JOURNAL OF VERBAL LEARNING AND VERBAL BEHAVIOR 21, 165-- 174 (1982)

Elaborative as an Explanation of Levels of Processing

GARY L. BRADSHAW AND JOHN R. ANDERSON Carnegie-Mellon University

Three experiments were performed to study how elaboration of structures affects accuracy and response latency. The experiments introduce a methodology which can independently manipulate the amount and type of elaborations given to subjects. Using this methodology, it was shown that integrated, highly elaborated memory traces were better recalled than either small unelaborated traces or large, poorly integrated traces. The results have implications for current analyses of levels of processing phenomena, and were found to support the encoding elaboration model (J. R. Anderson & L. M. Reder, Levels of process- ing in human memory. HiUsdale, N.J.: Erlbaum, 1979).

In their original conceptualization, Craik criminability of the memory trace (Lock- and Lockhart (1972) argued that memory hart, Craik, & Jacoby, 1976). traces were a record of the processing done Craik and Tulving (1975) performed a on a stimulus. Qualitatively different traces series of experiments which they inter- were said to result from processing a preted as showing both effects of depth of stimulus in different ways. Researchers processing and degree of elaboration. Rep- characterized processing tasks in terms of licating earlier research, they found better the level, or depth, of processing, Early performance for semantically processed studies (Hyde & Jenkins, 1973; Schulman, material than for shallow, physically pro- 1970) reported meaningful, semantic pro- cessed material. This result was obtained cessing of materials led to superior perfor- even when the shallow task took longer to mance on a subsequent recall or recognition perform than the semantic task. Craik and task than performance after shallow, non- Tulving felt that differences in elaboration semantic processing. This superiority was could not account for these results. Such a explained in terms of greater durability of conclusion was natural from a levels of pro- semantic traces (Kintsch, 1975), or less in- cessing framework, which regards memory terference between semantic traces (Wick- as simply a record of perception. Since the elgren, 1973) over other types of traces. shallow analysis took longer to perform, However, this simple analysis was inade- they argued that the memory trace must quate to account for differences in recall for have been more elaborate than the semantic material processed at the same, deep level analysis. They concluded that the memory (Bobrow & Bower, 1969). Later accounts traces were qualitatively different between expanded upon the original formulation by these tasks, due to processing items at dif- proposing that processing within a level ferent levels. could differ in terms of its degree of elab- Craik and Tulving observed differences oration (Craik & Tulving, 1975), or the dis- in recall for semantically processed items, contingent upon a yes or no response in the orienting task. They also found better per- This research was supported by Grant BNS78-17463 formance when subjects had to judge words from the National Science Foundation. Requests for in the context of more elaborate sentence reprints should be sent to John R. Anderson, Depart- frames. These results they thought re- ment of Psychology, Carnegie-Mellon University, Pittsburgh, Pa. 15213. We would like to thank Lynne flected the effect of degree of elaboration at Reder for her comments on an earlier draft of this a particular level. Craik and Tulving con- manuscript. cluded that two mechanisms were at work

165 0022-5371/82/020165-10502.00/0 Copyright O 1982 by AcademicPress, Inc. All rights of reproduction in any form reserved. 166 BRADSHAW AND ANDERSON

to produce the results observed. Processing elaboration models (Anderson & Reder, tasks with different levels affected the 1979). qualitative nature of the memory trace, The first class of models (Bransford et while differences which occurred at the al., 1979; Tulving, 1979) account for re- same level of processing were due to differ- duced performance for shallow processing ences in elaboration of the memory trace. by noting the information extracted in the Recently, a major series of criticisms semantic processing test is more consistent have been leveled at the original levels of with the retrieval requirements of the sur- processing formulation. Craik and Lock- prise recall task than information extracted hart (1972) claimed that subjects only en- in the other tests. Bransford et al. refer to code attributes of a stimulus item required this theory as transfer-appropriate pro- by the orienting task. A question about the cessing while Tulving sees this as an exten- sound of a word was said to lead to a trace sion of his encoding specificity principle. which encoded that attribute, but not in- To be somewhat neutral we will refer to this formation about the orthographic features type of account of depth of processing as or meaning of that word. Numerous ex- encoding-appropriateness. Our major crit- periments have been done (e.g., Bransford, icism of this approach is that it is simply a Franks, Morris, & Stein, 1979; Eysenck, summary of empirical data, rather than a 1979; Nelson, 1979) that indicate a memory model of the processes involved. As such, trace records multiple levels of information. it is unable to predict the effects of many For instance, Nelson, Reed, and McEvoy study manipulations including the ones to (1977) showed that phonetic or semantic be reported here. interference will lower recall independent The distinctiveness models (Bransford et of the original orienting task. Nelson (1979) al., 1979; Jacoby & Craik, 1979; Nelson, argued that orienting tasks have their effect 1979; Eysenck, 1979; Wickelgren, 1973) by emphasizing the attributes that are in- rest on the assertion that semantically pro- volved by them. Subjects automatically cessed traces are more distinct from one process items at many levels, but may di- another than are traces processed with re- rect further controlled processing to elabo- spect to surface properties like phonetic rate upon a particular attribute. structure. Calling on past research from the Most current models of levels of process- interference literature, these models predict ing phenomena (Jacoby & Craik, 1979; Nel- that more distinctive traces will be better son, 1979; Eysenck, 1979) accept the gen- recalled. Anderson and Reder (1979) argued eral formulation which has been developed that there are many situations where recall here: Subjects will automatically encode for highly distinctive events can be further most attributes of stimulus materials re- facilitated by elaborative processing. For gardless of the orienting task. Manipula- instance, Bower, Kadin, and Dueck (1975) tions in the orienting tasks are effective be- demonstrated better memory for distinctive cause they direct further controlled pro- drawings when the drawings were given cessing to different levels. Differences in meaningful interpretations. The experi- current models do not center around the ments to be reported here will contain a di- initial encoding of information, but rather rect contrast between the predictions of how the encoded information leads to dif- distinctiveness and elaboration. ferences in retrieval. There are three gen- Anderson and Reder (1979) proposed an eral types of explanations: models that em- elaboration model which could explain phasize the match between study and test depth-of-processing effects. According to (Bransford et al., 1979; Tulving, 1979), dis- this theory the to-be-remembered material tinctiveness models (Jacoby & Craik, 1979; and any elaborations are encoded into a Lockhart et al., 1976; Eysenck, 1979), and propositional network. Any particular en- ENCODING ELABORATION 167 coded proposition is fragile. There is a sig- in recall for the main fact like Newton be- nificant chance that the subject will not be came emotionally unstable and insecure as able to activate that proposition at test. a child. In the caused-by and resulted-in However, if the subject generated a mem- conditions subjects were provided with two ory episode which encoded a set of multiple other highly related facts. We assume that propositions that were partially redundant these facts are like the kinds of elaborations with the to-be-remembered information, he subjects would generate themselves. The or she would have a much better chance of unrelated condition also included two addi- recalling it at the time of test. There are two tional facts, but these facts had little con- basic ways that a more elaborate encoding nection to the target central fact. Finally, in can lead to better memory. One, which we the single condition, subjects studied just call network redundancy, involves using the main fact. alternate retrieval paths created by the The elaboration model predicts best re- elaborations. The second possibility, which call in the related condition. This is because is more appropriate to sentential material this condition creates the kind of network than single word material, is what we call and inferential redundancy that is postu- inferential redundancy. This refers to the lated to promote good recall. Memory per- fact that the subject may be able to infer the formance should be worst in the unrelated material he studied from remaining elab- conditions, according to the ACT theory, orations. because these facts only cause interference A critical assumption in the theory is that to the target fact. subjects find certain types of elaborations The ACT theory contrasts both with the easier to make at study, and are better able encoding-appropriate models and the dis- to interpret certain types of elaborations at tinctiveness models in that it predicts per- test. A person is best at elaborating that formance should be better in the related type of information with which he or she conditions than in the unrelated condition. has had the most experience. This is be- As noted earlier, it is hard to derive any cause one's elaborative abilities are a func- predictions from the encoding-appropriate- tion of what one knows about that domain. ness models, but it seems particularly un- Semantic processing tends to produce bet- likely that they should predict a difference ter memory than surface processing be- between the related and unrelated condi- cause one tends to have a richer knowledge tions, because the relationship between base about the semantic properties of con- study and test is identical for the two con- cepts than the surface properties of words. ditions. The distinctiveness model seems The present paper explores the hypothe- committed to predicting that performance sis that the redundancy, or interconnected- under the two conditions will be worse than ness, of propositions is the important factor that under the single condition, since the in improving recall. Previous studies of additional facts should lower the distinc- elaboration effects (e.g., Reder, 1979; Bob- tiveness of the central fact. If anything, the row & Bower, 1969) used instructional similarity between the facts in the related manipulations to vary the elaborative pro- conditions should reduce the distinctive- cess. By allowing each subject to generate ness of the central fact in these conditions, his or her own elaborations, the experi- leading to poorest performance in these menter cannot separate redundancy effects conditions. from the number of propositions in a mem- Latency measurements were collected in ory episode. all three experiments. ACT predictions for Table 1 illustrates the material that we such measurements are not clear. We pre- had our subjects study in the various ex- dicted that the unrelated condition should periments. In all cases we were interested be worse than the single or related condi- 168 BRADSHAW AND ANDERSON

TABLE 1 EXAMPLES OF STIMULI SEEN BY SUBJECTS

Caused-by condition Newton became emotionally unstable and insecure as a Child This fact was caused by: Newton's father died when he was born Newton's mother remarried and left him with his grandfather Resulted-in condition Newton became emotionally unstable and insecure as a child This fact resulted in: Newton became irrationally paranoid when challenged by colleagues Newton had a nervous breakdown when his mother died Unrelated condition Newton became emotionally unstable and insecure as a child This fact is unrelated to: Newton was appointed Warden of the London mint Newton went to Trinity College in Cambridge Single condition Newton became emotionally unstable and insecure as a child tions, but the ordering of related and single tive reaction times between the single and items is uncertain. It should take the sub- related conditions, and so will not be dis- ject longer to utilize the network of inferen- cussed further. tial redundancies in the related condition. On the other hand, subjects in the related EXPERIMENT 1 condition can still use direct retrieval if that is faster. So, their memory judgments will Method be some mixture of slow and fast decisions, with the judgment often determined by the Subjects. The subjects were 19 Car- fastest retrieval. Without unwarranted as- negie-Mellon undergraduates who either sumptions about the relative temporal dis- received course credit or were paid $3.00/ tributions of the two types of judgments, hour for their participation. ACT cannot make a more specific predic- Apparatus. The study materials were tion. presented to each subject individually via a A second difficulty in the interpretation cathode ray tube (CRT) terminal controlled of the reaction time data is due to a con- by a PDP 11/34 computer. Cued recall was founding in the experimental design. Reac- done as a paper and pencil test. Reaction tion times for the second session were al- time trials were again displayed on the CRT ways collected after a cued recall task, and responses were recorded via the termi- which might focus additional elaboration on nal keyboard. some of the items. The reaction time results Materials. The materials for the experi- are shown in Appendix 1 for all three ex- ment were a set of seven biographical facts periments. In general, reaction times were for each of 28 famous historical figures. The faster for single items than for unrelated historical people were chosen so that most items. This replicates a frequently-reported subjects should recognize each figure, but result which is known as the fan effect would have little actual knowledge about (Anderson, 1974, 1976; Hayes-Roth, 1977; the lives these people led. We decided to Lewis & Anderson, 1976; Moeser, 1979; use real people rather than making up arbi- Smith, Adams, & Schorr, 1978). The data trary names because we wanted our sub- do not reveal systematic trends in the rela- jects to treat the materials as naturally as ENCODING ELABORATION 169 possible. The memorization task should be condition. Whenever multiple facts were easier since learning the names of people in shown about a figure, subjects were explic- the experiment is not necessary. In addi- itly told the relationship between the cen- tion, we felt this design would lead to fewer tral fact and the other facts. Historical fig- confusions among responses during the ures were randomly assigned to conditions cued recall task. Most of the facts were such that seven figures appeared in each drawn from biographies of the people, and condition. Every subject received an inde- were true. Occasionally, a fact was altered pendent randomization of figures to condi- or fabricated to fit the constraints of the tions. experimental design. We felt the resulting Procedure. The experiment was con- plausibility might encourage subjects to ducted in two sessions, the second session treat the facts as real, instead of fictional was held exactly 1 week after the first. The curiosities of no import outside the labora- first session consisted of four tasks, per- tory. formed in a fixed order: A memorization Each subject saw only a subset of the phase for the materials; a dropout phase, facts about each historical figure. One of where subjects received additional practice the facts for each person, called the central on the materials; a cued recall phase, and a fact, was common to all four subsets. The reaction-time verification phase. In the relationship between this fact and the memorization phase, each set of facts about others shown in the subset, called the sup- a figure was presented one at a time on the portive facts, determined the condition of screen. Single fact sets were displayed for the set. The first condition, the caused-by 18 seconds, while all other sets were dis- condition, included two supportive facts played for 36 seconds. The subject was which were plausible reasons for the central asked to memorize the facts presented fact. The two causal facts were designed to about each person. be as unrelated to each other as possible, In the next part of the experiment, the and independently imply the central fact. dropout phase, the sentence was presented Thus, if subjects were able to recall either with the subject of the sentence replaced by causal fact None, they should have been the word Who or Whose. The subject re- able to generate inferences from that fact sponded to each query by typing in a name. which might contain the central fact. If they Feedback was given, then the subject was could recall both causal facts, the intersec- asked to specify how that fact was related tion of the plausible inferences should be a to others in its set. Again feedback was very restricted set which included the cen- given. If the subject correctly identified tral fact. In the second condition, the both the name and the condition of a fact, it resulted-in condition, two supportive facts was eliminated from the completion set. were each plausible outcomes of the central After all items had been tested, subjects fact. Again they were chosen to be as inde- were retested on items which remained in pendent of each other as possible. The third the completion set. The process was con- condition, the unrelated condition, included tinued until every fact was correctly iden- two supportive facts about each person tified once. which were unrelated to the central facts On completing the study phases, subjects and to each other. In the final condition, the were given a sheet of paper containing the central fact was shown alone. Historical names of each historical figure, and were figures were assigned to one of the four asked to write down all the facts they could conditions (single, caused-by, resulted-in, remember about each person. In addition, unrelated) on a random basis for each sub- they were asked to identify how the facts ject, with the constraint that each subject were related. Subjects were allowed to saw an equal number of figures in each complete this phase at their own pace. The 170 BRADSHAW AND ANDERSON

final task in the first session was a reaction and single items was not significant (t = time task. In this task, a sentence appeared 0.96). However, the mean of these items on the terminal screen, and subjects were took significantly longer to learn than re- asked to decide whether or not the fact was lated items (p < .01). Because of differ- one they had studied earlier. Foils were ences in supportive items across condi- constructed by randomly mis-pairing names tions, future analyses will only be reported and sentence stems. At the end of the reac- for central items. Unless otherwise men- tion time trials, subjects were reminded of tioned, analyses of all items showed a simi- the second session, but were not given any lar pattern to analyses of central items. information about the tasks in the second In the cued-recall task, items were scored session. for gist. Table 2 shows recall rates for both The second session consisted of two sessions by condition. An analysis of vari- tasks, a cued-recall task and a reaction time ance showed a significant effect of condi- task. Each of these tasks was similar to the tion type (p < .001). The effect of delay corresponding task in the first session. In approached significance (p < .086), while the reaction time task, the same foils were the interaction was not significant. Separate used, but all items were presented in a dif- orthogonal comparisons for each session ferent random order. showed recall rates for unrelated items to be lower than that of single items, which did Results and Discussion not differ from related items. Recall rates in In order to analyze the initial ease of the related and single conditions were quite learning particular items, we calculated the high. We had hoped to show that initial average number of trials for items in each learning was equivalent in all conditions. condition to be correctly completed in the However, trying to remember multiple, un- dropout phase. The mean number of trials related items about a single historical figure to dropout is 1.43, 1.37, 1.66, and 1.72 for was so difficult that, even though subjects the caused-by, resulted-in, unrelated, and received more practice on these items than single conditions, respectively. An analysis on related items, immediate recall was of variance revealed a significant main effect somewhat reduced. of item condition (p < .005). This analysis The results of this experiment were gen- was performed by averaging together erally supportive of the ACT elaboration scores for all facts which were presented in model. Clear differences were found be- a given condition. The experimental design tween the related and unrelated condition in required supportive facts to vary by condi- terms of recall of central facts. However, tions. Only the central facts were constant this experiment failed to find a significant for each possible condition (see Table 1). advantage of the related conditions over the The results could have been influenced by single condition in percentage recall. We spurious differences in supporting items in speculated that this might be due to high the various conditions. Therefore, a second levels of recall displayed in these condi- analysis was performed on the average tions. Therefore, the second experiment dropout rates of central items only. Drop- was designed to achieve lower levels of re- out rates in the conditions were 1.36, 1.39, call. 1.55, and 1.72 for the caused-by, resulted- in, unrelated, and single conditions, respec- EXPERIMENT 2 tively. An analysis of variance again showed Experiment 2 is basically a repetition of a significant effect of item condition (p < Experiment 1 with a slight change in the .01). Orthogonal comparisons showed no design. We decided to reduce initial learn- difference in dropout rates for related con- ing of items, to see if the performance on ditions. The difference between unrelated single and related sets was similar due to ENCODING ELABORATION 171

TABLE 2 PERCENTAGE RECALL RATES FOR CENTRAL ITEMS BY CONDITIONS AND DELAY

Condition

Experiment Caused-by Resulted-in Unrelated Single

Experiment 1 Immediate Recall 93 96 80 93 Week Recall 89 92 74 87 Experiment 2 Week Recall 70 75 45 61 Experiment 3 Week Recall 52 61 32 38

ceiling effects. We chose to omit the initial performance was similar to that obtained in cued recall task. Since initial recall levels Experiment l. Again they reflect difficulty were so high, these scores did not reveal in learning single and unrelated items. much about the effects of the experimental Table 2 shows cued recall rates for items manipulation. tested in the second session. An analysis of variance on recall rates revealed a signifi- Method cant effect of item condition (p < .001). As The method of Experiment 2 was similar in Experiment 1, recall rates in the unre- to that of Experiment 1, with differences lated condition were significantly poorer noted below. than in other conditions. Unlike Experi- Subjects. Subjects were 41 Carnegie- ment 1, recall rates for single items were Mellon University undergraduates who somewhat below the mean of related items. either received course credit or $3.00/hour Although this difference was not statisti- for their participation. cally significant (t = 1.33), the overall pat- Procedure. Experiment 2 took place in tern more closely conforms to predictions two sessions, like Experiment 1, 1 week made from the encoding elaboration model. apart. The only difference in procedure Also not significant, but consistent with took place in the first session, which did not Experiment 1, the resulted-in condition had have an initial cued recall task. Thus, the slightly higher performance than the first session had three phases: the study caused-by condition. phase, the dropout phase, and the reaction Experiment 2 thus replicated most of the time verification phase. The procedure in findings of the first experiment and is en- the second session was identical to the first; tirely consistent with the elaboration a cued-recall task followed by a reaction model. Again we observe the powerful ef- time task. fect of having to remember multiple unre- lated propositions in lowering the level of Results and Discussion recall. Although recall of central items in The average number of trials for central related conditions was not significantly items to be completed in the dropout phase higher than recall of single items, the trend are: 1.48, 1.46, 1.64, and 1.67 for the in Experiment 2 was clearly in this direc- caused-by, resulted-in, unrelated, and tion. Accordingly, we decided to reduce the single items, respectively. An analysis of level of initial learning still further in Ex- variance confirmed item condition to have a periment 3, to see if a significant separation significant main effect (p < .015). Dropout of these conditions could be achieved. 172 BRADSHAW AND ANDERSON

EXPERIMENT 3 a significant effect of item condition (p < Experiment 3 used the basic design de- .001). Orthogonal comparisons revealed the veloped in Experiments 1 and 2. Two single item recall rate was significantly changes to this design were made, how- lower than the mean of the related items (p ever. The level of practice with materials in < .001). The trend toward a separation of the first session was reduced still further, these conditions in Experiment 2 was and in the reaction time task administered strong enough to reach significance in the in the second session, a new set of foils was present experiment. The data are thus used. These foils were constructed to be completely in accord with the pattern pre- thematically related to targets. In general, dicted by the encoding elaboration model. the use of such foils prevents subjects from making reaction time decisions on the basis GENERAL DISCUSSION of thematic information (Reder & Ander- son, 1980), and so the fan effect should be The experiments reported here show a present for both related and unrelated items. consistent and powerful effect of item or- ganization on recall of materials. In every Method experiment, we found recall performance for the central fact to be best when it was The method of Experiment 3 was similar supported by related facts, next best when to that of Experiment 2, with differences alone, and worst when studied along with noted below. unrelated facts. This pattern of data sup- Subjects. Subjects were 28 Carnegie- ports the ACT model. As mentioned in the Mellon University undergraduates who introduction, both the encoding-appropri- either received course credit or $3.00/hour. ateness model and the distinctiveness model Procedure. Experiment 3 again took do not make clear predictions about the place in two sessions, 1 week apart. The ordering of related and unrelated condi- reaction time task was eliminated from the tions. In addition, the distinctiveness model first session. It consisted of two parts, the does not predict the superiority of the re- study phase and the dropout phase. The lated condition to the single condition in second session was identical to that of the terms of percent recall. Thus we feel the first experiments, except that the reaction elaboration model to be the most useful one time task used thematically related foils in- to explain depth of processing effects. stead of unrelated foils. An alternative formulation of the distinc- tiveness model might be that sets of traces Results and Discussion form integrated units, which are more or The average number of trials for central less distinctive. If this interpretation is ac- items to be completed in the dropout phase cepted, it is unclear why we should expect are: 1.53, 1.52, 1.70, and 1.72, for the differences between the unrelated and re- caused-by, resulted-in, unrelated, and lated sets of items. In fact, unrelated sets of single conditions, respectively. An analysis items should be highly unusual, and there- of variance did not reveal a significant ef- fore very memorable. It has been suggested fect for item condition, although the overall that this problem could be circumvented by pattern was similar to that obtained in the postulating that unrelated items are not first two experiments. collected into sets, while related items are. Table 2 shows cued recall rates for items However, this just points out the vagueness tested in the second session. An analysis of of current distinctiveness model formu- variance performed on recall rates revealed lations. ENCODING ELABORATION 173

The experiment found a consistent supe- by Reder and Anderson (1980). These au- riority of the resulted-in condition over the thors contrasted memory of text passages caused-by condition. Although the differ- with memory of summaries of those pas- ences never approached significance, they sages. In their text condition subjects represent an anomaly. Note that this com- studied the original text while in the sum- parison is based on the same central facts; mary condition the subjects studied sum- the conditions only differ in terms of the maries that just contained the main facts. supporting facts. We would like to think These summaries were between one-fourth that it reflected a weaker relationship and one-fifth as long as the original texts. among the facts in the caused-by condition. Reder and Anderson have consistently However, our efforts to establish that this found better recall in the summary condi- was the case have not been successful. We tion. had a separate group of 20 subjects rate the There are clear analogies between the mean interrelatedness of the triples of facts summary condition and our single condition in the caused-by and resulted-in condition. and the text and our related condition. We Resulted-in triples had an average relation- speculate that the reason why Reder and ship of 5.12 on a 7-point scale (7 was highly Anderson found the opposite result was related), whereas caused-by facts had a that their textbook facts were simply less mean of 5.37 on the same scale. Also, we related than our related facts. However, found little correlation between these rat- there are some potentially significant meth- ings for individual historical figures and re- odological differences between the Reder call performance on those figures (r = and Anderson situation and our own. In our -.01). Thus, our rating measures were fail- experiment, subjects had more study time ing to tap important factors influencing re- in the related condition than in the single call. It is perhaps not surprising that sub- condition whereas this tended not to be the jects have a poor sensitivity to the kind of case in the original Reder and Anderson interconnectedness that makes for good experiments. More recently, Reder and recall. Anderson (in press) have performed an ex- The results of these experiments are periment that showed study time is not the somewhat at odds with the results reported relevant factor.

APPENDIX REACTION TIMES (sec) AND PERCENT ERROR RATES FOR CENTRAL ITEMS BY CONDITIONS AND DELAY

Condition

Experiment Caused-by Resulted-in Unrelated Single Experiment 1 Immediate test 1.80 (3.7) 1.62 (7.5) 1.77 (7.5) 1.55 (2.2) Week test 1.77 (4.5) 1.67 (8.3) 1.68 (9.0) 1.54 (3.7) Experiment 2 Immediate test 1.60 (6.6) 1.52 (4.5) 1.77 (5.9) 1.60 (4.9) Week test 1.70 (6.3) 1.67 (7.3) 1.79 (9.1) 1.76 (5.2) Experiment 3 Week test (correct items) 2.08 (17.9) 2.10 (1.43) 2.15 (9.7) 2.11 (12.8) Week test (foil items) 2.20 (13.2) 2.34 (10.7) 2.17 (10.7) 2.14 (tl.2) 174 BRADSHAW AND ANDERSON

REFERENCES KINTSCH, W. Memory representations of text. In R. L. Solso (Ed.), Information processing and ANDERSON, J. R. Retrieval of propositional informa- cognition. Hillsdale, N.J.: Erlbaum, 1975. tion from long-term memory. Cognitive Psychol- LEWIS, C. H., & ANDERSON, J. R. Interference with ogy, 1974, 6, 451-474. real world knowledge. Cognitive Psychology, ANDERSON, J. R. Language, memory, and thought. 1976, 8, 311-335. Hillsdale, N.J.: Erlbaum, 1976. LOCKHART, R. S., CRAIK, F. I. M., & JACOBY, L. L. ANDERSON, J.R., & REDER, L. M. An elaborative Depth of processing, recognition and recall: Some processing explanation of depth of processing. In aspects of a general memory system. In J. Brown L. Cermak & F. Craik (Eds.), Levels of process- (Ed.), Recall and Recognition. London: Wiley, ing in human memory. HiUsdale, N.J.: Erlbanm, 1976. 1979. MOESER, S. D. The role of experimental design in in- BOBROW, S., & BOWER, G. H. Comprehension and vestigations of the fan effect. Journal of Experi- recall of sentences. Journal of Experimental Psy- mental Psychology: Human Learning and Mem- chology, 1969, 80, 455-461. ory, 1979, 5, 125-134. BOWER, G. H., KARLIN,-M.B.,-& DOECI~,--A. Com- NELSON, D. L. Remembering pictures and words: Ap- prehension and memory for pictures. Memory and pearance, significance, and name. In L. Cermak Cognition, 1975, 3, 216-220. & F. Craik (Eds.), Levels of processing in human BRANSFORD, J. D., FRANKS, J. J., MORRIS, C. D., & memory. Hillsdale, N.J.: Erlbanm, 1979. STEIN, B. S. Some general constraints on learning NELSON, D. L., REED, V. S., & MEEVoY, C. L. En- and memory research. In L. Cermak & F. Craik coding strategy and sensory and semantic inter- (Eds.), Levels of processing in human memory. ference. Memory and Cognition, 1977, 5, 462-467. Hillsdale, N.J.: Erlbaum, 1979. REDER, L. M. The role of elaborations in memory for CRAIK, F. I. M., & LOCKHART, R. S. Levels of pro- prose. Cognitive Psychology, 1979, 11, 221-234. cessing: A framework for memory research. REDER, L. M., & ANDERSON, J. R. A partial resolu- Journal of Verbal Learning and Verbal Behavior, 1972, 11, 671-684. tion of the paradox of interference: The role of integrating knowledge. Cognitive Psychology, CRAIK, F. I. M., & TULVING, E. Depth of processing 1980, 12, 447-472. and the retention of words in episodic memory. Journal of Experimental Psychology: General, REDER, L. M., & ANDERSON, J. R. Effects of spacing 1975, 104, 268-294. and embellishment on memory for the main points EYSENCK, M. W. Depth, elaboration, and distinctive- of a text. Memory and Cognition, in press. ness. In L. Cermak & F. Craik (Eds.), Levels of SCHULMAN, I-I. G. Encoding and retention of semantic processing in human memory. Hillsdale, N.J.: and phonemic information in short-term memory. Erlbaum, 1979. Journal of Verbal Learning and Verbal Behavior, HAYES-ROTH, B. Evolution of cognitive structures and 1970, 9, 499-508. processes. Psychological Review, 1977, 84, SMITH, E. E., ADAMS, N., & SCHORR, D. Fact re- 260-278. trieval and the paradox of interference. Cognitive HYDE, T. S., & JENKINS, J. J. Recall for words as a Psychology, 1978, 10, 438-464. function of semantic, graphic, and syntactic 'TuLVING, E. Relation between encoding specificity orienting tasks. Journal of Verbal Learning and and levels of processing. In L. Cermak & F. Craik Verbal Behavior, 1973, 12, 471-480. (Eds.), Levels of processing in human memory. JACOBY, L. L., & CRAIK, F. I. M. Effects of elabora- Hillsdale, N.J.: Eflbanm, 1979. tion of processing at encoding and retrieval: Trace WICKELGREN,W. A. The long and the short of mem- distinctiveness and recovery of initial content. In ory. Psychological Bulletin, 1973, 80, 425-438. L. Cermak & F. Craik (Eds.), Levels of process- ing in human memory. Hillsdale, N.J.: Erlbaum, 1979. (Received August 14, 1981)