During the first five training days where the patient had identity matching sessions using the same stimuli as in the conditional relation procedures, the patient maintained a high level of accuracy ranging from 88.6% to 100%, with an average of 96.3% (N = 19 sessions); six sessions were at 100%.
The percent correct for each of the 11 sessions of teaching the A → B relations is presented in Fig. . Data are shown for each of the three relations (A1 → B1, A2 → B2, and A3 → B3) and for all relations together. During the first five sessions with all relations presented in mixed order within sessions, the patient did not learn the relations and even was at 0% correct for the A2 → B2 trials in sessions 3 and 5. Therefore, the procedure was simplified to present only two of the relations within each session until the patient learned the relations. The percent correct immediately increased. When all three relations were again presented in mixed order in session 11, the patient scored at or above 90% correct on all three relations and had therefore learned the arbitrary conditional association between the type A stimuli (signs) and the type B stimuli (colored disks).
Figure 4 Percent correct for each of the A → B relations (1 through 3) and all three relations together. Numbers above the graph indicate which relations were presented in mixed order within a session. Along the X-axis, 0's indicate a score of zero, and (more ...)
Fig. shows percent correct for baseline and test trials for the symmetry, transitivity, and equivalence tests in steps 2 through 8. On the B → A symmetry tests, patient ER maintained a high accuracy on baseline trials and was 90% and 100% correct on probe trials showing clear evidence that the learned A → B relations were symmetrical without explicit training.
Figure 5 Percent correct for all trained and tested relations through steps 2 to 8. Relations above the graph indicate baseline relations and probe test relations. All indications are from Sample → Comparisons. Data are shown for both baseline trials and (more ...)
After the B → A symmetry test, the patient was taught the B → C relations (all three at the same time) in sessions 14 and 15 and learned all three in those two sessions in contrast to the slow learning of the A → B relations in sessions 1 though 11. On the C → B symmetry probe trials, patient ER scored 92% and 98% correct indicating that the taught B → C relations were symmetric. The A → C test for transitivity was given prematurely on session 18 by mistake (see procedure), and the patient scored at chance level on probe trials while maintaining the baseline at close to 100% correct. Because the patient had not been exposed to the A → B relations for some sessions, this relation was reintroduced in session 19. A new A → C transitivity test in the following session 20 showed nearly 100% correct on both baseline and probe trials indicating that the taught A → B and B → C relations were transitive.
To prepare the patient for the C → A equivalence test, two sessions presented all the necessary trial types (see Table ) in mixed order for sessions 25 and 26, and the overall accuracy was above 90%. On the two C → A equivalence tests, the patient scored 90% and 100% correct, respectively, indicating that the taught A → B and B → C relations were equivalent; the baseline accuracy was close to 95% in both sessions.
After the equivalence test, steps 2, 3, 4, 5 and 8 were repeated as step 9 over 11 sessions (sessions 29 – 39) to maintain the overall performance. Accuracy on baseline and probe trials was above 90% on each of these sessions, and data are not shown to save space.
The last test involved a choice between groups of stimuli; one group showed three equivalence class-related stimuli together while the other group showed three stimuli that did not constitute an equivalence class. All trials in these test sessions were of this type (two test types were presented, see procedure). Each test was presented twice. For the first two sessions, the accuracy was 94.4% and 98.3% for Test 1 and Test 2, respectively. For the next two sessions, the percent correct was 98.3 for both Test 1 and Test 2. That the overall correct selection was very close to 100% on all four sessions indicates that the patient easily identified and selected the group of stimuli that formed an equivalence class.
Because the patient received a smiley as positive feedback on all correct trials through the experiment, a natural question is whether the overall correct performance on probe test trials reflected learning during prior training sessions or reflected quick learning during the probe trials. To answer this question, the data were analyzed at the level of individual probe trials for the first session of each test type. Fig. shows correct or incorrect selection on each trial of probe testing (12 trials each test session) and in comparison also for the first 24 trials of the first session of A → B training and the first session of B → C training, where the stimulus relations were new to the patient. If the probed relations had to be learned during testing, one would expect probe trial performance to resemble performance during sessions with learning of new relations. The A → B and B → C relations were learned slowly over several sessions (11 sessions for the A → B relation and 2 sessions for the B → C relations, Figs. and ), and correct selections in the beginning of the first sessions of these relations did not lead to quick learning (Fig. ). In contrast, during probe trials in test sessions, the patient made very few if any mistakes from the beginning of probe trials, indicating that the relations the probes trial tested for were not learned during testing. Only during the first A → C probe test did the patient perform poorly, as already described above, and there was no indication of quick learning after one or two correct trials (Fig. ). Thus, the overall data suggest that the patient did not learn the test relations from feedback during the first few probe trials at the beginning of a test session. Instead, the performance on probe test trials indicated that the patient selected the correct stimulus during probe testing based on learning during prior training sessions.
Figure 6 To assess possible learning within sessions, performance on individual trials (correct or incorrect) is shown for selected sessions. Session number is indicated in parenthesis for each data set. For the first teaching sessions with A → B and B (more ...)