Psychology of Learning

Module 04: Classical Conditioning 1

Part 3: Generalization & Discrimination

Looking Back

In Parts 1 and 2, we established the foundations of classical conditioning. We learned Pavlov’s basic procedure, the key terminology (US, UR, CS, CR), and the factors affecting acquisition (number of pairings, US strength, CS-US timing). We explored extinction—how presenting the CS without the US reduces the CR—and spontaneous recovery, the mysterious reappearance of extinguished responses. We distinguished between appetitive and aversive conditioning and saw how impossible discrimination tasks can create experimental neurosis. Now we turn to more sophisticated aspects of classical conditioning: How do organisms respond to stimuli similar to the CS? How do they learn to differentiate among stimuli? Can a CS itself serve as a US for new learning?

Generalization: Responding to Similar Stimuli

Consider this statement: “If classical conditioning is going to serve an adaptive function, then the CR should also be elicited by stimuli that are similar to the CS that was conditioned.” Why is this true? Think about it from an evolutionary perspective.

If you learn that a specific predator sound signals danger, you benefit from responding to similar sounds—slightly different pitches or volumes—as also dangerous. Responding only to that exact sound frequency would be maladaptive. Environments vary; the next encounter won’t be identical to the first. Generalization allows organisms to apply learning flexibly across situations.

Generalization is the ability of stimuli that are similar to the CS to elicit a CR. The more similar a stimulus is to the original CS, the stronger the CR it elicits. Generalization produces a gradient—maximum responding to the original CS, decreasing responding as stimuli become less similar (Pavlov, 1927).

Pavlov’s original research on generalization involved conditioning a CR to one specific ticking speed of a metronome (CS). Following conditioning, he tested for CRs to different ticking rates. The more similar the ticking rate to the original CS, the stronger the CR. A metronome ticking at 60 beats per minute might produce maximum salivation, while 55 or 65 beats produce slightly less, and 40 or 80 beats produce even less (Pavlov, 1927).

Examples of Generalization

Visual Generalization: If presentation of a 550 nm light (green) was the CS that elicits a CR, then lights whose wavelength is similar to 550 nm should elicit similar responses. A 540 nm or 560 nm light elicits strong CRs, while a 450 nm (blue) or 650 nm (red) light elicits weaker CRs. The generalization gradient follows the visible spectrum (Guttman & Kalish, 1956).

Apple Generalization: If you’ve learned to associate Fuji apples with delicious taste, shouldn’t you generalize this knowledge to other apples? The degree of similarity between Fuji apples and a new apple determines how much generalization occurs. When presented with a Macintosh apple—similar in appearance, texture, and sweetness—high generalization is warranted. But when presented with a Granny Smith apple—green, tart, firm—you probably shouldn’t generalize as much. The CR (positive expectation) will be weaker (Shepard, 1987).

Haru the Dog—Generalization in Action: A personal example illustrates generalization beautifully. Haru, a Shih Tzu, loved people—an unconditioned reflex. Shih Tzus are bred specifically to be cute and love humans; it’s in their genetics. Through classical conditioning, Haru learned that the doorbell (CS) predicted visitors arriving (US), which triggered excitement (UR). Soon, the doorbell alone produced excitement—running to the door, tail wagging (CR). But Haru’s CR was evoked not just by the home doorbell. Doorbells on television had various effects on Haru, depending on how similar they were to the home doorbell. Very similar bells produced strong reactions (high generalization); dissimilar bells produced weaker reactions (less generalization). This is generalization and discrimination in action in everyday life!

Discrimination: Responding Selectively

Discrimination is the converse of generalization. The organism learns to respond to one stimulus but not others. Instead of making CRs to stimuli similar to the CS, responding is shown only to one stimulus, typically the CS that was conditioned to the US (Pavlov, 1927).

Discrimination is also adaptive. If one CS predicts the US but a similar stimulus does not, organisms benefit from distinguishing them. Humans learn numerous discriminations: which insects sting and which don’t, which mushrooms are edible and which are poisonous, which facial expressions signal friendliness and which signal hostility. Fine discriminations enhance survival (Spence, 1937).

How do researchers produce discrimination? If I present a 550 nm light (CS) and follow it with food (US), a CR (salivation) will develop. This light becomes a CS+ (excitatory CS). But if I start presenting other lights—say, 500 nm or 600 nm—without pairing them with food, the animal will stop responding to almost every light except 550 nm. These unpaired lights become CS- stimuli (inhibitory CSs). The organism discriminates among wavelengths (Guttman & Kalish, 1956).

Discrimination training narrows the generalization gradient. With repeated exposure to CS+ (paired with US) and CS- (not paired with US), organisms learn increasingly fine distinctions. What begins as broad generalization becomes precise discrimination.

Clinical Implications of Generalization

Recent research has revealed that overgeneralization of fear plays an important role in anxiety disorders. A systematic review and meta-analysis by Stegmann and colleagues (2022) examined fear generalization across multiple anxiety and stress-related disorders. Their analysis found robust evidence that patients with generalized anxiety disorder, panic disorder, and PTSD show broader fear generalization gradients compared to healthy controls—meaning they respond fearfully to a wider range of stimuli that only vaguely resemble the original threat. This overgeneralization, quantified with a small but reliable effect size, may explain why individuals with anxiety disorders experience fear in situations that are objectively safe. Understanding these generalization patterns has implications for treatment, suggesting that therapeutic interventions might benefit from targeting discrimination learning to help patients distinguish genuine threats from safe situations.

Higher-Order Conditioning: CSs Become USs

Consider this scenario: Last week you and your former best friend, Bill (CS), had a major argument (US) which made you very angry (UR). Now, whenever you see Bill, you become angry (CR). This is straightforward classical conditioning. But something else happens: On several recent occasions, you’ve seen Bill talking to John. Now you become angry when you see John, even though you have no reason to be angry with him. What’s going on?

Higher-order classical conditioning occurs when a previously neutral stimulus comes to elicit a CR after being paired with an already conditioned CS. The new stimulus has never been paired with the original US, yet it elicits a CR (Pavlov, 1927).

Here’s how it works. In the original conditioning, the US (violent argument) elicits the UR (anger), and Bill (CS1) is paired with arguments (US); result: Bill (CS1) now elicits anger (CR). Then Bill talks with John several times: Bill (CS1) elicits anger (CR), and John (CS2) is present during this anger; result: John (CS2) now elicits anger (CR), even though John was never present during arguments.

The original CS is called the first-order CS; it had direct contact with the US and is only one step removed from the US. The stimulus paired with the first-order CS is called the second-order CS; it’s two steps removed from the US. Bill is the first-order CS; John is the second-order CS (Pavlov, 1927).

Strength of Higher-Order Conditioning

Will the CR elicited by the second-order CS (John) be as strong as the CR shown to the first-order CS (Bill)? No. The CR elicited by the second-order CS will be weaker because the procedure used to establish higher-order conditioning is also the procedure for producing extinction—presenting the CS without the US. Each time John appears without an argument, extinction occurs (Rescorla, 1980).

Third and even fourth-order CSs are possible, but conditioning becomes very weak. Each additional step away from the original US produces weaker responding. Nevertheless, higher-order conditioning demonstrates how conditioning can spread through chains of associations, expanding the range of stimuli that elicit CRs.

Sensory Preconditioning: Associating Neutral Stimuli

Sensory preconditioning involves repeated pairing of two neutral stimuli before one of them is paired with a US. This is another mechanism, in addition to generalization and higher-order conditioning, by which stimuli that have never been paired with a US can come to elicit a CR (Brogden, 1939).

The procedure follows four steps. First, two neutral stimuli are paired: CS1 + CS2, and neither elicits any response yet—for example, a tone and a light are presented together repeatedly. Second, CS1 is paired with a US: CS1 + US → UR, and traditional classical conditioning occurs with CS1—the tone is paired with food. Third, CS1 now elicits a CR: CS1 → CR, which is standard conditioning—the tone elicits salivation. Fourth, sensory preconditioning emerges: CS2 → CR—the light, which was never paired with food, also elicits salivation because it was associated with the tone!

Real-World Example of Sensory Preconditioning

Assume a school child frequently sees dogs (CS1) on the playground at school (CS2). These two stimuli are associated—the child sees dogs and playground together regularly. One day the child is bitten (US) by a dog in his own front yard, producing fear and panic (UR). Now the child doesn’t want to go out on the playground for recess.

The dog (CS1) was paired with the painful bite (US) that elicited fear. Traditional conditioning occurred. But because dogs (CS1) and the playground (CS2) were previously associated, the playground also elicits fear (CR), even though the bite never occurred at school. Sensory preconditioning explains how phobias can spread to situations never directly associated with trauma (Rizley & Rescorla, 1972).

Sign Tracking: Investigating Predictive Stimuli

Sign tracking is the tendency to investigate and explore stimuli that predict relevant events (USs). Organisms direct attention and behavior toward CSs, treating them as significant objects in their own right (Brown & Jenkins, 1968).

Consider studying classical conditioning in pigeons. You’d have difficulty studying salivary CRs—pigeons don’t salivate like dogs. What response could you condition? Sign tracking provides the answer: A lighted disk is presented (CS), food is then presented (US), and the pigeon pecks at food (UR). After repeating this sequence, the pigeon begins exploring and pecking the disk (CR) when it lights up, even before food appears. The pigeon treats the CS as if it were the US, directing behavior toward this predictive stimulus (Brown & Jenkins, 1968).

This phenomenon, also called autoshaping, allows researchers to establish behaviors in pigeons that can then be studied using operant conditioning techniques. Sign tracking demonstrates that CSs don’t just elicit preparatory responses—they become objects of interest and investigation themselves.

Conditioned Emotional Responses: Emotions Through Association

Imagine a rat that has learned to press a lever for food. The rat performs this response well, making many lever presses during its daily 15-minute test session. During another session, the same rat receives several pairings of a flashing light (CS) and mild electric shock (US). What happens when the experimenter presents the flashing light CS during a lever-press session?

A conditioned emotional response (CER) occurs when the CS signals an aversive US and elicits emotional reactions that can suppress ongoing behavior. Because CERs can suppress behavior, the term conditioned suppression is also appropriate (Estes & Skinner, 1941).

When the light flashes, the rat will probably freeze—an innate defense response. Freezing is adaptive; it lowers the chances that a rodent will be noticed by a predator. The light (CS), through pairing with shock (US), now elicits fear and freezing (CR). This fear suppresses lever pressing. The rat stops responding entirely while the light is on (Estes & Skinner, 1941).

CERs demonstrate how classical conditioning shapes emotional experiences. Many human emotional responses—phobias, anxieties, pleasures—develop through classical conditioning. A stimulus associated with trauma elicits fear; a stimulus associated with pleasure elicits positive feelings. These conditioned emotions powerfully influence behavior.

Conditioned Inhibition: Signaling US Omission

So far, we’ve discussed conditioned excitation—situations where the CS signals US occurrence and produces a CR. But organisms also learn when stimuli signal US omission.

Conditioned excitation occurs when the presence of the CS predicts that the US will appear shortly. The CS excites or causes production of a CR (Pavlov, 1927).

Conditioned inhibition occurs when the presence of the CS signals that the US will not occur. The CS inhibits or prevents the expected response (Pavlov, 1927).

Pavlov’s Original Procedure for Conditioned Inhibition

Pavlov conditioned an excitatory CS (CS+) by pairing it with meat powder—standard classical conditioning. On other trials, he presented CS+ together with a second stimulus (CS-) but did not follow this stimulus pair with the US. The CS+ alone was followed by food; CS+ plus CS- together were not followed by food.

The organism learns that CS+ predicts food (excitation), but CS- signals food’s absence (inhibition). When CS- appears with CS+, it cancels the excitatory effect. The CS- becomes an active inhibitor—a safety signal indicating the US won’t occur (Rescorla, 1969).

Contingencies: Predictive Relationships

Contingency generally refers to the frequency that the CS and US are paired—the probability that the US will follow the CS. More specifically, contingency asks: Can an animal predict something about the US if the CS is present? (Rescorla, 1968).

Logic suggests three types of contingencies. Positive contingency: When CS is present, US occurrence probability increases; the CS is a CS+ (excitatory), signaling US occurrence. Zero contingency: The CS and US don’t occur together in any reliable or predictable manner; no conditioning takes place because the CS provides no information about the US. Negative contingency: When CS is present, US occurrence probability decreases; the CS is a CS- (inhibitory), signaling US omission.

This analysis reveals that conditioning is about information—organisms learn which stimuli predict important events. When stimuli provide no information (zero contingency), conditioning doesn’t occur (Rescorla, 1968).

Measuring Conditioned Inhibition

Measuring a CR to an excitatory CS is straightforward—count responses, measure their strength, assess resistance to extinction. But measuring something that doesn’t occur, like not responding to a CS-, poses problems. Researchers devised two clever procedures.

Retardation tests measure how quickly an inhibitory CS can be turned into an excitatory CS. Pair a CS- with a US and compare the number of pairings required for this CS- to become a CS+ with the pairings required for a neutral CS to become a CS+. Because of its initial inhibitory effect, the CS- retards formation of the new CS-US association. The CS- requires more pairings; the number of extra pairings measures conditioned inhibition (Rescorla, 1969).

Summation tests present a CS- and CS+ simultaneously to determine how much the CS- counters the CS+ effects. If the CS- is truly inhibitory, it should reduce responding below what the CS+ alone produces. The degree of suppression measures inhibition strength (Rescorla, 1969).

These procedures reveal that inhibition is an active process, not simply absence of excitation. Organisms actively learn safety signals—stimuli predicting danger’s absence.

Looking Forward

We’ve explored sophisticated aspects of classical conditioning: generalization and discrimination, higher-order conditioning, sensory preconditioning, sign tracking, conditioned emotional responses, and conditioned inhibition. These phenomena demonstrate classical conditioning’s power and flexibility as a fundamental mechanism by which organisms learn predictive relationships in their environments. This completes our survey of classical conditioning basics—Pavlov’s discovery, basic procedures and terminology, factors affecting acquisition and extinction, types of USs, and advanced conditioning phenomena. In Module 05, we’ll explore contemporary theories and research on classical conditioning, examining modern perspectives on what organisms actually learn during conditioning and how biological constraints shape what can be learned.