Memory & Cognition 1997,25 (2), 129-139

The verbal overshadowing effect: Why descriptions impair face recognition

CHAD S. DODSON and MARCIA K. JOHNSON Princeton University, Princeton, New Jersey and JONATHAN W. SCHOOLER University ofPittsburgh, Pittsburgh, Pennsylvania

Three experiments explored the verbal overshadowing effect, that is, the phenomenon that de­ scribing a previously seen face impairs recognition of this face. There were three main results: First, a verbal overshadowing effect was obtained both when subjects were provided with and when they gen­ erated a description of an earlier seen face. Second, instructing subjects at the time of test to be aware of potentially competing memories did not improve, and may even have worsened, recognition per­ formance when the subjects had generated a description of the targetface. However,these instructions improved performance and eliminated the verbal overshadowing effect when subjects were provided with someone else's description of the target face. Third, recognition of the target face was disrupted when subjects described a completely different face, such as their parent's face or a face of the oppo­ site sex. The results are discussed in relation to two potential mechanisms: source confusion between previously encoded visual and verbal representations of the face and a shift in processing of the test faces at recognition.

How well we remember an event depends in part on the quired during the slide sequence (Johnson, Hashtroudi, criteria we use to evaluate the information that comes to & Lindsay, 1993). mind (e.g, Dodson & Johnson, 1993; Hasher & Griffin, Although the misinformation effect is robust when 1978; Jacoby, Kelley, Brown, & Jasechko, 1989; Lindsay standard recognition tests are used, other studies have & Johnson, 1989; Raye, Johnson, & Taylor, 1980). The found fewer source confusions when subjects were given way we are oriented to examine our memory can affect the a source-monitoring test that oriented them to assess their kind and amount ofmemorial information that we use to memory in terms of the possible sources of the learned make a memoryjudgment. For instance, in the misleading­ information (Lindsay & Johnson, 1989; Zaragoza & Kosh­ suggestion paradigm (e.g., Lindsay & Johnson, 1989; mider, 1989; Zaragoza & Lane, 1994). That is, for each Loftus, Miller, & Burns, 1978; McCloskey & Zaragoza, item on the test, subjects needed to indicate whether it had 1985), subjects may view a slide sequence and then re­ been seen, read, both seen and read, or was new, Lindsay ceive an ostensibly accurate written description ofthe seen and Johnson (1989) suggested that criterial differences material. But, the description contains misleading infor­ between the recognition and source-monitoring tests might mation, introducing new items or referring to peripheral explain the differences found in the probability ofmak­ items that were seen in the slides as something else. On ing source confusions. Most recognition tests query sub­ a final recognition test for the seen information, subjects jects only about whether the target item was in the slide often respond on the basis ofthe misleading information, sequence, and the questions may be answered according claiming to have seen items that were only read about. In to the item's degree offamiliarity; the most familiar items essence, subjects confuse the sources ofthe information, are marked as having been seen in the slides. Incontrast, mistakenly reporting that the misinformation was ac- most source tests orient subjects to use different, perhaps stricter, criteria and to evaluate memory in terms of all the possible sources for the target item (e.g., was it read, seen, both read and seen, or new?). Accurate judgments This research was funded in part by a National Science Foundation ofthis sort cannot be based only on familiarity, but must graduate fellowship to the first author, Grants AG09253 and AG09744 rely on more specific memorial information (e.g., how from the National Institute on Aging to the second author, and a grant from the National Institute on Mental Health to the third author. For much and what type of perceptual detail and cognitive their helpful comments, we would like to thank Barb Chalfonte and operations did it include?) (see, e.g., Johnson et aI., 1993). other members ofthe Princeton Memory Lab: Jennifer Bonilla, Allison Another technique that seems to direct subjects away from Hermann, Scott Nolde, John Reeder, Sarah Smith, and Christine relying solely on familiarity is to warn them before they Voegeli. We would also like to thank Steve Lindsay, GeoffLoftus, and receive the standard recognition test that the postevent Mike McCloskey for their helpful comments. Correspondence should be directed to C. S. Dodson, Department of Psychology, University of narrative was inaccurate (e.g., Christiaansen & Ochalek, California, Berkeley, CA 94720 (e-mail: [email protected]). 1983), Inthis case, the misinformation effect diminishes,

129 Copyright 1997 Psychonomic Society, Inc, 130 DODSON, JOHNSON, AND SCHOOLER presumably because the warning induces subjects to dis­ cate that the visual representation is still available to and tinguish between their memory for what was read in the relied upon by the verbalization subjects under the proper narrative and their memory for what was seen in the slides. test conditions. Other paradigms have also demonstrated that subjects These results suggest that even under nonspeeded con­ sometimes fail to distinguish between memories of vi­ ditions, the number offalse recognitions might be reduced sual and verbal information under certain encoding and if subjects could be induced to examine their memories test conditions (e.g., Durso & Johnson, 1980; Intraub & more carefully, as in the misleading-suggestion and visuall Hoffman, 1992). For instance, in a study by Intraub and verbal scene paradigms, assuming that similar processes Hoffman, subjects initially viewed a sequence of unre­ account for the errors in these different paradigms (e.g., lated slides showing various ordinary scenes, such as a Dodson & Johnson, 1993; Lindsay & Johnson, 1989). That boy and his father at the zoo. The subjects then read a se­ is, when subjects describe a previously seen face, they ries ofparagraphs describing some ofthe scenes they had may imagine the face while they describe it, just as sub­ seen in the slides and some scenes that were new to them. jects may incidentally imagine misleading information On a final test, the subjects were presented with the pre­ or described scenes. During the recognition test, subjects viously experienced items and the novel items and were may confuse their memory ofhow they imagined the face asked to identify the source of each item. Intraub and with their memory of seeing the face, and rely on this Hoffman found that subjects often confused descriptions memory ofthe imagined face (a type offailure in reality of scenes that they had only read about with scenes that monitoring; Johnson & Raye, 1981). Presumably errors they had previously seen as pictures; they often claimed occur because the memory ofthe imagined face is less ac­ to have seen in picture form items that they had actually curate than the visual memory and thus provides less only read about. Evidently, when subjects read the para­ guidance during the test for choosing the target face. Ifso, graph they incidentally imaged the scene that was de­ the verbalization subjects might be more accurate, that is, picted, and then mistakenly interpreted their memory of fewer source confusions should occur, when they are ori­ this imaginal information as indicating that the item had ented during the recognition test to be aware ofthese sep­ actually been seen. However, Dodson and Johnson (1993) arate memory representations and are encouraged to rely showed that fewer source confusions occurred in this par­ on their more accurate visual representation ofthe face. adigm when subjects were given a source-monitoring Experiment 1 was designed to examine this source­ test that explicitly oriented them to examine their mem­ confusion explanation ofthe verbal overshadowing effect. ory for an item in terms ofall the possible sources. We used a video showing the target face and a corre­ Schooler and his colleagues (Fall shore & Schooler, sponding test slide that had been used by Schooler and 1995; Schooler & Engstler-Schooler, 1990; Schooler, Engstler-Schooler (1990, Studies 1, 2, 4, and 5) to dem­ Ryan, & Reder, 1996) have reported a phenomenon that, onstrate the verbal overshadowing effect. We included on the face ofit, appears similar to the false recognitions the verballself-paced-test and the control/self-paced-test obtained in the paradigms used by Loftus et al. (1978) conditions in order to replicate the impairment effect. We and Intraub and Hoffman (1992). Schooler and Engstler­ also included verballspeeded-test and control/speeded-test Schooler (1990) initially showed subjects a video of a conditions in which subjects had a limited amount oftime bank robbery, and subsequently directed halfofthe sub­ to make a recognition decision. Schooler and Engstler­ jects to describe the robber and the other half (the con­ Schooler (1990, Study 6) found improved performance trol subjects) to complete a filler activity. Schooler and in the verbal/speeded-test condition when an acquisition Engstler-Schooler found that subjects were worse at rec­ face was initially presented as a static slide. Ofinterest to ognizing the previously seen robber from a line-up ofsim­ us was whether the advantage of speeded responding ilar looking faces when the subjects had described his face would also occur when a face had been studied under than when they had not done so. Schooler and Engstler­ more natural, dynamic viewing conditions, such as during Schooler argued that the control subjects, who had not a video presentation. The few studies that have system­ described the face, based their recognition decisions on atically compared memory for faces presented via slides their visual memory of the face, whereas the verbaliza­ or videos have found no effect on subsequent recognition tion subjects were biased to rely on their memory ofhow performance (Bruce, 1994; Bruce & Valentine, 1988). they had described the face. Their verbal memory over­ Therefore, we expected to replicate Schooler and Engstler­ shadowed their visual memory. Apparently, the subjects Schooler's finding in the speeded-test condition. who described the face performed poorly when they re­ In addition, there were two manipulations that were in­ lied on their verbally biased representation, because the tended to make subjects who had described the face aware face had been described in only a general way and pro­ oftheir separate visual and verbal memory representations vided a weak basis for discriminating between the simi­ ofthe target face. In the verballignore condition, subjects lar looking faces at the test. However, Schooler and were told that these two memory representations might Engstler-Schooler found that the verbal overshadowing differ in how accurately they captured the face and that effect disappeared when subjects were forced to make a they should ignore their memories of what they wrote recognition decision within 5 sec ofseeing the test faces. and base their recognition decisions on what they saw. The results from the speeded-response condition indi- Christiaansen and Ochalek (1983) used a variation ofthis THE VERBAL OVERSHADOWING EFFECT 131

ignore manipulation in a study involving the misleading­ Before viewing the test slide, subjects in the verbal/ignore con­ suggestion paradigm, and found that subjects were more dition were instructed: accurate when they received instructions to ignore in­ We're interested in how awarepeople are of the differences between formation presented after the initial target event than their memoryfor seeinga faceand theirmemoryfor havingdescribed a face. It is rarethat thesetwomemories are identical, especially since when they did not. In another group (verbal/two-choice), people's descriptions rarely match(in all ways) what they described. we told subjects that we were interested in learning how When people see the original face again, these differencesbetween aware people were of differences between their visual theirinitialmemoryforthe faceandtheirmemoryfortheirdescription memories ofa face and their memories oftheir verbal de­ sometimes become apparent. In completing the followingtask, you should ignore your memoryfor how you describedthe face and only scriptions ofthe face. Accordingly, we asked subjects to rely on your memory for seeing the face. choose two faces during the recognition test: one that matched their visual memories and one that matched These subjects then saw the slide and were asked to put a "V" in the box that corresponded to their memory ofthe face of the robber. their verbal descriptions. As a control group for this last Before viewing the slide, the subjects in the verballtwo-choice condition, we asked another group ofsubjects, who had condition were told: not described the face, to choose two faces: one that We're interested in how awarepeople are of the differences between matched their memories ofthe robber's face and one that their memoryfor seeinga faceandtheirmemoryfor havingdescribed was the best alternative. On the basis of prior findings this face. It is rare that these two memoriesare identical,especially from other paradigms (Dodson & Johnson, 1993; Lind­ since people's descriptions rarely match (in all ways) what they de­ scribed. Whenpeopleseetheoriginalfaceagain,thesedifferences be­ say & Johnson, 1989; Zaragoza & Lane, 1994), we ex­ tweentheir initial memory for the face and their memoryfor theirde­ pected that both verbal/ignore and verbal/two-choice scriptionbecomeapparent. would encourage subjects to assess the properties oftheir These subjects were then asked to put a "V" in the box that memories carefully and thereby reduce the verbal over­ matched their memory ofthe face they had seen in the video and to shadowing effect. put a "0" in the box that matched their written descriptions. Sub­ jects in the control/two-choice condition were told to put a "V" in EXPERIMENT 1 the box that corresponded to their memory ofthe face ofthe robber and to put a "V2" in the box that represented the best alternative face that corresponded to their memory of the robber. In both of Method these two-choice conditions, recognition performance was com­ Subjects. The subjects were 140 male and female volunteers puted from only the "V" responses. who were Princeton students; each was paid for his/her participa­ All subjects were given the option ofsaying that the face was not tion. The subjects were assigned randomly to conditions, with 20 present. After choosing the box or boxes, depending on the test in­ subjects in each. structions, all subjects were then asked to indicate their confidence Procedure. All subjects viewed a 30-sec videotape depicting a in their choices on a scale from I (guessing) to 9 (certain). In the robber entering a bank, walking toward the camera, and demanding two-choice conditions, the subjects were directed to assess their "Just follow the instructions; don't press the alarm and you won't confidence in their "V" choices. get hurt." This was the same videotape segment that was used by Schooler and Engstler-Schooler (1990). All subjects then partici­ pated in a 20-min filler activity that consisted ofcompleting reading Results comprehension problems from the Graduate Record Examination Unless otherwise noted, the results were significant at (Educational Testing Service, 1989). Afterwards, all subjects were p< .05. assigned randomly to one ofseven conditions. In four ofthese con­ Table 1 presents both the mean percentage correct ditions (verbalization conditions), the subjects were encouraged to recognition and combined recognition and confidence spend 5 min writing a description of the previously seen robber. They were told to "try to use the full five minutes, describing in as scores for all conditions. This combination score, which much detail as possible the facial features ofthis robber." In the re­ has a range of 1 to 6, reflects the confidence and accu­ maining three conditions (control conditions), the subjects spent racy ofsubjects and penalizes subjects who express con­ 5 min completing more reading comprehension problems. Finally, fidence in inaccurate choices. The combined score was all subjects were shown a slide containing eight similar looking faces, including the face ofthe previously seen robber. They also received a sheet of paper with instructions appropriate for this condition. Table 1 There were eight boxes on the sheet that corresponded to the eight Recognition and Recognition Weighted by faces on the slide. Each subject was asked to put a "V" in the box Confidence Scores for Experiment 1 that represented the face that they had seen in the video. A more de­ Percentage Recognition tailed explanation ofeach condition follows. Correct X Confidence In the control/self-paced and the verbal/self-paced conditions, sub­ Recognition Scores SE jects were asked to choose the face that corresponded to their mem­ Self-paced test ory of the face ofthe robber. They were told to put a "V" in the box Control 80 4.95 .39 that matched their memory of it. Verbal 65 4.20 .51 In the control/speeded-test and verbal/speeded-test conditions, Speeded test the subjects were told before seeing the test slide that they would Control 70 4.30 .38 "have only five seconds to review the faces and to indicate if one of Verbal 35 3.20 .37 the faces matches" their memory of the face of the robber. It was Two-choice emphasized that they would seethe slide for only 5 sec and that they Control 65 4.25 .51 should decide immediately after viewing the slide. They were in­ Verbal 55 3.85 .49 structed to put a "V" in the box that corresponded to their memory Ignore of the robber. Verbal 45 3.30 .45 132 DODSON, JOHNSON, AND SCHOOLER computed in the following manner: When subjects made when they had earlier described the face than when they an incorrect choice and indicated the following level of had not described the face; (2) we did not replicate confidence, score = I (confidence is 7, 8, 9), score = 2 Schooler and Engstler-Schooler's (1990) finding that the (confidence is 4, 5, 6), and score = 3 (confidence is I, 2, verbal overshadowing effect is reduced under speeded­ 3); when subjects made a correct choice, score = 4 (con­ test conditions; and (3) contrary to our prediction, we did fidence is I, 2, 3), score = 5 (confidence is 4, 5, 6), not reduce the verbal overshadowing effect when we di­ score = 6 (confidence is 7, 8, 9). Although the combi­ rected subjects to be aware oftheir potentially competing nation score and the proportion correct recognition show visual and verbal representations ofthe target face (i.e., the same general pattern, as seen in Table I, group dif­ with the two-choice and ignore test conditions). In short, ferences were more apparent, statistically, with the com­ the verbal overshadowing effect persisted under speeded­ bination score, and therefore that measure was used in response conditions and under conditions designed to the following analyses. make subjects consider more carefully the information The first analysis examined two questions: First, does that was presumably driving their recognition judgments. describing a previously seen face impair subsequent rec­ Our failure to replicate Schooler and Engstler-School­ ognition ofthis face (the verbal overshadowing effect)? er's (1990) finding in the speeded-test conditions is not And second, does the verbal overshadowing effect dis­ idiosyncratic to our experiment. Read and Schooler (1994) appear under speeded-response conditions relative to non­ also found a verbal overshadowing effect under speeded­ speeded conditions? That is, with a static slide, Schooler test conditions when subjects saw a video at study. There and Engstler-Schooler (1990) found the verbal overshad­ are a number of reasons why subjects might have been owing effect in the self-paced conditions (.80 in the con­ affected differently by speeding up the test, depending trol condition and .50 in the verbalize condition), and they on whether they had seen a slide (Schooler & Engstler­ found no difference in recognition performance between Schooler, 1990) or a video (our Experiment I; Read & these two conditions when the test was completed under Schooler, 1994) at study. Static versus dynamic encoding speeded conditions (.76 in the control condition and .73 might result in qualitatively different underlying mem­ in the verbalize condition). A 2 X 2 analysis ofvariance ory representations. The memory representation for a (ANOVA) was conducted on the first four conditions in static face might be less complex than the memory rep­ Table I and yielded main effects ofcontrol versus verbal resentation for a face derived from dynamic information. [F(1,76) = 4.97, MSe = 3.44] and self-paced versus Accessing the simpler representation might be easier or speeded test [F(l,76) = 3.96,p < .051] and no interaction. more direct under speeded conditions than accessing the Subjects who had described the face were less accurate more complex representation. Alternatively, variations than subjects who had not described the face, and sub­ in the match between study and test conditions (i.e., jects were less accurate when they had to make speeded study slide and test slide or study video and test slide) recognition decisions than when their decisions were might influence the impact ofspecific retrieval manipu­ self-paced. Although we replicated the verbal overshad­ lations, such as speeded responding. That is, according owing effect in the self-paced conditions, we found no to the idea ofencoding specificity or transfer-appropri­ evidence that the verbal overshadowing effect was re­ ate processing (Morris, Bransford, & Franks, 1977; Tul­ duced under speeded as compared with self-paced test ving, 1983), accessing the visual representation should conditions. As seen in Table 1, our failure to replicate was be easier and perhaps less susceptible to verbal interfer­ not due to floor or ceiling effects (Loftus, 1978). Ifany­ ence when the study and test conditions are similar (i.e., thing, we found a larger verbal overshadowing effect in Schooler & Engstler-Schooler's use of a slide at study the speeded than in the self-paced conditions. and test) than when the test conditions are different (i.e., The second analysis tested our prediction that encour­ Read & Schooler's and our use ofa video at study and a aging subjects to examine their memories more carefully slide at test). While it is not possible to determine pre­ would reduce the verbal overshadowing effect. Contrary cisely why slide versus video may differentially interact to our expectations, subjects in the verbal/ignore condi­ with verbalization, we suspect that the mode ofpresen­ tion performed significantly less accurately than subjects tation is important in mediating verbal overshadowing in the control/self-paced condition [F(l,133) = 6.83, effects. MSe = 3.99] and worse, but not significantly so, than sub­ Our failure to reduce the verbal overshadowing effect jects in the verbal/self-paced condition [F(l, 133) = 2.03]. with the verbal/ignore and verbal/two-choice conditions Similarly, subjects in the verbal/two-choice condition was intriguing, because, in other similar paradigms, per­ were less accurate than subjects in the control/self-paced formance was better when subjects were oriented toward condition [F(I,133) = 3.04,p < .10], although they were examining their memory in terms ofpossible sources than not significantly worse than subjects in the control/two­ when subjects received tests that oriented them toward choice condition [F(l,133) = .40]. relying on overall familiarity (Dodson & Johnson, 1993; Lindsay & Johnson, 1989;Zaragoza & Lane, 1994). In the Discussion misleading-suggestion paradigm, for instance, subjects There were three main results to this study: (I) We rep­ who have received misleading postevent information per­ licated the verbal overshadowing effect-s-that is, the form worse than control subjects because they appear to finding that subjects are worse at recognizing a target face confuse their memories ofinformation from the slides and THE VERBAL OVERSHADOWING EFFECT 133 the narrative and incorrectly rely on the narrative informa­ faces are processed, then instructions to ignore the spe­ tion. There are fewer source confusions in the misleading­ cific contents of their verbalizations may fail to address suggestion paradigm when subjects are specifically di­ the source of subjects' difficulty, thereby failing to pre­ rected to be aware ofthe slides and the narrative as possible vent the verbal interference. In fact, orienting subjects to sources for a memory. assess their memories ofthe face in terms ofits possible Despite the similarities between the verbal overshadow­ sources (i.e., their having seen or described the face) may ing and the misleading-suggestion paradigms, specific have encouraged them to use a featural strategy during the confusion between previously encoded representations test and, consequently, worsened recognition perfor­ may not account for the errors in the verbal overshadow­ mance. From a source-monitoring point ofview, careful ing paradigm. Instead, an alternative explanation of the examination ofmemories should lead to better source ac­ verbal overshadowing effect is that writing a description curacy, but only if individuals are assessing the "right" of a previously seen face induces a change in the way qualities for the test. subjects examine the test faces (Schooler, 1989). The act of verbalization may induce a featural orientation at the EXPERIMENT 2 time oftest that is independent ofthe quality ofthe sub­ ject's description of the face. This featural orientation In Experiment 2, two mechanisms for a verbal overshad­ may take the form ofa process-of-elimination strategy in owing effect were examined: (1) source confusion be­ which particular faces are excl uded on the basis ofan ex­ tween competing memory representations may diminish amination of certain features, such as the shape of the recognition of the target face, and (2) recognition per­ nose or eyes. Subjects may rely on verbalizable features formance may be impaired by a holistic-to-featural shift at the expense ofnot considering nonverbal information. in the way the test faces are processed. Ifsource confusion The control subjects, on the other hand, may rely on a between two memory representations is a factor, subjects more holistic strategy during the test in which their mem­ should show better face recognition performance (i.e., a ory of the target face is globally matched with the alter­ reduced impairment effect) in conditions that presum­ native test faces. There is evidence that recognition per­ ably reduce the similarity between the target and poten­ formance is worse with a featural than with a holistic tially competing representations. strategy (e.g., Carey & Diamond, 1977; Diamond & Carey, Much research has shown that memories containing 1986; Dunning & Stern, 1994; Rhodes, Brake, Taylor, & similar characteristics are more likely to be confused with Tan, 1989; Stern & Dunning, 1994). Carey and Diamond each other than are memories containing different char­ showed that face recognition can involve global and fea­ acteristics (e.g., Dodson & Johnson, 1996; Ferguson, tural information, but that recognition performance is Hashtroudi, & Johnson, 1992; Foley, Johnson, & Raye, usually better when subjects rely on global rather than 1983; Johnson, Foley, & Leach, 1988; Lindsay & Johnson, on featural information (Carey & Diamond, 1977; Dia­ 1991; Lindsay, Johnson, & Kwon, 1991). Therefore, to mond & Carey, 1986). Similarly, people's ability to rec­ increase the distinctiveness between the original visual ognize faces is worse after judging the physical features memory and the subsequent verbal memory, we included (e.g., how large is the nose) than after judging the traits a condition in which subjects were provided with a de­ (e.g., honest?) ofa face (e.g., Baddeley, 1979; Bower & scription ofthe previously seen face as opposed to their Karlin, 1974; Wells & Hryciw, 1984; Wells & Turtle, generating a description ofthe face. Recently, Zaragoza 1987; Winograd, 1976). Presumably, judging traits in­ and Lane (1994) found that a similar manipulation af­ duces holistic processing of test faces. fected source-monitoring performance in the eyewitness Research suggests that featural processing oftest faces testimony paradigm. In their study, the subjects were may be closely linked to earlier verbalization of the tar­ more likely to remember the source ofthe misleading in­ get face. For example, Wells and Turtle (1987) found that formation (i.e., there were fewer source confusions) on a although trait encoding leads to superior face recogni­ source test when they had acquired this information by tion, featural encoding leads to superior verbal descrip­ reading a description as opposed to generating answers tions, as determined by independent raters. Ifverbaliza­ to questions. Zaragoza and Lane reasoned that the rep­ tion is associated with featural processing, and iffeatural resentation ofthe misleading information was more sim­ processing leads to poorer recognition performance, then ilar to the representation ofthe original visual event when it seems reasonable that verbalization might impair per­ subjects generated answers to questions than when they formance by encouraging a featural strategy at the time read a narrative, consequently causing more source con­ oftest. Consistent with this view, Fallshore and Schooler fusion. Therefore, we expected that, in our study, when (1995) found that the detrimental effects ofverbalization subjects read another's description of the target face, were lessened by two factors known to promote featural they should either experience less confusion, even in the processing at recognition: other race recognition (cf. absence ofthe ignore instructions, or be better able to ig­ Rhodes, Brake, Taylor, & Tan, 1989) and inversion (Di­ nore the competing representation once they had been amond & Carey, 1986). made aware ofits potential for confusion. If the effect of verbalization is not simply to cause A verbal overshadowing effect might also be a conse­ subjects to think about their verbal descriptions during quence ofa shift toward more featural and less global pro­ recognition but more generally to shift the way the test cessing induced by thinking about facial features. On 134 DODSON, JOHNSON, AND SCHOOLER

this view,reading someone else's description ofthe target 3 min to write a description ofthe previously seen face. They were face should produce a verbal overshadowing effect sim­ told to "write a detailed description of the person in the picture so ilarto that found in the standard verbalization (i.e., gener­ that someone reading the description would be able to identify the ate a description) condition. According to the processing­ person." They then took the recognition test. After the subjects in the verbal/ignore condition described the face shift explanation, it is unclear how the instructions to for 3 min, they were given the following instructions: ignore the generated or read description should affect We're interested in how awarepeopleare of the differences between performance. theirmemoryfor seeinga face andtheirmemoryfor having described Finally, a way to evaluate the relative contributions of thisface.Whenpeopleseetheoriginal faceagain,thesedifferences be­ these two mechanisms comes from the results ofa con­ tween their initial memoryfor the faceand their memory for their de­ scription sometimes become apparent. In completing the following dition in which subjects were instructed to describe a com­ task,you should ignoreyourmemoryfor how you describedthe face pletely different face (say, their parent's) after viewing andonly rely on your memory for seeing the face. the target face. Ifthe confusion is largely representation Each subject in the provide and provide/ignore conditions was centered, this describe-parent condition should not affect yoked with a subject in the corresponding verbal or verbal/ignore performance, because a parent's face is presumably very condition so that the provided descriptions were created by other dissimilar from the target's face and therefore should subjects. The subjects in the two provide conditions were told that produce little, ifany, interference. However, ifthe verbal "for the next 3 minutes you should read the following description, overshadowing effect is primarily caused by a processing made by somebody else, of the photo you just saw." After reading shift, then the describe-parent condition should impair the description, the subjects in the provide condition received the recognition of the target face, because the act of de­ recognition test. Before receiving the recognition test, the subjects in the provide/ scribing a parent's face, as in describing any face, tends ignore condition were given the following "ignore" instructions: to focus on features. The similarity ofthe face to the tar­ We're interested in how awarepeopleare of the differences between get face should not matter according to this explanation. theirmemoryfor seeinga faceandtheirmemoryfor havingreada de­ In sum, in addition to the standard control condition, scription oftbis face.Whenpeopleseetheoriginalfaceagain,thesedif­ some subjects either generated (i.e., the verbalize condi­ ferences between theirinitialmemoryforthe faceandtheirmemoryfor their description sometimes becomeapparent. In completingthe fol­ tion) or were provided with (i.e., the provide condition) lowing task, you shouldignoreyourmemoryfor howthe facewasde­ a description of the previously seen face. The descrip­ scribedand only rely on your memory for seeing the face. tions that were provided to subjects were actual descrip­ The subjects in the describe-parent condition were instructed to tions ofthe target face that had been generated by other spend 3 min writing "a detailed description of the face of one of subjects. Some subjects in both ofthese conditions, the your parents so that someone reading the description would be able generated and the provided, received the "ignore" in­ to identify your parent." They then took the recognition test. structions that were used in Experiment 1. We also di­ The recognition test for all subjects consisted ofsix similar look­ rected some subjects to describe their parent's face after ing faces, including a different photograph ofthe target face, and a viewing the target face. Finally, instead ofusing the video sheet ofpaper with six boxes corresponding to the six faces. Each subject was instructed to put a "V" in the box that represented the from Experiment 1, we showed subjects a photograph of face that they initially saw. They were told to choose the person that the target face. best resembled or was the person they initially saw. In contrast to the previous experiment, the "not present" option was not available. Method After making their choices, all subjects were asked to rate their con­ Subjects. There were 240 male and female Princeton students fidence in their choices on a scale from 1 (guessing) to 9 (certain). or visiting students in this study, with 40 subjects in each condition. Each subject was either paid or given course credit. Results and Discussion Materials. There were two different target faces with corre­ Table 2 presents the mean proportion correct recogni­ sponding recognition-test line-up slides. The photographs of the target faces had similar dimensions and were taken from the candid tion and the combined recognition and confidence scores and formal sections of college yearbooks. The photograph (ap­ for all groups. This combination score, which has a range proximately 4.5 em wide and 5.2 em high) presented at acquisition of 1 to 6, was determined in the same way as in Experi­ was a candid shot of each target face and was mounted on a card ment 1. There are 40 combination scores for each group, that was 15 em wide and 10 em high. In the recognition test, a for­ mal, professionally posed photograph of the target face was pre­ sented alongside five similar looking distractor photographs. Each Table 2 of the two sets of six test photographs was mounted on a separate Recognition and Recognition Weighted by card that was 18 em wide and 12.5 em high. Each of the six test Confidence Scores for Experiment 2 photographs was roughly 4 em wide and 5 cm high. The descrip­ Percentage Recognition tions provided to some subjects were generated by other matched Correct X Confidence subjects who were directed to describe the target face. Recognition Scores SE Procedure. Halfofthe subjects saw one target face; the remain­ Control 65 4.38 .35 ing subjects saw the other target face. The subjects initially viewed Describe parent 53 3.63 .34 a black-and-white photograph ofthe appropriate target for 5 sec and Verbal then read an unrelated essay for 5 min. The subjects were then ran­ Self-paced 53 3.72 .33 domly assigned to one ofsix conditions. The subjects in the control Ignore 38 3.23 .30 condition spent 3 min attempting to recall and write down the Provide names of the U.S. states and their capitals before taking the final Self-paced 45 3.48 .31 recognition test. The subjects in the verbalize condition were given Ignore 63 4.15 .30 THE VERBAL OVERSHADOWING EFFECT 135 except the verbalize condition, which has 39 scores be­ who were provided with these descriptions [r = .08; cause 1 subject failed to provide a confidence score. F( 1,78) < 1]. These results are consistent with similar re­ Analyses for the combined score are reported below. Be­ sults from Schooler and Engstler-Schooler (1990) and Fall­ cause the pattern was similar for both faces, the follow­ shore and Schooler (1995) and suggest that subjects were ing analyses are collapsed across this variable. not basing a recognition decision on their memory of the As stated earlier, ifverbal overshadowing stems from specific features that were mentioned in the description. confusion between representations, then performance In summary, there were three important findings in should improve when the visual memory representation Experiment 2. (1) Subjects were less accurate at recog­ is on average less similar to the verbal memory repre­ nizing the target face when they had described their par­ sentation (i.e., by describing a parent's face or by read­ ent's face (i.e., describe-parent group) than when they ing another's description of the target face) and when had not described a face (i.e., the control condition). subjects are oriented to notice differences between these (2) Reading a description ofa face can later impair face two representations (i.e., with the ignore instructions). recognition. And (3) the ignore manipulation was effec­ However, if neither ofthese manipulations improves per­ tive at improving recognition accuracy when subjects formance, it seems safe to assume that confusion between had been provided with descriptions of the previously representations does not contribute to the verbal over­ seen face but not when subjects had generated their own shadowing effect in this situation. Instead, a shift from pro­ descriptions of the target face. The fact that subjects were cessing the faces in a global manner to processing them not helped by the ignore instructions when they had in a feature by feature manner is a likely explanation. generated a description replicates the results from To examine whether generating a description of any Experiment 1. face impairs performance, we compared the control con­ Overall, these results provide evidence for two mech­ dition with the combined verbalize and describe-parent anisms that evidently can produce a verbal overshadowing conditions, and there was a verbal overshadowing effect effect. First, the finding that the verbal overshadowing ef­ [F(1,233) = 3.18, MSe = 4.13,p < .10]. As is clear from fect disappeared when subjects were directed to ignore Table 2, performance was no better when subjects de­ their memory of someone else's description suggests that scribed a parent than when they described the target face. inappropriate representations can be a source of con­ The following analyses examined two questions: fusion and that source monitoring can be improved with (1) Does the verbal overshadowing effect disappear when test instructions that tighten subjects' criteria while mak­ subjects read someone else's description of the target face? ing memory judgments (e.g., Dodson & Johnson, 1993; (2) What is the effect ofbeing warned to ignore the mem­ Lindsay & Johnson, 1989). Second, two other findings ory of the verbal material? An impairment effect does suggest that confusion from inappropriate representa­ not depend on the act of generating a description ofthe tions is not the only mechanism by which verbal descrip­ target face. The subjects in the provide condition were tions hurt recognition. If generated descriptions also less accurate than the subjects in the control condition produced recognition deficits from confusion between [F(I,233) = 3.90, MSe = 4.13]. As in Experiment 1, sub­ different representations, we would expect subjects to jects were not helped when they were instructed to ignore profit from ignore instructions (as did subjects who read their own verbal descriptions ofthe face: subjects in the provided descriptions) and to show more accurate recog­ verbal/ignore condition were less accurate than subjects nition ofthe target face when the generated descriptions in the control condition [F(1,233) = 6.37] and slightly were less similar to the target face. However,performance worse, but not significantly so, than subjects in the ver­ was not improved when subjects were directed to ignore balize condition [F(1,233) = 1.17]. memories oftheir own generated descriptions ofthe tar­ To investigate our prediction that the ignore instruc­ get face or when they described a parent's face rather tions are more likely to improve performance when sub­ than the target face. These results suggest that generating jects are provided with a description ofthe target face than verbal descriptions disrupts face recognition by shifting when they generate this description, we performed a 2 the overall strategy subjects use when examining the al­ (ignore/no ignore) X 2 (generate/provide) ANOVAon the ternative faces at test. In the General Discussion, we fur­ verbal self-paced and ignore and the provide self-paced ther discuss these two mechanisms and why they might and ignore conditions in Table 2. There were no reliable differentially come into play when verbal descriptions are main effects, but the predicted interaction was marginally read and generated. significant [F(1,155) = 3.53, MSe = 3.85, P < .07]. The ability to recognize the target face was not improved when EXPERIMENT 3 subjects were asked to ignore descriptions that they had generated (i.e., verbal/ignore condition) relative to sub­ The results ofExperiment 2 suggest that the negative jects in the standard verbal condition. The ignore in­ effects of verbalization on face recognition can occur structions, however, did improve recognition ofthe target even when it is not the target face that is described. This face when subjects had been provided with descriptions finding is consistent with the idea that, regardless of the ofthe target face. Lastly, there was no reliable correlation specific content ofthe verbally derived memory, gener­ between the performance of the subjects who generated ating a verbal description induces a shift from holistic/ a description of the target face and the yoked subjects visual processing ofthe test faces to featural/verbal pro- 136 DODSON, JOHNSON, AND SCHOOLER cessing. Although the impairment effect in Experiment 2 Table 3 of describing a parent's face was comparable to that of Recognition Rates for the Male and Female Faces describing the target face, the difference between this in Experiment 3 condition and control performance was only marginally Test Describe Describe Face Control Male Female significant. Because of the importance ofthis compari­ son for understanding the disruptive effects of verbal­ Male 57% 34% 37% Female 70% 51% 48% ization on recognition, a third experiment was conducted that used more subjects and focused exclusively on this critical comparison. In addition, in order to ensure that or the female face substantially impaired both recogni­ the task of describing the nontarget face was comparable tion of the male face [X2(2) = 8.14, P < .01] and recog­ to the task of describing the target face, we provided sub­ nition ofthe female face [X2(2) = 7.00, P < .01]. There jects with two unfamiliar faces (one male and one fe­ was no significant difference between the effects ofde­ male) and subsequently asked them to describe only one scribing the male and female face for recognition ofei­ ofthese faces (control subjects described neither face). ther the maleface [i.e., 34% vs. 37%; x2(l ) = .13] or the This paradigm thus enabled us to determine, for each female face [i.e., 51% vs. 48%; x2(l) = .14]. verbalization subject, the effect of verbalizing an unfa­ Verbalization impaired recognition ofthe nondescribed miliar face on recognizing both the described face and face as much as it did that of the described face. This out­ the nondescribed face. Ifthe specific content ofa verbal come, along with the negative effect ofdescribing a par­ description mediates the verbal overshadowing effect ent's face in Experiment 2, suggests that describing any (e.g., by establishing a representation that can potentially face will impair recognition of a target face. These results be confused with the target memory), then verbalization suggest that describing a face causes some generalized should affect recognition only of the face that was de­ shift in processing at recognition. Subjects in the control scribed. If, however,verbalization produces a more global group may rely on more global information during the test, shift in processing that is not specifically associated with whereas the verbalize subjects may rely on more featural the particular content ofthe verbalization, then verbaliz­ information (e.g., the shape of the eyebrows and lips). ing one face should impair recognition of both the de­ scribed face and the face that was not described. GENERAL DISCUSSION

Method Consistentwith previous research,the results from these Subjects. One-hundred and eighty-seven subjects from the Uni­ studies showed that face recognition suffers when sub­ versity of Pittsburgh participated in this experiment; they were as­ jects generate a description of a face they have seen (e.g., signed randomly to conditions. There were three conditions with 62 subjects in the describe-male condition, 62 in the describe-female Schooler & Engstler-Schooler, 1990). In addition, there condition, and 63 in the control condition. were three new findings: (1) The ability to recognize the Materials. The photographed male and female faces used in this target face was impaired not only when subjects had de­ study were two of the three used in Schooler and Engstler-Schooler's scribed the target face, but also when they had described (1990) Experiment 6. The faces presented at study were taken from a very familiar face-their parent's face-or when they the candid photo section from a university yearbook. The recogni­ had described a previously seen, but novel, face; (2) rec­ tion test included a different photograph of the target face taken from the formal section of the yearbook, alongside five similar ognition performance was disrupted when subjects were looking distractor faces. provided with a description of a previously seen face; Procedure. The subjects viewed, in succession, slides ofa male and (3) subjects were better able to ignore descriptions and a female face. Each slide was presented for 5 sec with a 5-sec that had been provided than those that they had gener­ interstimulus interval. All subjects were then given a 5-min filler ated. The first set ofresults suggests that when descrip­ activity of recalling and naming the states in the United States. Then tions are generated, there is little relationship between the subjects were assigned to one of three conditions: control group the content ofwhat has been described (i.e., whether it is (which continued the filler activity), the describe-male face, and the describe-female face. After 5 min, all subjects were given recog­ the target face, a parent's face, or a novel face) and sub­ nition tests for the male and female faces. The order of the original sequent recognition performance. These findings argue slide presentations and tests was counterbalanced across subjects. against a representation-centered explanation ofthe ver­ bal overshadowing effect that occurs when subjects gen­ Results and Discussion erate descriptions themselves. If subjects were confus­ Confidence measures were not taken with this data ing their representations of the perceived face and the set, so performance is based solely on accuracy. The per­ imagined description, we would expect recognition per­ centage of subjects choosing the correct face in each con­ formance to vary as a function ofthe similarity between dition is shown in Table 3. The effect of verbalization on the two representations (e.g., Johnson et al., 1988; also the verbalized and nonverbalized faces was assessed sep­ see Crowder, 1976, for a general discussion of the rela­ arately for the male and female faces in order to avoid vi­ tionship between similarity, interference, and forgetting). olating the assumption of independent observations re­ According to this view, we should have expected im­ quired by a chi-square test. As can be seen in Table 3, proved recognition of the target face when subjects had compared with the performance of subjects who did not described a very different face, such as their parent's face engage in verbalization, verbalization ofeither the male or a face ofthe opposite sex. This was not the case, how- THE VERBAL OVERSHADOWING EFFECT 137 ever. Apparently, the magnitude of the recognition­ scribed or read about a face: (l) A processing shift may be impairment effect is not affected by the degree of simi­ the most important mechanism of poor face recognition larity between the target and subsequent faces when sub­ when subjects have generated a description ofa face, and jects generate a description. It remains to be explored (2) confusion between representations from different how similarity affects recognition performance when sources may be the most important mechanism when sub­ subjects hear or read descriptions given by others. jects read a description ofa face. Generating descriptions Thus, as Schooler and his colleagues have suggested, offaces should invoke an analytic activity insofar as par­ verbal overshadowing may sometimes result from a shift ticular features of the face are identified and described. in the cognitive strategies that are applied at recognition This analytic set presumably carries over to the test and (e.g., Fallshore & Schooler, 1995; Schooler, 1989; shifts the strategy subjects use to examine the test faces Schooler et al., 1996). Unfortunately, the present find­ to a featural one. The featural strategy is not affected by ings do not reveal the precise nature ofthis apparent pro­ the ignore instructions, perhaps because these instruc­ cessing shift. One likely possibility, however, is that ver­ tions say nothing about how the test faces should be ex­ balization orients subjects to rely on verbalizable feature amined. Instead, the instructions orient subjects to dis­ information, which is less useful than the nonverbaliz­ criminate between potentially confusable representations able, holistic/configural information that control sub­ (i.e., subjects are instructed at test to ignore the verbal rep­ jects presumably rely on to make a recognition judgment resentation and to rely on the visual representation). Per­ (see Carey & Diamond, 1977; Diamond & Carey, 1986; haps recognition performance would improve ifsubjects Wells & Hryciw, 1984; Wells & Turtle, 1987). For in­ were instructed to use a different test strategy, such as a stance, Fallshore and Schooler (1995) provided evidence more holistic one. that verbalization influences individuals' relative reliance A different situation occurs when subjects read a de­ on featural versus holistic/configural information at rec­ scription of the face. In this case, subjects may not ana­ ognition. They showed that the detrimental effects of lyze so much as attempt to construct holistic images or verbalization were minimized under conditions designed impressions ofthe described faces. Because comprehend­ to hinder the use ofconfigural information (and encour­ ing a description promotes a holistic impression of the age the use offeatural information), such as the recogni­ face, these subjects maintain a holistic test strategy. How­ tion ofother-race faces (Rhodes, Brake, et al., 1989) and ever, constructed representations that are derived from the recognition ofinverted faces (e.g., Bartlett & Searcy, reading the description may produce confusion with per­ 1993). Apparently, generating descriptions disrupts recog­ ceptually derived representations from seeing the face and, nition only under conditions in which holistic processing consequently, cause recognition impairment, as in the is an effective default strategy. provide condition. The ignore instructions, however, can Although the verbal overshadowing effect persisted successfully orient subjects to distinguish between these when the similarity between the target face and described two representations, accounting for the improved per­ face was reduced (i.e., by describing either a parent's face formance in the provide/ignore condition. Because there or a different face ofthe opposite sex), another prediction appears to be no reliable relationship between the qual­ that follows from the idea that subjects confuse their vi­ ity ofthe description and recognition performance, these sually and verbally coded memories fared somewhat bet­ constructed images are probably derived from both the ter. We expected that conditions designed to make sub­ information in the description and the subjects' memory jects more aware ofalternative memory representations for the target face. would improve performance, as it does in other paradigms Our explanation of poor face recognition in terms of in which visual and verbal information are confused (e.g., both a change in test strategy (i.e., featural processing vs. Christiaansen & Ochalek, 1983; Dodson & Johnson, 1993; configural/holistic processing) and source confusion be­ Lindsay & Johnson, 1989). Consistent with this expec­ tween different representations (i.e., visual vs. verbal) tation, the verbal overshadowing effect disappeared when fits in a class ofsimilar characterizations ofrecognition subjects were told to ignore a description with which they performance. Dunning and Stern (1994) have suggested had been provided. The improvement in recognition that that accurate and inaccurate face recognition is associ­ is produced by instructions to ignore provided descriptions ated with different recognition strategies. They found parallels similar findings from the misleading-suggestion that subjects who agreed with statements indicating a re­ and visual/verbal scenes paradigms (Dodson & Johnson, liance on nonreportable, "automatic" processes (e.g., "his 1993; Lindsay & Johnson, 1989; Zaragoza & Lane, 1994). face 'popped out' at me") were more accurate than sub­ This suggests that there are similar mechanisms ofsource jects who agreed with statements associated with strate­ confusion in all three paradigms, because source moni­ gic processes (e.g., "I compared the photographs to each toring is improved when subjects are directed to attend to other in order to narrow the choices"). In this context, it aspects of memory that might differentiate memories seems possible that verbalization may increase subjects' from different sources. On the other hand, the verbal over­ reliance on reportable strategic processes (e.g., isolating shadowing effect was not reduced when subjects were told features) at the expense ofnonreportable, but more use­ to ignore descriptions that they had generated themselves. ful, automatic processes (e.g., holistic processing). In We are left with the speculation that two different mech­ addition, recent research by Brandimonte, Schooler, and anisms affect face recognition when subjects have de- Gabbino (in press) suggests that verbalization might also 138 DODSON, JOHNSON, AND SCHOOLER

shift individuals' relative reliance on verbal versus visual cess dissociation approach to recognition memory. Journal ofEx­ memory representations. This possibility is indicated by perimental Psychology: General, 125,181-194. the finding that verbal overshadowing effects in visual­ DUNNING, D., & STERN, L. B. (1994). Distinguishing accurate from in­ accurate eyewitness identifications via inquiries about decision pro­ form memory can be attenuated by reinstating the visual cesses. Journal ofPersonality & Social Psychology, 67, 818-835. context in which the stimuli were encoded. Clearly, ad­ DURSO, E T., & JOHNSON, M. K (1980). The effects oforienting tasks ditional research is necessary to fully isolate the precise on recognition, recall, and modality confusion ofpictures and words. nature ofthe effect ofverbalization. Ultimately, however, Journal ofVerbal Learning & VerbalBehavior, 19,416-429. EDUCATIONAL TESTING SERVICE (1989). Graduate Record Examina­ it seems likely that various alternative distinctions, such tion. Princeton, NJ: Author. as configural/featural (e.g., Carey & Diamond, 1977), FALLSHORE, M., & SCHOOLER, J. W.(1995). The verbal vulnerability of automatic/strategic (e.g., Dunning & Stern, 1994), and perceptual expertise. Journal ofExperimental Psychology: Learning. visual/verbal (e.g., Paivio, 1986), may be useful in help­ Memory, & Cognition, 21,1608-1623. ing to characterize the impact ofverbalization on recog­ FERGUSON, S., HASHTROUDI, S., & JOHNSON, M. K (1992). Age differ­ ences in using source-relevant cues. Psychology & Aging, 7, 443-452. nition performance (see Fallshore & Schooler, 1995). All FOLEY, M. A., JOHNSON, M. K, & RAYE, C. L. (1983). Age-related of these characterizations emphasize that remembering changes in confusion between memories for thoughts and memories is a product not only of what is stored and what is acti­ for speech. Child Development, 54, 51-60. vated at test, but also of how that information is evalu­ HASHER, L., & GRIFFIN, M. (1978). Reconstructive and reproductive processes in memory. Journal ofExperimental Psychology: Human ated (Johnson et al., 1993). Learning & Memory, 4, 318-330. Finally, our results have some practical implications INTRAUB, H., & HOFFMAN, J. E. (1992). Reading and visual memory: for understanding the experience of an eyewitness to a Remembering scenes that were never seen. American Journal ofPsy­ crime. Eyewitnesses are often asked to generate a descrip­ chology, 105,101-114. tion of whom they saw. Eyewitnesses also often hear JACOBY, L. L., KELLEY, C. M., BROWN, J., & JASECHKO, 1. (1989). Be­ coming famous overnight: Limits on the ability to avoid unconscious descriptions of the perpetrator from other eyewitnesses. influences of the past. Journal ofPersonality & Social Psychology, The present studies suggest that both the activity ofgen­ 56, 326-338. erating and the activity oflistening to descriptions ofan JOHNSON, M. K, FOLEY, M. A., & LEACH, K (1988). The consequences earlier event can hurt subsequent face-recognition per­ for memory of imagining in another person's voice. Memory & Cog­ nition, 16, 337-342. formance because oftwo different factors: (1) relying on JOHNSON, M. K, HASHTROUDI, S., & LINDSAY, D. S. (1993). Source ineffective criteria in examining a line-up offaces, such as monitoring. Psychological Bulletin, 114, 3-28. using a featural strategy; and (2) relying on an incorrect or JOHNSON, M. K., & RAYE, C. L. (1981). Reality monitoring. Psycho­ less accurate representation ofthe face for a recognition logical Review, 88, 67-85. judgment. LINDSAY, D. S., & JOHNSON, M. K (1989). Misleading suggestibility and memory for source. Memory & Cognition, 17, 349-358. LINDSAY, D. S., & JOHNSON, M. K (1991). Recognition memory and REFERENCES source monitoring. Bulletin ofthe Psychonomic Society, 29, 203-205. LINDSAY, D. S., JOHNSON, M. K, & KWON, P. (1991). Developmental BADDELEY, A. D. (1979). Applied cognitive and cognitive applied psy­ changes in memory source monitoring. Journal ofExperimental chology:The case offace recognition. In L.-G.Nilsson (Ed.), Perspec­ Child Psychology, 52, 297-318. tives on memory research (pp. 367-388). Hillsdale, NJ: Erlbaum. LOFTUS, E. E, MILLER, D. G., & BURNS, H. J. (1978). Semantic inte­ BARTLETT, J. c., & SEARCY, J. (1993). Inversion and configuration of gration of verbal information into a visual memory. Journal ofEx­ faces. Cognitive Psychology, 25, 281-316. perimental Psychology: Human Learning & Memory, 4, 19-31. BOWER, G. H., & KARLIN, M. B. (1974). Depth of processing pictures LOFTUS, G. R. (1978). On interpretation of interactions. Memory & Cog­ offaces and recognition memory. Journal ofExperimental Psychol­ nition, 6, 312-319. ogy, 103, 751-757. MCCLOSKEY, M., & ZARAGOZA, M. (1985). Misleading postevent in­ BRANDIMONTE, M. A., SCHOOLER, J. W., & GABBINO, P.(in press). At­ formation and memory for events: Arguments and evidence against tenuating verbal overshadowing through visual retrieval cues. Jour­ memory impairment hypotheses. Journal ofExperimental Psychol­ nal ofExperimental Psychology: Learning, Memory. & Cognition. ogy: General, 114, 3-18. BRUCE, V.(1994). Stability from variation: The case offace recognition. MORRIS, C. D., BRANSFORD, J. D., & FRANKS, J. J. (1977). Levels of pro­ Quarterly Journal ofExperimental Psychology, 47A, 5-28. cessing versus transfer appropriate processing. Journal of Verbal BRUCE, v., & VALENTINE, T.(1988). Whena nod's as good as a wink: The Learning & Verbal Behavior, 16, 519-533. role of dynamic information in facial recognition.In M. M. Gruneberg, PAIVIO, A. (1986). Mental representations: A dual coding approach. P.E. Morris, & R. N. Sykes (Eds.), Practical aspects ofmemory: Cur­ New York:Oxford University Press. rent researchand issues (Vol. I, pp, 169-174).Chichester,U.K.:Wiley. RAYE, C. L., JOHNSON, M. K., & TAYLOR, T. H. (1980). Is there something CAREY, S., & DIAMOND, R. (1977). From piecemeal to configurational special about memory for internally generated information? Memory representation offaces. Science, 195, 312-314. & Cognition, 8, 141-148. CHRISTIAANSEN, R. E., & OCHALEK, K (1983). Editing misleading in­ READ, D. J., & SCHOOLER, J. W. (1994, July). Verbalization decrements formation from memory: Evidence for the coexistence of original in long-term person identification. Paper presented at the Practical and postevent information. Memory & Cognition, 11,467-475. Aspects ofMemory Conference, Baltimore County, MD. CROWDER, R. G. (1976). Principles oflearning and memory. Hillsdale, RHODES, G., BRAKE, S., TAYLOR, K., & TAN, S. (1989). Expertise and NJ: Erlbaum. configural coding in face recognition. British Journal ofPsychology, DIAMOND, R., & CAREY, S. (1986). Why faces are and are not special: 80,313-331. An effect of expertise. Journal ofExperimental Psychology: Gen­ SCHOOLER, J. W. (1989, November). Verbalization can impair the non­ eral, 115, 107-117. verbal components ofvisual memories. Paper presented at the meet­ DODSON, C. S., & JOHNSON, M. K. (1993). The rate of false source at­ ing of the Psychonomic Society, Atlanta. tributions depends on how questions are asked. American Journal of SCHOOLER, J. W., & ENGSTLER-SCHOOLER, T. Y. (1990). Verbal over­ Psychology, 106,541-557. shadowingof visual memories: Somethingsare better leftunsaid. Cog­ DODSON, C. S., & JOHNSON, M. K (1996). Some problems with the pro- nitive Psychology, 22, 36-71. THE VERBAL OVERSHADOWING EFFECT 139

SCHOOLER, J. W., RYAN, R. S., & FALLSHORE, M. (1992, November). WELLS, G. L., & TuRTLE,J. W. (1987). What is the best way to encode Verbalizationmay induce excessively strategicface processing. Paper faces? In M. M. Gruneberg, P. E. Morris, & R. N. Sykes (Eds.), Prac­ presented at the Workshop on Memory Strategies: Theory and Appli­ tical aspects ofmemory: Current research & issues (Vol. I, pp. 163­ cation, Trieste. 168). New York: Wiley. SCHOOLER, J. W., RYAN, R. S., & REDER, L. M. (1996). The costs and WINOGRAD, E. (1976). Recognition memory for faces following benefits of verbally rehearsing memory for faces. In D. Herrmann, nine different judgments. Bulletin ofthe Psychonomic Society, 8, M. K. Johnson, C. McEvoy, C. Hertzog, & P.Hertel (Eds.), Basic and 419-421. applied memory: New findings (pp. 51-65). Hillsdale, NJ: Erlbaum. ZARAGOZA, M. S., & KOSHMIDER, J. w., III (1989). Misled subjects may STERN, L. B., & DUNNING, D. (1994). Distinguishing accurate from in­ know more than their performance implies. Journal ofExperimental accurate eyewitness identifications: A reality monitoring approach. Psychology: Learning. Memory, & Cognition, 15,246-255. In D. F. Ross, 1.D. Read, & M. P.Toglia (Eds.), Adult eyewitness tes­ ZARAGOZA, M. S., & LANE, S. M. (1994). Source misattributions and timony: Current trends and developments (pp. 273-299). New York: the suggestibility of eyewitness memory. Journal ofExperimental Springer-Verlag. Psychology: Learning. Memory, & Cognition, 20, 934-945. TuLVING, E. (1983). Elements ofepisodic memory. Oxford: Oxford University Press. WELLS, G. L., & HRYCIW, B. (1984). Memory for faces: Encoding and (Manuscript received March 27, 1995; retrieval operations. Memory & Cognition, 12,338-344. revision accepted for publication March 1,1996.)