Temporal aspects of storage and retrieval in free recall of categorized lists

Temporal aspects of storage and retrieval in free recall of categorized lists

JOURNAL OF VERBALLEARNING AND VERBALBEHAVIOR, 12,499-511 (1973) Temporal Aspects of Storage and Retrieval in Free Recall of Categorized Lists I GEORG...

955KB Sizes 2 Downloads 50 Views

JOURNAL OF VERBALLEARNING AND VERBALBEHAVIOR, 12,499-511 (1973)

Temporal Aspects of Storage and Retrieval in Free Recall of Categorized Lists I GEORGE KELLAS, M A R K H . ASHCRAFT, N A N C Y S. JOHNSON, AND SHERRY NEEDHAM

University of Kansas and Bureau of Chdd Research

The major purpose of the present research was to examine the temporal characteristics of storage and retrieval while manipulating variables related to the organizataonof free recall. In this context, two experiments evaluated the effects of a retrieval plan and degree of hst orgamzation. The provision of a specificretrieval plan led to a decrease in input processing time. In addition, a temporal analysis of recall provided support for the presenceof a category exit component an the retrieval of higher-order memory units. The general pattern of results was viewed as being consistent with a search model based on the retrieval of memory umts according to trace strength.

One of the central problems facing proponents of organization theory is to clarify the role of storage and retrieval processes in the organized output of free recall (cf., Postman, 1972; Wood, 1972). Slamecka (1968, 1969, 1972) has suggested that organized recall results from the orderly execution of a retrieval plan on independently stored items, while others (for example, Mandler, 1967, 1972; Tulving, 1968) have emphasized the influence of organization on the storage of list items. Attempts to examine storage and retrieval as independent processes, however, have met with only limited success. The interpretative problem of inferring storage processes from recall performance is simply that differences in recall may involve either storage or retrieval, or both (Slamecka, 1968). To the extent that a retrieval plan must be formulated and stored during list presentation, it may indeed be impossible to resolve storage-retrieval issues within the limitations of standard methodology.

It appears that the independent examination of storage and retrieval processes is contingent upon the employment of dependent measures capable of monitoring the flow of information into and out of the organism. Kellas and Butterfield (1971) have examined input processing through the use of a technique for measuring subject-controlled item presentation, which proved sensitive to different types of verbal material and recall requirements. The temporal profiles provided by this unobtrusive measure allow direct inferences concerning storage processes which are uncontaminated by output mechanisms. For assessing retrieval processes, Pollio, Richards, and Lucas (1969) have demonstrated the potential usefulness of interword-response times (IRTs), and this has been confirmed by other investigators (for example, Murdock & Okada, 1970; Patterson, Meltzer, & Mandler, 1971). The major purpose of the present investigation was to combine subject-controlled item exposure durations and 1RTs to evaluate the temporal flow of information a This investigation was partially supported by from input through output while manipulating P.H.S. Grants HD 00870 and 00183 from the National variables related to organized recall. SpeciInstitute of Child Health and Human Development. The third author was an Undergraduate Research fically, the effects of a retrieval plan and degree Participant, supported by the National Science of list organization on information storage and retrieval were evaluated. Foundation (GY 8921). Copyright © 1973 by Academic Press, Inc. 499 All rights of reproduction in any form reserved.

Printed in Great Bntam

500

KELLAS ET AL.

Slamecka (1968) has suggested that detection of the organized structure of a list is necessary for the formulation of an effective retrieval plan. When a random, rather than blocked, presentation method is employed, detection of the list structure is less complete, leading to a less efficient retrieval plan and a lower level of clustering during recall. As a test of Slamecka's hypothesis, D'Agostino (1969) provided subjects with knowledge of the list structure, assuming that this would facilitate the formation of a retrieval plan. The provision of general list information, prior to a blocked-random manipulation, was expected to reduce the blocked-random effect on cluster size. This should occur to the extent that detection of the list structure and the subsequent formation of a retrieval plan are sufficient to produce organized recall. D'Agostino's results, however, showed no reduction in the blocked-random effect as a function of providing subjects with general list information and, presumably, a retrieval plan. As a possible explanation, D'Agostino related the presentation order effect to "effective presentation time," rather than to differential detection of the list structure. For blocked presentation, conceptually related 1terns are immediately available for organization; whereas with random presentation, considerable time would be required to retrieve earlier list information essential for organization. Thus with equal presentation rates for blocked and random conditions, the effective presentation time would be greater for blocked presentation and would account for the facilitation in the number of items per category recalled. The hypothesis is supported by evidence indicating that the blocked-random effect is eliminated when a slower presentation rate is used with random presentation, relative to blocked conditions (Cofer, Bruce, & Reicher, 1966). Blocked-random presentation methods can be defined in terms of the number of category repetitions present in the experimental lists; that is, the number of times any word is

directly followed by another word from the same category during list presentation (cf. Puff, 1966). In the present investigation, use of the category repetition measure allowed an extended evaluation of the effects of systematically varying the presentation order along a continuum, ranging from maximum possible repetitions (blocked) to zero repetitions per list. A subject-paced task was employed which should not only allow subjects to detect the list structure but also provide the time necessary to utilize this information. On the basis of the presentation time hypothesis, one would predict that systematically decreasing the number of category repetition s would lead to increased processing time for storage with a subsequent elimination of the effect of category repetitions on cluster size during output. In addition to manipulating the number of category repetitions, the present investigation was designed to examine retrieval plan effects under conditions somewhat different than have been previously employed. Typically, researchers who have examined the influence of a retrieval plan on recall performance have suggested a plan to subjects, and the effects have either been minimal (Wood, 1972) or absent (D'Agostino, 1969). This may, in part, be related to individual differences in preferred strategies. Mandler (1969), for example, has reported that many subjects prefer to use a serial strategy to learn categorized lists even when the structural features are salient. Merely suggesting a retrieval plan may not be sufficient to establish the use of such a plan. The present approach differed in that subjects were required to use the plan offered by the experimenter. The retrieval plan employed here simply required recall by conceptual categories. By systematically varying the number of category repetitions under subjectpaced conditions and also requiring the use of a retrieval plan, it should be possible to examine separately the influence of effective presentation time and the effects of a retrieval plan on organized recall. If effective present-

TEMPORAL ASPECTS OF STORAGEAND RETRIEVAL ation time is the critical variable, one would predict increased processing time as the n u m b e r o f category repetitions decreases, for both retrieval plan and uninstructed conditions, leading to the elimination of cluster size differences for all levels o f category repetitions. If, on the other hand, the use o f a retrieval plan is the critical factor, one would predict equal cluster size across levels o f category repetitions only for subjects who had been given retrieval plan instructions. The temporal characteristics o f retrieval have been described recently by Pollio, Richards, and Lucas (1969). They showed that response emission has temporal characteristics corresponding to the retrieval o f higher- and lower-order m e m o r y units. Pollio et al. suggested that between-category I R T s represent search for and retrieval o f higherorder representations, whereas withincategory I R T s represent retrieval o f items within these units. Subsequent research by Patterson, Meltzer, and Mandler (1971) indicated that between-category I R T s are composed o f three components, (1) the time required to exit f r o m a category the subject has just recalled, (2) the time necessary to access a subsequent unrecalled category, and (3) the t~me required to access a word in the next category. The search for higher-order units, reflected by an exponential growth in betweencategory I R T s with successive categories output, was demonstrated to be related only to category access time. W h e n the category access c o m p o n e n t was eliminated by providing subjects with retrieval cues, betweencategory I R T s remained relatively constant at successive output positions, providing suggestive evidence that both category exit time and w o r d access time were independent of output position. As part of the retrieval plan instruction employed here, subjects were asked to attempt to exhaust a category before accessing a subsequent category. This should lead to an increase in between-category I R T s which could be attributed only to an increase in category exit time.

501

The first experiment reported here was designed to test the specific predictions concerning the retrieval plan and effective presentation time hypotheses. The more general intent was to clarify the role o f storage and retrieval processes in free recall by examining the temporal characteristics o f input and output processing. EXPERIMENT I Method Subjects. Eighty undergraduates, enrolled in introductory psychology courses at the Universxty of Kansas, participated for class credit. Subjects were randomly assigned to eight groups often subjects each. Materials and lists. Four exemplars from each of 40 categories were selected from the Battig and Montague (1969) norms. For all categories, the exemplars represented the most frequent responses to the category names. Ten lists were constructed by randomly sampling, without replacement, four categories successively until the sample was exhausted. Consequently, there were ten lists of four categories with four items per category. For each of the ten lists, the items were arranged to represent four levels of category repetitions. Category repetitions were defined as the number of times a word was followed by another word from the same category. The four levels were 0, 8, and 12 repetitions, plus a constrained random condition (R). By designating the four categories per list by the letters A through D, the following represents an example of the construction of 12, 8, and 0 repetitions, respectively: AAAABBBBCCCCDDDD, AABBCCDDAABBCCDD, ABCDABCDABCDABCD. The R condition was included to evaluate the effects of a more commonly employed presentation procedure. In the R condition, repetitions were free to vary with the exception that no more than two instances of a category could occupy adjacent positions. Over the ten lists, the number of repetitions per list m R ranged from zero to three, with a mean of two. Apparatus. A Sawyer Rotomatic slide projector was used to present the stimuli on a daylight projector screen. Exposure durations were measured by means of a response button which was programmed to the projector such that the subject controlled the viewing time of each slide. The latency between the button response and the visual presentation of the stimuli was approximately .4 ~: .05 sec. The duration of each stimulus presentation was automatically recorded and printed out by means of a pulse stream generator, parallel-entry control panel, and a Sodeco printout counter. Exposure durations were recorded to the

502

KELLAS E T AL.

nearest .05 sec. The stlmuh were typed onto transparency paper and fitted in Easymount 35-ram frames. Since all recall was oral, a Sony (model TC 252) tape recorder was employed to record each subject's recall Interword response times (IRTs) were measured by feeding subjects' taped recall into a single channel MFE (Mechanics for Electronics, Inc., model 20CAHA) heat-writing graphic recorder. Subjects' recall actuated a pen which indicated word onset and offset on paper moving at a rate of 25 mm/sec. All IRTs were calculated to include only the time between words. Word production ttme was excluded from this measure (cf. Polho, Rlchards & Lucas, 1969). Design and procedure. Subjects were assigned randomly to the ceils of a 2 × 4 factorial matrix representing two levels of Instruction and four levels of Repetition among category instances (12, 8, 0, and R) Ten subjects served in each unique combination of the treatment conditions. Half the subjects at each repetition level were given a retrieval plan which was required for recall This condition will be referred to as clustering instruction (CI) The subjects were told that they would see multiple lists, each of which would include four conceptual categories each (example provided). In addition, subjects were told that after presentation of the final item in each list they must recall the items according to category membershtp. They were further instructed that they could recall the categories and appropriate exemplars m any order they wished, but were to attempt to exhaust one category prior to recalling another. The remaining subjects were not given clustering instructions (NCI). Here, subjects were provided with typical free recall instructions which were void of information concerning hst structure. Finally, all subjects were given instructions concerning subject-paced presentation and told that a red field would serve as a signal for recall As soon as subjects either indicated they had finished recall or after a 30-sec interval of sdence, they were told to continue to the next hst.

Results and Discussion Temporal characteristics of storage. A 2 × 4 × 16 mixed analysis o f variance was performed on the mean exposure duration for each serial position averaged across the ten lists. The main effects corresponded to Type o f Instruction (CI vs NCI), Repetitions (12, 8, 0, and R), and Serial Position (1-16). In general, exposure durations increased across successive serial positions, F(15, 1080) = 19.92, p < . 0 0 1 . However, this result is

qualified by a higher-order interaction. As category instances became more disparate within the lists, exposure durations became substantially greater with the differences a m o n g 12, 8, 0, and R repetitions increasing across serial positions, F(45, 1080)= 1.52, p < .05. N o reliable difference was found between 0 and R conditions. The average results for the various treatments are presented in Figure 1. The patterning o f exposure durations for the 12-repetition condition (dotted curves) is o f particular interest. Subjects clearly exposed the last exemplar within successive categories longer than earlier exemplars. This patterning is in accord with the distributions of overt rehearsal reported by R u n d u s (1971, Experiment IV). Apparently, subjects pause longer on the terminal category items in order to rehearse previously presented 1terns according to category membership. The exposure measure, then, reflects a joint function of rehearsal set size and number of rehearsals. The results are not as obvious for conditions representing less than complete contiguity. The marked increase in processing time across serial positions m a y indicate the increment required to reorganize the list, as presented, according to semantic categories. However, this conclusion must be restricted to subjects not given a retrieval plan. Clustering Instructions led to a substantial decrement m exposure durations across serial positions relative to N C I , F(15,1080) = 3.39, p < .001. The number o f repetitions a m o n g category instances had no reliable effect on processing time when subjects had been provided with a retrieval plan. Nontemporal performance. A n initial analysis o f variance was conducted on the total correct responses per successive list in order to evaluate any possible list interactions with treatment conditions. The only significant result to emerge was a reliable main effect for List, F (9, 6 4 8 ) = 5.76, p < .001. Overall, performance improved slightly with successive lists. Accuracy level (out o f 16 possible items) for CI was 14.0, 13.3, 13.8, and

503

TEMPORAL ASPECTS OF STORAGE AND RETRIEVAL

NCI

CI 12.0

REPETITIONS • . . . . . . °,,*• 12

110 •

• 0

,.. ......

.,,

R

u•IO0 _

Z

9.0_

v

i7,./.

I.-~"~ :D

7.0~

/'h

6.0

.J,

u,~ 5.o

oa_ w

.

.-"

40

z "~ 3.o

s"

.,',.

Z 8o O

II. ~ll/ ,.~

~

,.-._. •

g

L'.. {...' ",:v'7""-,,,.-. ,:..

,

.~;..,

v

".....,

..

!' •"'-~. ."

%° ..."

/

"

/

~ I

7...

."

~

°

".ll

.,:",!

"... j

..;.I/.•

:<..

.."

i

-:" "... -

/ •

~"

.4

.

.." °

v .~

'is,..

20 1- I I I 2 o

I 4

I

I 6

I

I 8

r

I 10

I

I 12

I

I

I

14

I

l I I l I I I I I l I I I I I I

16

PRESENTATION

2

4

6

8

10

12

14

16

POSITION

FIG. 1. Mean exposure duration at each presentatxon position as a function of type of instruction and category repetitions.

12.8 (12, 8, 0, and R, respectively). For NCI the corresponding levels of performance were 12.4, 12.8, 12.8, and 14.2 correct responses. Extralist intrusions accounted for 1.1 ~ of the emitted responses and were unrelated to experimental conditions. Interestingly, of all intrusions representing instances of presented categories, 54 ~ occurred as the last item recalled within a category, suggesting a generative process following a breakdown in recall performance. A high level of clustering was observed in recall for all treatments. For present purposes, the data of interest were the mean size of the clusters according to their output positions. Only first entries into the successive categories were considered. Category reentries after the initial output positions accounted for 3.5 ~ of the words recalled. The average cluster size for all conditions is presented in the upper portion of Table 1. It should be noted here that the 12- and O-repetition conditions represent the

appropriate comparison groups for the research of D'Agostino (1969). Across all treatments the average cluster size was 3.26 items, which compares closely with the value reported by Polho et al. (1969). Subjects provided with a retrieval plan yielded significantly larger clusters than NCI groups, F(1, 72) = 2.81,p < .05. Output position also affected cluster size with early clusters being larger than later clusters, F(3, 216)= 85.78, p<.001. This finding supports the data reported by Smith, D'Agostino, and Reid (1970) and may be related to output interference effects on lower-order memory units. The interaction of Type of Instruction and Repetition was reliable also, F(3, 72) = 3.30, p < .05. While cluster size was independent of the level of repetitions for CI groups, a reliable decrement in cluster size across 12, 8, and 0 repetitions was obtained for NCI, followed by a sharp and unexpected increase for R-repetition, NCI.

KELLAS ET AL.

504

TABLE 1 MEAN CLUSTER SIZE FOR A L L TREATMENT CONDITIONS

Category output position Instruction condition

Repetition con&tion

1

2

3

4

Experiment I Clustering Instructions

12 8 0 R

3.87 3.81 3.69 3.52

3.64 3.60 3.62 3.35

3.71 3.52 3.63 3.21

2.80 2.47 2.84 2.62

3.50 3.35 3,44 3.18

12 8 0 R

3.83 3.44 2.90 3.63

3.56 3.28 2.76 3.53

3.42 3.19 2.75 3.42

2.50 2.44 2.40 3.16

3.33 3.09 2.70 3.44

No clustering Instructions Experiment II

12

3.83

3.74

3.46

2.67

3.42

List Information

8 0 R

3.40 3.14 3.44

3.33 3.16 3.38

3.22 3.16 3.41

2 06 2 48 2.46

3.00 2.98 3.17

12 8 0 R

3.81 3.42 3.00 3.59

3.73 3.25 2.95 3.49

3.73 3.16 2.82 3.57

3.11 2.52 2.61 2 77

3.60 3.09 2.84 3.36

N o list

Information

Temporal characteristics of retrieval. The temporal analysis of the retrieval time data was restricted also to the first entrance to each category. Due to experimenter error, data from 4 of the 80 subjects precluded temporal analysis. Three subjects were from the random presentation condition, NCI, and one from 0-NCI. Consequently, separate analyses were performed on the CI and N C I groups. Figure 2 shows the mean between- and within-category IRTs for the first four output positions, separately for the various conditions. N o between-category I R T was computed for the first category recalled since there is no acceptable procedure for measuring the latency for the first response. Between-category I R T s represent the average time to recall the first item from a category. Within-category IRTs represent the average I R T for the remaining words recalled at each of the four successive category output positions. Both measures were averaged across ten successive lists. The data depicted in Figure 2 represent means

based on 7849 observations, or 76 700 of the total possible observations. The results of both analyses were identical. Between-category IRTs were reliably longer than within-category IRTs, F(1, 36) = 149.17 and F(1, 3 2 ) = 90.29, p < .001 for CI and NCI, respectively. In addition, the magnitude of the difference between the types of IRTs systematically increased with output position, F(2, 72)=26.39 and F(2, 6 4 ) = 16.24, p < .001, again, respectively, for CI and NCI. The interaction resulted from the marked increase in between-category IRTs across successive positions. Only a slight increase occurred for within-category IRTs. Repetitions had no reliable effect on IRTs for either analysis. It was hypothesized that requiring a retrieval plan which encouraged subjects to exhaust a category during recall would increase category exit time. There is no reason to expect the requirement to affect either category access or word access times. Further, Patterson,

TEMPORAL ASPECTS OF STORAGE AND RETRIEVAL

Z0

NCI

CI

7.5

REPETITIONS e.......o 12

i

e--

65 6.0

m

5.5

m

I---

-.-.e

8

_- 0

• . . . . .



R

:.. ""... / P

///

4.5

I.u gO

--

_--

//

5.0

U

505

-

4.0 3.5

m

30

m

2.5

:'1,, t :" /,.," .',~.

:

/e

;/

/

./

/

./

i

,~l /

/

1.5_

/

/

:://+

i¢ ° s . /

2.0

7"--

: : :

.

Z

1.0_ .5

~e

~,~:-~_

_

o WITHIN BETWEEN

i 1

1

2

3

. . . .

I I 1

2

I I 2

3

I [ 3

4

TRANSITIONS

FIG. 2. M e a n within- a n d between-category I R T s across successwe o u t p u t p o s m o n s as a func t i on of type of m s t r u c t m n a n d c a t e g o r y repetitions.

Meltzer, and Mandler (1971) have suggested that category exit time should be independent of output position. A test of the hypothesis was made by directly comparing the betweencategory IRTs across output positions for CI and NCI groups. The hypothesis would predict a main effect for Instruction with no interaction with output position. These results were obtained. The requirement to exhaust a category significantly increased betweencategory IRTs relative to NCI, F(1,68) = 5.99, p < .05; the effect was approximately equal in magnitude at each output position. The preceding results have dealt with the differences between two types of IRTs representing retrieval of higher-order memory units and combined lower-order units, independent of output positions within categories. We will now examine the nature of intracategory retrieval and the changes in retrieval performance in relation to category output position. The average results are presented in Figure 3

for CI and NCI separately. The mean number of within-category IRTs is restricted by cluster size and, in general, the number of observations upon which the data points are based decrease as a function of positions within categories as well as category output position. The number of observations for each data point ranged from a high of 342 to a low of 134, which is substantially greater than in previous reports. Over all treatment conditions, within-category IRTs systematically increased with successive positions within categories, F(2,108) = 17.70, p <.001. The magnitude of the increment was greater for categories recalled later in the output sequence. This description is substantiated by a reliable interaction of position within category by category output position, F(6,324)=4.70, p < .001. Neither Instruction nor Repetitions systematically affected within-category IRTs. The similarity of the between-category performance in Figure 2 and the within-

506

KELLAS ET AL.

1.6 _ 1.4

Cl. Nc, ....

:

/

_

i

1.2

i_

,.-7. 1 . 0 _ U LLI 0.8 _ I--0.6 _

/

0.4_

/ /

0.2_ 0.0

f

I

I

I

I

I

~

I

I

I

I-2

2-3

3-4

I-2

2-3

3-4

I-2

2-3

3-4

I-2

L

I

2-3 3-4

II III IV I POSITIONS WITHIN CATEGORIES FOR THE FIRST 4 CATEGORIES

FIG 3. Mean IRTs across successwe output positions within categories as a function of type of instruction. category patterning o f Figure 3 suggests a commonality in the search and retrieval processes for higher- and lower-order units. This position was advanced by W o o d (1972), and the present data are in substantial agreement. EXPERIMENT II

The purposes o f Experiment II were to partially replicate the results of the first experiment and to evaluate the effects o f providing subjects with general list information. Apparently it has been assumed by some investigators (for example, D'Agostino, 1969; Slamecka, 1972; W o o d , 1972)that providing subjects with general list information is a sufficient condition for establishing a retrieval plan based on semantic relationshaps a m o n g items. Since it is not possible to require the use of a specific plan without simultaneously conveying information concerning list structure, the effects of the plan on storage time and organization in recall could be attributed to either variable, that is, the recall requirement per se, or simply facilitation arising f r o m knowledge of list structure. The present experiment examined this possibility. Metkod Subjects. E:ghty subjects enrolled in introductory psychology courses at the Umvers:ty of Kansas participated to fulfill a course requirement.

Destgn and procedure. Subjects were assigned randomly to the cells of a 2 x 4 factorial matrix representing two levels of Instruction and four levels of Repetition. Ten subjects served in each unique combination of the treatment condmons. Half of the subjects at each repet~non level were provided with general hst information (LI). The subjects were told that they would see multiple hsts. In addmon they were told that the items in each hst consisted of four conceptual categories. An example was provided with items not subsequently employed. The remaining subjects were not given hst information (NLI). Both LI and NLI groups were provided with standard free recall instructions. It should be noted that the group designated NLI is identical to NCI of Experiment I. In all other respects the procedure for Experiment II was the same as that of the previous investigation. Results and Discussion Temporal characteristics. The average exposure durations for the various treatments are graphically presented in Figure 4. Two important aspects o f the data are apparent. First, the effect o f category repetitions replicates fairly closely the results for N C I o f Experiment I. As the n u m b e r of repetitions decreased, increasingly greater study time was utilized across serial positions, F(45, 1080)= 3.32, p <.001. Second, in contrast to clustenng instructions, L I failed to influence average input processing time relative to N L I , F(1, 72) = 2.98, p > .05. This finding strongly suggests that list information per se does not provide a

507

TEMPORAL ASPECTS OF STORAGE AND RETRIEVAL

NLI

LI 120

REPETITIONS • . . . . . . . . . . • 12 o-.-- -- ~ 8 e 0 • . ....... ~ R

11.0



,.~

'~- I00 13 ku 1.0

J

'"..,.,: t

z9o

V

/

7 8 0 I

o ~7o

.I

~'x / / - 4 - -

"o

)

f

.s"

tll'\ /" I/ ~,I"

;

g '

3 o

".

\/t

.Y'--

"-'~,,:.'~

:

%:

"

"

.....

./!

0

x

-....

~...e

-..... .... m

2.0_

T

\//

-:::g-, i

o 40

/',..

I

I 2

I

I 4

I

I 6

I

I

I

8

I0

I

I 12

I

I

t

14

PRESENTATION

I 16

I

I I I I I I I I 2

4

6

8

I 10

I

I 12

I

I 14

I

.1 16

POSITION

FIG. 4. Mean exposure duration at each presentation posmon as a funcnon of type of instruction and category repetitions. sufficient condition for establishing a retrieval plan comparable to that employed by the CI groups in the first experiment. As before, the data representing completely blocked presentation clearly indicate subjects' use of the categorical structure of the lists for storage. In general, the number of pronounced peaks in the temporal profiles represent N-1 categories. The temporal depression for the final category corresponds with subjects' observed tendency to output the terminal category first followed by recall of the remaining categories. Nontemporal performance. An analysis performed on correct responses across successive lists indicated that performance improved slightly with practice, F(9, 648)= 5.40, p < .001. However, no reliable interactions including List and treatments were evident. Mean correct responding for LI was 13.7, 12.7, 13.5, and 13.4 for 12, 8, 0, and R repetitions, respectively. The corresponding leve!s of performance for NLI were 14.5, 13.3, 13.3,

and 13.8 correct responses. Extralist intrusions accounted for 1.4~ of the emitted items. Similar to the previous experiment, the majority &intrusions represented instances of categories actually presented. Of M1 such intrusions committed, 6 6 ~ occurred as the last response within a category. A final analysis was performed on average cluster size across output positions. Category reentries accounted for 6 ~ of the total words recalled, a slight increase over Experiment I. The average cluster size for the various treatments is presented in the lower portion of Table 1. The mean cluster size for all treatments was 3.20 items which is similar to the value of Experiment I (3.26). The analysis revealed that cluster size was directly related to the number of repetitions in the lists, that is, as instances were more contiguous, larger clusters were formed, F(3, 72) = 3.20, p < .05. In agreement with the earlier results, cluster size decreased across successive output positions, F(3,216) = 68.13, p < .001.

508

KELLA8ET AL.

List information had no facilitating effect on cluster size ( F < 1). Indeed, LI led to a reliable decrease, relative to NLI, in cluster size for the final output position,/7(3,216) = 3.12,p < .025. It seems reasonable to conclude that prowding information concerning list structure does not affect either input processmg time or the size of higher-order units. The only effect of list information, which is evident in Figure 4, was to decrease variability among subjects during storage. Since the major purpose of Experiment II was to evaluate the effects of general list information on storage time and recall organization, only a preliminary examination of IRTs was conducted. These data yielded results which were redundant wlth the uninstructed conditions of Experiment I, and consequently are not reported here. GENERAL DISCUSSION

The hypothesis concerning effective presentation time was only partially supported by the present results. Processing time for storage did systematically increase as conceptually related items occupied less contiguous positions during presentation. However, this effect was restricted to those conditions in which subjects received either no information concerning the list structure or only general list information. In addition, the temporal increment did not eliminate differences in cluster size as expected. Contrary to the results of D'Agostino (1969), a retrieval plan did eliminate repetition effects on cluster size as well as differences in processing time. It is quite clear that a retrieval plan has pronounced effects on storage of list items. As suggested by Mandler (1967), for organization to facilitate storage the list structure must not only be detected but utilized as well. Under experimenter-paced conditions where presentation rate is relatively fast, subjects may detect the nature of list organization but may have insufficient time to use this information for storing related items. The results of Experi-

ment II further indicate that general list information per se does not provide a sufficient condition for establishing a retrieval plan. No performance benefit accrued from simply providing general information. The mechanisms by which a retrieval plan facihtates storage are not made clear in the present research. However the results, in combination with previous research, suggest a testable hypothesis. First, it is assumed that a retrieval plan facilitates the unitization process. This assumption is supported by the finding that, when a retrieval plan was required, cluster size was independent of the number of category repetitions. The hypothesis states that successive list items, during presentation, serve the function of retrieval cues for earher presented ~tems, when a retrieval plan which requires immediate identification of category membership is enforced. To the extent that early list items have been encoded in specific relation to the appropriate superordinate label, the conditions are present via a mediaUon process for unitization of related items to occur. Hudson and Austin (1970) have demonstrated that category exemplars and superordinate labels are equally effective as retrieval cues. Recently, Lewis (1972) has provided evidence that unpresented category exemplars can serve the same function. Together, these consideratmns strongly support the retrieval cue value of list items during list presentation. If true, it may not be meaningful to investigate storage and retrieval processes as being coincidental with list presentation and recall phases of a task, respectively. This explanation of the unitization function of a retrieval cue can be extended to account for performance under various levels of repetition in the absence of a retrieval plan as well. According to the encoding specificity hypothesis (Thompson & Tulving, 1970), the relationship between a cue and a list item must be established during list presentation if it is to be effective. When subjects have not been given a specific retrieval plan, identification of higher-order units is most likely dependent

TEMPORAL ASPECTS OF STORAGE AND RETRIEVAL

509

upon noticing either attribute overlap or Meltzer, & Mandler, 1971) is the function semantic similarity among items. The rapidity describing between-category IRTs across with which the units are identified, then, will category output positions. Whereas the be a function of the contiguity of related items. earlier results showed an exponential growth Repetitions, in the present research, serve as function, the present data are best described one signal for the inherent organization of the by a linear function. The discrepancy may lists. However, as repetitions decrease, sub- simply reflect the fact that there were insuffijects must necessarily progress further into cient data points here to establish exponentialthe list to identify semantic categories. As a lty. The importance of the discrepancy is that result, the early list items will not have been the exponential growth function has been encoded in specific relation to a consistent assumed to reflect an underlying probabilistic retrieval cue, thereby omitting them from the search and retrieval strategy for higher-order unitization process. This would lead to memory units, which is based on random smaller cluster size as items become less samphng, with replacement. The units thus sampled are then subjected to a reentry check contiguous, as was demonstrated. Patterson, Meltzer, and Mandler (1971) to determine whether or not they had been have suggested that between-category IRTs previously recalled. An apparent relationship represent a combination of three components: exists m recent research which suggests that category exit time, category access time, and a model based solely on random sampling word access time. Evidence was provided that with replacement may not be sufficient to category access time could be eliminated by describe retrieval processes. The relationship giving subjects retrieval cues during recall. emerges from the patterning of IRTs, rehearThe effect of the retrieval cues was to omit the sals, and recall probabihties. Wood and higher-order search component of the re- Underwood (1967) have shown that the initial trieval process. The function describing items in successive categories have higher between-category IRTs across successive out- recall probabilities than terminal items. This put positions, under these conditions, became effect was independent of overall serial flat while remaining substantially greater in position and held for both blocked and magnitude than within-category IRTs. It was random presentation orders. Rundus' (1971, speculated that the remaining time difference Experiment IV) analysis of overt rehearsal between inter- and intracategory retrieval demonstrated that overall, initial category represented a composite of category exit and items are rehearsed more than terminal items word access times. Suggestive evidence was and that the terminal items in a early category presented indicating that providing subjects are rehearsed less than initial items in an with retrieval cues also served to increase immediately subsequent category. This patcategory exit time, presumably because this terning corresponds directly with the IRT information increased the likelihood of an patterns in the present research, and the recall extended, unsuccessful search for additional probability data of Wood and Underwood. items. The results of the present research fully If one assumes a search and retrieval model confirm the psychological reality of a category based on memory trace strength, the IRT exit component. The retrieval plan require- results would be predictable. Higher-order ment led to a reliable increase in between- units as well as their elements would be stored category IRTs which was independent of out- in a fixed sequence and subsequently retrieved put position, as predicted by Patterson et al. according to trace strength. This would be the The major discrepancy between the data case for retrieval of higher-order units as well reported here and previous research (that is, as retrieval of items within these units. With Pollio, Richards, & Lucas, 1969; Patterson, the inclusion of a recognition component, a

510

KELLAS ET AL.

strength model could account for both linear and exponential functions. I f subjects have been allowed sufficient time to process list items for storage, recall should increasingly represent organized output from a long-term store. W h e n items are retrieved in a fixed sequence determined by systematic patternings o f rehearsal, I R T s could be expected to produce a linear increasing function. This function would represent the decreasing trace strength o f units recalled later in the output sequence, which presumably have been rehearsed less than earlier units (cf. Rundus, 1971). The similarity o f the I R T patterns for both within and between categories in the present data suggest that this may, indeed, be an accurate description o f output processing for both higher- and lower-order units. However, when the strength o f a unit reaches some intermediate value which is below a subjective criterion for immediate output of the accessed unit, subjects would be more likely to subject the unit to a recognition check for list membership and the possibility o f prior recall. Such intermediate strength values could be expected when the output sequence becomes too long to be represented in a fixed sequence or when subjects have had insufficient time to organize items during presentation. The n u m b e r of reentry or reaccess checks could increase exponentially with the n u m b e r o f prior items recalled. A n ancillary expectation o f the strength model would be that successively output higher-order units would be less cohesive and c h u n k size should therefore systematically decrease across output positions. This was demonstrated here as well as by Smith, D'Agostino, and Reid (1970). While this analysis is post hoc and highly speculative, current research indicates that a modified strength model should be seriously considered. REFERENCES

BATTIG,W. F. & MONTAGUE,W. E. Category norms for verbal items in 56 categories: A replication

and extension of the Connecticut category norms Journal of Experimental Psychology Monograph, 1969, 80, No. 3, Part 2 COFER,C N., BRUCE,D. R. & REICHER,G. M Clustermg in free recall as a function of certain methodological variations Journal of Experimental Psychology, 1966, 71, 858-866. D'AGOSTINO,P. R The blocked-random effect in recall and recognition Journal of Verbal Learning and Verbal Behavior, 1969, 8, 815-820. HUDSON,R. L , & AUSTIN,J. B. Effect of context and category name on the recall of categorized word lists Journal of Experimental Psychology, 1970, 86, 43-47 KELLAS,G., & BtJTTERFIELD,E. C. Effect of response requirement and type of material on acqulsmon and retention performace in short-term memory Journal of Experimental Psychology, 1971, 88, 5O-56.

LEWIS, M. Q. Cue effectiveness in cued recall. Paper presented at the annual meeting of The Psychonomic Society, St Louis, November, 1972 MANDLER, G. Organization and memory. In K. W. Spence, J. A. Spence (Eds), The psychology of learning and motivation, Vol. 1. New York: Academic Press, 1967. MANDLER,G. Input variables and output strategies in free recall of categorized lists. American Journal of Psychology, 1969, 82, 531-539. MANDLER, G Organlzatmn and recognition. In E. Tulvlng, W Donaldson (Eds.), Organizatton of memory. New York: Academic Press, 1972. MURDOCK,B. B., Jr. & OKAD~, R Interresponse times in single-trial free recall. Journal of Expertmental Psychology, 1970, 86, 263-267. PATTERSON,K E, MELTZER,R. H., & MANDLER,G. Inter-response times in categorized free recall. Journal of Verbal Learning and Verbal Behavior, 1971, 10, 417-426. POLLIO,H. R., RICHARDS,S., • LUCAS,R. Temporal properties of category recall. Journal of Verbal Learning and Verbal Behavior, 1969, 8, 529-536. POSTMAN,L. A pragmatic view of organization theory. In E. Tulving, W. Donaldson (Eds), Organization of memory. New York: Academic Press, 1972.

PUFF, C. R. Clustering as a function of the sequential organization of stimulus word lists. Journal of Verbal Learning and Verbal Behavior, 1966, 5, 503-506. RUNDUS, D. Analysis of rehearsal processes in free recall. Journal of Experimental Psychology, 1971, 89, 63-77. SEAMECKA,N. J. An examination of trace storage m free recall. Journal of Experimental Psychology, 1968, 76, 504-513.

TEMPORAL ASPECTS OF STORAGEAND RETRIEVAL

511

SLAMECKA,N. J. Testing for assocmtlve storage m TULVING,E. Theoretical issues in free recall. In T. R. Dixon, D. L. Horton (Eds.), Verbal behavior multitrial free recall. Journal of Experimental and general behavior theory. Englewood Cliffs, Psychology, 1969, 81, 557-560. New Jersey: Prentice-Jail, 1968. SLAMECKA,N. J. The question of associative growth in the learning of categorized material. Journal of WOOD, G. Organizational processes and free recall. In E. Tulvmg, W. Donaldson (Eds), OrganVerbal Learning and Verbal Behavior, 1972, 11, izatlon of memory. New York: Academic Press, 324-332. SMITH,A. D., D'AGOSTINO,P. R., & REID,L. S. Output 1972. interference m long-term memory. Canadian WooD, G., t~ UNDERWOOD,B. J. Implicit responses and conceptual similarity. Journal of Verbal Learning Journal of Psychology, 1970, 23, 85-89. and VerbalBehavior, 1967, 6, 1-10. THOMPSON,D. M., & TULVING,E. Associative encoding and retrieval: Weak and strong cues. Journal of (RecelvedFebruary 12,1973) Experimental Psychology, 1970, 86, 255-262