files.eric.ed.govof list words to elicit the critical root word in a free association test. in...

SEMANTIC FALSE MEMORIES IN THE FORM OF DERIVED RELATIONAL INTRUSIONSFOLLOWING TRAINING

PAUL M. GUINTHER AND MICHAEL J. DOUGHER

UNIVERSITY OF NEW MEXICO

Contemporary behavior analytic research is making headway in characterizing memory phenomena thattypically have been characterized by cognitive models, and the current study extends this developmentby producing ‘‘false memories’’ in the form of functional equivalence responding. A match-to-sampletraining procedure was administered in order to encourage participants to treat groups of unrelatedEnglish words as being interchangeable. Following training, participants were presented with a list ofwords from within one of the groups for a free recall test and a recognition test. Results showed thatparticipants were more likely to falsely recall and recognize words that had been assigned to the samegroup as the list words during prior training, relative to words not assigned to the same group andrelative to words that co-occurred with list words. These results indicate that semantic relatedness can beexperimentally manipulated in order to produce specific false memories.

Key words: false memory, equivalence, semantic, intrusions, word lists, free recall, recognition,humans

_______________________________________________________________________________

Applied research on false memory phenom-ena has focused on settings in which uninten-tionally fictitious reports about past events canhave severe negative consequences for individ-uals and society. Both eyewitness testimony(e.g., Loftus, 1975; Loftus & Palmer, 1974) andself-reports of childhood sexual abuse (e.g.,Ceci & Friedman, 2000; Goodman & Clarke-Stewart, 1991; Lyon, 1995) have proven to besurprisingly fallible. Research in both of thesedomains has identified circumstances underwhich the accuracy of reporting is far belowwhat would be expected based on everydayexperience, suggesting that great care need betaken by police, therapists, or anyone whowishes to obtain accurate information withoutdistorting the recollection of subjective expe-riences. More generally, false memory phe-nomena are of interest to some researcherswho view these phenomena as being useful formodeling and exploring the properties ofcognitive schema (e.g., Brewer & Treyens,1981; Minsky, 1975; Schank & Abelson,1977), thereby extending the body of researchunderlying a major branch of therapeuticintervention (i.e. cognitive therapy; Beck,

1976). On a more basic level, cognitivepsychologists often study false memory phe-nomena in order to understand the relation-ship between semantic relatedness, associativestrength, cognitive processes or structures, andapparent errors in the encoding, storage, orretrieval of informational representations (forreviews see Brainerd & Reyna, 2005; Gallo,2006). The purpose of the current study is todevelop a parallel behavior analytic concep-tion of the relationship between semanticmeaning, associative strength, learning history,and false memory phenomena.

Cognitive psychologists often use the Deese-Roediger-McDermott paradigm (DRM; Deese,1959a, 1959b; Roediger & McDermott, 1995)to study the connection between semanticrelatedness, associative strength, and falsememories. During a DRM task, participantshear a list of study words that all have a strongassociation with a critical root word (e.g., thestudy words bed, rest, awake, tired, dream, snore,etc. are all strongly associated with the criticalroot word sleep). Following the study phase,participants perform a free recall task andoften falsely recall (i.e., ‘‘intrude’’) the crit-ical word even though it was never present-ed (e.g., Robinson & Roediger, 1997; Roediger& McDermott, 1995). Deese (1959a,1959b)found that across a variety of constructed lists,the probability that a given list’s critical rootword would be intruded was highly predictablefrom the list’s mean backward associativestrength, which was defined as the tendency

We thank Heather Dowdican and Quynh-Anh Bui fortheir assistance with data collection.

Correspondence concerning this article should bedirected to Paul Guinther, Department of Psychology, 1University of New Mexico, Logan Hall, MSC03 2220,Albuquerque, NM 87131-1161 (e-mail: [email protected]).

doi: 10.1901/jeab.2010.93-329

JOURNAL OF THE EXPERIMENTAL ANALYSIS OF BEHAVIOR 2010, 93, 329–347 NUMBER 3 (MAY)

329

of list words to elicit the critical root word in afree association test. In contrast, dense inter-item associative strength, or the tendency oflist words to elicit other list words in a freeassociation test, has been found to be predic-tive of greater levels of veridical recall andlower levels of false recall (Deese, 1959a,1959b; McEvoy, Nelson, & Komatsu, 1999).Other research suggests that there is a strongersemantic relationship (as measured by LatentSemantic Analysis; see Landauer & Dumais,1997) between intruded words and just-re-called words than there is between intrudedwords and remaining list words (Zaromb et al.2006; see also Howard & Kahana, 1999). Forexample, given a particular study list for freerecall (i.e., keys, farmer, wall, book, moon, island,etc.), a person who had just recalled the studyword ‘‘farmer’’ may subsequently intrude theword ‘‘rancher’’ on the basis of semanticrelatedness, even if this intrusion bears nostrong semantic relation with the remainingstudy words. Semantic relatedness also predictsthe order of veridical recall in episodicmemory tasks (Romney, Brewer, & Batchelder,1993; Steyvers, Shiffrin, & Nelson, 2005;Zaromb et al., 2006). Although cognitivefactors besides semantic relatedness or asso-ciative strength may also come into play duringmemory retrieval (e.g., Roediger, Watson,McDermott, & Gallo, 2001), considerable datasupport the cognitive interpretation that se-mantic relatedness and associative strengthplay important roles in both true and falsememory processes.

Given that semantic relatedness and associa-tive strength are predictive of false memoryphenomena, it is possible that that influencingthese variables may, in turn, influence falsememory phenomena. That is, insight into falsememory phenomena may be garnered bytreating semantic relatedness and associativestrength as dependent variables rather than asindependent variables (e.g., Barnes-Holmes etal., 2005; Dagenbach, Horst, & Carr, 1990;Davis, Geller, Rizzuto, & Kahana, 2008; Dur-gunoglu & Neely, 1987; Hayes & Bissett, 1998;McKoon & Ratcliff, 1979; Pecher & Raaij-makers, 1999). Before exploring the possibilityof directly manipulating semantic relatednessand associative strength as a means of induc-ing false memories, it is important to distin-guish between these two terms, and toconsider the overarching issue of ‘‘meaning’’

as construed from the perspectives of cognitivepsychology and behavior analysis.

Cognitive Construal of Meaning

Within the field of cognitive psychology,memory and meaning are defined in terms ofa computer metaphor, in which informationalrepresentations, which have referents in theworld, are stored in mental structures andretrieved by mental processes. There is, how-ever, no single underlying cognitive theoryregarding the nature of these hypotheticalrepresentations, structures, and processes. Assuch, the defining characteristics of semanticmeaning (i.e., meaning in the context oflanguage) varies according to which cognitivemodel one ascribes. Nevertheless, an overarch-ing commonality of many cognitive models isthe idea that informational representationscan be thought of as ‘‘nodes’’ that areconnected to each other in a cognitivenetwork (Anderson, 1983; Collins & Loftus,1975; Kawamoto, 1993; Masson, 1995; McNa-mara, 1992; Moss, Hare, Day, & Tyler, 1994;Plaut, 1995).

While a precise definition of semanticmeaning depends on how one construes thesecognitive networks, it is generally the case thatthe extent of connectivity between nodes isthought to arise from stimulus co-occurrence(see Hutchison, 2003). That is, the co-occur-rence of stimuli leads to the simultaneousactivation of their nodal representations, andsimultaneous activation increases the degreeof connectivity between the nodes. To theextent that the nodal representations ofstimuli are interconnected, or to the extentthat stimuli share nodal representations, theyare said to be semantically related, or to‘‘mean the same thing.’’ For example, it canbe said that cat and dog are semantically relatedon the grounds that they share associations (ortend to co-occur) with a variety of other stimuli(e.g., collars, veterinarians, allergies, fur, claws,etc.). Stimulus co-occurrence is so central tocognitive conceptions of semantic meaningand associative strength that it has served asthe foundation for meaning extraction inprominent formal modeling techniques suchas Latent Semantic Analysis (Foltz, 1996;Landauer & Dumais, 1997; Landauer, Foltz,& Laham, 1998) and Hyperspace Analogue toLanguage (Burgess, 1998; Lund & Burgess,1996; see also Spence & Owens, 1990).

330 PAUL M. GUINTHER and MICHAEL J. DOUGHER

The terms ‘‘semantic relatedness’’ and‘‘associative strength’’ have various operation-al definitions, and their technical definitionsalso vary across cognitive models. For example,in holistic cognitive models (e.g., Anderson,1983; Collins & Loftus, 1975; McNamara,1992), the two terms are fairly interchangeablewhereas in distributed cognitive models (e.g.,e.g., Kawamoto, 1993; Masson, 1995; Moss, etal., 1994; Plaut, 1995) the term ‘‘semanticrelatedness’’ is reserved for stimuli that sharephysical features. For present purposes, ratherthan ascribing associative strength to thedefinitions of any particular cognitive model,we can think of it as the degree of connectivityamong nodes arising from direct stimulus co-occurrence, whereas semantic relatedness canbe thought of as the degree of indirectconnectivity among nodes that is mediated byother nodes within the network. For example,lion and stripes do not co-occur frequently andtherefore are not highly associated. However,lion is associated with tiger and tiger is associatedwith stripes, so there is a semantic relationshipbetween lion and stripes that is mediated bytheir common association with tiger (see Balota& Lorch, 1986). In contrast, bread and butter co-occur frequently through phrasal contiguity(see Fodor, 1983) and therefore are highlyassociated, and they are also semanticallyrelated through a variety of mediating associ-ations (e.g., toast, jam, breakfast, etc.).

Behavior Analytic Construal of Meaning

Behavior analysts ground meaning, not inmental representations, but rather in theconditions under which behavior occurs(Day, 1969; Leigland, 2007; Skinner, 1953,1957). Sometimes the use of a particular wordis dictated by direct training, as when a child isreinforced for pointing to the printed word catin the presence of a picture of a cat. Othertimes the use of a word involves a derivedrelationship that has not been directly trained(see Hayes, Barnes-Homes, & Roche, 2001;Sidman, 1994). For example, a child who hasbeen directly reinforced for picking theprinted word cat in the presence of a pictureof a cat could also be directly reinforced forpicking the word gato in the presence of apicture of a cat. The child may then derive thatcat ‘‘means the same thing as’’ gato, as wouldbe evident if the child went on to pick theprinted word gato in the presence of the word

cat without any prior history of reinforcementfor such a behavior. While physical stimulusfeatures play a role in many behaviors (e.g.,generalized responding to perceptually similarstimuli), a principal characteristic of derivedrelationships is that they can occur betweenstimuli that bear no necessary physical rela-tionship to each other (e.g., words andpictures; words and other words; words andsounds; tastes and sounds; etc.). Thus, twostimuli can ‘‘mean the same thing’’ notbecause they refer to common physical fea-tures, but rather because they are functionallyinterchangeable and thereby produce compa-rable effects (Hayes, et al., 2001; Sidman, 1994;Sidman & Tailby, 1982).

In the laboratory, behavior analysts mayestablish semantic relations using match-to-sample (MTS; Barnes-Holmes et al., 2005;Hayes & Bissett, 1998; Sidman, 1994; Sidman& Tailby, 1982). On a given MTS training trial,a sample stimulus is presented along with twoor more comparison stimuli. Basic researchtypically employs arbitrary, unrelated stimuliduring MTS training, but for purposes ofillustration we will focus on stimuli that mostreaders will recognize as being related. Imag-ine that the sample on a given trial is a pictureof a cat, and the comparison stimuli are theprinted words cat, perro, and dog. Only one ofthese comparison stimuli ‘‘matches’’ the sam-ple, in the sense that its selection produces areinforcer (e.g., selection of the printed wordcat in the presence of the picture of the catproduces food, or praise, or the word ‘‘COR-RECT’’). On the next trial, the samplestimulus remains a picture of a cat, but thecomparison items now are the printed wordsgato, perro, and dog; in this instance theselection of the printed word gato is rein-forced. Later trials may include differentsamples (e.g., a picture of a dog) and differentcomparisons (e.g., the printed words dog, gato,and cat), whereupon different sample–com-parison selections are reinforced (e.g., theselection of the printed word dog in thepresence of the picture of the dog would bereinforced).

Following this kind of MTS training, animportant behavioral phenomenon known asstimulus equivalence can emerge in which aperson derives symmetric and transitive rela-tionships (Sidman, 1994; Sidman & Tailby,1982). In the example described above, in the

DRIFT PARADIGM 331

presence of a picture of a cat (A) picking theprinted word cat (B) is reinforced; this trainingarrangement can be denoted as ARB. The Aand B stimuli are said to be symmetricallyrelated if, without any further training, theindividual can select a picture of a cat in thepresence of the printed word cat (BRA).Similarly, direct training in which picking theprinted word gato (C) in the presence of apicture of a cat (A) is reinforced (ARC) mayproduce the emergent symmetric relationshipCRA. Moreover, direct training of ARB andARC may also yield the emergent transitiverelationships BRC and CRB (these derivedrelations are technically labeled as combinedsymmetry and transitivity, but for the sake ofbrevity we refer to them simply as transitiverelationships). When symmetric and transitiveresponding are exhibited with respect to a setof stimuli, the stimuli are said to participate ina stimulus equivalence class (Sidman, 1994;Sidman & Tailby, 1982). To the extent thatthey are now functionally interchangeable,these physically dissimilar and formerly unre-lated stimuli now ‘‘mean the same thing.’’

Our central assumption is that transitiverelations reflect the mediated semantic rela-tions described in the cognitive literature(Barnes-Holmes et al., 2005; Hayes & Bissett,1998). For example, recall that lion and stripesinfrequently co-occur in natural language, yetthey are related through the mediating itemtiger, as evidenced by findings that lion andstripes prime each other (e.g., Balota & Lorch1986). Likewise, B and C stimuli never co-occur during MTS training, yet they can cometo be functionally equivalent through theirdirectly trained relations with A. This func-tional equivalence reflects semantic related-ness, as evidenced by findings that B stimuliand C stimuli prime each other (Barnes-Holmes et al.; Hayes & Bissett). Thus, MTStraining can be used to establish semanticrelations among stimuli while holding theirunmediated associative strength relatively con-stant.

Origins of False Memories in Equivalence Classes

An interesting characteristic of stimulusequivalence classes (i.e., stimuli that meanthe same thing, are functionally interchange-able, or participate in a ‘‘relational frame ofcoordination’’; see Hayes et al., 2001) istransfer of stimulus function, in which a new

function is directly acquired by one classmember and then indirectly acquired byremaining members (Barnes, 1994; Donahoe& Palmer, 2004; Dougher & Markham, 1994,1996; Goldiamond, 1962, 1966; Hayes, 1991;Sidman, 1994). For example, suppose that viaMTS training the words cat and gato becomemembers of a common stimulus equivalenceclass (first row of Figure 1). Subsequently,suppose that the presentation of the word catis paired with an electric shock, such that thisword directly acquires a fear-eliciting function.Given that cat and gato are members of acommon stimulus equivalence class, the elicit-ing function that had been directly acquiredby the word cat is likely to be indirectlyacquired by the word gato, such that thepresentation of the word gato would elicit afear response even though the word gato wasnever directly paired with an electric shock.Following such a pattern of responding, itwould be said that the words cat and gato aremembers of a ‘‘functional equivalence class’’with respect to fear elicitation (see Donahoe &Palmer, 2004; Dougher & Markham, 1994,1996; Goldiamond, 1962, 1966).

We propose that certain semantic falsememory phenomena can be accounted for interms of functional equivalence responding.Just as an eliciting function can transferbetween the words cat and gato, a ‘‘remember-ing’’ function may transfer between studywords and nonstudy words that are function-ally equivalent to study words. That is, thefunctions of study words are altered by theinstruction to remember them, and nonstudywords that are functionally-equivalent to studywords may likewise undergo function alter-ation and be remembered (see Discussion).For example, the DRM list word rest may befunctionally equivalent to the critical rootword sleep for English-speaking participants,most likely because of a variety of learningexperiences that have accumulated duringmultiple interactions within an English-speak-ing community (see the second row ofFigure 1). Should the DRM list words cometo acquire a remembering function via instruc-tion (e.g., as manifest by correct recall andcorrect recognition of the word rest onsubsequent memory tests), functionally equiv-alent words that were not presented for studyalso may acquire this function (e.g., asmanifest by the false recall and false recogni-


tion of the word sleep). Theoretically, aremembering function could transfer fromstudy words to nonstudy words if there hasbeen a learning history that promotes thetransfer of function between the study wordsand nonstudy words, including a learninghistory sufficient for the formation of derivedrelationships (see the third row of Figure 1and Method). In order to test this possibility,we developed the Derived Relational Intru-sions Following Training (DRIFT) paradigm.

Our primary objective in developing theDRIFT paradigm was to manipulate specificenvironmental variables in order to producespecific false memory phenomena. Specifical-ly, we sought to use MTS training to influencesemantic false memory phenomena by creat-ing derived semantic relations. We wished toexamine if stimuli assigned to the same class asstudy list words (i.e., words assigned to besemantically related to study list words) wouldbe intruded or falsely recognized at greaterrates relative to words that had not beenassigned to the same class. Furthermore, sincecomparison stimuli are simultaneously pre-sented during MTS training, a co-occurrencethat could conceivably influence word associ-

ations and hence influence associative falsememory, we introduced a possible secondarysource of intrusions and false positive recog-nitions. Thus, the DRIFT paradigm was de-signed to demonstrate the effects of MTStraining on semantic false memory phenome-na while permitting exploratory analyses ofsemantic (i.e., derived or mediated) versusassociative (i.e., co-occurrence) effects.

METHOD

Participants

Sixty-six undergraduate psychology studentswho were recruited through a departmentalweb advertisement volunteered for the study inexchange for course credit. Data are omittedfor 9 individuals who were affected by equip-ment failure (n 5 4), did not comply withinstructions (n 5 4), or generated outcomesthat could not be categorized according to ournomothetic data analysis scheme (n 5 1),leaving N 5 57 participants.

Setting, Apparatus, and Materials

Participants completed the study in a quiet2 m 3 2 m laboratory in the Psychology

Fig. 1. Three plausible equivalence classes. Top row: An example from the text involving the words cat and gato and apicture of a cat. Middle row: An example from the DRM paradigm, which is commonly used to study false memory incognitive research. Bottom row: An illustration of the DRIFT paradigm. In all cases, see text for explanation.

DRIFT PARADIGM 333

Department at the University of New Mexico.The room contained a chair, a desk, and astandard desktop computer equipped with amouse for making responses and a 43.2-cm(17-inch) monitor on which stimuli werepresented. The computer program for collect-ing responses and presenting stimuli waswritten by the first author and is availableupon request. A total of 72 English words werechosen (from Wilson, 1988) as experimentalstimuli on the basis that there was no obvioussemantic theme uniting them; the words wereconcrete nouns five to eight letters in length(see Figure 2). While word norms are typicallycrucial for an understanding of semantic andassociative effects, our MTS training proce-dure involved the random assignment of wordsto conditions for each participant, therebystatistically neutralizing preexperimental se-mantic relations or associations. That is,barring extreme statistical improbabilities,differential dependent false memory effectsacross experimental conditions would be theproduct of the independent manipulation ofMTS learning history and could not beattributed to stimulus properties establishedprior to experimentation.

Procedure

After meeting the experimenter and signingstatements of informed consent, participantsbegan the study with a computerized MTStraining procedure. This training was intendedto establish three 24-member equivalenceclasses and consisted of an arbitrary, three-comparison, MTS task. Training followed aone-to-many format, with colored shapes (ared square, blue circle, or green pentagon)always serving as samples and the words ascomparisons. Participants read the followinginstructions on the computer screen:

Welcome to the experiment! For this first partof the experiment, you will be solving severalproblems. For each problem, you will be showna sample item on the top third of the screenand three selection items on the bottom thirdof the screen. Your job is to learn toconsistently pick the correct selection items,but we can’t tell you much about how to goabout doing this. What we can tell you is thatthe sample is important, and that for eachproblem you will be shown a happy face if youmake a correct selection or you will be shown asad face if you make an incorrect selection. To

solve the problems, you can make yourselection choices by using the 1, 2, or 3 keyon the keypad. To pick the item on the left,press the 1 key. To pick the item in the middle,press the 2 key. To pick the item on the right,press the 3 key. After you pick an item andreceive feedback, you can move on to the nextproblem by pressing the [SPACE BAR]. Afteryou have been through all of the regularproblems once, you will be given some ‘‘BonusProblems.’’ Even though the Bonus Problemshave correct answers, you will not be givenfeedback after you answer them! If you do notanswer the Bonus Problems correctly, you willgo through all of the regular problems again.You only have a limited number of chances totry to beat the Bonus Problems. This processwill repeat until you are able to answer theBonus Problems correctly or until you run outof chances, at which point you will move on tothe second part of the experiment. The personwho can beat the Bonus Problems with thefewest number of errors will win $100. Feel freeto take a break at any time, but please let theexperimenter know if you need to leave theroom. Pace yourself, use a good strategy, bepatient with yourself, and try your best to beatthe Bonus Problems! Once you have finishedreading this, please ask the experimenter anyquestions you may have, or otherwise let theexperimenter know that you are ready tobegin.

For each participant, a red square, a bluecircle, and a green pentagon were randomlyassigned as stimulus samples that are designat-ed here as Shape 1 (S1), Shape 2 (S2), andShape 3 (S3). For each participant, the 72stimulus comparison words were randomlyassigned to one of five Types, designated hereas T1, T2, T3, T4 and T5 (see Figure 2). T1words were presented for study during thememory test in the second phase of theexperiment, and they were consistently pairedwith T2 words as comparison stimuli duringMTS training (e.g., ANIMAL and GARDENalways co-occurred as comparison stimuliacross training blocks, as did ARTIST andGRAPH, etc.). Defined in terms of the frequen-cy of experimental co-occurrence, the T2words were therefore strong associates of T1words. However, the T2 words were notsemantically related to T1 words because theywere assigned to a different class. T3 wordswere semantically related to T1 words in thesense that they were assigned to the same classas T1 words during MTS training; T3 words


and T1 words never co-occurred (exceptduring tests of transitivity) and therefore werenot associated during MTS training trials. T4words were consistently paired with T3 wordsas comparison stimuli during MTS training(e.g., CLOUD and KNIFE always co-occurred ascomparison stimuli across training blocks, asdid DOCTOR and LEMON, etc.). T4 wordswere weak associates of the T1 words in thesense that they were never paired with T1words during MTS training. T5 words wererandomly presented as third comparisonstimuli along with T1–T2 and T3–T4 pairs;the particular comparison triplet that a T5word participated in was randomly determinedat the beginning of each block of trials. Forexample, a comparison triplet might consist ofthe words ANIMAL, GARDEN and PAINTduring the first block of training, but thetriplet might instead consist of the wordsANIMAL, GARDEN, and WINDOW during thesecond block of training. As such, T5 words

were moderate associates of T1 words, in thesense that they were inconsistently paired withT1 words as comparison stimuli during MTStraining trials.

Training (‘‘Regular Problems’’). On each MTStraining trial (see Figure 3), a sample shape(i.e., a red square, blue circle, or greenpentagon) appeared at the top of the comput-er screen and three comparison words ap-peared below the sample (i.e., a T1–T2 pairwith a random T5 word, or a T3–T4 pair with arandom T5 word). Participants were requiredto select one comparison using the 1, 2, or 3key on the computer keypad, at which point aline appeared under the selected word. Cor-rect responses were followed by the presenta-tion of a happy face beneath the selectedcomparison and a pleasant chime sound,whereas incorrect responses were followed bythe presentation of a sad face beneath theselected comparison and an unpleasant buzzsound. A correct response was scored if the

Fig. 2. The 72 words used in the study. Although shown in alphabetical order here, the words were randomlyassigned to one of five Types (T1–T5) for each participant. Superscripts denote specific randomly assigned words of agiven Type, and can be cross-referenced with Table 1. See text for details.

DRIFT PARADIGM 335

participant selected an S1 (T1 or T3) word inthe presence of an S1 sample, an S2 (T2 or T4)word in the presence of an S2 sample, or an S3(T5) word in the presence of an S3 sample. Anincorrect response was scored if the partici-pant’s selection did not match the sample.Following each comparison-stimulus selection,the visual feedback remained on the screenuntil the participant pressed the space bar, atwhich point a new problem was presented.

Each MTS training block consisted of 24trial sets of three trials each, corresponding tothe 24 possible word triplets within a block(i.e., T11, T21, and T5x; T12, T22, and T5x; …T312, T412, and T5x). Each trial set consisted ofan S1 sample problem, an S2 sample problem,and an S3 sample problem, with the sameword triplet (i.e., comparison stimuli) acrossthe set of trials. Thus, each training blockconsisted of 72 individual trials. The order oftrial set presentation and the assignment of T5words to comparison triplets were randomlydetermined at the beginning of each block,the order of problem type presentation withina set (i.e., S1, S2, and S3) was randomlydetermined at the beginning of each set, andthe left–middle–right positions of the compar-

ison stimuli were randomly determined at thebeginning of each trial.

Equivalence testing (‘‘Bonus Problems’’). EveryMTS training block of Regular Problems wasfollowed by a round of ‘‘Bonus Problems,’’which consisted of six symmetry trials and thensix transitivity trials (see Table 1), presented ina random order and in a format that resem-bled the Regular Problems but without feed-back. On symmetry trials, the sample was anexperimental word and the comparisons werethe three shapes. On transitivity trials, thesample and comparisons all were experimentalwords. Incorrect comparisons during transitiv-ity testing consisted of other stimuli specific tothat round of Bonus Problems, and one of thetwo possible incorrect comparisons from eachnonmatching class was randomly selected tobe presented as comparison stimuli (designat-ed by ‘‘or’’ in Table 1). Participants completedMTS training if they were able to achieve 100%accuracy on a round of these ‘‘Bonus Prob-lems.’’ New symmetry and transitivity trialswere administered following every block, suchthat participants were never tested on the samestimuli twice. Aside from this trial novelty,participants could not predict which random

Fig. 3. Examples of MTS training feedback following the selection of comparison stimuli. Left column: A squareserves as the sample (S1) with animal as the correct comparison. Middle column: A circle serves as the sample (S2) withgarden as the correct comparison. Right column: A pentagon serves as the sample (S3) with office as the correctcomparison. NOTE: Actual assignments of shapes and words were randomized for each participant and hence differentfor each participant. See text for further explanation.


relations would be tested during administra-tion of the Bonus Problems. Participants had12 chances (i.e., up to 12 training blocks) tomeet criterion on the Bonus Problems.

Study-list presentation and memory testing. Afterexhausting their 12 chances to pass a round ofBonus Problems or after passing a round ofBonus Problems with 100% accuracy, partici-pants were required to leave the laboratoryand take a break for approximately 5 min(participants actually returned 4 to10 minlater). Once participants returned, they were

told that they were entering a new phase of theexperiment, and were given the followinginstructions for the free recall task:

For the next part of the experiment, we willshow you some words that we want you toremember for a later memory test. The wordswill automatically appear one at a time on thescreen, so you don’t need to press any keys.Even though all of the words that you areabout to see will be familiar to you, when welater ask you to remember the MEMORY TESTWORDS, just remember the words that you are

Table 1

Sample and Comparison Stimuli Used to Test Symmetry and Transitivity.

Sample(Correct)

Comparison 1(Incorrect)

Comparison 2(Incorrect)

Comparison 3

Block 1 Bonus ProblemsSymmetry 1 T11 S1 S2 S3Symmetry 2 T31 S1 S2 S3Symmetry 3 T21 S2 S1 S3Symmetry 4 T41 S2 S1 S3Symmetry 5 T51 S3 S1 S2Symmetry 6 T513 S3 S1 S2Transitivity 1 T11 T31 T21 or T41 T51 or T513

Transitivity 2 T31 T11 T21 or T41 T51 or T513

















Blocks 4–12 Bonus Problems(Pattern Continues) … … … …

Note. Superscripts refer to specific individual words of a given Type, and can be cross-referenced with Figure 2.

DRIFT PARADIGM 337

about to see. Please ask the experimenter ifyou have any questions.

Once the experimenter had answered anyquestions, the 12 T1 study list words werepresented on the computer screen one at atime, below a presentation of the S1 stimulus,for 2 s each with a 2-s interstimulus interval.Following presentation of the list items,participants were shown a reminder screeninstructing them to ‘‘Remember the MEMORYTEST words! Please press the [SPACE BAR]now to continue…’’ Upon pressing the spacebar, participants were presented with theinstructions ‘‘Please read the following shortstory’’ along with a transcript of the AmericanIndian short story War of the Ghosts (fromBartlett, 1932). It took participants approxi-mately 2 min to read the short story (seeTable 2). At the conclusion of the story, theinstruction ‘‘Please let the experimenter knowthat you have finished reading the story’’ wasprinted on the screen. Once prompted, theexperimenter handed the participants a blanksheet of paper and a pen and asked theparticipants to write down as many of thememory test words as they could remember inthe next 4 min. The experimenter thenpressed a button on the computer thatadvanced the screen to show these sameinstructions, along with a timer that counteddown from 240 s. After the time expired, achime sounded and the experimenter collect-ed the participants’ recalled words. After

participants left at the end of the experiment,the experimenter manually entered the re-called words into the computer for scoring,making corrections for misspellings. Followingadministration of the free recall test, partici-pants were administered a recognition testbeginning with the presentation of the follow-ing instructions:

For the next part of the experiment, you willsee a word come up on the screen. If this wordis a MEMORY TEST word, click on the leftmouse button (YES). If the word you see is nota MEMORY TEST word, click on the rightmouse button (NO). Once you make yourresponse, a new word will appear for you tojudge. Even though all of the words that youare about to see will be familiar to you, onlychoose (YES) if the current word is a memorytest word. On average, only one out of everyfive words will be a memory test word. Wewould like you to make your selection asquickly as possible—we will be recording yourreaction time. However, you should also be asaccurate as you can! Please ask the experi-menter if you have any questions before youbegin. The test will begin by pressing the[SPACE BAR].

The specific instructions that were present-ed to each participant reflected a randomcounterbalancing of the left/right side towhich the YES and NO responses wereassigned, and the words YES and NO werepresented on either side of the bottom of thescreen throughout the recognition task in

Table 2

Demographic, MTS Training, and Memory Test Means (SDs) by Achievement Groups and forthe Total Sample.

Achievement Group

TotalFailed Partial Full

Age 19.85 (2.38) 19.73 (2.96) 19.34 (2.53) 19.56 (2.58)Males (n) 6 2 14 22Females (n) 7 13 15 35Education 13.00 (0.91) 12.73 (1.49) 12.72 (0.92) 12.79 (1.08)MTS Blocks 12.00 (0.00) 12.00 (0.00) 7.03 (2.68) 9.47 (3.14)MTS Total Errors 497.31 (53.96) 307.67 (98.06) 191.07 (72.13) 291.60 (144.07)MTS Final Accuracy 44.38 (14.98) 83.68 (11.66) 83.76 (7.67) 74.76 (19.72)Delay Time in Seconds 96.70 (21.89) 103.29 (24.51) 118.92 (39.69) 109.74 (33.68)T1 Correct Recall 3.92 (2.36) 5.60 (2.53) 6.83 (2.16) 5.84 (2.55)T1 Correct Recognitions 6.62 (2.29) 7.73 (3.08) 9.79 (1.80) 8.53 (2.64)

Notes. MTS Total Errors does not include errors on the Bonus Problems. MTS Final Accuracy refers to the percent ofcorrect trials on participants’ last MTS training block. Delay time refers to how long it took participants to read the shortstory between study list presentation and free recall.


accordance with this counterbalancing. A totalof 60 recognition trials were given in a randomorder, corresponding to each of the T1, T2,T3, and T4 words, along with half (randomlyselected) of the T5 words. On each trial, arandomly selected experimental word ap-peared on the center of the screen andremained until the participant made a re-sponse. After a response was made, the worddisappeared and a new word appeared on thenext trial 1 s later. Participants did not receivefeedback regarding the accuracy or speed oftheir recognition responses. After the lastrecognition word was judged, participantswere debriefed and thanked for their partici-pation. Depending on participant perfor-mance, the entire experimental procedurelasted no longer than 2 hr.

Post Hoc Separation into Groups

Because in most published studies falsememory phenomena are quite variable acrossparticipants, we chose to employ nomotheticanalyses and to analyze data from participantswho achieved different levels of success intraining. This approach was consistent withour view of semantic relatedness being afunction of experience. From this perspective,false-memory effects should be heavily depen-dent on training outcomes. Participants thuswere sorted by their MTS training and BonusProblems performances in order to assignthem to one of three groups, and data fromall three groups were analyzed.

The first (Failed) group (n 5 13) includedparticipants who had achieved less than 66%accuracy on their last block of MTS trials anddid not pass a set of random Bonus Problemswith 100% accuracy. The training experiencesof these individuals were not expected tocontribute to semantic relatedness or, there-fore, the emergence of false memory effects.The second (Partial Achievement) group (n 515) included participants who achieved limit-ed success during training, meaning greaterthan 66% accuracy on their last block of MTStrials and failure to pass a set of random BonusProblems with 100% accuracy. The trainingexperiences of these individuals were expectedto spawn some false memory effects. Finally,the third (Full Achievement) group (n 5 29)included those participants who had achievedgreater than 66% accuracy on their last blockof MTS trials and passed a set of random

Bonus Problems with 100% accuracy. Thetraining experiences of these individuals wereexpected to reliably spawn false memoryeffects.

Demographic characteristics of the threegroups and the total sample can be found inTable 2. According to one-way between-sub-jects ANOVAs (critical alpha 5 .05), thegroups did not differ significantly in terms ofage, F(2, 54) 5 0.21, p 5 .812, g2 5 .01,gender ratio, F(2, 54) 5 2.88, p 5 .065, g2 5.10, or years of education, F(2, 54) 5 0.31,p 5 .734, g2 5 .01.

RESULTS

MTS Performance Analyses

Table 2 shows performance summary datafor the three groups and for the total sample. Aseries of one-way between-subjects ANOVAswith post-hoc pairwise comparisons (criticalalpha 5 .05) were conducted to evaluatepossible differences among the three groupsin terms of MTS performance variables. Thegroups differed significantly in mean numberof MTS training blocks completed, F(2, 54) 547.19, p , .001, g2 5 .64. Specifically,participants in the full achievement groupwere able to pass the Bonus Problems beforeexhausting their 12 chances and thereforerequired fewer MTS training blocks than theother two groups; participants who failedtraining or with partial achievement did notpass the Bonus Problems and therefore wereexposed to exactly 12 MTS training blocks.There was also a significant difference amongthe three groups in mean total number oferrors (i.e., incorrect responses) made duringMTS training trials, F(2, 54) 5 72.55, p , .001,g2 5 .73, with significant differences among allthree groups reflecting a combination ofdifferential error rates and a differentialnumber of training blocks; error rates de-creased with increasing achievement. Therewas a significant difference among the threegroups in mean trial accuracy on the final blockof MTS training trials, F(2, 54) 5 67.17, p ,.001, g2 5 .71. Participants who failed trainingtended to achieve lower trial accuracy on theirfinal MTS training block than did participantsin the partial and full achievement groups; thepartial and full achievement groups wereequated for mean trial accuracy on their finalMTS training block by the experimenters.

DRIFT PARADIGM 339

Memory Analyses

There were no significant differencesamong the three groups in mean amount oftime required to read the short story duringthe delay period, F(2, 54) 5 2.45, p 5 .096,g2 5 .08. There was a significant differenceamong the three groups in mean number ofcorrectly recalled T1 study list words duringfree recall, F(2, 54) 5 7.25, p 5 .002, g2 5 .21,with participants in the full achievement grouphaving significantly greater mean correct recallthan participants who failed training. Similar-ly, participants in the full achievement grouphad significantly greater mean correct recog-nition of T1 words than participants who failedtraining, F(2, 54) 5 9.74, p , .001, g2 5 .27.These differences could reflect a facilitatingeffect of full achievement, or perhaps partic-ipants who were predisposed to full achieve-ment were also predisposed to correctly recalland recognize more words.

Free recall analyses. Perfect performance onthe free recall task would require participantsto recall all 12 of the T1 words withoutintruding any T2, T3, T4, or T5 words,indicating that a remembering function wasacquired by all of the T1 words with noindirect acquisition of the remembering func-tion by any other words. We anticipated thatsemantically related T3 words would be themost likely to be intruded, on the grounds thatthese words had been assigned to the sameclass as the T1 study list words during MTStraining and thereby would be most likely toindirectly acquire the remembering function.We also anticipated a relatively high probabil-ity of intrusions among T2 strong associates,since these words had consistently co-occurredwith T1 words throughout MTS training. Thelowest levels of intrusions were anticipated forT5 moderate associates and then for T4 weakassociates. In addition, we anticipated thatthese effects would be most pronouncedamong participants in the full achievementgroup and least pronounced among partici-pants who failed training. In order to correctfor the greater number of T5 words availablefor intruding, analyses were conducted afterfirst dividing the observed number of T5intrusions in half for each participant; theintrusion means in Figure 4 reflect this T5/2adjustment.

A 3 3 4 (group by word Type) mixed-designANOVA with a Greenhouse-Geisser correction

for sphericity indicated a nonsignificant maineffect of group, F(2, 54) 5 1.15, p 5 .33, partialg2 5 .04, and a significant main effect of wordType, F(1.49, 80.25) 5 27.20, p , .001, par-tial g2 5 .34. However, these main effects mustbe interpreted in light of a significant interac-tion between group and word Type, F(2.97,80.25) 5 4.90, p 5 .004, partial g2 5 .15. Post hocanalyses of the simple main effect of word Typeat each level of achievement group wereconducted using pairwise comparisons, a Bon-ferroni correction, and a critical alpha of .05.Among those participants who failed training,there were no significant differences in meannumber of T2, T3, T4, and T5/2 intrusions.Among those participants in the partialachievement group, the mean number of T3intrusions was significantly greater than themean number of T4 intrusions, but there wereno significant differences among the meannumber of T2, T4, and T5/2 intrusions.Among those participants in the full achieve-ment group, the mean number of T3 intru-sions was significantly greater than the meannumber of T2, T4, and T5/2 intrusions, whichdid not significantly differ from each other.The significant interaction suggests that suc-cessful MTS training may have produced botha semantic relatedness effect as well as a‘‘semantic suppression’’ effect. That is, itappears that not only did participants with

Fig. 4. Mean number of intrusions for each word Typefor the three MTS training achievement groups during thefree recall phase of the experiment. Error bars conveystandard error.


partial and full achievement tend to treat theT3 words as if these words were ‘‘the same as’’T1 words, they also tended to treat the T2, T4,and T5 words as if these words were not thesame as T1 words.

To summarize, differential word co-occur-rence during MTS training did not producedifferential frequencies of T2, T4, or T5/2associative intrusions (with a possible excep-tion of infrequent T4 intruding among partic-ipants in the partial achievement group).However, the results do indicate that seman-tically-related T3 words were most likely toindirectly acquire the T1 remembering func-tion, at least among those participants in thepartial and full achievement groups. Thus, athigher levels of MTS training achievement, theMTS training procedure was able to producespecific semantic false memory phenomena inthe form of elevated T3 intrusions.

Recognition analyses. Perfect performance onthe recognition task would require partici-pants to correctly recognize all 12 of the T1words without making any T2, T3, T4, or T5false positive recognitions, indicating that aremembering function was acquired by all ofthe T1 words with no indirect acquisition ofthe remembering function by any other words.We expected the results of the recognition testto run parallel to those of the free recall test.Since 12 words of each Type were presentedduring the recognition task, no adjustmentswere made to T5 means for recognitionanalyses. Figure 5 shows the mean number ofT2, T3, T4, and T5 false positive recognitionsfor each group.

A 3 3 4 (group by word Type) mixed-designANOVA with a Greenhouse-Geisser correctionfor sphericity indicated a significant maineffect of group, F(2, 54) 5 6.32, p 5 .003,partial g2 5 .19, and a significant main effectof word Type, F(1.46, 78.84) 5 25.57, p , .001,partial g2 5 .32. However, these main effectsmust be interpreted in light of a significantinteraction between group and word Type,F(2.92, 78.84) 5 6.27, p 5 .001, partial g2 5 .19.Post hoc analyses of the simple main effect ofword Type at each level of group wereconducted using pairwise comparisons, a Bon-ferroni correction, and a critical alpha of .05.Among those participants who failed training,there were no significant differences in meannumber of T2, T3, T4, and T5 false positiverecognitions. Among those participants in the

partial and full achievement groups, the meannumber of T3 intrusions was significantlygreater than the mean number of T2, T4, andT5 intrusions, which did not significantly differfrom each other. The recognition results weretherefore similar to the free recall results.

DISCUSSION

Many previous studies show that semanticrelatedness predicts the likelihood of memoryintrusions (e.g., Deese, 1959a, 1959b; McEvoy,et al., 1999; Roediger & McDermott, 1995;Zaromb et al., 2006). Most research on falsememory, however, has relied upon preexistingsemantic relatedness that was presumablybased in preexperimental linguistic experienc-es. By contrast, in the present investigation,semantic relatedness was built from theground up through MTS training, whichestablished derived semantic relations andpermitted the transfer of stimulus functions.By treating semantic relatedness as a depen-dent variable rather than an independentvariable, we were able to manipulate specificenvironmental variables in order to producespecific false memories.

Running parallel to naturalistic learninghistories, the results of this study indicate thatMTS training can be used as a tool to engineersemantic relationships that distort recall and

Fig. 5. Mean number of false positive recognitions foreach word Type for the three MTS training achievementgroups during the recognition phase of the experiment.Error bars convey standard error.

DRIFT PARADIGM 341

recognition of subsequent events: words as-signed to the same class as study list words weremore likely to be intruded and more likely tobe falsely recognized than other types ofwords. While there are undoubtedly manyMTS training arrangements that would notproduce this effect and other experimental ornaturalistic arrangements that would producethis effect, our findings demonstrate thatsemantic false memory effects can be engi-neered through the manipulation of environ-mental variables. As depicted in the third rowof Figure 1, we first used MTS training toestablish a group of T1 words and a group ofT3 words as being equivalent and thereforesemantically related, as demonstrated by testsof symmetry and transitivity. In the next phaseof the experiment, participants were shownthe T1 words and were asked to rememberthem for a later memory test. As evidenced bycorrect T1 recall and correct T1 recognitions,these instructions interacted with participants’learning histories to establish a rememberingfunction for the T1 words. Furthermore,because T1 words and T3 words had beenassigned to the same class during prior MTStraining, the remembering function was ac-quired indirectly and differentially by seman-tically-related T3 words. Specifically, relative towords that had not been assigned to the sameclass as T1 words during prior MTS training(i.e., T2, T4, and T5 words), more T3 wordswere intruded and falsely recognized. Thus,our results replicate previous findings thatsemantic relations are predictive of falsememories (see Brainerd & Reyna, 2005; Gallo,2006). Moreover, we influenced semanticrelations and thereby influenced false memo-ries.

However, the differential co-occurrence ofwords during MTS training did not result indifferential levels of associative intruding orfalse recognition. Specifically, T2 words andT5 words frequently co-occurred with T1words during MTS comparison presentations,yet these words were no more likely to beintruded or falsely recognized than T4 words,which never co-occurred with T1 words. Thesefindings were unexpected, given that stimulusco-occurrence is thought to increase associa-tive strength (Burgess, 1998; Fodor, 1983;Landauer & Dumais, 1997), and associativestrength is thought to increase the probabilityof intrusion and false recognition (e.g., Deese,

1959a, 1959b; McEvoy et al., 1999; Robinson &Roediger, 1997; Roediger & McDermott,1995).

Implications for Understanding False Memory

Our MTS training procedure producedsemantic behaviors that can be interpreted toreflect conceptual mediation (Balota & Lorch,1986), which is a central phenomenon sup-porting holistic models of semantic memory(see Hutchison, 2003). From a cognitiveperspective, the experimental procedure ap-pears to have established mediated semanticrelations (i.e., T1«S1«T3), similar to natu-rally occurring mediated semantic relations(e.g., lion«tiger«stripes). Thus, a semanticeffect was demonstrated despite the fact thatthe T1 and T3 words do not refer to commonphysical features, which would not be predict-ed from a distributed model of semanticmemory based in physical feature overlap(e.g., Kawamoto, 1993; Masson, 1995; Moss etal., 1994; Plaut, 1995; see Hutchison, 2003). Ofcourse, it remains possible that other falsememory phenomena may involve the physicalcharacteristics of stimuli, but the current falsememory phenomena would appear to bebased in derived, functional semantic relationsrather than referential feature overlap.

While the results of the current behavioranalytic investigation are consistent with cog-nitive explanations involving conceptual me-diation within holistic semantic networks, theydo not support the notion that associativestrength is the product of stimulus co-occur-rence (e.g., Burgess, 1988; Fodor, 1983; Land-auer & Dumais, 1997). For participants in thefull achievement group, the mean total num-ber of direct (T1–T3) and mediated (S1–T1and S1–T3) co-occurrences among the T1 andT3 words (Observed M 5 182.78) was smallerthan the mean number of co-occurrencesbetween T1 and T2 words (Minimum frequen-cy M 5 253.08), yet the vast majority ofintrusions and false positive recognitions wereT3 words. That is, among participants in thefull achievement group, the semantic relation-ship between T1 and T3 words reflected amore powerful effect than would be expectedon the basis of their relative associativestrength alone. We are aware of no cognitivemodel predicting a negative relationship be-tween level of co-occurrence and magnitude ofsemantic or associative effects. Thus, though


formal models based in stimulus co-occur-rence can be globally predictive of semanticand associative behaviors (e.g., Burgess, 1998;Landauer & Dumais, 1997), contingency fac-tors other than stimulus co-occurrence couldplay important roles in the formation oforganic semantic relations and associations.In particular, the present results indicate thatthe functional relationship between stimuli is amore important factor in determining seman-tic false memory phenomena than is stimulusco-occurrence.

In a related vein, Hutchison (2003) hasidentified several different topographies ofstimulus associations, such as synonyms, anto-nyms, category members, perceptually similaritems, and phrasal associates. The task ofcharacterizing these and other relations withincognitive models could be facilitated byanalyses of the environmental histories respon-sible for the formation of various forms ofrelatedness and association. For example, theDRIFT paradigm can be thought of as aprocedure for producing synonyms; variationsof this procedure could help illuminate thenature of conceptual mediation. To the extentthat different cognitive models make differentpredictions with respect to stimuli that vary instimulus functions, associations, and features,procedures designed to experimentally manip-ulate these variables may constitute usefultechnologies for cognitive researchers (seealso Carroll & Kirsner, 1982; Dagenbach etal., 1990; Davis et al., 2008; Durgunoglu &Neely, 1987; McKoon & Ratcliff, 1979; Neely &Durgunoglu, 1985). Conversely, there are avariety of semantic effects that are reliablypredicted using cognitive models (e.g., seeBrainerd & Reyna, 2005; Gallo, 2006; Hutch-ison, 2003), but have yet to be fully character-ized in terms of environmental manipulationsand derived relationships. The success of theDRIFT paradigm in producing specific seman-tic false memory phenomena represents anadvance in this direction, and suggests thatmany other avenues of research are availablefor exploration that can be mutually informedby cognitive and behavior analytic perspec-tives.

Methodological Issues

It could be construed that the current studywas not truly experimental because partici-pants were not randomly assigned to experi-

mental and control groups. That is, ratherthan providing an experimental group withtraining and withholding this training from acontrol group, all participants received train-ing and participant groups were formed posthoc (i.e., participants were assigned to groupsbased on their training achievement). Howev-er, it can also be construed that the study wasindeed experimental because stimuli wererandomly assigned to conditions. That is,participants had differential learning historieswith respect to different groups of words, andthese differential histories comprised experi-mental and control conditions within subjects.While it would not be difficult for futureresearchers to provide between-subjects exper-imental and control conditions, this approachmay not be especially informative. To illus-trate, including a between-subjects controlcondition would involve withholding trainingfrom participants in the control group, pre-senting them with a list of random studywords, and then seeing how many of theirrecalled words matched a different list ofrandom nonstudy words; one would expectthe resulting number of matching words to befairly constant at about zero. Rather thanproviding a between-subjects control condi-tion of this sort, resources may be betterallocated in determining those factors thatinfluence training achievement.

The current design does not rule out thepossibility that stimulus co-occurrence caninfluence semantic relatedness, associativestrength, or false memory phenomena. Forexample, verbally competent adults can formequivalence classes of simultaneously present-ed stimuli even in the absence of ongoingreinforcement for this behavior (Barnes,Smeets, & Leader, 1996; Leader, Barnes, &Smeets, 1996). One would expect that stimulusco-occurrence could influence behavior to theextent that there has been phylogenetic orontogenetic selection for such a relationship(see also Mitchell, De Houwer & Lovibond,2009), but neither of these variables wasmanipulated in the current design. Instead,we attempted to see if uncontrolled preexperi-mental histories of selection would interactwith experimental manipulations of stimulusco-occurrence to produce differential falsememory phenomena. However, we found nosuch effect. The null effect could have beenthe result of an insufficient preexperimental

DRIFT PARADIGM 343

history of selection for responding to stimulusco-occurrence, a lack of contextual cues forsuch responding, or could have resulted froma manipulation failure (i.e., a wider range ofdifferential co-occurrences would be necessaryto detect an effect). Alternatively, it may havebeen the case that there was a floor effect forassociation reflecting the observed semanticsuppression effect. While it would be difficultto manipulate phylogenetic selection, futureinvestigations could provide differential (orcontextually controlled) experimental learn-ing histories with respect to responding tostimulus co-occurrence, include conditionswith higher frequencies of stimulus co-occur-rence, and include conditions for whichstimulus co-occurrence is not confounded byderived relations or differential class member-ship (see also Leader & Barnes-Holmes, 2001).

A more serious concern regarding thecurrent design was the inclusion of tests ofsymmetry and transitivity prior to the admin-istration of memory tests. While stimulusequivalence is of great interest to manybehavior analysts, tests for stimulus equiva-lence introduced an experimental source ofdirect stimulus co-occurrence. Even thoughpassing these tests required the derivation ofstimulus relations, it is possible that the falsememory effects were directly mediated via thetests, which may have created linked, mediatedfunctional classes without requiring that thefalse memory performances be derived. How-ever, this interpretation seems improbable,given that equivalence tests involved co-occur-ring presentations of matching as well asnonmatching comparisons and no feedbackwas provided during equivalence testing; onecould not distinguish between classes on thebasis of co-occurrence alone. Furthermore, anumber of studies have shown that MTSrelational training procedures without tests ofstimulus equivalence are sufficient for thederived transfer or transformation of stimulusfunctions (see Barnes 1994; Barnes & Roche,1996; Barnes-Holmes, Barnes-Holmes, Smeets,Cullinan, & Leader, 2004), suggesting thatequivalence testing was not a necessary com-ponent for the production of false memoryphenomena in the current study (though itmay have catalyzed such responding). Indeed,we have conducted other studies, to bereported elsewhere, using modified versionsof the DRIFT paradigm that indicate that

stimulus equivalence testing is not necessaryfor the production of false memory phenom-ena following MTS training. Thus, the essen-tial training characteristics of the DRIFTparadigm appear to be those that permit thetransfer of stimulus function, whether throughlinked classes or through derived relation-ships.

Behavior Analytic Interpretation

While the MTS training procedure resultedin false memory performances that satisfy thedefinition of functional equivalence (see Do-nahoe & Palmer, 2004; Dougher & Markham,1994; Dougher & Markham, 1996; Goldia-mond, 1962, 1966), it also resulted in derivedsymmetry and transitivity performances thatsatisfy the definition of stimulus equivalence(see Sidman, 1971; Sidman, 1994; Sidman &Tailby 1982). However, this is not meant toimply that functional equivalence was causedby prior stimulus equivalence (see McIlvane &Dube, 1990). Rather, both types of respondingwere dependent behavioral variables that weredue to the MTS training, and both involvedthe transfer of function. For functional equiv-alence, the function that transferred to the T3words during free recall and recognition was aremembering function. For stimulus equiva-lence, the functions that transferred duringtests of symmetry and transitivity were therespective conditional and discriminative func-tions that had been directly trained duringMTS training (Hayes, 1994; Sidman & Tailby,1982). To further clarify, the false memoryphenomena exhibited in the current studywere not themselves instances of stimulusequivalence responding, since the free recalland recognition memory tests were not tests ofsymmetry and transitivity.

In line with Donahoe and Palmer (2004)and Skinner (1953; 1957), we assume that theexperimenter’s request to recall and recognizethe memory test words occasioned whateverprecurrent, covert operants the individualparticipants used to enhance recall andrecognition and, thereby, satisfy the implicitexperimental contingency. In the absence ofan appropriate experimental analysis, it is notpossible to stipulate the actual covert reper-toires used by the individual participants toenhance recall and recognition, so it isdifficult to specify the exact nature of thefunction that was acquired by the T1 study list


words. We have therefore opted to use thegeneral term ‘‘remembering function’’ torefer to the functions that differentially trans-ferred from T1 words to T3 words duringrecall and recognition, though the exactfunctions that transferred may have variedaccording to individual repertoires and ac-cording to type of remembering task.

In the case of the recognition task, the T1words functioned conditionally for the dis-criminative function of YES, while the T2–T5words functioned conditionally for the dis-criminative function of NO. The fact that theT3 words evoked significantly more falsepositive recognitions (i.e., YES responses) thanany of the other stimuli suggests that MTStraining facilitated transfer of the conditionalT1 function to same-class T3 words. Again,while false recognitions may well have beenmediated by private operant behaviors thathave not been elaborated in this account, thepattern of responding demonstrates the for-mation of a functional equivalence classconsisting of T1 and T3 words.

Our failure to find any differential associa-tive influence of T2, T4, or T5 co-occurrenceon intrusions or false positive recognitions issomewhat surprising in light of previousfindings that temporal contiguity influencesintrusion (e.g., Zaromb et al. 2006) anddemonstrations that concurrent pairing canlead to stimulus equivalence responding(Barnes et al., 1996; Leader et al., 1996;Tonneau, Arreola, & Martinez, 2006). The lackof associative effects (and presence of asemantic suppression effect) may be relatedto the nature of the experimental procedure,which may have resulted in three differentfunctional equivalence classes (i.e., S1, S2, andS3 words). This may have placed the threeclasses in a relation of nonequivalence ordistinction (e.g., see Hayes et al., 2001), andthat may have overridden the effects of co-occurrence. Specifically, across MTS trainingtrials, one comparison item was the correctanswer and the others were, by definition,wrong, and this might have served to ‘‘sepa-rate’’ these items functionally. Consistent withthis possibility, it was not uncommon inpostexperimental conversation for participantsto relate that they had considered writing downa word during the memory test, but thendecided not to write it because they realizedthat the word they were considering was a S2 or

S3 word. Because the relations among the threepotential classes were not specifically assessed,this account of the null associative findings isspeculative and awaits further research.

Our results contribute to the ongoingdevelopment of a satisfactory behavioral ac-count of semantic meaning and extend thisaccount to the domain of false memoryphenomena. Following the precedents ofHayes and Bissett (1998) and Barnes-Holmeset al. (2005), this study represents a furtherincrement towards establishing the viability ofa model of semantic meaning based onderived stimulus relations. While their resultsindicate that learning histories leading toequivalence relations can also produce seman-tic priming effects, our results similarly indi-cate that such histories can produce falsememory effects. These studies suggest thatbehavioral explorations of traditionally cogni-tive phenomena may be fruitful. It is our hopethat this line of research engenders anoptimistic appreciation of the interplay be-tween behavioral and cognitive perspectives onlearning, memory, and meaning.

REFERENCES

Anderson, J. R. (1983). A spreading activation theory ofmemory. Journal of Verbal Learning & Verbal Behavior,22, 261–295.

Balota, D. A., & Lorch, R. F. (1986). Depth of automaticspreading activation: Mediated priming effects inpronounciation but not in lexical decision. Journal ofExperimental Psychology: Learning, Memory, & Cognition,12, 336–345.

Barnes, D. (1994). Stimulus equivalence and relationalframe theory. The Psychological Record, 44, 91–124.

Barnes, D., & Roche, B. (1996). Relational frame theoryand stimulus equivalence are fundamentally different:A reply to Saunders’ commentary. The PsychologicalRecord, 46, 489–508.

Barnes, D., Smeets, P. M., & Leader, G. (1996). Newprocedures for establishing emergent matching per-formances in children and adults: Implications forstimulus equivalence. In T. R. Zentall, & P. M. Smeets(Eds.). Stimulus class formation in humans and animals(pp. 153–171). Amsterdam: Elsevier.

Barnes-Holmes, D., Barnes-Holmes, Y., Smeets, P. M.,Cullinan, V., & Leader, G. (2004). Relational frametheory and stimulus equivalence: Conceptual andprocedural issues. International Journal of Psychology andPsychological Therapy, 4, 181–214.

Barnes-Holmes, D., Staunton, C., Whelan, R., Barnes-Holmes, Y., Commins, S., Walsh, D., et al. (2005).Derived stimulus relations, semantic priming, andevent-related potentials: Testing a behavioral theoryof semantic networks. Journal of the ExperimentalAnalysis of Behavior, 84, 417–433.

DRIFT PARADIGM 345

Bartlett, F. C. (1932). Remembering: A study in experimentaland social psychology. Cambridge, UK: CambridgeUniversity Press.

Beck, A. T. (1976). Cognitive therapy and the emotionaldisorders. New York: International Universities Press.

Brainerd, C. J., & Reyna, V. F. (2005). The science of falsememory. Oxford: University Press.

Brewer, W. F., & Treyens, J. C. (1981). Role of schemata inmemory for places. Cognitive Psychology, 13, 207–230.

Burgess, C. (1998). From simple associations to thebuilding blocks of language: Modeling meaning inmemory with the HAL model. Behavior ResearchMethods, Instruments, & Computers, 30, 188–198.

Carroll, M., & Kirsner, K. (1982). Context and repeti-tion effects in lexical decision and recognitionmemory. Journal of Verbal Learning & Verbal Behavior,21, 55–69.

Ceci, S. J., & Friedman, R. D. (2000). The suggestibility ofchildren: Scientific research and legal implications.Cornell Law Review, 86, 33–108.

Collins, A. M., & Loftus, E. F. (1975). A spreadingactivation theory of semantic memory. PsychologicalReview, 82, 407–428.

Dagenbach, D., Horst, S., & Carr, T. H. (1990). Addingnew information to semantic memory: How muchlearning is enough to produce automatic priming?Journal of Experimental Psychology: Learning, Memory, &Cognition, 16, 581–591.

Davis, O. C., Geller, A. S., Rizzuto, D. S., & Kahana, M. J.(2008). Temporal associative processes revealed byintrusions in paired-associate recall. Psychonomic Bulle-tin & Review, 15, 64–69.

Day, W. F. (1969). On certain similarities between thephilosophical investigations of Ludwig Wittgensteinand the operationism of B. F. Skinner. Journal of theExperimental Analysis of Behavior, 12, 489–506.

Deese, J. (1959a). On the prediction of occurrence ofparticular verbal intrusions in immediate recall.Journal of Experimental Psychology, 58, 17–22.

Deese, J. (1959b). Influence of interitem associativestrength upon immediate free recall. PsychologicalReports, 5, 235–241.

Donahoe, J. W., & Palmer, D. C. (2004). Learning andComplex Behavior. Richmond: Ledgetop Publishing.

Dougher, M. J., & Markham, M. (1994). Stimulusequivalence, functional equivalence and the transferof function. In S. C. Hayes, L. J. Hayes, M. Sato, & K.Ono (Eds.). Behavior Analysis of Language and CognitionContext Press: Reno, NV.

Dougher, M. J., & Markham, M. R. (1996). Stimulus classesand the untrained acquisition of stimulus functions.In T. R. Zentall, & P. M. Smeets (Eds.). Stimulus classformation in humans and animals (pp. 137–152). NewYork: Elsevier.

Durgunoglu, A. Y., & Neely, J. H. (1987). On obtainingepisodic priming in a lexical decision task followingpaired-associate learning. Journal of Experimental Psy-chology: Learning, Memory, & Cognition, 13, 206–222.

Fodor, J. A. (1983). Modularity of mind. Cambridge, MA:MIT Press.

Foltz, P. W. (1996). Latent semantic analysis for text-basedresearch. Behavior Research Methods, Instruments, &Computers, 28, 197–202.

Gallo, D. A. (2006). Associative illusions of memory: Falsememory research in DRM and related tasks. New York:Psychology Press.

Goldiamond, I. (1962). Perception. In A. J. Bachrach(Ed.). Experimental foundations of clinical psychology (pp.280–340). New York: Wiley.

Goldiamond, I. (1966). Perception, language, and con-ceptualization rules. In G. Kleinmuntz (Ed.). Problemsolving (pp. 183–224). New York: Wiley.

Goodman, G. S., & Clarke-Stewart, A. (1991). Suggestibilityin children’s testimony: Implications for child sexualabuse investigations. In J. L. Doris (Ed.). Thesuggestibility of children’s recollections (pp. 92–105).Washington, DC: American Psychological Association.

Hayes, S. C. (1991). A relational control theory of stimulusequivalence. In L. J. Hayes, & P. N. Chase (Eds.).Dialogues on verbal behavior (pp. 19–40). Reno, NV:Context Press.

Hayes, S. C. (1994). Relational frame theory: A functionalapproach to verbal behavior. In S. C. Hayes, L. J. Hayes,M. Sato, & K. Ono (Eds.). Behavior analysis of languageand cognition (pp. 11–30). Reno, NV: Context Press.

Hayes, S. C., Barnes-Holmes, D., & Roche, B. (Eds.).(2001). Relational Frame Theory: A Post-Skinnerianaccount of human language and cognition. New York:Plenum Press.

Hayes, S. C., & Bissett, R. T. (1998). Derived stimulusrelations produce mediated and episodic priming. ThePsychological Record, 48, 617–630.

Howard, M. W., & Kahana, M. J. (1999). Contextualvariability and serial position effects in free recall.Journal of Experimental Psychology: Learning, Memory, andCognition, 25, 923–941.

Hutchison, K. A. (2003). Is semantic priming due toassociation strength or feature overlap? A microana-lytic review. Psychonomic Bulletin & Review, 10,785–813.

Kawamoto, A. H. (1993). Nonlinear dynamics in theresolution of lexical ambiguity: A parallel distributedprocessing account. Journal of Memory & Language, 32,474–516.

Landauer, T. K., & Dumais, S. T. (1997). Solution toPlato’s problem: The latent semantic analysis theoryof acquisition, induction, and representation ofknowledge. Psychological Review, 104, 211–240.

Landauer, T. K., Foltz, P. W., & Laham, D. (1998). Anintroduction to latent semantic analysis. DiscourseProcesses, 25, 259–284.

Leigland, S. (2007). Fifty years later: Comments on thefurther development of a science of verbal behavior.The Behavior Analyst Today, 8, 336–346.

Leader, G., Barnes, D., & Smeets, P. M. (1996). Establish-ing equivalence relations using a respondent-typetraining procedure. The Psychological Record, 46,685–706.

Leader, G., & Barnes-Holmes, D. (2001). Matching-to-sample and respondent-type training as methods forproducing equivalence relations: Isolating the criticalvariable. The Psychological Record, 51, 429–444.

Loftus, E. F. (1975). Leading questions and eyewitnessreport. Cognitive Psychology, 7, 560–572.

Loftus, E. F., & Palmer, J. C. (1974). Reconstruction ofautomobile destruction: An example of the interac-tion between language and memory. Journal of VerbalLearning and Verbal Behavior, 13, 585–589.

Lund, K., & Burgess, C. (1996). Producing high-dimen-sional semantic spaces from lexical co-occurrence.Behavior Research Methods, Instruments, & Computers, 28,203–208.


Lyon, T. D. (1995). False allegations and false denials inchild sexual abuse. Psychology, Public Policy, and Law, 1,429–437.

Masson, M. E. J. (1995). A distributed memory model ofsemantic priming. Journal of Experimental Psychology:Learning, Memory, & Cognition, 21, 3–23.

Mitchell, C., De Houwer, J., & Lovibond, P. (2009). Thepropositional nature of human associative learning.Behavioral and Brain Sciences, 32, 183–246.

McEvoy, C. L., Nelson, D. L., & Komatsu, T. (1999). Whatis the connection between true and false memories?The differential roles of interitem associations inrecall and recognition. Journal of Experimental Psychol-ogy: Learning, Memory, and Cognition, 25, 1177–1194.

McIlvane, W. J., & Dube, W. V. (1990). Do stimulus classesexist before they are tested? Analysis of Verbal Behavior,8, 13–17.

McKoon, G., & Ratcliff, R. (1979). Priming in episodic andsemantic memory. Journal of Verbal Learning & VerbalBehavior, 18, 463–480.

McNamara, T. P. (1992). Theories of priming: I. Associa-tive distance and lag. Journal of Experimental Psychology:Learning, Memory, & Cognition, 18, 1173–1190.

Minsky, M. (1975). A framework for representing knowl-edge. In P. H. Hinston (Ed.). The psychology of computervision. New York: McGraw-Hill.

Moss, H. E., Hare, M. L., Day, P., & Tyler, L. K. (1994). Adistributed memory model of the associative boost insemantic priming. Connection Science, 6, 413–427.

Neely, J. H., & Durgunoglu, A. Y. (1985). Dissociativeepisodic and semantic priming effects in episodicrecognition and lexical decision tasks. Journal ofMemory & Language, 24, 466–489.

Pecher, D., & Raaijmakers, J. G. W. (1999). Automaticpriming effects for new associations in lexical decisionand perceptual identification. Quarterly Journal ofExperimental Psychology: Human Experimental Psychology,52, 593–614.

Plaut, D. C. (1995). Double dissociation without modular-ity: Evidence from connectionist neuropsychology.Journal of Clinical & Experimental Neuropsychology, 17,291–321.

Robinson, K. J., & Roediger, H. L. (1997). Associativeprocesses in false recall and false recognition.Psychological Science, 8, 231–237.

Roediger, H. L., & McDermott, K. B. (1995). Creating falsememories: Remembering words not presented in lists.Journal of Experimental Psychology: Learning, Memory, andCognition, 21, 803–814.

Roediger, H. L., Watson, J. M., McDermott, K. B., & Gallo,D. A. (2001). Factors that determine false recall: Amultiple regression analysis. Psychonomic Bulletin andReview, 8, 385–407.

Romney, A. K., Brewer, D. D., & Batchelder, W. H. (1993).Predicting clustering from semantic structure. Psycho-logical Science, 4, 28–34.

Schank, R. C., & Abelson, R. (1977). Scripts, plans, goals,and understanding. Hillsdale, NJ: Erlbaum.

Sidman, M. (1971). Reading and auditory–visual equiva-lences. Journal of Speech and Hearing Research, 14, 5–13.

Sidman, M. (1994). Equivalence relations and behavior: Aresearch story. Boston: Authors Cooperative.

Sidman, M., & Tailby, W. (1982). Conditional discrimina-tion vs. matching to sample: An expansion of thetesting paradigm. Journal of the Experimental Analysis ofBehavior, 37, 5–22.

Skinner, B. F. (1953). Science and human behavior. New York:Macmillan.

Skinner, B. F. (1957). Verbal behavior. Englewood Cliffs, NJ:Prentice Hall.

Spence, D. P., & Owens, K. C. (1990). Lexical co-occurrence and association strength. Journal of Psycho-linguistic Research, 19, 317–330.

Styvers, M., Shiffrin, R. M., & Nelson, D. L. (2005). Wordassociation spaces for predicting semantic similarityeffects in episodic memory. In A. F. Healy (Ed.).Cognitive psychology and its applications: Festschrift inhonor of Lyle Bourne, Walter Kintsch, and ThomasLandauer (pp. 237–249). Washington, DC: AmericanPsychological Association.

Tonneau, F., Arreola, F., & Martinez, A. G. (2006).Function transformation without reinforcement. Jour-nal of the Experimental Analysis of Behavior, 85, 393–405.

Wilson, M. D. (1988). The MRC Psycholinguistic Data-base: Machine Readable Dictionary, Version 2. Behav-ioural Research Methods, Instruments, and Computers, 20,6–11.

Zaromb, F. M., Howard, M. W., Dolan, E. D., Sirotin, Y. B.,Tully, M., Wingfield, A., et al. (2006). Temporalassociations and prior-list intrusions in free recall.Journal of Experimental Psychology: Learning, Memory, andCognition, 32, 792–804.

Received: April 23, 2009Final Acceptance: January 6, 2010

DRIFT PARADIGM 347

files.eric.ed.govof list words to elicit the critical root word in a free association test. in...

Documents