Semantic Integration of Verbal Information into a Visual

Journal of Experimental Psychology;Human Learning and Memory1978, Vol. 4, No. 1, 19-31

Semantic Integration of Verbal Information into a Visual Memory

Elizabeth F. LoftusUniversity of Washington

David G. MillerUniversity of Houston

Helen J. BurnsUniversity of Washington

A total of 1,242 subjects, in five experiments plus a pilot study, saw a seriesof slides depicting a single auto-pedestrian accident. The purpose of theseexperiments was to investigate how information supplied after an eventinfluences a witness's memory for that event. Subjects were exposed to eitherconsistent, misleading, or irrelevant information after the accident event.Misleading information produced less accurate responding on both a yes-noand a two-alternative forced-choice recognition test. Further, misleadinginformation had a larger impact if introduced just prior to a final test ratherthan immediately after the initial event. The effects of misleading informa-tion cannot be accounted for by a simple demand-characteristics explanation.Overall, the results suggest that information to which a witness is exposedafter an event, whether that information is consistent or misleading, isintegrated into the witness's memory of the event.

Almost two centuries ago, ImmanuelKant (1781/1887) spoke of the humantendency to merge different experiences toform new concepts and ideas. That ten-dency has crucial implications for one'sability to report his or her experiencesaccurately. When one has witnessed animportant event, such as a crime or anaccident, one is occasionally exposed tosubsequent information that can influencethe memory of that event. This occurs evenwhen the initial event is largely visual andthe additional information is verbal innature (Loftus, 1975; Pezdek, 1977). Forinstance, in a previous study, subjects sawfilms of complex fast-moving events such

This research was supported by grants to thefirst author from the Urban Mass TransportationAdministration and from the National Institute ofMental Health. Special thanks are due to R.Abelson, W. Cole, C. MacLeod, G. Loftus, T.Nelson, E. Tulving, and S. Woods for reading andcommenting on an earlier draft of this article. R.Shiffrin suggested Experiment 2.

Requests for reprints should be sent to ElizabethF. Loftus, Department of Psychology, University ofWashington, Seattle, Washington 98195.

as automobile accidents or classroom dis-ruptions (Loftus, 1975). Immediately after-ward, the subjects were asked a series ofquestions, some of which were designed topresent accurate, consistent information(e.g., suggesting the existence of an objectthat did exist in the scene), while otherspresented misleading information (e.g.,suggesting the existence of an object thatdid not exist in the original scene). Thus,a subject might have been asked, "Howfast was the car going when it ran the stopsign?" when a stop sign actually did exist(Experiment 1). Or the subject might havebeen asked, "How fast was the whitesports car going when it passed the barnwhile traveling along the country road?"when no barn existed (Experiment 3).These subjects were subsequently askedwhether they had seen the presupposedobjects. It was found that such questionsincreased the likelihood that subjectswould later report having seen theseobjects. It was argued that the questionswere effective because they containedinformation—sometimes consistent, some-times misleading—which was integrated

Copyright 1978 by the American Psychological Association, Inc. All rights of reproduction in any form reserved.

19

20 E. LOFTUS, D. MILLER, AND H. BURNS

Figure I. Critical slides used in the acquisition series.

into the memorial representation of theevent, thereby causing a reconstruction oralteration of the actual information storedin memory.

In these earlier experiments, the originalevent was presented visually, the subse-quent information was introduced verballyvia questionnaires, and the final test wasalso verbal in nature. In the presentexperiments, a recognition procedure wasused; it involved showing a series of slidesdepicting a complex event and afterwardexposing subjects to verbal informationabout the event. This study phase wasfollowed by a recognition test in which thesubjects were presented with target picturesidentical to ones seen before and distractorpictures altered in some way. The firstreason for this change was that if onesubscribes to the view that verbal and visualinformation are stored separately, onecould argue that Loftus's (1975) final test,being verbal in nature, helped subjectsaccess the subsequent verbal information,thereby resulting in an incorrect response.

The second reason for using a recognitiontest procedure was that if recognition isassumed to be a relatively passive andsimple process of matching stimuli tospecific locations in a content-addressablestorage system, one would expect a re-presentation of the actual (or true) sceneto result in a match, whereas an alterationwould fail to match. In other words, if theoriginal visual scene is stored in memory,presenting the subject with the originalstimulus might result in a match betweenthe memory representation and the stimu-

lus. If the original scene had been trans-formed so that an altered version was storedin memory, presenting the subject with theoriginal stimulus would not result in amatch between the memorial representa-tion and the stimulus.

These considerations motivated thepresent series of studies. Before turning tothem, we describe a pilot study in somedetail, since the materials and procedureswere similar to those used in the remainingexperiments.

Pilot Experiment

In a pilot experiment (Loftus, Salzberg,Burns, & Sanders, Note 1), a series of 30color slides, depicting successive stages inan auto-pedestrian accident, was shownto 129 subjects. The auto was a red Datsunseen traveling along a side street towardan intersection having a stop sign for halfof the subjects and a yield sign for theremaining subjects. These two criticalslides are shown in Figure 1. The remainingslides show the Datsun turning right andknocking down a pedestrian who is crossingat the crosswalk. Immediately after view-ing the slides, the subjects answered aseries of 20 questions. For half of thesubjects, Question 17 was, "Did anothercar pass the red Datsun while it was stoppedat the stop sign?" The remaining subjectswere asked the same question with thewords "stop sign" replaced by "yieldsign." The assignment of subjects toconditions produced a factorial design inwhich half of the subjects received con-

VERBAL INFORMATION INTO VISUAL MEMORY 21

sistent or correct information, whereas theother half received misleading or incorrectinformation. All subjects then participatedin a 20-min filler activity, which requiredthem to read an unrelated short story andanswer some questions about it. Finally, ayes-no recognition test was administeredeither immediately or 1 week later. Thetwo critical slides (i.e., those containingthe stop and yield signs) were randomlyplaced in the recognition series in differentpositions for different groups of subjects.

The results indicated that relative to thecase in which consistent information isreceived, misleading information resultedin significantly fewer hits (correct recogni-tions of the slide actually seen) and slightlymore false alarms (false recognitions of theslide not actually seen). With misleadinginformation, the percentage of hits was71 and the percentage of false alarms was70, indicating that subjects had zeroability to discriminate the sign theyactually saw from the sign they did not see.

Some aspects of the data from this studypreclude a clear interpretation of theresults and beg for a variation in design.Most of the subjects responded "yes" tothe slide shown first in the recognitionseries, even though the opposite sign hadbeen seen and mentioned in the question-naire. This indicates that the two criticalslides are so similar that subjects failed tomake any distinction between them. Per-haps when the second slide appeared, somesubjects responded "yes" again, thinkingit was the same slide, while others feltobliged to respond "no," having alreadyresponded "yes" to the earlier slide. Forthese reasons, a forced-choice recognitiontest seemed necessary, since it elimmatesthe problem of successive recognition testsand forces the subjects to discriminatebetween the two critical slides.

Overview of the Experiments

In Experiment 1, subjects were presentedwith the acquisition series of slides, anintervening questionnaire, and a finalforced-choice recognition test. It is shownthat misleading information results in

substantially less accurate responding thandoes consistent information. Next, weconsider the possibility that subjects aresimply agreeing with the information intheir questionnaires, fully rememberingwhat they actually saw. Experiment 2 wasactually a demonstration designed to showthat the results thus far cannot be ex-plained simply by the demand charac-teristics of the procedure. In Experiment 3,we asked whether information presentedverbally has a different effect dependingon whether it is introduced immediatelyafter the initial event (i.e., at the beginningof the retention interval) or just prior tothe final test (i.e., at the end of the reten-tion interval). It was found that misleadinginformation has a greater impact whenpresented just prior to a recognition testrather than just after the initial event.Finally, we addressed the question ofwhether the verbally presented informationactually results in a transformation of anexisting representation or whether it issimply a supplementation phenomenon.To answer this issue, one needs to knowwhether the original sign entered memoryin the first place. If not, then the subsequentverbal information may simply introduce asign where none existed, supplementingthe existing memorial representation. Ifthe sign originally did get into memory,the subsequent information has causedeither an alteration in the original repre-sentation (i.e., one sign replaced the otherin memory) or the creation of a new,stronger representation that successfullycompetes with the original one, renderingthe latter so dramatically suppressed as tobe, for all intents and purposes, gone.Experiment 4, in conjunction with Experi-ment 3, indicates that the traffic sign isencoded by most subjects when they viewthe series of slides. Experiment 5 demon-strates the generality of the findings withother materials.

Experiment 1

Method

Subjects were 195 students from the Universityof Washington who participated in groups of various


sizes. With a few exceptions, the procedure wassimilar to that used in the pilot experiment. Thesubjects saw the same series of 30 color slides,seeing each slide for approximately 3 sec. Approxi-mately half of the subjects saw a slide depictinga small red Datsun stopped at a stop sign, whereasthe remaining subjects saw the car stopped at ayield sign. Immediately after viewing the acquisitionslides, the subjects filled out a questionnaire of 20questions. For half of the subjects, Question 17was, "Did another car pass the red Datsun while itwas stopped at the stop sign?" For the other half,the same question was asked with the words "stopsign" replaced with "yield sign." Thus, for 95subjects, the sign mentioned in the question wasthe sign that had actually been seen ; in other words,the question contained consistent information. Forthe remaining 100 subjects, the question containedmisleading information.

After completing the questionnaire, the subjectsparticipated in a 20-min filler activity that requiredthem to read an unrelated short story and answersome questions about it. Finally, a forced-choicerecognition test was administered. Using two slideprojectors, IS pairs of slides were presented, eachpair of slides being projected for approximately 8sec. One member of each pair was old and the otherwas new. For each pair, the subjects were asked toselect the slide that they had seen earlier. Thecritical pair was a slide depicting the red Datsunstopped at a stop sign and a nearly identical slidedepicting the Datsun at a yield sign. The slides thatthe subjects actually saw varied in the left and rightpositions.

Results

The percentage of times a subjectcorrectly selected the slide he or she hadseen before was 75 and 41, respectively,when the intervening question containedconsistent versus misleading information,Z = 4.72, p < .001. If 50% correct selec-tion is taken to represent chance guessingbehavior, subjects given consistent in-formation performed significantly betterthan chance, Z = 5.10, p < .001, whereasthose given misleading information per-formed significantly worse than chance,Z = 1.80, p < .05 (one-tailed test).

Experiment 2

Some time ago, Orne (1962) proposedthat certain aspects of any psychologicalexperiment may provide clues, or demandcharacteristics, that permit observant sub-jects to discern the experimental hypothesis.

Obliging subjects may then try to confirmthat hypothesis. In the context of thepresent paradigm, it is possible that someor all of the subjects not only rememberedwhat traffic sign they observed but alsoremembered what sign was presupposed ontheir questionnaire and then "went along"with what they believed to be the experi-mental hypothesis and chose the sign fromtheir questionnaire. A slightly differentversion of this position would argue thatat the time of the final test, subjects saidto themselves, "I think I saw a stop sign,but my questionnaire said 'yield sign,' so Iguess it must have been a yield sign."Experiment 2 was designed to investigatethis possibility.

Method

The method was similar to that of Experiment 1with a few exceptions. Ninety subjects saw the slideseries. Half of them saw a stop sign, and half ayield sign. Immediately after slides, the subjectsfilled out the questionnaire. For 30 subjects, thecritical question was, "Did another car pass thered Datsun while it was stopped at the intersection?"In other words, it did not mention a sign. For 30other subjects, the critical question mentioned astop sign, and for the remaining 30 it mentioned ayield sign. Thus, for one third of the subjects, thekey question contained a true presupposition; forone third, the presupposition was false; and for theremaining one third, the question made no referenceto a sign at all. A 20-min filler activity occurred,followed by a forced-choice recognition test.

Finally, the subject was given a "debriefingquestionnaire." It stated,

The study in which you have just been involvedwas designed to determine the effects of subse-quent information on eye-witness testimony. Inthe beginning, you saw a series of slides whichdepicted an accident. One of the slides containedeither a stop sign or a yield sign. Later you weregiven a questionnaire. One of the questions onthis questionnaire was worded to assume that youhad seen either a stop sign or a yield sign or else itcontained no information about what kind ofsign you saw.

Please indicate which sign you think you saw andwhat was assumed on your questionnaire.

I SawA stop signA yield sign

My QuestionnaireMentioned

A stop signA yield signNo sign.


Table 1Data from Experiment 2

Incorrect subjects onforced-choice test

Correct subjects onforced-choice test

Information given

ConsistentMisleadingNone

Weighted M

n

91711

% correct ondebriefing

questionnaire

22129

14

re

211319

% correct ondebriefing

questionnaire

52314243

This final debriefing questionnaire permitted asubject to claim, for example, that he or she hadseen a stop sign but that the questionnaire hadmentioned a yield sign. In other words, it gave thesubjects the opportunity to be completely "in-sightful" about their condition in the experiment.

Results

Of the 90 subjects who took the forced-choice recognition test, 53 chose thecorrect sign; 37 chose the incorrect sign.As in the previous experiment, accuracydepended on whether the subject had beengiven consistent, misleading, or no informa-tion on the intervening questionnaire. Thisrelationship can be seen in Table 1.

The subjects who chose the correct signduring the forced-choice test were morethan three times as likely as incorrectsubjects to be completely correct on thedebriefing questionnaire. Overall, 43% ofthe subjects choosing the correct signaccurately responded to the debriefingquestionnaire, whereas only 14% of theincorrect subjects were completely ac-curate, Z = 2.96, p < .01. Again, whetherthe subjects responded accurately to thedebriefing questionnaire depended onwhether they had been given consistent,misleading, or no information on theirintervening questionnaires.

Of central concern was the performanceof subjects who had been given misleadinginformation and who had subsequentlychosen incorrectly on their forced-choicetest. For example, they saw a stop sign,read that it was a yield sign, and subse-quently chose the yield sign on the forced-

choice test. These subjects were the oneswho may have been acting the way theexperimenter wanted them to act. Theymay have been deliberately choosing thesign mentioned on their questionnairealthough fully remembering what theysaw. Yet, when given the debriefingquestionnaire that afforded them theopportunity to say, "I think I saw the stopsign, but my questionnaire said yield,"only 12% did so.

Experiment 3

The issue that motivated Experiment 3was whether the information introducedsubsequent to an event has a differentimpact when it is introduced immediatelyafter the event than when it is introducedjust prior to the final test. To determinethis, we varied the time interval betweenthe initial slides and the final forced-choicetest. The intervening questionnaire waspresented either immediately after theacquisition slides or it was delayed untiljust prior to the final test.

Method

Subjects were 648 students from the Universityof Washington who either participated for coursecredit or were paid for their participation. Theyparticipated in groups of various sizes.

The procedure was nearly identical to that usedin Experiments 1 and 2, with the major variationsbeing the retention interval and the time of theintervening questionnaire. Subjects saw eachacquisition slide for approximately 3 sec. Half sawthe key slide that contained a stop sign, and halfsaw a yield sign. A questionnaire was administered,


Immediate Questionnaire Delayed Questionnaire1.00 r

0 20 1min. day

Retention IntervalType of Information

• ConsistentA None• Misleading

Figure 2. Proportion of correct responses as a function of retention interval displayed separatelyfor subjects given an immediate questionnaire and subjects given a delayed questionnaire inExperiment 3. (The curve parameter is type of information the subject received during theretention interval.)

followed by a forced-choice recognition test. Theforced-choice test occurred after a retention intervalof either 20 min, 1 day, 2 days, or 1 week, with 144subjects tested at each interval. Half of the subjectsat each retention interval answered the question-naire immediately after viewing the acquisitionslides (immediate questionnaire), and the otherhalf answered it just before the final forced-choicetest (delayed questionnaire). In addition, 72subjects saw the slides, received the questionnaireimmediately afterward, and immediately after thatwere given the forced-choice test. For purposes ofanalysis, we consider this group to have been testedat a retention interval of zero.1

Except at the zero retention interval, all subjectsread a short, unrelated "filler" story for 20 min andthen answered some questions about it. Subjectswho were given the immediate questionnaire com-pleted the filler activity after answering the question-naire. Subjects who were given the delayed question-naire completed the filler activity after viewing theacquisition slides.

Question 17 on the questionnaire was the criticalquestion. It mentioned either a stop sign, a yieldsign, or no sign at all. Equal numbers of subjectsreceived each version. Thus, one third of the sub-jects were given consistent information, one thirdwere given misleading information, and one thirdwere given no information at all relevant to atraffic sign.

In the final forced-choice recognition test, subjectswere asked to choose the slide they had seen beforeand give a confidence rating from 1 to 3, where 1indicated the subject was sure of the answer and 3indicated a guess.

Results and Discussion

Proportions of correct responses as afunction of retention interval are displayedseparately for subjects in different condi-tions in Figure 2. The data for subjectstested at a retention interval of zeroappear twice in Figure 2, once underimmediate questionnaire and once underdelayed questionnaire, because the question-naire occurred, by definition, both im-mediately after the slides and just prior tothe final test. In a sense, it was both animmediate and a delayed questionnaire.

Before presenting statistical analyses, weshall point out some major observations.First, for both the immediate and delayed

1 A better design would have orthogonally variedthe two critical intervals, namely, the intervalbetween the slides and the questionnaire and theinterval between the questionnaire and the recogni-tion test. However, such a design would haverequired nearly three times as many subjects toobtain reasonably stable proportions in each cell,and the authors' colleagues were already becomingdistressed at the rapidity with which these experi-ments were depleting the psychology department'ssubject pool. We doubt that any conclusions wouldbe changed as a result of the fuller design.


questionnaire, longer retention intervalsled to worse performance. Type of informa-tion given also had an effect: Relative to acontrol in which subjects were given noinformation, consistent information im-proved their performance and misleadinginformation hindered it. The functionsobtained when no relevant information wasgiven show the usual forgetting over time.By 2 days, subjects were performing atchance level. Immediately after viewingthe slides, however, there was relativelygood memory for them (up to 87% correct).

The first analysis considered only theimmediate-questionnaire data. A 5 (reten-tion intervals) X 3 (types of information)analysis of variance of the arc sine trans-formed proportions was conducted (Mos-teller & Tukey, 1949, p. 189). All F testsreported here are with MSe = .01 andp < .01, unless otherwise indicated. Theanalysis showed that longer retentionintervals led to less accurate performance,F(4t, oo) = 5.67. Further, the type ofinformation to which a subject was exposedaffected accuracy, F(2, ») = 50.19, andthere was an interaction between thesefactors, ^(8, °o) = 5.19. A test for mono-tonic trend for the subjects who were givenconsistent information yielded a significanttrend, F(l, oo) = 10.38. Similarly, thetrend was significant for subjects giveninconsistent and no information, F(l, oo)= 4.43 and F(l, oo) =43.13, respectively.

The second analysis considered the datafrom subjects who received a delayedquestionnaire. A 5 X 3 analysis of varianceof the arc sine transformed proportionsindicated that longer retention intervalsled to less accurate performance, ^(4, oo)= 13.37. Type of information and theinteraction were also significant, ^(2, oo)= 90.91, and ^(8, «) = 2.98, respectively.Again, the monotonic trends for each ofthe three types of information also reachedsignificance: F(l, oo) = 5.92 for subjectsgiven consistent information, 14.05 forinconsistent information, and 35.85 for noinformation (all ps < .05).

Consistent information. Not surpris-ingly, when a subject is exposed to informa-tion that essentially repeats information

previously encoded, recognition perform-ance is enhanced. With an immediatequestionnaire, the visual and verbal repeti-tions are massed, whereas with a delayedquestionnaire, they are spaced. Whereasin most memory tasks, successive repeti-tions affect memory less than do repetitionsthat are spaced apart in time (Hintzman,1976), this outcome was not obtained in thepresent experiment. A popular explanationfor the spacing effect is in terms of volun-tary attention. The subject chooses to payless attention to the second occurrence ofan item when it closely follows the firstoccurrence than he does when the intervalbetween the two is longer. In the presentcase, it appears as if the subject may havepaid more attention to the second occur-rence when it closely followed the first,resulting in memory enhancement that wasable to survive longer retention intervals.

Misleading information. When mislead-ing information occurs immediately afteran event, it has a different effect than whenit is delayed until just prior to the test.The immediate procedure results in anearly monotonically increasing function,whereas the delayed procedure leads to amonotonically decreasing function. Thisresult makes intuitive sense. When falseinformation is introduced immediately afteran event, it has its greatest impact soon.Therefore, when the test was immediate,such subjects performed well below chance.But after an interval of, say, 1 week, boththe event and the misleading informationapparently had faded such that the subjectperformed near chance levels. On the otherhand, when the misleading information wasdelayed, it was able to influence thesubjects' choice more effectively as thedelay increased. Presumably, the weakerthe original trace, the easier it is to alter.

To see more clearly the effects of animmediate versus a delayed questionnaire,we excluded the data for subjects tested at aretention interval of zero and collapsedthe data over the four remaining retentionintervals. The results of these computationsare shown in Figure 3. The proportioncorrect is presented as a function of thetype of information given, with the im-


1.00

.80

8 -60

.40

.20

| Immediate

i delayed

Consistent None Misleading

TVpe of information

Figure 3. Proportion of correct responses for subjectsgiven different types of information in Experiment 3.(Data for subjects given an immediate questionnaireare shown separately from data for those given adelayed questionnaire.)

mediate versus delayed questionnaire datashown separately. It is again evident thatthe delayed questionnaire had a largerimpact than the immediate one when thesubjects were given misleading information:When misleading information was intro-duced immediately after the incident,46% of the subjects were correct; however,when it was delayed until just prior to thefinal test, that percentage dropped to31.5%, Z = 2.06, p < .05.

We should mention here that Doolingand Christiaansen (1977) have found adifferent effect of misleading information.They found that such information had agreater effect on memory distortion whenit occurred before the retention intervalrather than afterward. As these investi-gators rightfully point out, there are somany differences between their experi-mental paradigm and ours that it isdifficult to essay a resolution of the differ-ence in results. Our subsequent manipu-lation focuses on one particular detail of thematerial to be remembered, and a peripheraldetail at that. In Dooling and Christiaan-sen's task, the subsequent informationconsists of the name of a famous personabout whom subjects already have a greatdeal of knowledge stored in memory.Unfortunately, neither they nor we havebeen able to come up with an appealing

hypothesis for why these paradigmaticdifferences should lead to different results.

Surprisingly, it appears that even whenthe questionnaire contained no informationrelevant to the traffic sign, performance onthis key item was somewhat better whensubjects were interrogated immediatelyafter the event rather than later. Althoughthis difference failed to reach significanceby a Z test involving all four retentionintervals, Z = 1.41, .10 < p < .20, it heldup for those retention intervals that,snowed some memory performance abovechance. For the 20-min and 1-day intervals,the immediate questionnaire had about a15% advantage over the delayed. Perhapsthe early questionnaire permitted thesubjects to review the incident in order toanswer questions about it, and in thecourse of this review, some of them re-freshed their memory for the traffic signeven though they were not specificallyqueried on this detail.

Confidence ratings. Recall that subjectsindicated how confident they were in theirresponses, circling "1" if they felt certainand "3" if they were guessing. The rating"2" was used for intermediate levels ofconfidence. Figure 4 illustrates how theseratings varied as a function of the type ofinformation a subject was exposed to, thetiming of that information, and whetherthe response was correct or incorrect.

A 3 X 2 X 2 unweighted-means analysisof variance (Winer, 1962, p. 241) wasperformed on all but the zero retention-interval data. This analysis included the576 subjects who were unambiguouslygiven either an immediate or a delayedquestionnaire. The error for all F tests is.493, and p < .01 unless otherwiseindicated.

Type of information affected confidence,F(2, 564) = 9.15, as did whether the sub-ject responded correctly or incorrectly,F(l, 564) = 23.64; in other words, subjectswere more confident if correct than ifincorrect (1.92 vs. 2.18). The main effect oftiming (whether the questionnaire wasanswered immediately or whether it wasdelayed) was not significant (F < 1). TheResponse Accuracy X Type of Information


(sure)

o>

0>o

Ioo

3

(guess)

Delayed Immediate Delayed li

Correct Incorrect

ConsistentCorrect Incorrect

NoneCorrect Incorrect

Misleading

Type of InformationFigure 4. Mean confidence ratings as a function of type of information given, immediate versusdelayed questionnaire, and correct versus incorrect responses in Experiment 3.

interaction was marginally significant,F(2, 564) = 2.71, .05 < p < .10, while theother two-way interactions were not(Fs < 1). Finally, the triple interactionreached significance, F(2, 564) = 5.01. Itis evident from Figure 4 that a subject'sconfidence is boosted by being told any-thing, whether it is true or not. Further,delaying misleading information raises con-fidence in incorrect responses above thecorresponding value associated with correctresponses.

To summarize the major results, thereappear to be two discernible consequencesof exposing a subject to misleading in-formation. First, the likelihood is loweredthat a subject will correctly recognize theobject previously seen. This is particularlytrue if the information is introduced justprior to the final test. Second, the mislead-ing information affects a subject's confi-dence rating. Generally subjects are moreconfident of their correct responses thantheir incorrect ones. However, when ex-posed to delayed misleading information,they are less confident of their correctresponses.

Experiment 4Loftus (1975) argued that the informa-

tion contained in a questionnaire influences

subsequent choices because that informa-tion is integrated into an existing memorialrepresentation and thereby causes analteration of that representation. This viewassumes that when a person sees the initialevent, the items of interest are actuallyencoded at the time of viewing. In thecontext of the present stimuli, this positionwould hold that when a person sees a stopsign, for example, the sign gets intomemory (i.e., is encoded). If a subsequentquestionnaire reports that the sign was ayield sign, that information might, accord-ing to this view, enter the memory systemand cause an alteration of the originalrepresentation. The subject can now beassumed to have a yield sign incorporatedinto his memorial representation of theevent.

A question arises as to whether the stopsign actually got into memory in the firstplace. If it did not, then the subsequentverbal information may simply be intro-ducing a sign where none existed. In otherwords, the existing memorial representa-tion of the accident is simply supplemented.On the other hand, if the sign was encodedinto memory, then the subsequent informa-tion may have caused what is functionallya transformation of the original representa-tion. Thus, it is theoretically important


(a)

(b)

Figure 5. Diagrams used in Experiment 4.

to determine whether subjects attend toand/or encode the sign. A portion of thedata from Experiment 3 suggests thatpeople do. Notice in Figure 2 that when noinformation is contained in the question-naire, subjects show some ability to dis-criminate the sign they saw from the onethey did not, up to and including a reten-tion interval of 1 day. For these subjects,the sign must have been encoded, otherwiseperformance would have been at chancelevel. Experiment 4 was designed to providea further test of whether subjects encodedthe sign they saw in the acquisition series.

Method

Ninety subjects were shown the same series ofslides described above, each slide for approximately3 sec. Following the series, they were given a sheetof paper with a diagram on it similar to that shownin either Figure 5a or Sb. Forty-five subjectsreceived Diagram Sa, and 45 received Sb. Theinstructions were to fill in as many details as couldbe remembered.

The reason for using two versions of the diagramstems from an observation made during a pilotstudy. Recall that the slides depict a red Datsuntraveling along a side street toward an intersection.From there the car turns right and knocks over apedestrian in the crosswalk. If the diagram containsno sketch of the car (Sa), the subjects tend toconcentrate their attention on details at the cross-walk, which is where the accident took place. Theymay have seen the sign at the corner, but do notdraw it, since it does not seem important to theaccident. What is needed is a way to focus theirattention on the intersection, and the placing of acar near the intersection as in Figure Sb appearedto be a way of accomplishing this. The experimentlasted less than 10 min.

Results

For purposes of analysis, we counted ascorrect the drawing made by any subjectwho either drew the sign he had seen orwrote its name. Over all, 45% of thesubjects indicated the correct sign. Of thosesubjects given the outline without a car(Figure 5a), 36% correctly drew the stopsign, while 32% correctly drew the yieldsign. Of those subjects given the outlinewith a car (Figure 5b), 60% correctlydrew the stop sign, while 52% correctlydrew the yield sign. An analysis of varianceof the arc sine transformed proportionsindicated that more subjects depicted asign when a car was used to direct theirattention to the intersection (Figure 5b)than when the diagram contained nocar (Sa), F(l, «) = 19.94, MSe = .01.Whether the subject had actually seen astop sign or a yield sign did not significantlyaffect the likelihood of drawing the correctsign, F(l, oo) = 1.51, MSe = .01, p < .20.The interaction also failed to reach signifi-cance (F < 1). Three of the 50 subjectswho saw a yield sign incorrectly drew astop sign in their diagram; none of the"stop sign" subjects drew a yield sign.

The results indicate that when subjectsview the particular series of slides usedthroughout these experiments, at leasthalf of them (and perhaps more) doencode the correct sign. The data fromsubjects given Diagram 5b (with a carto focus their attention on the intersection)indicate that over half have encoded thesign to the point of including it in theirdiagrams. Others may also have encodedit, but this was not revealed by the presentprocedure.

Experiment 5

The purpose of Experiment 5 was todemonstrate the generality of our studiesbeyond the single-stimulus pair used in theprevious studies.

Method

A new series of 20 color slides depicting an auto-pedestrian accident was shown to 80 subjects. A


male pedestrian is seen carrying some items in onehand and munching on an apple held with the other.He leaves a building and strolls toward a parkinglot. In the lot, a maroon Triumph backs out of aparking space and hits the pedestrian.

Four of the 20 slides were critical. One version ofeach critical slide contained a particular object(such as a pair of skis leaning against a tree), whilethe other version contained the identical slide witha changed detail (a shovel leaning against a tree).Each subject saw only one version of the criticalslides, and each critical slide was seen equally oftenacross subjects.

Following the slides, which had been seen at a3-sec rate, subjects completed a 10-min unrelatedfiller activity. Then they read a three-paragraphdescription of the slide series supposedly writtenby another individual who had been given muchmore time to view the slides. The descriptioncontained four critical sentences that either did ordid not mention the incorrect critical object. Forexample, if the subject had seen skis leaning againsta tree, his statement might include a sentence thatmentioned "the shovel leaning against the tree."The statements were designed so that the mentionor nonmention of a critical incorrect detail wascounterbalanced over subjects for the four criticalitems.

After an interval of 10 min, subjects were givena forced-choice recognition test. Using two slideprojectors, 10 pairs of slides were presented. The 4critical pairs were randomly intermixed with theremaining filler pairs. One member of each pairhad been seen before, whereas the other had not.The slides that the subject had actually seen variedin the left and right positions.

Results

The percentage of times a correct selec-tion occurred was 55.3 when the inter-vening statement contained misleadinginformation and 70.8 when it contained noinformation. For purposes of analysis, twoproportions were calculated for each of the4 critical slide pairs. One was the proportionof correct selections when misleading in-formation had intervened, and the other,when no information had intervened. A ttest for related measures indicated thatthe mean percentages (given above) werestatistically different from each other,*(3) = 9.34, SEditl = 1.66.

Discussion

The analysis of Experiment 5 permits usto generalize our findings beyond the singlestop-sign-yield-sign stimulus pair. In the

present experiment, subjects who saw aslide containing a particular detail, A,but who were given the information thatthe slide contained Detail B, were subse-quently more likely than control subjectsto select on a forced-choice recognition testa slide containing B rather than a slidewith A.

Note that even with misleading informa-tion, subjects were correct about 55% ofthe time, a figure that is much higher thanthe approximately 42% figure obtainedwith the stop-yield stimuli in Experiment3 after a comparable retention interval.There is probably good reason for this.Any particular object, such as a shovel,can assume many forms. The particularshovel that any subject imagines whilereading the story may not agree with theversion shown during the recognition test.A subject can then successfully reject theslide containing the shovel, not because heor she recognizes the other slide (containingthe skis) but because of not having seen theparticular shovel presented during therecognition test.

With common traffic signs, this wouldnot tend to happen. If a subject imaginesa stop sign while answering a question thatmentions a stop sign, the imagined signwill certainly match the stop sign thatwould be presented during the recognitiontest.

General Discussion

When a person witnesses an importantevent, he or she is often exposed to relatedinformation some time afterward. Thepurpose of the present experiments was toinvestigate how the subsequent informationinfluences memory for the original event.

In the pilot experiment, subjects saw aseries of slides depicting an accident, andafterwards they were exposed to a question-naire that contained either consistent ormisleading information about a particularaspect of the accident. The misleadinginformation caused less accurate respondingon a subsequent yes-no recognition test.Similarly, in Experiment 1, misleadinginformation resulted in poorer performance


on a forced-choice recognition test. Forexample, in one condition, subjects saw astop sign but a subsequent question sug-gested it was actually a yield sign. Sometime later they were given a forced-choicetest and asked to choose the sign theythought they had seen. Over half of thesesubjects incorrectly chose the yield sign.

It has been suggested that the reasonthis happens is that when the misleadinginformation is presented, it is introducedinto the memorial representation for theaccident and causes an alteration of thatrepresentation. Another interpretation isthat subjects are simply agreeing with theinformation contained in their question-naires, even though they actually rememberwhat they saw. This is a demand-charac-teristics explanation. Experiment 2 showedthat when subjects were told that theymight have been exposed to misleadinginformation and were asked to statewhether they thought they had, most ofthem persisted in claiming that they hadseen the incorrect item.

A second interpretation of the forced-choice results is that the original signinformation may not have been encodedin the first place. If it had not been encoded,then the subsequent question may haveintroduced a sign where none existed. Inother words, the phenomenon may be oneof supplementation. On the other hand, ifthe sign got into the original memory (i.e.,was encoded), then the subsequent infor-mation caused either an alteration in theoriginal representation or the creation of anew, stronger representation that competedwith the original representation. Experi-ment 4 showed that at least half of thesubjects encoded the initial sign to thepoint where it was included in a drawingthey made of the incident.

The paradigm used throughout thisresearch involves two critical time intervals:the time between the initial event and thepresentation of subsequent information andthe final test for recollection of the event.In Experiment 3, these intervals wereexamined. Subjects received their final testafter a retention interval of 0 min, 20 min,1 day, 2 days, or 1 week. The subsequent

information was introduced either im-mediately after the initial event or justprior to the final test. The usual retention-interval results were observed: poorerperformance after long intervals than aftershort ones. Of major interest was thefinding that misleading information hada larger impact if presented just prior to arecognition test rather than just after theinitial event.

We have noted two interpretations forour results, namely that either the subse-quent information alters the originalmemory or both the original and the newinformation reside in memory, and thenew competes with the old. Unfortunately,this extremely important issue cannot beresolved with the present data. Those whowish to maintain that the new informationproduces an alteration cannot prove thatthe earlier information will not one dayspontaneously reappear. Those who wishto hold that new and old information bothexist in memory will argue that a personwho responds on the basis of new informa-tion alone does so because the properretrieval cue or the right technique has notbeen used. The value of the present datalies in the fact that they clear up a numberof alternative explanations for previouslypublished phenomena. Furthermore, theyindicate something about the conditionsunder which new information is more orless likely to affect accuracy.

The present work bears some resemblanceto earlier work on the influence of verballabels on memory for visually presentedform stimuli. Much of the earlier work wasdesigned to test the Gestalt hypothesisthat progressive memory changes in thedirection of a "better" figure occur autono-mously. Riley (1962), in an excellentreview of that earlier literature, concludedthat the hypothesis of autonomous changeis probably not testable. Despite thisdrawback, the work on verbal labels wasuseful in revealing that reproductions andrecognition memory (Carmichael, Hogan,& Walter, 1932; Daniel, 1972) of simpleforms were affected by the labels appliedto those forms. The present work representsa much needed extension in that it reveals


that these effects occur not only withartificial forms but also with highlynaturalistic scenes under conditions thathave a high degree of ecological validity.Further, the present work convincinglydemonstrates both the integration of in-formation from more than one source intomemory and the use of that information toreconstruct a "memory" that was neveractually experienced.

Reference Note

1. Loftus, E. F., Salzberg, P. M., Burns, H. JL, &Sanders, R. K. Destruction of a visual memory byverbal information. Paper presented at the annualmeeting of the Psychonomic Society, Denver,November 1975.

References

Carmichael, L., Hogan, H. P., & Walter, A. A. Anexperimental study of the effect of language onthe reproduction of visually perceived form.Journal of Experimental Psychology, 1932, 15,73-86.

Daniel, T. C. Nature of the effect of verbal labelson recognition memory for form. Journal ofExperimental Psychology, 1972, 96, 152-157.

Dooling, D. J., & Christiaansen, R. E. Episodic

and semantic aspects of memory for prose.Journal of Experimental Psychology: HumanLearning and Memory, 1977, 3, 428-436.

Hintzman, D. L. Repetition and memory. In G. H.Bower (Ed.), The psychology of learning andmotivation (Vol. 10). New York: Academic Press,1976.

Kant, I. [Critique of pure reason] (Translated byJ. M. D. Meiklejohn). London: George Bell,1887. (Originally published, 1781.)

Loftus, E. F. Leading questions and the eyewitnessreport. Cognitive Psychology, 1975, 7, 560-572.

Mosteller, F., & Tukey, J. W. The uses and useful-ness of binomial probability paper. Journal of theAmerican Statistical Association, 1949, 44,174-212.

Orne, M. T. On the social psychology of the psycho-logical experiment: With particular reference todemand characteristics and their implications.American Psychologist, 1962, 17, 776-783.

Pezdek, K. Cross-modality semantic integration ofsentence and picture memory. Journal of Experi-mental Psychology: Human Learning and Memory,1977, 3, 515-524.

Riley, D. A. Memory for form. In L. Postman(Ed.), Psychology in the making. New York:Knopf, 1962.

Winer, B. J. Statistical principles in experimentaldesign. New York: McGraw-Hill, 1962.

Received May 26, 1977Revision received August 10, 1977- •

Documents

Semantic Integration of Verbal Information into a Visual