as R1 or R2. R1 was correct only following S1 and R2 only following S2. Incorrect choices produced a 10-sec blackout followed by the usual 5- sec ITI and a repeat of the trial. This correction procedure continued until the trial was correctly completed.
The experimental design involved four groups of four pigeons each. For Group DD (Differential-Differential), the color-to-line problem was programmed according to a Differential Outcomes procedure in which O1 was 3 sec access to grain while O2 was a 0.75 sec, moderately loud 1- kHz tone. On the geometric form trials for this group, the circle was always followed by the food whereas the triangle was always followed by the tone. For Group NN (Nondifferential-Nondifferential), the color-to-line problem was programmed according to the Nondifferential Outcomes procedure so that each kind of trial was consequated equally often by both kinds of outcome. Similarly, for this group the circle and triangle were each followed by food on a random half of the trials and by tone on the other half. The procedures for the remaining two groups, Group DN and Group ND, involved combinations of these procedures such that the first letter in the group designation identifies the procedure on the color-to-line problem and the second letter the procedure on the geometric form trials.
For the transfer test, S3 and S4 replaced S1 and S2, and all subjects were trained on the Differential Outcomes procedure where O1 occurred on S3 trials and O2 on S4 trials. R1 was correct only following S3 and R2 only following S4.
The results of the experiment are presented in Figure 8.2. Groups DD and DN acquired the original conditional discrimination significantly faster than did Groups ND and NN. Moreover, the insertion of a 1-sec delay between offset of the conditional cues and onset of the choice stimuli impeded performance of the latter groups more than the former. When the delay was removed, Groups DD and DN exhibited essentially errorless levels of performance while Groups ND and NN performed at around the 85-90% correct level. All of these results correspond to previous observations from our laboratory (cf. Petersonet al., 1980).
The results of most interest to the present discussion are those of the transfer test. As expected from the theory, significantly more accurate choosing of the correct alternatives obtained for Group DD than for the other groups. This was due, presumably, to S3 and S4 evoking E1 and E2 (as had S1 and S2 in original training), and, in turn, E1 and E2 cuing R1 and R2, respectively. It should be noted that there was some decrement in Group DD's performance at the outset of the transfer test. This suggests that either the expectancies did not have exclusive control' of R1 and R2 in original training (i.e., S1 and S2 exerted at least some direct stimulus control), or, alternatively, that if the expectancies did have exclusive control of the choice responses, then S3 and S4 did not invoke E1 and E2 as powerfully as S1 and S2 had. Further research is needed to resolve this issue. It is clear, however, that the superior transfer exhibited by Group DD relative to Group DN cannot be attributed simply to the former having learned to discriminate S3 and S4 during original training; since Group ND had the same S3-O1 and S4-O2 pairings during original training as Group DD, Group ND presumably would have learned to discriminate S3 and S4 too, and yet they did not show any particular benefit from this pre-