Journal of Memory and Language 44, 96–117 (2001) doi:10.1006/jmla.2000.2749, available online at http://www.idealibrary.com on

The Concreteness Effect in Implicit and Explicit Memory Tests

Maryellen Hamilton

Baruch College, City University of New York and Suparna Rajaram

State University of New York at Stony Brook

Four experiments were conducted to examine the concreteness effect in implicit and explicit memory measures. Experiment 1 replicated prior reports of an imagery effect on an implicit conceptual memory test. In Experiment 2, we confirmed our prediction of conceptual sensitivity of free recall, explicit general knowledge, explicit word fragment completion, and the implicit general knowledge tests with a levels of processing manipulation. Furthermore, although we obtained the concreteness effect (better memory for concrete than abstract nouns) in free recall and the explicit general knowledge test, we failed to find this effect in the implicit general knowledge test. Experi- ment 3 revealed that the failure to find the concreteness effect on the implicit general knowledge test was not attributable to combining two encoding manipulations in Experiment 2. In Experiment 4, we ruled out the possibility that the failure to find the concreteness effect in conceptual implicit memory may be related to the number of meaningful associates for targets. We discuss the implications of these findings within the context of the transfer appropriate processing framework (Roediger, Weldon, & Challis, 1989) and the dual-code hypothesis (Paivio, 1971, 1991) of memory. © 2001 Academic Press In this article, we focus on the nature of retical conceptualization of three well-known priming in conceptual memory tests. The exper- variables that are presumed to engender concep- iments we report were motivated by the theo- tual or meaning-based processing. These three variables are encoding pictures versus words, We thank Richard Delaney, Kevin Felker, William Gart- imaging an item’s referent versus reading land, Mana Heydarpour, Laurance Lemmerling, Kathy Le, words, and encoding concrete versus abstract Ellen Martin, Tara Price, Maura Rzeznikiewicz, Amanda nouns. These variables are assumed to share a Salomon, Jennifer Sanders, Sandra Siegel, Christine Ver- common mechanism in Paivio’s (1971, 1991) dino, and Carmine Vozzolo for their help in creating mate- rials and testing participants. We are grateful to Larry Ja- dual-code hypothesis. We considered the effects coby, Roddy Roediger, Mary Susan Weldon, and an of these variables on conceptual forms of ex- anonymous reviewer for providing helpful comments on an plicit and implicit memory with an empirical earlier version of this article. Parts of the research reported focus on the effects of concrete versus abstract in this article were presented at the annual meeting of the Eastern Psychological Association in Philadelphia, PA, nouns. 1996; Washington, DC, 1997; and at the annual meeting of Explicit and implicit memory tasks differ in the Psychonomic Society in Philadelphia, PA, 1997. This the test instructions given to the participants research was supported by NIH Grant R29MH57345-01 to (Graf & Schacter, 1985). Explicit memory tasks Suparna Rajaram. Address correspondence and reprint requests to Suparna such as recall and recognition require partici- Rajaram, Department of Psychology, SUNY at Stony pants to think back to the prior study episode Brook, Stony Brook, NY 11794-2500. Fax: (631) 632-7876. and perform the memory task at hand. In con- E-mail: [email protected], or to Maryellen trast, implicit memory tasks make no reference Hamilton, Department of Psychology, Saint Peter’s Col- lege, 2641 Kennedy Blvd., Jersey City, NJ 07306. E-mail: to the study episode and participants are re- [email protected]. quired to complete the task at hand with the first

0749-596X/01 $35.00 96 Copyright © 2001 by Academic Press All rights of reproduction in any form reserved. CONCRETENESS EFFECT 97 answer that comes to mind. According to the plicit memory tests such as free recall (Paivio, transfer appropriate processing framework pos- Rogers, & Smythe, 1968) and recognition tulated by Roediger and colleagues (Roediger, (Shepard, 1967; see also Madigan, 1983). Im- Weldon, & Challis, 1989), explicit memory aging the referent of a word produces better tasks typically entail considerable use of con- retention than nonimaged words (imagery ef- ceptual processing for retrieving the study epi- fect), and concrete nouns are better remembered sode (but also see Blaxton, 1989). As a result, than abstract nouns (concreteness effect) on ex- these tasks benefit from experimental manipu- plicit memory measures including free recall, lations which enhance conceptual processing of recognition, and paired associate learning (see information (see Roediger & McDermott, 1993 Paivio, 1971, 1983, for reviews). for a review). Implicit memory tasks can either In contrast to these outcomes, the effects of rely predominantly on perceptual processes or these variables on conceptual implicit tasks ap- on conceptual processes. For instance, implicit pear to be less clear and sometimes discrepant tasks such as word fragment completion (Blaxton, 1989; McDermott & Roediger, 1996; (_le_h_n_;solution: “elephant”) or word Nyberg & Nilsson, 1995; Weldon & Coyote, stem completion (ele______) require partic- 1996; Wippich & Mecklenbrauker, 1995). Spe- ipants to complete the word puzzle with the cifically, some researchers have failed to find a first solution that comes to mind and prefer- picture superiority effect on conceptual implicit entially rely on perceptual processing of in- memory tests (McDermott & Roediger, 1996; formation (see Roediger & McDermott, 1993 Weldon & Coyote, 1996), and yet, others have for a review). Here, we focus on a different reported a positive imagery effect on this same class of implicit memory tests that rely on class of tasks (Blaxton, 1989; Nyberg & Nils- conceptual processes. In these tests, no men- son, 1995; Wippich & Mecklenbrauker, 1995). tion is made of the study episode but implicit Our aim in this article is to empirically address memory is aided by the conceptual overlap these discrepancies in the results with concep- between study and test items. For instance, in tual implicit tests and provide a coherent theo- the implicit category production test, partici- retical description of these effects on explicit pants are provided with a conceptual cue at tasks as well as conceptual implicit memory test that bears no physical resemblance to the tasks. study item (e.g., for the studied item “ele- As mentioned earlier, the picture superiority phant,” the category label “animal” is pro- effect, the imagery effect, and the concreteness vided at test) and are asked to write down the effect are assumed to be related effects in exemplars of that category that come to mind. Paivio’s (1971, 1991) dual-code hypothesis. The increase in the generation of studied ex- According to the simplest interpretation of this emplars over nonstudied exemplars, or prim- hypothesis, there are two possible codes which ing, provides the measure of memory on these can be activated when an item is presented at implicit tests. Experimental manipulations study, verbal (logogens) or imaginal (imagens) that enhance conceptual processing improve (Paivio, 1978). These two codes are assumed to priming on these conceptual implicit tasks be additive in nature. It is postulated that pic- (Blaxton, 1989, Srinivas & Roediger, 1990). tures, items studied with imagery instructions, The effects of the three variables of present and concrete words activate both the verbal and interest, picture versus word processing, imag- the imaginal code, whereas simply reading a ing the referent versus reading words, and word only activates one code (verbal). There- studying concrete versus abstract nouns, have fore, the superior memory for pictures, items been well established on measures of concep- studied with imagery instructions, and concrete tual explicit memory and are straightforward items is the result of the additive contributions (see Paivio, 1971, 1983 for reviews). Specifi- of two codes to memory. cally, pictures are recalled or recognized better Under this possibility, the activation of two than words (picture superiority effect) on ex- codes requires cross-system, or referential, pro- 98 HAMILTON AND RAJARAM cessing (Paivio, 1971, 1991). Specifically, it is and types of processing (perceptual and concep- assumed that if the encoding of a picture is tual) so as to test the predictions of the transfer accompanied with invoking its name, both the appropriate processing framework (Roediger et verbal and imaginal systems become engaged al., 1989). Of these five tests, two conceptual through referential processing. Similarly, when tests, free recall and semantic cued recall (i.e., participants are required to image the pictorial target is cued with a semantically related word, representation of a word (e.g., sailboat), it is e.g., study: universe; test: cosmos____ ), were assumed that referential processing enables the explicit memory tests. The third test, graphemic activation of both codes. Finally, in the dual- cued recall, was an explicit perceptual test code hypothesis, it is also assumed that concrete where the target was cued with a perceptually words facilitate the activation of logogens and similar word (e.g., unversed______). The fourth imagens, whereas abstract words can only ac- test, word fragment completion, was an implicit cess logogens, and thus, referential processing perceptual test where participants were pro- is likely for the former and not the latter set of vided with a perceptually degraded version of items (see Paivio, 1983, for a review). the target(__iv__se)andwere asked to Thus, all three effects of interest are pre- provide the first solution that came to mind. The sumed to be guided by cross-system, or refer- fifth test, general knowledge, was an implicit ential, processing. Because referential process- conceptual test. Participants were provided with ing is assumed to be conceptual in nature conceptually related test questions (“What was (Paivio, 1991, pp. 280–281), the activation of the Big Bang said to have created?) which bore dual codes implies one type of conceptual pro- no physical resemblance to the studied words cessing of information. It should be noted that (universe) and were asked to answer with the referential processing represents only one type first solution they could think of to solve these of conceptual processing and that conceptual questions. processes during encoding and retrieval could As predicted by the transfer appropriate pro- be of different types depending on the experi- cessing framework (Roediger, Weldon, Stadler, mental manipulation employed to engender & Riegler, 1989), Blaxton found a typical im- conceptual processing. Within the context of the agery effect on all conceptual tests, explicit research reported here, our definition of concep- (free recall and semantic cued recall) or implicit tual processes at encoding will be restricted to (general knowledge). More items were recalled referential processing across the verbal and or questions answered with words studied with imaginal codes. Based on this assumed role of the imagery instructions than with words stud- referential processing in the dual-code hypoth- ied without the imagery instructions. In addi- esis, the picture superiority effect, the imagery tion, she found no imagery effect in either the effect and the concreteness effect may be con- explicit perceptual task (graphemic cued recall) sidered as conceptually driven effects on mem- or the implicit perceptual task (word fragment ory. As such, these three effects should emerge completion; see also Durgunoglu & Roediger, on memory tests that are classified as concep- 1987, for converging evidence). This finding of tual tests, both explicit and implicit, within the a positive imagery effect on a conceptual im- transfer appropriate processing framework. plicit test has been subsequently replicated with To examine this idea, Blaxton (1989, Exper- a category association task (Nyberg & Nilsson, iment 3) examined the imagery effect in five 1995) and a verb association task1 (Wippich & memory tasks. In this experiment, participants were asked to form mental images of the item’s 1 In the verb association task, participants studied action referent for half the items presented during phrases (i.e., holding the tire) within a text. During test they study. No specific study instructions were given were provided with verbs (half of which were from early action phrases) and asked to fill in the first associations that for the remaining items. Blaxton constructed came to mind (e.g., holding ______). Note that there is five different memory tests by combining dif- perceptual overlap between the study item and the test cue ferent test instructions (explicit and implicit) and therefore this task does not meet the criterion suggested CONCRETENESS EFFECT 99

Mecklenbrauker, 1995). Together, these find- 1995; Wippich & Mecklenbrauker, 1995) sup- ings support the notion that imagery entails ports both the dual-code hypothesis and the conceptual processing (Paivio, 1991) and that transfer appropriate processing framework. But this conceptual manipulation enhances perfor- on this view, the absence of a picture superiority mance on conceptual implicit memory tests. effect on conceptual implicit tasks (McDermott With respect to the picture superiority effect, & Roediger, 1996; Weldon & Coyote, 1996) is the reported findings are quite different from problematic for both accounts. those found with the imagery effect. In a recent To our knowledge, there are no published study, Weldon and Coyote (1996) consistently studies that have examined the effects of study- failed to find a picture superiority effect on two ing concrete versus abstract nouns on concep- implicit conceptual tests (category production tual implicit memory. This lacuna in the litera- and word association) such that pictures and ture becomes particularly striking in light of the words produced equivalent amounts of priming. discrepancy we have just discussed. Specifi- At the same time, a significant levels of pro- cally, positive imagery effects have been found cessing effect (Craik & Lockhart, 1972) was on implicit conceptual memory tests (Blaxton, found on both tests, indicating that the tests 1989; Nyberg & Nilsson, 1995; Wippich & were indeed conceptual in nature. Furthermore, Mecklenbrauker, 1995) but no picture superior- when the same two tests were given with ex- ity effect has been found on these tasks (Mc- plicit instructions, recall was better for items Dermott & Roediger, 1996; Weldon & Coyote, studied as pictures than for items studied as 1996). Therefore, in order to explore the reasons words. Similarly, McDermott and Roediger for this discrepancy it is essential to examine (1996) failed to find a picture superiority effect whether a theoretically related effect (i.e., the on the implicit conceptual test of category pro- concreteness effect) would be obtained on these duction. In fact, only one study (Nicolas, 1995) tasks. Based on the dual-code hypothesis, it is has reported a picture superiority effect on an reasonable to assume that superior memory for implicit conceptual test (category production). concrete nouns relative to abstract nouns likely It is not clear why Nicolas (1995) has obtained arises from the conceptual processing advan- a picture superiority effect on an implicit con- tage (arising from the cross-system or referen- ceptual test while Weldon and Coyote (1996) tial processing) for concrete nouns. Thus, ac- and McDermott and Roediger (1996) failed to cording to the transfer appropriate processing find one. Weldon and Coyote have suggested framework, an advantage for concrete over ab- that one possibility may be that he used inten- stract nouns should be obtained not only on tional study instructions, which may have led to explicit tasks such as free recall, recognition, explicit contamination on the implicit memory and paired associate learning (see Paivio, 1983, test. However, Weldon and Coyote and McDer- for a review), but also on conceptual implicit mott and Roediger’s failure to find the picture tasks. We set out to test this prediction. superiority effect is the focus of the current We selected the implicit general knowledge research since this finding seems to contradict task as our conceptual implicit task to examine the predictions of the transfer appropriate pro- the effects of the concrete/abstract variable. Our cessing framework. selection of this task was guided by the studies Thus, if we assume that imaging and picture that have tested the effects of theoretically re- processing both entail activation of two codes lated variables. In particular, we selected the through referential processing (Paivio, 1991), implicit general knowledge test as our concep- then the imagery effect on the conceptual im- tual implicit task because Blaxton (1989, Ex- plicit tasks (Blaxton, 1989; Nyberg & Nilsson, periment 3) reported a robust imagery effect on this task. Having selected this task, we first earlier for conceptual implicit tests (i.e., no perceptual over- sought to replicate the positive imagery effect lap between study and test cues). However, we include it (Blaxton, 1989) in our laboratory and with our here since the authors classify it a conceptual implicit task. subject pool. To this end, we used Blaxton’s 100 HAMILTON AND RAJARAM

(1989) general knowledge task as our implicit phase. All participants were presented with 40 conceptual test. It was expected that we would words during study for 5 s each. Twenty of the find a typical imagery effect; i.e., more ques- words were presented with imagery instruc- tions answered with items studied under imag- tions. Specifically, participants were instructed ery instructions than items studied without im- to form a mental image of the item’s referent. agery instructions. The participants were asked to hit a specially marked “Y” key as soon as they had formed a EXPERIMENT 1 mental representation of the item. They were told that the word may still remain on the screen Method for a predetermined amount of time and that Participants. Eighty-three undergraduates during this time they should keep the image in from SUNY at Stony Brook participated for their mind. The remaining 20 items were pre- credit as partial requirement for course work. sented with nonimaged instructions. Partici- The undergraduates were recruited from the pants were told to simply read the words. Again Psychology Department human subject pool they were instructed to hit the specially marked and informed consent to participate in the study “Y” key as soon as they had read the word. was obtained. Of these 83 participants, 35 took Participants were asked to read no slower or part in a norming study and 48 took part in the faster than they normally do. Once again they experiment. were instructed that the word would remain on Materials and norming. We used 35 partici- the screen for a predetermined amount of time pants to norm 132 general knowledge questions during which they should continue to read the selected from Blaxton’s (1989) materials for our word. Participants in conditions where imagery subject population. Blaxton’s materials were instructions were given first were told to make used because these materials had previously sure they did not image in the “read” condition produced a positive imagery effect on an im- as this would slow down their reading times. plicit conceptual test. From the 132 general The test phase was presented as a different knowledge questions, a pool of 80 items were experiment to help ensure that the participants selected based on the results of our norming did not associate the study and test phase. Par- study. The overall mean base-rate performance ticipants were told that for the second experi- for these items was 19%. ment they would have to answer trivia ques- Design. We used a 2 ϫ 2 within-subject tions. They were instructed that the majority of design. A given participant saw 40 words at the questions had multiple solutions, and there- study and the other 40 words served as new fore, there were no right or wrong answers. items at test. For each list of 40 words, half of Participants were instructed to answer the ques- the items were studied under imagery instruc- tions with the first word that comes to mind that tions and the other half of the items were simply answers the question appropriately. No mention read (nonimaged). The imaged and nonimaged of the prior study episode was made. All par- conditions were presented in a blocked fashion. ticipants took part in a practice session prior to Counterbalancing was done for item type (old/ the test. Test instructions and practice lasted new), study instructions (imaged/nonimaged), approximately 10 min. and the presentation order of study blocks (im- During the test phase the participants were aged followed by nonimaged vs nonimaged fol- presented with 80 general knowledge questions, lowed by imaged), thereby creating eight study one question at a time. Half of the questions had lists. target answers from the earlier study list (half Procedure. Participants were tested in groups from the imaged condition and half from non- of one to three. Items were presented on Zeos imaged condition). The other half of the ques- 486 computers using Schneider’s (1990) Micro- tions had nonstudied target answers. Each ques- Experimental Laboratory (MEL; Version 1.0). tion appeared on the computer screen, with the Participants took part in both a study and a test prompt, “If you have a solution, Press the Y CONCRETENESS EFFECT 101

FIG. 1. Mean proportion of questions answered (and standard errors) with studied imaged, studied nonim- aged, and new items in Experiment 1.

Key.” Participants were instructed to hit the items and the proportion of questions answered specially marked “Y” key as soon as they with target nonstudied items, for each encoding thought of a one word solution. Once they hit condition. Significant priming was obtained in the “Y” key, participants were instructed to type both the imaged condition [t1(47) ϭ 8.37, SE ϭ in their response. After typing in the response, .02; t2(79) ϭ 9.36, SE ϭ .02] and the nonim- or if the participants did not hit the “Y” key in aged condition [t1(47) ϭ 5.52, SE ϭ .02; the allotted time (20 s), the next question ap- t2(79) ϭ 4.66, SE ϭ .02]. In order to determine peared automatically. if there was an imagery effect, a within subject At the end of the testing phase, each partici- t test was performed between the priming (stud- pant was debriefed and questioned about their ied–nonstudied) proportions of imaged items awareness of any relationship between the study (M ϭ .21) and nonimaged items (M ϭ .09). and test phase. This comparison revealed a significant imagery effect [t1(47) ϭ 4.65, SE ϭ .02; t2(79) ϭ 4.81, Results and Discussion SE ϭ .02]. Thus, we replicated Blaxton’s None of the participants noticed a connection (1989) findings. Specifically, we found both between the study phase and the test phase. significant priming and a significant imagery Figure 1 illustrates the mean proportion of ques- effect on our implicit conceptual memory test of tions answered with target items for the studied general knowledge. imaged items, the studied nonimaged items, and Having replicated the imagery effect on the for new items. For all results reported, the alpha implicit conceptual test of general knowledge, level was set at p Ͻ .05. Unless otherwise we proceeded to examine our central question noted, the first statistical value (e.g., the F value of interest, i.e., the effect of studying concrete or the t value) represents the subject analysis, versus abstract nouns on conceptual implicit whereas the second statistical value represents memory. In Experiment 2, we included five the item analysis. A significant difference was memory tasks to examine the concreteness ef- found among the means [F1(2,141) ϭ 18.84, fect on memory. We included a free recall task

MSe ϭ .03; F2(2,237) ϭ 18.93, MSe ϭ .04]. To in order to replicate the well-established con- determine if priming occurred, planned compar- creteness effect on an explicit memory task ison t tests were performed between the propor- (Paivio & Csapo, 1969; see Paivio, 1983 for a tion of questions answered with target studied review). Our critical measure was the concep- 102 HAMILTON AND RAJARAM tual implicit task of general knowledge and its ness effect may also be expected on the explicit selection was based on the findings from previ- word fragment completion task. However, if ous studies (Blaxton, 1989) and the results of concreteness of words is not as strong a con- our Experiment 1. We also included an explicit ceptual manipulation as levels of processing, version of this task in order to examine the the concreteness effect may not occur on this concreteness effect on a conceptual explicit task predominately perceptual task (we return to this that met the retrieval intentionality criterion issue under General Discussion). (Schacter, Bowers, & Booker, 1989). That is, we sought to confirm the presence of a concrete- EXPERIMENT 2 ness effect in an explicit task that was identical to the conceptual implicit task in all respects Method except for the retrieval instructions. To our Participants. A total of 324 undergraduates knowledge, there is no published report of an from SUNY at Stony Brook participated in this explicit version of the general knowledge task. experiment for credit needed for partial fulfill- Finally, we also included two widely used per- ment of a course requirement. Of these, 108 ceptual tasks, the implicit word fragment com- participants took part in the norming study and pletion task and its explicit version of fragment 216 in one of the five experimental test condi- cued recall. The effect of encoding concrete and tions. Of these 216 participants, 40 each took abstract items has not been previously exam- part in four different tasks, free recall, question ined for this category of tasks. We did not cued recall, implicit word fragment completion, expect any effect on the implicit version of this and fragment cued recall. In the implicit general task because this task is at the perceptual end of knowledge task, 56 participants were tested. the processing continuum. While we included Design. In this experiment a 2 ϫ 2 ϫ 2 ϫ 5 an explicit version of the word fragment com- mixed-factorial design was used. Studied status pletion task we did not have a strong prediction (old vs new), item type (concrete vs abstract), for this task (we return to this issue shortly). and levels of processing (deep vs shallow) were In Experiment 2, we also included the levels manipulated as within-subject variables. Test of processing (Craik & Lockhart, 1972) manip- type (free recall, question cued recall, fragment ulation in our experiment to confirm the typical cued recall, implicit word fragment, or implicit assumptions regarding the processing nature of general knowledge) was manipulated as a be- these five tasks. Past research has documented a tween-subject variable. levels of processing effect on conceptual tasks From the pool of 80 critical items, we created but not on tasks that are predominantly percep- two lists of 40 items (20 abstract and 20 con- tual in nature (Challis & Sidhu, 1993; Hamann, crete). Each list was further divided so that half 1990; Srinivas & Roediger, 1990). However, the items (10 abstract and 10 concrete) were based on past research (Roediger, et al., 1992; shown with semantic encoding instructions and Weldon, Roediger, Beitel, & Johnston, 1995), the remaining half of the items were shown with we also predicted a levels of processing effect graphemic encoding instructions. The presenta- for fragment cued recall (i.e., the explicit per- tion of items was blocked with respect to se- ceptual task). A levels of processing effect is mantic and graphemic encoding. Altogether, believed to occur on this test, even though the eight possible study lists were created to cue is perceptual (word fragment), because ex- achieve appropriate counterbalancing. Thus, plicit retrieval in itself demands conceptual pro- across participants each item appeared as a stud- cessing (the participant needs to make a judg- ied or nonstudied item and for semantic and ment about whether an item has been seen graphemic encoding. In addition, the order for earlier). Therefore, this contamination of the these two encoding conditions was also bal- perceptual test with conceptual processing was anced across participants. expected to aid performance for semantically Materials and norming. All items, both at encoded words. On this logic, a small concrete- study and at test, were presented and data were CONCRETENESS EFFECT 103 collected on Zeos 486 computers by using we ensured that the nonstudied baseline hit rate Schneider’s (1990) Micro-Experimental Labo- for both the word fragments and the general ratory (MEL; Version 1.0) software system. Be- knowledge questions was, on average, 26%. cause our manipulation of interest (concrete- Furthermore, we ensured that our two critical ness) was inherent in our items, we needed to sets of items (abstract and concrete nouns) did select a new set of materials for Experiment 2. not statistically differ on their baseline perfor- A set of 80 nouns was selected from the Paivio, mance on word fragment completion, t(78) Ͻ 1, Yuille, and Madigan (1968) norms. Forty of or general knowledge questions, t(78) ϭ 1.32. these items were chosen to be highly concrete Our levels of processing manipulation was (scores above 6 on the 7-point concreteness based on a method used by Challis and Sidhu scale, M ϭ 6.74) and 40 items were chosen to (1993). For the graphemic (shallow) encoding be abstract (scores between 1.5 and 4.5 on the condition, participants were asked if a certain concreteness scale, M ϭ 2.69) in nature. Paivio letter was in the word (e.g., Does this word et al. also normed items on two additional contain a “c”?). Participants were never asked scales; imageability and meaningfulness. The about the same letter twice within a given list. imageability scale is a 7-point scale, whereas Approximately half of the answers were “yes” the meaningfulness scale reflects the number of and the other half were “no” responses. For the meaningful associates a participant generates semantic (deep) encoding condition, partici- for a given word. Since the imageability and pants were asked about the meaning of an item meaningfulness scales are highly correlated (e.g., Can you buy this?). A pool of 77 questions with the concreteness scale, concrete items were was created. Normative data showed that 38 of accompanied with high scores on both the im- the questions focused on concrete ideas (e.g., Is ageability (M ϭ 6.39) and the meaningfulness this a person?), whereas 39 questions focused (M ϭ 6.44) scales, and the abstract items were on abstract ideas (e.g., Is this desired?). Con- accompanied with low scores on the imageabil- crete questions were used for concrete items ity (M ϭ 3.62) and the meaningfulness (M ϭ while abstract questions were used for abstract 5.42) scales. All items were between 5 to 10 items. Each question appeared only once per letters in length and of medium frequency rang- study list and half of the questions required a ing from 10 to 49 (M ϭ 25) occurrences per “yes” response, whereas the other half required million words in print according to Thorndike a “no” response. and Lorge (1944). There was a significant dif- Our pilot data showed that single presenta- ference between concrete and abstract items on tion yielded weak priming on the implicit con- the concreteness scale, t(78) ϭ 26.50, SE ϭ .15, ceptual test. Therefore, the encoding procedure the imageability scale, t(78) ϭ 31.21, SE ϭ .09, was slightly different for the participants in the and the meaningfulness scale, t(78) ϭ 8.28, implicit general knowledge test. Following the SE ϭ .12. No significant differences were ob- precedence established in previous investiga- served between concrete and abstract items on tions (Challis & Sidhu, 1993; Weldon & Coy- both word length, t(78) ϭ 1.77, and word fre- ote, 1996), we used massed repetition during quency, t(78) ϭ 1.03. encoding in the implicit conceptual tests. Based For each item, both a word fragment and a on Challis and Sidhu’s (1993) parameters, each general knowledge question were created item appeared four times consecutively and four (e.g., target—strawberry; word fragment— separate questions were asked in each encoding st_a__e__y;general knowledge question— condition. For example, in the semantic encod- What fruit wears its seeds inside its skin?). ing condition the word “strawberry” was pre- These word fragments and general knowledge sented four times consecutively, and each time questions were normed to establish an accept- this word was followed by a different question able baseline rate of accuracy. One hundred (i.e., Is this made of wood?; Can this be found eight participants took part in one of four norm- in a garden?; Does this involve music?; Can this ing studies. Based on the four norming studies, grow?). An encoding question was used from 104 HAMILTON AND RAJARAM one to four times per study list. However, no than once and that each time an item was pre- question was used more than once for the same sented a new question would be asked. word. For each word, two of the questions had After the study condition, participants had a “yes” responses, whereas two of the questions 10-min retention interval where they played a had “no” responses. game of cards on the computer for 6 min and During retrieval, for all tests but free recall, were given test instructions and practice items participants were presented with the cues for the for the remaining 4 min.2 original 80 items. For a given participant, 40 Following the retention interval, participants items were studied, whereas the remaining 40 in the free recall condition were instructed to were new (nonstudied) items. In the implicit write down in any order all the words they could word fragment completion test and the explicit recall from the earlier study lists. They were fragment cued recall test, participants were pre- informed that the words could come from either sented with 80 word fragments. Similarly, for of the two lists that they saw earlier. Partici- the implicit general knowledge test and the ex- pants were given 7 min for the task. plicit question cued recall test, participants were For the remaining four tests, test items ap- presented with 80 general knowledge questions. peared on the computer screen one at a time The implicit and explicit tests differed only in with the prompt, “If you have a solution, press the instructions given. “Y.” When the participants pressed the “Y” key, Procedure. Participants were tested in groups they were asked to type in the solution and press of one to three. A given participant took part in the enter key. If the participant failed to press only one of five test conditions. In the free the “Y” key, the item remained on the screen for recall, explicit question cued recall, implicit 20 s and then timed out. word fragment completion, or the explicit word For the explicit tests (i.e., word fragment fragment cued recall test conditions, partici- cued recall and question cued recall), partici- pants saw single presentations of items in the pants were instructed to try to solve the items study phase. During the single presentation, presented in the allotted time. Participants were each word appeared on the computer screen for also informed that some of the cues could be 3 s. Following the word, a question that required solved with words that they had previously seen a yes/no response appeared on the screen for in either of the two encoding conditions. Partic- 2 s. Participants were asked to respond by hit- ipants were urged to solve only those cues for ting the specially marked “Y” or “N” keys which their solution was a word from the earlier (which were adjacent to each other on the key- studied lists. For the implicit tests participants board). In the shallow processing condition, were instructed to solve the items presented in participants were instructed to read the word the allotted time with the first word that came to carefully, pay attention to the letters, and an- mind. In the general knowledge tasks partici- swer the graphemic question. In the deep pro- pants were urged to give one word solutions. cessing condition, participants were instructed For all five groups, the entire procedure took to read the word carefully, pay attention to the less than 1 h. meaning of the word, and answer the semantic question. Results In the encoding procedure for the implicit In Fig. 2, we have collapsed the results across general knowledge test, the same manipulation the concrete/abstract variable to present the was used except that each word was presented priming and the levels of processing effects four times consecutively on the computer first. Figure 2 illustrates the mean proportion of screen with each presentation lasting 1 s fol- items retrieved for the nonstudied (new) condi- lowed by a question presented for 2 s. Partici- tion and both levels of processing encoding pants were given the same instructions as the single presentation groups except that they were 2 Participants in the free recall condition played the card informed that the same item may appear more game for 7 min because there was no practice phase. CONCRETENESS EFFECT 105

FIG. 2. Mean proportion of correct responses (and standard errors) for each item type for each of the five tasks used in Experiment 2. GK ϭ general knowledge task; WFC ϭ word fragment completion task. conditions for each of the five memory tests. processing effect was also observed on the ex- The levels of processing analyses and planned plicit word fragment completion test [t1(39) ϭ comparison t tests were performed on the cor- 2.75, SE ϭ .03, t2(79) ϭ 3.06, SE ϭ .03]. As rected measures of studied minus nonstudied expected, no levels of processing effect was items. observed for the implicit word fragment com- A typical levels of processing effect (i.e., pletion task [t1(39) Ͻ 1, t2(79) Ͻ 1]. greater retention for items studied under seman- Table 1 presents the studied, nonstudied, and tic processing when compared to items studied corrected scores for both concrete and abstract under graphemic instructions) was observed for nouns on each of the five tests. Figure 3 illus- all our conceptual tests [free recall, t1(39) ϭ trates the corrected recall and priming propor- 5.95, SE ϭ .02, t2(79) ϭ 4.95, SE ϭ .02; tions for both concrete and abstract nouns for explicit general knowledge test, t1(39) ϭ 7.6, each of the five memory tests. For every test but SE ϭ .03, t2(79) ϭ 8.66, SE ϭ .02; and implicit free recall, a 2 ϫ 2 ANOVA was performed general knowledge test, t1(55) ϭ 2.2, SE ϭ .02, with study status (old vs new) and item type t2(79) ϭ 2.40, SE ϭ .02]. A significant level of (concrete vs abstract) as factors. In addition, per

TABLE 1 Mean Proportion of Studied and Nonstudied Items Retrieved and Standard Deviations for Each Item Type in All Five Tasks in Experiment 2

Concrete Abstract

Task Studied Nonstudied St.-Nst. Studied Nonstudied St.-Nst.

Free Recall .23 (.12) 0 .23 (.12) .17 (.10) 0 .17 (.10) Explicit GK .35 (.17) .03 (.05) .32 (.18) .29 (.15) .04 (.06) .25 (.17) Explicit WFC .42 (.13) .06 (.07) .36 (.15) .41 (.16) .08 (.08) .33 (.18) Implicit WFC .48 (.13) .26 (.09) .22 (.14) .41 (.14) .24 (.12) .17 (.17) Implicit GK .32 (.15) .27 (.15) .05 (.16) .37 (.15) .30 (.14) .07 (.15)

Note. GK ϭ general knowledge task; WFC ϭ word fragment completion task. 106 HAMILTON AND RAJARAM

FIG. 3. Recall and priming proportions (and standard errors) for concrete and abstract nouns for each of the five tasks used in Experiment 2. GK ϭ general knowledge task; WFC ϭ word fragment completion task. our hypothesis, planned comparison t tests were effect of item type [F1(1,39) Ͻ 1; F2(1,78) Ͻ performed to detect priming and the concrete- 1]. No significant interaction was observed ness effect on our two implicit measures. [F1(1,39) Ͻ 1; F2(1,78) Ͻ 1]. Therefore, as Since there were no nonstudied items (intru- expected, there was no concreteness effect ob- sions) reported in free recall we did not use a served on the corrected priming scores in our 2 ϫ 2 ANOVA on these data. For the free recall explicit word fragment completion task task, significant differences were observed be- [t1(39) Ͻ 1; t2(78) Ͻ 1]. tween the proportion of studied concrete items In the implicit word fragment completion recalled and the proportion of studied abstract test, there was a main effect of study status items recalled (the concreteness effect) [t1(39) ϭ [F1(1,39) ϭ 152.71, MSe ϭ .01; F2(1,78) ϭ

2.71, SE ϭ .02, t2(78) ϭ 2.22, SE ϭ .03]. 93.35, MSe ϭ .02] and a main effect of item

In the general knowledge explicit task, a type by subject [F1(1,39) ϭ 5.23, MSe ϭ .01], main effect was found for study status but not by item [F2(1,78) Ͻ 1]. No significant

[F1(1,39) ϭ 140.63, MSe ϭ .02, F2(1,78) ϭ interaction was found [F1(1,39) ϭ 1.08,

147.94, MSe ϭ .02]. No main effect was found F2(1,78) ϭ 1.05]. Planned comparison t tests for item type [F1(1,39) ϭ 2.27, F2(1,78) Ͻ 1]. revealed significant priming for both the con- A significant interaction was found by subject crete items [t1(39) ϭ 9.89, SE ϭ .02; t2(39) ϭ

[F1(1,39) ϭ 7.01, MSe ϭ .05], but not by item 6.79, SE ϭ .03] and for abstract items [t1(39) ϭ [F2(1,78) ϭ 1.27]. Planned comparison t tests 6.81, SE ϭ .02; t2(39) ϭ 6.99, SE ϭ .03]. Per revealed that a significant concreteness effect our predictions, no concreteness effect was ob- was found on the corrected priming scores of served on the priming scores for the implicit the general knowledge explicit test by subject word fragment completion task [t1(39) ϭ 1.04; [t1(39) ϭ 2.64, SE ϭ .05], but not by item t2(78) ϭ 1.03]. Note that the small numerical [t2(78) ϭ 1.12], although note that the numer- trend for the concreteness effect in this task is ical difference was in the expected direction unlikely to be real because this trend became (concrete M ϭ .32; abstract M ϭ .26). Thus, as even smaller in the explicit version of the task expected, the overall pattern produced a con- and was not significant, as described above. creteness effect in this task. In the implicit general knowledge task, we For the explicit word fragment completion obtained a main effect of study status task, there was a significant main effect of study [F1(1,55) ϭ 14.16, MSe ϭ .01; F2(1,78) ϭ status [F1(1,39) ϭ 352.36, MSe ϭ .01; 14.21, MSe ϭ .01] and a main effect of item

F2(1,78) ϭ 279.50, MSe ϭ .02] and no main type by subject [F1(1,55) ϭ 10.81, MSe ϭ .01], CONCRETENESS EFFECT 107 but not by item [F2(1,78) Ͻ 1]. No significant predicted there was no levels of processing ef- interaction was observed [F1(1,55) Ͻ 1; fect on the implicit perceptual test. These find- F2(1,78) Ͻ 1]. Planned comparison t tests con- ings suggest that, according to the assumptions firmed that there was significant priming for of the transfer appropriate processing frame- both concrete items [t1(55) ϭ 1.93, SE ϭ .02; work, we identified the processing requirements t2(39) ϭ 1.99, SE ϭ .02] and abstract items of our tests correctly. Also, as predicted by [t1(55) ϭ 3.36, SE ϭ .02; t2(39) ϭ 3.56, SE ϭ transfer appropriate processing framework, we .02]. The most critical finding in this experi- did not find an effect of the conceptual manip- ment concerned the implicit general knowledge ulation of concrete and abstract nouns on our test; we did not find a significant difference perceptual tests. Furthermore, our conceptual between the priming scores for abstract items tests of free recall and the explicit question cued and the priming scores of concrete items on our recall did produce the concreteness effect. implicit general knowledge test [t1(55) Ͻ 1; The finding of major interest in this experi- t2(78) Ͻ 1]. Thus, we failed to find a concrete- ment is that we failed to obtain a concreteness ness effect in our conceptual implicit memory effect on the implicit general knowledge test. test. According to the transfer appropriate process- To help ensure that this failure to find a ing framework, a conceptual manipulation significant concreteness effect on our implicit should affect performance on a conceptual task. conceptual test was not due to the small levels We did find a levels of processing effect on the of priming observed in this task, we reanalyzed implicit general knowledge test, thereby illus- the data by dividing the data for high and low trating that the test is sensitive to a conceptual responders. Specifically, we examined whether manipulation. However, we failed to find a con- we would see a concreteness effect emerge if creteness effect on this task. In addition, a con- we only analyzed the performance of those par- creteness effect was found when the same test ticipants who showed relatively high levels of cues were given with explicit instructions priming. The data were divided in half accord- (question cued recall), illustrating that our fail- ing to the proportion of overall priming ob- ure to find a concreteness effect on our implicit served for each participant. A within subject t general knowledge test was not attributable to test was then performed on the data for the the nature of the test cues themselves. Before high-scoring participants in the implicit general we set out to describe the theoretical ramifica- knowledge test. However, no significant differ- tions of our findings, we sought to rule out two ence was found for the priming scores of con- empirical possibilities that may have counter- crete (M ϭ .13) and abstract nouns (M ϭ .16), acted a potential concreteness effect in our con- t(27) Ͻ 1, of these high responders. ceptual implicit memory test. The first of these two possibilities was tested in Experiment 3. Discussion One potential reason why we did not find a As predicted by the transfer appropriate pro- concreteness effect on our implicit conceptual cessing framework, there was a significant lev- test might be that we also included a levels of els of processing effect on all the conceptual processing manipulation at study. That is, in- tests (both explicit and implicit) and as we ex- cluding a levels of processing manipulation at pected on our explicit perceptual task.3 Also as study may have changed the way in which our items were encoded and, therefore, eliminated 3 We note here that even though we did obtain a levels of the concreteness effect. The possible mecha- processing effect on the explicit word fragment completion nism by which such elimination may occur is as task we failed to find a concreteness effect on this task. We follows. The concreteness effect is presumed to do not have a ready explanation for this discrepancy for the be due to a concrete items entailing more con- explicit word fragment completion task. Although, as pre- viously mentioned, we believe this finding may suggest that ceptual encoding compared to abstract items concreteness is not as strong a manipulation of conceptual (Paivio, 1971). However, specifically requiring processing as is levels of processing. participants to encode both concrete and ab- 108 HAMILTON AND RAJARAM

TABLE 2 Mean Proportion of Studied and Nonstudied Items Retrieved and Standard Deviations for Each Item Type in Experiments 3 and 4

Concrete Abstract

Studied Nonstudied St.-Nst. Studied Nonstudied St.-Nst.

Experiment 3 .34 (.16) .29 (.12) .05 (.15) .35 (.15) .29 (.12) .06 (.14) Experiment 4 .28 (.18) .23 (.12) .05 (.14) .30 (.14) .26 (.12) .04 (.16) stract nouns for meaning in our experiment may were interested in examining whether a con- have reduced the default differences in the con- creteness effect would arise on this test once ceptual encoding of these items. The explicit study instructions were changed. We did not use conceptual tests may have been less prone to the levels of processing manipulation; instead this problem because these tests utilize inten- participants were told that they were being tional retrieval. The use of a deliberate retrieval tested for their reading reaction times and that strategy may permit the subtle differences in the they should read no faster or slower than they conceptual and distinctive attributes of items at normally do. Each item was preceded by an encoding to emerge (see Roediger et al., 1992), asterisk for 1 s, and, as in Experiment 1, each thereby giving the concrete words an advantage item was presented for four massed repetitions over the abstract words despite the use of the of 1 s each. levels of processing encoding. To test this possibility, we conducted Exper- Results and Discussion iment 3 to examine whether the concreteness Table 2 contains the proportion of correct effect would arise on a conceptual implicit task studied and nonstudied questions for both con- if we did not use the levels of processing ma- crete and abstract items. There was a significant nipulation at study. Because the concreteness main effect of study status [F1(1,55) ϭ 13.28, effect is attributed to conceptual encoding of MS ϭ .01; F2(1,78) ϭ 14.65, MS ϭ .01] and concrete items (Paivio, 1971) and conceptual e e no main effect of item type [F1(1,55) Ͻ 1; tests should be sensitive to conceptual manipu- F2(1,78) Ͻ 1]. In addition, no significant inter- lations at encoding (Roediger et al., 1992), we Ͻ ϭ predicted a concreteness effect on our concep- action was observed [F1(1,55) 1; F2(1,78) tual implicit test once the use of the levels of 1.63]. As expected, planned comparison t tests processing manipulation at study was elimi- revealed significant priming for both concrete ϭ ϭ ϭ nated. items [t1(55) 2.44, SE .02; t2(39) 1.82, SE ϭ .02] and abstract items [t1(55) ϭ 3.14, EXPERIMENT 3 SE ϭ .02; t2(39) ϭ 3.44, SE ϭ .03]. However, there was no significant difference between the Method priming scores for concrete words (M ϭ .05) Participants. An additional group of 56 un- and the priming scores for abstract words (M ϭ dergraduates from SUNY at Stony Brook par- .06) [t1(55) Ͻ 1; t2(78) ϭ 1.77]. Once again, ticipated in the experiment to obtain credit we tried to eliminate the possibility that the needed for partial fulfillment of a course re- failure to observe the concreteness effect on our quirement. implicit measure was due to the low levels of Materials and procedure. All materials and priming observed. Therefore, we reanalyzed the procedures were identical to that of Experiment data only for that half of the participants who 2 with the following changes. We used only the showed higher levels of priming. No significant implicit general knowledge test because we difference was observed between the priming CONCRETENESS EFFECT 109 scores for these participants on concrete (M ϭ ber of meaningful associations for a particular .14) and abstract items (M ϭ .14) [t(27) Ͻ 1]. item. Specifically, the more meaningful associ- Thus, once again we failed to find a concrete- ations a particular item can produce, the more ness effect on a conceptual implicit test. potential incorrect solutions exist to our ques- Given that the default differences in the en- tion for that particular item. This problem is coding of concrete and abstract nouns did not unique to implicit open-ended tests because produce a concreteness effect in Experiment 3, these tests can be completed with any solution we considered the following possibility. One of and, therefore, competing responses may not be the important ways in which concrete and ab- suppressed as would be done in the explicit stract items differ is in the number of meaning- versions of the same tests. ful associations that can be generated for con- It should be noted that the meaningful asso- crete and abstract nouns (see Paivio et al.’s, ciates for concrete and abstract nouns do not 1968 norms). This difference may impair the differ on at least one other measure of mean- production of concrete words on open-ended ingfulness. Nelson and Schreiber (1992) re- tests such as the implicit general knowledge cently provided the meaning set size norms in test. which little differences were observed between As mentioned under “Method” in Experiment the meaning set size for abstract and concrete 2, in Paivio et al.’s (1968) norms, meaningful- nouns. In this approach, meaning set size is ness is measured by the mean number of asso- measured by having large groups of participants ciations participants generated for a given item. provide the first meaningful associate that The original pool of items that we used in comes to mind for a given item. The number of Experiments 2 and 3 were chosen so that the different items generated by a pool of partici- concrete items scored significantly higher than pants for a given item constitutes its meaning the abstract items on Paivio et al.’s imageabil- set size. Note that this measure of meaningful- ity, concreteness, and meaningfulness scales. ness differs from Paivio’s measure. In Paivio’s We used these criteria because these scales are multiple association measure of meaningful- highly correlated and because we wanted to ness, each participant provides many responses, create a large item set. However, these criteria whereas in Nelson and Schreiber’s single asso- led to the selection of concrete words that have ciation measure, each participant provides only more meaningful associations than the selected one response. abstract words and, therefore, could potentially Based on Nelson and Schreiber’s (1992) sin- provide more incorrect solutions. This latter gle association measure of meaningfulness, one point is elaborated below. would not expect response competition to play a All of our general knowledge questions had role in the production of concrete and abstract many possible solutions. Due to the nature of nouns on our conceptual implicit task. How- the questions, all of the possible solutions to a ever, based on Paivio’s multiple association given question were associated with each other. measure, response competition remains a possi- For example, our general knowledge question ble confound in our previous experiments. Fur- “What animal is a political symbol?” had the thermore, Nelson and Schreiber (1992) also target response “elephant.” However other pos- found that concrete and abstract nouns differed sible solutions to that question were donkey, on meaningfulness scales similar to those used eagle, and dove. Note that all possible solutions by Paivio even though the meaning set sizes of are associated with each other since they all concrete and abstract nouns did not differ. For belong to the category “animals.” Because our these reasons, the goal of Experiment 4 was to test was implicit (participants were instructed to determine whether the larger size of associated complete the item with the first word that came responses to concrete nouns had diminished the to mind), and many possible solutions to the production of target concrete words in an open- questions were associates of one another, the ended task such as implicit general knowledge intrusions may have been affected by the num- in our prior experiments. We tested this idea by 110 HAMILTON AND RAJARAM holding constant the meaningfulness scores for t2(58) Ͻ 1]. Once again, to help eliminate the concrete and abstract nouns from Paivio et al.’s possibility that the low levels of priming ob- (1968) norms. We predicted a concreteness ef- served on this task were masking the concrete- fect on our conceptual implicit task in Experi- ness effect, we reanalyzed the data for only ment 4 because this measure of meaningfulness those half of the participants who demonstrated was controlled. higher levels of priming. No significant differ- ence was found between priming scores for EXPERIMENT 4 concrete (M ϭ .15) and abstract items (M ϭ .13) for these participants [t(27) Ͻ 1]. Thus, we Method did not find a concreteness effect on the con- Participants. An additional group of 86 un- ceptual implicit test even when we equated for dergraduates from SUNY at Stony Brook par- the number of meaningful associations (Paivio ticipated for credit needed for partial fulfillment et al., 1968) for concrete and abstract words. of a course requirement. Of these 86 partici- pants, 30 took part in a norming study and 56 GENERAL DISCUSSION took part in the experiment. The goal of this research project was to ex- Materials and procedure. A pool of 60 items amine the effect of studying concrete versus (30 concrete and 30 abstract) was selected from abstract words on conceptual implicit memory. both Experiment 2 and from Paivio et al.’s Prior research has demonstrated that a related (1968) norms such that the abstract and concrete variable of imaging the referent of a word (see items only varied on the imageability and con- Paivio, 1983) produces greater priming on the creteness scales and not the meaningfulness implicit general knowledge test compared to scales [concrete M ϭ 5.82, abstract M ϭ 5.76; simply reading words (Blaxton, 1989). In con- t(58) Ͻ 1]. For each new item, a general knowl- trast, a related manipulation of studying pictures edge question was constructed and normed so does not produce greater priming than reading that the mean baseline rate of completion accu- words on another conceptual implicit memory racy for this pool of 60 items was 28%. All task of category association (Weldon & Coyote, procedures were the same as in Experiment 3 1996). Within this context, we explored the except that the participants saw 30 items at effects of concrete versus abstract words on study (15 concrete and 15 abstract) and 60 items different forms of memory. In Experiment 1 we at test (30 old and 30 new). replicated Blaxton’s (1989) finding of a positive imagery effect on the conceptual implicit mem- Results ory test of general knowledge. We then con- There was a significant main effect of study ducted three experiments to examine whether status [F1(1,55) ϭ 9.76, MSe ϭ .01; F2(1,58) ϭ the concreteness effect occurs on conceptual

9.42, MSe ϭ .01]. There was no main effect of implicit memory tests. To test this, we used the item type [F1(1,55) ϭ 2.99; F2(1,58) Ͻ 1]. implicit general knowledge test (Blaxton, There was also no interaction found 1989). In Experiment 2, we compared memory [F1(1,55) Ͻ 1; F2(1,58) Ͻ 1]. As can be seen in for concrete and abstract nouns in five different Table 2, we obtained significant priming for memory tasks that a priori make different in- concrete items [t1(55) ϭ 2.85, SE ϭ .02; structional and processing demands. These t2(29) ϭ 2.91, SE ϭ .02]. The priming was also tasks were free recall, explicit general knowl- significant for abstract items by subject edge test, implicit general knowledge test, ex- [t1(55) ϭ 1.92, SE ϭ .02], but not by item plicit word fragment completion test, and im- [t2(29) ϭ 1.64] (but note that the means were in plicit word fragment completion test. The the right direction). However, there was no sig- conceptual/perceptual processing demands of nificant difference between the priming scores these tasks were confirmed with a levels of for concrete words (M ϭ .05) and the priming processing manipulation at study. scores for abstract words (M ϭ .04) [t1(55) Ͻ 1; Specifically, free recall and explicit and im- CONCRETENESS EFFECT 111 plicit versions of the general knowledge test across Experiments 2, 3, and 4 in one within- produced significant levels of processing ef- subject t test. Even in this meta-analysis, we fects. The explicit version of the word fragment failed to detect a significant difference between completion test also produced a significant lev- concrete (M ϭ .05) and abstract items (M ϭ els of processing effect, as has been reported in .06) [t (167) Ͻ 1]. In addition, of the 168 other studies (Roediger et al., 1992). Further- participants, only 72 (or 43%) demonstrated more, as expected, the implicit version of the better performance for concrete over abstract perceptual word fragment completion task did nouns, 77 (or 46%) showed a reversal, and 19 not exhibit a levels of processing effect. These (or 11%) participants showed equal perfor- findings with the levels of processing manipu- mance for the two types of items. According to lation are completely consistent with the predic- a Wilcoxon matched pairs sign test, the differ- tions of the transfer appropriate processing ence in performance for concrete and abstract framework (Roediger, 1990; Roediger et al., nouns was not significant (p ϭ .66). Further- 1989, 1992). more, similar to the analysis we performed in With respect to our central question of inter- Experiments 2, 3, and 4, we divided the data est, the concrete/abstract variable produced pre- from all 168 participants in the implicit general dicted results with one important exception. The knowledge test in half and analyzed the priming predicted results included a concreteness effect scores of those participants who showed high on the explicit versions of the conceptual tasks, priming. We still failed to find a significant i.e., free recall and explicit general knowledge. difference between concrete (M ϭ .14) and ab- Furthermore, no concreteness effect was ob- stract items (M ϭ .15), t (83) Ͻ 1. tained on the predominantly perceptual tasks, To ensure that power was not an issue, we i.e., explicit and implicit word fragment com- also used Cohen’s (1988) power estimates. If pletion. However, contrary to the general expec- we assume a medium size effect of concrete- tations, we failed to find a concreteness effect ness, then in a one-tailed t test conducted with a on the implicit conceptual memory task of gen- within-subject design where the n is 56, accord- eral knowledge. ing to Cohen (1988; also see Aron & Aron We conducted two more experiments to ex- 1999, p. 278), we had 96% power in each of our amine two possible causes of this null effect. In analyses of the concreteness effect in the im- Experiment 3, we removed the levels of pro- plicit general knowledge test. If we assume a cessing manipulation on the reasoning that the small effect size for the concreteness effect on instructions to process both the concrete and the implicit measure, then according to Cohen abstract nouns at a deep (or a shallow level) (1988; and Aron & Aron, 1999) the number of may have inadvertently eliminated the default subjects needed for a one-tailed within-subject t processing differences between concrete and test with 80% power would be 172 participants. abstract nouns. However, our hypothesis that Our collapsed data had an n extremely close to the emergence of the default differences in the that (n ϭ 168) but there was no sign of a conceptual processing of concrete versus ab- concreteness effect (M ϭ .05 for concrete items, stract nouns would produce a concreteness ef- M ϭ .06 for abstract items) in these data. There- fect on the implicit general knowledge test, was fore, based on our collapsed data, and on Co- not supported in Experiment 3. In Experiment hen’s power estimates, we conclude that our 4, we equated the number of possible competing null effects are not a result of insufficient responses for concrete nouns compared to ab- power. stract nouns. However, this control also did not Our failure to obtain a concreteness effect in give rise to a concreteness effect. a conceptual implicit memory task is similar to In order to help ensure that we had sufficient the reports of a failure to obtain a picture supe- power for observing the concreteness effect, we riority effect in conceptual implicit memory analyzed the data for all 168 participants who tasks (McDermott & Roediger, 1996; Weldon took part in the implicit general knowledge test & Coyote, 1996). However, our findings stand 112 HAMILTON AND RAJARAM in sharp contrast to the report of a positive observed on their implicit conceptual tasks of imagery effect in an implicit general knowledge word association with strong associates, cate- test reported by Blaxton (1989) and observed in gory verification, and abstract/concrete classifi- Experiment 1 in the present series. Combined cation (i.e., memory measured by performance together, these findings are problematic for two latencies for classifying words as concrete or influential theories of memory, the transfer ap- abstract). propriate processing framework (Roediger, et Therefore, both studies illustrate that implicit al., 1989) and the dual-code hypothesis (Paivio, conceptual tasks may vary in their processing 1971, 1991). requirements. Although this is an important and Before we consider the theoretical implica- necessary consideration, this reasoning is not tions and possible theoretical resolutions of sufficient to account for the discrepancies in the these findings, it is important to reiterate the data we have discussed here. In Weldon and significance of the specific conceptual implicit Coyote’s (1996) and McDermott and Roedi- memory tasks used across these studies. In re- ger’s (1996) studies, no picture superiority ef- cent reports, researchers have speculated on the fect was obtained on the implicit tasks of word possibility that different conceptual implicit association or category production. In our tasks may vary subtly in their processing re- study, we failed to find a concreteness effect on quirements (Cabeza, 1994; Vaidya, Gabrieli, a different conceptual implicit test, that of gen- Keane, Monti, Gutierrez-Rivas, & Zarella, eral knowledge. We took care to select this task 1997; Weldon & Coyote, 1996). For example, because Blaxton (1989) originally reported a Cabeza (1994) demonstrated a dissociation be- positive effect of a related variable, i.e., the tween two implicit conceptual tasks, category imagery effect, on the implicit general knowl- association and free association (similar to the edge test and we have since replicated this find- word association task previously discussed), ing with this task. with two conceptual encoding conditions, gen- In this context, another important point needs erating category labels and generating word as- to be made regarding the implicit general sociations. Cabeza found that when participants knowledge task. Vaidya et al. (1997) have ar- encoded items by generating category labels, gued that the implicit general knowledge test is priming was better on the category association not a good measure of implicit memory because task compared to the word association task. The this task may be contaminated by explicit mem- opposite pattern of priming was observed for ory strategies. Our data do not support this the word association encoding condition. claim. We conducted both an explicit and an Cabeza’s findings suggest that differences may implicit version of the general knowledge task, exist both between the two types of conceptual and together, the findings support Blaxton’s processing at encoding and between the concep- (1989) original claim that the general knowl- tual processing used on the two implicit con- edge test can tap implicit memory. We found a ceptual tests. Note that his findings also support significant levels of processing effect on both the transfer appropriate processing framework the implicit and explicit versions of the general in that the more similar the processing overlap knowledge tests. Yet, the implicit version dis- was between study and test, the better perfor- sociated from the explicit version such that only mance was at test. the explicit version of the task produced a con- Vaidya et al. (1997) have also demonstrated creteness effect. Taken together, our findings that a conceptual manipulation at study (levels provide strong support for the proposal that the of processing) can differentially affect various general knowledge test is a useful measure of implicit conceptual tests. Specifically, a typical conceptual implicit memory. levels of processing effect was found on their In sum, our failure to find the concreteness implicit conceptual task of category association effect cannot be attributed to the selection of the and word association with weak associates. particular task. Instead, a broader and deeper However, no levels of processing effect was theoretical analysis is necessary to account for CONCRETENESS EFFECT 113 our results in conjunction with the results of the conceptual processing afforded by the referen- other studies cited here. The discrepancy be- tial processing but rather from the visual dis- tween finding a positive imagery effect on im- tinctiveness afforded by the activation of the plicit conceptual tasks and not finding either the imaginal code secondary to the referential pro- picture superiority effect or the concreteness cessing. effect raises questions about the underlying ba- Others have also considered some variations sis of these three effects. Specifically, two as- of this possibility. For example, Weldon and pects of this issue deserve consideration. One, Coyote (1996) have argued that the picture su- we need to consider whether the assumption periority effect in explicit memory arises from that these three effects are based on conceptual the visual distinctiveness of pictures relative to processing is a valid one. Two, if the conceptual words. This interpretation is similar to the ac- basis of these effects does seem to be a reason- count of the picture superiority effect in Nel- able assumption, then we need to consider son’s sensory-semantic model (Nelson, 1979; whether the underlying conceptual processing Nelson, Reed, & McEvoy, 1977). According to varies quantitatively or qualitatively across Nelson’s model, the perceptual representation these three effects. of pictures involves more unique and complex The assumption that the picture superiority visual features compared to words that are effect, the imagery effect, and the concreteness based on only 26 letters of the alphabet. This effect arise from conceptual processing was difference, according to Nelson’s model, gives based on the following logic. In the dual-code rise to the picture superiority effect. Weldon hypothesis, it is assumed that all three effects and Coyote (1996) have suggested that a picture arise from the activation of verbal as well as superiority effect emerges on explicit memory imaginal codes. This dual activation results tests because intentional efforts to distinguish from cross-system or referential processing between studied and nonstudied information which in turn is assumed to be conceptual in benefit from the distinctive aspects of studied nature. However, an alternate mechanism may information. However, the implicit tasks make also explain these three effects in explicit mem- no reference to the past and are completed with ory. These three effects may indeed arise from the first response that comes to mind and, there- cross-system processing but not because of the fore, distinctive information does not have a conceptual nature of such referential processing particular advantage relative to other informa- but because cross-system processing activates tion on implicit tests of memory (see also Mul- the imaginal code. Because the imaginal code is ligan, 1996). Weldon and Coyote’s (1996) fail- assumed to be a “mnemonically superior” code ure to find a picture superiority effect on in the dual-code hypothesis (Paivio & Csapo, implicit conceptual memory tasks confirm this 1973), it is the superiority of this code per se prediction. Thus, there is considerable support (subsequent to the cross-system processing) that for the possibility that at least the picture supe- gives rise to the picture superiority effect, the riority effect is not conceptually based. imagery effect, and the concreteness effect in A distinctiveness hypothesis has also been memory. A quick clarification is needed here considered to explain the concreteness effect with respect to the nature of the “mnemonic (Marschark & Hunt, 1989). Specifically, ac- superiority” of the imaginal code. Although cording to this claim the imagery associated Paivio considered the image code to be a mne- with concrete items causes them to be more monically superior code, it is not clear from distinct and it is this distinctiveness that drives Paivio’s description whether mnemonic superi- the concreteness effect. This assumption pre- ority arises from conceptual or perceptual prop- dicts better memory for concrete items over erties of the stimulus. The reasoning presented abstract items in explicit tasks. Although no here assumes that mnemonic superiority refers predictions were made in this context for per- to perceptually distinctive information. Thus, formance on implicit conceptual memory tests, the three said effects may arise not from the straightforward predictions can now be derived 114 HAMILTON AND RAJARAM based on Mulligan’s (1996) and Weldon and all three effects (i.e., picture superiority, con- Coyote’s (1996) ideas that perceptually distinc- creteness, and imagery) are the result of visual tive information enhances explicit memory but or perceptual distinctiveness. Instead, we pro- not implicit memory. Specifically, if the con- pose that these three effects can be explained by creteness effect arises from more distinctive combining the effects of visual distinctiveness encoding of concrete items (as suggested by of the imaginal code with the operation of con- Marschark & Hunt, 1989) and implicit concep- ceptual processes entailed by referential pro- tual tests are not sensitive to distinctive encod- cessing in the processing of concrete words and ing, then a concreteness effect should be ob- imaged items. tained on the explicit general knowledge test but Our proposal is derived from Paivio’s dual- not on the implicit version of this test. Our data code hypothesis in which both referential pro- confirm this prediction. We found a typical con- cessing as well as mnemonic superiority of the creteness effect on our explicit general knowl- imaginal code are considered to be the bases for edge test and we failed to find this effect on the the three effects under consideration. However, implicit version of this task. This “distinctive- we argue that the data from the conceptual ness” explanation is not in opposition to implicit memory tasks systematically specify Paivio’s dual-code hypothesis because, as noted the conditions under which referential processes earlier, the dual-code hypothesis assumes that versus the superiority of the imaginal code may the pictorial or imaginal code is a more superior differentially aid memory performance and the mnemonic code and could, on its own, produce mechanism by which each of the three effects better memory. To the extent that cross-system emerge. Specifically, we agree with Weldon and processing activates the pictorial or imaginal Coyote’s (1996) analysis that the picture supe- code, both the picture superiority effect and the riority effect results from the activation of the concreteness effect in explicit memory could be more visually distinctive (or “mnemonically su- the result of distinctiveness or the superiority of perior”) imaginal code. This proposal also gains the pictorial code. For this very reason, these support from the fact that in the picture/word two effects may fail to occur on the implicit manipulation, participants are presented with versions of these tasks. pictures that can directly contact the imaginal Note, however, that this explanation is not code (see also Nelson, 1979) without requiring sufficient to account for the imagery effect we extensive referential processing. In the case of obtained on the implicit general knowledge task processing of imaged items and concrete items, in Experiment 1 and others have reported in we argue that the conceptual or referential pro- prior research (Blaxton, 1989; Nyberg & Nils- cessing plays a larger role. One possibility as to son, 1995; Wippich & Mecklenbrauker, 1995). why imaging enhances conceptual priming but Specifically, if Marschark and Hunt (1989) concrete words do not may be that referential claim that imagery makes concrete items more processing is greater in magnitude during imag- distinct, then it is reasonable to assume that ing than during the encoding of concrete words. imaged items should also be more distinctive This possibility is supported by Paivio’s (1991) than read words. Furthermore, if conceptual im- suggestion that concrete words may contact plicit memory tests are not sensitive to this type imagens to a lesser extent than either pictures or of distinctiveness, then we should not find an the imagery instructions. For example, in ex- imagery effect on these tasks. plicit memory tasks, memory for concrete Thus, the imagery effect in conceptual im- words is found to be greater than that for ab- plicit tasks cannot be easily explained by the stract words (Paivio & Csapo, 1969, see also visual or perceptual distinctiveness interpreta- Paivio, 1983 for a review). In turn, explicit tion without some additional considerations. memory for imaged words is found to be greater Despite this complication with the interpreta- than for concrete words (see Paivio, 1983 for a tion of the imagery effect, it is not necessary to review). Finally, explicit memory for imaged entirely abandon the possible interpretation that words is equivalent to that for named pictures CONCRETENESS EFFECT 115

(Paivio & Csapo, 1973). Thus, even though the 1993; Roediger et al., 1989; Weldon, 1991; concreteness effect is mediated by imagery in Weldon & Roediger, 1987). However, as Mc- the dual-code hypothesis and, therefore, must Dermott and Roediger (1996) have pointed out, entail referential processing (which is concep- the same notion has not been widely utilized for tual in nature), it is reasonable to propose that conceptual implicit memory tests and processes. the referential processing involved in encoding Our findings in conjunction with those on the concrete words is less than that involved in picture/word manipulation and imagery manip- imagery. In sum, the extent of conceptual pro- ulation support the notion of such a conceptual cessing involved for different items may vary processing continuum. along a continuum such that abstract words, The theoretical analysis presented thus far concrete words, and imagery span the contin- focuses on possible quantitative differences in uum from lesser to greater degree of conceptual the conceptual processes because an explana- processing. tion based on differences in the amount of con- On this reasoning, although conceptual prim- ceptual processing fit the data on effects of ing is likely to be greater for imaged words than imagery and concreteness quite nicely. How- concrete words (as in Blaxton, 1989 and Exper- ever, it is important to bear in mind that con- iment 1 in the current article), enhanced con- ceptual processes can also differ qualitatively. ceptual priming should occur for concrete Within our own study we found that both the words relative to abstract words as well. That implicit general knowledge test and the explicit this “weaker” conceptual processing during en- fragment cued recall test were sensitive to one coding was not picked up on a conceptual im- type of conceptual processing (the levels of plicit task suggests that the dual-code hypothe- processing manipulation) but not another (the sis should be modified to explicitly include the concrete/abstract manipulation). Although the notion of a referential processing continuum research presented in this article does not defin- with respect to explicit as well as implicit mem- itively demonstrate whether the conceptual ory tasks. analyses engendered by these two manipula- A similar idea of a processing continuum has tions differ qualitatively or quantitatively, this already been adopted within the transfer appro- pattern of results does point to the potential priate processing framework. Roediger and col- differences in the sensitivity of tasks to qualita- leagues (McDermott & Roediger, 1996; Roedi- tive differences in meaning. Other recent re- ger & Blaxton, 1987; Roediger & McDermott, search earlier described in our article has also 1993; Roediger et al., 1989; Weldon, 1991; demonstrated that possible qualitative differ- Weldon & Roediger, 1987) have emphasized ences in conceptual encoding produce differen- that conceptual and perceptual properties of tial effects on conceptual memory tasks (Ca- tasks are not dichotomies but should be viewed beza, 1994; Vaidya et al., 1997). Thus, our as endpoints on a continuum. Therefore, these findings and those of others suggest that con- researchers suggest that there may be degrees or ceptual processing requirements may vary qual- gradations of “conceptual” or “perceptual” pro- itatively not only across tasks (as in Cabeza, cessing. In fact two separate continua may be 1994; Vaidya et al., 1997), but also across dif- necessary, i.e., a perceptual and a conceptual ferent encoding manipulations. Together, these continuum (Weldon, 1991) to account for mem- patterns of data point to the varieties of meaning ory performance, where each continuum may that guide cognitive, and in particular, mne- have its own endpoints of “least” and “most.” monic operations. A fuller exploration of these This idea has been previously used to explain ideas may prove to be a fruitful topic for future some of the inconsistent findings in perceptual research. tests and has become an accepted modification of the transfer appropriate processing frame- CONCLUSION work (McDermott & Roediger, 1996; Roediger We conclude by noting that, until recently, & Blaxton, 1987; Roediger & McDermott, our understanding of the relationship between 116 HAMILTON AND RAJARAM pictures, imagery, and concrete words was cessing: A framework for memory research. Journal of largely based on findings from a variety of Verbal Learning and Verbal Behavior, 11, 671–684. explicit memory tasks. However, the findings Durgunoglu, A. Y., & Roediger, H. L. (1987). Test differ- ences in accessing bilingual memory. Journal of Mem- from conceptual priming tasks in our study as ory and Language, 26, 377–391. well as in some previous studies (Blaxton, Graf, P., & Schacter, D. L. (1985). Implicit and explicit 1989; Roediger & McDermott, 1996, Weldon & memory for new associations in normal and amnesic Coyote, 1996) necessitate important modifica- subjects. Journal of Experimental Psychology: Learn- tions in our conceptualization of how pictures, ing, Memory, and Cognition, 11, 501–518. Hamann, S. B. (1990). Level of processing effects in con- images, and concrete words are related. Consis- ceptually driven implicit task. Journal of Experimental tent with the proposals of the dual-code hypoth- Psychology: Learning, Memory, and Cognition, 16, esis (Paivio, 1991) these findings support the 970–977. view that imaging requires stronger conceptual Madigan, S. (1983). Picture memory. In J. C. Yuille (Ed.), processing than processing of concrete words. Imagery, memory, and cognition: Essays in honor of With respect to the processing of pictures, we Allan Paivio (pp. 65–89). Hillsdale, NJ: Erlbaum. Marschark, M., & Hunt, R. R. (1989). A reexamination of agree with the views of Weldon and Coyote the role of imagery in learning and memory. Journal of (1996, also see Nelson, 1979) that the picture Experimental Psychology: Learning, Memory, and superiority effect arises largely from the visual Cognition, 15, 710–720. distinctive properties of the pictures rather than McDermott, K. B., & Roediger, H. L. (1996). Exact and from enhanced conceptual processing. With re- conceptual repetition dissociate conceptual memory tests: Problems for transfer appropriate processing the- spect to the processing of concrete words, find- ory. Canadian Journal of Experimental Psychology, ings from the present series of experiments 50, 57–71. clearly indicate that conceptual processing of Mulligan, N. W. (1996). The effects of perceptual interfer- concrete words is weaker than that found in ence at encoding on implicit memory, explicit mem- imaging the words. As we noted earlier, these ory, and memory for source. Journal of Experimental Psychology: Learning, Memory, and Cognition, 22, specifications are not entirely inconsistent with 1067–1087. the current versions of the dual-code hypothe- Nelson, D. L. (1979). Remembering pictures and words: sis. However, unlike the dual-code view, we Appearance, significance, and name. In L. S. Cermak have explicitly emphasized the role of distinc- & F. I. M. Craik (Eds.), Levels of processing in human tiveness in picture processing and the notion of memory (pp. 45–76). Hillsdale, NJ: Erlbaum. a continuum of conceptual processing to ex- Nelson, D. L., Reed, V. S., & McEvoy, C. L. (1977). Learning to order pictures and words: A model of plain the effect of imagery and concrete words sensory and semantic encoding. Journal of Experimen- on memory. tal Psychology: Human Learning and Memory, 3, 485– 497. REFERENCES Nelson, D. L., & Schreiber, T. A. (1992). Word concrete- Aron, A., & Aron, E. (1999). Statistics for psychology (2nd ness and word structure as independent determinants of ed.). New York: Prentice Hall. recall. Journal of Memory and Language, 31, 237–260. Blaxton, T. A. (1989). Investigating dissociations among Nicolas, S. (1995). The picture-superiority effect in cate- memory measure: Support for a transfer—appropriate gory-association tests. Psychological Research, 58, processing framework. Journal of Experimental Psy- 218–224. chology: Learning, Memory, and Cognition, 15, 657– Nyberg, L., & Nilsson, L. (1995). The role of enactment in 668. implicit and explicit memory. Psychological Research, Cabeza, R. (1994). A dissociation between two implicit 57, 215–219. conceptual tests supports the distinction between types Paivio, A. (1971). Imagery and verbal processes. New of conceptual processing. Psychonomic Bulletin & Re- York: Holt, Rinehart, Winston. view, 1, 505–508. Paivio, A. (1978). The relationship between verbal and Challis, B. H., & Sidhu, S. (1993). Dissociative effect of perceptual codes. In E. C. Caterette & M. P. Friedman massed repetition on implicit and explicit measures of (Eds.), Handbook of perception. Vol. IX: Perceptual memory. Journal of Experimental Psychology: Learn- processing. New York: Academic Press. ing, Memory, and Cognition, 19, 115–127. Paivio, A. (1983). The empirical case for dual coding. In Cohen, J. (1988). Statistical power analysis for the behav- J. C. Yuille (Ed.), Imagery, memory and cognition: ioral sciences (2nd ed.). Hillsdale, NJ: Erlbaum. Essays in honor of Allan Paivio (pp. 307–332). Hills- Craik, F. I. M., & Lockhart, R. S. (1972). Levels of pro- dale, NJ: Erlbaum Associates. CONCRETENESS EFFECT 117

Paivio, A. (1991). Dual coding theory: Retrospect and cur- Schneider, W. (1990). Micro-Experimental Laboratory rent status. Canadian Journal of Psychology, 45, 255– [Computer program]. Pittsburgh, PA: Psychology Soft- 287. ware Tools, Inc. Paivio, A., & Csapo, K. (1973). Picture superiority in free Shepard, R. N. (1967). Recognition memory for words, recall: Imagery or dual coding? Cognitive Psychology, sentences, and pictures. Journal of Verbal Learning 5, 176–206. and Verbal Behavior, 6, 156–163. Paivio, A., & Csapo, K. (1969). Concrete image and verbal Srinivas, K., & Roediger, H. L., III (1990). Classifying memory codes. Journal of Experimental Psychology, implicit memory tests: Category association and ana- 80, 279 –285. gram solution. Journal of Memory and Language, 29, Paivio, A., Rogers, T. B., & Symthe, P. C. (1968). Why are 389–412. pictures easier to recall than words? Psychonomic Sci- Thorndike, E. L., & Lorge, I. (1944). The teachers’ hand- ence, 11, 137–138. book of 30,000 words. New York: Columbia Univ. Paivio, A., Yuille, J. C., & Madigan, S. A. (1968). Con- Press. creteness, imagery and meaningfulness values for 925 Vaidya, C. J., Gabrieli, J. D. E., Keane, M. M., Monti, L. A., nouns. Journal of Experimental Psychology, 76, 1–25. Gutierrez-Rivas, H., & Zarella, M. M. (1997). Evi- Roediger, H. L., & Blaxton, T. A. (1987). Effects of varying dence for multiple mechanisms of conceptual priming modalities, surface features, and retention interval on on implicit memory tests. Journal of Experimental priming in word fragment completion. Memory and Psychology: Learning, Memory, and Cognition, 23, Cognition, 15, 379–388. 1324–1343. Roediger, H. L., III., & McDermott, K. B. (1993). Implicit Weldon, M. S. (1991). Mechanisms underlying priming on memory in normal human subjects. In H. Spinnler & F. perceptual tests. Journal of Experimental Psychology: Boller (Eds.), Handbook of neuropsychology (Vol. 8, Learning, Memory, and Cognition, 17, 526–541. pp. 63–131). Amsterdam: Elsevier. Weldon, M. S., & Coyote, K. C. (1996). Failure to find the Roediger, H. L., Weldon, M. S., & Challis, B. H. (1989). picture superiority effect in implicit conceptual mem- Explaining dissociations between implicit and explicit ory tests. Journal of Experimental Psychology: Learn- measures of retention: A processing account. In H. L. ing, Memory, and Cognition, 22, 670–686. Roediger & F. I. M. Craik (Eds.), Varieties of memory Weldon, M. S., & Roediger, H. L. (1987). Altering retrieval and consciousness: Essays in honour of Endel Tulving demands reverses the picture superiority effect, Mem- (pp. 3–14). Hillsdale, NJ: Erlbaum. ory & Cognition, 15, 269–280. Roediger, H. L., III, Weldon, M. S., Stadler, M. L., & Weldon, M. S., Roediger, H. L., Beitel, D. A., & Johnston, Riegler, G. L. (1992). Direct comparison of two im- T. R. (1995). Perceptual and conceptual process in plicit memory tests: Word fragment completion and implicit and explicit tests with picture fragment and word stem completion. Journal of Experimental Psy- word fragment cues. Journal of Memory and Lan- chology: Learning, Memory, and Cognition, 18, 1251– guage, 34, 268–285. 1269. Wippich, W., & Mecklenbrauker, S. (1995). Implicit mem- Schacter, D. L., Bowers, J., & Booker, J. (1989). Intention, ory for textual materials. Psychological Research, 57, awareness, and implicit memory: The retrieval inten- 131–141. tionality criterion. In S. Lewandowsky, J. C. Dunn, & K. Kirsner (Eds.), Implicit memory: Theoretical issues (Received August 18, 1999) (pp. 47–55). Hillsdale, NJ: Erlbaum. (Revision received June 27, 2000)