1) The document discusses an experiment that examines the effects of extinction and temporal variability in context conditioning. It uses a 2x2 factorial design where participants are conditioned to a cue in one context and then receive extinction training or no extinction in a different context along with fixed or variable intervals of unsignaled outcomes.
2) The expectation is that extinction training will lead to faster context conditioning compared to no extinction. Participants in fixed interval conditions are also expected to show predictive timing abilities during a probe trial compared to variable interval conditions.
3) The group that receives extinction training with fixed intervals is predicted to show the least suppression during the probe trial due to attention to contexts and temporal cues learned during the experiment.
A behavioral paradigm of extinction and temporal variability
1. Jade Dunstan
Professor Lamoureux
Independent Study
4 – 30 – 15
A Behavioral Paradigm of Extinction and Temporal Variability in Context Conditioning
Pairing cues and outcomes has been used since the time of Pavlov to condition people and
animals to exhibit certain behaviors. This cue-outcome relationship is mediated by timing, the cue
and outcome must occur close enough together in time for an association to be made and the cue
must occur before the outcome. Thus it appears that timing is everything, but what about context?
According to prior research the answer is yes, under specific conditions. When the cue-outcome
relationship is first established, a switch in context does not alter performance based on this
relationship. Stated another way, performance for a cue-outcome association made in context A
will be seen in context B as well if nothing else changes. However it is after extinction, when the
relationship between the CS and outcome is extinguished, that context becomes important.
Extinction occurs when the cue is repeatedly presented without the expected outcome, to the point
where the cue will no longer garner a response. Extinction is both second-learned information
(because it is learned after the acquisition of the cue-outcome relationship) and inhibitory
information (because it inhibits the performance of the response learned during the initial
acquisition).
In ABA renewal, where the cue-outcome relationship is established in context A,
extinguished in context B, and tested in context A, the pre-extinction performance is recovered
after returning to the original context (Bouton, 2004). According to the Attentional Theory of
Context Processing extinction will switch attention towards the context, making future learning
context dependent. It has been proposed that this phenomenon occurs because extinction is new
learning that makes the relationship of the CSto the US ambiguous (Bouton, Westbrook, Corcoran,
2. & Maren, 2004). This ambiguity arises because when a cue-outcome relationship is extinguished,
it is not completely eliminated. Thus extinction represents retroactive interference, in which the
learning performed in the second context interferes with the learning performed in the first context
(Bouton, 1993). In this situation the context becomes a cue that provides important insight into the
nature of this relationship. This concept of ambiguous cues resulting in context dependency
became the basis for Rosas and Callejas-Aguilera’s study on the effects of context change on the
acquisition and extinction of cues in human predictive learning.
In their first experiment, two groups were each trained to predict the probability of the
outcome given the cue (E-O1, where E = extinction) in Context A. They were then trained to
predict a relationship between a new cue and the outcome (P-O1, where p = positive cue), also in
Context A. The extinction group was trained on this P-O1 relationship while also receiving
extinction of the E-O1 relationship. For the control group, the E was replaced with F1 (F = filler),
which was presented without the outcome (O1). During the test phase, subjects in both groups
were asked to determine the probability of O1 given P1in a different (but still familiar) Context B.
The study was performed on computers with the contexts represented by two different restaurants,
E and P1 counterbalanced as cucumber and garlic, F1 as eggs, and O1 as diarrhea. Participants
indicated the probability that someone who ate a given food (E, P1, F1, or P2) would present with
diarrhea (O1) through clicking one of 21 buttons that indicated 5 point increments on a 0-100scale.
Phase 1 consisted of equal presentations of E-O1 and F1- for Context A and P2-O1and F1- in both
groups. In Phase 2 the acquisition followed the same procedure as Phase 1, except that P1 was
introduced as a new cue associated with O1. The outcome of E was changed in the extinction group
from E-O1 to E- and in the control group E was replaced by F1 and was not paired with O1. The
testing phases, in which the participants were asked for predictive judgments about the relationship
3. between P1 and O1 in Context B, occurred before Phase 1, between Phases 1 and 2, and after
Phase 2. In the last trial of acquisition participants in both groups showed high predictive
judgements in Context A, this effect persisted for the control group in context A, but was
diminished for those in the extinction group. This Group x Context interaction was significant,
suggesting that context change effects on retrieval of information may be affected by the attention
participants pay to the context in which the information was learned, which could be determined
by the ambiguity of that cue caused by new learning (Rosas & Callejas-Aguilera, 2006).
The second experiment they performed followed this same procedure, except with all target
cues (E, P1, and P2) appearing in both contexts in order to test the context change effects on first
learned vs. second learned information. In Test 1 (between Phases 1 and 2), the ratings were merely
a product of training, with higher ratings for the cues independently paired with O1 (E and P2)
than for the cue that had not yet received this training (P1). This remained the same for all contexts.
Context specificity of acquisition required that a different cue be extinguished. Cue E was not
affected by the context switch before extinction and P2 was not affected by the context change
until after extinction of E in a different context. These results suggest that the context switch seen
in human predictive learning results from a change towards attending to all contexts in the
experiment when extinction treatment starts (Rosas & Callejas-Aguilera, 2006).
Nelson & Lamoureux (2015) tested these findings through a videogame paradigm in which
participants controlled a spaceship that was fighting an enemy spaceship through firing torpedoes
(rapidly clicking the computer mouse). The contexts were two different galaxies, the cues were
different colored sensors, and the outcome was the enemy attacking, resulting in a timeout. The
timeout was calculated through a suppression ratio in which the amount of clicks in the 5 seconds
before the enemy attack was divided by 5 plus the average click rate in the 5 seconds prior to
4. stimulus presentation for the entirety of the game and then the ratio was multiplied by 120.
Message windows indicated the movement to a new galaxy, making the context switch salient.
There were six total conditions: No Extinction, With Extinction, Renewal After A, Renewal After
B, Renewal Before A, and Renewal Before B.
The game began for everyone in Context B with no cues. In Phase 1, all participants were
moved to Context A where they were conditioned using 8 trials with X (purple sensor). In Phase
2 all participants, except for those in the NoExtinction group, received extinction training in which
X was presented but not followed by an attack. The No Extinction group received no X
presentation (no visual changes during this time) with no attack. After Phase 2 training, the
Renewal Before A and Renewal Before B groups were tested with the presentation of X in Context
A or Context B respectively. The remaining 4 groups had an additional training trial (Phase 3) in
which T (yellow light) was conditioned in Context A. Testing occurred after Phase 3 in which the
Renewal After A and Renewal After B were tested with X in Context A and B, respectively. The
Extinction and No Extinction groups were tested with T in Context B.
The purpose of extinguishing X, or not, prior to conditioning T was necessary to test for
effects of previous extinction training on the importance of context on responding to a simple CS.
Based on the ATCP, this manipulation should have made responses to T more context specific as
a result of the added ambiguity to X. However, when tested in Context B there was no evidence
of such an effect, resulting in no significant difference in suppression rate for the No Extinction
and With Extinction groups. However, during conditioning of T, renewal with X was observed.
This suggests that, through the ATCP, the participants were attending to the contexts at the time
of encoding T. In addition, renewal with X was also seen during testing with T, providing more
support for the attending to contexts. Taken together, Nelson and Lamoureux’s behavioral model
5. was not able to recreate the effect of extinction of one cue (X) enhancing the contextual control of
an acquisitioned cue (T) seen in the predictive learning models of Rosas and Callejas-Aguilera,
despite the participants attending to the contexts during the test of T.
In their current study, Nelson & Lamoureux designed a 2x2 factorial design with the intent
of observing the effect of extinction and temporal variability in the context conditioning phase on
predictive behavioral responses. As seen in Table 1, during the acquisition phase all of the
participants received 8 presentations of the purple sensor in Context A followed by an enemy
attack. In the Extinction phase all participants were moved to Context B where half of the
participants received 8 presentations of the purple sensor followed by no enemy attack, while the
other half received no extinction. In this same context, all participants received 8 unsignaled enemy
attacks (USs). For half of the participants, the presentation of the enemy attacks occurs at fixed 20
second intervals, while the other half received variable intervals of these USs. The Probe Trial
consisted of 12 5-second bins in Context A (for all participants) to look for any evidence of timing
of the expected US that was trained during the phase of unsignaled USs.
Table 1. Experimental Design for CA2 (Context Attention 2)
Group Acquisition Extinction Context Conditioning Probe Trial
_______________________________________________________________
FI-Ext A: 8P+ B: 8P- B: 8 unsignaled USs (FI) A: 60sec
_______________________________________________________________
VI-Ext A: 8P+ B: 8P- B: 8 unsignaled USs (VI) A: 60sec
_______________________________________________________________
FI-No Ext A: 8P+ B: --- B: 8 unsignaled USs (FI) A: 60sec
_______________________________________________________________
VI-No Ext A: 8P+ B: ---- B: 8 unsignaled USs (VI) A: 60sec
_______________________________________________________________
Note. FI = Fixed Interval (20 second intervals). VI = Variable Interval. P = purple sensor.+ represents enemy attack,
- represents no enemy attack. A and B are different contexts represented by galaxies. Numbers in front of P
represent number of trials.
6. The expectation, based on the ATCP, is that participants in the extinction groups should
show faster conditioning to the context. This means that these participants will show less
suppression during the Probe Trial, because the unsignaled USs occurred in Context B, while the
Probe Trial occurs in Context A. Those who did not receive extinction will not be attending to
context, and therefore are expected to suppress their click rate in the Probe Trial. In terms of the
temporal factor, those in the fixed interval conditions should show behavior conducive to
predictive timing. Therefore, in the Probe Trial they should learn to expect the enemy attack every
20 seconds, and thus decrease their click rate only before each 20 second mark. Those who were
in the variable interval conditions will not come to expect the attacks with any sense of certainty,
and thus are likely to show more suppression. Thus, participants in the extinction condition with
fixed intervals should show the least suppression in the Probe Trial due to attention to the contexts
and temporal cues. By contrast, participants that did not receive extinction and received variable
intervals will not have learned the context or temporal cues necessary to predict when the enemy
will attack during the Probe Trial and should therefore show the most suppression.
7. Work Cited
Bouton, M. E. (2004). Context and behavioral processes in extinction. Learning & memory, 11(5),
485-494.
Bouton, M. E., Westbrook, R. F., Corcoran, K. A., & Maren, S. (2006). Contextual and temporal
modulation of extinction: behavioral and biological mechanisms. Biological psychiatry, 60(4),
352-360.
Nelson, J. B., & Lamoureux, J. A. (2015). Contextual control of conditioning is not affected by
extinction in a behavioral task with humans. Learning & behavior, 43(2), 163-178.
Rosas, J. M., & Callejas-Aguilera, J. E. (2006). Context switch effects on acquisition and
extinction in human predictive learning. Journal of Experimental Psychology: learning, Memory,
and cognition, 32(3), 461.