JOURNAL OF APPLIED BEHAVIOR ANALYSIS

ERRORLESS LEARNING: REINFORCEMENT CONTINGENCIES ANDSTIMULUS CONTROL TRANSFER IN DELAYED PROMPTING

PAUL E. TOUCHETrE AND JANE S. HOWARD

DEPARTMENT OF NEUROLOGY, HARVARD MEDICAL SCHOOL ANDCALIFORNIA STATE COLLEGE, STANISLAUS

Delayed prompting can produce errorless discrimination learning. There is inherent in the procedurea disparity in reinforcement density which favors unprompted over prompted responses. We usedthree schedules of reinforcement to investigate the impact of reinforcement probability on transferof stimulus control. One schedule of reinforcement was equal prior to and following a prompt(CRF/CRF), the second favored unprompted responses (CRF/FR3), and the third favored re-sponses following the prompt (FR3/CRF). Experimental questions concerned the probability oferrors, the probability of transfer, and the rate of transfer in the context of delayed prompting.Transfer was accelerated when reinforcement probability favored anticipatory responding. The schedulethat favored prompted responses did not prevent a shift to unprompted responding. Errors wereinfrequent across procedures. Reinforcement probability contributes to but does not entirely deter-mine transfer of stimulus control from a delayed prompt.DESCRIPTORS: fading, delay, prompts, schedules of reinforcement, retarded students

To prompt an appropriate response, teachers ofthe handicapped routinely model a correct re-sponse, point to the correct choice, provide an addedverbal stimulus, or use materials in which the cor-rect choice is distinctive by virtue of its location,size, or color. Prompting is the substitution of aneffective but inappropriate stimulus for an ineffec-tive but appropriate one. Prompts are stimuli thatcontrol the desired behavior, but that are not func-tionally related to the task. Unfortunately, it isoften the case that students who respond appro-priately when prompted, founder when the promptis removed. They evidence persistent dependenceon stimuli not intrinsic to the task. Once correctresponding has been initiated by prompting, theteacher's task is to maintain the response pattern

Authorship is equal. This research was supported in partby Grants HD 15871 and HD 004171.We are grateful to the administration and staff of the

Kalamazoo Valley Multihandicapped Center, Kalamazoo,Michigan; the Shelby Center, Livingston, California; and theEfficacy Research Institute, Framingham, Massachusetts.Special thanks for your assistance: Sunny Lippert, Lova Pack,David Ray, Mark Sandberg, Jerry Shook, Jacqueline Spen-cer, and Vincent Strully.

Reprints may be obtained from Paul E. Touchette, TheShriver Center, 200 Trapelo Road, Waltham, Massachusetts02254; or Jane Howard, Psychology Department, CaliforniaState College, Stanislaus, Turlock, California 95380.

while eliminating the prompt(s). Transfer of stim-ulus control from prompt to task-related stimuliis expected and relied on in virtually all instructionthat uses stimuli extrinsic to the final performance.

Premature removal of a prompt can initiate per-sistent incorrect response patterns which precludeacquisition of the target repertoire (Sidman &Stoddard, 1966; Touchette, 1968). On the otherhand, unnecessarily extended presentation ofprompts may foster dependence in the form ofpersistent selective attention to stimuli supplied bythe teacher. Applied research has not yet resolvedthe question of how to produce a successful tran-sition from prompted to unprompted responding.An ideal transition from prompted to un-

prompted responding will result in few or no errors.An error-free transition from instructional supportto unassisted competence can be very importantfor the severely handicapped learner. Terrace (1966,pp. 316-3 1 7) noted that pigeons that were taughta discrimination by differential reinforcement emit-ted agitated, emotional behavior whereas pigeonsthat were taught the same discrimination withouterrors were calm and attentive. Informal observa-tions and formal research suggest that the demandsof learning by trial and error can provoke problembehavior including apathy, tantrums, aggression,

175

1984)17,175-188 NUMBER2 (SUMMER 1984)

PAUL E. TOUCHETTE andJANE S. HOWARD

and self-injury, sometimes severe (Carr, Newsom,& Binkoff, 1980; Weeks & Gaylord-Ross, 1981).Effective instruction demands an understanding offactors that encourage the emergence of appropri-ate stimulus control during training.

If a prompt is a stimulus that reliably controlsa response, then the emergence of control by anotherstimulus associated with the prompt can be de-scribed as a transfer of control (Terrace, 1963;Touchette, 1971). There is an analogy betweenthe transfer of stimulus control and the emergenceof an operant response. In both cases the first in-stance can be encouraged by arranging the stu-dent's environment and history of reinforcement toimprove the likelihood that the desired entity willemerge. Its occurrence, however, cannot be guar-anteed. This topic has been discussed in depth bySkinner (1968, pp. 206-208) and by Ray andSidman (1970, pp. 193-194). Applied practi-tioners have recourse to few options when they arefaced with what Skinner called "the problem ofthe first instance." Data currently available are notsufficient to allow the selection of the best ap-proach.

In this study, we examined the impact of sched-ules of reinforcement that would be expected toencourage and discourage the transfer of controlfrom prompts to task-related stimuli. Reinforce-ment can and usually does maintain a stimulus-response relationship that already exists. It is lessdear what influence reinforcement may have onthe establishment of new stimulus-response rela-tions or the transfer of control from prompts totask-related stimuli (Huguenin & Touchette,1980).

Describing the conditions necessary to the trans-fer of stimulus control has been a more difficulttask than devising procedures that produce trans-fer. To evaluate the impact of reinforcement ontransfer it is necessary to use a paradigm that al-lows direct measurement of the point at whichtransfer occurs. Fading can produce errorless trans-fer of stimulus control, but the point of its occur-rence is obscured. The string of correct responsesthat accompanies a successful fading procedureprovides no information as to the point at which

task-intrinsic stimuli acquired control. In contrast,delayed prompting (Touchette, 1971) can produceerrorless learning and reveal the point of transfer.This technique allows the evaluation of differentialconsequences applied to prompted and unprompt-ed responses.

In delayed prompting the discriminative stimuliare presented at the beginning of a trial. A promptis provided after a delay. Responses occurring afterthe delay are presumed to be under the control ofthe added prompt. Response latencies shorter thanthe delay provide an indication that stimulus con-trol has shifted from the prompt to the discrimi-native stimuli that precede it. There can be a sub-stantial variation in reinforcement density per unitof time, dependent on whether correct responsesprecede or follow a delayed prompt. It is reason-able to postulate that an increase in reinforcementdensity, which results when correct responses an-ticipate the prompt, is a critical variable in pro-ducing and maintaining transfer of stimulus con-trol.An analysis of procedural variables that affect

transfer of stimulus control during errorless dis-crimination learning presents special problems inexperimental design. More than one discriminationmust be acquired under each condition to elimi-nate the possibility that the discriminative stimuliare the source of any variation observed. Rate oflearning must then be determined within individ-uals. As is customary, this study includes a pretestto verify that each discriminative performancelearned was not already in the student's repertoire.Evaluating the impact of the schedule of reinforce-ment on the point of stimulus control transfer re-quires a different reference. Uniform reinforcementof prompted and unprompted correct responses wasused to teach a number of successive discrimina-tions. These training sessions provided a referentfor comparing the rate of learning under two dis-parate schedules of reinforcement subsequentlyused.

The experimental questions addressed here haveboth applied and theoretical significance. Will anexaggerated discrepancy in reinforcement proba-bility favoring anticipatory responding produce

176

DELAYED PROMPTING

more expeditious transfer? Conversely, will transferbe prevented if reinforcement density is higher forresponses following a prompt? Is the probabilityof errors affected by the schedule of reinforcement?Are there uniform effects across severely retardedindividuals?

METHOD

StudentsThree students who had been diagnosed as se-

verely retarded and who attended an educationalcenter for multiply-handicapped children partici-pated in this study. All were able to follow one-step verbal instructions such as "put your handson the table" and "touch your nose."

Student 1 was 6-year-old boy with a Cattel es-timated mental age of 2 years, 6 months. Hisverbal repertoire included some mands, tacts, andintraverbals. He could follow a few familiar two-step commands in the context of daily events, e.g.,"go to the door and wait for me." He occasionallydisplayed self-abusive and aggressive behavior.

Student 2 was a 13-year-old girl diagnosed ashaving cerebral palsy and severe mental impair-ment. Her Leiter estimated IQ was 55. She had avery limited verbal repertoire, and she could notfollow commands involving more than a singleelement. She was cooperative and displayed no de-viant or disturbed behavior.

Student 3 was a 7-year-old boy whose Leiterestimated IQ was also 55. He could follow a fewone-step commands in a familiar context. He wasnonvocal, emitted unintelligible sounds, and com-municated primarily through gestures. He was co-operative and displayed no deviant behavior in theeducational center.

SettingData were collected at the training center that

the students attended daily. The students sat inthe work area to which they were normally as-signed for instruction. Experimental sessions tookplace during school hours with other students pres-ent. The experimenter sat facing the student acrossa table set at a convenient height.

Training sessions were conducted 5 days a weekand lasted for 20 to 30 minutes each. Tokensacquired during each training session were ex-

changed for a variety of toys, activities, and edibletreats when a student had collected six tokens.

Discriminative Stimuli

All three students were taught developmentallyappropriate visual discriminations. Four stimuluscards were placed before the students on each trial.The experimenter then instructed them to touchthe card that had been designated S+. For ex-

ample, a student might be shown cards containingthe letters M, B, U, and L and instructed "pointto B." The student then chose from among thefour stimuli by touching a card. The positions ofthe S+ and three S- stimuli were varied unpre-dictably on each trial. Criterion for mastery of eachdiscrimination task was nine correct unpromptedresponses within 10 consecutive trials in a singlesession.

Student 1 was presented with four letters of thealphabet on each trial. The letter designated S+was held constant until it had been learned. Letterswere assigned to three groups taking into accountthat each group should contain no two letters es-

pecially liable to visual confusion. For example,"M" and "W," "Q" and "O" were placed indifferent groups. Student 2 was more advanced,and was therefore taught to discriminate 19 four-letter words selected from the Popper word seriesand assigned to three groups as above. Student 3learned to discriminate letters and numerals. Table1 presents the stimuli used with each student dur-ing each of the three conditions of reinforcement.

The students were first taught to select a singlestimulus within a group, with other stimuli in thegroup serving as S-. After the student was ableto reliably select the first group member, that stim-ulus then served as S- during subsequent trainingon a random basis. No stimulus ever served as S-in a training condition other than the one to whichit was originally assigned. The stimulus groupsremained intact throughout training. Stimuli fromone group were not interchanged with members of

177

PAUL E. TOUCHETTE andJANE S. HOWARD

Table 1Training Stimuli and Schedules of Reinforcement in Effect

Condition A (CRF/CRF)

Si M B U T X L A ZS2 long this good like cold from someS3 M B U T L Z

Condition B (CRF/FR3)

Si S H K I 0 W P E GS2 down ride away walk call playS3 S H K 0 G V 4 5 7

Condition C (VR3/CRF)

SI N Q Y D F C R V JS2 come make look jump stop helpS3 N Q Y D F R J 3 2 8

another group. A group of stimuli correspondedto each of the training conditions (see Table 1).

PretestBefore each training condition the stimuli were

presented to each student to determine whether or

not the discriminations had already been learned.Four trials with each stimulus were presented. Noneof the students performed above the 50% levelwith any of the words, letters, or numerals. Duringpretest trials correct responses were reinforced witha token and verbal confirmation. Errors producedonly the next trial. To ensure appropriate attention,these trials were interspersed with discriminationtasks already in the student's repertoire (e.g. "touchyour ear"). Correct responses to these baseline tasksreceived verbal confirmation.

Schedules of ReinforcementThe independent variable in this study was the

schedule of reinforcement provided for each of twoclasses of responses: (a) correct responses occurringprior to the prompt and (b) correct responses co-

inciding with or occurring after the prompt. Plastictokens were delivered contingent on correct re-

sponses. Food, toys, and special privileges servedto maintain the reinforcing function of the tokens.The three schedules of reinforcement were as fol-lows.

Condition A. Correct responses both before and

after the prompt were reinforced on a continuousreinforcement schedule (CRF/CRF).

Condition B. Correct responses that occurredbefore the prompt were reinforced on a continuousschedule while responses following the prompt werereinforced on a fixed ratio of 3 (CRF/FR3).

Condition C. Correct responses that occurredbefore the prompt were reinforced on a fixed ratioof 3 and responses after the prompt on a contin-uous schedule of reinforcement (FR3/CRF).

General ProcedureA trial began when the experimenter gestured

toward the array of four stimulus cards and said"look here." When the student looked toward thecards, the experimenter said "point to (letter orword)." Initially, the experimenter pointed to thecorrect card simultaneously with the verbal instruc-tion to point to the letter or word. Tokens weredelivered according to the reinforcement schedulein effect. All correct responses throughout the studywere followed by a verbal confirmation "right" or"good." Four consecutive correct responses pro-duced a 0.5-second increase in the delay betweenthe verbal instruction and the prompt, up to amaximum of 5 seconds for Students 1 and 3, and8 seconds for Student 2. Student 2 was permitteda longer delay because of her poor fine motor co-ordination.

If the student did not respond before the pre-determined delay, the experimenter pointed to thecorrect card. If the student made a response beforethe delay was reached, the prompt was not deliv-ered on that trial. Incorrect responses always re-ceived the same consequence. The experimentersaid "no," removed the cards from the desktop,and looked away from the student for 10 seconds.Two consecutive incorrect anticipations resulted inthe delay being decreased to the shorter of thelatencies emitted by the student in these two trials.Subsequent correct responses produced an increasein the delay value of 0.5 until the value at whichthe errors occurred was equaled. The incrementrule previously in effect was then reinstated, andfour consecutive correct trials were required to in-crease the delay value 0.5 seconds. The delay fol-

178

DELAYED PROMPTING

lowing errors was reduced to reinstate appropriatestimulus control, because the purpose of this studywas to assess the impact of schedules of reinforce-ment on the transfer of control from an essentiallyerror-free baseline ofprompted responses. Criterionfor mastery of each discriminative performance wasnine correct, unprompted trials out of 10 within asingle session. Figure 1 contains a flowchart ofevents during a training session.

The 0.5-second delay increment was chosenoriginally (Touchette, 1971) because it providedseveral exposures to the prompt prior to the delayvalue exceeding the student's normal response la-tency. In a nonautomated field setting, a half sec-ond is the smallest interval value that an experi-menter or teacher can use. The presence of auditoryor visual time signals offered a potential distractionto both the teacher and student. Prior to the studythe experimenters practiced with a stopwatch untilaccurate 1-second counts were achieved using thefamiliar silent verbal aid "one Mississippi, twoMississippi . . ." This covert verbal pacing aid wasthe basis for timing the delivery of the prompt.Keeping track of a half second was achieved byacting before the verbal pacing aid was completed.The 0.5-second unit occurs only once during thefinal count of odd valued delays. It does not occurat all in even valued delays.To increase the probability of the students' dis-

criminating the several schedules of reinforcement,the color of the mat on which the cards were placed,the background color of the cards themselves, andthe color of the tokens dispensed were changed foreach condition.

Condition A (CRF/CRF) exemplifies the mostcommon version of delayed prompting where theprobability of reinforcement is equivalent whethera correct response precedes or follows the prompt.Acquisition under condition A served as a standardfor comparing the relative efficacy of condition B,which favored anticipatory responding, and con-dition C, which favored responding following theprompt. All the stimuli assigned to condition Awere learned first by all students. Following ac-quisition of six to eight condition A discrimina-tions, stimuli associated with conditions B and C

were presented in alternation. Acquisition of dis-criminations under each of these two experimentalconditions went on simultaneously. This approachallowed evaluation of conditions B and C in awithin-subjects design with minimal contamina-tion as a result of sequence. This type of designhas been called "multielement" (Sidman, 1960;Ulman & Sulzer-Azaroff, 1975) or "alternatingtreatments" (Barlow & Hayes, 1979).Random order test. Following acquisition of all

discriminations associated with a particular sched-ule of reinforcement, a test was conducted to de-termine whether the verbal stimuli spoken at theoutset of each trial had acquired control over thestudent's card selection. The S+ stimuli were theletters, words, or numerals learned under the pre-vailing contingency of reinforcement. The stimulusdesignated S+ was now changed on every trial.To be correct, students had to respond on the basisof the name of the stimulus spoken by the exper-imenter. This requirement was not in force duringoriginal training where a single stimulus remainedas S+ until criterion was reached. The contingen-cies in effect during test trials were the same asthose during criterion trials with the following ex-ception. The delay value was not shortened follow-ing an incorrect response nor was the trial pre-sented again. The pointing prompt was presentedif the student failed to respond prior to the max-imum delay value permitted for that student dur-ing training (5 seconds for Students 1 and 3, 8seconds for Student 2). Each discriminative stim-ulus in the group was presented five times in arandomized sequence of 20 trials.

Generalization test. When all three sets of dis-criminative stimuli under all three contingencies ofreinforcement had been acquired, a generalizationtest session was conducted. Reinforcement contin-gencies, card color, mat, and tokens for the gen-eralization session were those that had been presentduring baseline condition A (CRF/CRF). Thismeant that discriminative stimuli learned underconditions B and C were presented with coloredcards, mats, and tokens different from those intraining. Any one of these changes might havebeen sufficient to disrupt performance, even though

PAUL E. TOUCHETTE and JANE S. HOWARD

YES

Figure 1. Flowchart analysis of a training session.

180

DELAYED PROMPTING

the stimuli changed were not directly relevant tothe discrimination. Changing all three elements si-multaneously was most likely to decrease accuracy.In addition, all three groups of stimuli were com-bined so that some or all of the S- stimuli oneach trial had never previously been seen in com-bination with the S+ presented. Each stimulusserved as S+ three times during the session. Theprompting conditions in effect were the same asthose for the random order test.

Data Collection and ReliabilityThe data collected on each trial specified the

type of response as follows: (a) correct preprompt,(b) incorrect preprompt, (c) correct postprompt,(d) incorrect postprompt. Responses coinciding withthe moment of prompting were treated as post-prompt responses, but were identified in the datarecord as simultaneous.

Reliability checks occurred approximately onceevery 15 sessions. To ensure independence of ob-servation, the experimenter did not consequate thechild's responses during reliability checks until thesecond observer silently signaled that the responsehad been recorded. One of the four response cat-egories inevitably occurred on each trial. Agree-ments were trials in which both experimenter andobserver identified the outcome as falling in thesame category. Disagreements were trials in whichobserver and experimenter recorded different cat-egories. Percentage agreement equaled the totalagreements divided by the sum of agreements plusdisagreements multiplied by 100. Interobserver re-liability ranged from 88% to 100% over 11 reli-ability check sessions involving all three students.The overall mean was 94.5%.

RESULTS

Accuracy During TrainingAll three students responded at high levels of

accuracy across experimental conditions. Students'mean correct responses across conditions A, B, andC, respectively, were as follows: Student 1-91%,97%, 92%; Student 2-99%, 89%, 100%; Stu-

60

40

20

Q)

0

F-Q

.0

~t

40

20

40

20

*A(CRF/CRF)Alternating Treatments

I 0 B (CRF/FR3) NC (FR3/CRF)

I~ ~ ~~SI i \ . l

\/ *\I i If,'

IaI! lk .mIj o~ .o

-e

1 3 5 7 2 4 6 8 10 12 14 16 18

-~~~~ /Il S:2[>;Zw1~oId, <S2.

I OA A

3 5 7 2 4 6 8 10 12

S:3

1 35 24 6 810 121416 18

Successive DiscriminationsFigure 2. Trials to criterion under the three experimen-

tal conditions. Each point represents a discrimination taught.

dent 3-98%, 99%, 99%. Overall response ac-curacy averaged 96% and was uniformly high.Variation was insufficient to reveal any functionalrelationship between the accuracy of respondingand the contingencies of reinforcement in effect.

Trials to CriterionFigure 2 shows the number of trials to criterion

for each discrimination, by student. Student 1 av-eraged 37 trials to criterion (range = 22-65) dur-ing condition A (CRF/CRF). During condition B(CRF/FR3) he averaged 17 trials (range = 14-25). In condition C (FR3/CRF) his average was38 (range = 15-62). Student 2 presented a sim-ilar pattern with an initial average of 35 duringcondition A (range = 26-47). Her average duringcondition B was 21 (range = 17-25). ConditionC produced an average of 31 trials to criterion(range = 23-44). Student 3 produced a moreuniform result. During condition A he averaged

PAUL E. TOUCHETTE and JANE S. HOWARD

20 trials to criterion (range= 17-27). Duringcondition B his average was 18 (range = 14-2 5).During condition C his average was 23 (range =15-31).

These data reveal that fewer trials were requiredto reach criterion during condition B (CRF/FR3)for all three students. There is, however, someoverlap in all cases. For Student 3 the differencebetween conditions B and C is minimal. It is in-teresting to note that condition B reduced thenumber of trials to criterion relative to conditionA for Students 1 and 2, but there was a failure ofreciprocity. In no case did condition C (FR3/CRF)prolong acquisition relative to condition A (CRF/CRF). Student 3 acquired discriminations in allthree conditions at near the minimal (14) numberof trials to criterion. A criterion run required 10trials, and the first four trials were prompted atzero delay precluding correct responses in antici-pation of the prompt. For this student, conditionA produced an intermediate value with conditionB slightly lower and C slightly higher. The trendshere are the same as those for Students 1 and 2but the variation is so small as to suggest no im-portant difference among the three conditions.

Preprompt ResponsesGiven the overall high percentage of correct re-

sponses, the remaining issue of interest is the pro-portion of those responses that preceded and fol-lowed the prompt. The number of correctanticipatory responses was limited by the criterionfor acquisition. Large numbers of correct pre-prompt responses could occur only if a mixture ofpre- and postprompt responses persisted long afterthe initial appearance of anticipatory responding.Such was not the case. There was no major differ-ence in the scatter of anticipatory responses acrosstreatments. Figure 3 shows that Student 1 exhib-ited a comparable proportion of preprompt re-sponses during conditions A and B, with a largerproportion during condition C. Students 2 and 3showed little difference across conditions. Inspec-tion of these data reveals overlapping distributions,suggesting no dinically important difference among

80

60

(C)

.Qc(-)

Q)

QD

.O

.CL<3;z

cIq

Q)

11)

Q)Q)

_o_

40

20

80

60

40

20

OA(C RF/C RF)

9~~~~.... ..

Alternating Treatmentso B (CRF/FR3) a C (FR3/CRF)

.o. S:lP--or 1 /a 1,

6

1 3 5 7 2 4 6 8 10 12 14 16 18

pSKI,'eit4°Wt S:2

& ' >:;# *fi~v Ipi'pI I

3 5 7 2 4 6 8 10 12

I" / u 5:3

35 246810I2141618

3 5 2 4 6 8 10 12 14 16 18

Successive DiscriminationsFigure 3. Percentage of responses that were correct an-

ticipations under the three experimental conditions.

the three contingencies. There is only a weak sug-gestion that condition C (FR3/CRF) lowered theprobability of anticipatory responses, despite a 3to 1 disparity in reinforcement probability favoringpostprompt responses.

Responses following or simultaneous with thepresentation of the prompt were dassified as promptcontrolled. In the absence of any significant num-ber of errors, prompt-controlled responses neces-sarily constitute the inverse of preprompt re-sponses. The observations stated earlier thereforeapply to this class of responses as well.

182

Il

I

DELAYED PROMPTING

20cn:30wz

_jWe Z

C')

cr

w2

Si o0 S2=0 S3 =-

15 _

c'OF

5 _

0 IL- I

A B C(CRF/CRF) (CRF/FR3)(FR3/CRF)

CONDITION

Figure 4. Mean percentage of responses occurring si-multaneous with the presentation of the prompt, under con-ditions A, B, and C.

Simultaneous ResponsesCorrect responses that occurred in very dose

proximity to the delivery of the prompt were treat-ed as prompt controlled, in terms of consequences,but were recorded separately as simultaneous re-sponses. It is impossible to determine whether theseresponses were emitted in response to the earlystages of the teacher's prompt or on the basis ofthe discriminative stimuli alone. Figure 4 showsthat simultaneous responses were infrequent, rang-ing from 1.3% to 12% with an overall mean of5% and a mean of less than 4% during conditionsB and C.

Delay in Effect at TransferFigure 5 shows the delay values in effect during

the first of the 10 consecutive trials that definedthe transfer of stimulus control. For Student 1,delay values at which transfer series were initiatedwere much lower during condition B than duringconditions A or C. Condition C (FR3/CRF) pro-duced the greatest variability. Students 2 and 3again exhibited overlapping distributions with ini-tiation points generally lowest during condition B

CZ

C)

-owCZ

10

4-0

Q)0)

4.0

3.0

2.0

1.0

2.0

1.0

0 A(CRF/CRF)Alternating Treatments

0 B (CRF/FR3) EC (FR3/CRF)

!

*l j

PU,3 5 7 2 4 6 8 10 12 14 16 18

4.ii.II '1 p

i

"% II %%

l 6 01, . . . . .

S:2

3 5 7 2 4 6 8 10 12

|.).\ S:3I Q A2/b-<ot

, . P..~ ~~/ % 'N: i--

3 5 2 4 6 8 10 12 14 16 18

Successive DiscriminationsFigure 5. The delay in effect on the first of the 10 con-

secutive trials that defined the transfer of stimulus controlunder conditions A, B, and C.

(CRF/FR3). It is interesting to note that Student2's delay value never approached the maximum(8 s), an unexpected outcome in view of her sub-stantial movement impairment. Data collected priorto this study, using familiar objects as discrimi-native stimuli, suggested that minimum responselatencies for this student were between 3 and 4seconds. Student 3 anticipated at delays of 2 sec-onds or less in all but one instance. There is heavyoverlap in Student 3's data across conditions A,B, and C.

PAUL E. TOUCHETTE andJANE S. HOWARD

Table 2Percentage of Correct Responses During Post Acquisition

Tests

Verbal control Generalizationtest test

A B C A B C(%) (%) (%) (%) (%) (%)

Si 68 64 93 74 78 96S2 86 90 90 80 90 97S3 10 60 20 06 41 32

Testing for Verbal Control andGeneralization

In verbal control tests the stimulus designatedcorrect was changed on every trial, making it nec-

essary for the child to respond on the basis of theexperimenter's naming the correct choice. Test ses-

sions included all of the discriminative stimulitaught under one of the three experimental con-

ditions. Generalization sessions used stimuli fromall three conditions ignoring the groupings in whichthey were originally taught. This meant that thediscriminative stimuli were juxtaposed with novelS- stimuli. The question was whether accurate

responding would persist when the stimulus pre-

sentations were not exactly as they had been duringtraining. The percentages of correct preprompt re-

sponses during both types of sessions are shown inTable 2. The data are virtually idiosyncratic. Eachof the students tested produced a different pattern

of competency. Student 2 alone responded withhigh (80% or above) accuracy across all tests. Stu-

dent 1 was uniformly above chance, but performedwith high accuracy only with group C stimuli.Student 3 achieved accuracy above 50% only withgroup B stimuli. All three students designated fa-miliar objects named by the experimenter duringpretesting. This skill was uniformly present across

students. Learning of auditory-visual correspon-

dences during the study, however, was not uni-form. Student 3 learned little or nothing, Student2 learned nearly everything, and Student 1 fell inbetween. All three displayed similar profiles of ac-

curacy at the end of training under the generaliza-tion test conditions.

DISCUSSION

A disparity in reinforcement density is inherentin delayed prompting. Unprompted responses arecloser to the trial initiation, and prompted re-sponses can occur only after the delay has timedout. This means that when the schedule of rein-forcement is the same for prompted and un-prompted responses, the actual density of rein-forcement over time is higher when a responseprecedes the prompt. Instead of waiting for theprompt, the student can respond right after a trialhas been initiated and get reinforced a lot sooner.The disparity can be very large when long delaysare used. Delay values have ranged as high as 30seconds in laboratory studies (Touchette, 1971)but shorter delays are typically used in clinical ap-plications, as they were in this study.

Something about inserting a delay prior to thepresentation of a prompt encourages transfer ofstimulus control. It is not yet clear whether thedelay of reinforcement that accompanies post-prompt responding is responsible for transfer orwhether the shift in control is provoked by a pair-ing of prompt and task stimuli, which approxi-mates the Pavlovian paradigm (Denny & Adel-man, 1955; Johnson, 1981). If contrast inreinforcement density provokes transfer, the rate oftransfer should be sensitive to manipulations ofcontingencies that alter reinforcement density. Onthe other hand, reinforcement might serve only tomaintain responding. If reliable sequential pairingof task stimuli and prompt is sufficient to provokea shift in stimulus control, we would expect alteredreinforcement contingencies to have limited impacton the rate of transfer.

In this study, we compared the impact of threeschedules of reinforcement on rate of transfer andfrequency of errors during discrimination learningby delayed prompting. One schedule (CRF/CRF)was a standard for comparison, representing themost commonly used parameters of equal conse-

184

DELAYED PROMPTING

quences for correct responses preceding and follow-ing a prompt. The second schedule (CRF/FR3)exaggerated the disparity of reinforcement in favorof correct responses prior to the prompt. The thirdschedule of reinforcement (FR3/CRF) provided ahigher probability of reinforcement for responsesfollowing the prompt. The discriminative stimuli,instructional prompts, and setting were appropri-ate to the behavioral ecology of severely handi-capped adolescents.

The first question was whether learning wouldoccur under all three schedules and whether errorswould or would not occur. Surprisingly, all threeexperimental conditions produced successful trans-fer from the visual prompt (teacher's pointing) tothe discriminative stimuli (cards containing letter,words, or numbers), and error rates were uniformlylow. Students 1, 2, and 3 were correct on morethan 90% of all trials across conditions. Alteredcontingencies of reinforcement affected the tenden-cy to wait for or anticipate a prompt in some cases,but they did not significantly alter the probabilityof error. This finding is an important one becauselow error rates are highly desirable in teaching thelearning handicapped (Reese, Howard, & Rosen-berger, 1977; Sidman & Stoddard, 1966; Touch-ette, 1968), and a procedural variant that tradedearlier transfer for more errors would be contrain-dicated.Was there any difference in how quickly transfer

occurred? Figures 2 and 5 indicate that condtionB, which favored anticipatory responding (CRF/FR3) produced earlier transfer from prompt totask stimuli in Students 1 and 2. Student 3 alsotended to initiate anticipatory responding earlierwhen the CRF/FR3 schedules were in effect, butthe difference was so small as to be of no practicalimportance. Student 1, who was the slowest toanticipate under the equal probability schedule(condition A), was the most dramatically affectedwhen contingencies favored anticipatory respond-ing (condition B). Student 2 also required moretrials to criterion and longer delays prior to transferwhen the probability of reinforcement was equaland when contingencies favored waiting for the

prompt (condition C). The impact of the disparateschedules was intermediate for Student 2. She wasmore dramatically affected than Student 3, but lessso than Student 1. Student 3 had the fewest trialsto criterion, and transferred most rapidly undercondition A. His pattern of transfer was minimallyaffected by conditions B and C. Student 3 showedthe same tendencies as the other two individuals.He transferred most rapidly when contingenciesfavored responses preceding the prompt, and hetransferred a bit more slowly when contingenciesfavored responses following the prompt.

The rate of transfer can be accelerated by dis-parate response consequences favoring respondingbefore the prompt. In order for improvement tooccur, however, there must be room for that im-provement. Student 3 learned rapidly and efficient-ly under condition A and could therefore improvevery little under condition B. Students 1 and 2learned less efficiently under condition A, and hadmore room for improvement. It was surprising tofind that the accelerated transfer under the CRF/FR3 schedule was not accompanied by an increasein "guessing" or incorrect anticipatory responses.

Some factors limiting the generality of thesestatements should be noted. The sample of subjectswas small and not homogeneous, but this is typicalof the severely handicapped. The sequence of ex-perimental events was fixed, so that each studentwas initially exposed to the "traditional" proce-dure based on identical schedules of reinforcement.This provided a basis for comparing the conven-tional procedure's outcome with that of the twosubsequent training conditions. Condition A, how-ever, contaminated conditions B and C by givingall students an immediate prior history that mighthave been relevant. The results of procedures Band C show that condition B accelerated transfer,whereas condition C reversed this effect in an al-ternating treatments design. This outcome makesit dear that the differences between conditions B(CRF/FR3) and C (FR3/CRF) were not due sole-ly to increased exposure to stimuli, procedures, orconsequences.

Surprisingly, condition C produced acquisition

PAUL E. TOUCHETTE andJANE S. HOWARD

data that strongly resembled those obtained duringoriginal training with nondisparate consequences.Considering only Students 1 and 2, where vari-ability was evident, the number of trials to criterionduring condition C averaged 34.5, and duringcondition A averaged 36.2. Student 3 transferredin a similar course regardless of the schedule ofreinforcement. The failure of condition C to pre-vent or seriously inhibit transfer in any studentsuggests that reinforcement probability is only onefactor influencing the course of discriminationlearning by delayed prompting.

Condition B was superior in facilitating originaldiscrimination learning, but condition C producedthe best posttraining accuracy in Students 1 and2. The mean numbers of trials to acquisition werevirtually identical for conditions A and C, againruling out greater exposure as an explanation forthe verbal control test and generalization scores.Student 3 evidenced little control by any of theverbal stimuli during random order tests. This re-sult was unanticipated, and it was not possible tocarry out manipulations necessary to determine itsorigins post hoc.

All three students had previously evidenced somereceptive language competence. Each had beentested, and was able to comply with simple re-quests involving familiar objects and body parts(e.g. "Hands down please, touch your shoe . . .").They were under the control of common verbalstimuli at the outset of the study. The verbal re-quests that accompanied the visual discriminationsestablished in the course of the study did not ac-quire reliable control of responding. This stimulus-response relation, not explicitly trained, was notspontaneously available. Few assumptions of inci-dental learning are -reliably validated among theseverely learning handicapped. Each individualmust be tested to determine what, if any, learninghas occurred beyond that which was programmed(Marholin & Touchette, 1979; Rincover & Koe-gel, 1975). Desirable controlling relations canemerge spontaneously following training of relateddiscriminations (Sidman, 1971; Spradlin, Cotter,& Baxley, 1973) but most must be directly taught(Stokes & Baer, 1977). It has been suggested that

generalization and incidental learning followingerrorless discrimination acquisition may be a spe-cial case (Terrace, 1964; Schilmoeller, Schilmoell-er, Etzel, & LeBlanc, 1979). There are, however,only the most preliminary data available on thissubject.

In addition to providing an observable indica-tion of the transfer of stimulus control, currentevidence suggests that many instructional goals canbe achieved by inserting a delay prior to the deliv-ery of stimulus that controls the behavior of inter-est. The objective of delaying the prompt is to shiftcontrol from stimuli supplied by the instructor tostimuli inherent in the task. The classroom is onlyone possible setting. Halle, Marshall, and Spradlin(1979) used a 15-second prompt delay to provokeverbalizations from severely retarded children wait-ing to be served at a lunch counter. They used thedelay interval to bring behavior that they knew tobe available to the trainees under the control of itsnatural precursors. Those authors emphasized thestimulus function of the delay interval, which con-stituted an avoidable time-out prior to being servedlunch. It appears, however, that the delay of re-inforcement is not exclusively responsible for stim-ulus control transfer in this paradigm. Practicalapplications of delayed prompting need not awaita complete functional analysis of transfer amongsequenced stimuli.

The essence of delayed prompting is that a stim-ulus that controls the response of interest (a prompt)is presented concurrently with the task-relatedstimuli. A delay is then interposed between thepresentation of the task and the onset of theprompt. Stimulus materials do not appear to becritical and there is no need to develop elaborate,subtle, and sometimes unique graduated changesnecessary to effective fading (Etzel, LeBlanc, Schil-moeller, & Stella, 1981). A number of studieshave shown that this approach can teach usefuldiscriminations to handicapped and young unim-paired children (Johnson, 1978; Smeets & Striefel,1976; Stremmel-Campbell, Cantrell, & Halle,1977; Striefel, Bryan, & Aikins, 1974; Striefel,Wetherby, & Karlan, 1976). Parametric studieshave until now been conducted only with lower

186

DELAYED PROMPTING 187

organisms (Brown & Rilling, 1975; MacDonall &Marcucella, 1976).

Etzel and LeBlanc (1979) suggested that fadingis a more powerful instructional approach than de-layed prompting. Although this may be the case,there is little evidence currently available one wayor the other. A recent study by Bradley-Johnson,Johnson, and Sunderman (1983) failed to sub-stantiate this assertion in teaching readily confusedletter discriminations to preschool children. Eachtechnique has been shown to produce errorlesslearning under some circumstances, and each shouldbe considered as an alternative when the other fails.

The decision of whether to apply disparate oruniform consequences to pre- and postprompt re-sponses is a practical one. This study shows thatproviding higher reinforcement probability for an-ticipatory responses can produce more rapid trans-fer of stimulus control. The drawback in applica-tion is that the procedure is more complex whendifferent schedules of reinforcement apply to re-sponses before and after the prompt. The alter-native of using uniform consequences for all correctresponses does not prevent or seriously retard trans-fer of stimulus control, although more trials andlonger delays will result in some cases.

Students should be given an opportunity to re-spond prior to prompting (Hart & Risley, 1975;Halle, Baer, & Spradlin, 1981). We suggest thatthere is more to delayed prompting than offeringan opportunity to respond. Pausing prior to deliv-ering a prompt defers reinforcement of prompt-controlled behavior. We suggest that there is moreto delayed prompting than a differential in rein-forcement density favoring responses that precedethe prompt. In this study, when the probability ofreinforcement was three times as high for responsesfollowing the prompt, students still reliably shiftedto responding prior to the prompt. This is a shiftin stimulus control that occurred despite theschedule of reinforcement. It is reasonable to assert,given these data, that response consequences onlypartially determine the transfer of stimulus controlin this paradigm. Transfer appears to be provokedwhenever a suitable noncontrolling stimulus is re-liably followed by a controlling stimulus.

REFERENCESBarlow, D., & Hayes, S. (1979). Alternating treatments

design: One strategy for comparing the effects of twotreatments in a single subject. Journal of Applied Be-havior Analysis, 12, 199-210.

Bradley-Johnson, S., Johnson, C., & Sunderman, P. (1983).Comparison of delayed prompting and fading for teach-ing preschoolers easily confused letters & numbers. Jour-nal of School Psychology, 21, 327-335.

Brown, C., & Rilling, M. (1975). Stimulus delay and thereduction of error in the transfer of stimulus control.Animal Learning and Behavior, 4, 341-346.

Carr, T., Newsom, C., & Binkoff, J. (1980). Escape as afactor in the aggressive behavior of two retarded chil-dren. Journal of Applied Behavior Analysis, 13, 101-117.

Denny, M., & Adelman, H. (1955). Elicitation theory: 1.An analysis of two typical learning situations. Psycho-logical Review, 62, 290-296.

Etzel, B., & LeBlanc, J. (1979). The simplest treatmentalternative: The law of parsimony applied to choosingappropriate instructional control and errorless-learningprocedures for the difficult to teach child. Journal ofAutism and Developmental Disorders, 9, 361-382.

Etzel, B., LeBlanc,J., Schilmoeller, K., & Stella, M. (1981).Stimulus control procedures in the education of youngchildren. In S. W. Bijou and R. Ruiz (Eds.), Contri-butions of behavior modification to education (pp. 3-37). Hillsdale, NJ: Lawrence Erlbaum Associates.

Halle, J., Baer, D., & Spradlin, J. (1981). Teachers' gen-eralized use of delay as a stimulus control procedure toincrease language in handicapped children. Journal ofApplied Behavior Analysis, 14, 389-409.

Halle,J., Marshall, A., & Spradlin,J. (1979). Time delay:A technique to increase language use and facilitate gen-eralization in retarded children. Journal of Applied Be-havior Analysis, 12, 431-439.

Hart, B., & Risley, T. (1975). Incidental teaching of lan-guage in the preschool. Journal of Applied BehaviorAnalysis, 8, 411-420.

Huguenin, N., & Touchette, P. (1980). Visual attentionin retarded adults: Combining stimuli which control in-compatible behavior. Journal ofExperimental Analysisof Behavior, 33, 77-86.

Johnson, C. (1978). Errorless learning in a multihandi-capped adolescent. Education and Treatment of Chil-dren, 1, 25-33.

Johnson, C. (1981). Elicitation theory: II. Analysis of twomore learning situations. Academic Psychology Bulle-tin, 3, 119-130.

MacDonall, J., & Marcucella, H. (1976). Cross modaltransfer of stimulus control in the albino rat: A stimulusdelay procedure. Animal Learning and Behavior, 4,34 1-346.

Marholin, D., & Touchette, P. (1979). The role of stim-ulus control and response consequences. In A. Goldsteinand F. Kanfer (Eds.), Maximizing treatment gains (pp.303-351). New York: Academic Press.

Ray, B., & Sidman, M. (1970). Reinforcement and stim-

188 PAUL E. TOUCHETTE andJANE S. HOWARD

ulus control. In W. N. Schoenfeld (Ed.), The theory ofreinforcement schedules (pp. 187-214). New York:Meredith Corporation.

Reese, E., Howard, J., & Rosenberger, P. (1977). Behav-ioral procedures for assessing visual capabilities in non-verbal subjects. In B. C. Etzel, J. L. LeBlanc, and D. M.Baer (Eds.), New developments in behavioral research:Theory, method and applications. In honor of SidneyW. Bijou (pp. 279-301). Hillsdale, NJ: Lawrence Erl-baum Associates.

Rincover, A., & Koegel, R. (1975). Setting generality andstimulus control in autistic children. Journal of AppliedBehavior Analysis, 8, 235-246.

Schilmoeller, G., Schilmoeller, K., Etzel, B., & LeBlanc, J.(1979). Conditional discrimination responding aftererrorless and trial-and-error training. Journal of the Ex-perimental Analysis of Behavior, 31, 405-420.

Sidman, M. (1960). Tactics of scientific research. NewYork: Basic Books.

Sidman, M. (1971). Reading and auditory-visual equiv-alences. Journal of Speech and Hearing Research, 14,5-13.

Sidman, M., & Stoddard, L. (1966). Programming per-ception and learning for retarded children. In N. R. Ellis(Ed.), International review of research in mental re-tardation (Vol. 2) (pp. 151-208). New York: Academ-ic Press.

Skinner, B. F. (1968). The technology of teaching. NewYork: Meredith Corporation.

Smeets, P., & Striefel, S. (1976). Acquisition of sign read-ing by transfer of stimulus control in a retarded deafgirl. Journal of Mental Deficiency Research, 20, 197-205.

Spradlin, J., Cotter, V., & Baxley, N. (1973). Establishinga conditional discrimination without direct training: Astudy of transfer with retarded adolescents. AmericanJournal of Mental Deficiency, 77, 556-566.

Stokes, T., & Baer, D. (1977). An implicit technology ofgeneralization. Journal of Applied Behavior Analysis,10, 349-367.

Stremmel-Campbell, K., Cantrell, D., & Halle, J. (1977).Manual signing as a language system and as a speechinitiator for the nonverbal severely handicapped student.In E. Sontag (Ed.), Educational programming for theseverely and profoundly handicapped (pp. 335-347).Reston, VA: Council for Exceptional Children.

Striefel, S., Bryan, K., & Aikins, D. (1974). Transfer ofstimulus control from motor to verbal stimuli. Journalof Applied Behavior Analysis, 7, 123-135.

Striefel, S., Wetherby, B., & Karlan, G. (1976). Estab-lishing generalized verb-noun instruction-following skillsin retarded children. Journal of Experimental ChildPsychology, 22, 247-260.

Terrace, H. (1963). Discrimination learning with andwithout "errors."Journal ofthe Experimental Analysisof Behavior, 6, 223-232.

Terrace, H. (1964). Wavelength generalization after dis-crimination learning with and without errors. Science,144, 78-80.

Terrace, H. (1966). Stimulus control. In W. K. Honig(Ed.), Operant behavior: Areas of research and appli-cation (pp. 271-344). New York: Meredith.

Touchette, P. (1968). The effects of graduated stimuluschange on the acquisition of a simple discrimination inseverely retarded boys. Journal of the ExperimentalAnalysis of Behavior, 11, 39-48.

Touchette, P. (1971). Transfer of stimulus control: Mea-suring the moment of transfer. Journal of the Experi-mental Analysis of Behavior, 15, 347-354.

Ulman, J., & Sulzer-Azaroff, B. (1975). Multielementbaseline design in educational research. In E. Ramp &G. Semb (Eds.), Behavior analysis: Areas of researchand application (pp. 377-391). Englewood Cliffs, NJ:Prentice-Hall.

Weeks, M., & Gaylord-Ross, R. (1981). Task difficultyand aberrant behavior in severely handicapped students.Journal of Applied Behavior Analysis, 14, 449-463.

Received April 11, 1983Final acceptance January 23, 1984