JOURNAL
OF VERBAL
LEARNING
The A-B,
RUDOLPH
AND
VERBAL
7, 602-607 (1968)
BEHAVIOR
B-C, A-C Mediation Following A-C
Paradigm: Learning1
Recall
of A-B
W. SCHULZ, JANET R. LISTON~, AND GEORGE E. WEAVERS University
of Iowa,
Iowa
City,
Iowa
52240
Three interpretations of the fact that paired-associate (PA) performance on A-C in the A-B, B-C, A-C paradigm is usually superior to A-C performance in the A-B, D-C, A-C paradigm w’ere evaluated. A total of 60 Ss, 20 per cent condition, learned successively three PA lists conforming respectively to an A-B, B-C, A-C (Cond. B-C), an A-B, B-C, E-F (Cond. E-F) and an A-B, D-C, A-C (Cond. D-C) paradigm. Recall of A-B following A-C learning was then measured in all conditions. The expected levels of recall according to the mediational interpretation were Cond. B-C > Cond. E-F > Cond. D-C. The data were in accord with this expectation. Alternative interpretations, interference (Mandler and Earhard, 1964) or direct-association (Schwenn and Underwood, 1965) were not supported by the data.
Empirically, it has now been demonstrated with paired-associate (PA) learning on numerous occasions that, given the proper conditions, performance on A-C in an A-B, B-C, A-C transfer paradigm will be reliably superior to performance on the same A-C list in an A-B, D-C, A-C paradigm. Theoretically, it has usually been inferred from this finding that the facilitation of A-C performance under the so-called “chaining conditions” resulted from S’s ability to use B to mediate the learning of A-C. Recently, however, this interpretation has been challenged by the proposal that this result is amenable to at least two alternative interpretations. Moreover, these alternative interpretations have the advantage of being more parsimonious than the mediational interpretations. One of these alternatives has been proposed by Mandler and Earhard (1964). They maintain that superior performance on A-C under chaining conditions may be due, not to mediation, but to a decrease in 1 This research was supported by Grant GB2835 from the National Science Foundation. 2The second author was an NSF Undergraduate Research Participant (Grant GE6342) while assisting with the conduct of this research. 3 This research was conducted during the tenure of the third author’s NSF Predoctoral Fellowship. He is now at Florida State University. 602
interference from A-B associations by virtue of the weakening of B-A associations, via unlearning, during the acquisition of B-C. A second alternative is the one suggested by Schwenn and Underwood (1965). Though the learning of B-C is again implicated as the mechanism, its role is assumed to be quite different from the one assigned it by Mandler and Earhard. That is, B, via the backward association B-A, is presumed to elicit A in the presence of C and as a result direct A-C associations can be formed during B-C performance which will transfer positively to A-C performance. In sum, there are, then, for the same empirical result three alternative interpretations, one involving mediation, one decreased interference and, a third, direct association. While there is evidence favoring the mediational over the interference interpretation (e.g., Goulet and Postman, 1966; Schulz, Weaver and Ginsberg, 1965), there is also evidence favoring the latter over the former interpretation (e.g., Carlson, 1966; Earhard and Earhard, 1966) which cannot be gainsaid. Furthermore, the third interpretation, the direct-association hypothesis, appears not to have been evaluated independently of the data which led to its postulation. The concern, therefore, of the present research was to
RECALL
OF MEDIATORS
provide additional empirical evidence relevant for the assessment of the relative merits of these several possible interpretations. According to a mediational interpretation, the performance of A-C actually involves the “chain” of associations A- > B- > C. If this is, indeed, the case, then it follows that the performance of A-C should strengthen or at least maintain the strength of the requisite A-B and B-C associations. Thus, if a fourth stage requiring recall of B given A is added to the conventional A-B, B-C, A-C and A-B, D-C, A-C paradigms, it would be anticipated that A-B recall would be superior in the former when compared with the latter paradigm. The interference interpretation leads to the expectation of, if anything, poorer recall of A-B under the A-B, B-C, A-C than under the A-B, D-C, A-C paradigms because A-B is “extinguished” partially via unlearning of B-A in Stage II (Mandler and Earhard, 1964) of the former but not the latter paradigm. A-B is “extinguished” directly in both paradigms in Stage III according to this interpretation (Barnes and Underwood, 1959). A similar result may be anticipated from the direct-association hypothesis, though it is less clear what is happening to A-B in Stage II. Nevertheless, there is presumably no way for A-B to be maintained or strengthened in Stage III and, therefore, the strength of the A-B association should deteriorate during Stage III performance. A third condition, A-B, B-C, E-F, was employed as a control involving Stage II but not Stage III extinction of A-B. In short if B is serving as a mediator in Stage III of the A-B, B-C, A-C paradigm, then the expected order of Stage 1V recall, from best to worst, is as follows: A-B, B-C, A-C >A-B, B-C, E-F> A-B, D-C, A-C. Hereafter, these three paradigms will be called Cond. B-C, Cond. E-F and Cond. D-C, respectively. METHOD Materials and Procedure. Since ployed here, except for fourth-stage
the methods emA-B recall, were
603
identical in all essential respects to those described by Schulz, Weaver and Ginsberg (1965), only a brief review of these methods will be given here. With the A-B, B-C, A-C paradigm of Cond. B-C as a reference, the A terms were low-m Taylor (1959) paralogs while low-M Archer (1960) CVC trigrams served as C terms. Common nouns (T-L frequencies of AA) served as B terms. In Cond. D-C, the D terms were nouns appropriately comparable to the nouns used as B terms. Similarly, the E and F terms of Cond. E-F were selected from the same pools of materials as the A and C terms. Moreover, a pilot study was conducted to demonstrate that the A-C and E-F lists were equal in difficulty. All lists consisted of 10 pairs. Acquisition in Stages I, II and III proceeded with the study-test method. In Stage I, Ss recalled the nouns on test trials. The test trials in Stages II and III were of the multiple-choice variety where each stimulus term was presented with four alternatives, the correct response and three other list-member response terms. The alternatives were numbered 14 and Ss always responded with the number of the alternative they had chosen. The study interval was 2 set in all stages. The test interval was 2 set in Stage I, and 3 set in Stages II and III. Acquisition wascarried toacriterion of two successivecorrect trials in Stages I and II. Stage III consisted to ten alternated study and test trials. The lists were presented on a Stowe Memory Drum, with the order of presentation on study and test trials as well as the positions of the response alternatives in Stages II and III suitably randomized. Stage IV, the critical stage for the present purposes, was the same for all conditions and consisted of a singlerecall trial at a 1.5set rate on which only A terms were presented with S carefully instructed to attempt to recall the appropriate B terms. A rapid rate was employed to insure less than perfect recall under all conditions to preclude the possibility that a “ceiling effect” would obscure the differences among conditions. The interval between Stages III and IV was about 2 min (the time required to give the instructions for recall). .Subj~ts. The Ss were assigned to conditions randomlywiththerestrictionthattheNthSnot beassigned to a given condition until N-l Ss had been assigned to the other conditions. Proportional numbers of male and female Ss were assigned to each condition. If S failed to attain the criterion performance level within 20 trials in Stage I, the S was dismissed and replaced. Since a common first-list was employed in all conditions, a bias due to differential loss of Ss is unlikely to have occurred. Observations were made on 20 Ss per condition. A total of 60 Ss were used. All Ss were undergraduates at the University of Iowa for whom participation in experiments was a course requirement. The Ss were not naive with respect
604
SCHULZ,
LISTON,
AND
WEAVER
to verbal-learning experiments, but none had served previously in a transfer experiment.
p < .Ol. Since the order of these means was specified by the hypothesis, one-tailed probabilities are appropriate for the pairwise RESULTS comparisons. The latter revealed Cond. B-C The results summarized in Table 1 for to differ significantly from both Cond. E-F, Stages I, II, and III are fully in accord with t(57) = 1.94, p < .05, and Cond. D-C, t(57) = those observed previously under Conds. B-C 3.55, p < .Ol, while the difference between and D-C by Schulz, Weaver, and Ginsberg conditions E-F and D-C approached signifi(1965). In Stage I, performance did not differ cance, t(57) = 1.61, p < .10. The hypothesis reliably as a function of conditions (F< 1). that the facilitation of Stage III performance in The learning of B-C in Stage II required sig- Cond. B-C is attributable to the use of B as a nificantly, ~(58) = 4.66, p < .Ol, more trials mediator in the learning of A-C is clearly than the learning of D-C. Performance in favored. Cond. B-C was reliably superior in Stage III Since a contingency between performance in to that under Conds. E-F and D-C, t(58) = Stage III and Stage IV is central to the TABLE PERFORMANCE
IN STAGES I, II,
1 III,
AND IV OF EXPERIMENT
Trials to criterion Stage
I
1
Number correct II
III
IV
Cond.
Mean
SD
Mean
SD
Mean
SD
Mean
B-C
10.40 9.90 8.90
4.49 4.39 3.22
9.00 9.45 4.60
4.96 4.92 4.88
74.10” 61.75 57.85
10.82 15.59 13.34
7.70 6.50 5.50
E-F D-C
SD
1.95 1.67 2.19
LITotal number of correct responses over ten trials.
4.43, p < .Ol ; but, did not differ for Conds. E-F and D-C, t < 1. The latter result is a somewhat unexpected one, inasmuch as Cond. D-C involves an A-B, A-C relationship between Stages I and III while Cond. E-F does not. Critical for the present purpose is recall performance in Stage IV. As may be seen in Table 1, recall was best in Cond. B-C, poorest in Cond. D-C and intermediate in Cond. E-F. This is the order predicted by the mediation interpretation and contrary to expectations based on either the interference or directassociation interpretations. Statistically, the differences, overall, among these means may be regarded as significant, F(2, 59) = 6;40,
mediational interpretation, it will be productive to examine the present data further with the demonstration of such a contingency in Cond. B-C as the main objective. An analysis was made, therefore, of A-B recall as a joint function of A-B and A-C strength, as inferred from differences in degree of learning of individual items in Stages I and III. For each S in Conds. B-C and D-C, the ten pairs constituting the A-B list were ranked according to the number of correct responses given by S during the course of Stage I acquisition. A Rank of 5 was assigned the two pairs having been given correctly the greatest number of times, a Rank of 4 to the two pairs with the next highest number correct, etc. Ties were
RECALL
605
OF MEDIATORS
resolved randomly. Thus, five levels of A-B strength were defined. The same ranking procedure was applied to the A-C items of Stage III. We then posed the following question: What happens to A-B recall as a function of A-B strength when A-C is strong (Ranks 4 and 5) vs. the case in which A-C strength is weak (Ranks 1 and 2)? The answer to this question is displayed in Fig. 1. Though
research which is addressed specifically to this issue. DISCUSSION
Three interpretations of the facilitated performance commonly observed in the test stage of the A-B, B-C, A-C chaining paradigm have been given further empirical scrutiny. The results for Stage IV A-B recall strongly supported a mediational interpretation and failed to support the interference or directassociation hypotheses. Moreover, these results confirm directly those reported very recently by Horton and Wiley (1967a; 1967b) for Stage IV A-B recall under conditions A-C STRONG similar to the present ones. In addition to providing independent confirmation of Horton and Wiley’s observations, the results obtained here shed further light on an interpretative issue involved in both their results and ours. As in their studies, I I I I 1 I Stages II, III, and IV of this study constituted I 2 3 4 5 WEAK STROMi a B-C, A-C, A-B stimulus equivalence paraSTRENGTH OF A-S digm in Cond. B-C. Hence, there is the possiFIG. 1. StageIV A-B recall asa function of A-B and bility that mediation at the time of recall A-C strength in Conds. B-C and D-C. accounts for the superior Stage IV performthese data are not amenable to statistical ance under this condition. Though Horton analysis, the major trends apparent in Fig. 1 and Wiley were able to argue, based on other seem reasonably compelling and supportive evidence they obtained, that it was unlikely of a mediational interpretation. First, A-B that mediation at the time of recall accounted recall was a direct function of both A-B and for their results, their conditions (e.g., A-C strength. Second, the difference between involving learning by the anticipation method Cond. B-C and D-C is greatest when A-C was throughout and learning to a criterion of one strong. Third, there is a trend, albeit a slight errorless trial in Stage III) were considerably one, for A-C strength to “benefit” weak A-B more favorable than the present ones with items more than strong ones. respect to such an eventuality. Here, S’s The trends in the curves for Cond. D-C are Stage III learning was limited to ten trials also interesting in that they suggest that A-B (only 11 of the 20 Ss attained a criterion of one and A-C strength interacted in determining perfect recitation) and the C terms were merely A-B recall. Thus, interference appears to have recognized, never pronounced overtly; hence, been at a maximum when A-B and A-C were of A-C associations were not learned to as high a equal strength, both weak or both strong. degree, and C terms were undoubtedly less Though it is tempting to speculate further as to available, thereby reducing the likelihood of the meaning of these trends with respect to the mediation in Stage IV via the association of nature of the unlearning process which might A with C and C with B. Furthermore, if B-C have operated under nonmediated conditions, strength is used to index the strength of the such speculation had better await future C-B association, B-C strength and A-B recall
606
SCHULZ,
LISTON,
were unrelated in Cond. B-C, the recall percentages being 65.0 and 66.2 for high and low B-C strength, respectively. Insofar as B-C strength is a valid index of C-B strength, this result is at odds with an interpretation positing mediation in Stage IV. Lastly, the rapid presentation rate (1.5 sec.) used in Stage IV should have served to reduce still further the possibility of mediating during A-B recall (e.g., Schulz and Lovelace, 1964; Schulz and Weaver, 1968). Another aspect of the present results which deserves comment is the finding that A-B recall in Cond. E-F was reliably inferior to that in Cond. B-C. This result would seem to permit the inference that the strength of A-B associations was not merely maintained during A-C acquisition in Cond. B-C, but that these associations were strengthened additionally. The validity of this inference is contingent, of course, on the assumption that Cond. E-F represented an appropriate control, both for the effects of “unlearning” of A-B in Stage II and for “normal” forgetting while learning an unrelated list in Stage III. It seems, at least to us, that Cond. E-F met these control criteria. The demonstration of an increase in A-B strength during Stage III in Cond. B-C seems particularly crucial to the refutation of the direct-association hypothesis because a proponent of the hypothesis could maintain that A-B associations are not elicited in Cond. B-C during Stage III since direct A-C associations are already available to S. As a result, A-B associations might not be extinguished and the superiority of Cond. B-C over Cond. D-C in A-B recall would be predictable. Unfortunately, Horton and Wiley neglected to include an E-F control in their studies; hence, the extent to which the evidence regarding A-B recall in Conds. B-C and E-F is specific to the present conditions cannot be assessed. A firm conclusion on this issue must, therefore, await independent additional evidence such as a systematic increase in A-B recall under Cond. B-C as a function of
AND
WEAVER
increasing numbers of Stage III acquisition trials. It was noted in connection with the results for Stage III acquisition that Cond. D-C unexpectedly failed to be inferior to Cond. E-F even though Stages I and III of the former condition represent an A-B, A-C transfer paradigm while those of the latter do not (cf., Table 1). A similar result has been observed in an unpublished study involving paradigms of the form X-B, B-C, A-C and A-B, D-C, A-C. On the face of it, these results would seem to support Earhard and Mandler (1965) and, more recently, Earhard and Earhard’s (1967) contention that the procedures such as those employed here reduce task complexity and memory load, thereby presumably mitigating the influence of interference and forgetting factors. There are, however, certain other facts which are difficult to reconcile with this interpretation. First, if the use of multiple-choice test trials and “long,” 3 set, test intervals is to account for the lack of interference in Stage III, to what is the decrement in Stage II B-C performance to be attributed since these procedures were also used there ? Second, once A-B associations are “unlearned” in Stage II, as the decrement in B-C learning ostensibly indicates, how can S’s utilize extinguished associations to mediate A-C learning? Or could it be that the B-C decrement is entirely epiphenomenal with respect to the interpretation of Stage III performance inasmuch as it obtains both when pseudomediation effects are present and when they are absent (also cf., Goulet and Postman, 1966; Schulz et al., 1965)? Third, how is the congruence between Horton and Wiley’s results and the present ones for A-B recall to be accommodated ? These investigators did obtain a decrement in Stage III performance for Cond. D-C and used conventional procedures which, with the possible exception of the meaningfulness of their B terms, were highly similar to those proposed as requisite for the demonstration of pseudomediation. In sum, it seems increasingly evident that the
RECALL
applicability of the interference hypothesis is restricted to a very narrowly circumscribed set of conditions.
REFERENCES
re-evaluation of the meaningfulness of all possible CVC trigrams. Psychol. Monogr., 1960,74, No. 10 (Whole No. 497). BARNES, J. M., AND UNDERWOOD, B. J. “Fate” of first-list associations in transfer theory. J. exp.
ARCHER,
E. J. A
Psycho/., CARLSON,
1959,58,97-105.
J. G. Pseudomediation effects with no intertrial intervals. Psychon. Sci., 1966,5,77-78. EARHARD, B., AND EARHARD, M. Role of interference factors in three-stage mediation paradigms. J. exp. PsychoI., 1967,73,526-531. EARHARD, B., AND MANDLER, G. Pseudomediation: A reply and more data. Psychon. SC?., 1965,3,137138. GOULET,
607
OF MEDIATORS
L. R., AND POSTMAN, L. An experimental evaluation of the pseudomediation hypothesis. Psychon. Sci., 1966,4,163-164. HORTON, D. L., AND WILEY, R. E. Mediateassociation: Facilitation and interference. J. exp. Psycho/., 1967,73,636-638. (a).
D. L., AND WILEY, R. E. The effect of mediation on the retention and strength of previously formed associations. J. verb. Learn. oerb. Behav.,
HORTON,
1967,6,3&41. MANDLER, G.,
(b). EARHARD,
B. Pseudomediation: Is chaining an artifact? Psychon. Sci., 1964, 1 247-248. SCHULZ, R. W., AND LOVELACE, E. A. Mediation in verbal paired-associate learning: The role of temporal factors. Psychon. Sci., 1964,1,95-96. SCHULZ, R. W., AND WEAVER, G. E. The A-B, B-C, A-C mediation paradigm: The effects of variation in A-C study- and test-interval lengths and strength of A-B or B-C. J. exp. Psychol., 1968, 76,303-311. SCHULZ, R. W., WEAVER, G. E., and GINSBERG, S. Mediation with pseudomediation controlled : Chaining is not an artifact! Psychon. Sci., 1965, 2,169-l 70. SCHWENN, ELIZABETH, AND UNDERWOOD, B. J. Simulated similarity and mediation time in transfer. J. verb. Learn. verb. Behao., 1965, 4, 476483. TAYLOR, J. D. The meaningfulness of 320 words and paralogs. Unpublished doctoral dissertation, Duke Univer.. 1959. AND
(Received May 12, 1967)