WEEK 8 LECTURE: STIMULUS DISCRIMINATION AND GENERALISATION

Part 1 – Generalisation vs. Discrimination

• Stimulus control = any situation in which the probability, form, or intensity of behaviour changes systematically with the presence/absence of a particular stimulus.
• Classical Conditioning (CC): CR only when CS present.
• Instrumental Conditioning (IC): Response only in presence of discriminative stimulus (S+).
• Everyday examples: urge to smoke when drinking coffee; desire to drink alcohol around friends.

• Stimulus Discrimination
• Definition: organism responds differently to two or more stimuli.
• Illustration: different reactions when one’s mother vs. friend calls one’s name.
• Formal criterion: variations in CR as a function of variations in the eliciting stimulus.
• Reynolds (1961) pigeon study:
– Pigeons trained to peck a coloured, patterned key.
– Tested with components separately (e.g., colour-only vs. shape-only).
– Result: some birds controlled by colour, others by pattern ⇒ individual differences in selective attention & unpredictability of which stimulus element will dominate.

• Stimulus Generalisation
• Opposite of discrimination: similar responding to two or more stimuli.
• Discovered by Pavlov—dogs salivated to tones similar to training tone.
• Generalisation is graded by physical similarity to training stimulus.
• Generalisation Gradient: plot of response strength vs. systematic stimulus variation (e.g., wavelength).
– Steep gradient ⇒ strong stimulus control.
– Shallow gradient ⇒ weak control.

Part 2 – Factors That Influence Behavioural Control: Stimulus & Reinforcement Variables

• Multidimensional nature of stimuli
• Attributes: size, brightness, shape, colour, location, pitch, etc.
• More complex stimulus ⇒ more potential controlling elements (which one will win is empirical).

• Sensory capacity & orientation
• Stimulus must be detectable. Humans cannot hear >20{,}000\,\text{Hz}, but dogs can; colour-blind individuals cannot discriminate colours as CSs.
• Discrimination/generalisation tests reveal perceptual limits.

• Overshadowing & competition
• Compound stimuli: more salient element usurps control.
• May occur between distinct stimuli (light vs. tone) or between attributes of one object (colour vs. size).

• Reinforcer type modulates which modality dominates
• Foree & LoLordo (1973) pigeons: red light + tone compound →
– Group F (food): responding controlled mainly by light (visual).
– Group S (shock avoidance): responding controlled mainly by tone (auditory).
• Implication: visual cues pair better with appetitive US; auditory with aversive US (cf. Garcia’s ‘bright-noisy water’).

• Konorski’s quality–location effect
• Dogs trained with buzzer vs. metronome in different positions.
• Group 1 response = raise right vs. left leg (spatially distinct) → stronger control by spatial location.
• Group 2 response = raise leg vs. not (qualitatively distinct) → stronger control by sound identity.
• Take-home: stimulus dimension that maps most naturally onto response dimension gains control.

• Stimulus-Element approach (traditional)
• Assumes behaviour governed by separable components of a stimulus.
• Competition (overshadowing, blocking) viewed in terms of elements vying for associative strength.

Part 3 – Elemental vs. Configural Processing

• Configural-cue approach (Gestalt-inspired)
• Entire compound encoded as unique configuration; individual elements lose separate identity.
• Stimulus representation = pattern, not pieces.

• Pearce (1987) Configural Model
• Each learning situation stored as single node comprising CS + context.
• Generalisation governed by similarity between test pattern & stored pattern.
• Key phenomena:
– Generalisation decrement: removing an element from training compound → \downarrow CR.
– External inhibition: adding novel element → \downarrow CR.

Part 4 – Learning Factors Influencing Stimulus Control

• Discrimination procedures steer organism toward elemental or configural solutions.

• Positive Patterning (PP) vs. Negative Patterning (NP)
• PP: AB+ / A- / B- (can solve elementally).
• NP: C+ / D+ / CD- (requires configural processing because elements alone predict reinforcement).
• Successful NP indicates subject perceives CD as distinct whole.

• Relative Stimulus Validity (RSV)
• Training: AB+ / BC-
• Cannot be solved configurally; must track individual cues (A excitatory, C inhibitory).
• Demonstrates that discrimination history shapes attentional weighting.

• Stimulus Discrimination Training (SDT) basics
• In CC: \text{CS+} \Rightarrow \text{US}, \text{CS-} \nRightarrow \text{US}.
• In IC: \text{S+} \Rightarrow \text{Response} \Rightarrow \text{Reinforcer}, \text{S-} \Rightarrow \text{Response} \Rightarrow \varnothing.
• Simultaneous presentation (visual) eases learning; sequential harder, especially with inter-trial mask.

• Intradimensional discrimination
• S+ and S- differ along single continuum (e.g., wavelength).
• Experts (wine tasters, mechanics, shepherds) rely on refined intra-dimensional skills.
• Produces steeper generalisation gradients than no-training or extradimensional training.

• Lashley & Wade (1946)
• Argue generalisation is default; discrimination must be learned.
• Hence gradient shape ≈ reflection of learning history, not mere stimulus physics.

• Spence’s theory (excitatory & inhibitory gradients)
• Reinforcement at S+ → excitatory generalisation centred at S+.
• Non-reinforcement at S- → inhibitory gradient centred at S-.
• Net responding = summation of excitation – inhibition.
• Explains peak-shift effect: when S+ and S- close, inhibitory ‘shadow’ suppresses responses at S+, shifting peak away from S-.

• Peak-Shift empirical data (Hanson 1959)
• Pigeons: train S+ = 550\,\text{nm}; manipulate S-.
• Closer S- (e.g., 555\,\text{nm}) ⇒ larger shift & higher peak at stimulus <550\,\text{nm}.
• No shift when S- far or absent.

• Acquired Equivalence
• If two stimuli consistently paired with same outcome, they become functionally interchangeable.
• Experiment: Tone→food & Buzzer→food (Phase 1); Tone→shock (Phase 2).
– Group where buzzer previously paired with food shows CR to buzzer even though buzzer never paired with shock ⇒ transfer via equivalence class.
• Facilitates category learning (APPLE ≈ BANANA ⇒ ‘fruit’).

• Acquired Distinctiveness
• Training non-reinforced compounds AB- / AC- enhances discriminability between unique elements B vs. C.

Part 5 – Contextual Cues & Conditional Relations

• Contexts as stimuli
• Room, lighting, background odour can modulate behaviour without directly signalling reinforcement.
• Thomas et al. study:
– Context 1: S+ = vertical (90°), S- = horizontal (0°).
– Context 2: contingencies reversed (S+ = 0°, S- = 90°).
– Both contexts contained reinforcement & non-reinforcement equally; yet each context selectively retrieved its own CS–US relation ⇒ contextual modulation.

• Conditional (occasion-setting) relations
• Require 3 events: Modulator (M), Target cue (X), Outcome (US).
• Positive Occasion Setter (POS): \text{M}-!!\text{X}+ / \text{X}- → M enables X→US.
• Negative Occasion Setter (NOS): \text{M}-!!\text{X}- / \text{X}+ → M disables X→US.
• In instrumental tasks, discriminative stimulus (S+) is de facto occasion setter for response–reinforcer contingency.

Summary Key Points

• Stimulus control manifests as discrimination (respond differently) or generalisation (respond similarly).
• Control strength measured through generalisation gradients; steepness reflects sensitivity.
• Stimulus salience, sensory capacity, reinforcement type, and competition (overshadowing) determine which cue dominates.
• Elemental vs. configural processing: organisms can parse components or encode holistic patterns; procedure dictates strategy (e.g., NP forces configural representation).
• Learning histories (SDT, RSV, intra- vs. inter-dimensional training) sculpt attention & associative strength, producing phenomena such as peak-shift.
• Equivalence and distinctiveness training alter perceived similarity between cues, shaping generalisation.
• Contexts and occasion setters modulate cue–outcome relations without being direct predictors themselves.
• Understanding these principles aids in designing interventions (e.g., reducing relapse cues, enhancing category learning, tailoring exposure therapy).