Extraexperimental transfer in serial recall

Extraexperimental transfer in serial recall

JOURNAL 494-497 (1963) O F VERBAL L E A R N I N G AND VERBAL B E H A V I O R 9. 7 Extraexperimental Transfer in Serial Recall 1 GERALD R. M I L L E...

310KB Sizes 2 Downloads 101 Views

JOURNAL

494-497 (1963)

O F VERBAL L E A R N I N G AND VERBAL B E H A V I O R 9. 7

Extraexperimental Transfer in Serial Recall 1 GERALD R. M I L L E R 2

The Johns Hopkins University, Baltimore, Maryland To recall in order a list of words which exceeds his immediate memory span, S must have available associations between the successive elements which make up the list. For example, he must be able to recall groups of associated letters as a word. In addition, he must form associations between adjacent words so that the recall of one word will suggest the next word in the list. The ease with Which S can learn to recall such a list must depend on the extent to which the associations that S brings to the experiment correspond to the associations necessary to recall the list. Ease of learning is a function of the transfer from extraexperimental associations which have been built up by S's verbal behavior. Thus the old and widely accepted generalization that meaningful material is easier to learn than nonmeaningfnl material can be interpreted to mean that material to which there is positive transfer from S's verbal habits is easier to learn than material to which there is no positive transfer. If in a serial-recall experiment one uses the same words in all experimental lists and varies only the arrangement of words within the lists, meaningfulness measures such as m (Noble, 1952) or Association Value (Glaze, 1928) would predict no differences in learning because meaningfulness predictions are based on the sum of the values of the indi: 1 This paper is based on a doctoral dissertation submitted to The Johns Hopkins University in May, 1963. The author wishes to acknowledge the advice and guidance of Professor James Deese. 2 Now at Sul Ross State College, Alpine, Texas.

vidual words in each list. If each list has the same words, each list has the same meaningfulness value. In order to predict differences in learning and recall of lists which differ only in the arrangement of words within the lists, one must have a measure which reflects the strength of associations between words. One such measure is the overlap coefficient (Deese, 1962). It is computed by summing the direct associations and common associations between two words and dividing by twice the number of Ss in the sample. Essentially the same measure was proposed by Bousfield, Whitmarsh , and Berkowitz (1960), who called it mutual frequency score. The overlap coefficient is a measure which reflects not only the strength of direct associative connections between words (the degree to which they elicit each o t h e r ) b u t also reflects the extent of stimulus equivalence (Jenkins, 1959) between two words (the degree to which they elicit the same responses). Deese has shown that collections of words which have high overlap with one another produce highly organized clusters upon factor analysis. There are at least two ways to use the overlap coefficient to predict results obtained from serial lists which contain the same words in different arrangements. One way is to compute the mean of the overlap coefficients between adjacent pairs of words (O). The list arrangement with the highest mean O should be the easiest to learn and recall. Such a measure should not be confused with interitem associative strength (Deese, 1959). Unlike O, interitem associative strength is based only on

494

EXTRAEXPERIMENTAL

direct associations between words and is the same for any arrangement of the list. The O measure not only includes such direct associations but also includes common indirect associations and depends upon particular relations between adjacent items. The other way to use the overlap coefficient is to vary the extent to which Deese's factor clusters are kept intact (F). The list with the highest F, as measured by the number of intact factor clusters, should be the easiest to learn. Weingartner (1963) has shown that F is a significant predictor of the number of trials required to learn a list by serial anticipation. In Weingartner's experiment, however, there was a high correlation between O and F. In fact, the O measure predicts Weingartner's results about as well as the F measure. EXPERIMENT

I

The purpose of Exp. I was to see if both F and O predict ease of learning and recall when

lists are constructed so that F and O are orthogonal. This experiment was similar to Weingartner's except that a different method of presentation and a different measure of learning were used. Method Subjects. The Ss were 36 Johns Hopkins undergraduates fulfilling a course requirement. Materials. Four 16-word lists were constructed which consisted of the word butterfly and 15 highfrequency associates to butterfly. Table I shows that in the two lists designated as high F, the words are grouped into four 4-word clusters (e.g., moth, butterfly, insect, and bug form one of Deese's factor clusters; bird, wing, bees, and fly form another cluster, etc.). In the low F lists none of these 4word clusters was kept intact in the construction of the lists. These 4 lists represent extremes of F since it would have been possible to construct lists with 1, 2, or 3 clusters left intact. By restricting the range of O in these 4 lists from a possible 0.204-0.001 to 0.142-0.063, one can obtain a list with high O (0.140) which keeps all 4 factor clusters intact (high F ) ; a list with high O (0.142) which keeps none of the factors intact (low F ) ; a list with low O (0.078) and high F; and-a list with low O (0.063) and low

TRANSFER

EXP.

495

TABLE 1 LISTS WITH OVERLAP COEFFICIENTS

High F High 0

High F Low O

Low F High O

Low F Low O

moth 0.18 a butterfly 0.14 insect 0.51 bug 0.03 bird 0.44 wing 0.10 bees 0.07

insect 0.14 butterfly

color 0.03 sunshine

color 0.03 sunshine 0.00 bug 0.51 insect 0.00 garden 0.05 yellow 0.06 butterfly 0.00 sky 0.00 wing 0.00 blue 0.00 bees 0.01 summer 0.00 fly 0.17 moth 0.12 bird

fly 0.01 blue 0.31 color 0.08

sky 0.03 yellow 0.09 sunshine 0.06 summer 0.02 garden 0.03 spring

0,18

0,00

moth 0.08 bug 0.03 bird 0.04 fly 0.07 bees 0.10 wing 0.00 yellow 0.03 sky 0.08 color 0.31 blue 0.00 spring 0.03 garden 0.02 sunshine 0.06 summer

bug 0.51 insect 0.14 butterfly 0.03 garden 0.05 yellow 0.03 sky 0.46 blue 0.02 summer 0.01 bees 0.10 wing 0.44 bird 0.12 moth 0.17 fly 0.02 spring

0.00

spring

a The number between a pair of words is the overlap coefficient of that pair. F. The high F-high O list and the low F-low O list were taken directly from Weingartner's experiment. When F was scored dichotomously (high and low) ; the point-biserial correlation between F and O for the 4 lists was only 0.09. A warm-up list was constructed from slow and 13 high-frequency associates to slow. The 0 of this list was 0.014. Procedure. Each of the 4 experimental lists was given to a different group of 9 Ss. No S could be tested on more than one list because all the lists contained the same words. All Ss who received a particular list were tested at the same time. The Ss first received 2 free-recall trials on the warm-up list, and then 3 serial-recall trials on their experimental list. All lists, including the warm-up list, were

496

MILLER

presented at a 2.25-sec. rate by means of a 35mmslide projector with an automatic timing device. The Ss saw all the words in the list one at a time before they were required to respond. Pertinent instructions for the experimental list were: "Write down the words in the order in which they are presented. When you have gone through the list once, stop and raise your hand as a signal that you have finished. Do not go back and try to fill in blank spaces." The Ss responded by writing on carbon-backed

paper. The Ss used glass pencils which left no impression on the paper, so they could not see what they had written. When all Ss in a group indicated that they had finished writing, a trial was considered completed and a new trial was begun immediately. Length of trial varied from 76 to 115 sec., and there was no significant relation between the length of trials for a group and that group's performance on their list (tau = 0.33, p = 0.21).

Results Each S contributed two scores to a covariance analysis: his number of correct words, regardless of order, on the first warm-up trial (used as a control for learning ability) and as the dependent variable, his number of correct adjacent pairs of words, summed over all 3 trials. A pair of words had to appear in the same relative order as they were presented in order to be counted correct. For instance, if S recalled the 1st, 2nd, 3rd, 5th, and 4th words from his experimental list in that order, he was given credit for two correct pairs. The technique of presentation and testing used in this experiment is roughly analogous to the serial anticipation technique. TABLE 2 MEAN CORRECT PAIRS IN EXP. I High F

Low F

Overlap

Mean

SD"

Mean

SD

High O Low O

30.11 24.89

7.29 3.54

20.22 17.33

5.63 6.61

The covariance analysis (Ostle, 1954, Ch. 13) proved O to be a significant predictor (F ~- 13.77, d/ = 1, 31 p < 0.001) of the number of correctly recalled pairs. As in Weingartner's experiment, F also had a sig-

nificant effect (F --- 28.62, d] ~

1, 31 p <

o.ool). EXPERIMENT II

The purpose of Exp. I I was to show more directly that the pre-experimental associations which S brings to the experiment (as measured by O) can influence the recall of a list of words. More specifically, it was predicted that when S is allowed to choose his own order of recall, as in free recall, he will recall the words in orders such that there will be a positive correlation between the magnitude of the overlap coefficient of a given pair of words and the frequency with which that pair appears in adjacent position in recall. Thus, if S gives a particular word in free recall, the word he is most likely to recall next will be the word which has the highest associative overlap with the word he has just recalled.

Method Subjects. The Ss were 36 Johns Hopkins undergraduates who had not participated in Exp. I. Materials. Only one list of words was used: the low O-low F list from Exp. I (Table 1). Procedure. All Ss were given one free-recall trim exactly like the warm-up trial in Exp. I.

Results Each of the 166 adjacent pairs of words which was recalled by at least one of the 36 Ss, but had not appeared in adjacent position in the list as presented, contributed two scores to a correlation matrix: the overlap coefficient of the pair and the frequency of occurrence of the pair summed across all Ss. The correlation was thus between the overlap coefficient of pairs of words and their probability of occurrence in adjacent position in free recall. The distributions, projected on both axes, were J-shaped, so the Spearman rank-difference correlation coefficient was used. The correlation was in the predicted direction (rho 0.27, p < 0.001) but was small in magnitude. The small size of rho was in large measure due to the fact that 77 of the 166 pairs occurred only once in recall. The low predict-

EXTRAEXPERIMENTAL TRANSFER

ability is in part due to at least two causes: (a) the words in the list were so interrelated that S could associate to the whole complex of words instead of just the last word recalled, and (b) the list was so short that a good part of it could be held in immediate memory in the presented order so that little or no associative mediation was necessary for recall. DISCUSSION

497

of O in this case would allow only orders such as CBA which put A and B in adjacent positions. (2) A second explanation, suggested by Deese (1962), would be that the factor structure reflects the effects of some underlying categories of organization which might include nonverbal mediators and is thus more basic than the sum total of the associations between words, and that the distributions of associations underlying the overlap measure are derived from these categories.

The most important finding was that F and O could independently predict ease of serial SUMMARY learning. In Exp. I, the combined prediction Four groups, of 9 Ss each, learned the same of the simple effects of F and O accounted for 59.6% of the total variance when the effects 16 highly related words in 4 different serial of learning ability were held constant. The arrangements. The arrangement of words was interaction between F and O accounted for determined by two aspects of internal assoonly 1.3% of the same variance. Thus, even ciation: factor structure (F), derived from a though F and O are both based on the over- factor analysis of overlap coefficients; and lap coefficient, they can be said to measure mean overlap (O), also derived from overlap different aspects of internal structure of lists coefficients. Both F and O were shown to be separate and significant predictors of the of words. There are at least two explanations as to number of correctly recalled adjacent pairs what the two aspects of associative structure of words. The overlap coefficient was also measured by F and O might be. (1) It could shown to be a weak but significant predictor be that while O measures only the pre-experi- of the order of words in free recall. mental associative strength between adjacent REFERENCES pairs of words, F measures also the preBOUSt~IELD, W. A., W]tlTM:ARStt~ G. A, AND BERKOexperimental strength of the remaining remote WlTZ, H. Partial response identities in associaassociations. If so, it should be possible to tive clustering. J. gen. Psychol., 1960, 63t 233238. compute higher order O's (spanning intervening words) which, if weighted properly, DEESE, J. Influence of inter-item associative strength upon immediate free recall. Psychol. Rep., 1959, would account for all of the predictable vari5, 305-312. ance. It should be noted that an intact factor DEESE, 3". On the structure of associative meaning. cluster (as in the high F lists) increases the Psychol. Rev., 1962, 69, 161-175. chances for positive transfer by bringing to- GLAZE, J. A. The association value of non-sense syllables. J. genet. Psychol., 1928, 3fi, 255-269. gether the words which are most highly assoJEI~KINS, J. J. A study of mediated association. ciated with one another, but it also increases Studies o] verbal behavior: Report No. 2. the chances of negative transfer since all the Minneapolis: University of Minnesota, 1959. words in the cluster cannot be placed in a NOBLE, C. E. The role of stimulus meaning (m) in serial verbal learning. J. exp. Psychol., 1952, 43, position which is adjacent to each other word. 437 -446. For instance, if words A, B, and C all belong OSTLE, B. Statistics in research. Ames: Iowa State to the same cluster, but the strongest associaColl. Press, 1954. tion is between A and B, the arrangement WEINGARTNER, H. Associative structure and serial ACB might lead to some negative transfer learning. J. verb. Learn. verb. Behav., 1963, 2, 476-479. since there would be a tendency for A and B to move together in recall. The maximizing (Received 3,une 5, 1963)