Memory and organization: Category recall and retrieval capacity

Memory and organization: Category recall and retrieval capacity

JOURNAL OF VERBAL LEARNING AND VERBAL BEHAVIOR 11, 157-163 (1972) Memory and Organization: Category Recall and Retrieval Capacity 1 DAVID R. BASDEN2...

492KB Sizes 0 Downloads 31 Views

JOURNAL OF VERBAL LEARNING AND VERBAL BEHAVIOR

11, 157-163 (1972)

Memory and Organization: Category Recall and Retrieval Capacity 1 DAVID R. BASDEN2 AND JERRY HIGGINS Universtty of Cahfornia, Santa Barbara, Santa Barbara, Cahfornia 93106 Three experiments are reported concerning the hnear relationship between recall (R) and the number of categories (NC) formed in sorting of words In Experiment I, free recall learnmg preceded sorting into two to seven categorzes; the NC-R correlatzon was not sigmficant. In Experiment II, the category hmzt and sorting criterion were removed, and a s~gmficant NC-R correlation was observed In Experiment III, Ss sorted as m Experiment II and then recalled; the NC-R correlation was significant for the full range of NC (2-19). In wew of (a) the posmve correlation over the upper range of NC, and (b) the apparent absence of a hm~t on category retrieval, an alternative to Mandler's model was offered. In a recent series of papers, Mandler (1967, 1968, 1970) has proposed and tested a model of m e m o r y storage and remeval Briefly stated, the model holds that (a) presented words are stored in m e m o r y in an organizational scheme taking the form o f hierarchically arranged categories, (b) retrieval from storage proceeds from higher order categories through lower order categories to words and (c) storage and retrieval capacity at any level o f the hierarchy is h m a e d to about 5 ! 2 units per category. Thus m a two-level system subjects (Ss) can store and retrieve only about five categories and only about five words per category. Tests o f the model have been hmited to such a two-level situation, and have typically taken the form o f having Ss organize unrelated words into from two to seven categories and then recall the words they have organized. Recall order is free; however, the h~gh level of clustering leaves no d o u b t that recall is in terms of the S's organizational schema. Limiting the n u m b e r o f categories which m a y be formed to a m a x i m u m o f seven insures 1 Supported by a predoctoral National Institute of Mental Health fellowship to the semor author and a faculty research grant from the Umversxty of Califorma, Santa Barbara to the jumor author. The authors thank Barbara Basden, Jo Anne Hewat and Eleanor Johanson for their assistance. 2 Now at Fresno State College, Fresno, CA 93710.

that all categories formed will be recalled. Assuming that a sufficient n u m b e r o f words is revolved to insure five words per category, recall (R) should be a function o f the n u m b e r o f categories formed (NC). Thus a S w h o forms two categories should recall about 2 x 5 = 10 words, a S who forms three categories should recall about 3 × 5 = 15 words, etc. Mandler's studies have reliably shown a relatively strong hnear relationship between N C and R, as predicted by his model. Although Mandler's tests o f h~s model have been limited to situations m which categorization is required prior to recall, he has suggested that the model ~s also apphcable in the multitrial free recall situation (Mandler, 1967, p. 366). In other words the storage/remeval limitation o f 5 ± 2 units is proposed to hold for situations in which categorization is not explicitly required prior to recall. In an earlier paper, Higgins and Basden (1968) attempted to demonstrate the N C - R relationship in such situations. In each o f six experiments free recall learning was followed by a sorting task designed to reveal the n u m b e r o f categories Ss had formed during the learning o f the material. A l t h o u g h significant clustering in terms of sorted categories was observed m all experiments, in none did a significant N C - R correlation appear. One weakness m these studies was our failure to instruct the Ss to base 157

© 1972 by Academic Press, Ine

158

BASDEN AND HIGGINS

their sorting o n the category schema (ff any) they had formed d u r i n g learning. W i t h o u t specific instructions to the contrary the Ss p r o b a b l y made use of the sorting task to improve on existing o r g a m z a t t o n , therefore, m the first two experiments reported below Ss were instructed to sort m such a way as to reveal any groupings they had formed during learning EXPERIMENT I In this experiment Ss were given free recall training and were then asked to sort index cards bearing the words from the hst into from two to seven categories. To avoid requiring the S to sort words not i n c o r p o r a t e d into h~s schema, only those words recalled on the final free recall test were supphed for sorting. The sorting procedure was that discussed by M a n d l e r and Pearlstone (1966).

Method SubJects The Ss were 20 female introductory psychology students serving in partial fulfillment of course requirements at the Umverslty of California, Santa Barbara. Materials and apparatus. The word set employed was a 52-word list of mixed Thorndlke-Lorge frequency (Mandler & Pearlstone, 1966). Words were presented for learning using a Lafayette standard memory drum, and for sorting using 3 x 5 in. index cards. P~ocedure. During the learning phase Ss were given three presentation-test cycles with a different randomization of the words used for each Words were presented via memory drum at a two-sec rate; S read each word aloud as at appeared. During test phases, each word was written on a separate index card, with the test continuing until 60 sec had elapsed without the addition of a new word. A 30-sec interval during which S counted backward aloud by 7's was Interpolated between presentation and test as a recency control (Postman & Philhps, 1965). In the second phase, the S was informed that he was now to sort the words of the list mto groups in such a way as to reveal any groupings of related words he had employed as an aid to learning. He was further told that he must form no fewer than two and no more than seven groups. The actual sorting task proceeded as follows: a deck of cards bearing the words of the list was placed before the S, he removed the top card from the deck and placed it in one of seven sectors of the table, and continued this procedure

until all cards had been placed. At any one point only the top card in the deck and the last card placed in each sector were visible. When all cards had been placed, the categories were collected, a new deck of cards bearing the words in a different random order was provided, and the sorting procedure was repeated. This continued until the S had sorted the words of the list into the same categories twice an succession Th~s procedure is identical with that described by Mandler and Pearlstone (1966).

Results and Discussion M e a n recall on the final test was 30.2, with a standard deviation of 7 7. A mean of 3.3 sorting trials was required and the s t a n d a r d deviation was 0 8. The m e a n and standard deviation of the n u m b e r of categories formed were 5 8 a n d 1.4, respectively. The correlation between N C a n d R was essentially zero, r(18) = - 0 2 . p > . 1 0 . Significant clustering was observed, t(19) 5.34, p < .001, as tested by the method of Bousfield a n d Bousfield (1966). A clustering index c o m p a r a b l e to that employed by M a n d l e r (1967) was calculated using the formula, O(SCR) - E(SCR) divided by M ( S C R ) - E ( S C R ) , where O(SCR) a n d E(SCR) are as defined by Bousfield a n d Bousfield (1966), a n d M ( S C R ) is the maxim u m clustering possible (given by the n u m b e r of words recalled m i n u s the n u m b e r of categories represented in recall). Its m e a n value was .27; considerably lower t h a n the m e a n value of .53 o b t a i n e d in a replication of the M a n d l e r a n d Pearlstone experiment (Hlggms & Basden, 1968). Obviously, instructing the Ss to sort in such a way as to reveal o r g a m z a t m n formed d u r i n g learning was n o t sufficient to bring a b o u t a significant N C - R correlation, at least n o t for this procedure However, a s s u m i n g that Ss impose categorical o r g a n i z a t i o n d u r i n g m u l t l m a l free recall this lack of a n N C - R relationship could indicate that the 5 ± 2 limit on processing capacity does n o t apply in this situation. If Ss do not restrict themselves to the use of seven or fewer categories when orgamzing the material d u r i n g the learning phase, imposing a limit of 2-7 categories

MEMORY AND ORGANIZATION

159

arrangement. Thls replicatlve sort was introduced to provide E with reformation regarding the stabihty of the orgamzatlon achieved. After the S had completed the rephcatlve sort, he was required to label or otherwise describe the groupings estabhshed.

d u r i n g the sorting phase could only obscure a n y existing relationship between n u m b e r of categories sorted a n d n u m b e r of words recalled. Experiment II was u n d e r t a k e n to determine whether restricting the n u m b e r o f categories the Ss could use d u r i n g sorting was responsible for the absence of an N C - R relationship in Experiment I. I n addition, the r e q m r e m e n t that Ss meet a stabahty criterion in sorting was dropped since the only p u r p o s e of the sorting task in the present procedure Is to reveal existing organization.

Results and Discusston M e a n recall on Trial 3 was 28 6, with a standard deviation of 7.5, and the m e a n n u m ber of categories formed d u r i n g sorting was 8.7 with a standard deviation of 3.7. The n u m b e r of categories formed ranged from 2 to 18. The correlation between n u m b e r o f categories a n d n u m b e r of words recalled was significant, r ( 3 6 ) = . 5 8 , p < . 0 1 . Figure 1 confirms the linearity of this relationship t h r o u g h o u t the entire range of n u m b e r o f categories. Clustering was slgmficant, t ( 3 7 ) = 7.82, p < .00l ; the m e a n clustering index for Trial 3 recall was .42. The correlation between n u m b e r of categories and recall, though n o t as large as those reported by Mandler, stands m m a r k e d contrast to those observed in our earlier experiments. Specifically in c o m p a r i n g the results of Experiments I a n d II, it would appear that the restrictions inherent m the M a n d l e r sorting procedure do obscure the relationship in the m u l n t r l a l free recall situation. The level of clustering observed m the present experiment is of the same order of m a g n i t u d e as that observed using the M a n d l e r sort-recall paradigm, suggesting that cate-

EXPERIMENT II

Method The Ss (N = 38), matermls, apparatus, and learning phase of th~s experiment were of the same descr~ptmn as m Experiment I. Only the nature of the sorting task differentmted the two experiments. As m Experiment I, each S was g~ven the deck of index cards bearing the words of the hst he had recalled on the final test and was mstructed to sort the words into categories m such a way as to reveal any groupings he had employed as an aid to learmng However, no restriction was placed on the number of categorxes which could be used and S was permitted to rearrange cards until he was satisfied with his categorizaUon. This procedure d~ffers from that employed by Mandler and Pearlstone (1966) in a number of respects. In addition to the removal of a category limit, Ss were permitted access to all words simultaneously and were permitted to move cards from one category to another ff dissatlsfied with original placement. Although no criterion of sorting stabihty was reqmred of the S, he was given a second deck bearing the words m a different random order and was asked to rephcate h~s original category 5O

45 /

~35 ~3e -E~25 z~ ~2a

/ /

/

J

1Q r

r

l_r 5

]

,

Number

i

J t r , , f , 10 15 of C a t e g o r i e s

,

r

,

J 20

FIG. 1. Scattergram of the relationship between number of categories formed m sorting and number of words recalled, ftted by the regresslon hne, R = 1 1 NCR + 16.8.

160

BASDEN AND HIGGINS cedes recall. M o r e specifically, the p u r p o s e o f E x p e r i m e n t I I I was to d e t e r m i n e whether category retrieval w o u l d be limited to 5 ± 2 when a sorting task o f the type e m p l o y e d in E x p e r i m e n t II precedes recall.

gorlcal o r g a n i z a t i o n is e m p l o y e d to nearly the s a m e extent in the two situations One aspect o f the present experiment prevents a completely u n a m b i g u o u s Interpretation o f the results. W o r d recall, R, can be viewed as the p r o d u c t o f two factors: the n u m b e r o f categories recalled ( N C R ) a n d the n u m b e r o f instances recalled per category represented m recall (IPC) Since only recalled w o r d s were e m p l o y e d an the sorting task, the n u m b e r o f categories p r o d u c e d in sorting m o r e p r o p e r l y represents N C R than it does the n u m b e r o f categories f o r m e d during the learning phase. O f course, if Ss recall all categories f o r m e d during learning, then N C R will accurately reflect the n u m b e r o f such categories. However, when all categories f o r m e d are n o t represented in recall, the c o r r e l a t i o n between N C R a n d R m a y overestimate the N C - R relationship to some extent, since the N C R - R c o r r e l a t i o n in such cases w o u l d a m o u n t to a c o r r e l a t i o n between recall a n d one o f the factors c o m p r i s i n g it. A l t h o u g h the relatively large N C R values create some i n t e r p r e t a t i o n a l difficulties, they nevertheless do indicate that the 5 + 2 limit on retrieval capacity is n o t a p p l i c a b l e to the multitrlal free recall situation. P e r h a p s this h m i t is unique to the p a r t i c u l a r p a r a d i g m e m p l o y e d by M a n d l e r . The following experiment was u n d e r t a k e n to determine whether the storage/retrieval l i m i t a t i o n o f 5 ± 2 is characteristic o f p r o c e d u r e s in which sorting pre-

EXPERIMENT I I I

Method The Ss (N = 71) and the materials were of the same description as in Experiments I and II. The procedure was to give S a deck of 52 index cards bearing the words of the hst. He was instructed to sort the words into groups of related words based on some criterion involving the word as a unit. No limit was placed on the number of categories which could be used, Ss were permitted to rearrange words until satisfied with their orgamzatlon, and no hmit was placed on sorting time. The S was asked to repeat his arrangement and then to label his categories. In other words the sorting procedure was ldentacal with that of Experiment II with the single exception that the sorting deck contained all 52 words of the list Following the sorting phase S was asked to write as many of the words as he could remember in any order he found convenient. Recall proceeded until 60 sec had elapsed w,thout the addition of a new word.

Results M e a n N C was 8.6 with a s t a n d a r d d e v i a t i o n o f 4.7, a n d m e a n R was 30.9 with a s t a n d a r d d e v i a t i o n o f 9.4. T h e N C - R c o r r e l a t i o n was significant, r(69) = 65, p < . 0 1 ; F i g u r e 2 confirms the linearity o f this relationship t h r o u g h o u t the entire range o f N C (2-19). E a c h recall score was b r o k e n d o w n into the n u m b e r o f categories represented m recall a n d

50 !

~5 40

o ~ 3 5

:



,

"

,

°

S

"

°

15 10 L

~

I 5

Number

10

of Categories

15

20

FIG. 2. Scattergram of the relationship between number of categories formed in sorting and number of words recalled, fitted by the regression line, R = 1.3 NC + 19.9.

MEMORY AND ORGANIZATION

the mean number of instances recalled from represented categories. Mean N C R was 7.8 wlth a standard deviation of 3.6, and mean IPC was 4 0 with a standard deviation of 1.3. The N C R - R correlation was significant, r(69) = .81, p < .0l. Clustering was also significant, t(67)= 11.97, p < .001. (The data of three Ss who failed to write words in the order recalled were excluded from the clustering analysis.) The mean value of the clustering index was .60. To facilitate comparison with earher experiments the data of the 71 Ss were separated into those of the 31 forming seven or fewer categories (Group F) and those of the 40 forming eight or more categories (Group M). For G r o u p F, mean NC was 4.6 with a standard deviation of 1.6, and mean R was 24.7 with a standard deviatlon of 8.1. The N C - R correlation was significant, r ( 2 9 ) = .66, p < .005. For this group mean N C R was 4.4, with a standard deviation of 2.4, and mean IPC was 5.6, with a standard deviation of 1.2; the N C R - R correlation was also significant, r(29) = .73, p <.005. The mean clustering index was .44. For Group M mean N C was 11.8 with a standard deviation of 4.2, and mean R was 35.6 with a standard deviation of 8.1; the N C - R correlation was significant, r(38) = .47, p < .005. For this group mean N C R was 10.3 with a standard deviation of 3.3, and mean IPC was 3.5 with a standard deviation of 1.5; the N C R - R correlation was significant, r(38) =.77, p < . 0 0 5 . The mean clustering index was .74. Obviously the relationships which hold for the full range of NC values hold separately for the two subranges of the data represented by Group F and Group M. Discussion

Although both the free sorting and the Mandler sorting procedures appear to produce a linear N C - R relationship in the sortrecall paradigm, the basis of this relationship appears to be in need of further exploration. Apparently there is no fixed limit on the number of categories which can be formed and 6

161

recalled, as the mean and range of the N C R values demonstrate. Of particular importance is the existence of a significant N C - R correlation for Ss forming eight or more categories, demonstrating that recall increases as a function of NC throughout the entire range of NC. It follows that as N C increases the number of words placed in each category must decrease. If the number of words placed in each category decreases, then the number of words retrievable from each category must decrease. If N C R is fixed at seven, as suggested by Mandler, and IPC decreases as N C increases, then beyond some value of NC, recall (NCR times IPC) must begin to decrease. Thus according to the Mandler model, recall first increases as a function of N C but must eventually decrease as N C grows large. For a 52 word list the point of inflection should come somewhere between N C = 5 and N C = 10. No such point of inflection was observed in the present experiment. The reason for this is apparent, N C R was not hmited to seven but on the contrary increased uniformly with NC. One may ask first why Ss form so many categories in this experiment when in Mandler's experience (Mandler, 1967, Experiment G) most Ss decline to use more than seven when given the opportunity. The answer to this question probably has to do with the nature of the sorting task employed. In the Mandler procedure the S's task is to meet a criterion of sorting stability The most efficient technique for meeting this criterion would be to place two particular words in one category and all of the remaining words in a second. Fortunately most Ss are not this efficient, however, in our experience, they do rely upon clear-cut sorting schema such as part of speech, concrete versus abstract and the like. The use of this general type of category undoubtedly results in part from the necessity of meeting the stability criterion but also may reflect a tendency to adopt a sorting scheme before the range of semantic relationships existing among the words is apparent. By contrast, the free sorting method employed in

162

BASDEN AND HIGGINS

Experiments II and III obviates the S's use of easily applied sorting schemes and at the same time permits the S to examine the range of semantic relationships available prior to initial categorization. Frequently, Ss shuffled through the deck or arrayed all the cards on the table before collapsing words into categories. It is likely then that the difference in the demands of the two sorting tasks is responsible for the rather extended range of N C values observed in this experiment as compared with those of Mandler. A second question one may ask is given that Ss sort into many categories why are they able to retrieve such a large proportion of them. An answer within the framework of the Mandter model might be that Ss who formed many categories resorted to the use of higher-order hierarchies; that is, formed categories of categories. Mandler has argued that the use of such multilevel hierarchies permits circumvention of the processing capacity limitation. Although this argument is in no way diminished by the results of the present experiment, there is little or no evidence in the literature to indicate that Ss spontaneously form multilevel hierarchies from sets of unrelated words or that if they do this facilitates recall. A simpler conclusion would be that no fixed limit on retrieval capacity exists but rather that other things being equal N C R is a negatively accelerated function of NC. While this argument might'suffice as an explanation for the hlgh level of N C R observed, it does not account for the hnear relationship between N C and R To explain this relationship, we offer the following analysis. Given a set of randomly selected words it is probably easier to find pairs of related words than to find sets of three related words, to find sets of three related words than to find sets of four related words, etc. In essence what we are saying is that if Ss base their categories upon associatwe relationships among the words, then the smaller the category the more highly associated will be the members to one another. If Ss choose to form small categories comprising

highly interrelated words, they must of necessity form many m order to categorize all the words of the list. It has been established that recall in general (Deese, 1959) and that recall from represented categories in particular (Cohen, 1963) increases as a function of intermtem associative strength. In other words we are arguing that when categories are Sdefined recall increases as a function of NC because as N C increases and category size decreases the lnteritem associative strength of categories increases The relationship is then actually between recall and intemem assocmtlve strength. If it were possible to hold rateritem associative strength constant while varying NC within a constant list length the present analysis would lead to the prediction that recall would decrease as a function of NC. The present lr~terpretatlon is admittedly post hoc and m need of additional support However, it is consistent with interpretations of related studies and is therefore offered as a tentative explanation of the N C - R relationship observed here and elsewhere. REFERENCES BOUSFIELD, A. K. ~ BOUSFIELD, W A. Measurement

of clustering and of sequential constancies m repeated free recall. Psychologtcal Reports, 1966, 19, 935-942. COHEN,B. H. An investigation of recodmg in free recall. Journal of Experimental Psychology, 1963, 65, 368-376. DEESE, J. Influence of interltem assocmtive strength upon immediate free recall. PsychologgcalReports, 1959, 5, 305-312. HIGGINS, J., & BASDEN,D. Memory and orgamzation. Proceedmgs of the 76th Annual Conventgon of the American Psychological Assoclatwn, 1968, 3,

75-76. MANDLER,G. Orgamzatlon and memory. In K. W. Spence & J. T. Spence (Eds.), The psychology of learning and motwatlon: Advances m research and theory. Vol. 1. New York: Academic Press, 1967.

Pp. 327-372. MANDLER, G. Assocmt~on and organizatmn: Facts, fancies, and theories In T. R. Dixon & D. L. Horton (Eds), Verbal behavior and general behavior theory Englewood Cliffs, NJ: Prentice-Hall, 1968. Pp. 109-119.

MEMORY AND ORGANIZATION MANDLER,G. Words, hsts, and categories: An experimental v~ewof orgamzed memory. In J. L. Cowan (Ed.), Studws in thought and language. Tucson: Umverslty of Arizona Press, 1970. Pp. 99131. MANDLER,G., & PEARLSTONE,Z. Free and constrained concept learning and subsequent recall. Journal

163

of Verbal Learnmg and Verbal Behavior, 1966, 5, 126-131 POSTMAN,L., & PHILLIPS,L. W. Short-term temporal changes in free recall. Quarterly Journal of Expertmental Psychology, 1965, 17, 132-138. (Received August 4, 1971)