time delay procedures, manding was freedfrom control by verbal discriminative stimulisuch as "What do you want?" However, inmost cases (except in the time delay pro-cedure, where the reinforcer was assistancefrom a teacher) manding was still dependentupon the presence of nonverbal discrimi-native stimuli (the manded items), whichmakes the response part tact.Carr and Kologinsky (1983) trained autistic

children to mand for missing items, whichwere presumed to be reinforcers. Childrenwere imitatively prompted to mand for eachitem, then provided with the specific rein-forcer. Carr and Durand (1985) trained devel-opmentally delayed children to mand con-ventionally for adult attention or help witha task, as alternatives to inappropriatemands (problem behaviors). Appropriatemands were imitatively prompted. In both ofthese studies, trained mands were under thecontrol of the establishing operation; noverbal or nonverbal stimuli were involved(except an audience).The above studies appear relevant to Skin-

ner's analysis in two ways: First, theydemonstrated that the mand relation couldbe established. Second, the early problem ofrestricted stimulus control tentatively sup-ported the notion of functional independ-ence. When response forms were trained astacts or echoics, they apparently did notautomatically occur as mands ("spontane-ity" did not occur). This research was notexplicitly designed to assess functional inde-pendence, however, and the appropriateconditions to assess transfer may not havebeen present.

Recently, a few basic verbal behaviorstudies have demonstrated the functionalindependence of the verbal operants withnonhumans. Savage-Rumbaugh (1984)showed the separation of several verbalrepertories with two chimpanzees. She con-cluded "the procedures developed for apessupport Skinner's analysis ... and have

44 GENAE HALL AND MARK L. SUNDBERG

significant applied value for training men-tally retarded humans" (p. 249). Michael,Whitley and Hesse (1983), and Sundberg(1985) also demonstrated the separation ofthe verbal repertoires, but with pigeons.Birds who were taught to tact did not mandor emit intraverbal behavior until directlytrained.Several basic studies with humans have

directly assessed the functional independ-ence of the speaker and listener, and of theverbal operants. Lee (1981) assessed func-tional independence of verbal and nonverbalbehavior (speaker and listener repertoires)related to the same prepositional configura-tion of objects, with moderately retarded andelementary school children. It was foundthat initial training on nonverbal relationsnever led to verbal relations, but acquiringverbal relations sometimes led to nonverbalrelations (when nonverbal topographies hadbeen emitted during baseline, but underdefective stimulus control). After subjectswere trained in both verbal and nonverbalrelations, reinforced changes in verbalresponding led to collateral changes innonverbal responding, but not vice versa.Thus, transfer was unidirectional and relatedto the subjects' entering repertoires; func-tional independence was conditional.Lee and Pegler (1982) evaluated functional

independence of reading and spelling, withtwo fifth graders who could not read. Thisstudy assessed the effects of oral reading,reading overtraining, repeated testing onspelling and a "word exposure" procedureon correct spelling of the same words thatwere read. Two of these conditions led togreatly improved spelling, two did not; func-tional independence was inconsistent.Lamarre and Holland (1985) assessed

functional independence of mands and tactsrelated to the same prepositional configura-tion of objects, with nursery school children.Some subjects received mand training firstand others received tact training first, thentransfer to the untrained repertoire wastested. It was found that mands and tacts ofthe same form were functionally independ-ent, at the time of acquisition. After subjectswere trained in both verbal relations, rein-forced changes in the first repertoire led tocollateral changes in the second repertoire,with three of the nine subjects; functionalindependence was maintained with six sub-

jects. This inconsistency points to the needfor future research to elucidate the controll-ing variables for functional independence.The above research has raised further

questions related to effective mand trainingand the conditions under which functionalindependence occurs. One question relatedto effective mand training is whether it isnecesary to conduct training only in thenatural environment. By manipulating con-ditioned establishing operations the value ofconsequences can be rapidly changed allow-ing for more controlled training procedures.Training inthe natural environment is import-ant and usually effective, but many of thenatural UEOs build up slowly and, like mostCEOs in the natural environment, weakenfollowing the delivery of the specific rein-forcement characteristic of the mand. Pro-cedures conducted in naturalistic settingswere sufficient to teach manding, but perhapsthey were not necessary. Structured settingsmight avoid the problem of restrictedstimulus control by training the mand rela-tion (where the form of the response, bydefinition, is not controlled by a discrimi-native stimulus) and determine which rein-forcers are currently effective by manipulat-ing establishing operations.A second practical question concerns to

what extent mand training procedures allowthe experimenter to control the specifictopographies to be trained. Incidentalteaching and time delay procedures trainedmands for items or events which had alreadybeen established as reinforcers; they did notmanipulate the reinforcing effectiveness ofthese items or events. Procedures which didmanipulate establishing operations wouldallow mands for any objects or actions to betrained. With such procedures, the experi-menter might not have to "capitalize onopportunities, when they arose" to trainmanding; these opportunities would becreated.A third practical question is whether

manding for a variety of missing items orevents can be trained. With incidentalteaching and time delay procedures, themanded items were usually present and visi-ble. Manding was not dependent on verbaldiscriminative stimuli such as "What do youwant," but the presence of particular itemsserved as discriminative stimuli that mandsfor those items would be reinforced. Mand-

TEACHING MANDS 45

ingfor missing items would seem to be maxi-mally functional, since subjects would nothave to wait foritems to appearbefore mand-ingthem. Two applied studies (Carr&Kolog-insky, 1983; Carr & Durand, 1985) demon-strated that manding for missing items orevents couldbetrained. The firststudytrainedautistic children to mand for missing itemswhichwerepresumed tobe reinforcers; estab-lishing operations were not manipulated. Inthe secondstudy, establishingoperations weremanipulated; task difficulty was varied withdevelopmentally delayed children, toestablish the effectiveness ofadult attention orhelp as reinforcers. Then, manding for thosereinforcers was trained. This research mightbe extended by developing systematic pro-cedures to establish the effectiveness of awider range of items as reinforcers, so that avariety of mands could be trained.In addition to these applied questions, the

basic issue of functional independenceremains to be clarified. Lamarre and Holland(1985) found unambiguous evidence of func-tional independence of mands and tacts atthe time of acquisition, but Lee (1981) andLee and Pegler (1982) showed results thatwere less consistent. None of these studiesshowed that functional independence wasmaintained with all subjects after both reper-toires had been acquired. Inter-subject dif-ferences raised the question of how subjects'histories interact with the different condi-tions, to produce transfer or functionalindependence. Further research is needed,to identify the relevant variables.The purpose of the present study was to

assess the following research questions withtwo deaf, mentally impaired and languagedelayed students:

1. Whether mands can be trained in astructured setting, where the experimentercontrols the topographies to be trained bymanipulating conditioned establishingoperations.

2. Whether structured procedures will leadto generalized manding (where untrainedmands occur under the same as well as novelstimulus conditions, and trained mandsoccur under novel stimulus conditions).

3. Whether subjects' histories and condi-tioned establishing operations can be manip-ulated in order to create opportunities formanding (teach specific mands at specifictimes).

4. Whether establishing specific verbaltopographies under tact conditions is suffi-cient to produce emission of those sametopographies under mand conditions.

5. If establishing tact relations is insuffi-cient to produce mands, whether an imi-tative or tact prompting procedure will trainmands more effectively.

METHODSubjects and SettingThe subjects were two deaf students, one

male and one female, enrolled in a day train-ing program for the mentally and/or physic-allyimpaired. Inadditiontobeingprofoundlydeaf, they were diagnosed as severally men-tally impaired. They were 16 and 17 years oldrespectively, and were chosen because oftheir extremely low rates of mandingbehaviors. Both subjects had extensive tact-ing repertoires, and had received some train-ing on manding for items which were pre-sent, using a "mand frame" (e.g., "I want

."). However, they never mandedin the absence of a verbal discriminativestimulus; and, when asked, "What youwant?" in sign when no reinforcing objectswere present, each would emit a singlestereotypic response. Subject 1 (the male)would mand "walk" and subject 2 (thefemale) would mand "coloring book." Theyemitted no other mands for missing items oractions. The form of these stereotypic mand-ing responses did not appear to be controlledby what would function as strong reinforcersat the moment, since the students wouldmake these responses at any time during theday (early in the morning, before lunch,whenever they counted their tokens toexchange for back-up reinforcers).Observation and ReliabilityThe experimenter (the first author) con-

ducted all primary observations, and twoother staff members at the facility served asreliability observers. One class of behaviorsto be observed during pretraining, baseline,and training conditions consisted of manualsigns in American Sign Language (tacts ormands). Other responses which wereobserved, during pretraining only, wereactions which the subjects were required toindependently initiate with one object ineach of the four response chains. The relia-bility observers were shown the signs they

46 GENAE HALL AND MARK L. SUNDBERG

could not already tact, although both werefamiliar with sign language.During tact and mand probes, a trial was

scored as correct if the appropriate tact ormand occurred at any time during a 10-sec-ond observation period (if it was recogniz-able, it was scored as correct). For tacts, theobservation period began immediately afteran object to be tacted was presented. Formands, the observation period began at thepoint that the subject needed a missing itemto complete a chain of behavior leading to areinforcer.The response definitions for actions initi-

ated with one object for each chain (duringchain completion probes) are presented inthe Procedure section. During these probes,a trial was scored as correct if the subject initi-ated the appropriate action with the object,without assistance.

Reliability was taken during pretraining,both in chain completion sessions, wheredata were collected to establish that thestudents had met criterion on initiatingactions with particular objects in the chains,and in tact training sessions. Reliability wasalso taken during mand probes in baselineand training conditions. Reliability observerswere instructed to score correct responsesimmediately-the primary observer waitedone second, reinforced the mand by present-ing the item manded, and then scored theresponse. Incorrect responses were notscored. The primary and reliability observerswere seated at the same table as the student,about two yards away from each other, withtheir data sheets faced in different direc-tions.

DesignThe designs used were a multiple baseline

across subjects and behaviors (Baer, Wolf, &Risley, 1968) and a multielement design(Ulman & Sulzer-Azaroff, 1975). Subject 1received training on the first and secondpairs of mands sooner than Subject 2. Inaddition, within each subject, a first pair ofmands was trained before a second pair.Pairs of mands were also trained simul-taneously in a multielement design, to com-pare the effectiveness of tact and imitativeprompting procedures.Pre-trainingPrior to the baseline conditions, both sub-

jects were trained to criterion on the follow-ing prerequisite skills:

1. Both were taught to complete fourchains of behavior leading to a rein-forcer. The trainer provided someassistance in completing the chains, butthe subjects were required to independ-ently initiate the first action for eachobject that would later be trained as amand, at a criterion of 100% accuracyfor two consecutive days. Physicalguidance was used initially and wasfaded until criterion was reached. Thefour chains of behavior, the objects andactions involved, and the responsedefinition for "initiating the first actionwith the object" are presented below.Subject 1a. Making instant soup. The objects

involved were instant soup, hotwater, a bowl, and a spoon. Theactions consisted of tearing open thepackage of instant soup, pouring itinto a bowl, pouring hot water on itfrom a plug-in pot which hadalready been heated, stirring thesoup, and eating it with a spoon.The object from this chain thatwould later be trained as a mandwas the hot water. It was, therefore,required that the student initiate anaction with the water. This actionconsisted of picking up the pot andpouring at least one drop into thebowl.

b. Opening a can of fruit. The objectsinvolved were a can, can opener, abowl, and a spoon. The actions con-sisted of taking the can opener,opening the can, pouring the fruitinto the bowl, and eating the fruitwith a spoon. The object from thischain that was to be trained was thecan. Initiating an action with the canconsisted of touching the top of itwith the can opener.

c. Wiping up water spilled on the table (adry table was a reinforcer for the stu-dent, when water had been spilled).The objects involved were water anda paper towel. The chain consistedof the trainer spilling a small amountof water on the table, from a glassthat was filled to the top. The stud-ent then took the paper towel whichwas present and wiped up the water.The object to be trained was the

TEACHING MANDS 47

paper towel. Initiating an actionwith the paper towel consisted ofpicking it up and at least touchingthe water spilled on the table.

d. Operating a vending machine to getcandy. The objects involved were thevending machine and money (aquarter). The chain consisted of thestudent taking the quarter and put-ting it in the slot. The trainer thenassisted the student in pressing theright button to get a candy bar. Theobject to be trained was the money.Initiating an action with the moneyconsisted of taking it and makingcontact with the money slot on themachine.

Subject 2a. Making instant soup. The objects,

actions, and response definition forinitiating the action were the same asfor Subject 1.

b. Opening a can of fruit. Again, theobjects, actions, and responsedefinition were the same as for Sub-ject 1.

c. Making instant coffee. The objectsinvolved were instant coffee, hotwater, a cup, and a spoon. The chainconsisted of opening the jar of coffee,taking out a teaspoonful of coffee,putting it in the cup, pouring hotwater in the cup, and stirring thecoffee with a spoon. The object to betrained was the cup. Initiating anaction with the cup consisted of put-ting some (any) amount of instantcoffee in the cup.

d. Coloring a large picture. The objectsinvolved were a large, partly uncol-ored picture and colored pens. Thechain consisted of selecting a penfrom the container and coloring asmall section of the picture. Theobject to be trained was the pen.Initiating an action with the pen con-sisted of taking a pen and touchingthe picture with it.

2. Both subjects were tested to determinewhether they had correct tacts for allitems necessary to complete each chain.The experimenter successively present-ed each item and signed, "What that?"Correct responses were reinforced with

tokens. If a tact was incorrect, theexperimenter modeled the correctresponse and re-presented the dis-criminative stimulus. This correctionloop was repeated until the studentemitted a correct response without theimitative prompt. Tacts were eithertested or trained until a criterion of100% accuracy for two consecutive dayswas met.

BaselineBaseline consisted of mand probes, where

all items necessary to complete a chain,except for one item, were presented to thestudent. Then, the trainer manually signedthe specific discriminative stimulus to begineach chain. The following discriminativestimuli were used:

1. Soup chain "Make soup. Start."2. Can chain "Open. Start."3. Wiping table chain "Wipe. Start."4. Vending machine chain "Get candy.

Start."5. Coffee chain "Make coffee. Start."6. Coloring chain "Make picture. Start."After the discriminative stimulus was

given, the trainer provided some assistancefor the student in completing the chain upto the point that the missing item wasneeded. The assistance consisted of theminimal amount of physical guidancenecessary to ensure that the steps in thechain were completed. At the point wherethe missing item was needed to continue, thetrainer provided no assistance and waited 10seconds. If the missing item was mandedwithin this time period, it was presented; thestudent could complete the chain and obtainthe reinforcer. If an incorrect mand or noresponse occurred withinthe time period, allitems were removed and the next chain waspresented.TrainingMand probes continued during training

and were conducted the first time a par-ticular chain was presented, for each session.If students failed to mand within the10-second time period in the probe situation,training was conducted.Two training procedures, tact prompt and

imitative prompt, were compared in a multi-element design. Ofthe fourmands eventuallytrained for each subject, two were trainedonly with the tact prompt procedure; and

48 GENAE HALL AND MARK L. SUNDBERG

two only with the imitative prompt proced-ure. First, two mands were simultaneouslytrained (both in each session) using the dif-ferent procedures; and, after both weremastered, two more were simultaneouslytrained. The mand that was trained first eachsession was alternated randomly.The tact prompt procedure occurred as

follows. First, a mand probe was conducted,as described in the Baseline section. If thecorrect mand did not occur, the trainer pre-sented the missing item, signed, "Whatthat?" and the student tacted the item. If acorrect tact did not occur, the trainer mod-eled the topography and represented thequestion. After a correct tact, the item wasleft in front of the student, the trainer signed,"What you want?" and the student repeatedthe topography. If the correct response(partly mand and partly tact) did not occur,the trainer modeled it and represented thequestion. After a correct response to "Whatyou want?", all items were removed for 5seconds, then re-presented except for thepreviously missing item. At this point, thestimulus situation was identical to theprevious probe condition, where the studentcould mand without prompts and receive aspecific reinforcer (the missing object). If thecorrect mand did not occur, training was con-tinued by repeating the correction loop. Allcorrect responses during training, except cor-rect mands under unprompted probe condi-tions, were reinforced with tokens. Bothstudents were already on a token economyand tokens had been established as condi-tioned reinforcers. At the end of eachsession, tokens were exchanged for back-up reinforcers that the students typicallyearned, such as edibles or activities differentfrom those used in the chains.The imitative prompt procedure differed

from the tact prompt procedure in only twoways:

1. Instead of presenting the missing itemafter a failure to mand within 10seconds and signing "What that?", thetrainer did not present an item butinstead signed, "Do this."

2. After signing "Do this," the trainer pre-sented an imitative prompt. If a correctimitation of the topography occurred,the subject could then go on to the nextstep, which was, "What you want?" Ifnot, the trainer physically prompted

the response and represented theinstruction.

Generalization probesWhen the target response was a mand

which had been previously trained, stimulislightly different from those used in trainingwere presented. When the target responsewas a topography which had not beenpreviously trained as a mand (it had beenpreviously trained under tact, but not mand,conditions), stimuli which were the same asthose used in training were presented. In afew cases, the stimuli used in generalizationprobes and the target responses were bothdifferent. The specific stimuli and targetresponses are shown in Table 1 of the Resultssection.The general procedure for probes remained

the same as during baseline and training,except that the discriminative stimuli to beginthe coffee chain and the coloring chain forSubject 2 had to be slightly modified. Sincethe new target response for the coffee chainwas "coffee," the discriminative stimuluswas changed from "Make coffee. Start." to"Make drink. Start." So that an imitativeprompt would not be provided. Likewise,the new target response for the new coloringchain was "picture"; it was therefore neces-sary to change the discriminative stimu-lus from "Make picture. Start." to "Draw.Start."

RESULTSReliability was taken in pretraining, base-

line, and training conditions. In pretraining,reliability was taken in 49% of tact trainingsessions and 100% of chain completion ses-sions used to establish that subjects had metcriterion on the chains. Reliability was takenin 29% of baseline sessions and 29% of train-ing sessions. There were no instances ofdisagreement in any of the conditions.As shown in Figure 1 only a few mands

occurred during the baseline probes. Correctmands reliably occurred only after trainingwas implemented. Teaching topographiesunder tact conditions and arranging stimuliso that previously tacted objects currentlyfunctioned as reinforcers was not sufficientto produce the same topographies undermand conditions. This did not appear to bea tact maintenance problem, because whenthe tact prompt procedure was initiated after

TEACHING MANDS

I KEY: TYPE OFMANDTRAINING II II * IMITATIVE PROMPT PROCEDURE I

O--- TACT PROMPT PROCEDURE I.-I

BASELINE MAND TRAINING(FIRST PAIR):9-- WATER0--a CAN

."b'

.

I MANDBASELINE TRAINING -4CAN OPENERBASELINE I(FIRST PAIR): O--O CUP

MANDBASELINE TRAINING -MONEYB(SECOND PAIR): 0--0 PAPER TOWEL

1=.=I MAND

BASELINE ITRAINING * WATER|(SEC PR): 0-O PENS

A_ 1, --------Po . _ _ _

10CONSECUTIVE

1S 20SESSIONS

25 30

Figure 1. Performance on mand probes during baseline and training during both training conditions, for Subject 1and Subject 2.

0

S2

01

+

0l

S

I-w

0

o

*~-

zla

0U

0

az)E

S2+00'

1 5

49

I

I

I

I II d

50 GENAE HALL AND MARK L. SUNDBERGTable 1

Generalization Probes: Specific Stimuli Used,Target Responses (Mands) and Whether the Responses Were Correct

General i zationSession Stimulation during Generalization Probes Trained Responses Responses Correct?

Subject 1

Session 1 1. Soup chain items, plus SD Water Bowl 02. Can chain items, plus SD Can Can opener +3. Vending chain items, plus SD Money Money +

(different machine)4. Wiping table chain items, plus SD Paper towel Paper towel +

(different table)Session 2 1. Coke chain items (entirely different None for this chain Cup +

chain and SD--items were coke andcup; the SD was,,,"Drink pop. Start."

Session 3 1. Empty table, no SD. Reinforcement Soup, water, bowl, Soup, water, bowl, all +period after session. spoon, can, can spoon, can, can

opener, paper towel, opener, paper towel,walk, money, run, walk, money, run,sit, eat sit, eat

Subject 2

Session 1 1. Soup chain items, plus SD Water Bowl +2. Coffee chain items, plus SD Cup Coffee 0

(different SD)3. Can chain items, plus SD Can opener Can +4. Picture chain items, plus SD Pens Picture +

(different SD)

baseline, students made the correct tactsthree out of four times without prompts.Only one mand (the third trained for Sub-

ject 2) appeared to be acquired duringbaseline, after training for the first andsecond mands had been completed. Theincidences of correct manding of "water"during baseline for Subject 1 and "pen" forSubject 2 were not maintained. No other cor-rect manding responses occurred duringbaseline. After training on the four mandsfor each subject, correct responding acrosstime was maintained, with the exception ofone incorrect response for Subject 1, whichwas retrained in one trial.As shown in Table 1, three types of gen-

eralzations occurred after completionof training on all four mands. Subject 1emitted trained mands in the presence ofboth the same stimuli (except for the itemthat was missing from the chain) and dif-ferent stimuli with respect to the trainingconditions. He also emitted trained mandsin the presence of slightly different stimuli.Subject 2 emitted different mands in thepresence of the same stimuli (except for themissing item) and different mands in thepresence of different stimuli.In terms of comparing the two prompting

procedures, they produced similar results.Both were effective in training manding, andneither was markedly superior.

DISCUSSIONThe present results indicate that manding

for missing items can be trained using astructured procedure, where the experi-menter selects the mands to be trained. Thisprocedure manipulated subjects' historiesand conditioned establishing operations(Michael, 1982a, 1987) to teach specificmands at specific times. Subjects' historieswere manipulated by allowing them torepeatedly complete four chains of behavior(with some assistance) and contact certainreinforcers at the end. This pretrainingestablished the objects needed for chaincompletion as conditioned reinforcers, andestablished an invariant sequence ofresponses. After this history, it was possibleto manipulate conditioned establishingoperations by removing items which wereessential for chain completion. This pro-cedure established the momentary effec-tiveness of the missing objects as reinforcers,and topographies correlated with these miss-ing objects could then be prompted.A second finding was that direct mand

TEACHING MANDS 51

training was necessary, to establish reliablemanding responses. Teaching subjects to tactobjects during pretraining and establishingconditions under which those objects func-tioned as reinforcers during baseline werenot sufficient to produce manding for mis-sing items. These results (together withLamarre & Holland, 1985) would supportSkinner's (1957) interpretation that mandsand tacts are functionally independent at thetime of acquisition.There were indications that subjects were

affected by establishing operations duringthe mand baseline (after tacts were estab-lished), but simply did not have appropriatemanding responses. For example, Subject 1had to mand for the missing item "money"to operate a vending machine and get candy.Instead of emitting the conventional topo-graphy (which he reliably emitted as a tact),he tried putting his hand into the experi-menter's pocket. With the other chains, bothsubjects tried to get up and find the missingitems themselves, or they emitted unconven-tional topographies, like tapping the experi-menter's arm. Another possible unconven-tional mand occurred during baseline. Onthe wiping table chain, Subject 1 did nothave the conventional mand "paper towel"and aggressed toward the experimenter.There were additional indications that thechains led to strong reinforcers. Both subjectsate the consumables and engaged in rein-forcing activities quickly; they also repeatedmands vigorously when there was a shortdelay before presentation of a reinforcer.After four mands were trained for each

subject, transfer from tacts to mands andstimulus generalization were tested. Subject1 showed transfer from tacts to mands inthree out of three probes, and emitted twotrained mands in the presence of slightly dif-ferent stimuli. After a few transfer and gen-eralization sessions, Subject 1 also beganmanding before the discriminative stimulusto begin the chain was presented, and beforeany items were placed on the table. Aftereach mand, the manded item was presented;and he continued manding for all the nec-essary items to complete the chain. Whenthe items were all present, he made the signfor the discriminative stimulus to start thechain, and completed it. After the chain wascompleted and the reinforcer obtained, hemanded to engage in activities that generally

occurred after the session (walk, run, sit,eat). Training manding for missing objectsapparently led to manding for reinforcingactions, without further training.For Subject 2, one mand to be trained (the

third) emerged after the first two mandswere trained. (As with all mands that weretrained or tested, the topography had firstbeen acquired as a tact.) The fourth mandstill required direct training, however. Afterall four mands were established, transferfrom tacts to mands was assessed with dif-ferent items missing from the same chains.In three out of four probes, transfer wasfound; mands emerged without direct mandtraining. The occurrence of untrained mandsin slightly different stimulus situations wasalso assessed with Subject 2, when the dis-criminative stimuli for two of the chains werechanged. It was found that correct mandingstill occurred, under those conditions.For both subjects, these results indicate

that functional independence was presentwhen mands were initially trained, but itwas not maintained indefinitely. After initialtraining on four mands, both subjectsappeared able to mand any item that theycould tact, although additional data wouldbe necessary to firmly support this conclu-sion. Generalization data from both subjectsalso suggest that manding was not underrestricted stimulus control. Even thoughtraining was conducted in a structured set-ting, manding generalized to somewhatnovel stimulus situations.When compared in a multielement design,

the tact and imitative prompt procedureswere about equally effective. When onebehavior was trained using the imitativeprompt procedure, a second was simul-taneously trained using the tact prompt pro-cedure (in the same session, with the orderof presentation of each procedure random-ized). One potential advantage to the imi-tative prompt procedure would be that it isnot necessary to learn the tact for an itembefore manding can be acquired; imitation ofthe topography is sufficient. On the otherhand, a tact prompt has the advantage ofrequiring the subject to emit the responseform in the absence of a model.With one exception, all manding responses

were maintained at 100% accuracy once theyhad met the criterion of being emittedwithout prompts for two consecutive ses-

52 GENAE HALL AND MARK L. SUNDBERG

sions. The mand that did not initially main-tain with Subject 1 ("can") was retrained inone trial and maintained successfully afterthat point. Once responses were trained tocriterion, the time lapse between probes wasgradually lengthened. The maintenance datawere especially interesting for Subject 1,since he had a long history of inconsistentresponding during tact training in his regularacademic sessions (averaging around 70%)and of failing to maintain behaviors he hadpreviously acquired. Four months after term-ination of the study, his teacher reported thathe was still consistently manding for miss-ing items in novel stimulus situations (a dif-ferent room) and with novel trainers. Hewould mand "paper towel" and wipe thesink. He was also manding for activities (eat,walk, sit). Maintenance data remained at100% for Subject 2.There are several directions for future

research. First, the mand training pro-cedures could be refined. In the presentstudy, it appeared that the tact and imitativeprompt procedures might have been moreefficient if there had been fewer verbaldiscriminative stimuli. The subjects some-times made imitative responses to the dis-criminative stimuli rather than responding tothe tact or imitative prompts. For example,during the imitative prompt procedure, theimiative prompt "water" was preceded bythe discriminative stimulus "Do this."Instead of correctly responding to the imi-tative prompt, Subject 1 in particular wouldoften imitate "Do this." Such problemscould perhaps be avoided by using as fewdiscriminative stimuli as possible in thetraining procedure, especially with subjectswith strong imitative histories. The design ofthe study could also be improved by assess-ing transfer and generalization more sys-tematically. Maintenance could be assessedfor a longer period of time.Future research could further assess the

controlling variables for functional inde-pendence versus transfer between the verbaloperants. Perhaps mands could be trainedfirst with some subjects and tacts with others(as in Lamarre & Holland, 1985) and func-tional independence assessed at that point,before the second operant was trained. If thepresent procedures for training tacts andmands were used (with manipulation of sub-jects' histories and conditioned establishing

operations to ensure that mands wereindeed being trained), this would representan extension of Lamarre and Holland'sresearch. After the first operant was trained(either mands or tacts) and functional inde-pendence assessed at that point (at the timeof acquisition), each topography trainedas the first operant could successively betrained as the second operant. After eachwas trained as the second operant, remain-ing topographies trained only as the firstoperant could be assessed as the secondoperant (transfer could be tested). If trans-fer was produced using this procedure,one might train new topographies as oneoperant (mand or tact) and test whether theyoccurred as the other operant, without directtraining. This procedure might assess someof the necessary conditions for functionalindependence.

REFERENCESBaer, D., Wolf, M., & Risley, T. (1968). Some current

dimensions of applied behavior analysis. Journal ofApplied Behavior Analysis, 1, 91-97.

Carr, E. G., & Durand, V. M. (1985). Reducing behaviorproblems through functional communication train-ing. Journal ofApplied Behavior Analysis, 18, 111-126.

Carr, E. G., & Kologinsky, E. (1983). Acquisition of signlanguage by autistic children. II: Spontaneity andgeneralization effects. Journal of Applied BehaviorAnalysis, 16, 297-314.

Charlop, M. H., Schreibman, L., & Thibodeau, M. G.(1985). Increasing spontaneous verbal respondingin autistic children using a time delay procedure.Journal of Applied Behavior Analysis, 18, 155-166.

Halle, J. W., Baer, D. M., & Spradlin, J. E. (1981).Teacher's generalized use of delay as a stimulus con-trol procedure to increase language use in handi-capped children. Journal ofApplied BehaviorAnalysis,14, 389-409.

Halle, J. W., Marshall, A. M., and Spradlin, J. E. (1979).Time delay: A technique to increase language useand facilitate generalization in retarded children.Journal of Applied Behavior Analysis, 12, 431439.

Hart, B., & Risley, T. R. (1968). Establishing use of des-criptive adjectives in the spontaneous speech ofdisadvantaged preschool children. Journal ofAppliedBehavior Analysis, 1, 109-120.

Hart, B., & Risley, T. R. (1974). Using preschool materialsto modify the language of disadvantaged children.Journal of Applied Behavior Analysis, 7, 243-256.

Hart, B., & Risley, T. R. (1975). Incidental teaching oflanguage in the preschool. Journal ofApplied BehaviorAnalysis, 8, 411420.

Hart, B., & Risley, T. R. (1980). In vivo language inter-vention: Unanticipated general effects. Journal ofApplied Behavior Analysis, 13, 407-432.

Lamarre, J., & Holland, J. (1985). The functional inde-pendence of mands and tacts. Journal of the Experi-mental Analysis of Behavior, 43, 5-19.

TEACHING MANDS 53

Lee, V. L. (1981). Prepositional phrases spoken andheard. journal ofthe ExperimentalAnalysis ofBehavior,35, 227-242.

Lee, V. L., & Pegler, A. (1982). Effects on spelling oftraining children to read. Journal of the ExperimentalAnalysis of Behavior, 37, 311-322.

Lovaas, 0. I. (1977). The autistic child: Language develop-ment through behavior modification. New York: Irving-ton Publishers, Inc.

Michael, J. L. (1982a). Distinguishing between the dis-criminative and motivational functions of stimuli.Journal of the Experimental Analysis of Behavior, 37,149-155.

Michael, J. L. (1982b). Skinner's verbal operants: Somenew categories. The Analysis of Verbal Behavior, 1,1.

Michael, J. L. (1987). Advanced topics in behavior anal-ysis. Unpublished paper.

Michael, J. L., Whitley, P., & Hesse, B. E. (1983). Thepigeon parlance project. The Analysis of VerbalBehavior, 2, 6-9.

Peterson, N. (1978). An introduction to verbal behavior.Grand Rapids, MI: Behavior Associates, Inc.

Savage-Rumbaugh, S.E. (1984). Wrbal behavior at a pro-cedural level in the chimpanzee. Journal ofthe Experi-mental Analysis of Behavior. 41, 223-250.

Sidman, M. (1971). Reading and auditory-visual equiv-alences. Journal of Speech and Hearing Research, 14,5-17

Simic, J., & Bucher, B. (1980). Development of spon-taneous manding in nonverbal children. Journal ofApplied Behavior Analysis, 13, 523-528.

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

Sundberg, M. L. (1985). Teaching verbal behavior topigeons. The Analysis of Verbal Behavior, 3, 11-17.

Ulman, J. D., & Sulzer-Azaroff, B. (1975). Multielementbaseline design in educational research. In E. Ramp& G. Semb (Eds.), Behavioranalysis: Areas ofresearchand application. Englewood Cliffs, NJ: Prentice-Hall.

Vargas, E. (1986). Intraverbal behavior. In L. J. Parrott &P. N. Chase (Eds.), Psychological aspects of language:The West Virginia lectures (pp. 128-151). Springfield:C. C. Thomas.