2.3 Higher-Order Conditioning, Generalization & Discrimination

2.3 Higher-Order Conditioning, Generalization & Discrimination

Animals do not encounter environmental stimuli in sterile isolation. In the real world, domestic dogs operate within rich, dynamic sensory landscapes where multiple sights, sounds, and scents overlap, compete, and compound. Understanding how the brain processes complex stimulus configurations—and how prior conditioning alters future learning—is essential for diagnosing training errors and building reliable behavioral responses.

---

Second-Order and Higher-Order Conditioning

In standard primary conditioning (first-order conditioning), a neutral stimulus is paired directly with an unconditioned stimulus (NS + US → CS1 → CR). However, associative learning extends beyond primary biological reinforcers.

Higher-order conditioning is a process wherein an already established conditioned stimulus (CS1) functions in the role of an unconditioned stimulus to condition a completely novel neutral stimulus (CS2), without the primary unconditioned stimulus ever being presented.

Phase 1 (First-Order):    CS1 (Leash) + US (Outdoor Walk) ------> CR (Excitement)
Phase 2 (Second-Order):   CS2 (Running Shoes) + CS1 (Leash) ----> CR (Excitement)
Test Phase:               CS2 (Running Shoes Alone) ------------> CR (Excitement)

Mechanics and Inherent Limitations

• Second-Order Conditioning: An owner grabs running shoes (CS2) immediately before picking up the leash (CS1), which has historically led to walks (US). Soon, the sight of running shoes alone elicits intense anticipatory excitement (CR), despite running shoes never being worn on walks or directly paired with outdoor access.
• Third-Order Conditioning and Extinction Risks: In theory, a third stimulus (CS3) could be paired with CS2. However, in practice, conditioning beyond the second order is exceptionally fragile in non-human animals. During Phase 2, CS1 is repeatedly presented in the complete absence of the primary US. Consequently, higher-order pairings simultaneously initiate respondent extinction of CS1. If non-reinforced trials continue, the entire associative chain collapses.

---

Sensory Preconditioning

While higher-order conditioning pairs a new stimulus with an already conditioned stimulus, sensory preconditioning demonstrates that animals form latent cognitive associations between neutral environmental events without any overt reinforcement.

The Experimental Paradigm

1. Phase 1 (Preconditioning): Two completely neutral stimuli (NS1 and NS2) are repeatedly paired together in the environment (NS2 → NS1). Because neither stimulus carries biological significance, the animal displays only minor orienting behavior; no reward or punishment occurs.
2. Phase 2 (Conditioning): Stimulus NS1 is formally conditioned using a primary unconditioned stimulus (NS1 + US → CS1 → CR).
3. Test Phase: Stimulus NS2 is presented alone for the first time since Phase 1.

Result: The animal exhibits the conditioned response (CR) to NS2, despite NS2 having never been directly paired with the unconditioned stimulus or with an active conditioned stimulus.

Canine Training Example

A puppy regularly hears a distinct metal gate latch click (NS2) immediately followed by the home's front doorbell chime (NS1). At this stage, neither sound carries emotional valence. Months later, the owner pairs the front doorbell (CS1) with exciting arriving visitors (US), conditioning loud barking and frantic arousal (CR). The next day, the dog hears the distant garden gate latch click (CS2) and immediately begins barking frantically in the living room. Sensory preconditioning proves that dogs build complex cognitive maps of environmental contingencies passively.

Stimulus Generalization and Generalization Gradients

Once a conditioned stimulus reliably elicits a conditioned response, the animal does not restrict its responsiveness solely to that identical physical stimulus. Stimulus generalization is the automatic tendency for an organism to emit a conditioned response to novel stimuli that share physical, acoustic, or visual similarities with the original conditioned stimulus.

The Generalization Gradient

When response frequency or magnitude is plotted on a graph against systematic physical variations of the stimulus (such as sound frequency in Hertz, light wavelength, or visual size), the resulting plot is a generalization gradient:

• Steep Gradient: Indicates high discriminative precision. The animal responds vigorously to stimuli that closely match the training CS, but response rates drop precipitously as the stimulus varies even slightly.
• Flat / Broad Gradient: Indicates broad generalization and low discrimination. The animal responds similarly across a wide band of stimulus variations.

Response
Strength
   ^
   |            /\  <-- Steep Gradient (High Discrimination)
   |           /  \
   |       ---/----\--- <-- Flat Gradient (Broad Generalization)
   |      /   |    |   \
   +-----+----+----+----+-----> Stimulus Dimension (e.g., Audio Frequency)
             CS-   CS+  CS-

Clinical and Evolutionary Significance

Generalization possesses immense evolutionary survival value: a canine attacked by a rattlesnake must generalize that aversive association to all rattlesnakes, regardless of subtle differences in length, rattle pitch, or scale coloration. However, in domestic companion dogs, maladaptive generalization underlies severe behavioral pathologies:

• A puppy traumatized by a tall man wearing a wide-brimmed hat may generalize fear to all tall men, then to all people wearing hats, and ultimately to any person carrying an umbrella or wearing a bicycle helmet.

---

Stimulus Discrimination (Differential Conditioning)

Stimulus discrimination is the complementary behavioral process to generalization. It occurs when an animal learns to respond specifically to a target conditioned stimulus (CS+) while withholding or attenuating the response to alternative, non-reinforced stimuli (CS-).

Differential Conditioning Protocol

Discrimination is systematically developed through differential reinforcement:

• CS+ Trials: The target stimulus is consistently followed by the unconditioned stimulus (CS+ → US).
• CS- Trials: Perceptually similar alternative stimuli are presented without the unconditioned stimulus in extinction (CS- → No US).

Over successive trials, differential conditioning sharpens the generalization gradient, collapsing responding to non-target stimuli down to zero baseline while maintaining robust responding to the CS+.

Canine Applications

• Target Odor Detection: In scent detection (narcotics, explosives, conservation tracking), a dog must show sharp discrimination for the precise target volatile organic compounds (CS+) while ignoring closely related hydrocarbons, industrial packaging plastics, or competing food distractor odors (CS-).
• Household Management: Teaching a dog that a doorbell chime on television (CS-) carries no consequence, whereas the physical front door intercom (CS+) predicts visitor arrival.

---

Compound Stimuli: Overshadowing

When two or more distinct sensory stimuli are presented simultaneously as a compound package and paired with an unconditioned stimulus, they do not condition equally. Overshadowing occurs when one stimulus element possesses significantly greater salience (intensity, biological prominence, or perceptual conspicuousness) than the concurrent element, resulting in the more salient stimulus acquiring the vast majority of associative strength while the weaker stimulus acquires little or no conditioned responding.

Compound: [A (Salient) + B (Weak)] → US ===> Test A: Strong CR; Test B: No CR

The Rescorla-Wagner Explanation

Under the Rescorla-Wagner model (V_compound = V_A + V_B), the change in associative strength is divided between compound elements based on their individual salience parameters (α_A and α_B). If stimulus A has high salience (α_A = 0.8) and stimulus B has low salience (α_B = 0.1), stimulus A absorbs almost the entirety of λ, leaving stimulus B with negligible associative value.

The Classic Dog Training Blunder

A handler simultaneously gives a large physical hand gesture (e.g., raising an arm) while uttering a verbal command ("Sit"). Canines possess predatory evolutionary heritage highly attuned to biological motion, making physical gestures vastly more salient than human speech phonemes. The conspicuous visual movement overshadows the verbal word. When the handler eventually tests the verbal cue with their hands motionless behind their back, the dog stares blankly. The handler incorrectly labels the dog as "stubborn" or "disobedient," failing to recognize that the verbal cue never acquired associative strength.

---

Kamin's Blocking Effect

Discovered by American psychologist Leon Kamin in 1969, blocking is one of the most critical phenomena in learning science, conclusively proving that temporal contiguity is insufficient for learning.

Kamin's Experimental Design

1. Phase 1 (Prior Conditioning): Stimulus A (e.g., an 80 dB tone) is paired with an unconditioned stimulus (shock or food) until conditioning reaches asymptote (A → US, such that V_A = λ). Stimulus A fully predicts the US.
2. Phase 2 (Compound Conditioning): Stimulus A is presented simultaneously with a novel neutral stimulus B (e.g., a flashing light) as a compound stimulus, and paired with the exact same unconditioned stimulus ([A + B] → US).
3. Phase 3 (Testing): Stimulus B is presented alone to evaluate associative strength.

Result: Stimulus B elicits zero conditioned responding. Despite stimulus B undergoing dozens of contiguous, perfectly timed pairings with the unconditioned stimulus, prior conditioning to stimulus A completely "blocks" conditioning to stimulus B.

The Theoretical Mechanism: Zero Prediction Error

Applying the Rescorla-Wagner model illustrates why blocking occurs: ΔV_B = α_B β (λ - V_total)

In Phase 2, cumulative associative strength is already saturated by stimulus A (V_total = V_A = λ). Therefore: (λ - V_total) = (λ - λ) = 0

Because the unconditioned stimulus is already 100% anticipated by stimulus A, the arrival of the US carries zero surprise or prediction error. Without prediction error, the central nervous system does not update synaptic weights; stimulus B acquires zero associative strength.

Transfer of Stimulus Control in Dog Training

Blocking has massive ramifications for adding or changing cues in dog training. If a dog reliably responds to a hand signal (CS_A), a trainer cannot teach a new verbal cue (CS_B) by presenting the word and the hand signal at the same time ([Word + Gesture]). The established hand signal will block the novel verbal cue entirely.

To achieve successful transfer of stimulus control, the trainer must insert a clear temporal gap:

New Cue (Verbal) → 1-Second Pause → Established Cue (Gesture) → Behavior → Reward

By presenting the novel verbal cue prior to the established gesture, the verbal cue provides new predictive information forecasting the appearance of the gesture, allowing associative strength to transfer cleanly without blocking.

Complex Conditioning Phenomena Summary