Using the learning game, one subject (a prominent member of the Learning field) was trained on an A:R+/B:R- discrimination. In context A: the Red sensor indicated an attack (+), but in context B: the stimulus was presented without outcome (-). Finally, the subject was tested in context C:, a familiar context where the Red sensor had never before been encountered.
Points show responses per second to the sensor in the absence of the attack. Thus, on + trials they show the first 5-s before the attack occurs. On – trials they show all 20-s of responding to the stimulus as no attack occurs. Brown circles indicate responses per second during the first 5-s of the Red sensor, prior to the attack by the spaceship in context A: on each trial. White circles indicate responding to the Red sensor during all 20-s of each trial of its presentation in context B. Brown triangles at the far right indicate responding to the Red sensor during all 20-s of its presentation on each test trial in context C:.
The subject began to respond in anticipation of the attack on trials AR+3 and AR+4. On the next trial (BR-1), where the sensor was first encountered in context B, the subject responded considerably across the duration of the stimulus. The effect of the absence of the outcome was evident on the next trial (AR+ 5) in context A: where the subject responded minimally to the sensor. The effect of the attack on that trial was evident on the following trials (BR-2, BR-3) in context B. Here the subject responded to the sensor and showed some evidence that the attack was expected after five seconds. Responding peaked near the 5-s mark, and then began to decline.
On return to context A: the participant again anticipated the attack on the next three trials (AR+6 thru AR+8) as well as when the sensor was next encountered in context B (BR-4). The absence of the attack on that trial once more reduced performance on the following trial (AR+9). However, at that point the contexts appeared to acquire control. The sensor failed to elicit responding in context B on trials BR-5 thru BR-8 and elicited strong responding in context A (AR+10).
On test, the sensor elicited robust responding in context C, although it had never been presented there before. Of the two meanings (attack, no attack) that had appeared to come under the control of contexts A and B, only the latter (no attack) appeared to be under contextual control. Anticipation of attack in the presence of the sensor generalized to context C.
The effect is similar to ABC renewal where conditioning occurs in context A, extinction in context B, and testing in context C. The difference here is that rather than a phase of conditioning and a phase of extinction the two types of trials were intermixed in an explicit discrimination. Whether or not this training affects the extent to which the two associations conditioned to the sensor differentially transfer to context C, relative to a standard conditioning and extinction treatment, is presently under investigation.