Lectures 9&10: Pavlovian
Conditioning (Major Theories)
Learning, Psychology 3510
Fall, 2015
Professor Delamater
Pavlovian Learning
Three Key Questions
1. What are the major determinants of learning?
2. What is the content of learning?
3. How does learning get translated into performance?
Pavlovian Learning: Determining
Conditions
1.
2.
3.
4.
5.
6.
7.
8.
9.
Stimulus Novelty (CS, US)
Stimulus Intensity (CS, US)
Spatial Contiguity
Temporal Contiguity
Relative Temporal Contiguity
CS-US Contingency
US Surprise
Relative Cue Validity
CS-US Relevance (Belongingness)
There are lots of determinants of Pavlovian learning, as we have seen.
The key question now is to see if we can find some general
theoretical statements that could help explain why all of these
factors are important.
Rescorla-Wagner Model: Basis Ideas
1. Learning is quantified by a variable called
“associative strength.”
2. Associative strength is assumed to change on
each conditioning trial.
3. Learning will occur when the US is surprising, and
will also be modulated by the “saliency” of the stimuli.
4. The total expectation of the US will equal the sum
of the associative strengths of all the stimuli present
on a conditioning trial.
Rescorla-Wagner Model: Equation
VV)
CS
US
V
• Learning is viewed as a change in the associative
connection between CS and US representations.
• The Rescorla-Wagner model can be viewed as a
formula that illustrates how this connection changes.
Rescorla-Wagner Model: Equation
VV)
V: This refers to the change in associative strength on a conditioning trial.
,:
:
These refer to the salience of CS and US, respectively. 0 < , < 1
This refers to the value of the actual US presented on a conditioning
trial. Usually,  = 1 when the US is presented and  = 0 when it is
not presented. However, a more intense US will have a higher value
of  than a less intense US.
V:
This refers to the sum of the associative strengths of all stimuli present.
V):
This term refers to “US Surprise” (Actual US – Expected US)
Rescorla-Wagner Model: Acquisition
VV)
Rescorla-Wagner Model: Acquisition
VV)
alpha * beta = .2
Trial 1:
Trial 2:
Trial 3:
Trial 4:
Sum V
0
0.2
0.36
0.488
Delta V
0.2
0.16
0.128
0.1024
Trial n:
1
0
Rescorla-Wagner Model: Blocking
VA AV)
VB BV)
where V  VA + VB
Gp 1
Gp 2
Phase 1
A - US
A | US (u)
Phase 2
AB - US
AB - US
Test
B?
B?
VA =  by the end of Phase 1 for Gp 1.
VA + VB =  from the beginning of Phase 2, in Gp 1. Therefore VB = 0.
In other words, there is no new learning to B in Gp 1.
Rescorla-Wagner Model: Overexpectation
VA AV)
VB BV)
where V  VA + VB
VA = VB =  by the end of Phase 1.
VA + VB = 2 in Phase 2, but the US = .
So, (V) = - in Phase 2. This means that A and B will LOSE
associative strength because the US is “overexpected” on the trial.
Rescorla-Wagner Model: Conditioned
Inhibition & Extinction
VA AV)
A+, AXProcedure
VX XV)
where V  VA + VX
A+
AX-
X-
VA =  by the end of training.
VB = - by the end of training.
So, (V) = 0 on AX- trials by the end of training.
Both VA and Vx will lose their associative strengths in extinction.
Other Models: Mackintosh (1975), Pearce
& Hall (1980)
• These models also view learning in terms of changes
in associative strength
• But they suggest that CS salience, , changes with training.
• Mackintosh:  increases as the CS becomes the best predictor
• Pearce & Hall: decreases as the CS accurately predicts the US
Mackintosh Account of Blocking
Mackintosh (1975) & Blocking
Attention increases to the stimulus that best predicts
the outcome, but decreases if it is not the best predictor.
CS1-US
CS1 | US
CS1+CS2-US
CS1+CS2-US
CS2?
CS2?
• Attention to CS2 decreases during 2nd phase in the
pretrained group because CS1 is a better predictor
of the US.
• BUT, blocking should not occur on the first compound trial.
Mackintosh Account of Blocking
Mackintosh & Turner (1971) & Blocking
Attention increases to the stimulus that best predicts
the outcome, but decreases if it is not the best predictor.
Gp 1:
Gp 2:
Gp 3:
Tone – Sh
Tone – Sh
TL – Sh
TL – Sh
L?
L?
• Gp 3 should show blocking, but Gp 2 should shows
some learning to L because the strong shock was not fully
predicted.
Mackintosh Account of Blocking
Mackintosh & Turner (1971) & Blocking
Attention increases to the stimulus that best predicts
the outcome, but decreases if it is not the best predictor.
Gp 1:
Gp 2:
Gp 3:
Tone – Sh
Tone – Sh
Tone – Sh
TL – Sh
TL – Sh
TL – Sh
TL – Sh
• But what should happen in Gp 1?
L?
L?
L?
Mackintosh Account of Blocking
Mackintosh & Turner (1971) & Blocking
Attention increases to the stimulus that best predicts
the outcome, but decreases if it is not the best predictor.
Gp 1:
Gp 2:
Gp 3:
Tone – Sh
Tone – Sh
Tone – Sh
TL – Sh
TL – Sh
TL – Sh
TL – Sh
L?
L?
L?
• Gp 3 shows blocking, but Gp 2 shows some learning to L.
• However, Gp 1 shows very little learning to L because they
learned to ignore it as it was being blocked in Phase 2.
Pearce & Hall: Salience Reductions
Pearce & Hall (1980)
Attention decreases to the stimulus as it well predicts
the outcome.
Hall & Pearce (1979)
Gp 1: Tone - weak shock
Gp 2: Light - weak shock
Tone - Strong Shock
Tone - Strong Shock
• Attention to Tone decreases as it predicts weak shock
in the first phase of training in Gp 1
• Gp 1 learns to associate Tone with Strong Shock
slowly (inconsistent with Mackintosh)
Summary
• There are problems with all models
• But such models give us a framework for thinking about
how conditioning works, and also for organizing lots of
results from different experiments.
• These models, together, account for many of the facts
we considered last time (as determining conditions).
However, some of those phenomena are not fully
captured by these models (e.g., CS-US relevance,
spatial contiguity, temporal contiguity, etc).
• A more complete understanding will require additional
considerations.