Karen Stark

Temporal changes in interference

A series of five experiments investigating temporal changes in retroactive inhibition (RI) and proactive inhibition (PI) is reported. In the first four experiments changes in retention were measured by unpaced tests over intervals of the order of 20 min. Conditions of acquisition and methods of testing were varied with a view to identifying the factors determining absolute rises in first-list recall (“spontaneous recovery”) and decrements in the recall of interpolated lists (PI). Both of these phenomena were observed and were attributed to the dissipation of a set, established through interpolated learning, to give the most recently learned responses in recall. In the fifth experiment RI and PI were measured on paced tests after intervals of 20 min and 48 hrs. The amounts of RI and PI did not differ significantly after 20 min, and showed no further reliable changes over the 48-hr interval. The results indicate that the level of interference on paced tests is determined largely by the degree of reduction in response availability.

Studies of learning to learn: IV. Transfer from serial to paired-associate learning1

This experiment investigates the conditions determining transfer from serial to paired-associate learning. There were three conditions of transfer defined by the relationship between the successive lists. In Cond. E the paired-associate list consisted of all the pairs of adjacent items from the serial list; in Cond. C different items were used in the two lists, and in Cond. ER the paired associates were composed of nonadjacent members of the serial list. Under each condition of transfer there were an instructed group and an uninstructed group. After the end of serial learning the instructed group was informed about the relationship between the successive lists; the uninstructed group received no such information. All Ss served in three experimental cycles in each of which transfer from serial to paired-associated learning was measured. Significant amounts of positive and of negative transfer were obtained in Cond. E and Cond. ER, respectively. Instructions substantially enhanced positive transfer but failed to influence negative transfer. Positive transfer increased and negative transfer decreased as a function of practice. The results are consistent with the assumption that serial learning involves the development of sequential associations.

The role of response set in tests of unlearning

Evidence for the unlearning of first-list associations (A-B) during acquisition of a second list (A-C) has been obtained in unpaced tests of modified free recall (MMFR) in which S is required to reproduce both responses to each of the stimuli. The present experiment investigates the extent to which the loss of first-list associations in MMFR is attributable to a set established during IL and carried over into MMFR, viz., to restrict overt responses to those correct in the second list. In order to maximize the set to give first-list responses, a non-learning task was introduced between the end of IL and MMFR which required active recitation of all first-list associations. Comparison with appropriate control groups shows that (a) the set-inducing procedure increased the dominance of available first-list responses as indexed by the order of recall, but (b) had no effect on the number of such responses recalled, i.e., left the amount of RI unchanged. It appears unlikely, therefore, that a loss of set is responsible for the reduced availability of first-list associations in MMFR.

Proactive inhibition as a function of the conditions of transfer

Summary This study investigated long-term proactive inhibition as a function of the conditions of transfer during acquisition. The stimuli were CVC trigrams and the responses were English words. For Group A-B, A-C the successive responses to each stimulus were unrelated, whereas for Group A-B, A-B′ the two responses had strong preexperimental associations. These two conditions represent paradigms of negative and positive transfer, respectively. Both lists were learned to a criterion of one perfect trial and recall of the second list was tested after 48 hours. A Control Group learned and recalled a single list. Speed of List-2 learning was fastest for Group A-B′ and slowest for the Control Group, with Group A-C occupying an intermediate position. The divergence in the speed of acquisition was sufficiently great to produce substantial differences in the amount of overlearning of List-2 associations. Retention of List 2 was lower under both experimental conditions than under the control condition; but the amount of proactive inhibition based on the raw recall scores was significant only for Group A-B, A-C. It wa shown, however, that the differences in degree of List-2 learning resulted in a strong bias favoring superior recall by Group A-B′. When this bias is taken into account, there is evidence for proactive inhibition under conditions of maximal positive transfer as well of negative transfer.

Retroactive inhibition as a function of set during the interpolated task

Summary This experiment investigates retroactive inhibition in paired-associate learning as a function of Ss set during the interpolated task. The A-B, A-C paradigm was used, with two-syllable nouns as stimuli and responses in List 1 (A-B). There were two Work Groups and two Rest Groups in Part I of the experiment. The Work Groups learned a second list of paired associates (A-C), with each S generating his own set of interpolated responses. The Rest Groups were engaged in a series of mathematical problems. For purposes of manipulating set during the interpolated activity, one Rest Group and one Work Group were given explicit instructions about the later test of recall on List 1. The Instructed Work Group was also warned about the interfering effects of interpolated learning on recall. In Part II of the experiment two additional Work Groups, one Instructed and one Uninstructed, were required to learn the List 2 associations generated by the corresponding Ss in Part I. There were some systematic differences between the distributions of associations generated by the two Work Groups in Part I, with the Instructed Group favoring stimulus associates of a lower rank than the Uninstructed Group. Degree of List 2 learning was, however, nearly perfect under all conditions of interpolation. The amount of RI was significantly lower for the Instructed than the Uninstructed Work Groups, and this difference was greater when List 2 responses were prescribed than when they were generated by S. It is suggested that instructions lead to increased rehearsal which is more beneficial under conditions of work than of rest. Comparison of Parts I and II indicates that the effectiveness of instructions in reducing RI increases with the contextual similarity between successive tasks.

Transfer from serial to paired-associate learning: A reappraisal

Tests of the hypothesis of serial chaining have evaluated transfer from a serial list to the double-function paired-associate list derived from it. The limited transfer effects so far reported may have resulted from the inherent difficulty of the transfer task under conventional methods of testing. In the present experiments two conditions were employed to reduce the difficulty of the test list, and in particular interference from backward associations during paired-associate learning: (a) Ss were fully informed about the structure of the transfer list; (b) a 3-sec rather than 2-sec anticipation interval was used. Experiment I also evaluates the effect of word frequency upon transfer; Experiment II evaluates the effect of stimulus pronunciation during paired-associate learning. In both experiments there was highly significant transfer from serial to paired-associate learning, both on the first trial and as measured by trials to criterion. Neither word frequency nor stimulus pronunciation had significant effects. Most serial positions contributed to the transfer effects. The magnitude of the present transfer effects is interpreted as strong support for the hypothesis.

The role of associative mediation in retroactive inhibition and facilitation

The role of implicit mediating responses in producing retroactive inhibition and facilitation (RI and RF) was investigated. The degree of associative connection between the stimulus terms in successive paired-associate lists was varied. In Experiment I the response terms in the first and the second list were different and unrelated. There was no evidence of mediated interference in either the transfer or the recall stage. Negative transfer and RI were substantial when the stimuli were identical. In Experiment II the responses in the two lists remained the same. Both mediated facilitation and RF were found. Potential mediators appear to be activated only when they facilitate performance.

On the measurement of retroactive inhibition in the A-B, A-D paradigm by the multiple-choice method: Reply to merryman

Postman and Stark (Journal of Experimental Psychology, 1969 , 79 , 168–177) found no reliable evidence for associative unlearning under the A-B, A-D paradigm when retention was measured by the multiple-choice method. Merryman (Journal of Experimental Psychology, 1971 , 91 , 212–214) has advanced the hypothesis that the occurrence of associative unlearning was masked in the previous study because correct responses were mediated by backward associations which remained intact after interpolated learning. He presented experimental evidence purporting to demonstrate significant associative loss after bidirectional unlearning. The theoretical rationale and experimental method of Merrymans study are examined critically. A further investigation of the effects of bidirectional unlearning under the A-B, A-D paradigm is reported which failed to substantiate Merrymans hypothesis.