PsycholRes (1992) 54: 103-109
PsychologicalResearch Psycholo ische Forschung © Springer-Verlag 1992
The generation effect in primed word-fragment completion reexamined Ulrich Olofsson and Lars-G6ran Nilsson Departmentof Psychology,Universityof Ume~,S-901 87 Ume~,Sweden ReceivedJune28, 1991/AcceptedNovember14, 1991
Summary. Two experiments were designed to test a claim made by Gardiner (1988) that there are generation effects in implicit memory as measured by word-fragment completion. Subjects either read words at study or generated the words from fragments. As in previous research, fragments were completed to a greater extent if they were identical at study and test than if they differed. In Experiment 1 it was found that subjects could recognize explicitly the exact form of fragments that had been used for selfgeneration and distinguish these from other forms of fragments. An analysis of the contingency relations between recognition of fragments and fragment completion showed a high degree of dependence between the two tests. In Experiment 2 it was found that the match of surface features between study and test was a necessary, but not sufficient, condition to produce enhancement of priming. The results are interpreted as supporting the claim that generation does involve a data-driven component in addition to semantic elaboration.
Introduction Explicit memory and implicit memory are descriptive terms introduced by Graf and Schacter (1985) to make a distinction between forms of memory that require deliberate or conscious recollection of a study episode and those that do not require such a recollection. A long series of experiments has demonstrated dissociations between tasks considered as mainly measuring explicit memory and tasks seen as mainly reflecting implicit memory. For example, Squire, Shimamura, and Graf (1985) and Warrington and Weiskrantz (1970) have demonstrated that amnesic patients, who show very poor performance on standard-
This research was supportedby a grant from the SwedishCouncil for Researchin the Humanitiesand SocialSciencesto Lars-G6ranNilsson. Offprint requests to: U. Olofsson
memory tests, perform at essentially the same level as control subjects on implicit-memory tests. Others, using subjects from a normal population only, have demonstrated that variables that affect explicit-memory tasks such as recall and recognition have no effect on implicit-memory tasks such as word-fragment completion and stem completion. It is with this latter type of dissociation that the present study is concerned. More specifically, our point of departure is a claim that semantic elaboration, which is known to affect explicit memory, has little or no effect on implicit memory. For example, Graf and Mandler (1984) and Jacoby and Dallas (1981) have noted that levels of processing produce considerable effects on recall and recognition, but have little or no effect on implicit-memory tests such as perceptual identification and word-stem completion. On the basis of such findings, several researchers (e. g., Jacoby, 1983; Roediger & Blaxton, 1987 a; Roediger & Weldon, 1987) have argued that explicit memory is influenced by conceptually driven processing, whereas implicit memory is more sensitive to data-driven or surface processing. A distinction has recently been made between datadriven and conceptually driven implicit tasks. Tasks such as category generation and free association can reveal memory without requiring conscious recollection, but are sensitive to levels of processing manipulations and are thus conceptually driven. The present study is concerned exclusively with data-driven implicit tasks. In contrast to this processing view, Tulving (1985; Tulving, Schacter, & Stark, 1982; Tulving & Schacter, 1990) has argued that specific characteristics of different memory systems are responsible for performance differences in implicit- and explicit-memory tests. Tulving and his colleagues assume an episodic-memory system to be responsible for performance in explicit-memory tests such as recognition and recall, and a perceptual-representation system for an implicit-memory test such as word-fragment completion. Stated generally, the memory-systems view assumes that insensitivity to a levels of processing manipulation reflects the properties of an underlying system
104 (e. g., retrieval processes that are different from those of an episodic-memory system). The memory-systems view, as formulated by Tulving and colleagues, is therefore incompatible with the effects of semantic elaboration in fragment completion. The empirical findings mentioned on levels of processing and their theoretical interpretation led Gardiner (1988; 1989) to conduct a series of experiments on the generation effect in explicit and implicit memory. The generation effect refers to the phenomenon that items that have been self generated at study are recalled and recognized better than items that have simply been read at study (e. g., Gardiner & Hampton, 1985; Graf, 1980; McElroy & Slamecka, 1982; Rabinowitz & Craik, 1986; Slamecka & Graf, 1978). The predominant interpretation given to such data is that self-generation of items at study produces a deeper semantic processing of the items than simply reading them (e. g., Gardiner & Hampton, 1985). However, as was stated by Gardiner (1988), semantic or conceptual processing may not be the only consequence of generation of items at study. On the basis of the results of a study carried out by Glisky and Rabinowitz (1985), Gardiner argued that, in addition to semantic processing, a second factor, data-driven or surface-processing, might be involved in the generation effect. Previously, Kolers and Roediger (1984) had also argued that a second factor of this sort might be responsible for the additional "repetition of operations" enhancement of the generation effect. The basic design of Gardiner' s (1988) five experiments was to have subjects either read stimulus words at study or generate them from fragments and brief semantic cues. The test was a word-fragment completion test, in which the subjects were instructed to complete word fragments with the first word that came to mind. The critical manipulation involved conditions in which the subjects at test were given fragments that were either identical to, or different from, those given at study. Thus, a word such as ASSASSIN might have been presented as A--A--I- at study and either as A--A--I- or as -SS-SS-N at test. Gardiner's (1988) principal finding throughout his experiments was that generation effects did occur in fragment completion, provided that the fragments used at test were identical to those used for generation at study. No generation effects were obtained when fragments were different at study and test; that is, primed fragment completion was similar following generation and read instructions at study. Gardiner' s explanation of these results was that generation effects do occur in implicit memory when identical fragments are given at study and test, because data-driven processing is sensitive to surface information. Explicit memory, on the other hand, relies more on conceptually driven processing and is generally insensitive to shifts in surface information. The generation procedure is seen as involving both semantic elaboration and data-driven processing. In this sense Gardiner' s findings appear to be very useful in furthering the understanding of implicit memory. The procedure used by Gardiner (1988) does, however, permit an alternative interpretation of the results. It is possible that the subjects used explicit-memory strategies at test and that conscious retrieval was the source of the generation effect. It may have been the case that the sub-
jects in Gardiner's experiments recognized at least some of the fragments at test as having been presented at study. If this was so, the subjects may also have remembered the solution to the generation task at study, before solving the fragment-completion task at test. The generation effect in the condition with same fragments at study and test would then reflect the reliable, and already well-established, generation effect in explicit memory.
Experiment 1 Gardiner (1989) acknowledged the possibility of explicit memory causing the generation effect and attempted to control for it, in a second study, by carefully hiding the relation between study and test. To achieve this, the study list was presented to the subjects as a pilot study for another experiment: a filler task consisting of assorted verbal and semantic tests was given as the supposedly real experiment, and fragment completion was disguised as a "verbal skills" test. Gardiner found that although the subjects were reportedly unaware of the relation between the completion test and the study list, the selective enhancement of generation on implicit memory for identical fragments prevailed. The magnitude of the enhancement was also about the same as that previously found. The size of the priming effect was reduced by about 40%, however; a result that could perhaps be attributed to the longer retention interval and the presence of a distractor task in this second study. It should be noted that some priming effects in the 1988 series of experiments were unusually large (e. g.,.33 for the read condition and .45 for the generate same condition in Experiment 5). To summarize, Gardiner's (1989)results corroborate the conclusion that there are generation effects in implicit memory, given a high degree of similarity between study and test items. Experiment 1 was conducted with the purpose of testing the validity of Gardiner's (1988) explanation in a different manner. The basic procedure was the same, with the addition of a test for recognition of fragments that preceded the fragment-completion test. The objective was mainly to replicate Gardiner' s findings and to investigate whether the subjects could recognize the exact form of the fragments presented at study. If recognition performance for study fragments were high, this could be an explanation of the enhanced performance exhibited for completion of same fragments.
Method Sul~/ects.Nine female and seven male subjects, aged I7- 28 years (mean 21.8), were tested individually.They were undergraduate students at the University of Ume~ or secondaryhigh-school students who had volunteered as subjects for memoryresearch. They were paid the equivalentof DEM 14 for their participation.
Materials. The subjects were presented with cards containing a brief description, followedby a target word. The target was either a complete word or a word fragment (e. g., "Naval commander"followed by either "'CAPTAIN"or "C-P-A--").The fragmentswere either identical to those that would later be presented at test or different (e. g., "-A-T-I-"). Both forms of the fragments were designed to offer a single solution in the
105 Swedish language. The two forms of the fragments bad no more than 1 given letter in common.
Table 1. Proportions of fragments judged as old and identical to studied fragments, as a function of stimulus type. Stimulus Type
Design and procedure. There were 24 items in each of four sets. The items were rotated across conditions (i. e., Generate-same, Generatedifferent, and Read) and subjects, in such a way that each item appeared in all conditions and as distractors at test, over the experiment. Type of study condition varied within subjects. The subjects were instructed to read descriptions and words aloud, and were informed that the descriptions were cues that would help them solve the fragments. Each card was presented for t0 s. The order of presentation was randomized separately for each subject. If the subject failed to solve a fragment, the experimenter supplied the correct solution. When the last card had been presented, the experimenter chatted informally with the subject and asked for his or her age and occupation. After approximately 3 rain of conversation a recognition test was given. Half of the study items (i. e., 12 from each study condition or a total of 36) were presented as word fragments, together with 12 distractors (not previously presented) in random order on two sheets of paper. The subjects were instructed to mark with a plus sign those items they were convinced had occurred in exactly the same form in the study list (i. e., the same letters in the same positions - strict recognition criterion) and those they thought they had seen, but were less certain about, with a minus sign (lenient recognition criterion). The subjects were given 4 rain for this recognition test. The fragment-completion test followed immediately upon the fragment-recognition test. All study-list targets appeared in addition to the 12 distractors from the recognition test and 12 new distractors. The subjects were instructed to complete the word fragments with the first word that came to mind. The fragments were presented in random order on four test sheets. The subjects were told that this was a new task and that any relation to the preceding tests was irrelevant. They were given a maximum of 16 rain to complete a total of 96 fragments.
Results and discussion The subjects failed to generate target words for an average of 6.5 items (13%). The results obtained are described in three separate sections, one for the recognition test, one for the fragment-completion test, and one for the analysis of the dependence between the tests.
Recognition. Mean proportions of fragments judged as identical to fragments presented at study for each of the three study conditions and the false-alarm rate (for distractors) are presented in Table 1. Since the general pattern of data was similar for the strict and the the lenient recognition criterion, only the results from data sampled under the strict criterion will be presented. It can be seen that almost 20% of the Generatesame fragments were correctly recognized. For fragments that were different from those at study and for fragments presented as words in the read condition at study, proportions of recognition were considerably lower, .08 and .05, respectively. These two conditions represent types of false alarm that are higher than the pure false-alarm rate of .01. The degrees of freedom for all F values were adjusted according to the Geisser-Greenhouse procedure. A oneway ANOVA of the data collected under the strict recognition criterion yielded a highly significant effect of study condition, F (1,15) = 18.44,p <.002. Pair-wise comparisons with the Tukey HSD test, computed for repeated measures, showed that the Generate-same condition was significantly better than the other ,conditions, at p <.05. The difference
M
SD
Generate same
Generate different
Read
New
.18 .18
.08 .08
.05 .11
.01 .02
Note. Only "Generate same" are hits.
between the Generate-different and Read conditions was not significant. Thus, it seems that subjects can recognize the exact form of the fragments and discriminate between fragments presented in the same and in the other conditions.
Word-fragment completion. The results from the wordfragment completion test are presented in Figure 1, separately for those items that had been included in the fragment-recognition test and those that had not. As can be seen for the conditions Same, Different, and Read, there is a test-priming effect of, on the average, 5%. Moreover, the effect of study condition seems to be about the same for items that had and those that had not, been presented in the recognition test. A 2 (recognition test) x 3 (study condition) ANOVA was performed in order to examine the possibility of test priming. The analysis showed that there was neither a main effect of recognition-test appearance nor an interaction effect involving the recognition-test factor, p >.25. In the following analyses fragment-completion data are collapsed over the recognition-test factor. The baseline completion rate of each subject was subtracted from that subject's score in the different study conditions. Three planned comparisons were made on the basis of Gardiner's (1988) findings. A t-test adjusted for within-subjects comparisons was used. The difference between the Generate-same and Generate-different conditions fell short of significance at the c~= .05 level, t (15)= 1.52, p <.10. The difference between Generatesame and Read was significant, t (15) = 2.71,p <.01, but not between the Generate-different and Read conditions, t (15) = 1.10,p <.2. Since the difference between Generate-same and Generate-different conditions did not come out significant, as in Gardiner (1988), we only partially succeeded in replicating his findings. Dependence. In this final section of the results, we examine the dependence between recognition of fragments and fragment completion for the Generate-same condition. The responses to the 12 items that had appeared in both tests were arranged in a 2 × 2 contingency table and tested by Yule' s Q, a statistic recommended by Hayman and Tulving (1989) for measuring the degree of dependence between two successive tests. There were few cases of recognition without completion1 (.042) compared to the joint event of recognition and 1Joint probabilities are given in parenthesis.
106
I III Not in RnTest I [] In Rn Test
03 t(J3
E0 3 LL 0
0..
E ot0O 0 13.. 0 0..
Same Different Read
New
Fig. 1. Proportion of correctly completed fragments as a function of study condition and inclusion in recognition test, Experiment 1
completion (. 146). There were quite a number of cases of completion without recognition (.328), but successful completion was much more frequent than successful recognition (91 vs. 36). The joint event of failure both to recognize and to complete had the highest probability (.484). The value found, Q = .67, indicates a high degree of dependence. A significance test of Q based on the log odds ratio (Hayman & Tulving, 1989) proved it highly significant, Z 2 = 14.4, p <.001. The dependence obtained could originate from explicitmemory strategies used in completion. Alternatively, it could be due to data-driven processes involved in both recognition of fragments and completion of fragments that are sensitive to the match of surface features between study and test. If the latter explanation is true, then it is not recognition that produces the enhancement in completion, but priming that contributes to the recognition of fragments, perhaps by rendering the fragments easier to identify and to recognize. The size of priming effects obtained by Gardiner (1988) was almost twice as large as that of those obtained here and of those reported in Gardiner's (1989) second study. These unusually large effects suggest contributions from explicit memory. Nevertheless, the enhancement of Generate-same over Read conditions remains about the same in all studies (about. 12).
Experiment 2 Experiment 2 deals with the possibility that the results found by Gardiner (1988; 1989; Gardiner, Dawson, & Sutton, 1989) may be accounted for in transfer-appropriate processing terms alone, without reference to generation. In this experiment a control condition was added that made it possible to compare generate and read conditions with
surface features held constant between study and test. In Gardiner's procedure only the generate condition provided perfect surface-feature overlap in that the same fragments were presented at study and at test. In the present experiment fragments of the target words were also presented in the read condition, but the instruction induced a shallow level processing in order to unconfound generation and surface-feature similarity. As Gardiner himself pointed out (Gardiner et al. 1989), the degree of match between study and test items is critical for obtaining enhanced priming effects for self-generated study items. Since changes in surface variables are known to reduce priming effects, it is important to equate the study conditions on this dimension. For the explicit test we expected to replicate the generation effect - i. e., superior performance for the subjects who generate the target words to that of the subjects who only read the words. For the completion test we expected identical fragments to be completed more often than alternate fragments both in the generate condition and in the read condition, since changes in surface information should reduce priming effects. The most critical comparison was that between generation and read for identical fragments. If generation were superior to read, this would be a strong support for the dual-component explanation of the generation effect; if generation and read were equal, this would show that it is feature overlap between study and test and not self-generation that causes the selective enhancement reported by Gardiner (1988; 1989). We also expected generation and read to be equal for different fragments. Finally, we expected independence between recall and completion for both groups.
Method Subjects. Twenty-four undergraduate students, 16 female and 8 male, were randomly assignedto either of two groups and were tested individually. The age of the subjects ranged from 18 to 37, with a mean of 23.3 years. They were recruited partially from introductory psychology courses and partially through advertisements on the campus. They received an equivalent of DEM 14 for participating in the experiments. Materials. Seventy-five of the items from Experiment 1 were used; of these, 50 were presented in the study list and 25 were given as distractors in the fragment-completiontest. A Macintosh LC computer was used to present the study list. The program was designed to be similar to the presentation in Experiment 1. Two cards were shown in succession on the screen: the first contained either a description (generate condition) or a target word (read condition); the second contained a fragment of the target word. The text on the screen was black on white background; the font was monospacedCourier 12 points. The same font and the same size were used on the test sheets for fragment completion. Design and procedure. One group of subjects was required to solve the fragments at study (Generate group). The other group was shown the target words and was then required to check that the fragments were legitimate fragments of the target words, a "shallow processing" task (Read group). The descriptions/target words were displayed for 3 s and the fragments were displayed for 7 s. There was a 1-s interval between pairs, followed by a tone. The items were presented in random order for each subject and, as in Experiment 1, they were rotated so that each item appeared in all conditions, across subjects. Both lists had 6 buffers and 4 fillers in addition to the targets. The buffers occurred at the beginning and
107 at the end of the lists, and the fillers occurred at random positions in the lists. Buffer and filler items were excluded from statistical analyses. In the list given to the Read group the fragments associated with the fillers and buffers were made illegitimate. The reason for this was to make the checking task plausible to the subjects, by supplying about 13% actual misfit. The subjects were informed that the list was rather long and that a memory test would follow upon presentation. The total time required for presentation was about 12 min. Immediately after study the subjects were given 5 min for free recall of target items. A distractor task then followed, requiring the subjects to rate the familiarity of 104 words given to them on four sheets of paper. The mean time to complete this task was 7 min (SD = 1.3). Finally, there was a fragment-completion task with 75 items. One-third of the fragments were identical to those presented at study, one-third were alternative fragments that had no more than 1 letter position in common with those presented at study, and the final third were new fragments. The subjects were allowed 10 min for this task. It was emphasized that this was a new task and that the subjects should complete the fragments with the first word that came to mind, except foreign words and person names.
~D
E O o c
E
03 It. tO O Oo 13_
Same
Results and discussion After the experiment it was discovered that six target words also occurred in the familiarity-rating task, so that the subjects had been given two presentations of those items. To correct this error, these six items were excluded from further analyses. The three sets thus contained 23 items each. The mean number of failures to generate words was 6.75 (SD = 2.86), or 11.25% of the items. Failures in the checking task were rare, only 1.1%.
Free recall. The mean number of words recalled for the Generate condition was 12.1 (SD = 4.46) and for the Read condition 10.0 (SD = 4.73), respectively. The difference between the conditions of 2.1 items was significant, t (22) = 2.38, p <.05, demonstrating the typical facilitation of self-generation in an explicit-memory test.
Word-fragment completion. The data from the fragmentcompletion tests are shown in Figure 2. The baseline completion rate of each subject was subtracted from that subject's scores before the differences between conditions were tested. The hypotheses were tested by means of four t-tests. In the Generate group there was a significant difference between the same, M = 6.33, SD = 3.39, and different, M = 4.25, SD = 3.55, conditions, t (11) = 2.93, p<.01 (one-tailed), but not in the Read group, Msame 4.83, SD = 2.95 vs. Mdifferent 4.92, SD = 2.71, p >.2. The selective enhancement of generation on priming when surface information is held constant was thus obtained, but there was no enhancement of feature similarity per se. Contrary
Different
New
Fig. 2. Proportion of correctly completed fragments as a function of study condition, Experiment 2
to the hypothesis, the difference between Generate same, M = 6.33, SD = 3.39, and Read same, M = 4.83, SD = 2.95, conditions was significant, t ( 2 2 ) = 2.06, p <.05 (twotailed), showing that similarity was a necessary, but not sufficient, condition for the enhancement of priming. As was expected, there was no difference between the Generate-different, M = 4.25, SD = 3.55, and Read-different, M = 4.92, SD = 2.71 conditions, p >.2.
Dependence. As can be seen in Table 2, the Q value for the Generate condition is not significant, at o~ = .05, but the value for the Read condition is highly significant, although the association as reflected in the size of Q seems m o d e > ate. This finding was unexpected and is difficult to interpret. A possibility is that the dependence obtained for the Read group is due to the choice of explicit-memory test. The most common test for this type of contingency analysis is yes/no recognition, which given a better index on the content of memory than free recall. Since free recall was used here, one can suspect that fewer observations occur in the cells that require successful recall and higher frequencies in the cells that presuppose failure to recall, than would have been the case if recognition had been used to measure explicit memory. We may speculate that the test would have shown independence if recognition had been employed. Without any concern about ceiling effects, recognition might have been a better choice here.
Table 2. Joint probabilities of recall and word-fragment completion, experiment 2. Association
Joint probabilities Group
Rc, WFC
Rc, WFC
Rc, WFC
Rc, WFC
Q
Z2
Generate Read
0.132 0.121
0.142 0.106
0.287 0.276
0.437 0.496
0.173 0.344
3.17 11.69
Note. Rc is recalled target words, WFC is correctly completed fragments of target words.
108
General discussion In line with Gardiner (1988; 1989; Gardiner et al. 1989) the present experiments have demonstrated that generation effects occur in fragment completion, provided that the fragments used at test are identical to those used for generation at study. The order and magnitude of the effects were also comparable to those obtained in Gardiner' s experiments. Moreover, Experiment 1 demonstrated that subjects recognized same fragments better than the other types of fragments used. This reasoning holds true for those fragments that were rated by the subjects as having the exact form of the fragments that had been presented at study. Reasonably, this category of fragments should be of prime interest here, since it is this category that shows that subjects can differentiate the exact form of the fragments from any other form presented at study. With relaxation of this criterion, the variance increases; the tendency is the same, but the effect is not statistically reliable. It is to be expected that two successive episodic tests are dependent and that explicit and implicit tests are not dependent (Tulving et al. 1982; Hayman & Tulving, 1989). Hayman and Tulving (1989) demonstrated such contingent dissociation between recognition and fragment completion on the one hand, and recognition and cued recall of the same material on the other. The high degree of dependence between recognition and completion of those fragments that appeared in both tests can, however, be seen as reflecting common data-driven processes in the two tasks rather than the existence of an explicit-memory component in fragment completion. In Experiment 2 it was expected that the two "same" conditions would produce a superior performance compared to the "different" conditions. The fact that no effect of feature similarity was obtained in the read condition can probably be attributed to a single aspect of the procedure, namely that subjects had been exposed to the complete words before seeing the fragments. It is plausible that this exposure provided enough overlap in surface features, and so it made no difference whether the subjects had been presented with the actual fragments or with their alternative form. As was pointed out above, the memory-systems view cast in Tulving and Schacter's (1990) terms does not predict any facilitation of semantic elaboration on a task such as fragment completion, since fragment completion is regarded as being driven by a pre-semantic perceptual-representation system. If self-generation involves a data-driven component in addition to semantic elaboration, however, then it seems reasonable that a perceptual-memory system would benefit from generation. It would benefit to the extent that the data-driven processing enhances the perceptual record of an item represented in that memory system. It is possible that a test such as fragment completion is mixed on both the implicit/explicit and the datadriven/conceptually-driven dimensions. The laborious and problem-solving nature of the test may encourage explicit strategies and possibly associative processing. Roediger and Blaxton (1987 b) examined word-fragment completion on these grounds and argued that the task is mainly data-
driven, although less sensitive to surface features than more rapid implicit-memory tasks such as perceptual identification and word-stem completion. As a comparison, several authors have noted that recognition is mixed on these two dimensions. It has been argued that recognition involves something that could be labeled an implicit component and processes that are both data-driven and conceptually driven (cf. Mandler, 1980; Roediger & Blaxton, 1987b). However, Mandler does not use these terms; rather he refers to context-free recognition based on familiarity and an identification process based on context retrieval. In sum, we interpret our results as support for a two-factor explanation of the generation effect (Gardiner, 1988; Kolers & Roediger, 1984). That is, self-generation involves both general semantic elaboration and a specific data-driven component.
References Gardiner, J. M. (1988). Generationand priming effects in word fragment completion. Journal of Experimental Psychology." Learning, Memory, and Cognition, 14, 459 -501. Gardiner, J. M. (1989). A generation effect in memory without awareness. British Journal of Psychology, 80, 163-168. Gardiner, J. M., & Hampton, J. A. (1985). Semantic memory and the generation effect: Some tests of the lexical activation hypothesis. Journal of Experimental Psychology: Learning, Memory, and Cognition, 11, 732-741.
Gardiner, J. M., Dawson, A. J., & Sutton, E. A. (1989). Specificity and generality of enhanced priming effects for self-generated study items. American Journal of Psychology, 102, 295- 305. Glisky, E. L., & Rabinowitz, J. C. (1985). Enhancing the generation effect through repetition of operations. Journal of ExperimentaI Psychology: Learning, Memory, and Cognition, 11, 193 - 205. Graf, P. (1980). Two consequences of generating: Increased inter- and intraword organization of sentences. Journal of Verbal Learning and Verbal Behavior, 19, 316-327. Graf, P., & Mandler, G. (1984). Activation makes words more accessible, but not necessarilymore retrievable.Journal of Verbal Learning and Verbal Behaviol, 23, 553 - 568. Graf, P., & Schacter, D. L. (1985). Implicit and explicit memory for new associations in normal and amnesic subjects. Journal of Experimental Psychology: Learning, Memory, and Cognition, 11, 501 - 518. Hayman, C. A. G., & Tulving, E. (1989). Contingent dissociation between recognition and fragment completion: The method of triangulation. Journal of Experimental Psychology: Learning, Memory, and Cognition, 15, 228-240. Jacoby, L. L. (1983). Remembering the data: Analyzing interactive processes in reading. Journal of Verbal Learning and Verbal Behaviol; 22, 485-508. Jacoby, L. L., & Dallas, M. (1981). On the relationship between autobiographical memory and perceptual priming.Journal of Experimental Psychology: General, 110, 306- 340. Kolers, P. A., & Roediger, H. L., HI (1984). Procedures of mind. Journal of Verbal Learning and Verbal Behavior, 23, 425 -449. Mandler, G. (1980). Recognizing: The judgement of previous occurrence. Psychological Review, 87, 252- 271. McElroy, L. A., & Slamecka, N. J. (1982). Memorial consequences of generating nonwords: Implications of semantic-memoryinterpretations of the generationeffect. Journal of Verbal Learning and Verbal Behavior, 21, 249-259. Rabinowitz, J. C., & Craik, F. I. M. (1986). Specific enhancement effects associated with word generation. Journal of Memory and Language, 25, 226-237.
109 Roediger, H. L., III, & Blaxton, T. A. (1987 a). Retrieval modes produce dissociations in memory for surface information. In D. S. Gorfein & R. R. Hoffman (Eds.), Memory and cognitive processes: The Ebbinghaus centennial conference (pp 349-379). Hillsdale, NJ: Erlbaum. Roediger, H. L., III,& Blaxton, T. A. (1987 b). Effects of varying modality, surface features, and retention interval on priming in wordfragment completion. Memory & Cognition, 15, 379-388. Roediger, H. L., II],& Weldon, M. S. (1987). Reversing the picture superiority effect. In M. A. McDaniel & M. Pressley (Eds.), bnage~y
Squire, L. R., Shimamura, A. P., & Graf, P. (11985). Independence of recognition memory and priming effects. A neuropsychological analysis. Journal of Experimental Psychology: Learning, Memory, and Cognition, 1], 37-44. Tulving, E. (1985). How many memory systems are there? American Psychologist, 40, 385-398. Tuiving, E., & Schacter, D. L. (1990). Priming and human memory systems. Science, 247, 301-306. Tulving, E., Schacter, D. L., & Stark, H. A. (1982). Priming effects in word-fragment completion are independent of recognition memory.
and related processes; theories, individual differences, and applications (pp 151-174). New York: Springer-Verlag.
Journal of Experimental Psychology: Learning, Memory, and Cognition, 8, 336- 342.
Slamecka, N. J., & Graf, P. (1978). The generation effect: Delineation of a phenomenon. Journal of Experimental Psychology: Human Learning and Memory, 4, 592-604.
Warrington, E. K., & Weiskrantz, L. (1970). Amnesic syndrome: Consolidation or retrieval? Nature, 228, 628 630.
