Stimulus Discrimination & Generalization
• Discrimination: Responding differently to two or more stimuli
• Generalization: Responding in a similar fashion to two or more stimuli
• Generalization Gradient: a gradient of responding that is observed if that participant is tested
with stimuli that is increasingly different from stimuli present during training. (Diagram on the
• Prestudy: Trained pigeons to respond to a compound stimulus for food ( the stimuli was
combination of thing like triangle and circular red background)
• Procedure: Tested with the individual components of this stimuli
• Results: Stimulus parameters that controlled responding was not equivalent. For different
pigeons different stimuli became the cue for food and they responded more to that.
• Stimuli control: is demonstrated by the variation in responding (differential responding)
related to variation in the stimuli ( respond one way in one stimuli and then another way in
Stimulus Control of Spatial Search: Spetch & Mondloch, 1993
• Methods: Trained pigeons in a touchscreen task. Had to peck at a target area of 2 cm 2
• Target area defined by the landmarks. Tested to determine the landmarks that controlled the discrimination
• Results: Most birds showed control by only 1 or 2 of the landmarks, and some birds displayed surprisingly accurate search with a single
landmark. For individual birds, landmark-removal tests were very consistent with landmark-shift tests in indicating which landmark or
landmarks controlled search. However, the dominant landmark varied across birds. Manipulation of landmark color and shape revealed
that control was based exclusively on color.
What affects stimulus control?
Sensory Capacity and Orientation:
• Stimulus control depends on sensory capabilities of each animal. For example, could not use bat ultrasound as stimuli for human
• Also depends on the organisms’ orientation (i.e. you are not looking at it or facing it then chances are you are not going to be controlled
by it since you are not in contact with it)
Not All Stimuli Are Created/Discriminated Equivalently
• Some stimuli are more easily used as discriminative stimuli than others
• e.g., training songbirds to discriminate natural stimuli goes more quickly than training synthetic stimuli
• Can have overshadowing if the wrong stimulus is used as a stimulus (i.e. if stimuli hard to condition)
• Overshadowing: the interference with conditioning of stimuli because of the presence of another stimuli that is more easier to
• See slide 12.
• Overshadowing of Tones by Light: all the birds were given tone and their responses were measured. In Dark (when their was no light
present) the bird seem to become clearly under the stimuli control of the sound (specially around the 1.5Hz frequency we see a huge
• BUT, for the same tones, when light was also present, the responses were all straight lines suggesting light instead now controls the
responses and the variation is tone does not matter.
Overshadowing in Landmark Learning: Spetch, 1995
• Participants: Pigeons and undergraduates
• Touch screen task to study landmark learning
• Tested for stimulus control of landmarks and potential factors controlling overshadowing
• Results: when the birds were not overshadowed ▯ the responses were very steep at a particular point of the touch screen (thus showing a
clear stimulus control).
• But, when overshadowed ▯ the responses were all over the place thus suggesting that the landmarks now did not generated any stimulus
How do animals treat Compound Stimuli?
• Stimuluselement approach : organisms treat stimulus elements as distinct and separate features of the environment (i.e. sound will be
perceived the same way regardless of location)
• Configuralcue approach : the CCA assumes that organisms treat stimuli as integrated wholes (i.e. the sound of orchestra is very
different from the individual instruments we hear)
Effect of Different Reinforcers
• Degree of stimulus control also depends on the type of reinforcement used
• Some stimuli work better when contingent with positive reinforcement, others with negative reinforcement
Foree and LoLordo (1973) methods:
• Pigeons trained to press a treadle in the presence of a compound stimulus • In one condition the response was made to receive food. In the other condition the response was made to avoid shock▯Tested to see
which component of the stimulus controlled behavior
• Results: avoidance condition▯ animals are responding more to ToneResults 2: appetitive condition▯ ore responding when light is
• MP: Some stimuli work better when contingent with positive reinforcement, others with negative reinforcement
What about the Instrumental Response?
• Type of response required can affect stimulus control
• Study: Dobrzecka et al. (1966) trained dogs to use different instrumental
• Procedure: One group trained on location of response. (i.e. sound from the
front▯right food should be raised or the back ▯ left food should be ragroup
• Other group trained on quality (group 2) of response (either being a
buzzer▯raise leg or the metronome▯no leg)
• Results: In the test ▯ the locations of the sounds were switch ▯ they continued to
do depending on what they were trained (either by the quality or the location) to
make the correct response.
• MP: the qualitylocation effect: responses that are likely to be differentiated by
the location (left or right) are more likely to come under the control of spatial
feature of auditory cues. In contrast, responses that are differentiated by quality are more likely to come under the auditory cues.
Learning & Stimulus Control
• Different views about why generalization exists
• Pavlov: generalization caused by similarity of a stimulus to the original CS
• Lashley and Wade (1946): generalization gradients reflected the absence of learning
Discrimination Training and Stimulus Control
• To examine the degree of stimulus control, examine the generalization gradient ( the more steep it is the more the stimulus control)
• Need to determine the feature of discrimination procedure that controls the gradient
Landmark study by Jenkins and Harrison (1962) study:
Note: S+ ▯ mean you make a response. And S ▯mean do not make a response.
• Procedure: 3 groups of pigeons.
• Group 1: S+ ▯1000Hz tone that is reinforced if you peck and S was ▯no tone and no reinforcement.
• Group 2 ▯ S+ ▯1000Hz tone and were reinforced if you peck and S was ▯950Hz and no reinforcement
• Group 3 ▯ no discrimination training ( control)
• After the completion of the training the groups were tested for pecking response in various frequencies.
• Group 1 ▯ normal generalization curve was seen.
• Group 2 ▯ because it was intradimensional training (frequencies were uses) ▯ had the steepest generalization curve suggested a lot of stimulus
• Groups 3 ▯ no training to equal response to all of the frequencies.
• Note: the peak of the group 1 was higher then group 2 despite the steepness.
• MP: instrumental behaciour and the particular stimuli ( i.e. frequency) is more likely to gain control over the responding if the S+ and S
differ along the same dimension ( hense we see a steep line).
What the heck is learned in discrimination training?
• Spence (1936) thought that animals learned both about S+s and Ss. but simply observing that animal responded more to S+ then S is
not enough. – S+ = excitation ▯expected to acquire excitatory properties
– S = inhibition ▯ expected to acquire inhibitory properties.
– Generalization gradients(+/ )for different training regimes were looked at to test this hypothesis.
• Procedure: see the description in Fig 8.10:
• Group1: generalization gradients ▯excitation
• Group 2: The exact same thing/stimuli in his group is associated with i