Chapter 11
BEGINNINGS
In the harsh, icy land of the Canadian Arctic, hunting seals was a basic survival skill for the Inuit people. Before the arrival of Europeans and their rifles, the Inuit hunted with spears. To make a kill, they had to get close to their prey, which was not easy because there is little cover in the Arctic. Seals could see a man long before he posed a threat and easily slip down an ice hole or off the edge of an ice sheet and disappear into the ocean. The hunter’s solu- tion was to approach while imitating the seals. The hunter crouched low and crept forward, occasionally raising his head as the seals themselves do. Pro- ceeding slowly, if the hunter were lucky he got close enough to a seal to drive home his spear.
Edward Maurice (2005), who lived among the Inuit in the early days of the 20th century, tells us that one day a little Inuit boy was playing in the village. He pretended to be a hunter in pursuit of seals. He lay on the ground and crept along, now and then raising his head to look around, just as a hunter would do in imitation of a seal. Unfortunately, his play-hunting adventure ended tragically: The village sled dogs saw the boy and, evidently taking him for a seal, attacked and killed him.
This sad story illustrates three important learning concepts: generalization, discrimination, and stimulus control. Much of the research on these topics is done with pigeons and rats and involves pecking discs and pressing levers. Such experiments are essential to understanding these basic phenomena, but to students they often seem remote from everyday experience. As you read about these experiments, keep in mind the Inuit hunters, the seals, and the tragic story of the boy on the ice.
GENERALIZATON
Generalization is the tendency for the effects of a learning experience to spread. It is sometimes called transfer, since the effects of a learning experience “move.” (You get on bus 39, then transfer to bus 42.) Researchers have identi- fied four kinds of generalization (Cooper, Heron, & Heward, 2007), depending on where the learning moves:
Generalization across people (which might be called vicarious general- ization; see Chapter 10)
Generalization across time (also called response maintenance) could be considered the opposite of forgetting (see Chapter 12)
Generalization across behaviors (known as response generalization)
Generalization across situations (known as stimulus generalization)
A good deal of Chapter 10, on observational learning, was devoted to the first kind of generalization, the generalization of the learning experiences of a model to those of an observer. The following chapter, on forgetting, will deal largely with the second kind of generalization, the generalization of behavior over time. The third type, response generalization, is the tendency for changes in one behavior to spread to other behaviors. If a rat receives food after pressing a lever with its right front foot, it might then press the lever with its left front foot, or with its chin. Similarly, if a child is rewarded for expressing a willingness to share a toy, she is then more likely to actually share a toy (Barton & Ascione, 1979). All three of these kinds of general- ization are important, but the one that gets the most attention from both basic and applied researchers is stimulus generalization so we will consider it in some detail. From now on, when I use the term generalization, I mean stimulus generalization.
Stimulus generalization is the tendency for changes in behavior in one situation to spread to other situations. It is often defined as the tendency to respond to stimuli not present during training. This definition is perhaps a better fit with laboratory measures of generalization, where the environ- ment is made constant except for one or two features, such as a light or a tone. But stimuli never really occur alone; they are always part of a context, a situation. So it is fair to say that with generalization, what we learn in one situation carries over, or transfers to, a different situation. We could say that in generalization the behavior “travels” from one place to another. Some examples may clarify the phenomenon. In Pavlovian conditioning, a dog may learn to salivate to the sound of a tuning fork vibrating at 1,000 cycles per second (cps). After this training, the dog may then salivate to the sound of a tuning fork vibrating at, say, 900 cps to 1,100 cps, even though it was never exposed to these stimuli. The conditional response spreads, or gen- eralizes, to stimuli different from the CS.
The famous Watson and Rayner (1920) study (see Chapter 4) provides another example of the generalization of a conditional response. You will recall that Little Albert learned to fear a white rat. After establishing this fear, Watson and Rayner tested Albert to see whether other stimuli would frighten him. They presented Albert with a white rabbit, raw cotton, and a Santa Claus mask. None of these stimuli had been around when Albert learned to fear the rat, yet Albert was afraid of them, too. Albert’s fear had spread, or generalized, from the white rat to other white, furry objects.
Carl Hovland (1937) studied the generalization of fear conditioning in college students. He began by pairing a tone of a particular pitch with a mild electric shock; the UR was the galvanic skin response, or GSR (a measure of emotional arousal). After 16 pairings of the CS and US, Hovland then pre- sented four tones, including the CS. The results showed that the GSR spread from the original tone to the others; the less a stimulus resembled the CS, the weaker the CR was. When data on stimulus generalization are plotted on a graph, they yield a figure called a generalization gradient (Figure 11-1). Perhaps the first report of generalization following reinforcement came from Thorndike (1898). He observed that “a cat that has learned to escape from [box] A by clawing has, when put into [box] C or G, a greater tendency to claw at things than it instinctively had at the start” (p. 14). In other words, clawing generalized from box A to boxes C and G.
Those who followed Thorndike studied generalization in a more rigorous manner. In a classic study, Norman Guttman and Harry Kalish (1956) trained pigeons to peck a disc of a particular color and later gave them the opportunity to peck the disc when it was various colors, including the color used in training, for 30 seconds each. Pigeons pecked the disc most frequently when it was the color used during training, but they also pecked the disc when it was other colors. As the generalization gradient reveals, the more closely the disc resembled the training disc, the more often the birds pecked it. If a disc was almost the same color as the training disc, the birds pecked at it almost as much as if it were the training disc; if the disc was a very different color, the pigeons seldom touched it (see Figure 11-2).
Generalization is not restricted to highly specific acts such as disc pecking. Broader behavioral tendencies also generalize. You might recall from Chapter 6 that Robert Eisenberger and his colleagues found that rewarding people for making a strong effort on one task increases the level of effort made on other tasks, a phenomenon they call learned industriousness. This means that trying hard, if reinforced in one situation, may generalize to another situation, even though “trying hard” is not a specific act.
Studies of generalization usually involve the effects of reinforcement, but changes in behavior produced by extinction also spread beyond the learning situation. For example, R. E. P. Youtz (reported in Skinner, 1938) trained rats to press a horizontal lever for food and then put the behavior on extinction. After this, he tested the rats in a chamber with a vertical lever. He found that the effects of the extinction procedure reduced the rats’ tendency to press the new lever. Youtz trained other rats to press the vertical lever, then put the behavior on extinction and tested them on the horizontal lever. Again, he found that the effects of extinction spread to the new situation. Overall, the extinction procedure reduced the tendency to perform in a similar situation by 63%.
The suppression of behavior produced by punishment spreads in much the same way as the effects of reinforcement and extinction. Werner Honig and Robert Slivka (1964) trained pigeons to peck discs of various colors. When the birds were pecking all the colors at the same rate, the experimenters continued reinforcing disc pecks but also punished pecking whenever the disc was a particular color. The tendency to peck the disc when it was that color declined, of course, but so did the tendency to peck when the disc was other colors. The frequency of pecking varied systematically with the similarity of the disc to the punished color. Thus, the effects of punishment formed a generalization gradient like those seen following reinforcement and extinc- tion. Essentially the same phenomenon has been demonstrated in humans (O’Donnell & Crosbie, 1998). Generalization is a common phenomenon, but it cannot be taken for granted (Birnbrauer, 1968; Miller & Sloane, 1976; Wolf et al., 1987). For ex- ample, D. E. Ducharme and S. W. Holborn (1997) conducted a social skills training program with five hearing-impaired preschool children. The training program produced high, stable rates of social interaction in that setting, but the social skills did not generalize much to other settings.
Fortunately, researchers have identified ways of increasing the gener- alization of training effects (Baer, 1999; Cook & Mayer, 1988; Francisco & Hanley, 2012; Stokes & Baer, 1977; Stokes & Osne, 1989). One way is to pro- vide training in a wide variety of settings. If you want a pigeon to peck discs no matter what their color, reinforce disc pecking of a wide variety of colors. A related idea is to provide lots of examples. Another tactic is to vary the consequences. If you are reinforcing a behavior, vary the kind, amount, and schedule of reinforcers. Another tactic is to reinforce generalization when it occurs (see Generalized Therapy).The research on generalization and its enhancement has important implications in a variety of practical situations. Many educators assume that once a student understands a principle, such as the Pythagorean theorem or the refraction of light by water, the student should then be able to apply it in any situation. But the fact that a student can apply the Pythagorean theorem to a geometry problem posed by a teacher in a math class does not mean that he or she will apply it when constructing a toolbox in a garage. Managers make the same mistake when instructing workers in the operation of equipment or in following safety procedures. Parents also need to understand that just because their child has learned to look both ways before crossing a street does not mean they will do this when chasing a ball that rolls into the street (Miltenberger, 2009).
Doing things to increase the generalization of learning is important, but generalization is not always a desirable thing. A behavior that is useful in one situation is not always helpful in another. Thorndike (1898) noticed, for example, that a cat that had learned to escape from a box by pulling on a loop would later paw at the same spot—even though the loop had been removed! In the same way, a college student whose off-color jokes get big laughs in the dormitory may find that the same jokes are not appreciated at the family dinner table.
Carol Dweck and Dickon Repucci (1973) showed how generalization can work against a teacher and her students. Teachers first gave students unsolvable problems. Later these teachers gave the students problems that could be solved, but the students failed to solve them. The tendency to give up seems to have generalized from the first situation to the second. When a different teacher gave the students solvable problems, they were successful.
Generalization can also make problem behaviors more troublesome than they would otherwise be. For instance, if punching a large inflated doll is reinforced, children later tend to be more aggressive when interacting with other children (Walters & Brown, 1963). There is a substantial difference be- tween an inflated doll and a child, yet the behavior generalized from one to the other.
Sometimes the results of generalization are tragic. Every year in America’s national parks people are injured by bears. Most of these cases probably in- volve generalization. For example, bears will sometimes attack people when the people get near a blueberry patch or other source of food. The bear may have driven off bears and deer that compete with it for food, and this behav- ior may then generalize to people, though people are not usually competing for the bear’s food. The dogs that attacked the Inuit boy mentioned earlier were also generalizing. These dogs would readily attack a real seal and kill it, and this behavior generalized to the boy who pretended to be a seal. Humans also make tragic generalization errors. Many bear attacks occur when people approach wild bears as though they were pets. One man wanted to get a pho- tograph of his young daughter with a bear, so he urged her to go up to the bear with some food. The girl was lucky that she lost only a finger. Several years ago I read a newspaper item about the death of a skydiver. His parachute did not open, but there was nothing wrong with it. It turned out that he was us- ing a new parachute, and the release cord was not on the same side as it was on his old chute. As he fell to his death he tore at the chute in the area where the pull cord used to be. His behavior had generalized from the old suit to the new one.
Generalization also appears to be involved in some hate crimes. Following the horrific attacks against the United States by Arab extremists on September 11, 2001, many people of Arab descent living in the United States were assaulted. No doubt in most instances the victim’s only crime was physically resembling those who had committed crimes. Their attackers likely would have preferred to assault those involved in the 9/11 attacks, but those people were not available, so they attacked those who resembled them. Neal Miller (1948b) long ago provided evidence that such displaced aggression is due to stimulus generalization.
You can see that generalization is an important phenomenon, but it is only one side of a coin. The other side is discrimination.
DISCRIMINATION
Stimulus discrimination is the tendency for behavior to occur in certain situ- ations but not in others. It is often defined more narrowly as the tendency for behavior to occur in the presence of certain stimuli, but not in their absence. However, as mentioned earlier, stimuli are always part of a context, so the word situation is appropriate. You can see that discrimination and generalization are inversely related: The more discrimination, the less generalization, and vice versa. Generaliza- tion gradients therefore reflect the degree of discrimination. A relatively flat generalization gradient indicates little or no discrimination; a steep gradient indicates considerable discrimination (see Figure 11-3).
We saw earlier that the more a stimulus resembles the training stimulus, the greater will be the degree of generalization. It is therefore clear that the less similar a stimulus is to the training stimulus, the greater will be the de- gree of discrimination. A pigeon that has learned to peck a yellow disc for food is very likely to peck discs of similar color, but not so likely to peck a black disc. It is often possible, however, to establish a discrimination between very similar stimuli through discrimination training.
Any procedure for establishing a discrimination is called discrimination training. Discrimination can be established through both Pavlovian and operant procedures. In Pavlovian discrimination training, one conditional stimulus (designated CS+) is regularly paired with a US, and another (designated CS−) regularly appears alone. For example, we might put food into a dog’s mouth each time a buzzer sounds and give the dog nothing when a bell rings. The result will be that the dog salivates at the sound of the buzzer (the CS+) but not at the sound of the bell (CS−). At this point, we say that the dog discriminates between the buzzer and the bell—that is, it behaves differently in the two situations.
Pavlov (1927) conducted many experiments on discrimination training. In one, a dog saw a rotating object. Whenever the object rotated in a clockwise direction, the dog received food; when it rotated in the opposite direction, the dog got nothing. The dog soon discriminated: It salivated at the CS+ (clockwise rotation) but not at the CS− (counterclockwise rotation). Other experiments yielded similar results. Pavlov’s dogs learned to discriminate between different volumes of a particular sound, different pitches of a tone, different geometric forms, different temperatures, and so on. Sometimes the level of discrimina- tion achieved was remarkable. One dog learned to salivate at the sound of a metronome when it ticked at the rate of 100 beats a minute but not when it ticked at the rate of 96 times a minute.
In operant discrimination training, one stimulus, designated S+ or SD (pro- nounced ess-dee) typically indicates that a behavior will have reinforcing con- sequences, and another stimulus, S− or S∆ (pronounced ess-delta) indicates that the behavior will not have reinforcing consequences. SD and S∆ are both discriminative stimuli—that is, stimuli that signal different consequences for a behavior. To illustrate: We might arrange an experimental chamber so that a rat receives food each time it presses a lever, but only if a lamp is on. The result will be that when the lamp is on (SD), the rat presses the lever, and when the lamp is off (S∆) it does not press. At this point, we say that the rat discrimi- nates between the two situations. Discrimination training can take various forms. In simultaneous discrimi- nation training, the discriminative stimuli are presented at the same time. In a classic experiment, Karl Lashley (1930) put a rat on a small platform before two cards, placed side by side (see Figure 11-4). The cards are distinguishable in some way; for example, one may have horizontal stripes, the other vertical. One of the cards is locked in place; if the rat jumps at that card, it will then fall to the net below. The other card is not locked, so if the rat jumps toward it the card gives way and the rat lands on the other side. Whichever choice the rat makes, it is not hurt and is returned to the jumping stand for another trial. The difference is that if it jumps toward the correct card, it receives food. The position of the cards varies randomly so that the correct card will sometimes be on the right and sometimes on the left. The rat soon learns to jump toward the card that results in food.
In successive discrimination training, the SD and S∆ alternate, usually ran- domly. When the SD appears, the behavior is reinforced; when the S∆ appears, the behavior is not reinforced. Donald Dougherty and Paul Lewis (1991) used the successive procedure in doing one of the few discrimination training stud- ies in horses. The horses learned to press a lever (the same kind used with rats) using their lip. The researchers projected images of circles, one at a time, in front of the horses. One circle was 2.5 inches in diameter, the other 1.5 inches. Lever pressing when the larger circle appeared resulted in grain; pressing when the smaller circle appeared got the horse nothing. The horses learned to dis- criminate between the two circles. One horse, Lady Bay, learned to discrimi- nate very quickly (see Figure 11-5).
In a procedure called matching to sample (MTS), the task is to select from two or more alternatives (called comparison stimuli) the stimulus that matches a standard (the sample). The comparison stimuli include the SD —the stimulus that matches the sample—and one or more S∆. For example, a sample disc on one wall of an experimental chamber may be illuminated by either a red or a green light. On some trials the disc will be red, on some trials green. After a short time the sample disc goes dark, and two comparison discs, one red and one green, are illuminated. The SD + is the disc of the same color as the sample. If a pigeon pecks the comparison disc matching the sample, it receives food; if it pecks the other disc, it receives nothing. To obtain reinforcement, the bird must successfully discriminate between the disc that matches the sample (the SD) and the one that does not (the S∆).
The example of MTS just given is very simple, but the procedure can be more complicated. For example, a bird may be required to peck a disc that is different from the sample, a variation of MTS called oddity matching or mismatching. Increasing the number of variations in the sample and/or the comparison discs also may complicate the MTS procedure. For example, the sample may alternate among red, green, and blue, and the comparison discs may be red, green, blue, and yellow.
In the procedures described thus far, the animal or person undergoing training inevitably makes a number of mistakes. If you think about being a rat on Lashley’s jumping stand, you can see that there is no way you can tell which card to jump toward on your first trial. No matter how clever you are, you have a 50% chance of being wrong. Moreover, you could easily be wrong on the second trial as well because the correct choice could be the card with the horizontal lines or it could be the card on the left. Even a simple discrimi- nation can take some time to develop, and dozens of errors may be made along the way.
Herbert Terrace (1963a, 1963b, 1964, 1972) found that errors can be reduced through errorless discrimination training. In this procedure the S∆ is presented in very weak form and for short periods. For example, in training a pigeon to discriminate between a red disc (the SD) and a green disc (the S∆), Terrace (1963a) presented the red disc at full strength for three minutes at a time, but presented an unlit green disc for only five seconds. Pigeons are less likely to peck a dark disc than a bright one, and the shorter the time the disc is available, the less likely it is to be pecked. The result was that the S∆ was seldom pecked. Gradually, Terrace increased the duration and strength of the S∆ while reinforcing pecking at the SD. Finally, Terrace was able to present the green disc brightly lit for a prolonged time without the bird pecking it. With the Terrace procedure, a discrimination can be developed with few errors. This is important because errors tend to arouse undesirable emotional reactions. Birds trained in the traditional manner, for example, often stamp their feet or flap their wings when presented with the S∆. Birds trained with the errorless procedure merely watch the disc calmly until the SD reappears.
Errorless discrimination training has been put to good use outside the laboratory. Richard Powers and his colleagues (1970) found, for instance, that preschoolers learned a subtle color discrimination more quickly and with fewer errors when trained with the Terrace procedure. Children trained in the usual way also became emotionally upset during the S∆ periods: They banged hard on the lever and wandered around the room. In contrast, those who learned through the errorless procedure sat quietly when the S∆ was present, patiently waiting for the SD to appear.
Another way to improve the rate of discrimination learning is to vary the consequences. In an experiment by M. A. Trapold (1970), rats could press either of two levers. When a light came on, pressing the lever on the left was reinforced, but when a tone sounded, pressing the lever on the right paid off. However, in this experiment Trapold provided different consequences for the two responses. The reinforcer for pressing on the left lever was food; the reinforcer for press- ing on the right lever was sugar water. The result was that the rats learned to make the appropriate discrimination more quickly and achieved a higher level of accuracy than when reinforcers were the same for each response. This find- ing—improved performance in discrimination training as a result of different consequences—is called the differential outcomes effect (DOE) (Miyashita, Na- kajima, & Imada, 2000; Peterson & Trapold, 1980; Goeters et al., 1992).
Normally, training is more effective if reinforcers follow behavior immediately than if they are delayed. But what if we provide immediate reinforcement for one correct behavior and delayed reinforcement for another correct behavior? Will the DOE still hold? J. G. Carlson and Richard Wielkiewicz (1972) performed just such an experiment and found that the DOE did hold. Animals receiving immediate reinforcement for one behavior and delayed reinforcement for the other learned the discrimination faster than animals receiving immediate reinforcement for both behaviors. Thus, the way discrimination training is conducted can have pronounced effects on learning. The training procedures described are fairly simple, but the discriminations that can be produced with them are amazing. For example, in one experiment by Debra Porter (then an undergraduate student!) and Allen Neuringer (1984), two pigeons heard an excerpt from Bach’s Prelude in C Minor for flute and Hindemith’s Sonata, Opus 25, Number 1 for viola. When Bach was playing, pecking a disc resulted in food; when Hindemith played, pecking produced no food. By the end of the experiment, both birds responded correctly over 80% of the time (see Figure 11-6). In a second experiment, the researchers provided two discs and randomly alternated excerpts from Bach and Stravinsky. When the birds heard Bach, pecking the disc on the left resulted in food; when they heard Stravinsky, pecking the disc on the right got food. If a bird pecked the wrong disc (e.g., the left disc when Stravinsky aired), the result was no food and a penalty period when no music played and pecking had no effect. Four of the five pigeons in the study responded correctly over 70% of the time. Other research showed that fish can learn to discriminate between classical and blues music (Chase, 2001). In similar research, Shigeru Watanabe, Junko Sakamoto, and Masumi Wakita (1995) trained pigeons to discriminate between paintings by Picasso and those of Monet. Pecking on a disc resulted in access to grain, but only if a painting by the right artist was visible. For four birds this meant pecking when there was a Monet painting present; for four other birds, it meant pecking when there was a painting by Picasso. If a bird pecked when the wrong kind of painting was visible, it got nothing. The training continued until the birds responded correctly more than 90% of the time. After this the researchers ran tests using new pictures by the artists to determine the basis on which the birds discriminated between them. One possibility was differences in color: The two artists might have had different color preferences, and that might have been what the birds were responding to. To test this idea the researchers presented monochromatic paintings by both artists. It made little difference: Although the overall success rate declined slightly, all of the birds continued to discriminate between Picasso and Monet. Another idea was that the birds responded to differences in contour. Picasso’s paintings tend to have sharp edges, whereas those of Monet, like all impressionist paintings, have a softer look. To rule out contour as the basis for discrimination, the researchers presented the images out of focus. Again, there was an overall decline in the accuracy level, but all of the birds continued to discriminate even when the images were blurred. The researchers tested other possible cues, but it does not appear that the birds responded to a single feature of either kind of painting. How they could tell Picasso from Monet is not clear, but they did. In other experiments, pigeons learned to discriminate between different locations on a university campus (Honig & Stewart, 1988) and among the letters of the English alphabet (Blough, 1982), and rats learned to discriminate between spoken Dutch and Japanese (Toro, Trobalon, & Sebastian-Galles, 2005).
Discrimination learning is not merely an interesting laboratory phenomenon; it has important practical applications. Learning a second language in adulthood can be difficult because of differences in the sounds of the two languages. To the Japanese, for example, the English L sounds like the Japanese R. James McClelland, Julia Fiez, and Bruce McCandliss (2002) provided discrimination training to Japanese adults living in the United States. Training consisted of listening to an audiotape and indicating which of two words, one starting with L, the other with R, they heard. For some participants the two words were rock and lock; for others the words were road and load. Some participants got immediate feedback after each effort; others did not. The training consisted of only three 20-minute sessions, yet the participants showed marked improvement. (Those who did not get feedback showed far less progress.) This research and other work on discrim- ination training have important implications not only for second-language learning, but for education and training in general. Being an auto mechanic, for example, requires knowing the parts of an engine and how they relate to one another. Achieving that involves learning to discriminate among the various parts.
Discrimination training has also proved useful in training animals to help humans with a variety of tasks. For a dog to sniff out illegal drugs, for instance, requires discriminating among various fragrances. Animals with a keen sense of smell can also locate landmines. Alan Poling and others affiliated with APOPO (an organization devoted to the removal of landmines; Poling et al., 2011) describe how the African pouched rat, a large rodent, is trained to do this. The explosive most often present in landmines is TNT, so a key part of the rodent’s training involves discrimination between that explosive and other substances. The training begins in a lab, then goes to field training before the real work begins. The animals learn to detect TNT and to indicate its presence by scratching at the ground for five seconds.
When discrimination training is highly effective, the SD reliably predicts the appearance of the reinforced behavior. At that point, the behavior is said to be under stimulus control, our next topic.
STIMULUS CONTROL
Consider a rat that has learned to press a lever when a light is on but not when the light is off. In a sense, you can control the rat’s lever pressing with the light switch: Turn it on, and the rat presses the lever; turn it off, and the press- ing stops. When discrimination training brings behavior under the influence of discriminative stimuli, the behavior is said to be under stimulus control (for a review, see Thomas, 1991).
Rats are not the only creatures, of course, that have behavior under stimulus control. While you are driving, if you approach an intersection and the traffic light turns red, you move your foot to the brake pedal. When the light turns green, you move your foot to the gas pedal. Your behavior has, as the result of discrimination training, come under the influence of the traffic signal. Similarly, you tend to enter stores that have signs that say “Open” and walk past stores that are marked “Closed.” People respond to signs that say “Sale,” “Reduced Prices,” “Clearance,” “Going Out of Business,” and the like. Retailers know the influence exerted by such signs and use them to attract shoppers. You may have noticed that some retail stores are constantly “going out of business.”
Sometimes (perhaps always) stimulus control is exerted not by a single stimulus but by a complex array of stimuli. We behave differently at a for- mal ball than we do at a square dance, and some behavior that would be acceptable at a beach party is unacceptable at a dinner party. The differential control exerted by such situations probably has to do with a number of stim- uli including attire, furniture, food, and the behavior of other people present. When a youngster misbehaves, he often defends his actions by saying, “Well, everyone else was doing it!” This explanation is partly an appeal to the stimu- lus control exerted by the behavior of one’s peers.
The term stimulus control suggests that we are manipulated by our envi- ronment, but there is another way of looking at it: The discriminative stimuli involved actually give us a kind of power. Consider the rat that learns to press a lever when a light is on but not when it is off. The light is said to control the rat’s behavior, but the rat has also gained control: It no longer wastes time and energy pressing a lever when doing so is useless. Similarly, the behavior of motorists comes under the control of traffic lights and signs. But it is this stimulus control that enables us to travel more or less safely and efficiently. Without stimulus control, traffic jams would be routine and our highways would be deadly gauntlets. In fact, many traffic accidents are attributed to inattentive driving—and what is inattentive driving but a failure of driving behaviors to be under the control of appropriate stimuli?
An understanding of the control exerted by our immediate environment can also give us the power to change that environment in helpful ways. People who are overweight typically find it difficult to turn down delicious food; the items exert a measure of control over their behavior. But understanding this can help them avoid situations in which tempting foods are present. Many people who are overweight have dishes of candy about their homes; if they get rid of the candy, it has less opportunity to influence their behavior. Our environment exerts control over our behavior. Paradoxically, that can increase the control we have over our lives.
Now that you understand the basics of generalization, discrimination, and stimulus control, let us see how they can account for some important phenomena.
GENEREALIZATION, DISCRIMINATION AND STIMULUS CONTROL IN THE ANALYSIS OF BEHAVIOR
Research on generalization, discrimination, and stimulus control has changed the way we think about many aspects of our lives. We will consider three examples.
Mental Rotation as Generalization
Roger Shepard is a psychologist who has studied what he calls “mental rotation.” In a typical experiment (Cooper & Shepard, 1973), people were shown letters that had been rotated varying degrees from their normal, upright position and were asked whether the letters were backward (that is, mirror images of the original) or not. The result was that the greater the rotation, the longer it took people to answer. Shepard concludes from such data that people mentally rotate an “internal representation” or image of the letter until it is in its normal, upright position and then decide whether it is backward.
Although Shepard refers to the mental rotation of images, his data consist of the time it takes to react to rotated figures. It is interesting that when these data are plotted graphically, the resulting curve looks remarkably like a generalization gradient (Figure 11-7). Participants respond most quickly to the “training stimulus” (the letter they were trained in school to recog- nize); the less the stimulus resembles the training stimulus, the slower is the response.
In one experiment, Donna Reit and Brady Phelps (1996) used a computer program to train college students to discriminate between geometric shapes that did and did not match a sample. The items were rotated from the sample position by 0, 60, 120, 180, 240, or 300 degrees. The students received feed- back after each trial. When the researchers plotted the data for reaction times, the results formed a fairly typical generalization gradient (see Figure 11-8). In a second experiment, Phelps and Reit (1997) got nearly identical re- sults, except that with continued training the generalization gradients flat- tened. This is probably because students continued to receive feedback during testing and therefore improved their reaction times to rotated items. (They could not improve their performance on unrotated items much because they were already reacting to those items quite quickly.) In any case, these data clearly suggest that “mental rotation” data are generalization data. Phelps and Reit note that most of their students, like Shepard’s, reported that they solved the problems by “mentally rotating” the test stimuli. As Phelps and Reit point out, however, the subjective experience of mental rota- tion does not explain the differences in reaction times. A scientific explanation must point to physical features of the situation and to the learning history of the participant. The expression “mental rotation” at best identifies the covert behavior involved; it does not explain the participant’s performance.
Concept Formation as Discrimination Learning
The word concept usually refers to any class the members of which share one or more defining features. The defining features allow us to discriminate the members of one class from the members of another. For example, all spiders have eight legs; this distinguishes them from other animals, including insects, which have fewer than or more than eight legs.
A concept is not a thing, however, but, as Fred Keller and William Schoenfeld (1950) put it, “only a name for a kind of behavior.” They explain: “Strictly speaking, one does not have a concept, just as one does not have extinction—rather, one demonstrates conceptual behavior by acting in a cer- tain way” (p. 154). Concepts require both generalization and discrimination. One must generalize within the conceptual class and discriminate between that and other classes. So, for example, to understand the concept, spider, one must both recognize a variety of spiders when one sees them, including spiders one has never seen before, and distinguish between spiders and other critters, such as ants and aphids. As Keller and Schoenfeld put it, “Generalization within classes and discrimination between classes—this is the essence of concepts” (p. 154f).
One way concepts are learned is through discrimination training. In one study, Kenneth Spence (1937) trained chimpanzees to find food under one of two white metal covers that differed only in size. One chimp got a choice between covers that were 160 and 100 square centimeters. Whenever it chose the larger cover, it found food; whenever it chose the smaller cover, it found nothing. After the chimp had learned to choose the larger cover reliably, Spence presented it with new covers, identical to the first set except that the choice was now between covers that were 320 and 200 square centime- ters. We might expect that the chimp would select the 200 square centimeter cover because that one more closely resembled the cover that previously hid food. Instead, the chimp chose the larger cover. It had learned the concept “larger than.”
In a similar experiment, Wolfgang Kohler (1939) trained chickens to select the lighter of two gray squares. After training, he tested them with the light gray square that had always led to food and with a still lighter gray square they had never seen before. Again, we might expect the animals to select the original gray stimulus because that had previously led to food. In fact, the birds chose the new, lighter square. In this case, the concept is “lighter than.” Richard and Maria Malott (1970) used discrimination to teach pigeons the concept of sameness. In this study, two halves of a key were illuminated independently and could therefore have different colors. When both halves were the same color (either all red or all violet), pecking the key produced food; when the two halves were different colors (one half red, the other violet), pecking did not produce food. After the pigeons had learned this discrimina- tion, the researchers tested the birds on four new patterns: blue–blue, yellow– yellow, blue–yellow, and yellow–blue. Three out of the four pigeons pecked more often when the key halves were the same color than when they were different.
These results are impressive, but the kinds of concepts involved are far simpler than concepts such as house, boat, insect, and human. These concepts are large classes with tremendous variability within the class. Houses, for example, take a great many different forms. Can animals acquire such concepts through discrimination training?
Richard Herrnstein and his colleagues performed a series of brilliant experiments aimed at answering this question. In the first of these, Herrnstein and D. H. Loveland (1964) projected photographic images within a pigeon’s chamber. The images included countryside, cities, bodies of water, lawn, meadow, and so on. About half of the photographs included at least one per- son, while the others did not. If the bird pecked a disc when an image included a person, it received food; if it pecked when there was no person, it received nothing. The people in the photos were sometimes partly hidden by other objects. Sometimes there was one person, sometimes a group of people. Some of the humans were clothed, some partly clothed, some nude. They included males and females, adults and children, and people of different races. Thus, to get food for pecking, the birds had to respond to a wide variety of human images. Any normally functioning adult could easily perform the task, but could a pigeon? The answer was a clear, Yes.
Herrnstein and Loveland next manipulated variables to see if the pigeons might be discriminating on the basis of some stimulus that happened to be correlated with humans, such as color. These efforts supported the idea “that the pigeons were, in fact, looking for, and reacting to, images of people” (p. 551). Additional support for this idea came from the mistakes the birds made. They sometimes failed to peck when the human being in the photo was largely hidden, and they occasionally pecked when the picture contained items commonly associated with people, such as cars, boats, and houses. It seems likely that these are the same kinds of errors people would make in performing the same task. “The evidence for a concept,” the researchers con- clude, “is incontrovertible” (p. 551). A replication of this study got similar results (Herrnstein, Loveland, & Cable, 1976).
Some people were skeptical that pigeons grasped such complex concepts. S. L. Greene (1983) suggested that the pigeons might simply memorize the figures associated with reinforcement. If this were the case, the birds should fail to discriminate accurately when tested on pictures they had not seen be- fore. Herrnstein (1979) did an experiment in which the task was to respond when a photograph included one or more trees or parts of trees. In this case, the birds appeared to learn the concept “tree.” Contrary to Greene’s prediction, the birds pecked even when shown slides they had never seen before if those slides included trees (see also Edwards & Honig, 1987). The birds apparently re- sponded to features that define the category, which is the essence of a concept.
An experiment by Robert Allan (1990; see also Allan, 1993) adds support to the idea that pigeons can learn concepts. Allan trained birds to peck a panel on which photographs could be projected, and he provided equipment to record the segment of the panel the bird pecked. His reasoning was that if birds were discriminating on the basis of a conceptual feature, they would peck at the part of the photograph that contained the conceptual item. He projected 40 slides, 20 of which included pictures of humans, the concept to be learned. Birds received food periodically if they pecked when a human figure appeared. The result was that the birds not only learned to make the appropriate discrimination, they also tended to peck the part of the slide that included a human figure (see Figure 11-9). Allan writes that “as the position of the human form changes from one segment to another, the pigeons track this movement by pecking in the same segment.” In a sense, the bird points to the object in the concept category. In considering this finding, it is impor- tant to note that reinforcement was not contingent on pecking the part of the panel in which human figures appeared, yet that is what the birds did.
Researchers have demonstrated that pigeons and primates can master the concepts fish, cats, flowers, ships, oak leaves, cars, letters, and chairs, among others. Fifty years ago that idea would have been dismissed as absurd. The point of this is not that people are no better at grasping concepts than other animals. (So far as I know, no animal has yet mastered the concepts justice, entropy, or concept.) The point is that by viewing concepts as learned behavior, rather than “mental representations” (the phrase commonly used by philosophers [e.g., Carey, 2011] and some psychologists [e.g., Laurence & Margolis, 1999]), we are able to study them experimentally, learn how they are acquired, and put our improved understanding of them to practical use.
Smoking Relapse as Stimulus Control
Mark Twain, a lifelong cigar smoker, once quipped: “It’s easy to quit smoking. I’ve done it hundreds of times.” It is not hard to understand why people who have become addicted to nicotine continue to smoke. By some estimates, cigarette smoking is reinforced 73,000 times a year in a pack-a-day smoker (Lyons, 1991). The act of puffing on a cigarette has therefore been reinforced 730,000 times in a moderate smoker who has used cigarettes for ten years. For a heavy smoker (two packs a day or more), the number of reinforcements for that period is about one-and-a-half million. If each reinforcement increases the resistance of a behavior to change, then it is hardly surprising that people find it difficult to quit smoking.
But why do people who have given up smoking, and who are no longer under the influence of the physiological effects of nicotine, so often resume smoking? Quitting may be difficult because of the physiological effects of not maintaining the nicotine level, but taking up smoking again weeks or months after going through withdrawal seems to many people clear evidence of weak character. But, as we have seen, concepts like weak character do not explain puzzling behavior; they merely label it.
Smoking relapses become less puzzling when we realize that the physi- ological effects of smoking and of withdrawal are not the only factors involved in this behavior. In 1988, then Surgeon General C. Everett Koop concluded that “environmental factors, including drug-associated stimuli and social pressure, are important influences of initiation, patterns of use, quitting, and relapse to use of opioids, alcohol, nicotine, and other addicting drugs” (U.S. Department of Health and Human Services, 1988, p. 15). “Drug-associated stimuli” include environmental events that, because they have preceded tobacco use in the past, have acquired some degree of stimulus control over tobacco use. In other words, drug abuse, including smoking, is under stimulus control.
Ask smokers when they are most likely to smoke, and you are likely to be told on arising from bed in the morning, while having coffee, after eating, during work breaks (including the interval between classes), during or after stress (such as an exam or city driving), after physical exertion, when social- izing with friends, and so on (Buckalew & Gibson, 1984; Smith & Delprato, 1976). Because the use of tobacco and the reinforcing effects of nicotine have frequently occurred together in these situations, they have become discrimi- native stimuli for lighting a cigarette. And because smokers typically smoke throughout the day, many different situations become discriminative stimuli for smoking. Charles Lyons (1991) writes that “few other activities are so con- sistently and powerfully strengthened in such a wide range of temporal, situ- ational, and physical settings” (p. 218).
Most people have witnessed stimulus control in smokers, although they may not have realized it at the time. Imagine a moderate smoker who has just joined a group of people in casual conversation. A member of the group lights a cigarette. This act is a discriminative stimulus for smoking by others so even if our hypothetical smoker has recently smoked a cigarette, he may light up after seeing someone else do so. The smoker may explain this behavior by saying, “When I see someone else smoke, it makes me think of smoking, and then I have to have a cigarette.” Sometimes smokers report that cues that “remind them” of cigarettes also induce feelings of physiological deprivation or “craving.” But these thoughts and feelings do not explain the behavior of smoking any better than a lack of willpower does. The tendency for certain kinds of events to elicit smoking is explained by the history of reinforcement for smoking in the presence of those events.
Smoking in situations previously associated with smoking seems particu- larly likely to lead to an abrupt return to regular smoking. T. H. Brandon and colleagues (reported in Lyons, 1991) studied people who had quit smoking and who then had a single cigarette in a situation previously associated with smok- ing. Ninety-one percent of them soon became regular smokers again, nearly half within a day of the single cigarette. In another study, R. E. Bliss and colleagues (1989) found the presence of other people smoking commonly led to relapse.
The research on the role of stimulus control in smoking has important implications for those who would like to quit. It would appear that there are two basic approaches to preventing relapse. The former smoker can avoid sit- uations in which he or she often smoked in the past, thereby avoiding the ability of these situations to elicit smoking. Or the smoker can undergo train- ing to reduce the control these situations have over his or her behavior. It is extremely difficult, if not impossible, for a smoker to avoid all situations in which he or she has smoked; therefore, the best bet may be to undergo train- ing that will undermine the power of those situations. This might be done, for example, by gradually exposing the smoker to those situations while prevent- ing him or her from smoking. For example, a smoker who typically lights up after drinking coffee may have coffee in a therapist’s office without smoking. When this situation no longer arouses the urge to smoke, the same training might be repeated in the nonsmoking section of a restaurant. The training might continue with having a meal in the restaurant, with the therapist (or other supportive person) along to ensure that the smoker does not light up. The person who would quit smoking may need to undergo the same sort of treatment in each kind of situation in which he or she has often smoked in the past. Giving up smoking for good requires overcoming stimulus control.
The same may be said of many other habitual behaviors. We saw in an earlier chapter the role that the schedule of reinforcement exerts on gambling. But gambling is also a behavior that occurs in certain kinds of settings, and those settings and associated cues exert power over behavior. The same thing is true of overeating. Brian Wansink (2006), a marketing professor at Cornell University, talks about the effects of “hidden persuaders,” environmental cues for eating. He describes an experiment in which people ate soup from a bowl that automatically refilled as they ate. Normally an empty soup bowl is a cue to stop eating, but in this case there was no such cue. People tended to eat more than one bowl of soup, but some people ate much more—in some cases, more than a quart of soup. If an empty bowl or an empty plate is a discriminative stimulus to stop eating, then limiting the amount of food we serve ourselves can help reduce calorie consumption.
Whether it’s smoking, gambling, overeating, or some other behavior in which we participate to excess, the solution is not a matter of strengthening willpower; it is a matter of weakening the power of environmental cues.
THEORIES OF GENEREALIZATION AND DISCRIMINATION
Three theories of generalization and discrimination have dominated the field: those of Pavlov, Spence, and Lashley and Wade.
Pavlov’s Theory
Pavlov’s theory is physiological. He believed that discrimination training pro- duces physiological changes in the brain. Specifically, it establishes an area of excitation associated with the CS+ and an area of inhibition associated with the CS−. If a novel stimulus is similar to the CS+, it will excite an area of the brain near the CS+ area. The excitation will irradiate to the CS+ area and elicit the CR. Similarly, if a novel stimulus resembles the CS−, it will excite an area of the brain near the CS− area. The excitation of this area will irradiate to the CS− area and inhibit the CR. A similar explanation could be applied to general- ization and discrimination following operant learning.
Pavlov’s theory provides an intuitively appealing explanation and, wrapped as it is in physiology, it has the smell of science. Unfortunately, the physio- logical events were merely inferred from observed behavior. Pavlov presumed that irradiation of excitation occurred because generalization occurred, but there was no independent validation of its happening. The theory therefore suffered from circularity. Other theorists, most notably Kenneth Spence, have modified Pavlov’s ideas.
Spence’s Theory
Kenneth Spence (1936, 1937, 1960) put Pavlov’s physiology aside but kept the notions of excitation and inhibition.
Pairing a CS+with a US results in an increased tendency to respond to the CS+ and to stimuli resembling the CS+. Similarly, in operant learning, reinforcement for responding in the presence of an SD results in an increased tendency to respond not only to the SD but to similar stimuli. The generaliza- tion gradient that results is called an excitatory gradient. In the same way, presenting a CS− without the US results in a decreased tendency to respond to the CS− and to stimuli resembling the CS−. Likewise, withholding rein- forcement when operant behavior occurs in the presence of an S∆ results in a decreased tendency to respond to that stimulus and to similar stimuli. The generalization gradient that results is called an inhibitory gradient.
Spence proposed that the tendency to respond to any given stimulus was the result of the interaction of the increased and decreased tendencies to re- spond, as reflected in gradients of excitation and inhibition. Consider a dog that is trained to salivate at the sound of a high-pitched tone, and another that is trained not to salivate at the sound of a low-pitched tone. The first dog will show generalization of excitation around CS+; the second will show generalization of inhibition around CS−. We can plot the excitatory and inhibi- tory gradients that result and place them next to one another, as depicted in Figure 11-10. Notice that the two curves overlap. Discrimination training produces much the same effect within an individ- ual. That is, the increased tendency to respond to stimuli resembling the CS+ (or SD ) overlaps with the decreased tendency to respond to stimuli resembling CS− (or S∆). What Spence proposed was that the tendency to respond to a novel stimulus following discrimination training would be equal to the net differ- ence between the excitatory and inhibitory tendencies. In other words, the tendency to respond to a novel stimulus will be reduced by the tendency not to respond to that stimulus.
Consider a hypothetical experiment in which a pigeon is trained to peck an orange disc but not a red one. After training, we give the bird the opportu- nity to peck the disc when it is a variety of colors, from pale yellow to deep red. What color disc will it peck most often? We know that if the bird had merely received food for pecking the orange disc, it would peck that same color most often. But discrimination training, according to Spence, should re- sult in inhibition of the tendency to peck stimuli resembling the S∆. Spence’s theory therefore predicts that the peak of responding will not occur at the SD but at a stimulus further away from the S∆. In other words, the peak of re- sponding will not be on the orange disc but on one that is even less reddish.
This prediction, made in the 1930s, was actually confirmed in the 1950s in an experiment much like that just described. H. M. Hanson (1959) trained pigeons to peck a yellowish-green disc (550 nm, or nanometers, a measure of wavelength) and not to peck a slightly more yellowish (560 nm) disc.1 A control group of birds did not undergo discrimination training but did receive food for pecking the yellowish-green disc. After training, Hanson let the birds peck discs of various colors, from yellow to green. The control group showed a peak of responding to the discriminative stimulus. Birds that had received discrimination training, however, showed a shift away from the S∆; their peak of responding was to a stimulus of about 540 nm (see Figure 11-11). This phe- nomenon, called peak shift, has proved to be a robust phenomenon (Purtle, 1973; Thomas et al., 1991). The Lashley-wade Theory
Karl Lashley and M. Wade (1946) proposed an approach to generalization and discrimination that differs from those of Pavlov and Spence. These researchers argued that generalization gradients depend on prior experience with stimuli similar to those used in testing. Discrimination training increases the steepness of the generalization gradient because it teaches the animal to tell the difference between the SD and other stimuli. But the generalization gradient is not usually flat even in the absence of training. Why is this so if the gradient depends on training? The answer Lashley and Wade give is that the animal has undergone a kind of discrimination training in the course of its everyday life. A pigeon, for example, learns to discriminate colors long before a researcher trains it to peck a red disc. The more experience a pigeon has had with colors, especially those resembling the SD, the steeper its generalization gradient will be; the less experi- ence the bird has had, the flatter the gradient will be.
The theory implies that if an animal is prevented from having any expe- rience with a certain kind of stimulus, such as color, its behavior following training will be affected. If such a color-naïve animal is trained to respond in the presence of a red disc, for example, it will later respond just as frequently to a green disc. In other words, its gradient of generalization will be flat.
Several researchers attempted to test this hypothesis. In a typical experiment, animals were reared from birth in the dark to deprive them of experiences with color. Then they were trained to respond to a stimulus such as a green disc. After this, the animals were tested for generalization by pre- senting them with discs of other colors and noting the extent to which they discriminated. The results were then compared to those obtained from animals that had been reared normally. If the gradients of the color-deprived animals were flatter, the Lashley–Wade theory was supported; if rearing in the dark made no difference in the shape of the gradient, this argued against the theory.
Unfortunately, the results of such experiments have been ambiguous, with one study tending to support the theory and another tending to under- mine it. Moreover, interpretation of the results is subject to argument. When there is no difference in the gradients of deprived and normally reared animals, proponents of the Lashley–Wade theory argue that the rearing procedure did not entirely preclude experience with the relevant stimuli. When deprivation produces a flat gradient, opponents of the theory argue that the deprivation procedure damaged the eyes of the animals so that their physical capacity for discriminating colors has been limited. The Lashley–Wade theory needs a stronger test than deprivation studies can provide.
If the theory is valid, one argument holds that depriving an animal of all experience with a stimulus should not be necessary; merely restricting its experience with the stimulus during training should be sufficient to sup- port the theory. To test this idea, Herbert Jenkins and Robert Harrison (1960) trained pigeons to peck a disc. Some pigeons heard a tone periodically; peck- ing was reinforced in the presence of the tone but not during periods of quiet. Other pigeons heard the same tone without interruption. In both cases, then, disc pecking was reinforced in the presence of a tone; but in one case, there were periods of silence during which pecking was not reinforced. Next, the experimenters tested all the pigeons for generalization to other tones and to periods of silence. They found that those pigeons that had been exposed to periodic tone were much less likely to peck the disc during periods of silence than when the tone sounded. The other pigeons pecked the disc just as much when the tone was on as when it was off. This much is to be expected because the birds that heard the tone constantly had no opportunity to discriminate, whereas those that heard the tone periodically did. But what happened when the pigeons were exposed to different tones, sounds that neither group had heard before? The pigeons that had learned to discriminate between periods of tone and periods of silence also discriminated between the original tone and other tones. Pigeons that had received reinforcement during constant sound did not discriminate between tones (see Figure 11-12). These results are just what the Lashley–Wade theory predicts. Not all tests of the Lashley–Wade theory have yielded positive results, but it is now generally acknowledged that the steepness of a generalization gradi- ent depends to some extent on the experience the participant has had with the relevant stimuli before training.
No theory of discrimination and generalization has won universal sup- port, but they have all spurred basic research. Some of this research has pro- vided insights into problems of theoretical and practical importance.