Stimulus generalization vs discrimination: the difference, and how one turns into the other

Stimulus generalization occurs when a learned response also occurs to stimuli that are similar to the trained stimulus. Stimulus discrimination occurs when the learner responds to the trained stimulus but not to similar stimuli. A dog that is conditioned to one tone and also salivates to a slightly higher tone shows generalization. A dog that salivates to the trained tone and not to the higher tone shows discrimination. Generalization comes first. Discrimination is learned, usually when the similar stimulus is never reinforced.

The two definitions are short. Three things make the exam questions difficult. Textbooks often call the terms opposites, but one test can show both. The word "discrimination" also has a social meaning that has no connection with conditioning. And a third term, response generalization, sounds almost the same. This article explains each term from the original sources: Pavlov's dogs, Little Albert, pigeons and horses. It ends with a method to sort exam scenarios.

• Both terms apply to classical and to operant conditioning. The APA Dictionary of Psychology defines stimulus generalization for conditioning that is "either operant or classical".
• Pavlov saw generalization before discrimination. He found that discrimination needs contrast: a similar stimulus that is never reinforced.
• Generalization is a matter of degree. The more similar a new stimulus is to the trained stimulus, the stronger the response. A graph of this relation is a generalization gradient.
• Discrimination training changes generalization. After the training, the strongest response can move past the trained stimulus. This change is the peak shift.
• Little Albert's record shows generalization and discrimination in the same session. His later fear of the rabbit and the dog was partly direct conditioning.

The difference in one table

The open textbook Psychology 2e from OpenStax gives an everyday example. Tiger, a cat, has learned that the sound of the electric can opener comes before her food. If the electric mixer sounds very similar to the can opener, "Tiger may come running". That is generalization. If Tiger learns that the mixer never brings food and stays where she is, that is discrimination.

A quick memory aid: to generalize is to treat many things as one general class. To discriminate is to tell things apart.

What Pavlov found: generalization first, discrimination by contrast

Ivan Pavlov described both processes in Lecture VII of Conditioned Reflexes, published in English in 1927. He wrote that the analysis of stimuli "is always preceded by what we call a 'period of generalization'".

His example used tones. When a tone with a pitch of "1000 d.v." became a conditioned stimulus, "many other tones spontaneously acquire similar properties". The effect became weaker as the pitch moved further from the trained tone. Touch worked in the same way. A touch on one area of skin became a conditioned stimulus, and touches on other areas also produced some response. The response decreased with the distance from the trained area.

Pavlov did not treat this spread as an error. He explained it by the fact that "natural stimuli are in most cases not rigidly constant". A response that also occurs to similar stimuli lets the animal respond to these natural variations.

Then he asked how discrimination develops. His laboratory first tried one method: present the conditioned stimulus with reinforcement many times. The method did not work. Pavlov reported that "no absolute differentiation was ever obtained by the use of the first method, even though the stimulus was repeated with reinforcement over a thousand times".

The second method worked. The laboratory reinforced the trained stimulus and also presented a similar stimulus that was never reinforced. Even "a single unreinforced application of an allied stimulus" led to "a rapid development of differentiation". Pavlov wrote: "The method of contrast is now always employed in our experiments".

Three details help with exam questions:

1. 1.Pavlov used the word differentiation for most of the process. He called the process that suppresses the response to the similar stimulus differential inhibition.
2. 2.He counted differential inhibition as his fourth type of internal inhibition. The first type is extinction, which also follows from non-reinforcement. Our article on schedules of reinforcement explains extinction in operant conditioning.
3. 3.Discrimination did not develop in a straight line. When the laboratory first tried a neighbouring tone, the response was often weak. With more unreinforced trials, the response increased until it was equal to the response to the trained tone. Then it decreased, "falling finally to a permanent zero".

The generalization gradient

Generalization is not all or nothing. The APA Dictionary defines a generalization gradient as "a graph marking the similarity or difference between two stimuli versus the similarity or difference in their elicited responses". It adds: "In general, the more similar two stimuli, the more similar the responses."

Think of the graph for Pavlov's tones. The horizontal axis shows the test tones, from low to high. The vertical axis shows the strength of the response. The highest point is at the trained tone, and the curve falls on the two sides.

Later researchers used generalization tests as a standard technique. A 1956 paper by Guttman and Kalish on "discriminability and spectral generalization in the pigeon" was a starting point. Spectral means that the stimuli were colours of light. In 1981, Honig and Urcuioli reviewed the 25 years of research on operant stimulus generalization since that paper.

The gradient also shows why the two terms are not strict opposites. OpenStax calls stimulus generalization "the opposite of stimulus discrimination". The APA Dictionary entry on stimulus generalization is more exact. If the response to a new stimulus is the same as the response to the original stimulus, generalization is complete. But "If responding is different enough to be detected, discrimination is evident as well as generalization."

So a dog that salivates half as much to a new tone shows both processes. It responds, which is generalization. It responds less, which is discrimination.

Peak shift: discrimination training moves the peak

Discrimination training does more than stop the response to one similar stimulus. It can change the shape of the whole gradient. The APA Dictionary describes this change as the peak shift.

The training uses two stimuli on one dimension, for example a bright light that is reinforced and a dim light that is not. After this training, the strongest response moves "away from the less favorable stimulus". It moves to a point beyond the reinforced stimulus. In the APA example, the peak moves from the bright light to a very bright light.

A 1991 study with horses by Dougherty and Lewis shows the effect. The stimuli were nine black circles from 0.5 to 4.5 inches across. The horses learned to press a lever with the lip when a 2.5-inch circle was shown. In discrimination training, grain followed presses with the 2.5-inch circle and never followed presses with a 1.5-inch circle. Three horses reached zero responses to the 1.5-inch circle in fewer than 15 sessions.

The results show what the training did to generalization:

• Horses tested before discrimination training produced symmetrical gradients. The response decreased evenly on the two sides of the 2.5-inch circle.
• Horses tested after discrimination training produced asymmetrical gradients. In the words of the authors, "The peak of these gradients shifted away from the negative stimulus."

The negative stimulus was the smaller circle. So the peak moved toward the larger circles, past the circle that had earned the grain.

Little Albert: what the record shows

Little Albert is the standard textbook example of generalization. OpenStax summarises the case in one line: "he became afraid of other furry things". The 1920 paper by John Watson and Rosalie Rayner gives more detail, and the detail matters for the two terms.

Watson and Rayner did not use the word generalization. They asked whether there would be "a transfer to other animals or other objects". They presented a white rat together with the loud sound of a struck steel bar. After seven of these joint stimulations, Albert showed a complete fear reaction to the rat. Five days later, at 11 months and 15 days of age, the researchers tested other objects.

The table shows both terms in one session. The fear spread to the rabbit, the dog, the fur coat and the mask. That is generalization. The fear did not spread to the blocks or to the hair of the other observers. That is discrimination. The reaction also varied in strength, from a strong reaction to the rabbit to a mild one to the cotton wool.

Summaries often omit the next step. Five days after this test, Watson and Rayner paired the rabbit and the dog directly with the noise. They wrote that "this was the first time any effort had been made to directly condition response to the dog and rabbit". So Albert's later fear of the rabbit and the dog was partly direct conditioning. For these two animals, only the tests before that pairing show generalization in a clean form. The fur coat, the cotton wool and the Santa Claus mask were never paired with the noise.

The record also shows a limit of the study. The last test came 31 days after the session with the direct pairing. At that test, the mask and the fur coat still produced fear, although neither was ever paired with the noise. Albert left the hospital on the same day, and the researchers never tried to remove the fear.

Discrimination in operant conditioning

Pavlov, Watson and Rayner worked with classical conditioning. But the two terms also apply to operant conditioning, where a behaviour is learned from its consequences. Our article on classical and operant conditioning explains the two kinds.

In operant conditioning, the stimulus does not cause the response. It signals when the response will be reinforced. The APA Dictionary calls this kind of stimulus a discriminative stimulus. Its symbol is SD, an S with a raised D, so do not confuse it with SD for standard deviation. It increases the probability of a response "because of a previous history of differential reinforcement in the presence of that stimulus".

The APA example uses a pigeon. Pecks are reinforced when the key is red and not when the key is green. The red light becomes an SD, and the pigeon learns to peck only when the key is red. Generalization also occurs in operant conditioning. A pigeon reinforced for pecks at one colour also pecks at similar colours, which is the spectral generalization that Guttman and Kalish studied.

An SD signals that a response will now be reinforced. Our article on reinforcement and punishment explains the four kinds of consequence that operant conditioning uses.

When generalization is the goal, and when it is a problem

In Pavlov's laboratory, generalization was the default, and discrimination was the skill to train. Applied behaviour analysis (ABA) often has the opposite goal. A therapist who teaches a skill in a clinic wants the skill to appear at home and at school too.

Stokes and Baer described this difference in a 1977 review. They wrote that discrimination "has been understood as an active process", with a developed technology. Generalization, in contrast, was seen as "the natural result of failing to practice a discrimination technology adequately". It remained "a passive concept almost devoid of a technology". The authors argued that generalization "is equally deserving of an active conceptualization and technology". They sorted the studies into nine groups, from "Train and Hope" to "Train Sufficient Exemplars".

ABA also uses a term that students confuse with stimulus generalization. In response generalization, the stimulus stays the same and the response varies. The APA Dictionary lists it under induction: reinforcement of some forms of a behaviour also makes "similar but nonreinforced forms" more probable. In its example, lever presses with a force of 0.2 to 0.3 newtons are reinforced, and weaker and stronger presses also increase.

Generalization can also cause harm. A fear that spreads to more and more stimuli limits what a person does. A 2015 systematic review by Dymond and colleagues calls fear generalization "a defining feature of anxiety disorders".

A 2010 study by Lissek and colleagues measured it. The researchers tested 19 patients with panic disorder and 19 healthy comparison subjects. The stimuli were 10 rings of graded size on a computer screen. The ring at one extreme was a conditioned danger cue, and the ring at the other extreme was a safety cue. The researchers measured fear with the startle blink reflex.

• In the patients, fear generalized to rings up to three units different from the danger cue.
• In the comparison subjects, fear generalized only to rings one unit different.

In a 2012 review, Lissek called fear generalization "one of the more robust conditioning markers of clinical anxiety". These are laboratory findings about groups. They cannot show whether one person has an anxiety disorder. If fear of a growing range of places, objects or situations affects your daily life, talk to a doctor or a psychologist.

Classifying exam scenarios

Most exam items describe one learner and one new stimulus. Two questions sort them:

1. 1.What changed: the stimulus or the response? If the response changed and the stimulus stayed the same, the term is response generalization.
2. 2.If the stimulus changed, does the learner respond to the similar stimulus? If yes, the term is stimulus generalization. If the learner responds only to the trained stimulus, the term is stimulus discrimination.

Watch for signal words in the scenario. "Never reinforced" or "not followed by" usually points to discrimination. "Similar" or "also" usually points to generalization.

How to keep the terms apart

Four memory aids help:

• Generalize: treat many things as one general class. The learner responds to the whole class.
• Discriminate: tell things apart. The learner responds to one member of the class.
• Generalization comes without effort. Discrimination needs contrast, as Pavlov found.
• If the stimulus stays the same and the response changes, the term is response generalization.

A memory aid helps you only if you practise with it. Our guide to memorising psychology terminology explains the two study methods with the strongest evidence. The article on spaced repetition explains why short sessions on different days are better than one long session.

Psychology World is built for this kind of practice. You look up a term and save it to one of your own lists. Then you drill the list as flashcards, by typing the answer, or as multiple-choice questions once the list has four or more terms. The four steps of the learning loop and what each study mode does are on the home page.