A framework for reproductive memory tasks

Scandinavian Journal of Psychology, 1983,24, 45-56

A framework for reproductive memory tasks JERKER RONNBERG

University of Urned, Sweden

Ronnberg, J.: A framework for reproductive memory tasks. Scandinavian Journal of Psychology, 1983.24. 45-56. The present paper presents a framework for reproductive memory tasks. It is proposed that the notion of predictability of presentation is of prime importance to performance in reproductive tasks and that this notion may be decomposed into three critical dimensions, viz. the item expectancy, the type-ofrecall and recall expectancy dimensions. The first dimension refers to the manipulation of inter-item intervals; the second to backward, forward or free recall and the third to when recall is prompted. Moreover, a study was conducted, wherein these three dimensions were combined. The main results demonstrated superior performance for prerecency items (the first four items) with fast inter-item intervals in the beginning of list presentation and also reliable effects for recency items (the last four items). Precueing and postcueing (recall expectancy) interacted with the item expectancy dimension and also with the type of recall expectancy dimension in systematic ways. The data obtained could not be accounted for a traditional encoding, storage, retrieval or rehearsal terms, whereas the predictability notion seemed more viable. Finally, implications for a general interactionistic theory were also discussed, where the interplay between cognitive skills and predictability characteristics of the task constitute the basic factors. J . Ronnberg, Department of Psychology, University of Umed, Rddhusesplanaden 2,

S-90247 Umed, Sweden.

The present paper attempts to introduce a framework for analysis of stimulus situations commonly used in memory tasks. The first part of the paper consists of a theoretical analysis of the concept of predictability and the second part of an empirical section and the third part consists of a general discussion based on the proposed predictability notion. A predictability analysis The aim of this paper is to study the potential importance of the concept of predictability and its application to what has been denoted reproductive memory tasks (see Ronnberg, 1980a, b, 1981, 1982), see also Garling, 1981). A reproductive memory task is the traditional memory task where it is demanded of the subject to reproduce (e. g. by means of recognition or recall) as accurately as possible the items on a preceding memory list. As argued by Ronnberg (1980a), the notion of predictability constitutes a potent tool for analyzing

and understanding the subjects’ behavior in a reproductive memory task. It has for instance been demonstrated (Ronnberg, 1980a, 1981) that predictability of presentation, as manipulated by temporal variations between items in a list of items, provoke quite different patterns of recall compared to the standard method of presenting items with constant inter-item intervals. The data obtained were also found t o be at variance with traditional models for the serial position curve. Furthermore, it is here important to note that this type of manipulation concerns the aspect of the concept of predictability which can be denoted predictability of presentation as contrasted to predictability of structure. Predictability of structure refers to the structure and coherence of stories (e.g. Mandler, 1978), predictability of content of the stimulus material and the like. Given these distinctions, we are now in the position to elaborate on the particular aspects of predictability denoted predictability of presentation, Scandinauian Journal of Psychology, 24

46

J . Ronnberg

Table 1 . The time (sec) elapsed for each of eight presentation conditions, since the start of list presentation, when an item in a given serial position is about to be presented Serial position

Condition 1

2

3

0 0 0 0

3.0 4.5 5.0 1.o

6.0 7.0 9.5 2.5

0 0

D I

0

1.3 1.9 2.1 0.5

2.6

R

4

5

6

7

8

9

10

9.0 9.0 13.5 4.5

12.0 12.0 17.0 7.0

15.0 13.5 20.0 10.0

18.0 14.5 22.5 13.5

21.0 19.5 24.5 17.5

24.0 23.5 26.0 22.0

27.0 27.0 27.0 27.0

3.9 3.9 5.7 2.1

5.2 5.2 7.2 3.2

6.5 5.9 8.5 4.5

7.8 6.4 9.6 6.0

9.1 8.5 10.5 7.7

10.4 10.2 11.2 9.6

11.7 11.7 11.7 11.7

Slow

C R D I Fast

C

0

3.0 4.0 1.2

Predictability of presentation Based on the previous studies (Ronnberg, 1980a, b , 1981), it is here postulated that three dimensions constitute basic ingredients for how a subject behaves with respect to a reproductive task. First, the timing of stimuli is of fundamental importance to the recall of a list of to-be-remembered items (TBR-items). Ronnberg (1980a, 1981) has demonstrated that recall probability covaries with this type of manipulation. The types of conditions used in those studies were linearly increasing or decreasing inter-item intervals, random and constant inter-item intervals. It was generally found that condition D (decreasing) lists are superior to Condition I (increasing) lists for recency items when recall signals (indicating when the subject can start recalling the list of items) follow immediately after list termination. However, when the recall signal occurs at the expected point in time, according to the function governing the inter-item intervals (i.e. not necessarily immediately after list termination), Condition I has been found to be superior to Condition D lists. Condition R (random) and C (constant) have never been found to be superior to Conditions I and D when those conditions were most predictable, but have ranked as second or third condition in a quite orderly manner. For a presentation of the conditions, see Table 1. The fact that the list presentation conditions interact with the timing of the recall signal, implicates that a second dimension operates on the subject’s ability to recall the TBR items. This second dimension is here denoted recall expectancy, while the first dimension is denoted item expectancy. Scandinavian Journal of Psychology, 24

The third dimension, which also has been shown to be critical to recall is here denoted type o f r e -

call expectancy. Ronnberg (1981) demonstrated a larger superiority for condition D compared to Condition I when a backward recall strategy was used than was obtained in the previous experiments. With a fonvard recall strategy Condition D was also superior. The main impact of the above cited studies has been that they falsify traditional encoding (Craik & Lockhart, 1972), storage (Atkinson & Shiffrin, 1968), rehearsal (Rundus, 1971) and retrieval (Tulving, 1968) models in three ways. First, these models d o not predict that presentation rate should affect the recall of recency items (cf. Metcalfe & Murdock, 1978), which has been demonstrated for e.g. Conditions D and I in all studies. Second, they do not predict that Condition D, which actually has a much faster presentation rate for recency items (see Table 1 ) than does the traditionally employed Condition C, should facilitate recall to a larger extent. Third, these models do not predict interactions with the type of recall dimension as was obtained in Ronnberg (1981). Hence, we have a three-dimensional space which is a hypothetical representation of the predictability by presentation concept. More specifically, we have a space where high and low expectancies for recall and type of recall interact in systematic ways with the item expectancy dimension. Moreover, it is assumed that the parameters operating within this stimulus space have certain definite consequences for the interaction with the cognitive sytem. First, it is postulated that the item

Reproductive memory tasks

expectancy dimension in basic to the cognitive system in that it constitutes a necessary condition for successful retrieval, i.e. the function of high item expectancy is to structure input (which of course also is true for predictability of structure). The structuring process is practically unlimited as long as dual attention tasks are not imposed on the subject and as long as item expectancy is high. A high item expectancy also leads to more appropriate encoding and storage of items. However, the real bottleneck in this system will be contingent on the application of retrieval operations, which in turn depend on the recall expectancy and type of recall expectancy parameters. Without appropriate knowledge of when retrieval can start and in what type of output format, retrieval operations will loose in efficiency and have a deleterious effect on recall. Output order is an essential factor when dealing with for instance modality of presentation (e.g. Nilsson, Wright & Murdock, 1975, 1979) and also when dealing with skilled performance (see Ronnberg, dhngren & Nilsson, 1981, 1982). Oral versus written recall also interact with type of skill, as represented by blind and hearing handicapped subjects (Ronnberg & Nilsson, 1982a, b ) . Still, if optimal conditions for retrieval are at hand, the allocation of resources for retrieval must be considered limited. As will be discussed in further detail later, a person’s skills will always set a boundary to the interaction with a certain type of temporal presentation or type of material at any given point in time. However, the concern of the present study is to study the interplay between the two, given conditionsof high and low item expectancies. A comparison between a predictability view and traditional conceptions The restrictions on the cognitive system envisaged by encoding (Craik & Lockhart, 1972) and storage theorists (Atkinson & Shiffrin, 1968) are clearly at odds with the present proposal. In Craik & Lockhart’s (1972) formulation there is no room for predictability effects to occur. Their main assumption was that acquisition activity, as induced by an orienting task, clearly maps on to subsequent memory record. Later research has shown that this .1: 1 mapping is wrong. Nelson (1979) has shown that other attributes than those conceived of in the early “levels” experiments are operative, since e.g. “case” classification of printed words does not produce zero recall scores. Nelson, Reed & Mc-

47

Evoy (1977) showed that sensory and semantic interference were equal and did not interact with time for encoding. Neither did rehearsal or imaginal instructions alter the interference effects. A vast amount of research generated by Bransford and collegues (see Bransford, 1979) demonstrate that the interaction with retrieval on the one hand and “transfer-appropriatenes” on the other will determine level of recall and not level of encoding per se. Bransford et. al. also argue for the notion of skills of the learner as an important determinant interacting with the levels continuum. Neither are the restrictions put forward by Atkinson & Shiffrin (1968) and adherents compatible with the present framework. The restictions are set by the compartmentalization of information storage and therefore search processes for stored items depend on these stores. Except for the criticism of the search metaphor (e.g. Bransford et al., 1977) the present proposal does not postulate separate stores. Successful storage will depend on high item expectancy and subsequent “search processes” at retrieval will hence be due to recall expectancy and type of recall expectancy. Given that the recall expectancies are of fundamental importance, traditional retrieval theories (Tulving, 1968; Tulving, 1972; Tulving & Watkins, 1975) differ from this proposal in that they have focused on the accessibility of the memory trace per se (via cueing techniques). According to this framework it is equally important to consider how the subject is attuned to the recall demands, which will set a “base-rate” for utilization of aids such as cues. Independent evidence can readily be interpreted within this framework. With aprecueing procedure, i.e. when expectancy of recall is high, Nilsson, Wright & Murdock (1979) found that the modality effect (i.e. when auditorily presented Condition C lists provoke better recall for terminal items as opposed to visually presented items) was smaller than when recall was postcued. Precueing and postcueing for type of recall (backwardlforward) also seems to interact with Condition C stimuli (Hinrichs, 1968; Hinrichs & Grunke, 1975; Morton, 1969; Wood, & Hinrichs, 1971). Data on stimulus suffi effects can also be nicely captured by the recall expectancy dimension (Crowder, 1978; Crowder & Morton, 1969; Morton, 1970; Morton, Crowder & Prussian, 1971). The general procedure within this paradigm has been to inform Scandinavian Journal of Psychology, 24

48

J . Ronnberg

the subjects, that after the last item presented, an extra (suffix) item was also to be presented. This item could either be ignored or otherwise viewed as a recall cue. When comparing auditory and visual presentation modes, the auditory superiority is very small as compared to other reproductive paradigms investigating this modality effect. Hence, this redundant suffix serves as a recall cue and therefore increases predictability, whereas the interaction with the auditory and visual systems is masked. Pertinent to this hypothesis is the fact that Frankish (1973) has found that the auditory superiority effects are not limited to final span position if very small but consistent pauses are introduced. An other interesting recall manipulation was introduced by Shallice (1975), where a signal was presented before the fifteenth item in a list of twenty items. Data indicated that recall of the last five items increased substantially compared to a control group with no signal, while recall of prerecency items remained equal for both groups. Both groups also had received instructions designed to maximize recall of the list, but they had to start recalling the terminal items first. This finding is consistent with a predictability view and does falsify a strict capacity notion (Broadbent, 1971), i.e. for the very same portion of the list, recall is boosted with no apparent aid of recall order, instructions or the like, except for an increase of item expectancy. Although not discussed in predictability terms, list length manipulations can be considered to be an aspect of the recall expectancy dimension. The size of the modality effect has been demonstrated to be larger with unknown (unpredictable) list lengths (Nilsson, Ohlsson, Ronnberg, 1977, 1980). This means that the interaction between the auditory and visual systems is unmasked (cf. suffix effects) and the auditory systems can benefit more from its temporal character. Moreover, Watkins & Watkins (1974) have argued that the obtained negative recency effect (Craik, 1970), where terminal items in immediate free recall tend to be forgotten with the highest probability in final free recall, is counteracted when subjects do not know about list length (see also Mascarinec & Brown, 1974). When item expectancy is high, and when list length is known, a maintenance rehearsal strategy will be employed hence keeping the items in a highly accessible and superficial state, leading to negative recency. Other evidence coherent with the item expectScandinavian Journal of Psychology, 24

ancy dimension are experiments demonstrating the potency of temporal pauses (Bower & Springston, 1970). Ryan (1969) found that these temporal patterns had to be grouped regularly (in a predictable fashion), otherwise they caused recall decrements. Predictions The predictions derived for the experiment are based on the shift in situational demands that the cueing procedure imposes. In the postcueing condition the subjects does not actually know whether to backward recall or forward recall before termination of the list. This means that in the precueing condition, the subject will, to a relatively large extent, take advantage of the item expectancy condition which is most beneficial for prerecency items since this condition gives a chance to allocate resources for retrieval operations. Hence, the precueing should provoke larger differences between conditions than postcueing for prerecency items. This also implies that the precueing condition should provoke larger differences for recency items due to the advantage of knowledge of required recall strategy. In essence, we have a joint manipulation of recall expectancy (pre/postcueing) and type or recall expectancy (recall order). Skill in list learning (naive contra trained subjects) was not manipulated, since the aim of the study was to elucidate how the postulated "bottleneck" at retrieval can be manipulated. Hence, naive subjects were used throughout the study. Moreover, due to this postulated retrieval limitation, in conjunction with the extreme uncertainty from list to list (as manipulated by list length variations and item expectancy), it is expected that prerecency performance in general will be much more affected than recency performance (cf. Baddeley & Hitch, 1977) compared to previous experiments on predictability and memory (Ronnberg, 1980a, 1981). Given that the prelpost cueing manipulation is very potent in this experiment, it can be expected that the postcueing condition simulates the unpredictable presentations of recall signals in Ronnberg (1980a), where it was found that Condition D lists were superior to ConditionZ lists for terminal items. Now, since uncertainty is assumed to be extended to prerecency items it is logical to predict that Condition Z lists will produce better recall for prerecency items (since they constitute the mirror

image of Condition D lists) and Condition D lists for recency items. This effect should also be potentiated with a forward recall strategy. For recency items, this strategy represents a simple read-out strategy and with an auditory mode of presentation (see Nilsson, Wright & Murdock, 1975, 1979) a forward output order is optimal for terminal items. Precuering on the other hand, should induce a relatively more predictable situation and hence simulate conditions with predictable recall signals (cf. Ronnberg, 1980a). In those conditions, it was found that Condition I lists were superior for recency items. Hence, Condition I performance is expected to be superior and especially for a backward strategy. For prerecency items, a forward recall of Condition I lists should provoke a large superiority for Condition I lists, given that the general uncertainty between lists forces the subject to concentrate more on the prerecency items. (This also implies that more processing resources are exhausted and therefore a forward recall order should be worse for recency items compared to the postcueing condition.) The basic alternative to a predictability concept are the traditional models based on encoding, storage or retrieval operations (Murdock & Metcalfe, 1978). All these models predict a superior prerecency performance for longer inter-item intervals and no presentation rate effect for recency items. No interactions with cueing or recall order have been explicitly posited. Thus, the purpose of the present experiment was to study the three-way interaction between the three proposed predictability dimensions and also to use the obtained data to test current models for the serial position curve. METHOD

Subjects Forty students at the University of Uppsala, in the age of 20-30 years, participated in the experiment on a voluntary basis.

Materials All items were "unrelated" nouns and drawn from a Swedish word-pool (Nilsson, 1973).

Design Ten subjects were randomly assigned to one of four groups. Two of the groups received their recall instruction by means of a precueing procedure (Regroups) and the other two by means of postcueing 4-831944

(Post-groups). Further, one of the Pre- and Postgroups received Condition I lists and the other Preand Postgroups received condition D lists. Moreover, each subject received 16 experimental lists of which 8 lists were 7-item lists and 8 laitem lists. Of these 8 lists, four were slow lists (see Table I ) and four fast lists. Two of both types of lists were recalled in a forward manner and two in a backward manner. To further potentiate uncertainty about type of recall, there were actually another four lists, one fast and one slow variant for each list length, which were free recalled. These lists were randomly distributed among the 16 experimental lists and were considered dummy lists and also omitted from statistical treatment. The distribution of forward and backward recall was also random with the restriction that no more than two list types in succession were to be recalled in the same recall order. In accordance with Ronnberg ( 1 9 8 0 ~and ) as can be seen in Table 1, the inter-item intervals governing presentation of Condition I and D lists were constructed from two components. The acutal time an item was presented was always 0.5 seconds and as each new item was presented an increment/decrement was addedhbstracted to/from the inter-item interval. In the fast condition the increment/ decrement was 0.2 seconds and in the slow condition it was 0.5 seconds. It should also be noted that the fast condition was compressed so that the first and last interval did not include a 0.1 seconds increment/decrement for Condition I and D, respectively. The main focus of the present study was on laitem lists, but 7-item lists were also included in the experiment. As is evident from Table 1, there is a confounding with total-time of presentation for the 7-item lists, but previous data (Ronnberg, 1980a) suggest that this confounding is not critical for fast presentations. The introduction of 7-item lists also served the purpose of potentiating item expectancy effects. The order of presentation was random with the restriction that each type of list presentation always was followed by another type of presentation. In addition, the subjects within each of the four main groups were randomly divided into two subgroups, five subjects receiving one randomization of the lists, and the other five subjects received the reversed list order. The eight conditions corresponding to the three dimensions are denoted with the help of PrelPost; FIB and D and I. For instance, a postcued presentation of I lists that were recalled in a backward manner is denoted Post-BI. Thus, the design was a 2x2x2x list length (where the number of levels in the serial position variable refers to the actual list length subjectd to analysis) ~ 2 ~ 2 x 1 0 split-plot, design, referring to the Recall Expectancy, Item Expectancy, Type of Recall Expectancy, Serial Position, Total-time, Replication and Subjects factors, respectively. The between-subjects variables are the recall expectancy and item expectancy variables, while the type of recall, serial position, total-time and replication variables constituted within-subjects variables.

Procedure. Prior to list presentation the subjects were informed about the nature of the different item expectancy functions and Scandinavian Journal of Psychology, 24

50 1.0

.9 .8

J. Ronnberg

-

P)

5.6

-

.!.5

.

3.4

-

n

\

*

.3

-

.2

.

P

,’?

?

-

s.7 I

Precue

Pos tcue

\ \

\

.I

1

2

3

4

5

6

7

8

9

10

I

2

3

4

C

7

8

9

I0

serial position

s e r i a l Dosition Forward, decreasing

+-*Forward,

increasing

Backward, decreasing

~-~

Backward. increasing

Fig. 1. Proportion of words correctly recalled as a function of serial position, item expectancy, type of recall and recall expectancy.

how they should deal with the list when recalling in a backward or forward manner. With a forward recall order the subjects were instructed to start with the first item in the list and then proceed with the second etc.-With a backward recall order the subjects were required to begin with the last item in the list and then work backwards with the next last item etc. When the subjects were unable to remember one or more items they were to order those that they did remember as accurately as possible. Immediately after termination of each list a recall signal (knock-sound) was presented, whereafter the subjects wrote the words down in a topdown order on a page in a response booklet. Although, an immediate recall signal favoured Condition D recency items in Ronnberg (1980a), it is here assumed that prelpostcueing manipulations of recall expectancy will override that effect, especially since this is a between-groups factor. Moreover, this effect was very small for the same types of lists as in this experiment. However, in Experiment 4 in that study, where three lists lengths were used, the effect was substantially larger. The recall interval was 20 seconds and at the end of the interval the ordinal number of the next list was spoken, followed by five silent seconds before presentation of the list. List presentation was auditory and done by means of a Tandberg tape recorder. The words in the lists were spoken out normally. Moreover, the subjects received their instructions “forward” or “backward” by means of a small sign shown by the experimenter, either Scandinavian Journal of Psychology, 24

before (actually after the recall interval was terminated and after the ordinal number of the list was read to the subject) or after (actually coinciding with the recall signal) list presentation. If no sign was shown, the subjects were to free recall the words (dummy lists).

RESULTS AND DISCUSSION As c a n be observed in Fig. 1, dramatic effects have been obtained for prerecency portions of the serial position curve. The precueing condition provokes larger differences between conditions for both prerecency and recency items than the postcueing condition. The Pre-FZ condition is superior for prerecency portions in both cueing conditions. Regarding recency performance, the Pre-BZ condition seems slightly superior to the Pre-BD condition. In the postcueing condition, the Post-BD condition appears t o be marginally superior t o the BZ condition and the same relation appears t o hold true for the P o s t - F conditions. To trace the recall performance further, data w a s partitioned o n the basis of the slow and fast presentation conditions. By inspection of Fig. 2 and 3, it can be seen that

Reproductive memory tasks

51

FAST

Backward

Forward

1.0

L

1.0

-,

-

.9

.9

.8-

.8 -

.7.

Y

u

;. 6 -

.6-

u c

5 .5.

.o - 5 Y

Y

g .4 -

g hP'

.3

-

.2

-

a

.3 .2

.7-

-

I -

. I _

1 I

2

1

1

I

,

1

I

I

l

l

3

4 5 6 7 s e ria l p o s i t i o n

8

9

10

1

I

I

I

2

3

4

I

I

I

5 6 7 serial position

I

1

I

8

9

10

Precue. decreasing

+-+

Precue, i n c r e a s i n g

bo Postcue. decreasing Cr-Q

Postcue. i n c r e a s i n g

Fig. 2. Proportion of words correctly recalled as a func-

tion of a fast presentation, serial position, item expectancy, type of recall and recall expectancy. a fast presentation causes an even larger difference between the Pre-FI and Pre-FD conditions for prerecency items than does a slow presentation. The same potentiation holds true for the postcueing comparison. Regarding recency items, the Pre-BIPre-BD contrast is substantially larger for the slow condition as compared to the fast condition. In the postcueing condition, the fast condition appears to provoke a larger difference between the Post-BD and Post-BI condition than for the slow condition and the same holds true for the Post-FLkPost-FI cornparison. On the whole, a slow presentation is more susceptible to strategic variations due to the saw-tooth like pattern for the prerecency items and this is especially true for the forward recall order conditions. The data were subjected to an ANOVA. The following effects were found to be significant: The main effect of Item Expectancy, Total-time and Serial Position were statistically reliable sources of variation with F(1,36)=10.59, p