Tutorial Stimulus Control: Part II

The Behavior Analyst 1995, 18, 253-269 No. 2 (Fall)

Tutorial Stimulus Control: Part II

James A. Dinsmoor Indiana University

The second part of my tutorial stresses the systematic importance of two parameters of discrimi- nation training: (a) the magnitude of the physical difference between the positive and the negative stimulus (disparity) and (b) the magnitude of the difference between the positive stimulus, in par- ticular, and the background stimulation (salience). It then examines the role these variables play in such complex phenomena as blocking and overshadowing, progressive discrimination training, and the transfer of control by fading. It concludes by considering concept formation and imitation, which are important forms of application, and recent work on equivalence relations. Key words: stimulus control, disparity, salience, blocking, overshadowing, transfer, fading, con-

cept formation, imitation, equivalence relations

The first part of this tutorial dealt with the basic principles that account for the acquisition of stimulus control under what is conventionally known as discrimination training. Among the sa- lient points that were raised were sug- gestions that (a) control by antecedent stimuli is just as important in operant as it is in respondent behavior; (b) Pav- lov’s conditional stimulus is a discrim- inative stimulus; (c) stimulus general- ization is not a behavioral process in- dependent of and antagonistic to dis- crimination but is simply another way of describing control by antecedent stimuli; and (d) the increases in control that occur during discrimination train- ing can be attributed to more frequent and more prolonged observation of the relevant stimuli accompanied, it is pre- sumed, by concomitant changes in at- tention.

Note that in conventional discrimi- nation training, the subject is exposed to repeated alternations between the positive stimulus (S+), which is ac- companied by a schedule of reinforce- ment, and an alternative stimulus (S-), which is not accompanied by rein- forcement. This alternation establishes

Address correspondence to James A. Dins- moor, Department of Psychology, Indiana Uni- versity, Bloomington, Indiana 47405.

a positive correlation between the S+ and the primary reinforcer that selects the one relevant stimulus out of the many that impinge upon the organism and transforms it into a conditioned re- inforcer of observing behavior. Note also that if the subject is exposed ex- clusively to S+ training or to a single period of S+ training followed by a single period of S- training rather than to a continued alternation, control is less adequate (Honig, Thomas, & Gutt- man, 1959; Yarczower & Switalski, 1969). Similarly, in a compound dis- crimination like that studied by Blough (1969), if one dimension is left contin- uously at its positive value for a num- ber of sessions, control by that dimen- sion is reduced and control by the other dimension enhanced; when the proce- dure returns to the previous alternation between positive and negative stimuli, the discriminative performance returns to its normal level. The second part of my tutorial will

necessarily be less tightly integrated. Taking off from the foundation laid in Part I, it extends the treatment of stim- ulus control to a discussion of two of its most important parameters and to several more complex patterns that seem to hold special significance for basic theorizing and practical applica- tion.

253

254 JAMES A. DINSMOOR

STIMULUS PARAMETERS There are a number of variables that

influence the rate at which the subject learns to discriminate between stimuli and the level of performance that will ultimately be attained. However, some of these, like the physical quality of the stimulus (e.g., wavelength of light, en- ergy of sound), the topography of the response, and individual and species differences among different subjects, do not readily lend themselves to the formulation of general laws of behav- ior. Of greater interest from a system- atic point of view are those parameters that enter into a variety of behavioral paradigms. There are two parameters, in particular, that will appear and reap- pear in subsequent accounts of topics like overshadowing, blocking, the easy-to-hard effect, fading, and con- cept formation. These are (a) the mag- nitude of the difference in physical units between the positive and the neg- ative stimulus, sometimes subsumed by the phrase stimulus disparity, and (b) the magnitude of the difference be- tween the discriminative stimuli and the background stimulation, subsumed as stimulus salience.

Stimulus Disparity Clearcut evidence for the role played

by the disparity between the two stim- uli may be found in a series of early studies, using rats as subjects, con- ducted by Rosemary Pierrel and her as- sociates at Brown University. Bar pressing was reinforced on a variable- interval schedule in the presence of the positive stimulus but not in the pres- ence of the negative stimulus. In Pier- rel, Sherman, Blue, and Hegge (1970), for example, the discriminative stimuli were pulsed tones of 4 kHz that dif- fered in intensity. The difference be- tween the positive and the negative stimulus was set at 10, 20, 30, or 40 dB, with the higher intensity serving as the positive stimulus for half the groups and as the negative stimulus for the other half. Also, for half the ani- mals, independently assigned, the low-

er intensity was set at 60 dB, with the higher intensity determined by the magnitude of the difference between the two; for the other half, the higher intensity was set at 100 dB, with the lower intensity determined by the mag- nitude of the difference. (Because there was only one 60-100 group and one 100-60 group, there were 14 rather than 16 groups in all, plus a special control group.)

It seems obvious that very small physical differences between the posi- tive stimulus (SD or S+) and the neg- ative stimulus (S^ or S-) must be dif- ficult to discriminate. Up to some limit, at least, larger differences should pro- mote faster acquisition and a larger ul- timate difference between the two per- formances. This expectation is borne out by a series of plots tracing the course of the discrimination index (multiplied by 10) as a function of the hours of training. In all four panels of Figure 1, the discrimination is slowest to develop and attains the lowest final level in groups for which the stimuli differ by only 10 dB; groups for which the difference is 20 dB do somewhat better; groups for which the difference is 30 or 40 dB differ less during ac- quisition and tend to converge, but at still higher levels of performance. With pigeons as subjects, Hanson (1959) found a similar relation between the magnitude of the difference in wave- length and the time required for the de- velopment of a discrimination. Again the slope of the function was steep at small values but decreased as larger values were approached.

Stimulus Salience

Further examples of the effects of stimulus disparity would not be hard to find, but the magnitude of the differ- ence between the discriminative stim- uli and their background stimulation (salience) poses more of a problem. This is a dimension that does not come up for consideration within the older response-strengthening-and-weakening theories of discrimination learning.

STIMULUS CONTROL 255

A B

900 100

, ~ ~ OR IN LCSO

[V ~~ ~ ~ ~ ~ ~ ~ ~ ~ _

700/ :! ~~60- 707!-6

X600 60 -80 —- 80-m60–

F 60Pct 100 – 60 z o 500 P

o900 V

spcfew ndcbl.(erdcdfo ire ta. 90 cprgh Socet fo th xermna

SD ofDBA 100- 90- a90 – 00-

600 loo~0- 80– 80 – 100— 100 – 70-”-70 – tO …..-

100-~~~~~~~~606- 100- 500

32 96 160 224 32 96 160 224

HOURS IN BLOCKS OF 8

Figure 1. Percentage of responding in SD (S+), multiplied by 10, as a function of hours of training. Each point represents the mean of 4 rats during an 8-hr session. The stimuli for each group are specified in decibels. (Reproduced from Pierrel et al., 1970; copyright Society for the Experimental Analysis of Behavior.)

However, when investigators are guid- ed by an interest in the role of observ- ing or attention, it becomes a more likely candidate. In my laboratory, we became aware of the importance of sa- lience while studying the effects of

stimulus disparity on the relative rates of pecking two observing keys (Dins- moor, Sears, & Dout, 1976). In our first experiment, the birds pecked a key that produced large increases or decreases in illumination as discriminative stim-

256 JAMES A. DINSMOOR

uli at higher rates than they pecked a key that produced smaller increases or decreases in illumination. In our sec- ond experiment, however, we varied the magnitude of change for the posi- tive and the negative stimuli indepen- dently. We discovered that the positive relation between magnitude of change and rate of pecking stemmed entirely from the positive stimulus: For the negative stimulus, the function was negative in slope. That is, larger changes in stimulation reduced the rate. On returning to the experimental literature, we were led to the conclu- sion that in most previous examina- tions of stimulus difference, the mag- nitude of the difference between the two stimuli had been confounded with the magnitude of the difference be- tween the discriminative stimuli and their background. That is, the disparity betwen the stimuli had been confound- ed with their salience.

At the time, we were not aware that a study of discrimination in which the salience of the stimuli had indeed been varied independently of other factors had already been conducted by John- son (1970). A student of Cumming, Johnson was interested in the role of attention in discrimination learning, and one of the parameters he had var- ied, as part of a larger study of selec- tive control, was the brightness of a white line displayed on the pigeon’s key. Pecks when the line was vertical (S+) were reinforced on a random-in- terval schedule, but pecks when the line was horizontal (S-) were never reinforced. The rate at which the birds learned the discriminaton between these two orientations was a function of the brightness of the line.

Still blithely unaware of Johnson’s study, Dinsmoor, Mueller, Martin, and Bowe (1982) chose as their controlling stimulus a black line bisecting a white key. Pecks on the key produced grain on a variable-interval schedule that al- ternated with an extinction schedule. Ordinarily, the line remained horizon- tal in its alignment, regardless of which of these schedules was operating. But

by depressing a low-lying “perch,” similar to the conventional cross-bar for rats, the pigeon produced stimuli that were correlated with the compo- nent schedules. During periods when the variable-interval schedule was in effect, the stimulus was, for two dif- ferent groups, a clockwise tilt of 150 or a clockwise tilt of 300 (S+); during pe- riods when the grain was being with- held, it was a counterclockwise tilt of the same magnitude (S-). For half of the subjects, then, the total difference between the two tilts was 30°, and for the other half, it was 600. These differ- ences represented small and large dis- parities, respectively, between the two stimuli. To vary the salience of the stimuli,

the experimenters used an arrangement similar to that of Johnson (1970), re- ducing the contrast between the black line, regardless of its tilt, and the white surround. This was accomplished by the simultaneous lighting of two pro- jection units, one of which displayed an image of the line on the key and the other of which produced only a uni- form white field. Both lamps were re- duced to 50% of their normal intensity. Thus, the salience of the stimuli could be manipulated without affecting their disparity. When the image of the line was projected at its maximal contrast by a single projector cell, 7 of the 8 birds in that group learned both to hold down the perch and to discriminate in their rate of pecking between the re- sulting stimuli. When the second cell was lighted, the line, now gray, was readily visible to the human eye and presumably to that of the pigeon. Nev- ertheless, only 1 of the 8 birds in the other group gave any indication that its behavior was influenced by the tilt of the line. Obviously, the salience of the stimuli was an important determinant of the level of observing and of the consequent level of discrimination. The effects of disparity were not as clear as those for salience, but in a sub- sequent study, using an entirely differ- ent design, Dinsmoor et al. (1983) found that both of these dimensions af-

STIMULUS CONTROL 257

fected the rate at which pigeons pecked an observing key.

SELECTIVE CONTROL

It was Pavlov (1927/1960, pp. 141ff.) who first called attention to the reduction in the effectiveness of a stim- ulus that sometimes occurred when that stimulus was presented in tandem with another stimulus. He found that when two conditional stimuli were reg- ularly presented together, in the same temporal relation to each presentation of the unconditional stimulus, one member of the pair often came to as- sume complete control over the amount of saliva that was secreted, to the partial or complete exclusion of the other stimulus. That is, when tested alone, the second stimulus was ineffec- tive. On further examination, he found that it was the relative intensity of the two conditional stimuli (CSs) that de- termined which of them prevailed and the extent to which the effectiveness of the less intense stimulus was reduced. When the stimuli were of equal inten- sity (e.g., two tones of equal loudness), the phenomenon did not appear: The two stimuli were followed by re- sponses of equal magnitude. Pavlov therefore spoke of the more intense CS as ”obscuring” or “overshadowing” the less intense CS.

Years later, at the Miami Symposium on the Prediction of Behavior, Kamin (1968) presented a series of experi- mental comparisons that provided con- vincing evidence not only for over- shadowing but also for another phe- nomenon, also based on the simulta- neous presentation of two stimuli. In this case, the dominance of one stim- ulus over the other was not established by its intensity but by prior training. If the subject was trained first with a sin- gle CS and later with a compound that included the original and some other stimulus, the added CS turned out to be ineffective. Kamin spoke of the pri- or training with the first stimulus as “blocking” the subsequent acquisition of control by the second stimulus. In

the same year, a set of experiments published by Wagner, Logan, Haber- landt, and Price (1968) implicated yet a third variable, the consistency with which each member of the stimulus compound was followed by the uncon- ditional stimulus (US). A CS that was always followed by the US reduced the effectiveness of a stimulus that was followed by the US on only half of its presentations. The common thread running through

the loss of effectiveness in each of these instances is that both stimuli are regularly presented at the same time and therefore in the same temporal lo- cus with regard to the unconditional stimulus. Other ways of describing the relation between the two CSs are to say that they covary, that they are con- founded, or that they duplicate one an- other. Cognitive psychologists often re- fer to the second stimulus as “redun- dant,” meaning that it provides no ad- ditional information concerning the arrival of the US (i.e., no improvement in the prediction of the US over that provided by the first stimulus). Both stimuli convey the same meaning. In courses on composition, the word re- dundant is sometimes used to refer to the type of error manifested by a speaker who proclaims that “We’ve won four straight in a row.” In a row conveys the same information as straight, and the repetition is jarring. (Sometimes, however, a more remote and less conspicuous redundancy is ef- fective in expository writing.)

Blocking and overshadowing have also been observed with operant be- havior. The primary difference in pro- cedure is that instead of the stimulus control being based exclusively on the presence versus the absence of a CS, as in respondent conditioning, in op- erant work the discrimination is rec- ognized, and both a positive stimulus and an explicit negative stimulus are normally provided. In a study by vom Saal and Jenkins (1970), for example, pecking was initially reinforced in the presence of green illumination of the key but not in the presence of red; later

258 JAMES A. DINSMOOR

a tone and a noise were added to the procedure, covarying with the first pair of stimuli; and finally, control by the tone and noise alone was compared with control by those same stimuli in a group of birds that had not been pre- trained with the red and the green. The prior training on the red-green discrim- ination was found to have blocked the acquisition of control by the tone and the noise.

In a study of overshadowing, Miles and Jenkins (1973) reinforced pecking in the presence of a bright light and a tone of 1000 Hz (S+) but not in the presence of a light of lower intensity and a white noise (S-). The intensity of the second light varied from group to group. In subsequent tests, the tone had relatively little influence over the responding by birds in the group that had received the largest difference in illumination (brightest light vs. total darkness) but exerted much more con- trol with birds that had been trained with smaller differences in illumina- tion. The easier the light discrimina- tion, the more it overshadowed the tone. The harder the light discrimina- tion, the less it overshadowed the tone.

Note that in this experiment the vari- able that determined the amount of overshadowing was not the absolute intensity of the S+ (which remained the same from group to group) but the size of the difference between that S+ and the corresponding S- (stimulus disparity). This may have been the crit- ical variable in the respondent work as well, because the absolute intensity of the CS (which was assumed to be the relevant dimension) was always con- founded with the magnitude of its dif- ference from the absence of the CS (disparity). Recognizing that the con- ditional stimulus is a discriminative stimulus enables us to formulate broad- er, more general principles of behavior.

Blocking and overshadowing have figured prominently in theoretical dis- cussions about Pavlovian conditioning, but at a more applied level they have largely been ignored. I suspect that their manifestations appear fairly fre-

quently in everyday life and that over a period of time a good many illustra- tions may be found. For the present, one example that comes to mind is a phenomenon, well known to social psychologists, in which individual members of an outgroup who share a prominent physical characteristic are perceived as “all looking alike.” (For a recent list of citations, see Anthony, Copper, & Mullen, 1992.) That is, for each individual, the characteristic com- mon to all members of the group over- shadows the characteristics distinctive of that one individual and makes dif- ferent members of the group more dif- ficult to discriminate. Two major types of explanation

have been offered for blocking and overshadowing. Rescorla and Wagner (1972) have proposed an elegantly simple mathematical model of Pavlov- ian conditioning, suggested directly by Kamin’s (1968) data, that accounts for these phenomena in terms of a ceiling on the level of associative strength. When one stimulus is relatively intense or is presented earlier in training, the overall strength of conditioning based on that stimulus approaches an asymp- tote, and little additional conditioning can occur to a second stimulus that is less intense or is introduced later in training. Theoretical descriptions of this type (see also Revusky, 1971), however, do not account for instances in which overshadowing has been demonstrated on the first trial of con- ditioning (e.g., Mackintosh & Reese, 1979). The other type of explanation that

has been popular in discussions of blocking and overshadowing relates the reduction in effectiveness to a fail- ure to observe or to attend to the sec- ond stimulus or pair of stimuli (e.g., Mackintosh, 1974, pp. 585ff.). Some support for this interpretation may be found in experiments on observing be- havior. Two replications of such an ex- periment were conducted in my labo- ratory. In the first one, some of the data were disrupted when we were required to remove the birds from their usual

STIMULUS CONTROL 259

living quarters. Then the second one became to a certain extent redundant scientifically when a similar study was published by Blanchard (1977). In our experiments, which were somewhat more complex than Blanchard’s, we provided the pigeon with an observing key that altered the tilt of a white line displayed on a colored ground. During periods when food was being pro- grammed on a variable-interval sched- ule, pecking the observing key inter- mittently produced a shift in the ori- entation of the line, rotating it 450 from the horizontal; during periods when no food was scheduled (extinction), the shift was in the opposite direction. Meanwhile, the colored backgrounds switched back and forth on a schedule that was unrelated to the schedules of food delivery and the direction of the rotation produced by pecking the ob- serving key. On some sessions the al- ternating colors differed by a large amount, and on other sessions they dif- fered by a small amount.

After the rate of pecking had stabi- lized, however, we synchronized the alternations of the colors with the al- ternations between the variable-inter- val schedule and extinction and with the direction taken by the line. At this point, the changes in the orientation of the line produced by pecking the ob- serving key became completely redun- dant: Whenever such a change was produced, its relation to the delivery of food was exactly the same as that of a color already displayed on the key. As would be expected, the rate of pecking the observing key in the presence of one of the S – colors (pecking that pro- duced only S- tilts) dropped precipi- tously. Much more important was the fact that the rate of pecking in the pres- ence of one of the S+ colors also de- clined. In other words, less observing behavior was maintained by the line tilts as conditioned reinforcers when they covaried with another, more effec- tive set of stimuli. Moreover, in accord with conventional experiments on overshadowing, when the colors that differed by a large amount were dis-

played on the key, the rate of pecking dropped more than when the colors that differed by a small amount were displayed.

At a broad and general level, then, a model based on principles derived from the study of observing can ac- count for the effects of continued train- ing, as well as for the overshadowing sometimes noted on the initial trial. On the other hand, it is difficult to see how any model based on the observation of or attention to one stimulus rather than another could handle the results ob- tained by Revusky (197 1), which mim- ic blocking and overshadowing despite the absence of any temporal overlap between the two conditional stimuli. Perhaps no single factor can account for all of the data obtained in experi- ments on blocking and overshadowing.

Progressive Discrimination

Another phenomenon that highlights the importance of the magnitude of the difference between the stimuli to be discriminated (disparity) is known var- iously as progressive discrimination training, transfer along a continuum, or the easy-to-hard effect. It has long been recognized that the most efficient way to train a subject to discriminate between two stimuli that lie close to- gether along the physical continuum is to begin with stimuli that are farther apart and then, in successive steps, to close the gap (e.g., James, 1890, p. 515; Montessori, 1912, p. 184). In oth- er words, initial training with easy stimuli from the same dimension ac- complishes more than an equal period of training with the difficult pair on which the subject is eventually to be tested. In Pavlov’s laboratory, this technique was employed to train a dog to discriminate between a light gray and a white circle and to train another to discriminate between a fairly well- rounded ellipse and a perfect circle (Pavlov, 1927/1960, pp. 121ff.). Even before bar-pressing or key-

pecking techniques came into common use, systematic data were reported with

260 JAMES A. DINSMOOR

relatively crude procedures, typically involving choices between test cards of differing shades of gray. The study that has been most widely cited, historical- ly, is one reported by Lawrence (1952). In his procedure, the rat was forced to jump across a gap that separated the starting box from a pair of goal com- partments lined with cardboard in dif- fering shades of gray. One group of subjects was trained throughout with two highly similar grays, the same as those used in the final test; two groups started with the lightest and the darkest gray and were shifted to the final test pair at different points in their training; and finally, one group received its first 10 trials with the most disparate pair of grays, the next 10 trials with a less dis- parate pair, the third 10 trials with a still less disparate pair, and its last 50 trials with the final test pair. All the groups that began with the large dif- ference performed better during their last 50 trials than did the group that began with the difficult discrimination. The last group, which shifted in a se- ries of steps, made the fewest errors of all the groups. Lawrence suggested that the easy stimuli facilitated later learn- ing of the hard stimuli because they helped “the animal to isolate function- ally the relevant stimulus dimension from all the other background and ir- relevant cues” (Lawrence, 1952, p. 516). Both the magnitude of the dis- parity between the positive and the negative stimulus and the salience of the positive stimulus, which is con- founded with its disparity in many of these studies, contribute to the effec- tive reinforcement of relevant observ- ing behavior (Dinsmoor et al., 1983).

Fading

In contrast to the research on block- ing and overshadowing, which was concerned exclusively with an eluci- dation of the basic principles of con- ditioning, the original impetus to re- search on fading seems to have ema- nated from attempts to fashion effec- tive techniques for use in programmed

instruction. Early papers (Cook, 1960; Skinner, 1958) describe a procedure then known as “vanishing”: In succes- sive steps, letters were eliminated from words to be spelled by the subject, words were deleted from passages to be recited from memory, or labels for various geographical features were re- moved from a map. The subject con- tinued to respond correctly, despite the elimination of the original stimuli, even when the only remaining cues might be those generated by chains of ongoing behavior.

Almost immediately, there was a change in terminology, and the same technique came to be known as “fad- ing” (Holland, 1960). Then, not long after that, investigators using nonhu- man subjects began studying fading in the conditioning laboratory. Using a highly specialized training procedure, Terrace (1963) had initially established a discrimination between red and green illumination of the pigeon’s key. Al- though he did not point out the paral- lel, the red and the green were equiv- alent to the stimuli that were already effective when a human trainee began his or her program of instruction. Then a vertical line was superimposed on the red S+ and a horizontal line on the green S -. Gradually, in successive steps, the brightness of the red and the green was reduced, until the key be- came totally dark except for the lines. In other words, the red and the green illumination were faded. As the sali- ence of the colors was reduced, the birds made less and less use of those stimuli and gradually came to depend on the direction taken by the line. In two replications of this transfer proce- dure, a total of 4 birds learned the ver- tical-horizontal discrimination without ever pecking in the presence of the S- (i.e., without making a single error).

Other techniques, used as experi- mental controls, proved to be far less effective. For example, if the red and the green were completely removed from the key at the time that the ver- tical and the horizontal lines were in- troduced, so that there was no temporal

STIMULUS CONTROL 261

overlap between the two sets of stim- uli, hundreds of S- responses occurred before the birds learned the vertical- horizontal discrimination. When the two sets of stimuli overlapped in time but nothing was done during this time to reduce the effectiveness of the red and the green, intermediate results were obtained. Something was evi- dently learned with regard to the lines during the period when they were su- perimposed on the colors, but not as much. It was the gradual reduction in the salience of the colors that forced the birds in the fading group to switch to the lines during the period when both sets of stimuli were available. By the time the colors were completely gone, the birds had already learned to discriminate accurately on the basis of the lines. (For an alternative procedure in which the original stimuli were de- layed in onset rather than reduced in salience, see Touchette, 1971.)

There is another parallel that was not obvious at the time Terrace conducted his work: His red and green illumina- tion were also comparable to the pre- trained or inherently more effective stimuli subsequently used in operant studies of blocking and overshadow- ing. In light of this correspondence, Terrace’s (1963) fading procedure can be viewed as a method of overcoming the overshadowing or undoing the blocking of one pair of stimuli by an- other. By reducing the influence of the originally dominant pair, fading allows the stimuli that have been blocked or overshadowed to gain effectiveness. The advantage offered by Terrace’s

(1963) fading procedure was that dur- ing the prior training, highly effective stimuli were available on the key. In contrast to the lines, which were more localized, the colors covered the entire surface of the key, and anywhere the pigeon might initially look while peck- ing that key it was likely to see the color. Early contact with a stimulus correlated with reinforcement pro- duced early acquisition and a high lev- el of performance of the behavior by which the bird observed that stimulus.

Later, when the lines were introduced, the birds had already learned to look at and attend to stimuli displayed on the key. The lines were displayed in the same general location as the colors (i.e., on the surface of the key). Then, as the salience of the colors was re- duced, their value as reinforcers de- clined (Dinsmoor et al., 1982, 1983) and the bird shifted from observing the colors to observing the lines.

Evidence confirming this interpreta- tion comes from a study conducted by Fields (1978). Again the pigeons were pretrained to discriminate between red and green illumination of the key. Then white lines of differing orienta- tion were superimposed on the red and green backgrounds. Between blocks of training trials, Fields inserted probe tri- als on which he presented the same white lines, but on a dark background. These trials enabled him to track the acquisition of the line discrimination during the course of training. The most revealing aspect of Fields’

(1978) procedure was that with one group of subjects he faded both the red and the green, with another group he faded only the red (S+), leaving the green at its original intensity, and with a third group he faded only the green (S-), leaving the red at its original in- tensity. Interpretation of the difference in results between the first two of these groups is complex and need not con- cern us here, but the interesting finding occurred with the third group. Data from a large number of experiments in- dicate that it is the S+ that reinforces observing (for a review of these data, see Dinsmoor, 1983). Accordingly, an interpretation in terms of the observing paradigm predicts, uniquely I think, that it is fading of the S+ that is re- quired to transfer control of pecking from the colors to the direction taken by the lines. Fading of the S – should have no effect. And that is the result that Fields obtained. “Attenuation of the S – alone … did not produce stim- ulus control by the line-tilt dimension” (Fields, 1978, p. 126). His Figure 1 showed no pecking to either S+ or S –

262 JAMES A. DINSMOOR

probes when it was the green (S-) il- lumination that was faded.

CONCEPT FORMATION

When, in everyday conversation, people speak of someone “possessing” a certain concept, it is taken for granted that they are referring to some kind of unobservable content or state of the mind. But when we examine the mat- ter, we find that the observation that leads them to use the term concept is that the person responds in the same way to a set of objects or relations that have some characteristic or set of char- acteristics in common but does not re- spond in that same way to other objects or relations. All birds are called birds, for example, but no snakes, squirrels, or sheep. As Keller and Schoenfeld put it, “Generalization within classes and discrimination between classes” (1950, p. 155).

If we can verbalize what it is that the various instances have in common, we may be able to define the concept, but often this can be very difficult. Try defining what is meant even by such familiar categories as a “dog,” for ex- ample, or a “human being.” Are you sure you did not include wolves or coyotes in the first instance or chim- panzees in the second? Did you in- clude feral children? Definition is sec- ondary to the behavior: In essence, it is a description of that behavior, al- though it can provide in turn a rule for increasing the accuracy of that behav- ior or for passing it along to another person (e.g., Skinner, 1969, pp. 121- 125, 136-142).

Laboratory research on the forma- tion of concepts arose from an interest in the processes of human thought and has continued for many years in almost complete isolation from research on the formation of discriminations. The general nature of the procedure is il- lustrated by an experiment conducted by Trabasso (1963). College students were presented with drawings of flow- ers, in a randomized sequence, and were asked to classify each drawing by

calling it an A or a B. Four different varieties of flower were used, and each of these varied also in the shape of its leaves, the angle of the leaves to the stem, and the number of leaves on ei- ther side. For most of the subjects, the angle with the stem was the relevant feature (i.e., the discriminative stimu- lus), determining whether A or B was the response to be reinforced. Trabasso found that such strategies as holding other features constant, rather than al- lowing them to vary, and, as in ordi- nary discrimination training, increasing the difference between the positive and the negative angles (i.e., disparity), or increasing their difference from the background stimulation by coloring them red (i.e., salience) all led to faster acquisition of the correct categoriza- tion. These findings suggest that con- cept formation is very similar to dis- crimination training. In particular, the effect of highlighting the angles to in- crease their salience indicates that a major part of the subject’s task is to learn to observe the appropriate feature within the field of stimulation.

Traditionally, as I have indicated, psychologists have treated experiments dealing with the formation of discrim- inations and experiments dealing with the formation of concepts as two sep- arate and distinct fields of inquiry. But the close relation between the two has long been recognized by some writers (e.g., Keller & Schoenfeld, 1950), and whether the traditional distinction is justified is open to question.

It is true that most of the work on discrimination learning has been car- ried out with rats, pigeons, and mon- keys, whereas most of the experiments on concept formation have been con- ducted with human subjects. This re- flects the differences in the historical origins and rationales of the two types of research. But there is no hard and fast rule: Sometimes.humans are used to study discrimination learning, and sometimes other species are used to study the acquisition of concepts.

Another common difference be- tween the two types of experiment is

STIMULUS CONTROL 263

probably a result of the first. Because an entire set of verbal responses can readily and quickly be established with members of the human species, human subjects may be asked to acquire sev- eral different concepts at the same time. This practice enables the experi- menter to collect more data from each subject. When other species are used, however, it takes much longer to estab- lish each response topography, and the use of any substantial number of cate- gories becomes impractical, even in those experiments designated as stud- ies of concept formation. To respond or not to respond is usually the question. A third feature, which may also re-

late to the historic use of human sub- jects, is that in experiments on concept formation the distinction between the positive and the negative stimulus is often very subtle. Human subjects learn simple, gross discriminations so rapidly that variations in their perfor- mance created by manipulating some independent variable might be too small to detect, and subtle differences are necessary to collect meaningful data. In this case, the historical feature may have been extended to serve as an implicit definition for classifying the two types of research: Even when non- human species are used, subtlety of the properties distinguishing between the positive and the negative instances (e.g., Herrnstein, Loveland, & Cable, 1976) appears to be characteristic of those experiments that are considered to be examples of concept formation. But this is a difference in a parametric value, not a difference in the underly- ing process.

Finally, the most meaningful differ- ence between discrimination and con- cept formation may be that in experi- ments on discrimination, all aspects of the environment other than the alter- nation between the positive and the negative stimulus are kept as uniform as possible (“held constant”), so that they will produce the least possible fluctuation in the results. In experi- ments on concept formation, by con- trast, other aspects of the stimulus ob-

ject are deliberately varied. In this re- spect, experiments on concept forma- tion are more closely related to what is colloquially called “real life.” They provide us with a persuasive bridge be- tween the stripped down simplicity of the laboratory and the rich complexity of the natural setting.

IMITATION

There is another form of stimulus control that is not well understood but that merits consideration because it is believed to play an enormous role in training the young of many species, in- cluding our own, to behave like their parents and other adults. The observa- tion that leads us to speak of imitation is that the stimulus setting the occasion for a particular response or set of re- sponses is the performance of approx- imately the same pattern of behavior by another individual, usually of the same species. Such a correspondence between two patterns of behavior is so common that it is frequently assumed that it must be the result of some in- trinsic connection between the behav- ior of the second organism and that of the first. But not all behavior is imitat- ed. In most cases in which imitation is said to occur, the correspondence be- tween the two sets of behavior appears to be historical in origin. It is a product of the contingencies of reinforcement.

It has sometimes been suggested (e.g., Bandura, 1965) that imitation is an alternative mode of learning, by means of which a response can be es- tablished without the use of reinforce- ment (“no-trial learning”). In some sense, at a global level, this may be true, but it is not clear that any new principle of behavior is required to ex- plain Bandura’s results.

In exploring this issue, it may be helpful to distinguish between a com- plex pattern of physical activity like swimming the breast stroke, riding a bicycle, driving an automobile, or per- forming a new dance step, and the con- stituent movements that make it up (see Guthrie, 1952, pp. 27-28, or Skin-

264 JAMES A. DINSMOOR

ner, 1953, p. 94). Modeling is akin to verbal instruction, and, indeed, might be considered another form of the same process. If the constituent responses are already a part of the individual’s repertoire (by prior differentiation) and some kind of imitative control has al- ready been established, an appropriate sequence or an effective combination of responses can often be called forth by the corresponding behavior of a model. Sometimes this is all that is needed. In other cases, it may still serve as the initial step: Once some ap- proximation to the appropriate behav- ior has occurred, the overall pattern can be maintained and can subsequent- ly be refined by selective reinforce- ment. But if the constituent responses are not in the individual’s repertoire to begin with, they cannot be established simply by imitation. Who among us has not struggled in vain to match the sounds that have been demonstrated to us by the teacher of a foreign language, only to find that the closest approxi- mations within our vocal repertoire leave much to be desired? Who among us has not tried to copy someone else’s drawing, only to discover that much more training was necessary before a satisfactory replica could be produced? The process by which the correspon-

dence between stimulus and response is originally established was illustrated more than half a century ago in an ex- periment by Miller and Dollard (1940). All training was conducted on a special maze, shaped like the letter T. First, several “leader” rats were trained to discriminate between black cards and white cards, half of them to enter the arm of the maze with the black card and half to enter the arm with the white. The only function of these lead- er rats, however, was to provide dis- criminative stimuli for the rats that were to follow them. When the leader rats had learned to

turn in the direction indicated by the appropriate card, they were used to train the “follower” rats to imitate a specific item of behavior. First, a leader was placed in the start box at the be-

ginning of the stem, with the follower in a second box immediately behind it. When the leader was released from its start box, it promptly ran to the choice point at the intersection of the T and turned in the designated direction. At the end of the arm, it received its usual allotment of food. The follower rat was released immediately behind it. If the follower rat turned in the same direc- tion as its leader, its response was re- inforced with food in a special recep- tacle uncovered only after the leader rat had passed over it. If the follower rat turned in the other direction, how- ever, its response was not reinforced.

For the follower rat, then, the leader turning to the right was a discrimina- tive stimulus for turning to the right, and the leader turning to the left was a discriminative stimulus for turning to the left. As might be expected, the fol- lower rats had no difficulty in master- ing this discrimination. With respect to this one response, their behavior matched or imitated the behavior of their leaders. However, in terms of learning principles, there was nothing unique about the resulting correspon- dence between the behavior of the 2 animals. To demonstrate this point, Miller and Dollard (1940) trained an- other group of follower rats in a pattern of stimulus control that was exactly the opposite of imitation: Turn left when the leader turns right and turn right when the leader turns left. The two groups learned their tasks with equal facility. Following a partner in ball- room dancing may provide an illustra- tion of this reversed relationship, but such a pattern of stimulus control is not commonly reinforced in the outside world.

Apparently there is more to imita- tion, however, than establishing dis- criminative control over a single re- sponse. When a number of correspon- dences have been reinforced between the actions of an experimental subject and the actions of a model, the corre- spondence itself may become a gov- erning factor in the relation between the two actions, extending to new to-

STIMULUS CONTROL 265

pographies of behavior. Baer, Peterson, and Sherman (1967) demonstrated this with 3 profoundly retarded children who originally showed no tendency to imitate. In the beginning, the experi- menters used manual guidance and se- lective reinforcement with food to es- tablish a series of physical actions like raising the left arm, tapping a table, or moving the arm in a circular motion, each of which followed a like action by the experimenter. One of the things these authors

showed was that the behavioral corre- spondence could itself be placed under the control of some other stimulus in what was therefore known as a condi- tional discrimination. In their study, re- inforcement of the child’s action was conditional upon the prior presentation of the verbal stimulus, “Do this,” fol- lowed by a demonstration of the de- sired response. Stimuli in the presence of which correspondence is reinforced determine when imitation will occur and when it will not occur. In some studies, the choice of a model has de- termined when correspondence would be reinforced.

In the Baer et al. (1967) study, 130 different responses were employed. Eventually the children began to imi- tate new actions, demonstrated for the first time. Some of these were repeated on a number of test trials, still without reinforcement, interspersed among training trials on which other actions were reinforced. The performance of these nonreinforced actions continued to depend, however, on the general ten- dency to perform responses demon- strated by the experimenter. When re- inforcement of the main body of re- sponses was replaced by differential reinforcement of other behavior, the never-reinforced test actions also dropped out. Both sets of behavior were restored with restoration of the original reinforcement schedule.

Similarity to the behavior of the model may also serve as a conditioned reinforcer. In another experiment, Lo- vaas, Berberich, Perloff, and Schaeffer (1966) reinforced the production of

English words by each of 2 autistic children, in response to those same words presented by an experimenter. Then some Norwegian words were slipped into the sequence, without re- inforcement; with continued reinforce- ment of the English words, the pronun- ciation of the Norwegian words grad- ually improved. These data suggest that the increasing similarity between the sounds produced by the child and the sounds presented by the experi- menter had itself become reinforcing, leading to successive approximations of those sounds by the children.

STIMULUS EQUIVALENCE

In his presidential address to the Midwestern Psychological Associa- tion, Skinner (1950) included brief re- sumes of several pieces of work con- ducted in his laboratory that were nev- er subsequently reported in greater de- tail. Among these was a procedure that he called “matching to sample,” pre- sented in an attempt to strip some of the surplus meaning from what would nowadays be called “cognitive” de- scriptions of choice and other complex patterns of behavior. The pigeon was confronted with

three keys, arranged in a horizontal row. First, on a given trial, the middle key was illuminated with a sample col- or, perhaps red or green. When the bird pecked that key, the sample color was extinguished and the keys on either side were lighted with the comparison colors. One of these colors was the same as the sample, and the other was different. In a strict matching-to-sam- ple procedure, it was a peck on the same-color key that was reinforced. In what later came to be known as oddity matching, it was a peck on the differ- ent-color key that was reinforced. With a delay introduced between the dark- ening of the center key and the lighting of the side keys, Skinner’s procedure was widely used by more cognitively oriented psychologists to analyze the processes involved in responding to past events (memory).

266 JAMES A. DINSMOOR

Years later, Sidman (1971, 1994) adapted Skinner’s procedure to the task of teaching a severely retarded boy to read English text. Prior to the proce- dure in question, this boy had already learned correspondences in both direc- tions between spoken words and pic- tures. That is, he had learned to choose the pictures that illustrated the words presented as sample stimuli and to pro- duce the correct vocal responses to (name) the pictures. During the train- ing procedure, the sample stimuli were words spoken by the experimenter, and the comparison stimuli were printed words. Somewhat to Sidman’s surprise, after his subject learned to choose the printed word that corresponded to each spoken word, not only was he able to proceed in the opposite direction, pro- ducing the correct oral responses to printed text (reading aloud) but, with- out further training, was able to choose pictures corresponding to the printed words (reading comprehension). Estab- lishing an equivalence between the au- ditory and the visual forms of a series of words had extended the subject’s ability to select the correct pictures from the original spoken words to their printed equivalents.

Struck by these initial findings, Sid- man launched a major program of re- search designed to explore their theo- retical ramifications (Sidman, 1994). Because there is now a burgeoning lit- erature on the topic, it may be impor- tant to consider some terminological issues. First, Sidman objected to the use of the term matching to sample as a description of Skinner’s procedure, on the grounds that this phrase implied a generalized behavioral outcome that did not necessarily result from that pro- cedure. In its place, he substituted the phrase conditional discrimination, first applied by Cumming and Berryman (1965). Generically, the term condi- tional discrimination covers more than the matching-to-sample paradigm (see Yarczower, 1971), but in the Cumming and Berryman usage an additional, ex- traneous stimulus (e.g., a red sample or a green sample) determines which of

the alternative stimuli (in this case, red or green comparison stimuli) in a dis- crimination is positive (i.e., followed by reinforcement of the response). The relation between comparison stimulus and reinforcer is selected by or condi- tional upon the nature of the sample stimulus. When the sample stimulus is red, pecking red is reinforced and pecking green is not reinforced; when the sample is green, pecking red is not reinforced and pecking green is rein- forced.

Sidman (1994) has also suggested the use of the term conditional stimu- lus to refer to the role of the sample stimulus in this type of discrimination, but such a usage could lead to termi- nological confusion, because that term is widely used in Pavlovian condition- ing to refer to the stimulus that ac- quires its effectiveness through its tem- poral relation to the unconditional stimulus. The terms instructional stim- ulus (Cumming & Berryman, 1965) or contextual stimulus (Sidman, 1994) have also been suggested; the former term has intuitive appeal, and the latter term has the advantage of emphasizing that stimuli and responses usually do not pair off in a simplistic one-to-one correspondence, as sometimes implied in early writings by Watson (1919), for example, but are characteristically re- lated in a way that depends on what other stimuli are present (context).

Although an exposition of the nature of the procedure we are considering has been greatly simplified by the use of physically identical colors as ex- emplars, in establishing equivalence classes the experimenter utilizes stim- uli that bear no necessary resemblance to each other and establishes their cor- respondence through the program of training. A quantity of objects, for ex- ample, and the corresponding numeral begin as arbitrary, unrelated stimuli, but the standard relations characteristic of a given language can be established through this type of training. Eight ob- jects can be made equivalent to the digit “8” and the binary number “1000” and the sound “ate” and the

STIMULUS CONTROL 267

printed word “eight” and even the Spanish word “ocho,” thus forming a class, all of whose members may be said for our purposes to be equivalent. When they become members of the same class, they are for many purposes interchangeable. If we respond in the same way to these physically very dif- ferent stimuli, they may be said to have the same meaning.

Sidman (1994) has laid out three cri- teria, all of which are necessary to the mathematical definition of an equiva- lence relation among a set of stimuli. Reflexivity requires that, without spe- cific instruction or training, the subject choose each stimulus in the list as a comparison in reaction to that same stimulus as a sample (i.e., generalized identity matching). Symmetry requires that the subject perform correctly when the roles of sample and comparison stimuli are reversed. Transitivity in- volves three stimuli in a sequence. Once “if a, then b” and “if b, then c” have been established, then “if a, then c” must emerge without further in- struction or training. The apparent significance of equiv-

alence classes for what are sometimes known as higher mental processes is currently attracting a considerable amount of attention within the behav- ior-analytic community. Quite a bit of research has recently been reported, ac- companied by a large amount of theo- rizing. However, it remains a relatively new field of investigation, still in flux, and the available information does not as yet permit us to be sure how it fits into a broader, more systematic frame- work. One active program of research

stems from Fields and Verhave’s (1987) analysis of the parameters that define the internal structure of an equivalence class. In a review of the relevant literature, Fields, Adams, and Verhave (1993) examined the effects of two of these parameters. Directionality refers to whether a stimulus serves as a sample or a comparison in the pro- cess of training. In several studies, this relation was found to influence the

likelihood of class formation, the for- mal characteristics of specific emergent relations, and the degree of transfer be- tween stimuli. Nodes are formally de- fined as individual stimuli that are linked by the program of training to more than one other stimulus within the particular equivalence class; they may be thought of as intervening steps or mediators of the relation between stimuli that were not presented togeth- er during the training. And nodal dis- tance refers to the number of such steps that lie between the stimuli that are to be tested. In a number of studies, the authors found, this parameter had a consistent influence on the subjects’ performances on a number of different types of test. A point of emphasis in their review was that functions ac- quired by one stimulus in an equiva- lence class do not transfer equally, but rather differentially, to other members of the class. The relatedness of the stimuli is affected by the directionality of training and is an inverse function of nodal distance. The relation between the formation

of equivalence classes and earlier, sim- pler forms of discrimination learning remains the subject of wide-ranging discussion. Still in dispute, for exam- ple, is the question of whether emer- gent relations and stimulus classes can be demonstrated with nonhuman sub- jects (see Schusterman & Kastak, 1993; Zentall & Urcuioli, 1993). Sid- man (1994) has suggested that the for- mation of equivalence classes through conditional discrimination training may not be reducible to or explicable by other processes with which we are already familiar but may be a wholly independent phenomenon grounded in the evolution of the human species. For those of us in search of systematic laws of behavior, this is not a very satisfying solution. Hayes (1991), on the other hand, has suggested that equivalence classes can be traced to a preexperi- mental history of training in relational responding. At the present time, such issues are far from being resolved.

268 JAMES A. DINSMOOR

REFERENCES Anthony, T., Copper, C., & Mullen, B. (1992).

Cross-racial facial identification: A social cog- nitive integration. Personality and Social Psy- chology Bulletin, 18, 296-301.

Baer, D. M., Peterson, R. E, & Sherman, J. A. (1967). The development of imitation by re- inforcing behavioral similarity to a model. Journal of the Experimental Analysis of Be- havior, 10, 405-416.

Bandura, A. (1965). Vicarious processes: A case of no-trial learning. In L. Berkowitz (Ed.), Advances in experimental social psy- chology (Vol. 2, pp. 1-55). New York: Aca- demic Press.

Blanchard, R. (1977). Control of observing by antecedent stimuli. Learning and Motivation, 8, 569-580.

Blough, D. S. (1969). Attention shifts in a maintained discrimination. Science, 166, 125- 126.

Cook, D. A. (1960). On vanishing stimuli in instructional material. Journal of the Experi- mental Analysis of Behavior, 3, 292.

Cumming, W. W., & Berryman, R. (1965). The complex discriminated operant: Studies of matching-to-sample and related problems. In D. I. Mostofsky (Ed.), Stimulus generalization (pp. 284-330). Stanford, CA: Stanford Uni- versity Press.

Dinsmoor, J. A. (1983). Observing and condi- tioned reinforcement. Behavioral and Brain Sciences, 6, 693-704.

Dinsmoor, J. A., Bowe, C. A., Dout, D. L., Mar- tin, L. T., Mueller, K. L., & Workman, J. D. (1983). Separating the effects of salience and disparity on the rate of observing. Journal of the Experimental Analysis of Behavior, 40, 253-264.

Dinsmoor, J. A., Mueller, K. L., Martin, L. T., & Bowe, C. A. (1982). The acquisition of observing. Journal of the Experimental Anal- ysis of Behavior, 38, 249-263.

Dinsmoor, J. A., Sears, G. W., & Dout, D. L. (1976). Observing as a function of stimulus difference. Journal of Experimental Psychol- ogy: Animal Behavior Processes, 2, 154-162.

Fields, L. (1978). Fading and errorless transfer in successive discriminations. Journal of the Experimental Analysis of Behavior, 30, 123- 128.

Fields, L., Adams, B. J., & Verhave, T. (1993). The effects of equivalence class structure on test performances. Psychological Record, 43, 697-712.

Fields, L., & Verhave, T. (1987). The structure of equivalence classes. Journal of the Exper- imental Analysis of Behavior, 48, 317-332.

Guthrie, E. R. (1952). The psychology of learn- ing (Rev. ed.). New York: Harper & Row.

Hanson, H. M. (1959). Effects of discrimination training on stimulus generalization. Journal of Experimental Psychology, 58, 321-334.

Hayes, S. C. (1991). A relational control tieory of stimulus equivalence. In L. J. Hayes & P

N. Chase (Eds.), Dialogues on verbal behav- ior (pp. 19-40). Reno, NV: Context Press.

Herrnstein, R. J., Loveland, D. H., & Cable, C. (1976). Natural concepts in pigeons. Journal ofExperimental Psychology: Animal Behavior Processes, 2, 285-302.

Holland, J. G. (1960). Teaching machines: An application of principles from the laboratory. Journal of the Experimental Analysis of Be- havior, 3, 275-287.

Honig, W. K., Thomas, D. R., & Guttman, N. (1959). Differential effects of continuous ex- tinction and discrimination training on the generalization gradient. Journal of Experi- mental Psychology, 58, 145-152.

James, W. (1890). Principles of psychology. New York: Holt, Rinehardt, & Winston.

Johnson, D. F (1970). Determiners of selective stimulus control in the pigeon. Journal of Comparative and Physiological Psychology, 70, 298-307.

Kamin, L. J. (1968). “Attention-like” processes in classical conditioning. In M. R. Jones (Ed.), Miami symposium on the prediction of behav- ior, 1967: Aversive stimulation (pp. 9-3 1). Coral Gables, FL: University of Miami Press.

Keller, F S., & Schoenfeld, W. N. (1950). Prin- ciples ofpsychology: A systematic text in the science of behavior. New York: Appleton- Century-Crofts.

Lawrence, D. H. (1952). The transfer of a dis- crimination along a continuum. Journal of Comparative and Physiological Psychology, 45, 511-516.

Lovaas, 0. I., Berberich, J. P., Perloff, B. F, & Schaeffer, B. (1966). Acquisition of imitative speech by schizophrenic children. Science, 151, 705-707.

Mackintosh, N. J. (1974). The psychology of animal learning. London: Academic Press.

Mackintosh, N. J., & Reese, B. (1979). One- trial overshadowing. Quarterly Journal of Ex- perimental Psychology, 31, 519-526.

Miles, C. G., & Jenkins, H. M. (1973). Over- shadowing in operant conditioning as a func- tion of discriminability. Learning and Moti- vation, 4, 11-27.

Miller, N. E., & Dollard, J. (1940). Social learning and imitation. New Haven, CT: Yale University Press.

Montessori, M. (1912). The Montessori method. New York: Frederick A. Stokes.

Pavlov, I. P (1960). Conditioned reflexes: An investigation of the physiological activity of the cerebral cortex (G. V. Anrep, Trans.). New York: Dover. (Original work published 1927)

Pierrel, R., Sherman, J. G., Blue, S., & Hegge, F W. (1970). Auditory discrimination: A three-variable analysis of intensity effects. Journal of the Experimental Analysis of Be- havior, 13, 17-35.

Rescorla, A. A., & Wagner, A. R. (1972). A theory of Pavlovian conditioning: Variations in the effectiveness of reinforcement and non- reinforcement. In A. H. Black & W. F Prokasy (Eds.), Classical conditioning: Vol. 2. Current

STIMULUS CONTROL 269

research and theory (pp. 64-99). New York: Appleton-Century-Crofts.

Revusky, S. H. (1971). The role of interference in association over a delay. In W. K. Honig & P. H. R. James (Eds.), Animal memory (pp. 155-213). New York: Academic Press.

Schusterman, R. J., & Kastak, D. (1993). A California sea lion (Zalophus californianus) is capable of forming equivalence relations. Psy- chological Record, 43, 823-839.

Sidman, M. (1971). Reading and auditory-vi- sual equivalences. Journal of Speech and Hearing Research, 14, 5-13.

Sidman, M. (1994). Equivalence relations and behavior: A research story. Boston: Authors Cooperative.

Skinner, B. F (1950). Are theories of learning necessary? Psychological Review, 57, 193- 216.

Skinner, B. F (1953). Science and human be- havior. New York: Macmillan.

Skinner, B. F (1958). Teaching machines. Sci- ence, 128, 969-977.

Skinner, B. F (1969). Contingencies of rein- forcement: A theoretical analysis. New York: Appleton-Century-Crofts.

Terrace, H. S. (1963). Errorless transfer of a discrimination across two continua. Journal of the Experimental Analysis of Behavior, 6, 223-232.

Touchette, P E. (1971). Transfer of stimulus control: Measuring the moment of transfer. Journal of the Experimental Analysis of Be- havior, 15, 347-354.

Trabasso, T. (1963). Stimulus emphasis and all- or-none learning in concept identification. Journal ofExperimental Psychology, 65, 398- 406.

vom Saal, W., & Jenkins, H. M. (1970). Block- ing the development of stimulus control. Learning and Motivation, 1, 52-64.

Wagner, A. R., Logan, E A., Haberlandt, K., & Price, T (1968). Stimulus selection in animal discrimination learning. Journal of Experi- mental Psychology, 76, 171-180.

Watson, J. B. (1919). Psychology from the standpoint ofa behaviorist. Philadelphia: Lip- pincott.

Yarczower, M. (1971). Stimulus control during conditional discrimination. Journal of the Ex- perimental Analysis of Behavior, 16, 89-94.

Yarczower, M., & Switalski, R. (1969). Stimu- lus control in the goldfish after massed ex- tinction. Journal of the Experimental Analysis of Behavior, 12, 565-570.

Zentall, T. R., & Urcuioli, P. J. (1993). Emer- gent relations in the formation of stimulus classes by pigeons. Psychological Record, 43, 795-810.