Back to Section Contents

Back to General Contents

Chapter VIII

STIMULUS CONTROL, PART 2:

TRAINING PROCEDURES
Contents
Introduction: discrimination procedures, generalization, psychophysics

Operant discrimination procedures: prerequisites

Potentiation
Magazine training
Some special considerations
Control and RD-RM separation
Discriminative linkage
Response topography
D ∆
Standard S -S procedure
Extinction and consequences
D ∆
Establishment of S -S discrimination
Simultaneous presentation

Oddity procedure

Establishment of oddity discrimination

Prior conditions
The chain
Discriminative linkage
Terminal discrimination training
Successive presentation

Match to sample procedure

Sample-match interval
The chain
Discriminative linkage
Terminal discrimination (matching) training
Successive presentation (Local Contents continued on next page)
Programming errorless discrimination: fading
Magazine training and discriminative linkage
D ∆
Errorless programming of S -S discrimination
Specifying the terminal discrimination
Assessing the relevant initial discriminative repertoire
The program

Errorless programming of Oddity discrimination

Specifying the terminal discrimination
Assessing the relevant initial discriminative repertoire
Program One
Program Two

Errorless programming of match to sample

Specifying the terminal discrimination
Assessing the relevant initial discriminative repertoire
The program
Program evaluation
Teaching reading and visual match to auditory sample

Errorless programming of adjustment

Other fading of stimulus control

Measurement of stimulus control

(Back to Contents)

Back to General Contents

 INTRODUCTION:
DISCRIMINATION PROCEDURES,
GENERALIZATION, PSYCHOPHYSICS

The following sections will be devoted to procedures that establish operant

discrimination. These include the traditional operant procedures, as well as newly developed
programming procedures that establish discrimination without error. We shall also consider
extensions to abstraction, conceptualization, and verbal behavior.

Generalization is often brought up in the context of discussions on discrimination. In

complete generalization, a response that is established under one set of conditions continues
when those conditions are changed. For example, if the doorbell rings and we pick up the
telephone, we are considered as generalizing, across an auditory dimension. It will be noted that
we are not discriminating between the two tones, or that we are confused. Generalization has
been, accordingly, regarded as the reverse or opposite of discrimination. Some writers have
asked whether both terms are necessary, since by attaching appropriate consequences, we can
“improve” the discrimination, or reduce the generalization, between doorbell and telephone bell,
which is saying the same thing. Other writers have argued that generalization is extremely
important since it deals with the extension of behavior, learned under one condition, to
conditions where it was not established. A child who is bitten by a spaniel may avoid petting a
German shepherd. He has never seen this dog before. The conditions under which this transfer
occurs have been the subject of much generalization research. The reader will recognize the
similarity between this discussion of generalization and our discussion of the stimulus class.
Needless to say, such generalization, or inclusion in the same stimulus class, may be useful, as in
the dog case, or a nuisance, as in the telephone case, or downright embarrassing, as when the
child greets a traveling salesman with “dada.” Generalization has been used as an explanatory
principle to account for such behavior under novel circumstances.

Generalization curves often relate response strength to the topographic similarity between
the novel situation and the original one. If we are trained to respond to a telephone bell with one
pattern of ringing, to what extent will we respond to a slightly different pattern or to a doorbell,
an alarm clock, and so on? In psychophysical research, the observer is instructed to identify a
certain tone by a specified response. Other tones may then be presented. To what extent will she
confuse them with the initial tone? Obviously, the more different they are, the less often she will
be confused. A curve maybe drawn relating accuracy to the topographic stimulus separations.
The similarity between the generalization experiment and the psychophysical one should be
noted. The languages and concepts are quite different, however, and the psychophysical
investigator will neither discuss nor explain his results in terms of generalization to novel
situations. He may use the terms detection and confusion, with one representing absence of the
other, quite analogously to the use of discrimination and generalization in the learning literature.
The psychophysicist may design his experiments in terms of decision processes. Applied to the
telephone situation, a decision analysis suggests that the observer may respond in at least two
ways. She may classify the ringing sound as a telephone signal, or may classify it as a different
noise, for example, an alarm clock. When the telephone rings, she can go to the telephone or to
the alarm clock, as she can when the alarm sounds. In addition to the distinctiveness of the
stimuli, the consequence of the “decision” will enter into it. If she expects a very important
phone call, for example, she may make many telephone responses to the alarm.

The relation of psychophysical theory and procedures to generalization must await our
exposition of psychophysics that, it will be reiterated in passing, handles the functional relations
and data of generalization without ever resorting to the term. However, generalization research
and discrimination research have been associated with each other, and because of such
association, we shall consider the relevant generalization procedures when we discuss
discrimination procedures. The utility of the term, generalization, will be discussed separately.

A general rule of thumb that distinguishes the areas of discrimination and generalization
may be abstracted from the following examples. An experimenter establishes discrimination
between green and red, that is, he obtains differential responding. In generalization research, he
may seek to ascertain what happens when stimuli between green and red are presented. Indeed, a
continuum can be established consisting of green, green-yellow, yellow, yellow-red, and red,
with numerous points between. The investigator may also employ greens and reds outside the
red-green limits on the continuum (blue-green, red-violet). If we obtained a peak rate at red (it
D ∆
was S ), and a very low rate at green (it was S ), we might obtain a high, but not peak rate at
red-yellow; somewhat lower rate might be obtained at yellow, and so on. This would produce a
curve called a generalization gradient, as shown in the illustration.

Now, we might want the slope of this curve to

be more steep. Stated otherwise, we want greater
differences in responding between the green and red
ends of the continuum. We might now introduce
procedures to sharpen the discrimination between the
green and red ends of the continuum. Our procedure
D
would still keep red as S , but we might also make
∆
green-yellow S . The very low rate may now appear at
green-yellow, with responding at green now possibly
lower than before. We have extended the boundaries of
∆
the S stimulus class. We could similarly extend the
D
boundaries of the S stimulus class, by reinforcing responses when stimuli to the right of red
appear. The sharpened discrimination would be depicted by a steepened generalization gradient,
as indicated in the illustration. It will be noted that we can, in this case, speak interchangeably of
discrimination and generalization. The situation may also be described in terms of stimulus class
and abstraction, in this case, warm colors and cool colors. Procedures for establishing such
discriminations will be the subject of the next few sections. Since stimulus class is involved, the
procedures will not be restricted to colors and forms, but will be extended to the establishment
and sharpening of the complex concepts so meaningful to many human investigators.

Generalization procedures are closely linked to psychophysical procedures. A body of

procedures has been developed in psychophysics, called psychophysical methods. These
methods have proven important not only in the study of discrimination and perception, but also
in the study of the complex concepts used in mental tests and measurement. Many of these
procedures are analogous to the operant procedures that we shall now present. However, the
psychophysical methods stem from a different historical context, in answer to problems that have
been formulated differently, that have required different emphases, and that have dictated
different areas of instrumentation and procedural analyses. To the extent that there is overlap
between operant and psychophysical research, these differences suggest that both areas may
learn from each other, and can adopt refinements that the other has already worked out. The
differences also suggest that the two areas had best be approached separately. Accordingly, our
discussion of psychophysical methods will be deferred to later sections. After these have been
presented, we shall consider the overlap, and the topic of generalization.
(Back to Contents)

OPERANT DISCRIMINATION PROCEDURES:

PREREQUISITES
Although some operant discrimination procedures differ considerably from each other,
there are commonalities to them implied in the common term, operant. In, this section, we shall
discuss these commonalities, some of which are relevant to setting up any operant experiment.

While we can put a person in a room and tell him to “go to it,” we cannot put a pigeon in
a box and expect it to respond appropriately in a discrimination task. A colleague once trained
two pigeons in such tasks, and then sent the apparatus and the birds, by express, to another
university, where they were used in a classroom demonstration. One of the pigeons subsequently
died. The lecturer at that school then substituted a pigeon he bought locally. He was indignant
when the bird did not perform and returned the equipment as defective. He also questioned the
generality of operant methodology, since it did not even work with all pigeons. He might also
have questioned the usefulness of books, since when they are given to people who have not
learned to read, the printed word does not produce the behavior it is supposed to. The lecturer
was unaware of the prerequisites to discrimination behavior, that will be the subject of this
section.

Whether we are training a person or a pigeon, or whether we are assessing discrimination

D ∆
in a person or a pigeon, we shall have, in the simplest case; an S and S , with differential
D
responding to each. Where the S does control behavior, the following chain will be considered
D
as the paradigm. The S , say red, is presented on a key for a pigeon, and on a small panel for a
child. Pecking the key or pressing the panel when they are red, and not otherwise, defines
discrimination.

The vertical line we have drawn splits the chain into two parts. To the left of the line, the
stimulus controls the behavior we are studying, namely, discrimination. To the right of the line,
the behaviors deal with obtaining the reinforcer. (We could have extended this side to include
eating, but for our present purposes this is not necessary.) These are discriminative behaviors in
that they are controlled by the magazine sounds (and other stimuli), but with respect to our
investigation, they are not the dependent variable, but are among the supporting behaviors
necessary. They are labeled RM, or magazine behaviors (that actually comprise a chain that we
are condensing into a response). Since the discrimination we are investigating is established and
maintained by differential reinforcement, the magazine behaviors, that are more closely related
D
to reinforcement, are established first. In Premack’s terms, the initial S sets the occasion for

behaviors, RD, which are reinforced by getting (RM) food. If the food is not a reinforcer, the
whole chain may collapse. Accordingly, we start out with the reinforcer. Our training sequence
will be:

These will be considered, in order, under the headings of (1) Potentiation, (2) Magazine
training, and (3) Discriminative linkage. We shall discuss these in terms of commonalities and
relevance to all the operant discrimination procedures.
(Back to Contents)

1. Potentiation: First, of course, we must find a reinforcer that will maintain the
behavior we are interested in. With animals, it is often simple to use food. The pigeon is
deprived of food until he is at eighty-percent of normal free-feeding body weight. Water may
also be used with its corresponding deprivation. Some aquatic animals present a problem, since
frogs and turtles can go without food for extended periods of time. In one set of procedures,
access to air was made their reinforcer. Intracranial stimulation has also been used. These
reinforcers require special apparatus and training procedures.

With children, food and candy have been used as reinforcers, and money and tokens have
been used with adults as well. Praise or approval may be offered. Many children and adults
have had a history of success and its consequences, and success alone, or doing the assigned
task, may suffice. Behavior may also be used as a reinforcer.

None of these reinforcers is a reinforcer per se, but all require potentiation. The reader’s
attention is called to those sections of the preceding chapter that deal with the various reinforcers
available, and the ways of potentiating them.
(Back to Contents)
 2. Magazine training: Once we know that we have a potent reinforcer, our next task is to
establish stimulus control over the magazine behavior. Stated otherwise, we want the organism
to go to, or reach for, the reinforcer only upon signal. Otherwise, he may spend all his time
hanging around the food magazine. A more important reason for such stimulus control is that
the signal can then be used as a linked reinforcer that is presented only when the appropriate
discriminative response is made. The general term for this training, that bridges the gap between
the discriminative response and the magazine behavior, is magazine training. Where the animal
is trained to go to the food magazine when food is presented, the term is being used literally, of
course. Where there is no literal food magazine, but where a bridge must be established between
discrimination and consummation, the term is used as a metaphor. For example, the mother who
says, “You may come to me for a big kiss if you do it right,” is engaging in magazine training by
instructional means.

The type of magazine training will vary with the species, the reinforcer, the apparatus,
and the task. In magazine training a pigeon, the food magazine is initially open and illuminated;
the lights in the rest of the chamber are dimmed. After the pigeon has eaten for a few seconds,
the magazine is closed, and the illumination of the overhead lights is raised. Thereafter, the
magazine is opened and closed at different times and when the pigeon is in different places.
Opening of the magazine is accompanied by visual stimuli (overhead lights dim, magazine
illuminated) and sounds (buzzers, clicks, thumps). When the pigeon goes to the magazine as
soon as these changes occur, and not otherwise, magazine training is considered established; the
changes are now conditioned reinforcers. We can use them in the establishment of
discrimination.

In magazine training a child, an M&M candy may be rolled down a dispenser, with
accompanying sound effects. Verbal instructions may be used; the child may be told that when
the sound effects occur, an M&M is on its way. The student may be told that the approval of the
laboratory assistant is critical for passing the course. Such use of instructional control saves
considerable time in human magazine training.

Once the conditioned reinforcers seem to control the magazine response, the
investigator’s attention may then be turned toward establishing the discriminative response. The
instructional control that may establish magazine training should be distinguished from the
instructional control that may be used to establish discrimination, for example, “Press the panel
only when it lights up red.”

Magazine training is facilitated when the conditioned reinforcers are “loud and clear.”
As the reader will recall, in obedience training of dogs, the trainer is urged to let the dog “know”
in no uncertain terms that he has made the appropriate response. The trainer is urged to throw
himself around the dog’s neck, to hug and pat him, and to say “Good dog!” If he can also
provide something to eat, all the better. The dog is not distracted by this procedure, nor does he
merely learn the difference between reinforcement and nonreinforcement. The differences
between these two may actually facilitate learning the related discrimination contingency, whose
learning is the purpose of the differential reinforcement procedure. The reader is reminded of the
video game, where the appropriate response may involve pushing pairs of buttons just so, in an
exquisite discrimination of one’s own behavior. Here, any noise or whisper may be distracting.
But once the “game piece” runs true, the view screen lights up, fireworks may be presented, and
the marine hymn may be played. Video game machines suffer no loss of customers. Instructions
may be used in conjunction with such conditioned reinforcers: “When you are correct, the
fireworks will start.”

Magazine training typically involves providing occasional free reinforcers. In the pigeon
case, the magazine is often initially left open until it is used for some time. With children, a few
candies may be given away at the outset. A good teacher in a classroom may supply
encouragement of any behaviors. In Sidman’s project with children with mental retardation, it
will be recalled, the tokens were first given away free, and were then collected in a candy
exchange. In the project with juvenile delinquents, it will be recalled, Cohen initially allowed
free access to his pool-hall. In a later project, where, quality of food was a differential
reinforcer, all entering students were given a free week on Class A. Thereafter, they had to pay.
Throughout the project, all Sunday meals were Class A. People who may miss out on the finer
things often may not know what they are missing. The Sunday case brought this to the attention
of the delinquents, along with the posted signs listing the menus of all the meals for the
forthcoming week. A World War I song ran: “How’re you gonna keep ‘em down on the farm,
after they’ve seen Paree?” Presumably, unless they had first seen Paree, they would not be
interested in returning. A first exposure has to be made to get the system moving, but how do we
get them to go there in the first place, when they don’t know what they’re missing? Hence the
free sample, the introductory offer, and other such examples of “pump-priming.”
(Back to Contents)

Some special considerations: A procedure which works well with one species or
individual -- or one reinforcer, or one apparatus, or one task -- will not necessarily work well
with another. Every situation can be considered a special situation, requiring special treatment.
The aim is to produce the same functional results known as magazine training.

In the human case, the special treatment often involves use of instructions. It should
D
always be remembered that these are S s, and require differential reinforcement to maintain their
control. To return to the military example, sound out your orders loud and clear -- and back
them up.

Magazine training may not be needed at all under some special conditions. Where air
was used as a reinforcer for amphibians, the water level was up to the ceiling. Reinforcement
consisted of suddenly lowering that level by opening a valve. Here the animal did not have to be
magazine trained to go to the reinforcer, since it was brought to him. In an alternative version, a
trapdoor in the ceiling was opened. This allowed the animal to get air. Magazine training was
required here to train the animal to go to the trapdoor when the lights and sounds changed.

The reader may ask: Why complicate the situation, and have magazine training at all?
Why not have the discriminative response also be the magazine one? For example, we can form
a red-green discrimination using intracranial stimulation. When the red light goes on, and the
animal presses, he immediately sets ICS. He does not then have to go to an ICS magazine. Why
not try this procedure for other reinforcers, as, well? Ethological investigators of chicks’ color
preferences have used grains of different colors to assess such preferences. If one wishes to
establish color discrimination in pigeons, why not color the grains differently, and have the red
grains glued permanently to the panel with the green ones removable? We do learn to
discriminate red from green apples, and ripe from unripe fruit in general. Such discrimination
can be rapidly acquired, and finds use in shaping the discriminative response (to be discussed
later in this section). Why not make the discriminative response and the magazine response one
and the same, condense the chain, and eliminate the necessity for magazine training? We shall
consider this question from the viewpoint of the control provided, and thereby the analysis
permitted.
(Back to Contents)

Control and RD -RM separation: Although the delivery of reinforcement is more

immediate when the discriminative response is the magazine response, than when the
discriminative response is followed by a separate magazine response, we shall see that separating
the two responses provides for greater control. For one thing, different data are obtained. We
shall, in this context, review the ICS procedure developed by Pliskoff. It will be recalled, that
when food is used as a reinforcer, RD is followed by RM, but when ICS is used as a reinforcer
they are identical:

Parenthetically, it will be noted that after we get food, we eat it. This is called the
r
consummatory response, or RC. Premack considers the RC of eating, rather than the S of food

itself, as the reinforcer. What RC follows ICS delivery is not known. For our present purposes,

as we shall see the RC difference is not important.

The ICS chain is much simpler than the food chain. However, using this ICS procedure,
it proved impossible to replicate many of the data on the effects of schedules of food and water
reinforcement. It was difficult to maintain behavior on a large FR, a long temporal schedule, or a
high variable schedule. The control by this reinforcer, when it was given, was effective. The
perseveration of its control, when it was absent in the nonreinforced sections of these schedules,
was, however, very poor. Reinforcers that provide control only when they are on CRF (or close
to it) are of limited usefulness, both in the laboratory and outside it. This difference between ICS
and other reinforcers led to considerable theorizing and speculation about brain areas,
physiological as opposed to other reinforcers, and so on.

However, inspection of the diagram tells us that the two chains differ not only in
reinforcers used, but in procedures. Pliskoff noted that in one case, RD and RM were separated,
but in the other they were not. To ascertain which of the differences was critical (reinforcers or
procedures), he separated RD and RM in an ICS experiment:

This chain is functionally equivalent to the food chain, or the inherent chains used for
most other reinforcers. Behavior has been maintained under complex schedules using these
reinforcers. When Pliskoff used the same chain for ICS, it, too, could maintain behavior under
complex schedules. The difference was in chains, not reinforcers. The critical difference in the
chains was the separation of RD and RM. Separation provides greater control. This supplies one

answer to the question: Why not combine RD and RM? Parenthetically, it also indicates the
advantage of a procedural over a conceptual or process analysis. The process of discrimination
is going on in both chains, and we would call the procedures in both cases discrimination
procedures. The differences obtained in results can not be explained on the basis of differences
in discriminative processes, but can be related to procedural differences.

Where the RD and RM are combined, as they were in the ICS case, we could not separate
the two functions. Separating the two suggests that we can take any arbitrary response and use it
as RD. In conditioning terms, we can hitch anything to anything, and we are not bound by

limitations in the RM or reinforcer to choose our discriminative response. We can use any
response that suits our design. If we are interested in analysis and measurement, we can choose
or even develop a response that is amenable to such computation.
(Back to Contents)
 3. Discriminative linkage: Once the pigeon goes to the food magazine when the buzzer
is sounded, and not otherwise, we may start to establish our desired discrimination. The pigeon
is already discriminating buzzer from no buzzer, and food magazine from other parts of the
apparatus. Without further ado, we could start discrimination training for different buzzers. This
might involve presenting food when certain buzzers are sounded, and having the pigeon go on a
wild goose chase to the (empty) magazine when others are sounded. The buzzers might vary in
sound pressure level (loudness), in location, and so on. By capitalizing on behavior already
established in the repertoire, we can train in discrimination very rapidly.

We shall, however, consider the more complex case where we use pecking a key to study
visual discrimination. The key is used because it is part of standard apparatus, and its arbitrary
nature provides us with the advantages just discussed. We are not limited to the topography of
the conditioned reinforcers and can arbitrarily use any dimensions. Further, we can study
different discriminations using the same response.

Our terminal requirement is to have the pigeon peck the disk when it is red and not when
it is green. This assumes that he is already pecking the disk, something he has not been doing up
to now. He has merely been running to the magazine upon signal. Our temporal program is as
follows:

1. Establish magazine control.

2. Get him to peck disk, and not peck elsewhere

3. Get him to peck disk when red, and not when green.

Step 1 has been established. It will be noted that from 2 to 3 we are going from the
general to the specific. Step 3, our terminal requirement, is the subject of later sections on
discrimination training. The present section is concerned with linking the magazine control of 1
to the discriminative control of 3, or discriminative linkage.

To get the pigeon to peck the disk, we have a powerful device at our disposal. This is the
r
set of stimulus changes that always accompanied S during magazine training, and that are now
so powerful that they control running to the food magazine whenever they are presented. Since
they exert such control, we can now make them contingent upon some behavior. Whatever
behavior produces them will be reinforced thereby. The response requirement for reinforcement
will be pecking at a disk on the wall. This will activate the magazine and produce the
conditioned reinforcers with the rapidity of the electrical impulses involved. Accordingly, once
the appropriate response is made, it is reinforced immediately, and is likely to recur again. The
problem is to get that first response to occur. Once that response is established, we can start
discrimination training, the subject of the next sections.
(Back to Contents)

There are several general procedures available to produce this first response. Which
procedures are used will vary with the individual, the species, the apparatus, and the state of the
art. The procedures may be outlined as follows:

a. Single-step method. Here, the pigeon is left to his own devices in the experimental
chamber. The experimenter checks periodically to see if he has performed. In setting up the
chamber, an effort is made to emphasize the disk -- for example, the chamber may be dark, with
the key illuminated. Normally, during a 24 hour period, the animal will peck at least once, since
the operant level is greater than zero, and once is all that is needed if he is well magazine-trained.
Needless to say, in this procedural variant of learning the hard way, or learning by discovery, the
pigeon is required to be a rugged individualist. If he does not succeed, he may die of starvation.

b. Shaping. This is a programming procedure, and is the most commonly used procedure
for animals. Indeed, the initial impetus for programmed instruction in humans was generated by
extensions from this procedure. Here, the experimenter will start with the animal’s initial
repertoire, observing him carefully. This may be standing in a corner. When any response is
made that is closer to the terminal repertoire of pecking the key than the ones currently being
exhibited, the experimenter presses a hand switch, which presents the conditioned reinforcers.
The next response requirement is even more in the desired direction, and successive
approximations to the key are reinforced. The entire process, in the hands of a skilled
investigator, may take no more than five minutes with a magazine-trained but otherwise naive
bird.

Here, the experimenter follows the bird very carefully, and like all good teachers, puts
himself under the bird’s control, adjusting the requirements in accord with the bird’s changing
behavior. The experimenter must always be on the alert, and the procedure is called
“hand-shaping.” Photocells have been used to replace the human eye. When a given beam is
broken, reinforcement is obtained, and the beam series is programmed. Ultrasonic devices and
proximity detectors could be used in a similar manner.

Another programming suggestion has been programming the equipment, rather than the
organism, through direct shaping. This would involve a system of constraints that are removed
in a programmed manner. The pigeon might be in a small enclosure within the apparatus, which
gradually expands in the direction of the key. While this may not be entirely practical with
pigeons, the procedure is often used in research with humans, where a mask may cover the
material, gradually being removed in the appropriate direction, to establish movement in that
direction.

c. Baiting. Here, the experimenter leaves a trail of grain leading to the key, with grain on
the key itself. Anyone who has observed ants knows how effective this is. The details that
might make this a generally effective procedure with pigeons have not been worked out.

d. Capitalization on existing reflexes. When a momentary shock is presented to a rat, the

reflex is a strong startle response. The animal may stiffen, jump, and then return to its normal
position. If a bar is nearby, he may hit the bar on the return, thereby producing the conditioned
reinforcers and food. The procedure is especially useful to establish avoidance or escape
behavior under discriminative control. Reflexes will vary with the species.

e. Stimulus change. If a white light goes on over the food as one of the conditioned
reinforcers during magazine training, and the key is then illuminated white once such training
has been established, the probability that it will be pecked is increased. With rats a retractable
lever has been used for the same successful effect. Here, once magazine training has been
established, a lever is suddenly protruded into the box and left there. While it may be argued that
having both the magazine and key appear white links them into the same stimulus class and
facilitates substitution, it is difficult to make this argument for the lever behavior. Its rapid
establishment has been explained in terms of exploratory behavior or response class. The latter
term refers to the fact that pressing levers is in the same response class as other manipulatory
behaviors that characterize rats.

f. Instructional control. Children and adults also have a long history of results obtained
when they press or otherwise manipulate objects. Stated otherwise, buttons are to press. The
button instructs us to press, as a wet paint sign does to touch, or, a knothole to look through. If a
child is trained to go for candy in a dispenser, and a previously retracted button now pops out, he
may immediately press it.

Instructional control may also take the form of verbal instructions, both spoken and
written. Imitation and modeling may also be used to establish that first response. In establishing
imitative behavior with recalcitrant children, one investigator put his hand on his head, and
literally had to force the child’s hand, whereupon a candy was given to the child. Getting the
child to imitate became progressively easier, and the requirements were then increased so that
speech was established. There have been attempts similarly to force an animal’s first response,
but the emotional effects produced have usually terminated the procedure. The children so
treated have also exhibited emotional behaviors, but these have diminished over the course of
training, and have been replaced by smiles and laughter as the child’s progress accelerates.
(Back to Contents)

4. Response topography: The choice of the response in the discrimination experiment is

also critical. If the discriminative stimuli on the key are visual, it is difficult for the pigeon not to
attend to them as he responds on the key. The standard rat response is a lever press. He need not
look at a display as he responds. It has been difficult to get as precise visual control over rat
behavior as it has been over pigeon behavior. To what extent these differences are species
differences, and to what extent they are governed by the apparatus remains to be seen.

Accordingly, apparatus notes form a critical part of a journal devoted to discrimination

training. A response device has been designed for a rat that requires him to nudge it with his
nose; the aim is to facilitate visual discrimination. The manipulandum for a child may be the
display apparatus itself. Instead of requiring the child to choose among five buttons that
correspond to five entries in a display, the display itself may be five small translucent panels,
each of which may be pressed, or a touch-sensitive computer screen. A compromise may
involve having a large picture on an immovable panel, with choice panels directly underneath.
The changes in response apparatus may be programmed.

The “talking typewriter” developed by Moore was an electric typewriter that capitalized
on all of these possibilities. Hitting the correct key activated the machine and produced an
electrical whirr and jump. The reinforcement, of course, was making this adult equipment go.
Once the child was “hooked,” the differential reinforcement of go, no-go was systematically
related to a program. In some cases, the correct key was the one that was illuminated. Here, the
panel procedure was employed. In other cases, the, correct key was the letter corresponding to
one presented on a screen above, or on the paper. Or it, might be the letter that was sounded by a
tape recorder. Or it might be part of the name of an unnamed object pictured on the screen (a
cow), or part of a word in the child’s own story. In these cases, we move to complex linguistic
discriminative training, by way of discriminative linkage, from magazine training with a child,
capitalizing upon his fascination for adult equipment and making things go.

In the case of the pigeon, we have to build in much of the history, as we do for the child
with retardation. The precise details of the acquisition process are of importance to other
humans when we wish to establish entirely new repertoires, or where a terminal deficit requires
trouble-shooting to ascertain where, along the line, the break occurred.

The sections that follow classify the basic operant procedures for establishing and
D ∆
maintaining discrimination into four classes. These are the standard S -S procedure, the oddity
procedure, the match to sample procedure, and the adjusting procedures. A fifth section, dealing
with errorless programming, cuts across all four procedures.

In each of these sections we are assuming that the prerequisites to discrimination training,
that have been the subject of this section, have been met. The organism is at the discrimination
manipulandum, and her response is strong. It is up to us to channel it further, and to let her know
clearly what is expected of her henceforth. We not only expect a new repertoire of her, but we
also place the demand on ourselves that we be competent in getting her there.
(Back to Contents)
 STANDARD SD- S∆ PROCEDURE
D ∆
The standard S - S procedure is defined by presenting two stimulus alternatives. One is
D ∆
S and the other is S . The alternatives may be presented simultaneously or successively. They
may be colors, forms, etc.; they may be complex presentations; they may be words:

Stated formally, the standard SD-S∆ procedure involves discrimination between two
members of two stimulus classes. Discrimination is defined by the fact that in the presence of a
member from one class, one behavior occurs, but in the presence of the other, it does not. Of all
trees in the garden you may eat, but of the fruit of the tree of knowledge (of good and evil), you
may not. The story of the Garden of Eden, like that of Pandora’s box, involved the establishment
of discrimination through instructional control, whose effectiveness was not maintained. Once
the fruit was eaten, innocence was lost. This loss consisted in the establishment of the new
stimulus classes of good and evil, along with corresponding behaviors. The rules for inclusion in
these classes have been a problem since, with different social groups ordering the events
differently. In addition, behaviors which the moral dispensers of reinforcement define as S∆
often produce reinforcers from other dispensers which are better potentiated than those in the
societal SD class. The reinforcers are often more immediate, more direct, more frequent, may
have lesser response requirements, and so on.

In the laboratory, stimulus control is one of the classical ways to assess discrimination in
animals. If we present a red key and a green key, with differential reinforcement, the color
associated with reinforcement will rapidly acquire discriminative control over a pigeon’s
behavior, but not over a dog’s, when other bases for discrimination, such as brightness, are ruled
out. Accordingly, we state that dogs are colorblind. Outside the laboratory, the procedure has
been used to assess intelligence and the ability to conceptualize in certain ways, as suggested by
the illustrations at the beginning of this section.
(Back to Contents)
D ∆
Extinction and consequences: S and S are defined by the presence or absence of
consequences attached to responding in the presence of each, with such reinforcers as food,
D
money, points, grades, approval being attached to S responding, and no such consequences, or
∆ D ∆
extinction, being attached to S responding. In a variant of the S -S procedure, consequences
∆
are attached to S responses, or errors. These may include time-out, where the equipment shuts
off for a period and is inoperative; deduction of points for errors or even a head-on collision with
some obstacle. One laboratory device used before operant procedures were developed is known
as a jumping stand. here, a rat stood on a small diving board. He was required to jump at a wall
containing two doors. One might have a square on it and the other a triangle. If he jumped to
the correct door, it swung open and he landed safely on a platform behind. If he jumped into the
incorrect door, which was locked, he would literally collide with it head on. To get him to
respond again, he often had to be blasted from the jumping stand with air.

Current operant procedures do not require such expenditure of energy to indicate choice.
A lever press, a key, a button, a mark on an examination sheet, a word, may suffice.
Nevertheless, the consequences of error must be considered in any experiment designed to
D ∆
permit errors. And the classical S - S procedures typically involve numerous inappropriate
choices, or errors. Although the (topographically) same triangle and circle will be presented
D ∆
when the S –S consequences are food-no food, as opposed to food-collision, the course of
acquisition of discrimination in the two cases is likely to be different, and different procedures
will be required to maintain the behaviors necessary for learning. The Theory of Signal
Detection tells us that the terminal discriminative behaviors will also be different.
∆
Technically, S associated with extinction; the association of a square with a collision
D D
defines the square as an S for punishment. In this case, the discrimination is between S
r ∆ a
(triangle) –RÆS , and S (square) RÆS . Our present discussion will be concerned with the
D ∆ D ∆
simpler procedure of S -S discrimination. The S –S procedures will be considered later as an
extension.
(Back to Contents)
D ∆
Establishment of S -S discrimination: In the animal laboratory, instructional control
over the appropriate dimension (that is, the dimension that is the subject of our investigation)
cannot often be established in advance, nor can the animal be told that if he responds
appropriately, he will get food. Accordingly, such controls must be built into the experimental
procedure.
D ∆
We shall first consider the case of successive discrimination, where periods of S and S
∆
follow each other. The response key is red during SD and green during S . When the
discrimination is well established, the sequences can be specified as follows:
 ∆
When S appears, the pigeon waits for the key to turn red before pecking. The
consummatory behaviors connected with eating have been omitted for brevity. It will be noted
r
that this is a chain. One response depends upon another. The S that maintains prior behavior,
D D r
also serves as an S for the behavior of the next link. The S of the red light actually is an S for
the R of going back to the key. It might be said that when the food magazine is withdrawn, the
pigeon looks forward to a red light on the key. He may hope he does not get a green one.
D ∆ D ∆
Every S in this chain implies an S . The S of the clicks and lights implies an S of no
D ∆
clicks nor lights. The S of the magazine withdrawal implies an S of magazine presence (the
∆
S is for leaving the magazine and returning to the key). These discriminations were established
during the magazine training and discriminative linkage discussed in the preceding section. Our
concern here is with the payload or terminal discrimination.

We shall assume that during the prior training, the key had always been red. Behavior
during this all-red phase of discriminative linkage may be diagramed as follows:

(To simplify, we have omitted the response of returning to the key when the magazine is withdrawn.)

D ∆
Once this stimulus control is established (S -- R, S -- 0), the next step in the standard
procedures is introduced.
 The next step constitutes occasionally turning the red light off in the key, and making it
green:

D ∆
Where form discrimination is involved, S may include a triangle, S a circle. And so
on.

With a verbal human subject, we can, of course, instruct the subject to respond when red
and not when green. We can use imitation, modeling, or other procedures. Each of these assume
a long history of prior training upon which we can capitalize. If the history is not there, we may
be in for some difficulties. In the animal cases, we provide the history. In the
Herrnstein-Loveland experiment, where pigeons abstracted the human presence, the
experimenters reported that they thought the history of such abstraction had been present prior to
the experiment. The history the experimenters established was one of getting the pigeons to use
the experimental apparatus in accord with that past history. Stated otherwise, the behaviors may
be in the repertoire, but other variables are negating their occurrence. When this occurs, a child
may be called negative, or it may be assumed that he lacks the repertoire.

The foregoing are the basic procedures for establishing discrimination using the classical
D ∆ ∆
S -S method. The organism will make numerous S responses, or errors, and will be subjected
to a considerable number of extinction trials. We may use a variety of indices of discrimination.
In the classical operant procedure, response rate is the dependent variable. A rate measure of
 ∆ D
discrimination is the S /S ratio, that compares rates of responding under the two conditions.
Perfect discrimination, of course, will be indicated by a ratio of .00. If rate is discarded as a
measure, and only one response is allowed for each presentation, the index of discrimination may
involve the length of a run of correct choices. For example, it may be assumed that
discrimination has been established when the subject makes 7 correct responses in a row, since
the chance probability of such a run is less than 0.0l. Or, as in a true-false series, the proportion
of correct responses as compared to total presentations, or responses, may also be used.

The procedures described hold as well with people. When we substitute instructions for
some of the more detailed nonverbal procedures, the instructions, to be effective, should require
behavior, which will then be reinforced. The sequence may be treated as a chain, with the link
that is closest to reinforcement being established first. This link may then be made contingent on
another behavioral link, and so on progressively (the number and topography of the links may
vary with the apparatus). When the behaviors are run off, the behavior established last will
occur first. In the present case, we are interested in the discriminative behavior. It is maintained
by an extended chain following it. This behavior may be so well established that we do not even
think of the maintaining chain. The good photographic interpreter works to detect missile
emplacements, whose discovery reinforces his vigilance responses. If he finds none continually,
his vigilance may extinguish. But we state that he is task-oriented and governed by doing a good
job, as opposed to the double agent whose espionage is at the disposal of the highest bidder. The
photographic interpreter’s interpretations are part of a chain including superior officers’ orders,
bombardment and destruction of enemy installations, support of one’s own country, and so on.
If he does not support his country, or is opposed to destruction of enemy installations, the
discriminative behavior may not be precise. In most human cases, we take the chain for granted,
and concentrate on the task ahead, but every so often, the interpreter may be chewed out by a
superior in terms referring to the chain, e.g., “Are you trying to lose us the war?” The appeal
may extend to the loved ones at home. Needless to say, such verbal resort to implied chains may
be no substitute for the establishment of an effective maintaining chain.

On occasion, a subject may be shown a sequence from beginning to end: “First you do
this, then this, then this . . . and then you collect your pay.” The conditions under which
such forward sequencing may be substituted for the backward sequencing of chains, or
may be preferable to them, have not been systematically investigated.

Curricula, the forward sequences, and chains, the backward sequences, will be discussed
separately in a later chapter.
(Back to Contents)

Simultaneous presentation: The successive procedure can be considered as one in

which we train the organism when to respond. In the pigeon case, the correct time was
when red was presented, and not when green was presented. For such a procedure, one
key can be used. The child learns to discriminate his mother’s frown from her smile. An
alternative procedure to assay discrimination is to train the organism where to respond.
D
Here we would present two keys simultaneously, one red and one green, with red S , and
∆ D ∆
green S . Discrimination would be defined, as before, by S -- R, S -- 0.

With regard to the chain, magazine training is established first, as before. We now move
backward through the chain to establish discriminative linkage. Since we have two keys,
and the pigeon cannot peck both simultaneously, we have several options open to us.

One procedure involves (1) establishing a response to one key only. This key has always
D D
been red, and is S . The other key has always been dark. Once S controls behavior, we
(2) switch the red to the other key. A response to the same key as before will not produce
reinforcement. When the red key again controls behavior, we switch its position again.
We do so until only the red key controls behavior. We now (3) illuminate the other key
∆
with green, which is S . When this key controls no behavior, and the red key does, (4)
the positions of the red and green are reversed. We wait until control is evident, and may
reverse again. Eventually red-green discrimination will be established. Key position has
been eliminated as a possible basis for discrimination. It will be observed that this
procedure requires many more steps in the chain than the successive (single-key)
procedure. It will also be noted that a change is introduced only when the disruptive
effects of the previous change have been replaced by a steady state of stimulus control.
Errorless programming, or fading, will be considered later. It follows this rationale
except that it is designed to produce no disruption when a change is made. Stated
 otherwise, there is a steady state of stimulus control as the sequence of differing stimuli
D ∆
progresses. The classical S -S procedure is partly programmed in that it waits for a
steady state before introducing change. The completely unprogrammed procedure is the
single-step, where all the changes, that is, the terminal repertoires,, are required at once.

Other options may be used. Both red and green may be presented at the onset, with
shaping toward only one. This is analogous to Step 3 of the preceding procedure. One
D
such control is established, the position of the colors may be reversed, with red the S , in
a repetition of Step 4, that involves continual switching until discrimination is
established. Yet another option is to present both red and green at the onset and establish
∆
responding to both. Once such control is established, green is made S , in accord with
Step 3, followed by Step 4. This last procedure may be combined with baiting, with the
D
bait attached only to S . The reader is invited to design other procedures. There are
many royal roads leading to simultaneous discrimination learning.
(Back to Contents)

ODDITY PROCEDURE
The oddity procedure is defined by presenting several stimulus alternatives. One
D ∆
alternative is S , and the remaining alternatives are S . The alternatives may be presented
simultaneously or successively. They may be colors, forms, etc.; they may be complex
presentations; they may be words. The display in an oddity problem is an extension of the
D ∆
standard S -S procedure. Instead of two alternatives, there are three or more. The addition of
∆ D ∆
at least one other S to the standard S -S procedure produces behavioral differences
disproportionate to the change.
 Stated formally, the oddity procedure involves discrimination between several members
drawn from two stimulus classes. One of the stimulus classes is represented by only one
member, while the other class is represented by more than one. Discrimination is defined by the
fact that behavior occurs in the presence of the single-membered class, but not in the presence of
the members of the other class. Stated otherwise, in the array presented, behavior is controlled
by the odd stimulus, or the one that is different. The procedure is of interest since the
establishment of instructional control by oddity can cut across a variety of dimensions, that is,
one can “pick the one that’s different, regardless of how it is different.” If such instructional
control is well-established, we may present all kinds of different dimensions without disrupting
behavior. A series may include oddities based on form, color, size, and so on, as in the familiar
intelligence test items. The procedure may also be used to assay whether or not behavior can be
controlled along a specified dimension. A colorblind child, for example, who has been
performing appropriately when the oddities were size and form, may not select the green circle
when it is presented with three rose ones.

Instructions to pick the one that stands out, the one that is correct, etc., utilize this
discrimination procedure, as does a blinking red light over an open manhole in an otherwise
motionless and safe countryside.

The stimulus class that is excluded need not be represented by topographically identical
elements, but by elements which vary and are linked only by their common membership in the
class, as some of the illustrations that opened this section indicate. We can have a series such as
D ∆
potato, radish, onion, lemon, celery. Control by S -S differences involved in selection of the
∆
odd one implies control both by the commonality in the S class and the distinguishing
D
difference of the S element. The flexibility of the oddity procedure and its relation to the
logical operations of definition by inclusion-exclusion make the procedure useful for concept
formation research and for intelligence tests. The procedure might be used to study the
conditions necessary to establish and maintain such logical behaviors.
∆
Where there are only three stimuli in an array, both the commonality in the S stimuli and
D
their difference from S are logically co-defined by the same third element. For example, it is
D ∆
difficult to make S -S assignments to the pair a, B. However, the addition of one element will
provide the separation in the oddity procedure. If we add R, we have a, B, R, and our groupings
D
are a and BR, with a the S . If we add r, we have a, B, r, and our groupings are ar and B, with B
D ∆ D
the S . The third element establishes both the S commonality and the S difference.
∆
If we have more than two S elements, the extra elements are theoretically redundant.
∆ D
However, they may help define the S commonality, and thereby the S difference. The number
of elements needed to define the commonality will vary with the degree of instructional control.
The oddity in 2, 3, 4 may be 3 (respond to the noneven number), or 4 (respond to the square).
Adding a 5 makes the correct choice more likely, and adding 11 and 13 and 17 and 19, make the
selection of the nonprime 4 all the more likely. As a matter of fact, we might be interested in the
number of numbers we had to give you before you caught on. Just what it is we would be
measuring, we do not know. It may be our ineptness in setting up the series, or the number of
different dimensional differences that might have provided control, or your prime difficulties,
and so on. Needless to say, these alternate explanations may also be applied to intelligence tests,
where the score is often ascribed to the individual’s strengths and weaknesses. Prior training is
critical in all such behavioral tasks, where the investigator requires the abstraction of the
contingency rules. Even if the contingency rules are already in the subject’s repertoire, people
may differ in terms of how rapidly they pick the rule that is appropriate. In all events, to infer a
generalized ability or aptitude for abstraction on the basis of performance on oddity tasks, and to
compare subjects thereby, would, without appropriate control of the other variables, seem to be
questionable.
D
In the laboratory, control of responding by S is used to assess discrimination in animals
and humans, just as it is used in the schoolroom to assess intelligence. Either dimensional or
instructional (abstractive) control maybe assessed by the oddity procedure. The present section
will be concerned with specification of the procedures whereby such control is established and
maintained. The animal laboratory will serve as a model here, and will be considered in detail
for the same reason it was before: the fine details tell us what is necessary and what may have
been overlooked by broader procedures that incorporate the details in a manner that makes them
not readily available for inspection. The details may be relevant for establishment of concept
formation and abstractive behaviors in children, or in any population who for some reason has
not as yet come under such control in certain areas.

Oddity discrimination may be of two types: simultaneous and successive. The

simultaneous cases are more typical, and were presented in the opening illustration. An example
of the successive procedure would be having us say to you: beep, beep, beep, buzz. Since it
may be argued that even simultaneous visual presentations involve successive scanning, it should
be noted that simultaneous and successive are defined procedurally, namely, by what the
experimenter does, not by what the observer does. We shall consider the simultaneous visual
case first.
(Back to Contents)

∆
Establishment of oddity discrimination: Numerous S responses, or errors, are made
D
during establishment of oddity discrimination. The S consequences may be as varied as they
D ∆ ∆
are in standard S -S research, and for S , extinction is typically used.
 With verbal human subjects, verbal instructional control may be established rapidly
through simply telling the subject to select the odd one, or the one that differs. As was
mentioned earlier, this instruction is a general one that is applicable to practically any dimension.
However, the abstraction of classifying by difference-similarity underlies this instruction, and
such an abstraction, along the dimensions appropriate to reinforcement, may not be in the
organism’s functional repertoire. Or even if it is, it may not be controlling the required
behaviors. In such cases, verbal instructional control may not suffice. Of course, in the animal
laboratory it seldom does, and we must build such control into the experiment.

Let us assume we wish to establish oddity discrimination along a color dimension with
pigeons. The apparatus has three keys on the wall, arranged in a row. The sequence of
behaviors is for the pigeon to wait at the keys, peck the one whose color differs from the other
two, get his food from the magazine, and return to the colored keys. The terminal behavior we
want can be diagramed as follows:

This defines oddity control for us. If the odd-same dimension is color, then red-green
discrimination is defined by:
 Prior conditions: We shall assume that the same prior conditions discussed previously
have been met. These include, having a potent reinforcer, apparatus that is appropriate to our
purpose, appropriate constant stimuli, magazine training, and the like.
(Back to Contents)

The chain: The discriminative operant, once established, is part of a chain. The terminal
chain we shall move towards is the following:

D ∆
This chain is identical to the chain in the standard S -S procedure, except that an
∆
additional S has been added to that basic building block. The same general statements apply
here: the behavior, once established, runs itself off from left to right. Its order of establishment
D ∆
reads from right to left. The procedures are straightforward copies of the standard S -S
procedure, until we come to the R which produces the conditioned reinforcer. This is the
discriminative response which requires linkage to the rest of the chain before it can be brought
under the stimulus control of the oddity problem.
(Back to Contents)

Discriminative linkage: The reader will recall that when only one key was used in
D ∆
successive S - S discrimination, the shaping procedure was rather simple. When a second key
was introduced in simultaneous discrimination, there arose a whole variety of experimenter
options. Introducing a third and fourth key increases these options all the more, and the training
procedures are not standardized. This is an area in which systematic research is needed to see
which training procedure is most effective under what conditions; the conditions may include the
D ∆
dimensions of S -S similarity-difference, the history of the organism or its ancestors, the
number of alternatives, and the like.

One procedure is to have separate training and discrimination chambers. The training
D ∆
chamber has only one key to which the pigeon is shaped as in standard S -S discrimination, the
D
S being an illuminated key.

Once this key controls pecking, the pigeon is transferred to an oddity chamber, that
contains several keys. He is immediately exposed to an oddity problem; that is, one key differs
from the rest, and pecking only to that one gets reinforced. The type of oddity problem will
govern the training key. Where the oddity problem is color, the training key is never one of the
colors to be used in the discrimination experiments. The reason is clear: if the oddity situation
∆
is Blue-RED-Blue, and the training key was Blue, the pigeon will select S in the oddity
D
problem. If he has been trained on Red, he will select S , which is desirable. However, what the
experimenter gains here in transfer time will be lost when he switches the oddity problem to
∆
Red-Red-BLUE. Here the animal will respond to S . Accordingly, the training stimulus should
be one equally removed from all possibilities -- a white key, in this case.

Other procedures are possible as well. These would involve training in the discrimination
chamber itself. Here, the experimenter might shape to peck at an illuminated white key, with the
others dark. Once it controls behavior, the illumination is shifted to another key, with the others
dark. When illumination alone controls behavior, the oddity stimuli are then introduced. We
invite the readers to propose other procedures, and guess what difficulties they will encounter.
Highly sophisticated procedures that provide some systematization and the opportunity to ask
interesting questions will be presented in the section on programming. The problem of
establishment of oddity, as we mentioned earlier, is a critical one since it may provide us with
information on the acquisition of the exclusion-inclusion method of classification, an important
method in logical analysis. It is unfortunate that the variables affecting oddity acquisition have
not been systematically explored (except, perhaps, in programmed instruction, direct instruction,
and other areas of instructional design).
 Instructions for a verbal human subject may simplify the training procedure. We may
start out telling the subject to press a key or panel; when he does so, the conditioned reinforcer is
presented, and the reinforcer is available. We may use imitative control, or modeling for the
same effect. In some cases, we may put the child’s hand on the key or panel. Once this behavior
is established, we may then introduce an array of panels, and instruct her to press the one that is
different, or which does not belong. Some examples may be given, as in intelligence tests.
Occasionally, small keys or buttons are used, whose locations correspond to some pictures or
words on a screen. There may be a row of pictures with a row of buttons underneath. Here, the
subject must first also be trained in the concept of correspondence of keys to pictures, and then,
to press the key corresponding to the picture that is different, or that does not belong. Adding
this requirement of correspondence, however, increases difficulties considerably, since it
requires the addition of extra links in the chain. Further, the organism can manipulate the keys
or buttons without necessarily observing the target. It will be recalled that this is one of the
difficulties in the rat apparatus, and has raised questions concerning the results obtained.
Children may dawdle and look away, and superstitious behavior may be reinforced. If there are
four keys, and the child behaves in this chance manner, she will be on a VR schedule that
becomes difficult to extinguish. Accordingly, in most oddity research, the control is simplified
by combining manipulandum and presentation target.
(Back to Contents)

Terminal discrimination training: Once the subject is responding to the manipulandum

that is appropriate for an oddity task, we can turn our attention to training in the oddity
abstraction. This abstraction is defined as control by the contingency rule of responding to that
event which differs from other events in that it has no other members in its stimulus class, or to
the event which has no match, or as the subject might state the rule, to the odd event, regardless
of what makes it odd. The abstraction may be established in two ways: continual change and
post-control change.

Continual change. Here, a particular oddity situation is presented, say Blue, Blue, RED.
The subject responds, and the next presentation may be Green, YELLOW, Green. The following
presentation is Red, Red, VIOLET. All the combinations and permutations to be used are
presented in a sequence. There will be numerous errors in the acquisition of the abstraction of
responding to the presentation that has no match, and the series required for such acquisition may
consist of hundreds of combinations. On a school-teaching level, this is the case of hundreds of
different examples being thrown at the student; after a response by a student, a new example is
given. The teacher might feel that the student has caught on when he gets, say, five examples in
a row correct.

Post-control change. Here, a particular oddity problem is presented, say Blue, Blue,
RED. The subject responds, and the next presentation is the same problem. It is continually
presented until the third position, RED, controls behavior. The position of the red stimulus will
then be switched until it reexerts control. Another position may then be utilized. Positions may
then be switched with each presentation. When red alone exerts control, the next presentation
may be Green, YELLOW, Green. When YELLOW alone similarly acquires control, another
combination is presented, and so on. On a school-teaching level, a new example is not given
until the student has mastered the present one.

The continual change and post-control change procedures may be combined by

presenting Blue, Blue, RED on the first trial; RED, Blue, Blue on the second; Blue, RED, Blue
on the third; and so on. The procedure is continual change in that the positions are changing
before there is mastery, but is post-control change in that new color combinations will be
presented only after the RED-Blue oddity combination is mastered.

The oddities presented above are all color oddities. For these there may be substituted
form oddities, concept oddities, or combinations of all oddities. For example, oddity
discrimination along a form dimension was used in the 1960’s for chimpanzees in space. Each
correct choice presented the next trial almost immediately. Food was provided on an FR
schedule. So rapid was the behavior that it was almost impossible to follow on a motion picture
screen. The stimulus control demonstrated, in this case, that the chimpanzee could be alert
despite radical changes in gravitational force.

The fading procedure, as we shall see shortly, also combines both continual change and
post-control change procedures. A change is made only when control has been established, but
the sequence is such that each presentation provides control, and each presentation is a change
from the preceding. Fading represents the application of errorless programming to this area.
 Successive presentation: The procedures discussed thus far involve selection of oddity
from stimuli which are simultaneously presented, for example, 0 X 0 0. Such stimuli may also
be presented singly, and in succession. Here the subject must indicate when the odd one was
presented. Successive presentation is especially useful in auditory discrimination; it is far more
cumbersome to discriminate among simultaneous sounds than sights. Hence the successive
procedure.

In the successive procedure, it is often impossible to respond when the oddity is

presented. If the oddity is on the first or second presentation, this will not become evident until
the third, at which point the opportunity to respond will have gone by, and first and second
presentations cannot therefore be used as occasions for responding. The problem is evaded by
having the entire series go by, with a verbal response required at the end of the series, such as
First, Second, Third, . . . or a row of buttons. This procedure introduces variables such as
memory, the effects of the sequence on judgment and its changes, and the problem of training the
organism in the correspondence of responses to stimuli, an issue we have already discussed.
With animals, the manipulanda are often withdrawn during the series. The development of
procedures that seek to overcome the various problems raised by this method, which is the major
one used, has a long history in psychophysics, and reliable results have been obtained when
appropriate controls have been instituted.

An alternative procedure combines successive with simultaneous discrimination. Here a

single visual stimulus is presented, 0. Two are then presented simultaneously, X, 0. One of
them matches the preceding single stimulus, and the other does not. The responding to the odd
stimulus is reinforced.

It is evident that the identical procedure can be used to have the subject select the
matching stimulus, rather than the odd one. This procedure is called Match to Sample, and is
considered one of the basic procedures defining perception. It is the subject of the next section.

Conditional discrimination may be established using both Oddity and Match to Sample.
In the presence of a red light, the contingency rule for reinforcement may be to select the odd,
and in the presence of a green light, the contingency rule may be to select the match. The
oddity-match dimensions may vary with each presentation. This procedure has established very
precise stimulus control by red and green. Here, red and green are instructions on which
instruction to employ. Stated otherwise, they are superordinate instructions.

Although both Oddity and Match to Sample seem to involve the same judgmental
processes, with the subject seemingly distinguishing both similarity and difference, and with the
only difference seeming to be the response, in actuality the two procedures produce different
behavioral results, as we shall see. Why this should be is a matter not of logical controversy, but
of experimental analysis of differences in the discriminative behaviors involved, their
establishment, and their maintenance.

(Back to Contents)

MATCH TO SAMPLE PROCEDURE

The match to sample procedure is defined by presenting a sample and then presenting
two other stimuli. The stimulus which matches the sample is S~, and the one which does not, or
∆
is odd, is S .

Within this basic framework, variations are possible. Only two keys may be used, with
D ∆
the sample appearing in either one. The sample goes out, and S -S appear in both keys. Three
D ∆
keys may be used, with the sample in the center. The sample may stay on when the S -S are
D ∆
presented. It may be withdrawn when S -S are presented. There may be a delay between the
D ∆ D ∆
withdrawal of the sample and the presentation of S -S . In all of the foregoing, S -S are
presented simultaneously; they may also be presented successively. The stimuli may be colors,
forms, etc.; they may be complex presentations; they may be words. The sample is in the center
in all of the following:

The matches presented in the foregoing illustration involve only two choices. Needless
to say, more alternatives may be made available, as in a multiple choice examination.

Stated formally, the match to sample procedure involves discrimination between two or
more members of two stimulus classes. One of the members is in the same stimulus class as
another element (the sample), and the other(s) is not. Discrimination is defined by the fact that
behavior occurs in the presence of the member of the stimulus class defined by the sample, but
D
not in the presence of the other(s). The sample presentation acts as an instructional S i, whose
D ∆
control over behavior is assessed by discrimination between the S d and S d presented. The
D
training procedure and experimental situation provide a superordinate S i to “match to sample,”
D
and the sample S i instructs the observer what the class is to be. The superordinate rule may be
given in advance through verbal instructions, or such instructional control may emerge as an
abstraction through continual presentation. In all events, we have two rules for generating
reinforced behavior.

It is this instructional complexity which makes the match to sample procedure a very
useful and general one. The match to sample procedure may be used in almost any form of
discrimination and also across sensory modalities, as when the teacher says “Horse,” and the
child picks the picture of the horse from a set also containing a cow, dog, and house. Or the
picture is presented, and the child selects the word horse from among others. The word Lee can
be presented on tape, and the reader picks the written Rhee or Lee, and may thereby be trained to
♠
discriminate these sounds, that are in the same stimulus class in many Asian languages.

(We could also use oddity, by having the teacher say “horse,” and requiring the child to
pick a non-horse. This has its limitations for teaching.)

The match to sample may also be used in sequential presentation, where the sample, a
tone, is presented, and the listener must press a button when a sound matching it is presented.
This has been extended as well to visual presentation of geometric figures.
(Back to Contents)
D ∆
Sample-match interval: Where S -S are presented simultaneously, they may be
presented (a) while the sample is still present (concurrent match), (b) as the sample is withdrawn
(zero delay), or (c) some time after the sample has been withdrawn (nonzero delay).

Concurrent match: In the concurrent match, the sequence of presentations leading to

discrimination, is the following:
 The correct choice in match to sample is L. The correct choice in oddity would be R. As
we have indicated, these tasks seem to be logically equivalent: select the one that matches
(which requires isolation of the non-match from match), and select the one that does not match
(which requires isolation of the match from non-match). It is what one reports and what one
does not report that seems transposed. The principle seems the same.

However, much as it is tempting to draw principles, the two different procedures produce
different results. The illustration below depicts curves from two different groups of birds
learning to acquire the same color discriminations. For some birds, oddity was used, and for
others, match to
sample. The group
curves are presented
for convenience only.
Data with individual
animals provide the
same results. They
indicate that oddity
discrimination is
acquired very
gradually, and match to sample much more rapidly, for the same animals. The match to sample
curve has a sharp inflection. Such curves are often called “insight” curves, and may reflect a
switch in instructional control from hitherto inappropriate instructions to the appropriate one,
under conditions where response rate under the appropriate control will be high. In the match to
sample, at some point, the instructional control of matching takes over, the stimuli are clearly
separable at that point, and it is possible that the sharp inflection represents similar “insight” by
the birds. Why such shifting does not occur during oddity training cannot be answered by these
data. Nevertheless, different results are produced when the two procedures are used.
(Back to Contents)
 Zero delay: In zero delay, the sequence of presentations leading to discrimination is the
following:

The correct choice is, of course, L. We could also use two keys:

The correct choices are, respectively, L and R. There are two other ways to present these.
The reader is invited to fill in the “etc.”

It should be noted that we have not changed the alternatives in the matching presentation.
D ∆
Merely by changing the sample, we have reversed the S -S relation. The reader will recall the
airplane presentations discussed earlier. These were unchanged, but behavior to them was
changed by instructions. This reiterates the instructional nature of the control exerted by the
sample.
 Suppose, instead of making L and R correct, as in the foregoing, we make R and L
correct. We do this simply by altering our reinforcement contingency. We have now turned the
situation into an oddity problem of zero delay. The two situations appear identical.

The transposable relation between match to sample and oddity disappears when we use
more than three keys, our minimum for oddity. In the following case, the sample is the top key:

In Presentation 1, M is correct, since the triangle with the vertical axis which was in the
sample is matched thereby; in 2, L is correct, since the match is axis right; in 3, M is correct
since the match is axis left.

It will be noted again that we have not changed the alternatives of the matching
D
presentation. Merely by changing the sample, we have changed which one is S , and which two
∆
are S . This flexibility is not as practical in the oddity situation. To see why this is so, we shall
make presentation of sample and match concurrent:
 On the basis of oddity, either L or R is odd. However, only M
can be the match. The more alternatives we present, the more
economical the matching procedure becomes. Where there are
five stimuli (one being the sample), any of 3 will be odd, but
only one will be the match. In a case where there are limitless
choices, a pairing match will be odd, just as identical twins
stand out.

A rule for stimulus class inclusion also implies a rule

for exclusion, and vice versa. Logically, both oddity and match
imply both rules. Practically, however, in the oddity procedure, it appears as if the rule for
exclusion is explicit (pick the one that’s different) with the inclusion rule implicit, and in match
to sample, the reverse appears to be the case (pick the one that matches). Students who are
“test-wise” learn to apply these rules very rapidly. Rather than go through all five choices on a
multiple choice question, they learn to look for classes immediately and are able to zero-in on the
excluded or included choice almost as soon as it appears. Where the correct answer appears on
the fifth choice, there is no advantage to such behavior. However, in a typical test, the fifth one
is correct only 20% of the time. On other questions, the test-wise students save the time that
others spend considering all alternatives. The students also learn rapidly whether and how to
match or look for oddity, depending on the nature of the question.

Nonzero delay: In nonzero delay, the sequence of events leading to discrimination is the
following:
 The correct choice is, of course, L, and all the variations discussed with the other
procedures apply here as well.

It is apparent that this is an ideal procedure for study of memory. Memory may be
related to the types of stimuli presented, as well as to the maintaining conditions studied in
operant research.

When the delay is increased inordinately, there will be “forgetting of the sample,” that is,
a breakdown in matching. The subject is accused of having a “short memory span.” One
ingenious application of operant procedures has been the use of temporal adjustment procedures.
Here each correct match provides for an increase in the delay between the disappearance of the
sample and the presentation of the match. Each incorrect match provides for a decrease in this
interval. Eventually, the subject’s behavior adjusts the interval around some point, which has a
very stable range between the amount of delay that is too long, and the amount of delay that is
too easy. The procedure is similar to Verhave’s adjusting ratio, and has been used to study
drugs, RNA, memory, and learning. Adjusting procedures will be discussed in greater detail in
the next section.

If, using any of the procedures, the sample is presented only momentarily, appropriate
matching, as Brady has suggested, would indicate attentiveness. Since the matching procedure
typically involves two match keys, the two keys can readily be converted into “Yes-No”
responses of psychophysics, with Signal Detection Theory readily applicable in the study of
decision processes in discrimination, memory, and attention.

We shall now consider the specific procedures necessary to establish matching to sample.
The prior conditions and magazine training are identical to those discussed up to now. The chain
D ∆
is a different one, and consists of other links added to the building block of the standard S - S
procedure.
(Back to Contents)
 The chain: There will be a sequence of behaviors that relate to the discriminative
matching operant. If discrimination is well-established, the sequences can be specified as
follows:

(Back to Contents)

Discriminative linkage: Here we wish to get the pigeon to peck at the keys so that the
matching behavior can be established. Instructions for a verbal human subject are so simple that
the linkage stage is often omitted, with the matching stage introduced immediately. When we
omit linkage, we may simply tell the subject to press the panel that matches a sample or to type
the words presented on a page. Where we use the linkage stage, we may start out telling the
subject simply to press a key, or panel; this provides the conditioned reinforcers. We may use
imitative control or modeling for the same effect. In the “talking typewriter” of O.K. Moore, one
of the typewriter keys was illuminated. Striking this key had the consequence of typing the letter
that was struck. (Discriminative training followed. This involved providing reinforcement only
if the letter struck matched a letter presented). Illuminating this particular key in the typewriter
capitalized upon response to oddity, of course, and there was no reason why one could not
capitalize upon this behavior in establishing discriminative linkage for prospective match to
sample. The subject did not realize that the inclusion-exclusion rules were going to be
transposed on him, and simply proceeded in ignorance of this logical nicety.

With animals, abstractional control must be established during the experiment.

The chamber has three keys, with the center one being the only one illuminated. If the
match will be on the basis of color, the key may be white. The pigeon is shaped as in standard
D ∆
S -S discrimination. Pecking the key produces the conditioned reinforcers, and ultimately
food. Once this key controls pecking, it is turned off and a side key is illuminated. Once this
key controls pecking, it is turned off and the other side key is illuminated. Which key it is that is
illuminated is then varied systematically, and once illumination alone controls pecking, we are
ready for terminal discrimination training. An alternative is to have the center key colored in
shaping. It changes colors, and also positions. It is often convenient to have separate chambers,
one wired for discriminative linkage, and the other wired for match to sample training. One
animal can be matching while the other is learning the prerequisites. On the other hand, we may
wish to perform both functions in the same chamber.
(Back to Contents)

Terminal discrimination (matching) training: The pigeon is now under the control of
whichever position is illuminated. The procedures for establishing match to sample are varied,
as was the case with the oddity procedures.

In one procedure, the center key is now the only one illuminated. It is illuminated in
some color other than white. When the pigeon pecks this key (and he may not do so
immediately because of the stimulus change), it goes off and the two side keys are simul-
taneously illuminated, one in a matching color and the other in a different one. Pecking the
match produces the conditioned reinforcers; pecking the other does not. The pigeon has
match-to-sample thrust on him immediately, so to speak. Numerous errors are made, but as we
saw previously, when instructional control does occur, it occurs very rapidly. Rather than zero
delay, we could have started off with concurrent presentation. Here, pecking the center key
would turn on the side keys, and the pigeon would be confronted with an array of three keys.
The concurrent procedure has the advantage of presenting the match and sample simultaneously,
but the disadvantage of increasing the likelihood of perseveration at the center, which is off in
the zero delay procedure.
 Another option available to the experimenter is continual change or post-control change.
We can vary the nature of the match from one presentation to the next, independently of correct
responding, or wait until the organism has mastered on type of match before going on to the
next. These alternatives were considered in detail in the section on oddity.

The backward chaining procedure may be used to establish discrimination. Here

behavior is initially shaped to one of the side keys, the only one illuminated. When this side key
controls behavior, the other side key is illuminated instead. When this exerts control, either is
illuminated. Control is now by either illuminated side key. The center key alone is now
illuminated. When the pigeon pecks here, this peck illuminates one of the side keys. Since this
key already controls behavior, he pecks here and the food is presented. Which side key follows
the center key is switched back and forth in some unsystematic manner. The chain now in effect
may be depicted as follows:

When the sequence is established of pecking center, then side, color is now introduced,
with both keys the same color, that may vary with different presentations. The change is now
made so that pecking the center key illuminates both side keys, one in a color different from the
center. Pecking this key will have no effect. Pecking the side key that matches the center will
produce food. Pecking the odd key, then the match, may establish a superstitious chain. Such
chains are quite persistent. This is a problem that recurs whenever there is opportunity for error,
and special procedures have been developed to handle it. These procedures will be discussed in
the section on problems of stimulus control. An alternative is not to have errors. This will be
discussed in the section on errorless programming.
(Back to Contents)
 Successive presentation: In the matches we have been discussing, SD and S∆ are
simultaneously presented. They may also follow each other, one at a time. In the terminal
behavior of successive match to sample, a red sample may be presented in one panel. In an
adjacent panel, a green stimulus is presented for a brief period. It is then replaced by a yellow
one; this is replaced by a red one. Discrimination is considered established if the organism
continually presses the matching panel when SD is presented. The sample may be concurrent
with SD or S∆ as in the foregoing example. We might also use zero delay. Here the sample
would go off when a stimulus in the matching panel went on. It would go on again as that
stimulus went off, staying on for a while until the next stimulus was presented, and so on. The
reader is invited to suggest the various procedures whereby such matching can be established.

Successive presentation requires the organism to respond at an appropriate time and

place, as the field he is observing changes. It is also possible to teach an organism to change his
field to produce such a match, and these adjusting procedures are the subject of the next section.
(Back to Contents)

PROGRAMMING ERRORLESS
DISCRIMINATION: FADING

The present section will be devoted to the presentation of procedures whereby

discrimination can be established without error. In each of the foregoing four sections,
considerable error, or S∆ responding, was involved in the establishment of discrimination. The
present section will recapitulate these four procedures, and demonstrate how they may be
programmed errorlessly.

The reader will recall that each new procedure rested upon the preceding one, and
extended it somewhat. In the method basic to them all, the classical SD-S∆ procedure, S∆
responding has been so prevalent that extinction of S∆ responding has been regarded as the
“hallmark of discrimination.” The reader will recall that in the acquisition of red-green
discrimination, numerous errors were made to green. Knowing what not to do has been
considered as important as knowing what to do, and integral to its acquisition. The reader is
referred to the previous chapter on errorless programming for a discussion of the effects of
extinction and error on the behaviors of learners, investigators, theories, and systems of research
and training.

We shall now present procedures whereby discrimination may be established without the
organism making a single S∆ response. That such procedures have been developed challenges
the various statements made about the necessity of error for discrimination learning. If learning
what is correct can occur without extinction of incorrect responses, then we can sidestep the
extinction problem. Where such extinction is the “hallmark of discrimination,” it is a hallway in
the school of hard knocks -- which is not the only academic institution in which one can learn.

The errorless procedures, which are the new school, have also produced results that are at
odds with previous findings considered typical of discrimination. These will be considered in a
separate section. The procedures are also less time-consuming than the error-laden procedures.
Using errorless procedures, it has been possible to establish discriminations hitherto considered
extremely difficult or impossible. The difficulty had previously been assigned to the problem or
organism or both, and the results obtained from errorless discrimination procedures suggest that
the difficulty may be a function of the teaching method, instead.

Before discussing each of the four procedures in detail, the commonalities should be
noted. Errorless discrimination training is the application of programming principles to
discrimination. A terminal discrimination, that is, the goal, is first specified by the investigator.
She then assesses the current discriminative repertoire of the organism with respect to the
terminal repertoire. This repertoire may be related to genetic variables common to a species, or
specific to an individual (color blindness in dog and man, respectively), to past environments
(musical training), or to experimenter procedures. Operant programming does not ignore species
and individual differences; on the contrary, it requires intimate knowledge about them. Once the
experimenter has assessed the current repertoire, she now introduces a program that gradually
alters the current repertoire to the terminal one. In shaping, the sequence consists of a program
of reinforcement of successive response ensembles, the succession being dictated by increasing
presence of behavior along a criterion dimension. On the stimulus side, or discriminative
programming, the investigator reinforces responses to SD in stimulus ensembles that successively
differ, the succession being dictated by increasing approach to the terminal discrimination. To
proceed without error, each new SD-S∆ difference is close enough to the preceding one to
maintain control by SD, but slightly different, and it is these differences that provide the
direction. Either the SD or S∆ or both may be gradually changed, and the procedure is called
fading, from the analogy of fading out a color gradually. Like the glass that can be called half
empty or half full, we shall use the term for gradual decrement (fading out) or increment (fading
in) along any discriminative dimension -- color, form, concepts. On occasion, as we shall see,
the program may not move directly toward the terminal repertoire, but may employ intervening
dimensions, that are not involved in the terminal repertoire, but seem necessary for its
establishment, as when we use, then remove the scaffold that helped us build the house.

The reader will recall that in our presentation of the basic discriminative training
procedures, various steps were involved, and a new step was often added when behavior had
been brought under discriminative control in the preceding step. If a new step introduced errors,
the investigator waited until there were no more before she introduced the next one. And so on.
In the fading procedures, this logic is carried through and extended -- each step that is introduced
provides immediate control without error, so that the step following it can then be introduced
immediately. Rather than waiting until a chapter is mastered before introducing the next, in
fading we introduce a line at a time, and in such a manner that each one is automatically
mastered. As we shall see, these fading procedures may be extended to all four of the procedures
presented, and to all the various problems to which they are addressed. We can teach mastery
not only of simple discriminations, but of complex concepts and abstractions. Whether we use
fading or other procedures for discriminative change will be governed, needless to say, by our
technology, the nature and requirements of the task, and the repertoire of our learner.
(Back to Contents)

Magazine training and discriminative linkage: In each of the following applications of

fading, the prerequisites for discrimination are assumed. These include control of constant
stimuli, potentiation, magazine training, and discriminative linkage. Where discrimination
training enters into any of these (the key may be SD with respect to an S∆ wall), fading
procedures can be employed. Here standard SD-S∆ fading will be used, and the reader is referred
to the formal presentation that follows. Where the procedures involve response alteration
(shaping to peck the key, for example), errorless shaping procedures may be used. These
prerequisites underlie all four of the major procedures. Our present concern will be with the
discrimination that distinguishes the typical SD-S∆ from oddity, match to sample, and adjustment.
(Back to Contents)

ERRORLESS PROGRAMMING OF
SD-S∆ DISCRIMINATION
In this procedure, it will be recalled, discrimination is defined by the occurrence of
behavior in the presence of an element from one stimulus class (SD), and its absence in the
presence of an element from another stimulus class (S∆). The definition holds for the errorless
procedure, as well. It is the procedure for establishing such discrimination that differs.

The first systematic research in this area is by Terrace, who cites prior studies in which,
for example, discrimination of two narrowly separated grays was learned errorlessly by starting
off with one black and the other white. They were then gradually changed to almost equal grays,
and discrimination was transferred from black-white to two grays without error. The procedure
so described incorporates the major features of the errorless discrimination to be discussed. The
systematic exploration of variables relevant to it begins with the work of Terrace. This research
was concerned with isolation of relevant variables in a systematic manner. Our report of the
research will differ from the actual account in that it abstracts from the account the basic
procedures used in the various experiments, and presents them as if they were one experiment
with one aim, namely, explication of a procedure.
(Back to Contents)

Specifying the terminal discrimination: The task we shall consider is the establishment
of discrimination between a vertical and horizontal line. The terminal discrimination may be
defined as follows:
 To establish such discrimination (once the pigeon has been magazine-trained and
shaped), the typical SD -S∆ procedure confronts the pigeon with these terminal requirements
immediately, as described below:

This task typically requires thousands of trials, and has been considered difficult for
pigeons. The pigeon makes errors continually, and seems to make no headway for extended
periods. Under similar situations, children have given up, and blamed themselves (and have
considered themselves stupid), or have blamed the problem (it’s too difficult), or both (it’s too
difficult for me). The failure is often accompanied by emotional concomitants whose nature, like
the other responses, rests upon past experience.
(Back to Contents)

Assessing the relevant initial discriminative repertoire: Since we are going to

program the establishment of such control, we must first assess the initial repertoire. It is
comparatively easy to train a pigeon to peck at a colored disk, to train him to discriminate light
from dark, and red from green. A program implies progressions, and the progressions in the
initial experiments by Terrace, as we shall see, at each step capitalized upon such knowledge
about species and experimentally acquired behaviors.
(Back to Contents)
 The program: The pigeon was magazine trained and shaped to peck at a key that was
always illuminated red. A colored key was used rather than a white one because of program
requirements that will shortly be evident. The discrimination, when established, can be
described as follows:

A change is now introduced: the key flickers momentarily. The illumination of the key
is briefly interrupted. There is no pecking during this change. As a matter of fact, the pigeon
jerks away. Whether this pattern is attributable to stimulus change, or the fact that the first jerk
backwards is followed (and therefore reinforced) by the key going back to red, or an ethological
response to sudden darkness, we do not know. The change suffices to interrupt the pecking, that
resumes when SD is returned. Control of behavior is maintained by the red light, and differential
behavior results:
 The off-period of the flicker is, with each reinforcement, now gradually increased until it
is the same duration as the red-on periods, namely, 30 seconds. Control by red alone is
maintained:

Discrimination has been established between (red) light on and off, with no errors. (Had
there been errors, this would have indicated to the experimenter that he had gone too fast, and he
might backtrack).

Since SD will remain constant (red key, on 30 seconds) during the next phases, we shall
present only the changes in S∆ that were instituted next:
 The investigator introduced faint green into the off-period, dropping the time as he did so
to ensure control by SD. The green is now made increasingly brighter, until it matches the red in
brightness, and the duration is increased to 30 seconds. Red-green discrimination has been
established without error. The reader will recall that the example we chose in our previous
presentation of the standard SD-S∆ procedure was red-green discrimination. The same
discrimination can thus be established with or without error.

The next step is to superimpose the vertical and horizontal lines upon red and green,
respectively. This can be done errorlessly in a single step, so a single step is used. Durations
will be omitted below, since they will not be changed.

The colors are now faded out, leaving only a vertical white line on a dark key, and a
horizontal white line on an equally dark key, as shown below.
 Discrimination has been established between differently-tilted lines, without error. At
least six steps were required. The reader will note that this procedure is considerably more
complicated than the standard one, where the terminal discrimination was required at the outset.
The two procedures may be compared as follows:

The differences in programming effort required of the experimenter in the two situations
is obvious. The standard situation is a cinch to program. The animal is put into the box
to learn on his own, to sink or swim. The errorless program requires considerably more
ingenuity, effort, and preparation. It requires continual monitoring of the effect of every
change made by the investigator, on the behavior of the learner. On the other hand, the
experimenter who uses the standard procedure purchases his own leisure at the expense
 of the time, effort, and extinction effects of the learner, as well as of the complexity of
problems that may be taught. If the learners are children in a school room, and the
instructor is interested in teaching them by these trial and error methods, the time, effort,
and extinction of the learners may be matched by the time, effort, and extinction of the
teacher in this apparently thankless job. Where, however, trial and success procedures
are substituted, the task is not so formidable either for teacher or learners. Once such
programs are developed, they can be plugged into an automatic programming device (a
computer), leaving the experimenter or instructor free to do something else. However, it
is costly to develop and refine such programs.

An example of the programming of a human concept beyond the repertoire of pigeons is

provided by the following program, which has been used for over 7,000 people in audiences of
different sizes, including one person. The audience is told that one pair of stimuli at a time will
be drawn on the blackboard, and
every person is to select either the
first or second member of the pair
by raising his hand when it is read
off.

For Presentation 1, the

audience typically splits 50-50 on
successive show of hands, and is
then told: “B was correct.” For
Presentation 2, more pick B. Some
pick D. The audience is then told:
“B was correct.” And so on. At
Presentation 6, the letters are read
as words. It will be noted that from
1-4, B is correct. Then extra letters
are added from 5-7. Recognizable
words are added at 7 (at 6 if the
audience knows Latin). The
dimension of male-female names is introduced; B is embedded in the male names from 7-13. At
14 and 15, the discriminative control of B is faded out; SD and S∆ are no longer differentiated by
B, since this is present now in both members of each pair. At 16, there is no B, and SD and S∆
are defined by male-female names. At 19, there is reversal of the original SD-S∆ dimension, that
had been replaced by male-female names. B controls no responding here.

The program presented is not completely errorless, although with refinement it could be
made so. Rather than having A-B as indicated in 1, B alone might be presented earlier, and other
changes made. The point is that fading procedures can be used for the establishment of
discriminative control by the same abstract concepts that the standard SD-S∆ procedure can be
used for, as was indicated in the illustration that introduced that section.
(Back to Contents)

ERRORLESS PROGRAMMING OF
ODDITY DISCRIMINATION
In the oddity procedure, it will be recalled, discrimination is defined by the occurrence of
behavior in the presence of the member of that stimulus class represented by only one
element (the odd one), SD, and the absence of behavior in the presence of members of the
stimulus class represented by more than one element, S∆. This definition holds for the
errorless procedure, as well.

A series of studies by Sidman and his associates utilize fading procedures to establish
oddity discrimination between ellipses and circles. The subjects were nonverbal children
with severe retardation.
Specifying the terminal discrimination: The terminal discrimination was to be between
ellipses and circles, where either was the odd stimulus. This can be represented as
follows:

A very fine discrimination was desired.

(Back to Contents)

Assessing the relevant initial discriminative repertoire: The children with mental
retardation could readily respond differentially to dark and light, and could push buttons
when directed to do so. It was, however, extremely difficult to teach them verbally to
pick an odd stimulus in an array. Rather than attempt to develop such instructional
r
control immediately, the investigators decided to use an oddity procedure. The S was a
small piece of candy. Initially, any press against a large vertical panel was reinforced.
When this behavior was established, the panel was subdivided into nine equal panels, as
in a “tic-tac-toe” game. Pressing any of the eight outer panels was reinforced.
(Back to Contents)

Program One: One of the eight panels was now illuminated brightly; a circle was
projected on it. This illuminated panel rapidly acquired control over behavior; the
investigators were probably capitalizing on a past history of responding to illuminated
objects. The position of the circle was varied and continued to control behavior. The
discrimination may be depicted as follows:

A flat ellipse was now projected dimly on each of the remaining seven S∆ panels. The
illumination of S∆ then gradually increased until it was equal in brightness to SD, the
circle panel. The brightness difference having been faded out, the only basis for
classification and oddity discrimination was now circle versus flat ellipse, as indicated
below:

Although many of the children progressed through the program with virtually no errors,
for some children the task proved to be too difficult. Rather than placing the blame on
the child, and assigning him a failing grade on this item in an intelligence test, Sidman
went back to revise his program. He reasoned that he had tried to accomplish two things
simultaneously. He had started with both a circle and a bright panel as SD, and with
seven dark panels as S∆. As the brightness of S∆ had been increased, the flat ellipse on
the S∆ panels had also been faded in. He then revised the program to make these two
separate phases. First, he faded out brightness differences. Then he faded in ellipses.
The revised program was as follows:

The preceding program had been steps 1, 3. All

children now learned the circle-ellipse discrimination
with ease. The children who had failed to classify
and discriminate appropriately before, were now able
to make the appropriate classification and
discrimination. Apparently, they could classify
appropriately -- if taught to do so appropriately.

Sidman was now ready to determine how fine a

discrimination the children could make. From one
trial to the next, the vertical dimension of the flat
ellipses was gradually increased, making them look
more and more like circles. The child was now
errorlessly differentiating a circle (eccentricity =
1.00) from an array of ellipses (eccentricity = .95)
almost identical to it. Interestingly, “experimental
neurosis” did not accompany this difficult
discrimination.
 The entire program (before revision) is summarized in the accompanying illustration.
The reader is invited to select the odd one from H. The children with mental retardation were
able to do so.
(Back to Contents)

Program Two: Sidman was now interested in reversing the discrimination, that is, have
the children discriminate an odd ellipse from many circles. The reader may logically argue that
presenting one ellipse and many circles, and differentially reinforcing appropriately, should do
the task. However logical this may be, the children’s responding to the circle had been
reinforced, and they continued to do so, although there were now seven circles.

Conceivably, they thought that one of the circles was right, but they couldn’t tell which
one. If, instead of immediate reversal, the fading procedure is continued from one circle/seven
ellipses, to circle = ellipse, to seven circles/one ellipse, there is a period of built-in error.
Besides, it is not fading alone that is critical; it is the use of fading in a program which starts out
with stimulus control that is critical. Stimulus control is lost during the condition of equality
between circle and ellipses. Accordingly, Sidman moved away from near-equality. He
“regressed” the program to gross differences between the one circle and seven ellipses. Rather
than fading directly to equality and then reversal, he faded to intervening dimensions.

First, in the presence of seven flat ellipses, the circle was gradually straightened out on
four sides and transformed into a square with each reinforcement. The children readily
discriminated this circle-square from the ellipses in the other seven panels. The square was now
the oddity SD.

The S∆ ellipses were now gradually transformed into circles, by lengthening the minor
axis. The square continued to control behavior.

The SD square was now gradually transformed into a rectangle, by decreasing its height.
The single rectangle continued to control behavior.

The SD rectangle was now transformed into an ellipse. First its corners were rounded out,
then the top and bottom. Finally, SD was an ellipse while S∆ was a circle. Discrimination had
been reversed without error.
 To ascertain whether the ellipse was SD, or whether it was merely the odd stimulus that
was SD, Sidman threw in one circle and seven ellipses. The children chose one of the seven
ellipses. Discrimination was on the basis of circle-ellipse, with ellipse SD, rather than on the
basis of oddity, although the oddity procedure had been used to establish such discrimination.

The entire reversal program is presented in the accompanying illustration. At Points 5

and 43, the investigator sought to ascertain whether the behavior was under the stimulus control
of oddity or the specific single stimulus. Such assays into the basis of behavior are called
probes, as they are in
chemistry, where a
thermometer is inserted
into a cauldron to find out
what is going on, in this
case, the temperature.
Probes are to be
distinguished from tests.
The former is in the area of
assessment, and suggests
what stage in the program
is to be applied. The latter
is in the area of diagnosis,
which is used to classify
individuals.

Discrimination and reversal training have been the focus of considerable laboratory
research in learning and unlearning. It has been argued that learning and unlearning involve
different aptitudes. Where unlearning has been difficult, it has been argued that the subjects are
rigid or inflexible. This has been related to retardation, hence Sidman’s choice of the reversal
problem. The fading procedures presented suggest that the numerous errors that have
accompanied such problems are unnecessary. The errors and the difficulty of learning and
unlearning using other procedures have needlessly restricted research, since it has been assumed
that the problem is an extremely difficult one beyond the scope of many subjects.
(Back to Contents)
 ERRORLESS PROGRAMMING OF
MATCH TO SAMPLE
In the match-to-sample procedure, it will be recalled, discrimination is defined by the
D
occurrence of behavior in the presence of a stimulus, S , that is in the same class as another
stimulus presented (the sample), and by the absence of behavior in the presence of the stimulus,
∆
S , which is not in the same class. This definition holds for the errorless procedure as well.

An experiment by Moore and Goldiamond used fading procedures to establish match to

sample discrimination between triangles differing slightly in degree of rotation. The subjects
were preschool children for whom the task was quite difficult when, standard match to sample
procedures were used.
(Back to Contents)

Specifying the terminal discrimination: The terminal match was to a narrow isosceles
triangle. The sample was pointed straight down, or tilted slightly from straight down either to
the left or right. The matching choices always included all three. Each choice was presented in a
small window. The position of the correct choice varied, being occasionally in the first, second,
or third window. Discrimination could be represented as follows:

Zero-delay match was desired. When the sample went out, the three choices immediately
went on.
(Back to Contents)
 Assessing the relevant initial discriminative repertoire: The nursery school children
could discriminate light from dark, and could press a button when told to do so. If told to match
the sample, many errors occurred and then considerable time was taken in acquisition of the task,
even when reinforcers, small trinkets or candies, were dispensed liberally. The child was told to
observe the picture in the small window on top the apparatus and to see which way it pointed. It
was then turned off, simultaneously illuminating three choice windows beneath. He was
instructed to touch the match and a correct touch was reinforced by having a light go on above it
immediately; the child was also given the trinket or candy. An incorrect choice shut off the
apparatus until the next trial.
(Back to Contents)

The Program: Only the correct choice panel was illuminated when the sample went off.
This immediately controlled responding. After three such presentations, the voltage in the two
incorrect windows was increased (thereby increasing their brightness) to 35% of the correct one.
The voltage was then increased in a small step with each reinforcement until the incorrect
windows were equal in brightness to the correct one, at which point only the rotation could
govern responding. The fading program can be described in the following manner:

The discrimination established by brightness difference was maintained in its absence,

thereby transferring stimulus control from brightness to form, in an errorless sequence.
(Back to Contents)
 Program evaluation: The investigators evaluated the effectiveness of their program, as
compared to standard match-to-sample, where the subject is confronted with the terminal
discrimination immediately. They did this by shifting from one procedure to another. In the
accompanying illustration, Subject 1 (curve on the left) was started out on the

terminal discrimination, sink or swim. He sank. At A, fading was introduced, and he proceeded
errorlessly to B, where the terminal discrimination was reinforced before the fading program was
complete. Accuracy immediately deteriorated to chance values. At C, the fading program was
reinstated, and the child continued with only two errors during the remainder of the program. At
D, the terminal discrimination is in effect. It is errorless, in contrast with conditions BC, where
the same requirements were not met. This record contrasts with that of Subject 6, who was
started using the fading procedure, and was continued on it without interruption. By T, the
progression had reached 1.00 and was maintained without error. Discrimination was established
with only two errors, and in the minimal time. His condensed performance contrasts sharply
with that of Subject 6 (and the other subjects, not shown), and the similarities of his record to
those of other subjects during the fading series, prompt the suggestion that it was the fading
procedures themselves that were involved in the rapid establishment of discrimination. It should
be noted that the learning of this subject was the most “perfect” -- with the least practice. It is
the program, rather than the amount of practice, that makes perfect. In this experiment, the
fading procedures were instated and eliminated in other subjects as well, and accuracy was
established and reversed in functional relation to the procedures. The procedures were also used
to establish discrimination of selected letters of the alphabet.
(Back to Contents)

Teaching reading and visual match to auditory sample: Among the consequences of
writing is reading the written material, and O.K. Moore capitalized upon this response-
specific-reinforcer in teaching children to read and write simultaneously, using an electric
typewriter. The machine provides an important consequence of its own: it can fail to work if the
wrong key is pressed.

On a screen there appears a picture of a cow with the letters C-O-W underneath. The
child must match these letters on the screen by striking the keys, C-O-W. To control appropriate
responding immediately, these are the only ones illuminated (difference from other letters of the
keyboard, that is, oddity, is providing an assist here.) The illumination on the keys is then faded
out, transferring control of match from light-dark to the forms of the letters themselves. The
letters on the screen are now faded out, transferring control from matching letters-to-letters to
matching letters-to-pictures. Stated otherwise, the child is typing the name of the object
pictured. At other sessions, a word is sounded by a tape recorder as its letters are presented on a
screen. The letters on the screen are faded out, and control is transferred from matching letters to
letters to matching letters to spoken words. Stated otherwise, the child is typing from dictation.
As part of the program, he also types from his own dictation, that is, from words he himself says
aloud. He then fades out the loudness so that he is typing words he says to himself, that is, his
own stories. First graders put out their own newspaper and can read Alice in Wonderland, as
well as other books with words they have never seen in print. Their spoken vocabulary becomes
their reading and writing vocabulary, in contrast to other procedures that severely restrict the
reading vocabulary and thereby bar the child from the reinforcements of the interesting world of
child literature.
(Back to Contents)
 ERRORLESS PROGRAMMING OF
ADJUSTMENT
In the adjusting procedures, discrimination is defined by match to some criterion, where
the matching stimulus is adjusted by the subject. At the present writing, errorless programming
has not been extended to this procedure, but there is no reason why it could not be. Accordingly,
we shall consider how it might be applied.

What strikes one immediately about the adjusting procedures is that the subject is already
fading the stimulus he is adjusting until it matches some criterion. The role of errorless training
would, accordingly, involve fading the definition of the match and its establishment. Stated
otherwise, we would start out with an obvious match already in the organism’s repertoire, and
gradually fade that to the less obvious, and more difficult, match that is the terminal
discrimination.

In the adjusting-to-yellow task, we could start out with large steps between the adjusting
stimulus and the yellow. A peck at the right key would turn the adjusting stimulus from a deep
blue to a yellow that matches the yellow that is present. We would then insert a blue-green
between these two, then more and more intervening colors; we could also gradually eliminate the
extremes, starting with green-yellow. If we wished to program absolute adjustment, we could
make the yellow standard progressively smaller, or have it appear less and less frequently. The
reader is referred to our original discussion, which presented a procedure that incorporated many
programming features.

If we wish the pigeon to rotate a horizontal line so that it matches a vertical standard, the
apparatus will be set up so that each peck presents a frame containing a line slightly rotated from
the preceding one, in the direction of the standard. Following Terrace’s procedure, we might
first establish adjustment of green to yellow. Having done so, we now superimpose upon each
frame in the color progression, a line slightly rotated in the rotation progression. We now fade
out the colors, and transfer control to rotation. Whether or not this will work, we cannot say,
since Sidman’s experience indicates that sometimes more steps are necessary than are called for
by such a logical analysis. Nevertheless, the reader is invited to suggest how we might program
a color adjustment without error, using, say, flicker or darkness from which to transfer, as
Terrace did.
(Back to Contents)

OTHER FADING OF STIMULUS CONTROL

The techniques and procedures of fading have been bases for successful persuasion since
recorded history began. Practice precedes science. In the field of learning, fading is comparable
to the “phasing out” and programming of change in industrial and other social settings. Here,
too, the consideration is change that is programmed to produce minimal disruption, that might
otherwise be considerable.

In the psychiatric and penal context, “halfway houses” have been suggested and have
been put to use in certain problem areas to provide a transition from the hospital to the home,
that is, to provide for stimulus control that is somewhat removed from that provided by the
institution but that is closer to the terminal control. In psychotherapy, the therapist often tries to
assume less and less control as therapy progresses; graduate educational systems often fade out
their control, setting the conditions for increasing control by the professional requirements of the
task. An anthropologist reports that in order to get certain American Indian tribes to attend
church, missionaries from a religious order integrated parts of the Indian ritual into the Christian
ritual, gradually replacing pagan elements with Christian ones (Easter eggs are vestiges of early
fertility rituals). Identification with the church became so strong that when the religious order
was withdrawn by edict, the Indians withstood armed attack by forces of the central government.

The explicit use of fading is not, accordingly, a discovery of the experimental analysis of
behavior. The contribution here would be the procedures for making explicit the entire program,
including the basic behavioral specification of the terminal discrimination, the relevant current
repertoire, and each of the stimulus requirements that specify the intervening steps, as well as the
behavioral contingency relations to consequences. In probably no other area of investigation
have these been so well spelled out as in programmed instruction. Any serious discussion of this
field would require a separate book. An example from a program by Holland will serve as an
illustration. The
program is in
neuroanatomy, and is
for medical students.
Frames A, B, and C
are taken from
successive stages of
the program. At each
step, the student is
required to identify the
spatial locations of
different anatomical
structures. After
frame C, the picture
itself is faded out; the
student can now
discuss the spatial
positions without an
external visual
representation. In a
sense, a “private map”
of the structures
involved has been
programmed. The reader will recall the corks, ping-gong balls and sticks used to “internalize”
vectors in a previous example. Features of the Montessori method can be considered implicit
programs to develop mathematical abstraction. One of the sequences starts out with the
assumption that children are less abstract than adults. Disagreement with this assumption has
caused some psychologists to overlook the fact that it is being used to explain a procedure that
can be related to programming, namely, that contact with balls and sticks is more likely to be
part of the child’s initial repertoire than is contact with circles and lines. The sequences then
move increasingly in the direction of mathematical relations.

An implied fading procedure may be involved in a clinical procedure, called

“desensitization.” Where the patient reports, for example, a fear of going up elevators (called a
phobic reaction), the therapist supplies instructions for him to relax, to imagine what he is told,
and to report when he feels anxious -- usually by raising a finger. He is then told to imagine
going into an elevator, going up the first floor, and so on; with the experimenter continuing in
this progression until the patient raises his finger. The therapist stops here, instructs the subject
to relax again until the finger drops, and then resumes his instructions of going up the imagined
elevator until the finger goes up again, and so on. (Instead of the elevator progression, which is a
simple linear one, the subject may be asked to establish a progression of his own behaviors in
picking up a laboratory rat. The procedure is then applied to this progression.) Wolpe’s
explanation is that the respondent anxiety is being extinguished, or desensitized, at each step in
the progression. If we consider the instructions to imagine the elevator as controlling such
imagery, then the procedures progressively increase the range of the imagery that controls
relaxation. Starting out with minimal elevation, relaxation is brought under the control of
progressively more elevation in this implicit fading procedure. The explanations and procedures
seem quite straightforward, but it has been argued that the simplicity is only apparent, and that
complex explanations and procedures are being concealed by the ready acceptance of certain
implicit assumptions. It is assumed, for example, that the anxiety indicator response is a valid
one (we shall discuss indicator validity in the sections on psychophysics). Other assumptions are
also beyond the scope of the present discussion, and will be considered elsewhere.

Fading has been increasingly applied in the explicit manner required by programming, to
human areas of research other than formal programmed instruction. In research with chronic
stutterers, for example, Goldiamond established a new speech pattern completely devoid of
stuttering by using delayed auditory feedback. The subject read aloud and heard himself 250
milliseconds (ms) after he had produced the sounds. This was part of a program to produce a
novel and prolonged speech pattern (the subject re-e-e-ead li-i-iike thi-i-i-is). Once the new
pattern was established, the delay was faded out from 250 ms to 00 ms over several days, and the
new pattern continued. It was then speeded up to produce extremely rapid and well-articulated
reading, devoid of stuttering. The subject was reading alone, from material presented on a
screen. In successive stages this was changed to conversation with a person outside the booth.
For many of the subjects, each change produced a breakdown in the new pattern. Whenever this
occurred, the delayed feedback was reinstated, returning the speech to its well-articulated
pattern. The delay was then faded out, transferring its control to the new situation. The program
could have been established without breakdown at all by reinstating delay at each transfer point,
but here, there would have been conflict with the time requirements of the patients, which called
for trying to make the program work as rapidly as possible by omitting steps and backtracking
where the behavior itself indicated the step was too large. It should be noted that, as in explicit
programming, it was the objective record of the behavior itself that provided this information,
rather than intuitive or “clinical” judgment. These procedures have been extended to carry the
behavior over outside the laboratory, into everyday conversation and verbal behavior.

Sherman, working with psychotic patients who had not spoken for many years, had them
repeat sentences verbatim. He then left out more and more of the sentence, fading out his own
control and transferring it to the speaker. Another clinical investigator reported treating a school
phobia in this manner. The therapist and child daily took longer walks in the direction of the
school, finally coming into the class on opening day. He stayed with the child all morning.
Everyday thereafter, he excused himself for increasing periods, and faded himself out entirely.

In Sherman’s investigation, reinforcers were explicitly provided contingent upon each

successful completion. In the school case, we must assume that the therapist and class provided
social reinforcers that were contingent on walking with the therapist initially, or staying at
school, later. The behavioral requirement to produce them was altered every day, as were the
discriminative stimuli. In both cases, changing discriminative control is intertwined with
changing behaviors and yet other changes. This type of confounding is often one of the facts of
life outside the laboratory. It indicates why the laboratory is a ready source of isolation of
relevant functional relations, and the testing and development of procedures. Procedures so
developed may then be applied, outside, where it is more difficult to ascribe the results obtained
to them.
(Back to Contents)
 MEASUREMENT OF STIMULUS CONTROL
The absence of error poses a measurement problem in programming. Where behaviors
are categorized as correct and incorrect, more desirable and less desirable, or into other discrete
evaluative categories, the categories may be used to measure discrimination or stimulus control.
In academic examinations, grades may be based on numbers of correct responses, or ratios of
correct to incorrect, or to totals (that include errors). For rats running mazes, a criterion may be
defined in such terms as “ten correct in a row.” Task difficulty may be measured by differences
in such scores. In fading programs, there are no errors. If no mistakes are made, how can we
assess the difficulty of a program, or the comparative complexity of different concepts and types
of stimulus control?

This unit will consider the evaluation of stimulus control when errorless procedures are
used. We have already considered the evaluation of the learner. His grade is how far he has
advanced. If he is at Step 70, next semester he starts 71. If he is at Step 90, next semester he
starts 91. If he transfers into our school, we probe his performance to see what step we start him
on, and what programs we use. The amount of time required to reach comparable points in the
program may also be used to compare individuals, and to compare programs, giving us slow and
rapid learners and programs.

A more accurate measure is probably the number of steps required to reach the same
point. All things being equal, the program with the smaller number of steps is the better one.
The number of errorless steps required to attain the terminal repertoire may also be used to assess
the difficulty of the task. Given two different tasks with steps of the same size, the task that has
more errorless steps is the one that is more difficult to develop. This difficulty measure may also
be used for assessment of an individual. Certain people with aphasia do not rhyme readily. As
they recover from their head wounds, they compose more rhymes. When they are completely
recovered, they rhyme readily. Let us assume that a programmer develops a program that can
teach any person with aphasia to rhyme. The programmer may have to run the person with
aphasia from a fresh head wound through all one thousand steps of the program to obtain the
terminal behavior. He may run another person with aphasia who has partly recuperated through
only the last five hundred steps. The person completely recovered from aphasia may, need only
to be given the last ten steps. It should be noted that we are talking of procedural difficulty, that
is, the difficulty that the investigator faces in programming the subject’s repertoire without error.
It may be faster for the subject to take a few large steps, but he may also make mistakes. This
interplay requires further investigation.

If difficulty is defined by the number of steps, then complexity may be defined by the
number of dimensional changes required. It will be recalled that Terrace faded a dark period into
the red presentations, producing a light (red)-dark discrimination. He then faded green into the
dark, producing a red-green discrimination. He then superimposed vertical and horizontal lines
upon the colors, and faded the colors out, producing line-tilt discrimination. These are disparate
dimensions, and each dimension requires a different number of steps. The number of dimensions
might define the complexity of the task. Some tasks might be far more simple (less complex)
than others, although far more difficult (more steps).

It should be evident that the complexity and difficulty of establishing any repertoire will
depend on the relevant current repertoire of the organism. Complexity and difficulty cannot
therefore be considered independently of the repertoire.
(Back to Top)

♠
The sounds 1 and w are in the same stimulus classes in many Western languages, e.g., the
Polish letters l and ł, and the French phonetic change from terminal l to w in the plural (cheval,
chevaux). The sounds r and w are often in the same stimulus class in English, and the Asian
substitution of r and w is not confined to these languages.