JOURNAL OF VERBALLEARNING AND VERBALBEHAVIORS, 489-495 (1964)
The Development of Constancies in Sequential Organization during Repeated Free Recall PUFF, AND T . M . COWAN 1
W . A. BOUSFIELD, C. R.
University o/ Connecticut, Storrs, Connecticut
Tulving (1962) reported a useful method for measuring the tendencies of Ss to develop constancies in the subjectively determined sequential organization of their repeated recall of a list of words which were unrelated in the sense that E was unaware of the sources of organization available to Ss. His results indicated a monotonic increase in subjective organization when the serial order of the words was systematically varied from trial to trial. The present study was undertaken to determine the results of using a modified form of Tulving's method combined with a very different statistical treatment o f the data. Prior research on subjective organization in free recall has centered on clustering. As Tulving has indicated, these studies have employed a method essentially opposite to the one he developed since they have depended on the use of pre-experimentally organized lists of words. Typically, E has prepared the lists so as to comprise subgroups of items chosen on the basis of their representing coordinate categories. A list of words thus assembled is then presented in a random order for free recall. Clustering is shown in the tendency of Ss to recall the items in sequences or runs, with each run comprising items in the ~ m e E-determined category. Several examples of lists successfully used for inducing clustering may be cited. Jenkins 1 This research was supported by a grant to the senior author from the Research Foundation of the University of Connecticut. 489
and Russell (1952) employed a list containing S-R pairs chosen from free associational norms. Bousfield (1953) used words representing taxonomic categories such as animals, names, professions, and vegetables. Cofer (1959) used sets of synonyms. The favored explanation of clustering is that Ss tend to bring the separated but related items together on the basis of their inter-item associative connections. This interpretation was first tested by Jenkins and Russell (1952) by their selection of directly associated words for their stimulus list. As Marshall and Cofer (1963) have pointed out, however, clustering appears also to depend on indirect or mediated associations. In considering the properties of words capable of determining the order of their emission, mention should certainly be made of frequency of usage. This has received formal recognition in the spew hypothesis of Underwood and Schulz (1960). There are, of course, other properties of words which may reasonably be assumed to provide a basis for the ordering of emitted responses. Some of these properties are general and some are probably idiosyncratic. The inference suggested by available evidence is, as Postman (1963) pointed out, that it is virtually impossible to compile a list of unrelated words. He states, "There always remains a multiplicity of formal, grammatical, and semantic dimensions which provide a basis for sequential linkages." With a wide assortment of bases on which S can organize his recall there is also a wide range over
490
BOUSFIELD, PUFF, AND COWAN
which he may depart from a n y single basis selected in advance by E. Tulving's method is such as to provide an objective measure of the incidence of sequential organization on a n y basis S m a y choose. Because of its potentiality as a needed tool for the study of organizational processes, it merits being tested for its sensitivity. The present study was planned to provide such a test, and the results indicate the method should be useful for asse~ing both general trends and individual differences in subjectively determined organizational processes.
writing" appeared in the apparatus and stop when /~ said "Stop." A new data sheet was provided for each trial. The starting signal appeared 1 sec after the appearance of the last item on a list, and a limit of 30 sec was allowed for writing the recalled items. A new trial began immediately after each recall. A period of 2 rain, however, elapsed between the completion of the practice and the beginning of the experimental session. The S was told when he had correctly recalled all the words. It was also indicated that the experiment would continue while S was to try to increase his speed in reproducing the items. Regularly thereafter, E urged S to go faster, and showed him his recorded time. The learning of the experimental list was continued to the criterion of five consecutive errorless recalls.
METHOD
RESULTS
Subjects. The~Ss were 20 undergraduate students
The data of each S were scored so as to provide four measures: (a) the n u m b e r of words recalled on each trial; (b) the number of trials required to reach the criterion of five successive errorless recalls; (c) the number of what are here termed obtained intertrial repetitions, O - I T R ' s , for each successive pair tonight, enemy, subtract, turtle, big, theory, legal, of trials; (d) the number of expected interand family. Five randomizations of the 10 items were prepared, and each randomized list was typed on a trial repetitions, E - I T R ' s , for each successive separate tape. Five Gerbrands memory drums, one pair of trials as computed by a new formula. for each tape, served to expose the items at a 1-sec An O - I T R may be said to have occurred rate. Three randomizations of a practice list of 10 whenever a pair of items occurring consecumale first names were similarly prepared. The tively in recall on trial a was repeated in randomizations were derived from a table of random numbers. This procedure for the randomizations recall on trial n -~: 1. I n the case of the 10differed from that of Tulving, who had a different item stimulus word list used in the present order of his words for each of 16 trials, and con- study, the maximum of nine O - I T R ' s o c structed his sequences so as to avoid second-order or curred whenever S o n two successive trials higher redundancies. Additionally, a stopwatch was recalled all the words in the same order. The used for. timing each recall of the list, and dittoed data sheets were prepared for use by Ss in writing counting of the O - I T R ' s was facilitated b y using a 10 X 10 matrix with the rows reprethe items they recalled. Procedure. The experimental session began with senting the words in the n t h position and .the separate randomizations of the practice list, and the columns representing the words in the was followed by the separate randomizations of the (n + 1)th position. There is a n important experimental items. The instructions indicated a difference between this matrix and that of series of words would appear, one after another, in the window of the memory drum facing S. The task Tulving: He added .a symbol for the blank was to learn all the words, and after each presenta- position of no word immediately preceding tion of the list to write them as rapidly as possible the first word in S's recall, and another symbol on the data sheet. The same words would appear for ~the blank position following the last word. on successive trials but their order of presentation would change. It was emphasized that the order in Thus if his type of matrix had been used here which the words were recalled was of no consequence. there would have been 11 rows and 11 colThe recall was to start when the words, "Start umns. I n terms of the matrix analysis an I T R in psychology who were paid for their services. Stimulus Words and Apparatus. The stimulus words comprised the first 10 of the 15 items in List No. 18 compiled by Deese (1959) on the basis of their having zero interitem associative strength according to the Minnesota norms of Russell and Jenkins (1954). These words were as follows: lake, ride,
ORGANIZATION IN RECALL
may be said to have occurred whenever a given cell was checked twice in succession as a consequence of recording the data for a pair of successive trials, n and n + 1. The measure introduced here for evaluating the significance of the obtained subj ective organization, as represented by the O-ITR's, was the expected number of intertrial repetitions, or E-ITR's. The model for the E - I T R ' s was that of two successive random drawings from the same pool of W items, where each drawing was made without replacement, and where the remaining items were equally available at all stages of both drawings. The pool of W items corresponded to the stimulus-word list and the number of items in each drawing represented the number of words recalled. The formula derived from this model for the expected number of intertrial repetitions for any pair of trials, n and n + 1, was as follows:2 E-ITR'S
(h--l)
(k--l)
w (w--l)
In this expression, h ~_-the number of words recalled on trial n, the first of two consecutive recall trials; k - - t h e number of words recalled on trial n-Jr 1, the second of two consecutive recalls; and w -~ tt~e number of stimulus words presented for recall. Thus if 10 words were presented for recall, and S recalled 5 on trial n and 8 on trial n - I - l , the number of E - I T R ' s is 0.311. I t may be noted that the formula gives the same number of E - I T R ' s regardless of which 5 words are recalled on trial n and which 8 words are recalled on trial n ~ - 1 . In other words, according to the present model, the amount of overlap in the particular words recalled on the two trials does not serve as a parameter of E-ITR's. The measure E - I T R ' s is not a substitute for Tulving's measure of subjective organization, SO, which he defined as the ratio of the actual to the maximum pos2 The measure presented here was developed by A. K. Bousfield.
491
sible organization. He derived the formula for SO from a measure of redundancy based on information theory.. In the present method the obtained organization is evaluated in terms of its deviation from the amount that might be expected on the basis of random sampling alone. A preliminary examination of the data indicated a high degree of variability in several characteristics. The trials required for reaching the criterion of five successive errorless recalls ranged from 8 to 39 with a mean of 22.05 and an SD of 8.00. The variability in trials to criterion was apparently due to a high incidence of response oscillation, for repeatedly having recalled all 10 stimulus words, Ss would drop to 9 or 8. There also appeared to be wide differences in the trends in the development of organization as represented by the O-ITR's. Four Ss reached the maximum of 9, which as a chance occurrence has a p of 1/10. The five lowest O - I T R values for the last trials were either 1 or 2. In each of these instances equal or higher O - I T R values had been attained o n earlier trials. Because of the variability in the trials to criterion, the measures were treated in units of Vincentized tenths so that comparable stages of performance might better be examined across Ss. First, the mean values of the basic measures as they were divided into decile groups of recall trials were computed for each S. These means were then used in the various analyses of the trends revealed by the data. Table 1 shows the group means of the basic measures. From inspection it appears that while the number of words recalled by the group showed the expected tendency toward negative acceleration, the group progression of O - I T R ' s tended to show relatively large increments in the second half of the learning period. The nature of the changes in O-ITR's during the learning may be seen in the examples shown in Fig. 1. The choice of Ss No. 1, No. 18,
492
BOUSFIELD~ PUFF~ AND COWAN TABLE 1 GROUP ~V~EA~SOF BASIC MEASURES FOR SUCCESSIVEDECILES OF RECALL Deciles
Measure
1
2
3
4
5
6
7
8
9
10
Words recalled
5.50
8.44
8.85
8.46
8.90
9.24
8.99
9.45
9.69
10.00
Obtained I T R ' s
0.46
0.98
1.12
1.15
1.22
1,28
2.36
2.53
3.76
4.05
Expected I T R ' s
0.36
0.65
0.70
0.66
0.71
0.72
0.80
0.80
0.86
0.90
and No. 19 as examples was based on their being representative of the individual differences in performance. As shown in Fig. 1, S No. 18 attained the same amount of organization on the last three deciles as on the third. Several statistical analyses were performed on the differences between the group means of the O-ITR's and E-ITR's shown in Table 1. These O-E differences, as they are here labeled, were tested by combining the data for each of three decile blocks.. The results indicated the O-E difference for the block combining the first three deciles was significant (t(19) - - 2.50, p < 0.05). The difference for the middle block of four deciles was also significant (t(19) - - 3.65, p < 0.01), and a significant O-E difference (t(19)-~6.30, p < 0.001), was found for the last block of three deciles. It was thus established that the amount of subjective organization 9
~8 o
w
: H
: SNO. I S NO. 18
/
2
0
•
I
2
3
TRIALS
4
5
6
7
IN DECILE
8
9
I0
BLOCKS
Fxo, 1. Mean number of obtained intertrial repetitions for pairs of successive trials as a function of trials grouped in deeile blocks.
was significantly greater than chance during all three decile blocks of learning. Additional analyses were undertaken to determine whether there was a progressive increase in subjective organization through the three blocks of deciles. The first of these consisted of testing between the O-E differences for the three decile blocks. The results indicated that the incidence of subjective organization greater than chance for the first block of deciles was significantly less than that for the second block (t(19) - - 2.14, p < 0.05). The O-E differences for the second block of deciles were in turn significantly less than for the last block (t(19) - - 5.84~ p < 0.001). The finding of a progressive increase in subjective organization during the successive recalls was also indicated by the results of applying Page's (1963) nonparametric L-test of ordered treatment means. The data for this analysis were the O-E differences for each decile and every S. For each S the deciles were assigned ranks according to these differences. The ranks per decile were then summed over Ss. The calculations based on these data gave results showing a significant monotonic development of subjective organization from the first through the last decile (L(10,20) - - 7020, p < 0.001). The final two analyses of the data were exploratory tests of Postman's (1963) inference of the virtual impossibility of compiling a list of words which would be free of cues for the ordering of their emission. The first of these involved an analysis of the extent to which the Ss showed similarity in the order in which they recalled the 10 stimu-
ORGANIZATION IN RECALL
lus words on the final criterion trial. Ranks were assigned to the words according to their order of emission for each S. Kendall's coefficient of concordance was then calculated for these data. The result indicated a significant amount of agreement among Ss in the final order in which they recalled the words (W(9) - - 0.131, p ~ 0.01). The second exploratory analysis was a simple test of the possibility that the rank order of emission of the words on the last trial was related to the frequencies of usage of the words given in the Thorndike-Lorge (1944) L-count. This count was based on a sample of approximately 4 ~ million words. The range of the counts for the i0 stimulus words was from 6 for subtract to 1000 or more for big and ]amily. The mean rank order of emission of the stimulus words on the final trial ranged from 3.05 for big to 7.45 for turtle. The rank order correlation of these two measures of t h e ordering of the stimulus wordS was not significant (rho(9) ~ 0.342, p ~ 0.05). It was therefore unlikely that the tendency toward agreement in t h e ordering of the emitted responses depended on the frequencies of usage of the words. DISCUSSION
The results showed that a modified form of Tulving's (1962) method was effective for inducing subjectively determined sequential organization in the repeated free recall of randomized words which had been rated as having zero interitem associative strength. In spite of the differences between Tulving's experiment and the present one, no major discrepancies in the general characteristics of the development of subjective organization had been anticipated. While his measure, SO, is not directly comparable to the O-ITR's of the present study, it appears there is a critical difference between.his findings and ours. His fitted curve for mean SO based on blocks of two and three trials shows a negative acceleration. Most of the increase in our
493
O-ITR's is positively accelerated. A tentative explanation is offered for this difference. The present experiment used a 10-word stimulus list, and learning was continued to the stringent criterion of five errorless trials. Tulving used 16 words and all his Ss were limited to 16 trials. It is likely that our method was such as to result in the occurrence of a substantial amount of response learning relatively early during the period of learning. It may be inferred, furthermore, that the positive acceleration in organization indicated by our data took place after an appreciable consolidation of the response learning. It is a tenable inference that if the learning of Tulving's subjects had been continued to the same criterion as we used, his curve for SO might have become positively accelerated. The positive acceleration could not, of course, continue indefinitely. Regardless of the measure of organization, its curve of development would eventually tend to flatten. A measure of support for these suppositions may be derived from a close inspection of the mean O-ITR's shown in Table 1. If the means through the sixth decile block only are considered, the tend may be described a s negatively accelerated and thus similar to Tulving's curve for SO. The rapid increase in O-ITR's begins in the seventh decile block, and the rate of increase declines noticeably from the ninth t o ' t h e tenth block. The inference regarding the lack of a positive acceleration in Tulving's SO can and should be checked experimentally. Further comparisons of Tulving's results with those of the present study are of questionable value because of differences in method. One of the major theoretical problems involved in the interpretation of the development of subjective organization of the type considered here is that of specifying the processes responsible for the effect. In this undertaking several speculations may be offered. First, it may be assumed that when a given
494
BOUSFIELD, PUFF, A N D
list of items is presented in a serial order for learning, one of the consequences is the establishment of intralist serial associations. Such associations to given items in the list are both adjacent and remote, and they extend in both the forward and backward directions. Generally the forward are stronger than the backward, and the adjacent stronger than the remote. The strengths of these associations, furthermore, reflect the influence of serial position effects. If the given list is repeatedly presented in a constant serial order, the incremental theory indicates the various strengths should increase as a function of the number o'f presentations of the list. In the present experiment, however, the order in which the words were presented was not constant, but changed from trial to trial. A major effect of the changing order of presentation was that the various intralist serial associations progressively became more similar in strength. It would follow that if the stimulus words had intrinsic sequential dimensions, even though they were small, an ordering effect could take place because the competitive advantages of specific words due to favored serial positions tended to become inconsequential. A very different effect of the learning should also be recognized. The repeated presentation of the words, with each presentation occurring in a 'different order, should produce interference, effects on subjective organization. This follows from the assumption that during the learning, each word should in effect become a stimulus cue for an increasing number of competing responses. While such competition should interfere with subjective organization, it may further be assumed that the disruptive consequences of the interference should become reduced during the learning as: (a) the intralist serial associations deriving from the order in which the words are presented become equalized in strength; (b) increments of associative strength are added to the intrinsic connections between the words. This
COWAN
occurs when S repeatedly emits the words in their intrinsically determined order. It may be speculated that the interference effects considered here were responsible for the apparently high incidence of response oscillation occurring during learning. An additional consequence of the changing order of presentation of the stimulus words merits consideration. Not only should each word become a stimulus cue for an increasing number of competing responses, but in addition the number of stimulus cues for the recall of each word should increase during the learning. Perhaps the latter effect should within limits increase the number of items freely recalled. These speculations are intended to be heuristic and obviously they constitute an incomplete theoretical account. The formulation of an adequate theory of the development of subjective organization requires additional evidence, and this undertaking should benefit from Tulving's method. SUNflVfARY
A modified form of Tulving's method was used for appraising the development of subjectively determined sequential organization in free recall. Multiple randomizations were prepared for 10 stimulus words which were unrelated in the sense of lacking interitem associative strength according to free associational norms. Free recall was obtained after each presentation of the words. The unit of organization was the intertrial repetition which occurred whenever a pair of items recalled consecutively on trial n was repeated on trial n + 1. A formula was proposed for estimating the number of these units expected by chance. A monotonic increase in organization was found as the words were learned to the criterion of five consecutive errorless recalls. REFERENCES BOUSFIELD, W. A. The occurrence of clustering in the recall of randomly arranged associates. J. gen, Psychol., !953, $.9, 229-240.
ORGANIZATION IN RECALL
COFER,C. N. A study of clustering in free recall based on synonyms. J. gen. Psychol., 1959, 60, 3-10. DEESE, J. Influence of inter-item associative strength upon immediate free recall. Psychol. Reps., 1959, 5, 305-312. JENKINS, J. J., AND RUSSELL, W. A. Associative clustering during recall. ]. abnorm, soc. Psychol., 1952, 47, 818-821. MARS~a~Lr~, G. R., AND COFER, C. N. Association indices as measures of word relatedness: a summary and comparison of ten methods. Y. verb. Learn. verb. Behav., 1963, 1, 408-421. PACE, E. B. Ordered hypotheses for multiple treatments: a significar~¢e test for linear ranks. J. Amer. Stat. Ass., 1963, fiB, 216-230. POST~C~AN, L. Does interference theory predict too much forgetting? J. verb. Learn. verb. Behav., 1963, 2, 40-48.
495
RUSSELL, W. A., AND JEWKINS, J. J. The complete Minnesota norms /or responses to 100 words ]rom the Kent-Rosano~ Association Tests. Tech. Rep. No. 11, Contract N8ONR-66216 between Office of Naval Research and the University of Minnesota, 1954. THORNDXKE, E. L., AND LORCE, I. The teacher's word book o] 30,000 words. New York: Teachers College, Columbia Univer., 1944. TuLvmo, E. Subjective organization in free recall of "unrelated" words. Psychol. Rev., 1962, 69, 344-354. UNDERWOOD, B. J., AND SCHULZ, R. W. Meaning]ulness and verbal learning. Philadelphia: Lippincott, 1960. (Received January 14, 1964)