Chapter 10
Generalization, Discrimination, and
Stimulus Control
Variability
• Changing conditions
• Adaptive learning must adapt
• Transfer behaviour across situations
Generalization
• Tendency for a learned behaviour to occur
in the presence of stimuli not present during
training
Discrimination
• Tendency for a learned behaviour to occur
in the presence of certain stimuli, but not in
their absence
Stimulus Control
• Stimuli come to exert influence over
behaviour
• Application of generalization and
discrimination
• CS+ and CS• S+ and S– S+ indicates more reinforcing outcome, S- less
reinforcing outcome
Discrimination Training
• Any procedure that establishes the ability to
discriminate between stimuli
• Process by which stimulus control is
established
Examples
• Pavlovian
– Producing CR for stimulus slightly different
than trained CS
• Operant
– Train in one puzzle box, test in variant box
Generalization
• Not a given
• Can increase generalization by training in a
variety of settings
• Generalization not always appropriate or
useful (e.g., generalizing violence from
video game to real world)
Generalization Gradients
•
•
•
•
Measure of generalization/discrimination
Respond to stimuli more like trained stimuli
Train on one stimulus, test on others
Techniques/methodologies
Probe Trials
• Insert occasional unreinforced test stimulus
training stimulus
(reinforced)
trials
probe stimulus
(unreinforced)
• Won’t extinguish
Extinction Blocks
•
•
•
•
Train stimulus to asymptote
Blocks of extinction trials
Each stimulus presented once/block
Extinction constant across stimuli
Extinction
Blocks
# of responses
4+3+1 = 8
20+15+10 = 45
15+11+5 = 30
12+7+3 = 22
3+1+0 = 4
Training
Block 1
Generalization Gradient
4 20 15 12 3
50
7
3 1 15 11
Block 3
0 5 10 1 3
and so on...
Responses
Block 2
40
30
20
10
Light wavelengths
Response rate
Response rate
Response rate
Reading a Generalization
Gradient
Stim. continuum
Flat:
No discrimination/
high generalization
Stim. continuum
Stim. continuum
Broad:
Narrow:
Some discrimination/ High discrimination/
some generalization low generalization
Semantic Generalization
• Doesn’t have to be a perceptual stimuli
• Generalization of abstract feature
• Adults ate candy (US) to salivate (UR) while
shown words (style, urn, freeze, surf)
• Shown homophones (stile, earn, frieze, serf)
• Shown synonyms (fashion, vase, chill, wave)
• CRs for homophones, but very strong CRs for
synonyms
Generalization Post Extinction
• Operant training, then extinction
• Produces reduction in generalization to
other stimuli
Generalization of Punishment
• Suppression of behaviour via punishment
also generalizes
• Honig & Slivka (1964)
• Pigeons peck plain disk, get reinforced
• Peck, e.g., green disk, get shocked
• Test with other coloured disks
• Greatest reduction of pecking to “greener”
colours
Different Discrimination Training
Techniques
Presence/Absence Training
• Successive Discrimination Training
– S+ & S- alternate randomly (S+ --> reinf., S- -->
extintion)
• Simultaneous Discrimination Training
– S+ & S- presented at same time
• Matching to Sample (MTS)
– Select from 2+ alternatives
(comparison stimuli) the stimulus
that is the same as the sample
Training
Testing
S+
• Mismatching
– Like MTS, but pick comparison
stimulus not like sample
• Delayed Matching to Sample
(DMTS)
– Like MTS, but delay between
presentation of sample and
choice
Training
Testing
--Delay period-S+
Errorless Discrimination Training
• Previous techniques slow
• Many mistakes where S- selected
• Present S+ as normal, but start S- at low
salience (short time and “faint”)
• Gradually increase salience of S- to equal S+
• Quick, relatively little frustration for Schoice, greater discrimination learned
Differential Outcomes Effect
•
•
•
•
Reinforcers available for different responses
But, reinforcers are different
Learn multiple three-term contingencies
Can produce faster and stronger
discrimination training than basic forms
Theories of Generalization and
Discrimination
Pavlov’s Theory
• Physiological interpretation
• Species influenced
• Discrimination training produces establishes
areas of activation in brain
• CS+ --> excitatory regions
• CS- --> inhibitory regions
Activation
• Stimuli similar to CS+ will excite parts of
brain close to CS+ area
• Dissimilar stimuli will not activate CS+
area
• Result is CR or no CR, respectively
Inferential Interpretation
• Theory based on inference from observed
behaviour
• No independent validation of brain area
generation through conditioning
• Physical proximity of brain areas not
needed for response generation
Spence’s Theory
• Opponent process theory
• Excitatory (CS+ or S+) and inhibitory (CSor S-) gradients
• Net sum effect of gradients
• Resultant behaviour
Peak Shift
• Change in generalization gradient
• Peak level of responding
• Shift in peak level of responding away from
S+ in direction opposite S-
Peak Shift
Responses
shift direction
Control (S+ only)
Exp. 1 (S+ & S-)
S+
S-
Peak Shift: Shift Away from SS+ SNet
gradient
+15
+10
+5
Excitatory
gradient
+15
+10
+5
Inhibitory
gradient
-5
-10
-15
Support for Spence’s Theory?
• Honig et al. (1963)
Responses
– Excitatory and inhibitory gradients
Group 1
S+
SGroup 2
S+
S-
Lashley-Wade Theory
• Generalization gradients depend on prior
experience with stimuli similar to those
used in testing
• Discrimination training --> discrimination
because it teaches subjects to tell the
difference between S+ and other stimuli
• Everyday experiences produce
discrimination learning
Predictions
• Previous experience with stimuli will make
discrimination training of those stimuli
easier
• Lack of previous experience will make
subsequent training harder
Standard Design
• Rear animals under specific environmental
condition
– e.g., darkness so no experience with colours
• Give S+/S- training
• Test for generalization gradient
• If gradient of perceptually deprived subjects
flatter than normally reared subjects, then
support for Lashley-Wade theory
Results
• Ambiguous
• Possibility that special rearing environment
produces neurological damage
Jenkins & Harrison (1960)
• Group 1 pigeons
– S+ (tone) --> reinf., S- (quiet) --> no reinf.
• Group 2 pigeons
– S+ (tone) --> reinf., no S- (i.e., tone always on)
• Test both groups for generalization to other
tones and to periods of silence
Results
• Group 1 birds
– Less likely to respond during silent periods
– Show standard generalization gradient to tones
• Group 2 birds
– Responded same amount during tone or silence
– Flat generalization gradient (i.e., no
discrimination of tones)
• Supports Lashley-Wade theory
Theories
• Pavlov’s
– Lacks support
• Spence and Lashley-Wade
– Both have situations that support and contradict
predictions
Applications
Concept Formation
• Concept: any class of things sharing one or
more defining features
• Defining features allow discrimination
between stimuli within class and outside
class
• Concepts can be learned through
discrimination training
Herrnstein’s Studies
• Stimuli from natural environment
• Train/test many stimuli
– Positive and negative instances
• Pigeons, 80 pictures
– Tree/no tree = positive/negative instances
• Learn discrimination easily
• Generalization test
– Supports concept formation, not memorization
Concepts of Absolute or Relative
• Concept of absolute
– Learn individual stimuli
– Specify features of members of class
• Concept of relative:
– Learn relationship between stimuli
– Degrees of similarity of features of class
members
Example
Transposition
• Transfer relational rule to new stimuli set
• Kohler (1939)
Training
S+
S-
Test
transfer
absolute
Stimulus Control
• Absolute stimulus control
– Successive discrimination tasks
• Relational stimulus control
– Simultaneous discrimination tasks
• Animals do whatever is easiest
Smoking Relapse
• Smoking gives frequent reinforcement
• But, not only physiological effects of
nicotine
• Social reinforcement
• Environmental factors become conditioned
as S+ for smoking
• Smoke in many situations, strong
generalization
Experimental Neuroses
• When not possible to distinguish between
stimuli in discrimination conditions
• Can’t establish stimulus control
• Consumer situations
– Frustration
– No-choice as option