Bruce R. Ekstrand

An analysis of intralist similarity in verbal learning with experiments on conceptual similarity

An analysis was made of four subprocesses of verbal learning which may be expected to vary as intralist similarity (meaningful, formal, or conceptual) is manipulated in a PA list. Expectations were stated as to the role of each subprocess in over-all PA learning when similarity is manipulated among stimulus terms and among response terms independently. In the initial experiments, lists were constructed to eliminate two of the four subprocesses, leaving response learning and associative interference as the primary factors varying with conceptual similarity. The experiments showed that similarity among all stimulus terms was deleterious to learning while similarity among response terms had no effect on learning. These effects were independent of word frequency. Further experiments, devised to study response learning and associative learning independently, gave plausibility to the notion that with similarity obtaining among response terms, the positive effects of response learning balance the negative effects of associative interference for these materials. A final experiment varied the number of concepts within a list to examine the influence of a further hypothesized factor, S-R limitation. The overt-error data indicated the presence of this subprocess varying in extent as the number of concepts varied. However, the manner in which it contributed to over-all PA learning could not be ascertained.

Knowledge of rights and wrongs in verbal-discrimination learning.

Summary The purpose of this experiment was to determine what information S acquires about right and wrong items in verbal-discrimination learning. Following learning of one list, two groups of Ss were given a new list in which the right items from the first list were retained with new wrong items inserted (R. Cond.). For two other groups new right items were inserted with the wrong items remaining the same as in the first list (Cond. W). Two other groups served as controls, having completely new items on the second list. For each condition, two anticipation intervals were used on the second list (1.5 and 3 sec.). The results showed that under the R Cond. performance was essentially perfect on the second list; even with the short anticipation interval Ss averaged only slightly over one error across 10 trials. Under the W Cond. performance on the second list was initially very high but improvement beyond this point was very slow. With the short anticipation interval under the W. Cond., performance was inferior to the control on later trials although it was initially much superior. The results were interpreted by a frequency-discrimination hypothesis.

A note on measuring response learning during paired-associate learning

An analysis of a technique commonly used to measure the relative lengths of the two stages of paired-associate learning (response learning and associative learning) was presented. This technique involves using the “first given” (FG) as a measure of the duration of the response-learning stage. FG is the number of trials before S gives a response anywhere in the list, disregarding the appropriateness of the stimulus to which it is given. On the basis of studies existing in the verbal-learning literature, it was shown that FG overestimates the time required for response learning, probably because S s wait to give a response until a certain level of confidence about the correctness of the response is reached. This level was termed the confidence threshold. The amount of interference present in the learning situation is likely to be one of the determinants of the time required to reach the threshold. In order to measure accurately the response-learning stage, a procedure similar to that of Underwood et al. (1959) may be followed: PA learning is stopped at various points and response learning is measured by free recall with a “paced” procedure. A second alternative is the use of free learning to study the effects of variables on the response-learning stage of PA learning.

Solving words as anagrams

Two experiments investigated the effect of spacing on the solution of word and nonword anagrams. In word problems the five letters already constitute a word and the solution is another unrelated word. The results indicate that word problems are more difficult than nonword problems and that spacing probably does not affect anagram solution.

Paced versus unpaced recall in free learning

Summary This study examined free learning with paced and unpaced recall. In paced recall, S “anticipated” items at a 2-sec. rate; in unpaced recall 60 sec. were allowed to write down the responses. High- and low-similarity lists were used with three study and three recall trials given. The results show that the high-similarity list was better recalled than the low-similarity list and higher recall occurred with unpaced than with paced recall. However, similarity did not interact with method of recall. Although unpaced recall produced better performance than paced recall statistically speaking, in an absolute sense the differences were not great. The number of correct responses given with paced recall was 89%, 90%, and 95% of the number given with unpaced recall for three different lists. An analysis of recall as a function of word frequency gave suggestive, but not definitive evidence that the influence of frequency is greater under unpaced than under paced recall.

Free learning and recall as a function of unit-sequence and letter-sequence interference.

Free learning was used in two experiments to test the Underwood and Postman (1960) theory of long-term retention. Experiment I compared the forgetting of high- and low-meaningful word pairs and high- and low-meaningful letter pairs. The second experiment repeated the comparison of the word pairs with the addition of a free association test just prior to 24-hour recall. This test was designed to “arouse” all of Ss associative systems directly in the experimental context. The results of both experiments showed no differences in retention between the high- and low-meaningful materials. This finding does not agree with the Underwood and Postman theory. Further analyses revealed that free learning is not forgotten appreciably faster than paired-associate or serial learning. Although the theory has had remarkable success in retroactive and proactive situations, it is inadequate in its present form for predicting retention of a single list learned by naive Ss.

Feature frequency and the acquisition of natural concepts.

Subjects were presented with 100 schematic faces that conformed to a particular frequency distribution of features and then were asked to make typicality judgments. Method of acquisition varied across the six conditions tested. Both absolute ratings and paired comparisons of typicality revealed a linear relationship between summed feature frequency and degree of category membership. The relationship was invariant across sequential and paired presentation of training faces as well as instructions to organize the concept. Subjects required to make typicality judgments during acquisition displayed a weak relationship between frequency and category organization unless they were given feedback after each judgment. The results support a feature-frequency interpretation of natural concept learning.

Differentiation among stimuli as a factor in transfer performance

It was assumed that in learning a paired-associate list, S must establish differentiation among stimuli. If this differentiation transfers to a second list, a positive factor should be present in learning the second list of the A-B, A-C paradigm which is not present in A-B, C-D. The magnitude of the positive factor should increase directly with stimulus similarity, hence an interaction between paradigm and similarity in transfer performance was expected. Lists of high and low formal stimulus similarity were used. Although stimulus similarity had an enormous effect on first-list learning, the evidence for a transfer of stimulus differentiation in three experiments was slight.

Linguistic associations and retention

Four experiments were concerned with the role of linguistic associations in producing interference in recall. In the first, strong associates were crossed (inappropriately paired) in a paired-associate list, and recall after 24 hr was compared with that for control lists. Although the crossed-associate list was more difficult to learn than the control lists, recall was superior. Evidence indicated the superiority was due to higher response availability. In Exp. II, connectives were used as stimuli in the critical list. Learning was slower than for the control lists but retention was not depressed. In Exp. III three interfering lists were used to study PI in the 24-hr recall of a fourth list. All lists had the same stimuli. For the E Group the pairs in the first three lists were associates; this was not true for the C Group. To equalize performance for the two groups on the first three lists the E Group was given one anticipation trial on each, the C Group five trials. Both groups had six trials on the fourth. Recall for the E Group was twice as high as that for the C Group and the magnitude of the recall (68%) indicated little PI. In Exp. IV the same lists were used but the E Group was given five trials on each of the first three lists. Recall for the E Group was 22%, and did not differ from the C Group. It was suggested that differentiation based on differences in trial frequency between the list to be recalled and the interfering lists was in part responsible for the high recall of the E Group in Exp. III.

Response-term integration

Lists of 9 trigrams, varying from common words to consonant syllables, were presented for free learning. In build-up procedures, some groups first learned single letters of the trigrams, then bigrams, and finally trigrams in which each bigram appeared. Other groups learned only bigrams before these became a part of the trigrams, and still other groups learned the trigrams only. The build-up procedures produced a small amount of facilitation in learning the difficult trigrams only, and this did not differ for groups starting with single letters and for those starting with bigrams. The facilitation in trigram learning was accompanied by a decrease in integrative errors.