journal of experimental psychology: general · the repetition effect. experiment 2 used separate...

Journal of Experimental Psychology:General

VOL. 112, No. 3 SEPTEMBER 1983

Episodic and Lexical Contributions to theRepetition Effect in Word Identification

Timothy C. Feustel, Richard M. Shiffrin, and Aita SalasooIndiana University

SUMMARY

The repetition effect refers to the finding that the speed and accuracy of naming avisually presented word is enhanced by a single prior presentation of the word. A newtechnique was developed to study this phenomenon: The visual signal-to-noise ratio ofa printed item in a field of masks was slowly increased. When accuracy was of interest,the increase ceased at a predetermined time; when latency was of interest, the increasecontinued until the printed item could be named. Experiment 1 tested the validity ofthe new accuracy technique against a more traditional threshold measure of ease ofidentification, in which the item is presented for a single brief exposure, followed bya mask. When performance levels at the first presentation were equated for the twotechniques and for both words and nonwords, the repetition effect was equal for thetechniques and slightly stronger for words than for nonwords. In Experiment la psy-chometric functions for first presentations were obtained, giving accuracy as a functionof final exposure duration. A large interaction was seen with the traditional techniqueyielding superior performance for words than the new technique, but the reverse wastrue for nonwords. In Experiment 2 the latency version of the new technique was used:The difference in'the latencies necessary for word and nonword identification was foundto be additive to the difference due to repeated presentations. Taken together, the resultsof the experiments suggest separate contributions of lexical status and presentations tothe repetition effect. Experiment 2 used separate groups for words and nonwords, butthe word-nonword difference was, if anything, increased when mixed lists of words andnonwords were used in Experiment 3. This result rules out certain guessing bias in-terpretations of the word-nonword differences. In Experiments 1,2, and 5, lag betweenrepetitions had at most a small and nonsignificant effect on identification accuracy andlatency. However, in Experiment 5, lag between repetitions had a large effect on rec-,ognition performance. In Experiments 1 and 3, shifting case between presentations ofidentical items produced a very small decrease in the repetition effect, suggesting aminimal role for low-level physical features in the repetition effect. In Experiments 2and 4, orthographic similarity (i.e., spelling overlap) of new items to previously presenteditems not sharing a common morpheme was studied. A small (sometimes significant)facilitation of identification of such new items was observed. This result suggests thatletter-name clusters play some role in the repetition effect.

A model was developed that outlines the relative contributions of episodic traces forparticular events, and of uriitized representations of words in semantic memory, to therepetition effect in word and nonword identification. The unitization that characterizesidentification of words and that is missing for nonwords plays a prominent role in themodel. Specifically, the repetition effect is attributed to the presence of episodic memorytraces that are assumed to increase uniformly the speed and accuracy of both word

Copyright 1983 by the American Psychological Association, Inc.

309

310 T. FEUSTEL, R. SHIFFRIN, AND ,A. SALASOO

and nonword identification. On the other hand, the observed superiority of word overnonword identification and the interaction of the size of this effect with the new andtraditional tasks are attributed to the availability of unitized memory representationsthat allow fast, accurate identification responses for the words.

In this article we attempt to study somefactors that produce the "repetition effect"in identification, decision, and recognitiontasks. In the present context the repetitioneffect refers to the enhancing effect of a singlepresentation of an item on a subsequent testof that same item. The enhancement is usu-ally substantial and has been demonstratedin a variety of different paradigms. For ex-ample, the probability of correct identifica-tion of a briefly presented word is muchhigher for words that have been presentedearlier (Carroll & Kirsner, 1982; Jacoby,1983, in press; Jacoby & Dallas, 1981; Jacoby&Witherspoon, 1982; Morton, 1979b;Murrell& Morton, 1974). In lexical decision, in whichsubjects are asked to decide whether a pre-sented letter string is a word, a single priorexperience with a given word decreases boththe latencies and errors for subsequent testswith the same word (Forbach, Stanners, &Hochhaus, 1974; Scarborough, Cortese, &Scarborough, 1977; Scarborough, Gerard, &Cortese, 1979). Similar effects have been foundin a wide variety of tasks, including word frag-ment completion (Tulving, Schacter, & Stark,1982), reading inverted text (Kolers, 1976),spelling homophones (Jacoby & Witherspoon,1982), solving anagrams (Jacoby & Dallas,1981), and free association (Kihlstrom, 1980).In some instances, the effects of a single pre-sentation of a stimulus could be observed

This research was supported by Public Health ServiceGrant MH12717 to Richard M. Shiffrin and by an In-diana University Pre-Doctoral Grant in Aid to TimothyC. Feustel.

We would like to thank Richard N. Aslin, Larry L.Jacoby, Lloyd R. Peterson, and David B. Pisoni for theircomments on earlier versions of this article. Jerry For-shee and David Link were responsible for much of thesoftware and hardware used to complete the experi-ments.

Timothy C. Feustel is now at American Bell, Lincroft,New Jersey.

Requests for reprints should be sent to Richard M.Shiffrin, Department of Psychology, Indiana University,Bloomington, Indiana 47405.

hours, or even days, later (Jacoby, 1983; Jacoby& Dallas, 1981; Tulving et al., 1982).

The repetition effect is important becauseit lies at the empirical and theoretical inter-face between what have come to be knownas episodic and semantic memory (Tulving,1972). The distinction was motivated by anemergent body of literature that dealt withmemory for abstract knowledge—knowledgeabout words, linguistic structure, facts, rela-tions among concepts, categorical relations,and the like. The study of this abstract, or"semantic," memory was contrasted with thestudy of episodic memory: the study of theability to remember individual events thatoccurred within the confines of a specifictemporal-contextual environment. In effect,the dichotomy distinguishes between mem-ory for specific instances, or episodes, andmemory of a more general character.

Two largely distinct areas of research maybe defined according to this distinction, eachemploying different tasks, concepts, andmodels. The study of episodic memory hasbecome identified with tasks such as list recallor recognition memory, in which a subject isasked to decide, say, whether ajword has beenpresented on an earlier list of words. Semanticmemory, on the other hand, has been studiedwith tasks such as word identification or lexicaldecision, in which subjects are asked aboutthe identity of a word or whether a string ofletters constitutes a word. These tasks do notlogically require access to memory for indi-vidual events. Only limited contact has beenmade between these two fields of endeavor.However, item repetition is an important ex-ception; it has been studied extensively in bothtypes of task, sometimes in the same experi-ment.

Theoretically, repetition effects have beenattributed to semantic factors by some re-searchers and to episodic factors by others.In one type of "semantic" account, repetitioneffects are attributed to residual activationsof abstract memory representations, or"logogens," which are responsible for pro-

REPETITION EFFECTS 311

during phonological codes for words (Mor-ton, 1969, 1979a, 1979b). The idea is thateach time a logogen is activated, its thresholdis temporarily lowered so that on subsequentpresentations, reactivation is more rapid.One of the important properties of the Ip-gogens is that they are assumed to be abstractrepresentations that do not retain informa-tion about individual instantiations of theirreferents. In effect, the logogen system, asdenned by Morton, is a component of se-mantic memory.

Much of the research with semantic mem-ory tasks has been consistent with the logogenmodel. For example, in threshold identifi-cation tasks, large repetition effects are ob-served with repeated words that differ inphysical appearance, as in upper- versus low-ercase (Jacoby & Witherspoon, 1982) orhand- versus typewritten (Morton, 1979b).The transfer of facilitation between items thatdiffer in case has also been found in lexicaldecision (Scarborough et al., 1977). Theseresults have been taken as evidence for theabstract character of logogens. Similarly, inlexical decision, although repeated words en-joy persistent repetition effects, repeated non-words are only slightly facilitated (Scarbor-ough et al., 1977), not at all facilitated (For-bach et al., 1974), or even inhibited (McKoon& Ratcliff, 1979). Findings such as these areconsistent with the logogen model becausenonwords are not represented in the system,and thus there is no logogen, or "node," formaintaining activations.

Perhaps the most ubiquitous evidence thatthe repetition effect is tapping semantic,rather than episodic, memory is the findingthat repeating an item has a much differenteffect on recognition memory, an episodictask, than on semantic tasks. Although therepetition effect may persist essentially with-out change for days in semantic tasks, rec-ognition performance for the same materialsdeteriorates over comparatively short periodsof time (Jacoby & Dallas, 1981; Jacoby &Witherspoon, 1982; Scarborough etal., 1977;Scarborough et al., 1979; Tulving et al.,1982). In fact, Jacoby and Witherspoon haveoffered evidence that the size of the repetitioneffect in identification does not depend onthe response given to that item on an earliertest of recognition memory. Similarly, Tulv-

ing and his associates (Tulving et al., 1982)have provided evidence of stochastic inde-pendence between performance on a recog-nition memory task and a fragment comple-tion task. These findings have promptedmany investigators to suggest that two in-dependent memory systems are tapped bythe two types of task, that is, episodic andsemantic memory. Unfortunately, there areempirical and methodological problems withthis interpretation.

First, the facilitation observed betweenphysically different instantiations of the sameword in the visual domain does not generalizevery strongly to repetitions across the audi-tory and visual modalities (Jacoby & Dallas,1981; Morton, 1979b; Winnick & Daniel,1970). In fact, even within the visual domain,there is no transfer between a printed wordand a picture of its referent (Scarborough etal., 1979). Thus, if the logogens are respon-sible for the repetition effect, there must bedifferent logogen systems for the differentmodalities. Morton (1979b) adopted this so-lution to account for transfer failure betweenvision and audition. Of course, the pictureresults would require a similar solution forpictoral versus written stimuli. Clearly, thisproliferation of logogen systems violates thespirit of parsimony.

The second problem with the conclusionthat the repetition effect resides in semanticmemory is the finding that when changes areinduced in the size of the effect in identifi-cation tasks, parallel changes are sometimesobserved in recognition memory perfor-mance. For example, Jacoby (1983) foundthat as he increased the proportion of wordsthat were common to both the study and testlists, the repetition effect was larger. Thus, asis true for recognition memory (e.g., Jacoby,1972), list context can affect identificationperformance. Similarly, although repetitioneffects in semantic tasks appear to be verypersistent, both identification and lexical de-cision do yield observable performance de-creases over retention intervals that also pro-duce decreases in recognition memory per-formance (Jacoby, 1983; Scarborough et al.,1977).

Another difficulty revolves around thequestion of independence in performancebetween identification and recognition mem-

312 T. FEUSTEL, R. SHIFFRIN, AND A. SALASOO

ory tasks, and the implication that this in-dependence reflects different memory sys-tems (Tulving et al., 1982). The finding inquestion is that the probability that a wordis correctly identified in a threshold task isindependent of whether a correct recognitionjudgment had been made on that same wordin an earlier test of recognition memory (Ja-coby & Witherspoon, 1982). This result doesnot preclude the possibility that the two tasksshare some common mechanisms that arenot involved in the process whereby a rec-ognition decision is made. For example, if aword must be identified before a recognitionjudgment about it can be made (not an un-reasonable assumption), then the identifica-tion process would be completely embeddedwithin the recognition process, with recog-nition requiring an additional decision stage.Thus, the independence observed betweenrecognition responses and identification fa-cilitation bears on the type of processing nec-essary for performing the two tasks andshould not be interpreted as reflecting theoperation of two different memory systems.

In fact, recently Jacoby (1983, in press) hassuggested that the identification process is in-timately involved in recognition memoryperformance. His idea is that the ease orfluency with which a word is identified con-stitutes the feeling of "familiarity" that hasbeen proposed by some experimenters as animportant component of the recognition pro-cess (e.g., Atkinson & Juola, 1973, 1974;Mandler, 1980). That is, the more rapidly anitem is identified, the more familiar it seemsand, consequently, the more likely that it willbe judged as having occurred earlier. Unfor-tunately, the single-trial threshold identifi-cation technique is not ideally suited for in-vestigating this possibility because it does notyield a measure of the relative fluency ofidentification of an item: A word is eitheridentified or it is not.

Finally, the finding in lexical decision thatrepetition effects are not observed with non-words is suspect for methodological reasons:Whether or not an item is a word is con-founded with the response. That is, the sub-ject must always respond "yes" to a word and"no" to a nonword. For first presentationsof words and nonwords this is not a problem.The first time a nonword is presented there

is (presumably) no memory representation,,and the subjects can base their negative re-sponses on the failure to contact a memorytrace. But what about repeated presenta-tions? If the episodic trace from the previouspresentation is accessed, it might be expectedto inhibit a no response because the nonwordhas now gained wordlike qualities: It seemsfamiliar. Thus, even if the episodic trace fromthe prior presentation facilitates the analysisof a nonword letter string, the facilitationcould be offset by the requirement of an in-congruous response. In fact, McKoon andRatcliff (1979) obtained results supportingthis hypothesis: Repeated nonwords were re-jected more slowly and less accurately thannew nonwords in their lexical decision task.

This response confound is particularly dis-turbing because, in other respects, lexical de-cision is ideally suited for exploring the in-dependence question for repetition effects.That is, because lexical decision provides areaction time measure, the method of addi-tive factors (Sternberg, 1966, 1969a, 1969b)could be used to determine whether the ef-fects of repetition are additive to the effectsof the lexicality of an item. Unfortunately,any interpretation of the Word-Nonword XRepetition interaction in lexical decision iscomplicated by the inherent response con-found in the task.

In light of this array of seemingly incon-sistent results and interpretations, it is nottoo surprising that some contention existswith respect to the locus of the repetitioneffect. For example, Morton (1979b) andScarborough et al. (1979) have posited se-mantic memory as the locus of the repetitioneffect. In contrast, Jacoby (1983, in press) hassuggested that the effect derives from memoryfor specific temporal-contextual episodes.Tulving et al. (1982) have proposed yet a thirdmemory system, neither episodic nor seman-tic, to account for repetition effects in theirfragment completion task.

In the present article, we propose a modelthat we believe reconciles some of these dis-parate views of the repetition effect. Themodel is based on the results of several ex-periments conducted with a new task. Thetask we developed combines some of the fea-tures of other tasks, such as lexical decisionand threshold identification, that have pre-


viously been used to investigate repetitioneffects. First, the task yields unconfoundedprocessing time data for the identification ofwords and nonwords. This not only providesan (unambiguous) test of additive factors, butalso confers the advantage of a relative mea-sure of the amount of facilitation for itemsin an identification task. Thus, the probabil-ity of a correct recognition memory responsecan be conditionalized on the relative easeof identification for the same item on a singletrial. Any correlation would indicate someinterdependency between the two processes.

Second, the task is modifiable to allow testsof accuracy of identification at threshold.This modification, then, can be comparedwith the single-trial threshold identificationtechnique. Thus, our new task and its mod-ification allow direct comparisons with boththe lexical decision and identification resultsin the literature and provide a basis for com-parison between the two types of tasks withrespect to performance changes due to rep-etitions.

In the General Method section below, thenew tasks are described. Essentially, they in-volve a slow "fading in" (i.e., increasing thevisual signal-to-noise ratio) of a word or non-word embedded in a field of masks.1 If thedisplay is terminated when the item in thefield of masks is not readily discernable, thetask yields a probability of correct identifi-cation. On the other hand, the display can beterminated by the subject when the item be-comes clear enough to be identified unam-biguously. In this case, a latency, or process-ing-time measure, results.

In Experiment 1, the probability of correctidentification of repeated words and non-words was investigated using both the accu-racy version of the new task and the tradi-tional threshold identification task. Experi-ments 2 and 3 used the latency version of thenew task to address the issue of the relation-ship between lexicality and the effect of rep-etitions. Finally, Experiments 4 and 5 ex-tended the results of the earlier experimentsand investigated the relationship betweenease of identification and recognition mem-ory using the latency version of the task.

In addition, several of the experiments in-clude manipulations of similarity betweenitems on the list. With these manipulations

we aimed to determine the level of processingat which repetitions exert their influence:How closely related do items need to be inorder to obtain facilitative effects betweenpresentations of related items? For example,large repetition, or priming, effects betweenitems that have several letters in commonwould suggest a mechanism for the effectbased solely on overlap of physical features.Conversely, large facilitatory effects betweenitems that are physically disparate but retaina high degree of conceptual identity (e.g.,upper- vs. lowercase instantiations of thesame word) would suggest the involvementof a higher level of abstraction in the repe-tition effect.

General Method

Because the experiments are similar, this GeneralMethod section describes tjie common procedure, ap-paratus, and stimulus-generation techniques. Any devia-tions from the general procedure, and the specific designand stimuli for individual experiments, are describedseparately before each experiment.

SubjectsAll the subjects were undergraduate psychology stu-

dents at Indiana University. They were either paid orwere participating in partial fulfillment of an introduc-tory psychology course requirement. All subjects re-ported normal or corrected-to-normal vision and werenative speakers of English.

Stimuli and Apparatus

The subjects were tested individually. They wereseated in a small testing booth in a dimly lit (.002 fL)room. All stimuli and prompts were presented on a Tek-tronix Type 602 fast phosphor (P-15) display scope. Thescreen was situated about 60 cm in front of the subjectjust below eye level. In addition to the display scope, thebooth contained a 16-button response box, a small dis-play unit for visual feedback, and a microphone. Stim-ulus presentation and response collection were con-trolled by a PDF 11/34 computer located in an adjacentroom.

The display consisted of a row of eight characters (let-ters and/or random dot masks) presented horizontallyacross the center of the screen. The letter characters wereupper- and lowercase roman letters. Both the letters andthe masks were formed by specifying the coordinates ofdots in a rectangular grid 48 dots high X 32 dots wide.The mean number of dots needed to form a single letter

1 The idea for the tasks we used grew out of discussionswith William A. Johnston, who developed a similar taskfor his research on attention.


was 43. Although they were formed from discrete dots,the line segments in the letters appeared continuous. Themasks were formed by randomly positioning 43 dots inthe grid. Five different masks were constructed in thisfashion.

When displayed on the scope, the individual charac-ters subtended about ,33° of visual angle horizontallyand .49° vertically. The spacing between the centers ofthe characters was .94°. Thus, for an eight-characterdisplay (the maximum size used) the total angle sub-tended was 6.91 °. The estimated luminance directionalenergy for each dot constituting the characters was about.5 candle-microsecond (see Sperling, 1971). The back-ground luminance of the screen was .0004 fL.

ProcedureA single experimental trial consisted of three phases:

identification, self-scoring, and in some experiments, rec-ognition. The trials were embedded in a long continuouslist. Depending on the individual experiments, the iden-tification phase of a trial could consist of any one ofthree different display configurations. The first type ofdisplay configuration was the traditional task, which con-sisted of a single brief presentation of a target item im-mediately followed by a mask. This technique will sub-sequently be referred to as discrete threshold identifica-tion (DTI). The second type of display consisted of aseries of DTI display configurations presented in veryrapid succession with the duration of the stimulus itemrelative to the mask increasing by a small amount witheach successive, presentation. The sequence of displayswas terminated when the ratio of the item duration tothe mask duration was such that the item could only-beidentified some proportion of the time (20% to 80%).This technique will be called continuous threshold iden-tification (CTI). The third type of display configurationallowed the collection of latency data. This technique

was identical to the CTI technique except that the displaysequence was allowed to continue until the subjects in-dicated that they knew what the item was. Thus, thedependent measure was the length of time the subjectsallowed the display sequence to continue before theymade their identification responses—a latency measure.This technique will subsequently be called continuousthreshold latency identification (CTLI). The CTLI taskwill be described first.

Each individual presentation of a word and mask dur-ing one portion of one trial of CTLI is called & frame.The frames were of a fixed duration of 100 msec. Themaximum number of frames in a series was 50. This isillustrated across the top of Figure 1. The manner inwhich the content of the display varied with successiveframes is shown in the bottom of the figure. The firstframe consisted entirely of eight masks and served as atemporal and spatial warning signal. At the beginningof the second frame the word was presented for 2 msecand was immediately followed by masks that stayed onfor the remaining 98 msec in the frame. In the thirdframe the word was displayed for 4 msec before it wasmasked, in the fourth, 6 msec, and so on until in the lastframe only the word was presented.

Depending on the individual experiments, the itemswere either left-justified or centered in the row of masks.Item lengths varied from three to eight letters. When theitem was fewer than eight letters long, the remainingletter positions were filled with masks as shown in Figure1. Within the portion of the frame that was occupiedonly by masks, the five different masks were randomlyplaced in the available (nonletter) positions with the re-striction that any single mask was never used more thantwice. Between frames these masks were randomly re-positioned. The phenomenological appearance of thedisplay was one of a rapidly flickering row of constantlychanging dot masks out of which a word would graduallyemerge.

T R I A L

T ( H E

F R A M E

100 J_ JL

H O R D K 9 I J I

ICO -*- J- 100

Figure 1. Schematic representation of the continuous threshold identification task. (Across the top of thefigure is shown the time course [in msec] of a single trial. The bottom half of the figure shows how thecontent of successive frames varied as the trial progressed.)


The subjects' task was to push a button on the re-sponse box as soon as they could identify the word andsimultaneously to say the word aloud. The verbal re-sponses were not recorded. However, the subjects weretold that they were being recorded and that the, record-ings would be checked for accuracy at a later time. Thisprocedure was adopted to ensure that the subsequentself-scoring of the identification errors would be accu-rate. In addition, subjects were monitored to ensure thatthey were responding according to instructions. The sub-jects used the index finger of their preferred hand tomake this initial identification response.

As soon as the response was made, the display wasterminated. After a blank interval of 500 msec, the itemthat had actually been presented under the mask wasdisplayed for 500 msec. This allowed the subjects to ver-ify whether they had correctly identified the item. Im-mediately following the offset of the item, the promptCORRECT appeared on the screen. The subjects then in-dicated whether they had made a correct identificationby pushing one of two buttons labeled yes and no. Anyother response was ignored. On trials for which the sub-jects made no identification response, an error was re-corded and the scoring prompt was omitted. These trialswere eliminated from all subsequent identification timeanalyses. The identification time recorded for each trialwas taken to be the duration of the word in the frameduring which the identification response was made.

Because the CTI and DTI tasks were designed to yielda probability, rather than a latency, measure, the pro-cedure was somewhat different for these trials. For bothtypes of trials the display began with the prompt READYpresented across the center of the screen. When the readysignal was displayed, pushing either the yes or no buttonwould start the trial. When the button was pushed, theready signal was erased and the screen remained blankfor 500 msec. If the trial was a CTI trial, the blank in-terval was immediately followed by the same display se-quence described above except that the display was ter-minated at a predetermined frame. If the trial was a DTItrial, the blank interval was followed by a single presen-tation of the stimulus item and mask. The total displayduration of the stimulus and the mask for the DTI trialswas 100 msec.

In both the DTI and CTI trials, the display of itemand mask was immediately followed by the appearanceof a small question mark in the center of the screen.When the question mark appeared, the subjects at-tempted to identify the item that had been presented bysaying it aloud. Once they made their identification re-sponse, or if they had no guess as to what the item was,they again pushed either of the response buttons. Thisbutton press terminated the question mark, and, 500msec later, the item that had been presented was dis-played for 500 msec so that the subjects could verifywhether they had responded correctly. The screen thenprompted the subjects with the word CORRECT, and theyscored themselves by pushing the appropriate responsebutton. Immediately after the scoring response, the readysignal for the next trial was displayed.

In the experiments for which recognition responseswere required, the computer responded to the button presswith the prompt ON LIST. Here the subjects were asked toindicate, by pushing the appropriate button, whether thepresented item had appeared on an earlier trial on the

list. For these recognition judgments, feedback was pro-vided: A green light flashed for 100 msec for a correctresponse, and a red light for an incorrect response. Thescreen was then erased, and 1 sec later the next trial began.

For the experiments in which the CTLI task was used,the instructions emphasized the speed of identificationbut also encouraged the subjects to make as few, iden-tification errors as possible. An arbitrary criterion of aminimum 85% correct identification, over all trials, wasadopted. 'Subjects who failed to satisfy the criterion werereplaced.

Experiment 1

The CTI and DTI tasks yield the proba-bility of correct identification given a fixedamount of information under discrete orcontinuous presentation procedures. CTLIyields a latency measure and is not directlycomparable to the other two tasks. The firstexperiment was designed primarily to ascer-tain whether the CTI and DTI tasks are tap-ping the same underlying mechanisms. Laterstudies look into the relationship betweenthese two tasks and the CTLI approach.

Aside from the repetition effect itself,1 theCTI and DTI paradigms were compared withrespect to their sensitivity to changes in thephysical appearance of repeated items. Ja-coby and Witherspoon (1982), using casechanges, and Morton (1979b), using hand-versus typewritten words, found that al*though some differences were observed, therepetition effect persisted across such changes.As a check on the validity of the new tech-nique, changes in case were incorporated intothe design of the experiment. In addition,because data on nonword repetition effectshave been inconclusive and few, a between-group manipulation of lexicality (i.e., wordvs. nonword) was made. Thus, the experi-ment provided several variables across whichthe two techniques could be compared.

Method

One group of subjects was tested on a list of words,and another group was tested with nonwords. The wordsand nonwords were presented either two or four times,with the last presentation of any given item always in theopposite case from the earlier presentation(s). Half of thetested items were presented using the continuous (CTI)technique and half were presented with the discrete(DTI) technique.

Subjects. The subjects were 69 students drawn fromthe population described in the General Method section.Five subjects were replaced for failing to meet an iden-


tification criterion of between about 20% and 80% cor-rect for first presentations of itenis on the experimentallists.

Stimuli and procedure. Two groups of subjects weretested: a word group and a nonword group. The nonwordswere derived from the words by replacing a single pseudo-randomly chosen letter with a letter of similar frequencyto form a pronounceable letter string, as judged by threeindependent observers. In all other respects, the list-gen-eration procedure was identical for the two groups. Forthis reason, only the construction of the word list is de-scribed in detail.

In addition to" the word-nonword condition, whichwas between subjects, three other variables were manip-ulated within subjects: number of presentations (two vs.four) of the same items, changes in the case (uppercasevs. lowercase) of the items between repetitions, and thetwo test types (CTI vs. DTI). For these conditions, 16high-frequency (A or AA) words from the Thorndikeand Lorge (1944) frequency counts were chosen.

The CTI and DTI trials were randomly intermixed onthe experimental lists. For any given subject, 8 of the 16words were presented using the CTI task, and 8 were pre-sented with the DTI task. Of these 8, 4 were presentedtwice and 4 were presented four times. At both presentationlevels, the last occurrence of the words was in the oppositecase from the earlier presentations. Thus, for the wordspresented twice, if the first presentation was in uppercaseletters, the second was in lowercase letters and vice versa.Similarly, in the four-presentation condition, the fourthoccurrence of a given word was in the opposite case fromall of the previous occurrences. Two of the 4 wordsat each Presentation X Test-Type combination changedfrom upper- to lowercase and 2 changed from lower- touppercase.

The number of words intervening between repetitionof the same words (lag) was controlled for the criticalitems. Repetitions of 1 of the 2 words at each of the eightcombinations of conditions were presented at a short lagof 1 to 5 intervening words. Repetitions of the other wordwere separated by a longer lag of 21 to 25 interveningwords. The actual number of intervening words was variedpseudo-randomly within the specified intervals so that themean lag fell at or near the center of the interval.

A master list was constructed, which specified the listpositions occupied by each of the 16 words and theirrepetitions. The same master list was used for all thesubjects. Between subjects, the 16 words were rotatedthrough the 16 sets of list positions. Thus, across subjects,each of the 16 words occurred an equal number of timesin each of the combinations of conditions.

The critical words were embedded in a long list of 152total items. The first presentations of the critical wordswere spread as evenly as possible throughout the list inan attempt to minimize any confounding between theeffects of repetitions and list position. In addition, noneof the critical items occurred within the first 20 list po-sitions. This was done to minimize any possible primacyeffects. The primacy buffer and remaining list positionswere filled with 60 high-frequency filler words. The case,number of presentations, and test type of the fillers werecounterbalanced. The lag between repeated fillers wasnot controlled.

An additional set of 32 items was used for the practice

list. The practice list served two purposes. First, it allowedthe subjects to become familiar with the apparatus andprocedure. Second, it served to give an estimate of theidentification thresholds for the individual subjects. Forthe words, the practice exposure time for both the CTIand DTI trials was 36 msec. (For the nonword practicelist, the exposure duration for both tasks was set at 48msec.) These times were chosen because they were nearthe average threshold for most individuals tested in apilot experiment.

Recall that for the DTI trials, an exposure durationof 36 msec means that the word was flashed on the screenfor 36 msec and was immediately covered by masks for64 msec. For a CTI trial, however, an exposure durationof 36 msec means that the continuous display terminatedin the frame where the duration of the word was 36 msec.Thus, the total exposure time of the word was longer inthe CTI task (i.e., 36 + 34 + 32 + . . . + 2 msec) thanin the DTI task (i.e., 36 msec).

The performance of the subjects on, the practice listwas used to estimate their thresholds via an algorithmbased on pilot work. For each subject, separate estimateswere made for the DTI and CTI trials. These (admittedlyrough) estimates were then used as the exposure timeson the experimental list. In spite of the individual thresh-old adjustments, performance on the experimental listvaried substantially between subjects. In order to avoidany ceiling or floor effects, five Subjects who failed to fallwithin a range of 20% to 80% correct identification forthe first presentations of items on the experimental listwere replaced. Recognition responses were not collectedin this experiment.

During the instructions, the subjects were told that theexperimental list would contain repetitions of itenis andthat some items would be presented in uppercase lettersand some would be lowercase. They were not, however,told that repeated items might be presented in differentcases. In addition, the subjects in the nonword groupwere told that they would be seeing pronounceable non-words and that they should treat these nonwords as realEnglish words that they had never before seen. None ofthe subjects had any difficulty pronouncing the non-words during the practice list.

Results and Discussion

Unless otherwise noted, the significancelevel adopted for all statistical tests was .01.An overall analysis indicated that the lagmanipulation had no effect on the identifi-cation probabilities of repeated items, F( 1,62) = 2.79, p > .10. For CTI, short lags re-sulted in correct response probabilities forrepeated items that were .08, higher than atlong lags. The comparable difference for DTIwas .024. For this reason, the data were col-lapsed across the two lags for all subsequentanalyses. The collapsed means are shown inFigure 2. The top figure shows the probabilityof correct identification of words as a func-


tion of the number of presentations. The leftand right panels are for the two-presentation(P2) and four-presentation (P4) conditions.The filled symbols are the results for the DTItask, and the open symbols are for the CTItask. Recall that for both the P2 and P4 con-ditions, the last presentation was always inthe opposite case from the earlier presenta-tions. The bottom of the figure shows the re-sults for the nonword group.

Separate analyses were conducted on theP2 and P4 conditions. The P2 analysis re-vealed a main effect of presentations, P(l,62) = 22.17. Repeated items were more ac-curately identified than first presentations ofthose same items. This was true in spite ofthe fact that the second presentation was inthe opposite case from the first presentation.In fact, for the words tested with CTI and thenonwords for both tasks, there was a smalltrend for the second presentations to be iden-tified more accurately than the second pre-sentations of unshifted words in the P4 con-dition. The effects of test type (CTI vs. DTI)and lexicality (word vs. nonword) were notsignificant (both Fs < 1.0). None of the in-teractions were significant.

Thus, both test types yielded similar re-sults, and the identification probabilities forthe words were not different from those ofthe nonwords. The failure to find any differ-ences for at least the first presentations acrosslexicality and test type was not unexpectedbecause the exposure durations were adjustedfor each subject so that performance wouldbe roughly equivalent across these variables.(However, the differences in the adjustedtimes are, in themselves, of interest, and arediscussed in detail below.)

The analysis for the P4 condition alsoshowed a repetition effect, F(3, 186) = 47.99.Repeated items were more accurately iden-tified than new items. Again, the effects oflexicality and test type were not significant,F(l, 62) = 1.72, p > .19, and F(l, 62) < 1.0,respectively. However, there was a significantinteraction between lexicality and presenta-tions for this condition, F(3, 186) = 3.50,p < .02. This interaction is apparent in Figure2 from the fact that the identification accu-racy rose more rapidly with increasing pre-sentations for the words than for the non-

w n R n g

,80

,60

2 1 2

P R E S E N T A T I O N

K O N H Q R D S

o cn• DTI


Figure 2. The probability of correct identification ofwords (top) and nonwords (bottom) as a function ofpresentations in Experiment 1. (The two panels in eachfigure are for the two- [P2] and four- [P4] presentationconditions. The open symbols are for the continuousthreshold identification [CTI] task, and the closed are forthe discrete threshold identification [DTI] task. In all con-ditions the item presented last was in the opposite case asthe earlier items. The error bars are one standard errorof the mean.)

words. None of the other interactions weresignificant.

Because the Lexicality X Presentation in-teraction was significant, post hoc analyseswere done for the four presentations in thetwo groups. The analyses confirmed that therepetition effect was significant for both thewords and the nonwords, F(3, 93) = 32.49and F(3, 93)= 16.40, respectively. Therewere no differences between the tasks in ei-ther group (both Fs < 1.0). Thus, there werereliable repetition effects for both words andnonwords, and the repetition effect was notdifferentially affected by the two tasks (CTIand DTI).

The small decrease in accuracy at thefourth presentation for both test types in theword group and for the CTI task in the non-


word group indicates that there may be asmall decrement in the repetition effect whenthe repeated items are not physically iden-tical. Unfortunately, unlike the P2 condition,there were no items on the lists that werepresented four times without a shift in case.Thus, it was not possible to determine whetherchanging the case caused a significant de-crease in the repetition effect relative to un-shifted items at,the same presentation level.The apparent decreases observed between thethird, unshifted presentation and the fourth,shifted presentation were not significant (allFs < 1.0). Although it could not be deter-mined whether the change in case caused asignificant decrease in the repetition effectrelative to unshifted items, it was neverthelesstrue that the change did not eliminate theeffects of the prior presentations of the items.Scheffe's paired comparisons showed that theshifted words were identified with greater ac-curacy than first-presented words for both theCTI and DTI tests, F(3,29) = 21.34, p < .02,and F(3, 29) = 33.68, respectively. This wasalso true for the nonwords, F(3,29) = 19.53,p < .03, and F(3, 29) = 25.80, for the CTIand DTI tests, respectively. This result is inagreement with the results of other investi-gators who varied the display characteristicsof repeated words (e.g., Jacoby & Wither-spoon, 1982; Morton, 1979b).

In summary, the two techniques yieldstrikingly similar results for both words andnonwords when the exposure durations areadjusted to equate performance. In addition,although it was found that the magnitude ofthe repetition effect was greater for wordsthan nonwords, there was still a substantialincrease in the probability of identificationof repeated nonwords. Changing the case ofboth the words and nonwords between rep-etitions had at most a small effect on the fa-cilitation due to repetitions.2

The lack of a substantial effect of changingthe,physical appearance of items betweenpresentations suggests that the memory rep-resentation being contacted is more abstractthan a single episodic trace consisting of a"physical copy" of the item. However, the factthat a repetition effect was observed for thenonwords is problematic for certain theoriesthat posit abstract memory representations(e.g., logogens or semantic "nodes") as the

sole locus of the repetition effect. In order fortheories of this sort to account for these re-sults, a single presentation of a novel stimulus(e.g., a nonword) must be sufficient to estab-lish the representation. If this is true, thereseems little point in making a strong-distinc-tion between episodic and semantic memoryrepresentations. Thus, the repetition effectfor the nonwords, coupled with the casechange results, starts to call into question thesemantic node, or logogen, view of the rep-etition effect.

Except for the repetition effect, the datadiscussed so far do not allow any compari-sons to be made between words and non-words. This, of course, is because the expo-sure durations of the items were adjusted in-dividually for the subjects in the two groupsso that performance was roughly equated at50%. However, any large differences betweenthe groups should be manifested in the dif-ferences in the amount of time that the sub-jects needed in order to identify the wordsand nonwords with equal accuracy. In orderto see whether any differences were apparentfor the duration adjustments, the mean ad-justed times and resulting correct identifi-cation probabilities of first presentations oflist items for the two groups were calculated.These data are shown in Table 1 for both theDTI and CTI tasks.

Note, first, that the performance on boththe CTI and DTI tasks was very close to 50%in all cases, as designed. Second, given thealmost equal identification probabilities, therewere large differences in the adjusted timesbetween the words and nonwords for boththe DTI and CTI tasks, f(62) = 12.14 andt(62) = 9,51, respectively. Much more timewas needed to attain 50% accuracy for thenonwords than for the words. This difference,of course, was not unexpected: Many factorscould be contributing to the enhanced iden-tification of words relative to nonwords. Forexample, the identification of constituent ele-ments of words may be perceptually en-

2 In this and subsequent experiments, the performanceon lowercase items was somewhat lower than on upper-case items, presumably because the internal details ofthe lowercase letters were more severely masked. How-ever, although absolute performance levels were lower,the magnitude of the repetition effect did not differ ap-preciably.


hanced by excitatory interaction with preex-isting lexical nodes. This is a "top-down"view of identification. Another possibility isthat the subjects' frequent past experiencewith words has resulted in highly unitizedmemory representations that allow auto-matic responses to words without recourseto analyses of the individual letters (e.g.,Drewnowski & Healy, 1982). The more in-teresting and unexpected finding was that thetwo tasks seemed to require very differentamounts of time for identification of wordsrelative to the nonwords. For the words, theadjusted frame times were shorter for theDTI task than for the CTI task, t(31) = 9.41.For the nonwords the opposite was true: TheCTI task required significantly less time,f(31) = 7.14. This effect was very consistentacross subjects. Out of 32 subjects in the wordgroup, 30 had lower adjusted times for theDTI task. For the nonwords, 28 out of 32had lower adjusted times for the CTI task.Furthermore, the differences in the adjustedtimes were paralleled by opposite (althoughsmall) differences in the identification prob-abilities. That is, for the words, the DTI taskresulted in greater accuracy at shorter timesarid the CTI task yielded lower accuracy atlonger times. Precisely the opposite was truefor the nonwords.

Of course, it could be argued that the ad-justed times shown in Table 1 are not mean-ingful—that within this range of times, frametime would not affect accuracy. If so, no con-clusions from these data would be possible.To test this possibility, additional subjectswere tested with words and nonwords thatvaried in exposure duration within the twolists. This was done to ensure that changesin the exposure times would result in changesin the identification probabilities over therange of times used in the experiment. Ad-ditionally, and more important, the resultswill be used to generate a model for the effectspresented throughout this article.

Experiment la

MethodThe exposure duration of first-presented items was

varied within subjects. One group of subjects was testedwith words, and one was tested with nonwords. For eachgroup, half the items were tested with the CTI and halfwith the DTI technique.

Table 1Mean Adjusted Exposure Times for theSubjects in the Word and NonwordGroups in Experiment 1

Task type

CTI DTI

Group M M

WordNonword

37.7542.44

.45/ '

.5431.1348.25

.53

.48

Note, P = probability of correct identification for firstpresentations of items on the lists; CTI = continuousthreshold identification; DTI = discrete threshold iden-tification.

Stimuli and procedure. Four subjects were tested ona list of 180 words, and four were tested on a list of 180nonwords. The first 20 list positions in both lists servedas a primacy buffer. Within each list, half of the re-maining 160 words were tested with the DTI task, andhalf with the CTI task. Four different exposure durationswere tested for each task. In the word group, the exposuredurations for the DTI task were 24,28,32, and 36 msec.The CTI durations were 30, 34, 38, and 42 msec. Forthe nonword group, the DTI durations were 44, 52, 60,and 68 msec, and the CTI durations were 36, 44, 52,and 60 msec. Twenty items were tested at each exposureduration. The range was chosen to be wider for the non-words because between-subject variability tended to begreater than for the words. A larger range was necessaryto get a reasonably accurate estimate of the psychometricfunction.

The items tested in the two tasks were exclusive. Thatis, an item that was assigned to the DTI test was neverused as a CTI item. However, across subjects, the itemswere used equally often (once) at each duration withineach task. The items were always presented in lowercaseletters. In all other respects, the procedure was the sameas in the previous experiment.'


Figure 3 shows the resulting psychometricfunctions for the two tasks in the two groups.The figure shows the probability of correctidentification as a function of exposure du-ration. The curves drawn through the pointsare the best fitting logistic response functionsfor the data shown.3 The logistic fits will beused later for model development.

3'The logistic response function is commonly used forpsychometric data. It closely approximates the cumu-lative normal distribution but is computationally much •simpler. It is given by

Pi1 exp(/>o


20

E X P O S U R E D U R A T I O N (M S £ 0

Figure 3. The probability of identification as a function of duration for the continuous threshold iden-tification (CTI) and discrete threshold identification (DTI) tasks used in Experiment la. (The circles arefor the word group and the squares are for the nonword group. The filled symbols are for the DTI itemsand the open symbols are for the CTI items. The error bars are one standard error of the mean.)

The psychometric functions confirm andclarify the findings from the adjustment data.First, the words were identified more accu-rately with shorter exposure durations thanthe nonwords for both DTI and CTI trials.Second, the CTItask was more difficult thanthe DTI task for the words, whereas the con-verse was true for the nonwords. Third, it isapparent from the figure that the perfor-mance on the DTI task was affected by theword-nonword manipulation to a muchgreater extent than was performance on theCTI task. This was true not only for absoluteperformance differences between words andnonwords but also for the difference in therate of change of performance with increas-ing exposure duration. The slopes of the wordand nonword functions are more similar forthe CTI than DTI task.

where # is the estimated proportion of positive responsesat x(, and 6p and b, are parameters. The parameters forthe curves in Figure 3 "were, for the DTI words, bo =-4.62, hi = .176; for DTI nonwords, b0 = -3.79,fc, = .06; for CTI words, ba = -6.45, bt - .15; and forCTI nonwords, b0 = -6.49, bt = .131.

Can both the word and nonword effects beaccounted for by a model positing incre-ments in activation of abstract semantic rep-resentations (e.g., logogens) as the sole basisfor repetition effects? We think this would bedifficult. In order to account fbr the nonwordrepetition effect, logogens would have to beestablished on the first presentation of a non-word. However, these new logogens must nec-essarily be weaker than those for words toaccount for the large word advantage: Wordsare always more readily identified than non-words. Furthermore, repetitions of nonwordsmust cause less of an increment in activationof these new logogens, because the repetitioneffect is smaller for the nonwords. If this weretrue, there would then be no obvious way toexplain the Lexicality X Task interaction seenin the exposure durations needed to reachthreshold. We therefore prefer an explanationin which the repetition effect is due to sup-port from episodic memory images, and thelexical difference is due to the availability ofa unitized code for words that can produceautomatic, rapid identification on the basisof a brief sample of near-threshold infor-


mation. In one sense, this interpretation islike a logogen model, except that the newlycreated nonword logogens are not capable ofproducing a unitized identification response.This argument is described in more detailbelow and again in broad theoretical contextin the General Discussion section.

Given that the above analysis is correct,some explanation must be made as to whythe CTI and DTI tasks interact so stronglywith the lexicality of the tested items. Anyexplanation must take into account the dif-ference in the masking characteristics of theDTI and CTI procedures. The DTI task hasno masks prior to the single display. On theother hand, the masks at the end of eachframe in the CTI task are forward masks forthe letter strings at the start of the succeedingframe. Forward visual masking causes deg-radation of the displays and hence shouldcause an overall increase in difficulty for theCTI task relative to the DTI task.

The possibility that masking differencesbetween the two tasks may be largely re-sponsible for the Task X Lexicality interac-tion is suggested by related findings in re-search on the "word superiority effect." Theword superiority effect refers to the findingthat letters are more easily identified whenthey occur in a word context than when theyare presented alone or in a nonword (e.g.,Reicher, 1969). This advantage, however,seems to be largely dependent on the visualcharacteristics of the display (Johnston &McClelland, 1973;Massaro&Klitzke, 1979).With high-contrast displays and a patterned,backward mask, the effect is quite large. Onthe other hand, the advantage is largely elim-inated when the displays are low in contrastor the targets are otherwise indistinct.

In light of these findings, we suggest thatthe Task X Lexicality interaction is due tomasking differences between CTI and DTI.In the CTI task it might be expected thatperformance is limited by the quality of theavailable information in the display. That is,the fact that the presented item is a word isnot helpful if the constituent letters cannotbe discerned because of forward masking. Inthe DTI task, however, the letters are sharply(although briefly) presented. Here the limit-ing factor is not the quality of the display butrather the time available in which to identify

or label the presented string before the sen-sory impressions of the constituent elementshave faded. Thus, the advantage for wordsmay be due to an automatic code generatedfrom a unitized memory representation, anadvantage greatly accentuated when only asingle brief stimulus display is available. Inthe CTI task, although the quality of the dis-play is poorer, the items are presented againand again for longer periods of time so mem-ory limitations are less of a contributing fac-tor: Features forgotten after one frame canbe reinstated in the next. What becomes im-portant, then, is the discriminability of theindividual letters in the display. Thus, for theCTI task, the word-nonword difference ismuch smaller.

Whether or not the above analysis is cor-rect, the Task X Lexicality interaction sug-gests intriguing possibilities for use as a re-search tool. That is, it should be possible todetermine whether a tested item is contactingan integrated, unitized memory representa-tion (i.e., a node or logogen) by determiningwhether threshold identification of the itemis more likely under DTI or CTI conditions.4

As exciting as this possibility is, it is periph-eral to the present purposes and is not pur-sued in this article.

In summary, four important conclusionscan be drawn from the results discussedabove. First, given appropriate adjustmentsin overall exposure duration, the CTI and DTItasks produce similar results as a function ofrepetitions and lexicality. Therefore, the fac-tors producing the sharp differences betweenCTI and DTI (presumably the rapid forget-ting of features in DTI) probably do not in-teract strongly with factors producing the rep-etition effects. Second, changes in the phys-ical appearance (case) of repeated items hadonly a small effect on identification facilita-tion, precluding any process based solely onthe physical identity of repetitions as the lo-cus of the effect. Third, large repetition effectswere found with repeated nonwords, and

4 Suppose, for example, that one wished to determinehow many presentations of a novel item, under whatkind of presentation schedule, were necessary to establisha unitized trace that can respond quickly and readily. Bypresenting the items in question and testing equivalentitems under CTI and DTI procedures, one could deter-mine the answer to this question.


there was also a large interaction between lex-icality and the CTI and DTI tasks (in termsof the exposure durations required for equiv-alent performance). These results are difficultto reconcile with the idea that the action ofthe repetition effect is on the activation of alexical memory representation, independentof contributions from individual episodicmemory images. Fourth, even though bothwords and nonwords showed repetition ef-fects, the words were always identified moreeasily than the nonwords, suggesting the pos-sibility of a unitized identification responsefor the words. '

The fact that words are identified moreeasily than nonwords, although both showrepetition effects, suggests that these two ef-fects result from the operation of differentprocesses. The CTLI task may be well suitedfor exploring this possibility because it yieldsa measure indicative of processing time.Thus, it should be possible to determine ifthe effects of lexicality and repetitions areadditive. As Sternberg (1966, 1969a, 1969b)has pointed out, additivity of processingtimes suggests the involvement of differentcontributory stages or processes.

Another important and related questionconcerns the stage of processing at which rep-etitions have their effect. We have alreadyargued that increments to logogens by them-selves do not explain the effect. On the otherhand, the small effects of case shifts seem torule out facilitation due to residual activa-tions only at the physical feature level. If thelocus of the repetition effect is not at the levelof physical features or logogens, where is it?

One possibility is that the facilitation oc-curs at the level of individual letters or con-figurations of letters. However, because theexact shape of the letters (i.e., upper- vs. low-ercase) is of little consequence to the repe-tition effect, it is perhaps better to adopt thephrases letter name and letter cluster whenreferring to these levels of analysis. Thephrase letter name is used here to indicatethat the distinction between individual lettersis preserved, whereas the exact physical shapeof the letters is not.

Some evidence for an encoding level withthese characteristics has already been found.Jacoby and Witherspoon (1982) have re-ported that although switching the presen-

tation modality of words from auditory tovisual virtually eliminates any identificationfacilitation, the repetition effect is reinstatedif the subjects are required to spell the au-ditorily presented words aloud. It seems ,asif it is enough to make the subjects aware ofthe letters that constitute a word in order todemonstrate a repetition effect.

In another experiment, using a discretepresentation technique, Murrell and Morton(1974) found partial identification facilita-tion for words that were similar in spellingto previously presented words but only whenthe spelling overlap was accompanied by acommon base morpheme (e.g., "bored" and"boring"). When there was no morphemicoverlap (e.g., "bored" and "born"), there wasa trend toward facilitation, but it was not sig-nificant. In that experiment, the overlap inspelling was always at the beginning of thewords, and nonwords were not tested. Ex-periment 2 was therefore designed, in part,to elaborate on these results. An overlap con-dition similar to that of Murrell and Mor-ton's was included to investigate the possi-bility of identification facilitation at the letter-name or letter-cluster level.

Experiment 2

Experiment 2 used words and nonwordsthat were different but shared letter clustersto study the role of spelling similarity on therepetition effect in word identification. TheCTLI task was used and identical repetitionsof words and nonwords were! included, sothat the question of additivity between therepetition effect and lexicality could be pur-sued.

Method

Two groups of subjects, a word group and a nonwordgroup, were tested using the CTLI technique. For bothgroups, some of the items presented differed by one ortwo letters from items that had been presented twice inearlier list positions. The locus of the different letters waseither at the beginning or ending of the items. The iden-tification times of these "overlapping" items were com-pared with those for items that were unrelated to anyearlier presentations and to those for third presentationsof identical items.

Subjects. The subjects were 42 students from thesame population described earlier. Of these, 19 weretested in the word condition and 23 were tested in thenonword condition. Three of the subjects tested with


words were replaced for failing to reach the 85% criterionfor correct identification. Fot the nonword condition, 7such subjects were replaced.

Stimuli and procedure. For the word condition, 32pairs of related words between three and seven letterslong were generated. Of these 32 pairs, there were 8 ineach of four conditions: Bl, B2, El, and E2 (B for be-ginning; E for ending). In the Bl condition, one letterwas added to or deleted from the beginning of one of thepair members to form another, semantically distinct,word (e.g., SHIP and HIP). In the B2 condition, the twowords from the pairs differed by two letters at their be-ginnings (e.g., TROUT and OUT). In conditions El andE2, similar manipulations were performed at the endsof the words. These four conditions will subsequently becalled the overlap conditions.

In all cases the word pairs were carefully chosen sothat they did not share a common morpheme. Thus,none of the pairs differed only by the addition or deletionof a prefix or suffix. This was done so that any identi-fication facilitation that might occur between the wordpairs could be attributed to their physical or ortho-graphic similarity rather than to semantic relatedness.

The same overlap conditions were used in the nonwordlist. The nonwords were generated from the words bychanging one letter that was common to both membersof an overlap pair.- All of the nonwords generated in thisfashion were judged pronounceable by three independentobservers. The shorter of the pair members for both thewords and nonwords varied between three and five let-ters.

The word-nonword variable was tested between sub-jects. The lists used in the word and nonwqrd groupswere generated identically. Within each list, four of thepair members for each letter-overlap level were used ascontrol and four were used as experimental items. Forthe control items one member of each pair appearedthree times on the list. Of the four control items, twowere the shorter member of the pair and two were thelonger member. One of these two was randomly assignedto a short lag of 10 to 15 intervening items betweenrepetitions and one was assigned to a long lag of 40 to50 items.

The experimental items were treated identically to thecontrols except that on the third presentation the other,related, member of the, pair of items was presented. Forconvenience, the item from a given related pair that waspresented first on the list will subsequently be called theinitialitem. The other pair member, which was presentedlater in the list, will be called the derived item. The wordfrom the pair that was used as the initial word was coun-terbalanced across subjects. Between subjects the pairswere rotated through the experimental and control con-ditions.

An additional 60 filler items, presented once or twice,were used. The total list length was 216 items, and thefirst 20 list positions included none of the experimentalor control items. A practice list of 20 items was givenprior to the experimental list. For the subjects in theword group, the practice list consisted of 20 proper nouns(country names). The nonword subjects received a prac-tice list of pronounceable nonwords derived from thecountry names.

All the items were presented in uppercase letters. Be-cause the overlap between the initial and derived items

occurred both at the beginnings and ends of the words,the items were centered in the field of eight masks. Inall other respects, the procedure was identical to that ofExperiment 1.


For the word group, the identification re-sults for correct responses in the four overlapconditions are shown in Figure 4. The topfigure shows the identification times for theexperimental and "control words. The fourpanels are for the four different overlap con-ditions. In the figure the two presentationsof the initial items from the experimentalpairs have been pooled together with the firsttwo presentations of the control items forclarity. The derived items and the third pre-sentation of the control items are shown sep-arately. The bottom figure shows the errorrates for these same words. The comparabledata for the nonword group is shown in Fig-ure 5.

A preliminary analysis showed no effectfor the lag manipulation, F(l, 30) = 3.64,p > .07. On the average, the repetition effectfor words repeated at long lags was slightlylarger (1.13 sec) than for repetitions at shortlags. For the nonwords, short lags resulted ina larger (1.38 sec) effect than long lags. There-fore, the data were collapsed across the lagvariable for subsequent analyses.

The control data were analyzed first. Inthis and all subsequent analyses, only theidentification times for the correct responseswere included. Because the overlap variablewas irrelevant for the controls (they wereidentical repetitions of the items), the controldata were collapsed across the overlap con-ditions. In the overall analysis for the iden-tification times, there were main effects ofrepetition, F(2, 60) = 99.85, and lexical sta-tus, F\l, 30) = 5.14, p < .04. The effect ofrepetitions was due to the fact that the iden-tification time decreased with successive pre-sentations of identical items. Post hoc anal-yses confirmed that this was true for both thewords, F(2, 30) = 36.47, and nonwords, F(2,30) = 67.19. The main effect of lexical statusreflects the finding that the words were morerapidly identified than the nonwords. Therewas no interaction between repetitions andlexical status, F(2, 60) = 1.22, p > .30. Asimilar analysis for the control error rates


showed no effect of repetitions, F(2, 60) =2.01, p > . 14. The effect of lexical status andthe interaction were also not significant forthe errors (both Fs < 1.0). Thus, the repeti-tion effects found in Experiment 1 were rep-licated for both the words and nonwords us-ing the CTLI task.

Next, the identification times for the ex-perimental items were analyzed. Of course,the interesting comparison here is betweenthe first presentation of the initial item andthe single presentation of the derived item.For this reason, the second presentations ofthe initial items were not included in any ofthe analyses for the experimental condition.

The analysis showed that there were nosignificant differences between the four over-lap conditions for the identification times,F(3, 90) < 1.0. Thus, whether the initial andderived items differed by one or two lettersand the location of the intact letter clustersthat they shared did not affect the facilitationin identification time. Further, no differences

were found between derived items that con-sisted of additions or deletions of their initialitems, F(l, 30) < 1.0. However, the deriveditems were identified significantly faster thanthe initial items overall, F(l, 30) = 9.96, and,as with the control items, the words weremore rapidly identified than the nonwords,F(l, 30) = 4.69,;? < .04.

Because the effect of lexicality was signif-icant, separate post hoc analyses were con-ducted on the word and nonword data todetermine whether the facilitation found inthe overall tests was present for both thewords and nonwords. These analyses showedthat the derived nonwords were identifiedmore rapidly than the initial nonwords, F(\,15) = 7.18,p < .02. However, the same com-parison for the words fell short of signifi-cance, F(\, 15) = 2.99, p > .09. This resultis very similar to Murrell and Morton's(1974) finding for words that did not sharea common base morpheme. That is, for thewords, spelling overlap produced only a non-

k

E2i i i

1 2 1 2 3 1 2 3P R E S E N T A T I O N

r ' 1. , 1

f ,ft A9 9

i i i i i iv;

. i i i1 2 3 1 2 3 1 2 3 1 2 3

P R E S E N T A T I O N » IDENT.CALO DERIVED

£O.Qt

Figure 4. Identification results for the words in Experiment 2. Top: The four panels show the identificationtimes for the four overlap conditions. (In this and all subsequent figures, identification time refers to theamount of time [in msec] the stimulus item occupied in the frame during which the identification responsewas made. Within each panel, the closed symbols include the identical repetitions of the control wordsand the first two [identical] presentations of the experimental words. The open symbol is for the derived[experimental] word. The error bars are one standard error of the mean.) Bottom: Error probabilitiesassociated with the identification times.


£ 58

"• 56Z

° M

E 50I—

5 18

S LEl

1 2 3 1 2 3P R E S E N T A T I O N

ffi0

UJ

o .05

£ 0.0

-

- K^»• i '.*:

f-4>* *V4i •

f / ti i i

1 2 3 1 2 3 1 2 3 1 2 3 ,

P R E S E N T A T I O N » IDENTICALO DERIVED

Figure 5. Identification results for the nonwords in Experiment 2. Top: The four panels show the iden-tification times for the four overlap conditions. (Within each panel, the closed symbols include the identicalrepetitions of the control nonwords and the first two presentations of the [identical] experimental non-words. The open symbol is for the derived [experimental] nonword. The error bars are one standard errorof the mean.) Bottom: Error probabilities associated with the identification times.

significant trend toward facilitation for thederived items.

Although the facilitation for the wordsfailed to reach significance, in all cases thetrend was for the derived words to be iden-tified more rapidly than the initial words(Murrell & Morton, 1974, have also founda trend in this direction). This pattern of re-sults suggests that the kind of similarity thatexists between words that share letter clustersmay play some part in the large facilitationthat is typically found between repetitions ofidentical items. However, it is also apparentthat the shared letter clusters cannot entirelyaccount for the repetition effect. Similar re-sults and conclusions apply to the nonworddata.5

The error analyses revealed more errors forthe derived items relative to the first presen-tation of the initial items for both words andnonwords, F(l, 15) = 6.64, p < .03, and F(l,15) = 5.28, p < .04, respectively. As was truefor the identification times, the error rates forboth the words and nonwords were unaf-

fected by the overlap manipulation, F < 1.0and F(3, 45) = 1.42, p > .25, respectively.

Thus, although the derived items that werecorrectly identified enjoyed some facilitation,there was a concurrent elevation in the prob-ability of an identification error. This wasprobably due to a tendency for the subjectsto identify incorrectly the derived items asthe previously presented initial item. Unfor-tunately, because the responses were not re-corded, the only basis for this hypothesis isanecdotal. However, if this interpretation iscorrect, then the rise in errors for the deriveditems is further evidence that the identifica-tion process is contacting the episodic tracesleft by the prior presentations of the initial

5 If the slight difference in facilitation for the wordsand nonwords is real, it could be argued that the differ-ence is due to inhibition between words that does notexist for nonwords. Although the present data are tooweak to draw any strong conclusions concerning thispoint, such inhibitory connections between words areprobably fairly weak.


items. This is especially true for the derivednonwords, because the initial nonwords pre-sumably are less likely to have a stored se-mantic representation.

Although the overall error analysis showedno differences between the four overlap con-ditions, it is apparent from Figure 5 that therise in errors for the derived nonwords wasdue entirely to the two conditions in whichthe nonoverlapping portion of the items wasat the ends. In fact, for the El condition, therise in errors was significant, F(3,12) = 6.56,p < .05, and for the E2 nonwords, it ap-proached significance, F(3, 12) = 3.44, p <.06. There were no significant differences forthe JB1 and B2 conditions (both Fs < 1.0).This differential effect on 'errors, which wasnot observed for the words, is probably dueto the way in which the subjects were ana-lyzing the nonwords. That is, if the subjectswere analyzing the nonwords from left toright, in a letter-by-letter fashion, more errorswould be expected for items that differedfrom earlier items only in the later letter po-sitions. Presumably, for the nonwords thatdiffered at initial letter positions, the first let-ters analyzed could be used to eliminate theearlier presented similar item from the set ofpossible responses. By the same reasoning,the lack of a similar^ differential effect for thewords suggests that they are processed in aunitized fashion rather than serially from leftto right (see Drewnowski & Healy, 1982, fora similar interpretation of frequency effectsin identification).

In summary, there were large repetitioneffects for both words and .nonwords. In ad-dition, partial facilitation of (derived) non-words was observed when they were precededon the list by (initial) nonwords that shareda common letter cluster, suggesting that atleast some of the facilitation is occurring atthe level of letters or letter clusters. A similartrend was found for the words, although itfailed to reach significance. Concurrentlywith the partial facilitation of identificationtimes, there was a rise in identification errorsfor the derived words and nonwords in mostconditions, suggesting that the episodic tracesof the prior occurrences of the initial itemscontributed to the identification of the de-rived items.6 This interpretation is especiallyconvincing for nonwords, where no preex-

58

250

42

• NONWORDS

• WORDS

P R E S E N T A T I O NFigure 6. Identification times for the control words andnonwords in Experiment 2. (The data have been col-lapsed across the four overlap conditions. The error barsare one standard error of the mean.)

perimental semantic representation is avail-able.

The issue of additivity between repetitionsand lexicality was examined next. Figure 6shows the identification times for the threeidentical presentations of the control wordsand nonwords collapsed across the overlapconditions. Note that the two curves are verynearly parallel. As noted above, the interac-tion fell far short of significance, F(2, 60) =1.22, p > .30. That is, the effect of .lexicalityremained constant across presentation levels:The words were identified with about 6 msec(3 frames) less exposure than the nonwords,regardless of the presentation level. This ad-

6 Of course, the rise in errors makes the interpretationof the latency results problematical: The possibility ofa speed-accuracy trade-off cannot be discounted. How-ever, in an analysis of only those subjects with very lowerror rates (less than 3%), no change was observed in thefacilitation effect for derived items. Furthermore, in Ex-periment 4, no rise in errors was found under similarconditions (see Figure 8).


ditivity suggests that different processingstages are involved in the two effects.

Why were the effects of repetition and lex-icality additive in this experiment and inter-active in Experiment 1? The answer may liein the differences between the tasks in the twoexperiments. In Experiment 1 the exposuredurations were held constant, and the prob-ability of correct identification was the de-pendent variable of interest. Thus, the sub-jects' responses were based on only the partialinformation available in a very brief display.In the present experiment, however, the sub-jects were holding their identification prob-abilities relatively constant near 100%, andidentification time was the dependent vari-able of interest. Thus, their responses werebased not on partial information but on suf-ficient information to allow positive identi-fication. The implication of the difference inthe tasks will be discussed more fully in theGeneral Discussion section, where a modelof word-nonword repetition effects is elab-orated.

Experiment 3

The results discussed so far provide evi-dence that the effects of repetition and lexi-cality are derived from the operation of twodifferent processes. However, one issue espe-cially needs further consideration. In the in-troduction, it was argued that the failure tofind repetition effects for nonwords in lexicaldecision tasks (Forbach et al., 1974; Scar-borough et al., 1977, 1979) was probably dueto a confounding between lexicality and thedifferent responses required for words andnonwords in a lexical decision task (e.g.,McKoon & Ratcliff, 1979). There is, however,another difference between the CTLI taskused here and the lexical decision task. Thatis, lexical decision necessarily involves theuse of mixed lists of words and nonwords.

In Experiments 1 and 2 the lexicality ma-nipulation was between groups using sepa-rate, unmixed lists. It is possible that the useof unmixed lists might have allowed the sub-jects to adopt special strategies for identifi-cation of words and nonwords that were notavailable when the lists were mixed. For ex-ample, the subjects in the word group in Ex-periment 2 knew that only a word response

was acceptable. They could have "guessed"appropriate words on the basis of partial in-formation. However, this strategy would notbe possible in a mixed list because partialinformation about presented items wouldnot be sufficient to specify a single item oreven a set of items: Subjects would have towait until every letter was clearly visible be-fore they could decide that an item was aword or a nonword. Thus, the word-non-word difference might be eliminated in amixed list. On the other hand, a model basedon readily available, unitized responses forwords would predict that mixed lists wouldyield results similar to those in the unmixedlists. Experiment 3 tested these possibilities.In addition to the mixing of words and non-words, changes in case between repetitionswere included in the design of Experiment3 to ensure that the results obtained for caseshifts in the CTI task of Experiment 1 wouldgeneralize to the CTLI task here.

MethodMixed lists of words and nonwords were used in a

CTLI procedure. Each of the words and nonwords werepresented four times. For half the items, the last presen-tation always displayed letters that were in the oppositecase from the earlier presentations.

Subjects. The subjects were 18 students from thesame population used in the previous experiments. Twosubjects were replaced for failing to reach the 85% cri-terion for correct identification.

Stimuli and procedure. From the stimuli used in Ex-periment 2, 32 nonwords, together with the words fromwhich they were derived, were chosen. These 32 pairswere divided into two exclusive sets of 16. The nonwordsfrom one group of 16 and the words from the other groupserved as critical items in the two lexicality conditions.

Of the 16 items in each group, half served as exper-imental items and half were used as controls. Both thecontrol and experimental items were presented a totalof four times on the list. For the experimental words andnonwords, the fourth presentation of any given item wasalways in the opposite case from the previous three pre-sentations, as in the P4 condition in Experiment 1. Fourof the eight experimental items (chosen randomly)switched from upper- to lowercase, and four switchedfrom lower- to uppercase. For the controls, all four pre-sentations were in uppercase letters for half of the itemsand in lowercase letters for the remaining half. The itempairs were rotated through the conditions between sub-jects so that each pair contributed to the word and non-word conditions an equal number of times. The itemsalso served equally as experimental and control items,

As in the earlier experiments, a master list specifyingthe list positions of the items in the various conditionswas constructed and the same master list was used for


all subjects. The lag between presentations of criticalitems was varied pseudo-randonily within an interval of10-30 intervening items. An additional 24 pairs of wordsand nonwords were used as filler items. The case andnumber of presentations of the fillers were counterbal-anced. The lag between fillers was not controlled.

The resulting list length was 168 items with the first20 items constituting a primacy buffer. The proportionsof words to nonwords and critical to filler items were thesame in the first and second halves of the list. In addition,the critical items used in the various combinations ofconditions were evenly distributed between the two listhalves. A 20-item practice list, using the same words andnonwords as the practice lists in Experiment 2, was givenat the beginning of each session. The instructions werethe same as in Experiment 2 except, of course, that thesubjects were made aware of the fact that both wordsand nonwords would be presented.


The identification times and error rates asa function of presentations are shown in Fig-ure 7. In the figure the first three presenta-tions of the experimental items (which didnot change case) are pooled with the controlitems. On the fourth presentation, the (ex-perimental) items that changed case areshown separately from the control items thatdid not.

Analysis of variance revealed significanteffects of both lexicality, F(l, 15) = 64.17,and presentations, F(3, 45) = 162.53. Thewords were identified more rapidly than thenonwords, and identification became fasterwith repetitions. Thus, the major results ofExperiment 2 were replicated. In addition,however, there was a significant interactionbetween lexicality and presentations, F(3,45) = 25.53, which was not found in Exper-iment 2. This interaction is apparent in thefigure from the fact that the decrease in iden-tification time with repetitions was larger forthe nonwords than for the words. This dif-ference is especially obvious between the firstand second presentations. If anything, themixing of words and nonwords has increasedrather than decreased the difference in per-formance between them.

The other comparison of interest is thatbetween the shifted (experimental) and un-shifted (control) fourth presentations for thewords and non^vords. Any differences be-tween these points would indicate an effectof changing the physical appearance of theitems between repetitions. Planned compar-

isons indicated that there was a significantdecrement in the repetition effect with caseshift only for the nonwords, F(3, 12) = 4.94,p < .05. A similar trend for the words fellshort of significance, F(3, 12) = 2.37, p >.10. There were no other significant effects.

Although highly variable, the errors par-*

62

58

50

• IDENTICAL WORDS• IDENTICAL NCNWORDSO SHIFTED WORDSD SHIFTED NCNWORDS

2 3


.25

.20

.15

.05-

Q.9-1 2 3 4


Figure 7. Top: Identification times for the experimentaland control words and nonwords in Experiment 3. (Thefilled symbols include the control items and the first three[identical] presentations of the experimental items. Theopen symbol is for the fourth presentation of the ex-perimental items that were in the opposite case from theearlier items. The error bars are one standard error ofthe mean.) Bottom: Error probabilities associated withthe identification times.


alleled the identification times to a large ex-tent. There were more errors for nonwordsthan for words, F(l, 15) = 10.79. Similarly,fewer errors were made, in general, asthe number of presentations increased,F(3, 45) = 8.79. There were no other signif-icant effects for the errors.

These results confirm much of what wasfound in the first two experiments. First, al-though there is some indication that the phys-ical appearance of the item is affecting its easeof identification, it is obvious that the iden-tification of the case-shifted items is stillgreatly facilitated relative to the first presen-tations of items on the list. Second, mixingthe list did not eliminate the difference be-tween words and nonwords. In fact, if any-thing, the nonwords enjoyed an even largerfacilitation, due to mixing, than did thewords. Thus, the repetition effects themselvesand the word-nonword differences cannot beattributed to the use of unmixed lists.

One new result is the finding that the rep-etition and lexicality effects were not additivefor the mixed lists. The interaction is dueprimarily to the elevated identification timesfor the first presentations of the nonwordsand the low identification times for the fourthpresentation of the nonwords, relative to thewords. In other respects, the identificationtimes are quite comparable to those foundin Experiment 2. The effect of the fourthpresentation may well be due to a floor effectfor the words. The reason for the elevatednonword identification times at the first pre-sentation is somewhat more difficult to ex-plain. Because words and nonwords weremixed, the set of possible candidates available(in memory) for the first presentations ofnonwords would include any words that theyresembled. The activation of similar wordsmight compete with the mechanism respon-sible for producing a response to the non-word. On the second presentation, however,an episodic trace of the first presentation hasbeen established. Because this trace exactlyresembles the test item, a nonword responseis now available, and competition from sim-ilar words might be less severe. In conditionssuch as Experiment 2, such competitionmight not occur for the nonwords becauseany similar word candidates could be elim-inated from consideration on an a priori ba-

sis: The subjects know that there are no wordson the list. Of course, the notion that a non-word response is available from an episodictrace does not imply that this response occursautomatically in a unitized fashion. Psycho-metric functions would have to be examinedusing CTI and DTI techniques at presenta-tions beyond the first to test this possibility.7

In summary, the results from the first threeexperiments suggest that episodic represen-tations of items play a large part in the rep-etition effect observed in identification. Theinteraction observed in Experiment 3 some-what weakens the force of the conclusionsreached from the data of Experiment 2 be-cause additivity was not observed. Nonethe-less, the weight of evidence in all the studiessupports the notion that somewhat differentfactors produce the word-nonword differ-ence and the repetition effects. We hypoth-esize that an automatic generation of a uni-tized code is responsible for much of theword-nonword difference, whereas contri-butions from episodic traces are responsiblefor much of the repetition effect.

Experiment 4

The word-nonword results from the firstthree experiments provide evidence thatmemory for individual events plays a role inthe identification process. However, this is notthe only method of addressing the issue ofepisodic involvement in word identification.Recently, Jacoby (1983, in press) has providedcorroborative evidence of a different sort. Aswas mentioned in the introduction, his ap-proach has been to look for effects on iden-tification accuracy of variables that haveknown effects on recognition memory. If the

, identification process involves memory for in-dividual events, then changes in identificationperformance for repeated items should be ac-companied by parallel changes in recognitionmemory, a task that necessarily involves accessto single episodes.

7 In subsequent work we have collected such psycho-metric functions comparable to Figure 3, for presenta-tions after the first. At least for the first three presenta-tions, the strong Task X Lexicality interaction remains,suggesting that a strongly unitized response to the non-words in DTI and CTI tasks takes longer than three pre-sentations to develop.


The present experiment investigated thepossibility of parallel effects on the two tasksfor the overlap variable of Experiment 2. Ifthe partial facilitation for the derived itemsin Experiment 2 was due to the prior episodicoccurrences of the initial items, then it mightbe expected that the derived items would alsotend to elicit more false alarms in a test ofrecognition memory. Furthermore, if epi-sodic memory is important for both tasks,then both the partial facilitation effect inidentification and any tendency to make falsealarms to the derived items in recognitionshould change in a similar fashion as thenumber of prior occurrences of the initialitem changes. For this reason, the number ofpresentations of the initial item before thederived item was varied in Experiment 4.

Method

Subjects viewed a long list of words presented with aCTLI procedure. Immediately following the identifica-tion response for each word, the subjects made recog-nition judgments as to whether the word had been pre-sented on an earlier trial on the list. Embedded in thelist of words were initial-derived pairs as in Experiment2. The number of presentations of the initial word priorto the presentation of the derived word varied betweenone and four.

Subjects. The subjects were 35 students from thesame population as the earlier experiments. Three sub-jects were replaced because they failed to reach 85% ac-curacy in their identification responses.

Stimuli and procedure. As described in the GeneralMethod section, a continuous-identification/recognitionparadigm was used in this experiment. Thus, both rec-ognition accuracy and identification times were the de-pendent variables of interest. The'stimuli were 32 pairsof words with the same relationship as the word pairsin the El condition of Experiment 2. That is, the twomembers of the pair overlapped in spelling-except for asingle final letter (e.g., HUG, HUGE). The length of theshorter word in the pair was always three letters. Fromthese 32 pairs, 16 were selected pseudo-randomly foreach subject with the restriction that, across subjects, all32 pairs were used an equal number of times.

The overlap condition was similar to the overlap con-dition of Experiment 2, except that the number of rep-etitions of the initial word was varied between one andfour before the derived word was introduced. Thus, thetotal number of presentations of either member from agiven pair varied between two and five with the last pre-sentation switching to the derived pair member. Four ofthe 16 pairs were assigned to each of these four presen-tation levels. Within-a given pair, the case in which thetwo words were presented was always the same. However,in each presentation level, 2 pairs were presented in up-percase and 2 in lowercase letters. The presentation orderof the pair members was counterbalanced as in Exper-

iment 2. The lag between pair members in both condi-tions was held constant within an interval of 20 to 25intervening items.

An additional 36 words, counterbalanced for case andnumber of presentations, were used as fillers. The lengthof the resulting experimental list was 136 items. The first20 positions on the list included only filler items. Thesubjects were given a practice list of:15 proper nouns(men's names). During the instructions the subjects weretold that half of the words would be in uppercase lettersand half in lowercase letters. In all other respects theprocedure was identical to that of Experiments 2 and 3.


Identification. The identification timesand error rates for the overlap condition aredisplayed in Figure 8. The four panels in eachfigure are for the four presentation levels: 2,3,4, and 5 total presentations of either of thewords from the pair. In each panel the lastpresentation is of the derived word.

As in Experiment 2, the critical compari-son for these means is between the first pre-sentations of the initial and the counterpartderived words. The analysis of these timesshowed that there was an ovejrall differencebetween the identification times for the initialand derived words, F(\, 31) = 4.77, p < .04,as well as a significant difference between thepresentation levels, F(3, 93) = 3.41, p < .03.More important, however, the interactionwas also significant, F(3, 93) = 6.02. That is,the number of prior presentations of the ini-tial word had a differential effect on the iden-tification facilitation seen with the derivedword. In general, relatively more facilitationoccurred for the derived word as the numberof presentations of the initial increased. How-ever, the difference between the identificationtimes for the first presentation of the initialword and the derived word was significantonly when the initial word was presentedthree, F( 1,30) = 7.16, p < .04,] or four times,F(l, 30) = 6.56, p < .05, prior to the occur-rence of the derived word.

The error data are shown in the bottompanel of Figure 8. The analysis comparingthe errors for the first presentation of the ini-tial words and the derived words showed noeffect between the pair members and no effectfor the presentation level (both Fs < 1.0),The interaction was also not significant,£1(3,93)= 1.34, p> .26.

Thus, there was no elevation in the errorrates for the derived word relative to the first


presentation of the initial word. This was trueeven for the three and four presentation levelswhere significant facilitation was found foridentification of the derived words: UnlikeExperiment 2, there was no indication of aspeed-accuracy trade-off for these data.

The fact that the identification times werenot significantly facilitated for the derivedwords when the initial word had been pre-sented one or two times (or for the words inExperiment 2) indicates that the facilitationbetween visually similar words may be some-what fragile. Nevertheless, it is clear that theeffect is real and that as the number of rep-etitions of the initial word increases, the iden-tification of the derived word becomes faster.

In summary, the results replicate the find-ing from Experiment 2 that the partial facil-itation of the derived words stems from theprior presentations of orthographically sim-ilar (initial) words. Of course, the partial fa-cilitation observed with these related wordsis always much smaller than that observedwith repetitions of identical items, presum-

ably because the initial and derived wordsare, after all, not identical.

Recognition. The false alarm probabilitiesfor the words/in the overlap condition areshown in Table 2. Only the false alarms forthe first presentation of the initial word andthe presentation of the derived word, at eachlevel, are tabulated. These data include thewords for which identification errors weremade, because the recognition judgmentswere always made for a clearly presenteditem. The mean hit rates for identical repe-titions of the initial words were .862, .977,and .945 for the second, third, and fourthpresentations, respectively. An analysis per-formed on these false alarms yielded a sig-nificant effect on recognition between the ini-tial and derived words, F(l, 31) = 22.81.Neither the presentation level nor the inter-action was significant, F(3,93) < 1.0 and F(3,93) = 1.07, p > .36, respectively. Thus, thesubjects were much more likely to say, in-correctly, that the derived word had beenpreviously presented than they were to say,

52

H 50

u 18

- L


S ,15O

«.BUJ

Q .05

.£0,0

1 2 3 1


1 2 3

Figure 8: Identification results for the overlap condition of Experiment 4. Top: Identification time for theinitial and derived words at the four presentation levels. (For each of the presentation levels, the lastpresentation of a pair member was of the derived word. Prior presentations were of the first and repeatedoccurrences of the initial word. The error bars are one standard error of the mean.) Bottqm: Error ratesassociated with the identification times.


Table 2 ,Recognition Probabilities for the OverlapCondition of Experiment 4

Pairmember

InitialMSE

DerivedMSE

Presentation of initial words

1

.070

.037

.140

.032

2

.063

.024

.133

.026

3

.039

.024

.188

.030

4

.039

.035

.133

.027

Note. Shown are the false-alarm rates for the first pre-sentation of the initial word and the derived word for thefour presentation levels of the initial word.

incorrectly, that the initial word had occurredbefore.

As was true for the identification times,recognition performance was significantlyaffected by the overlap in spelling betweenthe initial and derived words. This result isconsistent with several findings in the liter-ature that show that recognition responsesare very sensitive to a variety of types of re-latedness between words. These findings in-clude such variables as associativeness (Un-derwood, 1965), antonymity (Fillenbaum,1969), synonymity (Anisfeld & Knapp, 1968),and modality changes (Hintzman, Block, &Summers, 1973). More directly related to thepresent findings are those of Wallace (1968)who found elevated false alarms to words thatwere orthographically related to earlier pre-sented words. Thus, the effect of the initial-derived manipulation on recognition was notunexpected.

The relationship between identificationtimes and false alarms is also of interest. Sev-eral investigators (e.g., Atkinson & Juola,1973, 1974; Mandler, 1980) have suggestedthat recognition responses are based jointlyon the outcomes of a search process and a(usually ill-defined) "familiarity" evaluationprocess. That is, to the extent that testeditems seem familiar, they are judged as havingbeen previously presented. Recently, Jacoby(1983; Jacoby & Dallas, 1981) has suggestedthat the familiarity mechanism in recogni-tion is derived directly from the ease withwhich an item is identified. That is, if an item

is more easily or rapidly identified, it "seems"more familiar, and consequently, a positiverecognition decision is more likely.

This interpretation predicts that as theidentification times for the derived wordsbecome faster (as a function of increasingpresentations of the initial word), the falsealarms to the derived words 'should risemonotonically. The data in Table 2 do not.support this interpretation. However, thepower in this experiment was low. Neverthe-less, the fact that the derived words exhibitboth higher false-alarm rates and lower iden-tification times than the first presentations ofthe initial items (r = -.43) suggests that rec-ognition and identification share at leastsome common mechanisms. The nature andextent of the commonality between the twotasks is discussed in more detail in Experi-ment 5.

Experiment 5

The results from the previous experimentssuggest that the repetition effect in identifi-cation is supported by stored episodic events.However, there is a puzzle. Episodic supportis generally thought to give rise to lag effectsor memory loss, yet one of the reasons thatthe repetition effect has received so muchattention is that it tends to persevere overextended retention intervals (e.g., Jacoby,1983; Jacoby & Dallas, 1981; Scarborough etal., 1977, 1979). Similarly, we also failed tofind significant loss in the magnitude of therepetition effect over lag. Specifically in Ex-periment 1, using CTI and DTI, no differencein the identification probabilities for repeateditems was found between lags of 1-5 and 21-25 intervening items. In Experiment 2, usingCTLI, lags of 10-15 and 40-45 also failed toproduce a difference in identification times.However, in those experiments, lag was notthe primary variable, and consequently, thetests lacked power. The present experimentwas aimed at comparing the effects of lag onrecognition and identification time using theCTLI/recognition paradigm. This comparisonshould also allow evaluation of Jacoby's hy-pothesis (e.g., Jacoby & Dallas, 1981) that easeof identification is a contributory factor to rec-ognition. Any relationship between identifi-cation time and recognition accuracy might


provide insight into the possible role of epi-sodic traces in the repetition effect in identi-fication.

Method

A list of words was presented that contained itemsthat were repeated after lags of 1-5, 10-15, 45-55, or120-130 intervening items. There were three total pre-sentations of the items in the three shorter lag conditions.The words at the longest lags were presented only twice.As in Experiment 4, each word on the list was testedwith CTLI for identification fluency and for recognitionaccuracy.

Subjects. The subjects were 34 students drawn fromthe same pool as the previous experiments. Two subjectswere replaced for failure to reach 85% accuracy on theidentification task.

Stimuli and procedure. The lag between repetitionsof words on the list was the independent variable. Thewords used in the lag condition were 24 high-frequency(A or AA) nouns from the Thorndike and Lorge (1944)word counts. The lengths of the words varied betweenfour and six letters. For each subject, the 24 words wererandomly divided into four groups of 6 words each.These four groups were assigned to four levels of lag:1-5, 10-15, 45-55, and 120-130 items intervening be-tween successive presentations of the same word. Forlags 1-5, 10-15, and 45-55 there were three presenta-tions of each of the 6 words within the levels. In the 120-130 lag, however, each word was presented only twiceso that the list would not be excessively long. Betweensubjects the words were rotated through the conditions.The items in the lag condition were distributed as evenlyas possible throughout the list. However, for obvious rea-sons, it was not possible to present first occurrences ofitems in the long lag conditions (45-55 and 120-130)in positions near the end of the list.

An additional 94 words were used as fillers. The num-ber of presentations of the fillers was chosen so that thenumber (120) of newly presented items equaled thenumber of items presented earlier. This was done tominimize the possibility of response bias for the recog-nition judgments. The lag for the filler items was notcontrolled. The first 10 words on the list constituted aprimacy buffer. The subjects saw the same 15-word prac-tice list that was used in Experiment 4. For both thepractice and experimental lists, all the words were pre-sented in uppercase letters.


Identification. The top panel in Figure 9shows the mean correct identification timesas a function of the four lag conditions. Thethree curves are for the first, second, andthird presentations of identical items. Thebottom panel shows the error probabilitiesassociated with the identification responses.

Because there was a different number of

presentations in the 120-130 lag conditionthan in the other lag conditions, two separateanalyses were performed on both the iden-tification times and error probabilities. Thefirst wa^for the first and second presentationsat all four lags, and the second was for allthree presentations at the three shorter lags.

The analyses confirmed the large effect ofrepetitions on identification times evidentin the figure, F(l, 31) = 165.02 andF(2, 62) = 132.77 for the four-lag and three-lag analyses, respectively. There were signif-icant effects of lag for the four-lag, F(3, 93) =4.16, and three-lag analyses, F(2, 62) = 5.00.Analyses of the simple main effects showedthat the lag effect was due primarily to therise in identification time across lags for thesecond presentations, F(3, 93) = 3.79. Therewere no significant lag differences for first,F(3, 93) = 2.33, p > .07, or third, F(2, 62) <1.0, presentations. Although the change inidentification time with lag for second pre-sentations suggests that there was some dec-rement in the repetition effect over time, thePresentation X Lag interaction for the firsttwo presentations failed to reach significance,F(3, 93) = 1.97, p > .12. That is, the iden-tification time difference between the first andsecond presentations did not differ signifi-cantly over lag. Thus, there is, at best, onlyweak evidence for an effect of lag.

The results for the error rates closely par-allel those for the identification times. Be-cause the error rates for the second and thirdpresentations were essentially zero at all lags,statistical analyses will not be reported.

Recognition. The results for the recogni-tion judgments are shown in Table 3. Shownin the table are the miss rates for the secondand third presentations of repeated items asa function of lag. There was a large effect oflag for the second presentation, F(3, 93) =14.58, as well as a significant interaction be-tween presentation and lag over the threeshorter lags, F(2, 62) = 24.27. The miss ratesfor the third presentations are all near zero(a floor effect). For the first presentations ofcritical items, the correct rejection rates were.91, .94, .97, and .96 for words in the 1-5,10-15, 45-55, and 120-130 lag conditions,respectively. These results show that as lagincreases, the probability that a previouslypresented word will be recognized falls dra-


48-

46

2 40

-r

.15

.10

,05

0.0

1-5 10-]5 45-55 120-130L A G

1-5 1D-15 45-55 120-130L A G

Figure 9. Top: Mean identification time as a function of lag between presentations in Experiment 5. (Thethree curves are for the first, second, and third presentations of identical words. The error; bars are onestandard error of the mean.) Bottom: The probabilities of identification errors associated with the iden-tification time results.

matically.8 This is riot an unexpected finding.The fact that forgetting occurs over time inrecognition tasks is a well-documented phe-

Table 3Miss Rates for the Second and Third

' Presentations of the Words in the Four LagConditions of Experiment 5

Lag (intervening items)

Presentation 1-5 10-15 45-55 120-130

SecondMSE

ThirdM'SE

.042

.012

.031

.015

.047

.015

.026

.013

.188

.029

.022

.013 .

.172

.028

——

Note. There were no third presentations at the longestlag.

nomenon. The fact that the miss rates seemto be close in value at the two short lags andclose in value at the two long lags is certainlyunexpected, and we have no compelling ex-planation to account for such; a finding.

In summary, a large decrement with lagwas found in recognition performance, andif there was a diminution with lag for the

8 One possible problem in analyzing the lag effects forboth recognition and identification is the inherent con-found between lag and serial position: Repeated itemsmust necessarily be presented in later list positions thanfirst presentations of the same items. However, an in-spection of the identification times andirecognition prob-abilities for only those items that occurred within a re-stricted range of serial positions (80-120) yielded a pat-tern of results not appreciably different from that foundin the overall analysis. Thus, relative serial positionseems not to have been a contributory factor to the re-sults shown in Figure 9.


repetition effect in identification, it was verysmall. This pattern of results is similar tothose found by other investigators using dif-ferent tasks (e.g., Jacoby, 1983; Jacoby & Dal-las, 1981; Scarborough et al., 1977).

The failure to find a large lag effect forrepeated items in identification apparentlyposes a problem for the idea that the effectderives from access to episodic traces duringthe identification process. Recognition mem-ory (logically) requires access to episodictraces, and recognition shows large effects ofretention interval. If identification also relieson access to individual episodes, then whyare comparable performance decrements notobserved for the repetition effect in identifi-cation?. Jacoby (1983) has proposed one possiblesolution. He has suggested that the differencesobserved between recognition and identifica-tion are due to differences in the kind of in-formation required for accurate performancein the two task?. For example, recognitionmemory is commonly thought to rely jointlyon two processes: an evaluation of the "fa-miliarity" of a tested item plus an extendedmemory search (Atkinson & Juola, 1973,1974; Mandler, 1980). Mandler (1980) hassuggested that the.familiarity stage of recog-nition relies on "intra-item integration," aquality derived from characteristics of the itemitself. Identification also relies on informationinherent in the item (e.g., pronunciation, let-ters, letter clusters). The similarity betweenidentification and recognition may thereforebe based on similarities between the identifi-cation process and the familiarity (nonsearch)phase of recognition.

The search phase of recognition, on theother hand, may rely on information that isnot integral to the tested item. Recognitionrequires that contextual information asso-ciated with an item be retrieved. To do this,subjects must use information about the orig-inal list context as a retrieval cue. Thus, rec-ognition failure over time may result from aprogressive mismatch between the test andstorage contexts (e.g., Anderson & Bower,1972; Bower, 1972). If identification does notrely on these search processes, then no greaterlag effect in identification would be predictedthan would be predicted in recognition dueto the familiarity phase. Thus, if the famil-

iarity phase in recognition does not show lageffects, then lag effects would not be seen inidentification either. Such an explanation hasa number of problems, however. If this weretrue, for example, a long delay would ensurethat recognition performance would dependmainly on familiarity (which does not decay)rather than search (which does). Not muchevidence supports the view that search pro-cesses and associated effects disappear at longlags in recognition tests. In fact, Mandler(1980) has presented arguments and evidencethat exactly the opposite is the case—thatsearch processes become predominant atlong test delays.

We next consider an explanation that relieson different types of retrieval cues and in-formation used in identification and recog-nition. According to this hypothesis, the ep-isodic traces stored in memory can be ac-cessed by several types of cues. In recognition,contextual cues are used, that change overtime, so lag effects are observed. In identifi-cation, however, the cues that give access tothe episodic images are not temporally vari-able. They include such features as the phys-ical characteristics of the test stimulus, rela-tionships among those features, and infor-mation at the letter level of analysis. Becausethese types of cues do not change much overtime, only small lag effects would be pre-dicted. It might then be asked, Why do sub-jects not use these same cues in recognitionand eliminate lag effects? The answer seemsclear: Such cues are used and they succeedin part. The tested items in recognition areidentified, and identification is probably morerapid for target items than for distractors.However, identification is not the goal in arecognition task. Rather, access to temporal-contextual information is needed for accu-rate performance, and temporal-contextualcues are needed to gain access to this kindof information (e.g., Bower, 1972).

The above analysis does not preclude thepossibility that the information used in iden-tification plays some role in recognition, asJacoby and his associates (Jacoby, 1983, inpress; Jacoby & Dallas, 1981) have suggested.In order to probe this issue, a correlationalanalysis between recognition accuracy andidentification time was done on all of theitems on the list that were presented three


times. Both the critical lag items and the filleritems were included in this analysis.

The relationships are shown in Figure 10.The top panel shows the relationship betweenthe probability of a correct rejection of firstpresentations of the words and their identi-fication times. The identification times werepooled across blocks of five consecutiveframes. Extreme frame times, in which few

responses were obtained, were not includedin the analysis. The numbers in brackets be-low each point are the number of observa-tions that constitute the point.

Notice that the correct rejection rate rosewith the time required to identify the word.That is, when the word took more time toidentify, the probability of a no response forthe recognition judgment increased. The

1.00

LLJ

.90

(35)

(130)

OO

32-40 42-50 52-60 62-70 72-SO

I D E N T I F I C A T I O N T I M E

1.00

.90 _ (53)

(564)

(35)

(63) .

(202)

22-30 32-40 42-50

I D E N T I F I C A T I O N T I M E

62-70

Figure 10. Top: The correlation between the probability of correct rejection of first occurrences of wordsand the identification time for the words in Experiment 5. Bottom: The correlation between hit rates andidentification time for repeated items on the list (second and third presentations). (The number in bracketsbelow each point is the number of observations that constituted the point.)


Pearson product-moment correlation be-tween the weighted means was +.78. Thebottom panel shows the correlations betweenthe hit rates and identification times for sec-ond and third presentations of the words. Thesame trend is present here as for the first pre-sentations. As identification took longer, a noresponse (which for a repeated item is an in-correct response) was more likely. Becausethe recognition responses are plotted in termsof hit rates, the slope is in the opposite di-rection than for correct rejections of first pre-sentations. The correlations were —.84 and—.21 for the second and third presentations,respectively. Thus, it is clear that as identi-fication became faster, the subjects tended toreport that they had seen the item previously,regardless of whether that was true.

Although the evidence is correlational anddoes not imply any causality, this result isconsistent with Jacoby's proposal that theidentification ease or fluency for an item con-tributes to the recognition judgment madeabout that item. That is, the subjects couldbe using the ease of identification ,as onesource of evidence that the word had oc-curred on a recent list. For several reasons,however, it seems unlikely that this is a majorfactor. First, large differences in identificationtimes were accompanied by changes of onlya few percentage points in recognition ac-curacy. Second, identification times varyidiosyncratically for different words. Thus,relative "perceptual fluency" is an unreliablecue to list membership. Finally, the small in-crease in identification time with increasinglag hardly seems adequate to account for thelarge decrease in recognition accuracy. Al-though identification fluency may play somerole in recognition, it seems likely that theretrieval of the context associated with epi-sodic events is the more important determi-nant of recognition performance.

General Discussion

The experiments discussed above raise thepossibility that the repetition effect in wordidentification is derived largely from memoryfor individual episodes. First, large repetitioneffects were observed from nonwords in thefirst three experiments. Unless it is assumedthat a single experience with a nonword is

sufficient to establish an abstract memoryrepresentation, it is difficult to reconcile thisresult with models that posit activations ofabstract unitized memory traces (i.e., logo-gens) as the sole locus of the repetition effect.Even if a single presentation is assumed suf-ficient to establish a logogen, the additive re-sult in Experiment 2 and the Task X Lexi-cality interaction for exposure durations inExperiment 1 remain unexplained. These re-sults suggest that the repetition effect is de-rived from the operation of a process that haslittle to do with general knowledge aboutwhat constitutes a word.

Second, the elevated errors and faster iden-tification time for derived items in Experi-ments 2 and 4 suggest that memory for theprior occurrences of the initial items was ac-cessed by the identification process. This ar-gument is particularly convincing for thenonwords, because any explanation based on"spreading activation" between orthograph-ically related lexical nodes is unlikely for non-word stimuli.

Our emphasis on episodic contributionsto the repetition effect should not be takenas support for a model positing a sharp struc-tural distinction between episodic and se-mantic memory or a model in which seman-tic knowledge does not play a role in wordidentification. We would be quite comfort-able with a model like Jacoby's (1983, inpress) in which semantic knowledge is anemergent property of amalgamations of in-dividual episodic traces. Furthermore, wordidentification may well be the result of a com-bination of system response to semantic andepisodic images even if these are thought ofas different. The response could result froma race between episodic traces and semantictraces or from some synergistic interactionbetween the two.

We do wish to make a sharp distinctionbetween one characteristic of semantic codesfor words (or logogens) and another charac-teristic of episodic codes for nonwords: thepresence versus absence of a unitized re-sponse. The results of Experiment la, par-ticularly, provide evidence that the word-nonword differences may be due to an (au-tomatic) generation of a unitized code forwords that is not available for (unfamiliar)nonwords. We suggest that it is the avail-


ability of a unitized code that appears largelyindependent of the factors producing repe-tition effects.9

Our emphasis on the importance of theunitization factor leads us to a model thatdiffers in certain respects from a model ofidentification and repetition recently pro-posed by Jacoby (1983, in press). In his ac-count, the episodic-semantic distinction isobviated, and differences observed betweenidentification and recognition are attributedto differences in the requirements of the twotasks. In many respects the model elaboratedbelow is consistent with Jacoby's approach,but we retain part of the distinction betweenepisodic and semantic memory. As alreadyindicated, we argue that lexical codes andepisodic representations, each contribute toidentification facilitation in separate ways.This seems necessary not only to account forthe large differences in the identificationtimes of words and nonwords (particularlyin Experiment 3 where the mixed list resultspreclude any response bias explanation of theword-nonword difference) but also the Lex-icality X Task interaction found in Experi-ment I (which suggests a qualitative differ-ence between the identification of words andnonwords).

The discussion of the model is divided intotwo parts. In the first part, it is shown thatwith appropriate assumptions the additivitybetween the effects of lexicality and repeti-tions that was found in Experiment 2 usingCTLI, but was not found in Experiment 1using CTI and DTI, can be predicted fromthe psychometric functions obtained in Ex-periment la (Figure 3). This analysis, inturn, provides a basis for isolating the relativecontributions of information of a lexical na-ture (i.e., unitization, automatic response)from information specific to temporal-con-textual episodes. In the second part, a generalframework for repetition effects in identifi-cation is proposed, which illustrates the roleof various component mechanisms in theidentification process.

The Model

Lexicality and repetitions. In the model,we assume that when an item is presented,eyidence about the identity of the item begins

to accumulate in the perceptual system.When a sufficient amount of evidence hasaccrued to specify a single item, an identifi-cation response ensues. This "informationgrowth" assumption is similar to proposalsmade by other investigators to account forfindings in recognition memory (e.g., Rat-cliff, 1978) and letter perception (e.g.,McClelland & Rumelhart, 1981; Rumelhart& McClelland, 1981).

Once this framework has been adopted, itis possible to generate an explanation of theadditivity of the effects of lexicality and rep-etitions under CTLI in Experiment 2 and thenonadditivity of these factors under DTI andCTI in Experiment 1. Consider first the re-sults from Experiment 1. Recall that with thetechniques used in that experiment, the stim-uli (i.e., words and nonwords) were all pre-sented at, or near, threshold. That is, the dis-play was terminated at some point when onlypartial information about the stimulus wasavailable to the subjects. The responses were,of course, based on the extent ;and quality ofthe available information. If we assume thatthe probability of identifying a given item ata given time is proportional to the amountof information that has accrued about theitem at that time, then the rate at which theinformation accrues is proportional to theslopes of the psychometric functions thatwere obtained (Figure 3). Once this assump-tion is made, then the time course of infor-mation growth can be depicted graphicallyas in the top panel of Figure 11. The func-tions in Figure 11 are the best fitting logisticresponse functions that are shown in Fig-ure 3.

Figure 11 also shows how the relationshipbetween performance and exposure durationwould be obtained given the informationgrowth functions. Interrupting the display ata time that would yield about 50% accuracy(as was attempted in Experiment 1) is rep-resented in the figure by the horizontal line

9 However, unitization is only one characteristic of lex-ical codes—there are many others as well. Thus it maybe possible to present nonwords sufficiently many timesthat the letter strings become unitized and act like wordsin the CTI and DTI paradigms, but it would not beappropriate to conclude that a lexical code has been gen-erated.


drawn through the growth functions at the50% crossover point. The projections ontothe abscissa of the intersections of this linewith the growth functions would then rep-resent the times needed in the two tasks toobtain 50% accuracy for the words and non-words.

What, then, is the effect of repetitions onthe growth functions? Consider the possibilitythat each presentation of an item results inthe storage of a complex memory trace con-sisting of the item and the presentation con-text in which it is embedded (Raaijmakers& Shiffrin, 1981). Suppose further that thenext time that item is presented the rate ofaccrual of information is uniformly in-creased by a constant amount of time be-cause of the presence of episodic image(s).

The resulting growth functions are illustratedin the bottom panel of Figure 11.

Here the curves are shifted to the left bya constant to represent the increased pro-cessing efficiency due to the episodic trace.Because of the shift, relatively ttiore infor-mation will accrue at a given exposure du-ration than for a first presentation. This ad-ditional information is represented in thefigure by the differences between theintersections of the (constant) times with thefunctions at the first (i.e., 50% line) and sec-ond presentations. The extent of the infor-mation increase is thus a joint function ofthe amount of the shift and the slopes of thegrowth functions. Thus, to the extent that theslopes of the functions differ, the amount ofavailable information will increase differen-

nri CTI

x:oc.o

u.o

oK.

T I HEDTIMDHD

CTI Cf! DTIWOUD fCNHOUD NONMOUD

T I M EDTIwono

CTI CTI DTIWORD NOHHORO

Figure 11, Top: Schematic representation of the growth of information over time for first presentationsof words and nonwords in the continuous threshold identification (CTI) and discrete threshold identifi-cation (DTI) tasks. (The curves are the best-fitting logistic functions from Figure 3. The projections ontothe abscissa of the intersections of the 50% line with the functions indicate the amount of time necessaryto obtain 50% accuracy for the two tasks.) Bottom: Shift of the functions to the left for repeated items(i.e., the decrease in processing time as a result of contributions from the episodic trace established atthe first presentation). (The deltas [A] indicate the gain, between presentations, in the amount of infor-mation available at the constant exposure durations.)


tially with repetitions for the words and non-words in the two different tasks.

The difference in the amount of additionalinformation made available by repeating theitems is, in turn, reflected in the identifica-tion probabilities from Experiment 1 (seeFigure 2). The probability of identificationincreases more rapidly with repetitions underconditions that yield a steeper growth func-tion. In fact, to a large extent, the differencesin the Identification X Presentation functionsin Figure 2 mirror the slope differences in thepsychometric functions in Figures 3 and 11.

The above analysis carries with it the im-plicit assumption that the lexicality of anitem has at least one effect on its identifica-tion that is independent of the effects of rep-etition. That is, as has already been suggested,the presence of an episodic image speeds theidentification process more or less uniformly,whereas the fact that an item is a word tendsto increase the ease with which it is encodedinto a form more resistant to decay. The dif-ference in encoding ease is reflected in thedifference in the shapes and levels of the func-tions, and the change in the speed of theidentification process due to repetitions isrepresented by the temporal shift of the func-tions. This hypothesis is bolstered by the re-sults from the control conditions of Experi-ment 2 (Figure 6), suggesting that the effects

of lexicality are additive to the effects of rep-etition, at least for three presentations.

As mentioned earlier, the reason why ad-ditivity was present in the results from Ex-periment 2, although lexicality and repeti-tions clearly interacted in Experiment 1, hasto do with the differences between the tasksused in the two experiments. In Experiment1 the exposure durations were held constantfor a given subject, and the probability ofidentification was the dependent variable.Thus, the error rates changed, so the slopesof the functions became crucial. In Experi-ment 2, as well as in the other experiments,the subjects were encouraged to keep theirerror rates fixed. (Actually, they were keptquite low, but the level at which the errorsare kept is not crucial to the argument.) Thus,the identification performance was being heldrelatively constant, at a high level, and iden-tification time was varying. This is mani-fested as a constant time difference, whenrepetitions cause all the curves to shift (leftin Figure 12) by a constant amount of time.This is illustrated in Figure 12 for the firstand second presentations of word and non-word items. Notice that the difference be-tween the two functions is unaffected by theshift to the left: The identification time dif-ference between the words and nonwords willstay constant between repetitions even though

FIRST PRESENTATIONSECOND PRESENTATION

T I M E 'WORD NCNWORD WORD NONWORD

SECOND FIRST

Figure 12. The growth functions for the first and second presentations of words and nonwords in thecontinuous threshold identification task. (The projections onto the abscissa show the identification timesat asymptotic performance levels. Under the assumption that the presence of an episodic trace uniformlyincreases the speed of analysis for both the words and nonwords, the difference in response times betweenthe two stimulus types remains constant across repetitions.)


both are more rapidly identified. Note thatthe constant time shift would be found evenif performance were not at asymptote—anyhorizontal line (constant accuracy) drawnthrough Figure 12 would result in such a con-stancy. (Of course, it is possible that the sub-jects find it easier to maintain equal accuracyunder "low error" instructions that produceasymptotic performance.)

Thus, we have a tenable hypothesis aboutthe differences between the effects of lexical-ity and repetitions on identification. In thenext section, we propose a general frameworkfor the repetition effect in identification thatoutlines some probable components of theprocess and the way in which these compo-nents might interact with unitized lexicalcodes and episodic memory representationsof the target items.

A framework for the repetition effect. Thedata discussed above, together with resultsfrom other experiments, suggest a model ofthe identification process that would operateas follows: First, assume that there is a mech-anism that analyzes the features of a letterstring through a hierarchically organized se-ries of levels. At each successive level, a morecomplex organizational structure is imposedon the incoming stimulus. The end result ofthis analysis is a name or label for the letterstring—the identification response. The no-tion that word identification is accomplishedthrough a hierarchically organized informa-tion-processing system is not original. Severalexisting models make similar assumptions(e.g., Adams, 1979; Johnston & McClelland,1980; McClelland, 1976; McClelland & Ru-melhart, 1981; Rumelhart & McClelland,1981, 1982). What is unique, however, is thetype of information stored in memory thatis accessed when the items are presented foridentification.

As the identification process progresses, itbegins recruiting information in memorythat is associated to the particular featuraland contextual elements of the stimulus item.This information includes both general se-mantic or lexical knowledge about words,transitional probabilities between letters, andso on, and information that is associated withspecific temporal-contextual episodes. Thedegree to which the stored information is re-cruited or activated is a function of its as-

R E S P O N S E

SECrec

WNTICCRY

WORD CODE :

PREEXPERIMENTAL

TRACES

WORD NAME AND

PRONUNCIATION

'

LETTER

CLUSTERS

'

LETTERNA*ES

'

PHYSICAL

LETTER

'PHYSICALFEATURES

i

ElM

"*

PISODICEMSRY

EPIS0DIC

TRACES

A

I T E M C O N T E X T

P R O B E

Figure 13. Schematic representation of the analytic pathleading to an identification response for words. (Eachlevel of the hierarchy interacts with either episodic orpreexperimental semantic memory or both.)

sociation to the particular stimulus and con-text that are presented for test.

For words, the operation of this analyticmechanism is depicted schematically in Fig-ure 13. When the word is presented, it beginsto ascend an analytic path represented by thecentral column of the figure. At each levelepisodic memory representations that areconsistent with the presented word are acti-vated. Simultaneously, contextual features ofthe test situation begin to activate stored ep-isodic images that contain contextual ele-ments common to the test context. Thus, theactivation of the episodic information is sen-sitive to both the item being tested and thecontext in which the test occurs.

The distinction between episodic and se-mantic memory, represented in the figure bythe dashed vertical line, is not intended toimply two separate, noninteracting memorysystems. That is, the memory traces of epi-sodic occurrences of the items include infor-

342 T. FEUSTEL, R: SHIFFRIN, AND A. SALASOO

mation about those occurrences generallyassumed to be semantic in nature: pronun-ciation, spelling, meaning, and so on. Thus,in this regard our model is similar to thatproposed by Jacoby (1983, in press). The dis-tinction is intended to be of significance onlyin that it makes clear the separate contribu-tions of a unitized code for words and episodicand contextual information generated by theexperimental presentations.

Similarly, the levels represented in the an-alytic path are not intended to be inclusiveof the kinds of knowledge brought to bear onthe stimulus word. Other sources of infor-mation not examined in our paper, for ex-ample, word shape, pronunciation regularity,and so forth, may well be involved in theidentification process. However, the repre-sented levels do seem to constitute a mini-mum requirement in light of the results re-ported here and elsewhere. Thus, the inclu-sion of a "letter name" level, as distinct froma "physical letter" level, reflects the findingthat changes in case or type font have littleeffect on the magnitude of the repetition ef-fect. Similarly, the "letter cluster" level is in-cluded because of the repetition effect ob-served between items that share common lit-eral configurations. The "word name" levelreflects our belief that one important com-ponent in the identification task is the deci-sion that a string of letters composes the en-tire stimulus word.

The levels in the ahalytic path make con-tact with the episodic traces that are consis-tent with the activations within the levels.The interaction between the images and thelevels is similar to the kind of interaction sug-gested by McClelland and Rumelhart (1981).(It may prove necessary to assume that theconnections are much more strongly excit-atory than inhibitory, because words thatoverlap in orthography sometimes facilitatebut never inhibit each other in our work.) Tothe extent that an episodic image is consistentwith the activations at any given level, it willitself become active. This activation, in turn,feeds back to the perceptual level, increasingthe excitation of consistent features withinthe level and, consequently, speeding theanalysis. This interactive quality is repre-sented in the figure by the double-headedarrows interconnecting the various compo-

nents. Dashed connections reflect interac-tions for which empirical evidence is weakor nonexistent. Thus, for example, the con-nection between episodic memory and the"physical letter" level reflects the finding thatcase changes have only a slight effect on therepetition effect, and the connection betweenepisodic memory and word pronunciationsignifies that no link exists before the firstpresentation during the session.

The excitation of the episodic images isalso influenced by the overlap between thetest context and the context stored with theepisodic trace. The ability of the test contextto influence the identification process is nec-essary for any model of repetition effects. Thesmall effect of lag, as well as^the effect of listcontext on the identification process reportedby Jacoby and his associates (Jacoby, 1983;Jacoby & Witherspoon, 1982), demonstratesthe influence of contextual factors at the timeof test.

The analytic process also interacts with thepreexperimental traces, including codes like"logogens" and lexical codes in the case ofwords. These contain unitized representa-tions that are responsible for automaticallyproducing a phonological code and/or namecode when appropriate.

How does the model handle repetition ef-fects for nonwords? Logically, at least, thedifference between words and nonwords isthat the words have preexisting lexical rep-resentations and the nonwords do not. Inother respects, the analytic system elaboratedabove will process an incoming nonword let-ter string in the same fashion as a word letterstring. The major contribution of the seman-tic store is that it provides the identificationsystem with a readily available unitized re-sponse. For pronounceable nonwords, thisreadily available response is nonexistent andmust be constructed analytically from thesubjects' knowledge of orthography and let-ter-to-sound rules.

The differences are illustrated in Figure 14.The primary change is a mechanism thatproduces a phenological response analyti-cally. (The phrase letter-to-sound rules is usedhere more for convenience than as an en-dorsement of any particular model of pho-nological encoding.) Because, the synthesis ofa phonological representation presumably


takes time, it is the absence of a readily avail-able response based on preexperimental uni-tized codes that causes the constant differ-ence between the identification times ofwords and nonwords observed in Experiment2. Although the present data do not bear onthe automatism of these processes, the word-nonword difference could be construed as amanifestation of the difference between anautomatic versus controlled process (Schnei-der & Shiffrin, 1977; Shiffrin & Schneider,1977). The poor performance of the subjectson nonwords in the DTI task could reflecta capacity limitation for the (controlled) pro-cess responsible for constructing a phonolog-ical representation of the nonwords. Thewords, then, would not be subject to this lim-itation.

Notice that, in the.figure, the semanticstore is still contacted for the nonword stim-uli. This, of course, is because the spread ofactivations through the hierarchy is largelyautomatic, and the system has no a prioriway of "knowing" that the incoming letterstring is a nonword. Thus, the representa-tions of words that are similar to the non-words are partially activated. The partial ac-tivations may contribute (by analogy) to theprocess of constructing a pronunciation ofthe nonword. Mechanisms that allow knowl-edge about the pronunciation of words tocontribute to the construction of a pronun-ciation for nonwords have been suggested byother investigators (Glushko, 1979, 1981;McClelland & Rumelhart, 1981). However,such an assumption can also account for thetendency, reported by some subjects, to re-spond to nonword stimuli with word re-sponses.

It is important to note that the differencebetween episodic and semantic memory (rep-resented by the difference between the twosides of Figures 13 and 14) is not relevant tothe position that we are adopting with respectto the repetition effect for words and non-words. Note that both semantic and episodicmemory components are contacted duringthe identification of nonwords as well as ofwords. What is important is the manner inwhich an identification response is producedfor word and nonword stimuli. For words, aunitized response code is available, whereasfor nonwords, an additional phonological

R E S P O N S E

SEWrerc

WT1CRY

PREEXPERIMENTAL

TRACES

J

NONWORD

PRONUNCIATION

' '

LETTER TO

SOUND RULES

'

LETTER '

CLUSTERS

'

LETTERNW-ES

'

PHYSICAL

LETTER

11

PHYSICAL

•FEATURES

I.TE

Ec

£«•••&

f S

C- -3

PISODICEMORY

EPIS0DIC

TRAC 'ES

t

1

M C O N T E X T

P R O B E

Figure 14. Schematic representation of the analytic pathleading to an identification response for nonwords. (Itdeviates from the path for words depicted in Figure 13by the inclusion of an additional process responsible forconstructing a pronunciation of nonwords and the ex-clusion of a unitized response from Ithe semantic store.)

construction process is required to producea response. It is this additional analytic pro-cess (letter-to-sound rules) that is responsiblefor the (additive) difference in the identifi-cation of words and nonwords. We are sug-gesting that repetitions affect processing alongthe analytic pathways in both Figures 13 and14, whereas lexicality is manifested in thedifference between the two figures—the pho-nological receding stage for nonwords andthe availability of a relatively "automatic"and unitized response for words.

How can the model resolve the contradic-.tory results that have sometimes been re-ported for the repetition effects? For example,why should there be facilitation between vi-sually presented items that differ substan-tially in physical appearance (e.g., case) but


no facilitation between auditory and visualpresentations of items (e.g., Jacoby & Dallas,1981; Morton, 1979b)? The answer probablylies in the type of information that is pre-served between successive presentations ofitems. For example, with visual presenta-tions, it could be argued that the importantcues for access to episodic traces are suchthings as the visual characteristics of the dis-play and the configuration of letters in theitem. When the first presentation is auditory,information of this sort is not present (and,hence, not stored), and consequently, the ep-isodic trace is not contacted by the second,visual presentation. In support of this expla-nation, it is interesting to note that when sub-jects are required to spell auditory presen-tations of words, subsequent visual identifi-cation of those words is facilitated (Jacoby& Witherspoon, 1982).

In lexical decision, the failure to find fa-cilitation for repeated nonwords (e.g., For-bach et al., 1974; Scarborough et al., 1977)is probably due to a somewhat different, butrelated, problem. It seems likely that thisfinding is derived from the facility with whichrepeated nonwords contact episodic memoryrepresentations rather than the failure to ac-cess lexical or semantic representations. Sup-pose, for example, that the decision about thelexicality of a letter string is based partiallyon the failure to access a memory trace. Ifthe episodic trace from the first presentationof a nonword is accessed the second time thenonword is presented, then one reliablesource of evidence about the item's lexicalstatus is no longer effective. The subjectsmust resort to other, perhaps less efficient,means of determining whether the item is aword. McKoon and Ratcliff's (1979) findingthat repeated nonwords are rejected moreslowly and less accurately than first occur-rences of nonwords in lexical decision is con-sistent with this interpretation. Thus, the fail-ure to find enhancing repetition effects fornonwords in lexical decision can be taken asevidence for, rather than against, the idea thatthe repetition effect involves access to mem-ory for single prior events.

If it is accepted that memory for episodicevents plays a large role in word identifica-tion, then what account can be given in themodel of the independence that is sometimes

observed between identification and recog-nition memory (e.g., Jacoby & Witherspoon,1982)? As was discussed in Experiment 5,such independence can be accounted for bydifferences in the kind of information re-quired for performance on the different tasksand the retrieval cues used to access that in-formation. One hypothesis would posit thatrecognition requires access to temporal-con-textual information associated with a re-peated item and that identification requiresaccess to information about structural as-pects of the item. Thus, the decision in rec-ognition is based on the use of different cuesto access different aspects of the memory rep-resentation than those accessed in identifi-cation. The differential effects of lag on thetwo tasks (e.g., Jacoby, 1983, in press; Jacoby& Dallas, 1981) probably results from suchdifferences in the important cues. The cuesthat are effective in identification may not besubject to the deleterious effects of temporaldelay. The important recognition cues, how-ever, could be subject to such effects.

In summary, we suggest that both recog-nition memory and the repetition effect inword identification are manifestations of thepresence of a stored episodic image. Thus,recognition should not be adopted as the onlytask diagnostic of memory for specific epi-sodes. Tasks such as word identification,which do not logically require the retrievalof the context in which an event occurred,can nevertheless reveal the influence of thatevent. Jacoby and Witherspoon (1982) havecharacterized the difference between thesetwo manifestations of episodic memory asthe difference between memory "with aware-ness" and memory "without awareness."

In our model, repetitions produce episodictraces that support the identification process,thereby speeding the analysis of both wordsand nonwords. Whether this process operatesas a retrieval phenomenon or as a result ofactivation of stored episodic traces is not yetclear. Although word and nonword identifi-cations are similarly facilitated by repetitionsin our model, the perceptual processing path-ways resulting in correct pronunciation andidentification responses to words and non-words differ vastly. Words have integratedrepresentations in semantic memory that al-low identification to be made relatively


quickly, on what may be an automatic basis:A unitized code can be provided for a per-ceived set of features before the features areforgotten. The nonwords are dealt with bysome sort of phonological construction pro-cess, and feature forgetting may take placeto a considerable degree at various stages ofthe identification process.

References

Adams, M. J. Models of word recognition. CognitivePsychology, 1979, 11, 133-176.

Anderson, J. R., & Bower, G. H. Recognition and re-trieval processes in free recall. Psychological Review,1972, 79, 97-123.

Anisfeld, M., & Knapp, M. Association, synonymity,and directionality in false recognition. Journal of Ex-perimental Psychology, 1968, 77, 171-179.

Atkinson, R. C., & Juola, J. F. Factors influencing speedand accuracy of word recognition. In S. Kornblum(Ed.), Attention and performance IV. New York: Ac-ademic Press, 1973.

Atkinson, R. C., & Juola, J. F. Search and decision pro-cesses in recognition memory. In D. H. Rrantz, R. C.Atkinson, R. D. Luce, & P. Suppes (Eds.), Contem-porary developments in mathematical psychology (Vol.1). San Francisco: Freeman, 1974.

Bower, G. H. Stimulus sampling theory of encoding vari-ability. In A. W. Melton & E. Martin (Eds.), Codingprocesses in human memory. Washington D.C.: Win-ston, 1972.

Carroll, M., & Kirsner, K. Context and repetition effectsin lexical decision and recognition memory. Journalof Verbal Learning and Verbal Behavior, 1982,21, 55-69.

Drewnowski, A., & Healy, A. F. Phonetic factors in letterdetection: A reevaluation. Memory & Cognition, 1982,10, 145-154.

Fillenbaum, S. Words as feature complexes: False rec-ognition of antonyms and synonyms. Journal of Ex-perimental Psychology, 1969, 82, 400-402.

Forbach, G. B., Stanners, R. F, & Hochhaus, L. Repe-tition and practice effects in a lexical decision task.Memory & Cognition, 1974, 2, 337-339.

Glushko, R. J. The organization and activation of or-thographic knowledge in reading aloud. Journal' ofExperimental Psychology: Human Perception and Per-formance, 1979,5,674-691.

Glushko, R. J. Principles for pronouncing print: Thepsychology of phonography. In A. M. Lesgold &C. A. Perfetti (Eds.), Interactive processes in reading.Hillsdale, N.J.: Erlbaum, 1981.

Hintzman, D. L., Block, R. A., & Summers, J. J. Mo-dality tags and memory for repetitions: Locus of thespacing effect. Journal of Verbal Learning and VerbalBehavior, 1973, 12, 229-238.

Jacoby, L. L. Effects of organization on recognitionmemory. Journal of Experimental Psychology, 1972,92, 325-331.

Jacoby, L. L. Perceptual enhancement: Persistent effectsof an experience. Journal of Experimental Psychology:Learning, Memory, and Cognition, 1983, 9, 21-38. ,

Jacoby, L. L. Remembering the data: Analyzing inter-active processes in reading. Journal of Verbal Learningand Verbal Behavior, in press.

Jacoby, L. L., & Dallas, M. On the relationship betweenautobiographical memory and perceptual learning.Journal of Experimental Psychology: General, 1981,770, 306-340.

Jacoby, L. L., & Witherspoon, D. Remembering withoutawareness. Canadian Journal of Psychology, 1982,56,300-324.

Johnston, J. C., & McClelland, J. L. Visual factors inword perception. Perception & Psychophysics, 1973,14, 365-370.

Johnston, J. C., & McClelland, J. L. Experimental testsof a hierarchical model of word identification. Journalof Verbal Learning and Verbal Behavior, 1980, 19,503-524.

Kihlstrom, J. F. Posthypnotic amnesia for recentlylearned material: Interactions with "semantic" and"episodic" memory. Cognitive Psychology, 1980, 12,227-251.

Kolers, P. A. Reading a year later. Journal of Experi-mental Psychology: Human Learning and Memory,1976, 2, 554-565.

Mandler, G. Recognizing: The judgment of previous oc-currence. Psychological Review, 1980, 87, 252-271.

Massaro, D. W., & Klitzke, D. The role of lateral maskingand orthographic structure in letter and word recog-nition. Acta Psychologica, 1919,43, 413-426.

McClelland, J. L. Preliminary letter identification in theperception of words and nonwords. Journal of Ex-perimental Psychology: Human Perception and Perfor-mance, 1976, 7, 80-91.

McClelland, J. L., & Rumelhart, D. E. An interactiveactivation model of context effects in letter perception:Part 1. An account of basic findings. PsychologicalReview, 1981,55, 375-407.

McKoon, G., & Ratcliff, R. Priming in episodic andsemantic memory. Journal of Verbal Learning andVerbal Behavior, 1979, 18, 463-480.

Morton, J. Interaction of information in word recog-nition. Psychological Review, 1969, 76, 165-178.

Morton, J. Facilitation in word recognition: Experimentscausing change in the logogen model. In P. A. Kolers,M. E. Wrolstal, & H. Bouma (Eds.), Processing visiblelanguage I. New York: Plenum Press, 1979. (a)

Morton, J. Word recognition. In J. C. Marshall & J. Morton(Eds.), Psycholinguistics 2: Structures and processes.Cambridge, Mass.: MIT Press, 1979. (b)

Murrell, G. A., & Morton, J. Word recognition andmorphemic structure. Journal of Experimental Psy-chology, 1974, 702, 962-968.

Raaijmakers, J. A., & Shiffrin, R. M. Search of associa-tive memory. Psychological Review, 1981,88, 93-134.

Ratcliff, R. A theory of memory retrieval. PsychologicalReview, 1978, 85, 59-108.

Reicher, G. M. Perceptual recognition as a function ofmeaningfulness of stimulus material. Journal of Ex-perimental Psychology, 1969, 81, 274-280.

Rumelhart, D. E., & McClelland, J. L. Interactive pro-cessing through spreading activation. In A. M. Lesgold& C. A. Perfetti (Eds.), Interactive processes in reading.Hillsdale, N.J.: Erlbaum, 1981.

Rumelhart, D. E., & McClelland, J. L. An interactiveactivation model of context effects in letter perception:


Part 2. the contextual enhancement effect and sometests and extensions of the model. Psychological Re-view, 1982, 89, 60-94.

Scarborough, D. L., Cortese, C., & Scarborough, H. Fre-quency and repetition effects in lexical memory. Jour-nal of Experimental Psychology: Human Perceptionand Performance, 1977, 3, 1-17.

Scarborough, D. L., Gerard, L., & Cortese, C. Accessinglexical memory: The transfer of word repetition effectsacross task and modality. Memory & Cognition, 1979,7, 3-12.

Schneider, W., & Shiffrin, R. M. Controlled and auto-matic information processing: I. Detection, search,and attention. Psychological Review, 1977, 84, 1-66.

Shiffrin, R. M., & Schneider, W. Controlled and auto-matic information processing: II. Perceptual learning,automatic attending, and a general theory. Psycholog-ical Review, 1977, 84, 127-190.

Sperling, G. The description and luminous calibrationof cathode ray oscilloscope visual displays. BehaviorResearch Methods and Instrumentation, 1971,5, 148-151.

Sternberg, S. High-speed scanning in human memory.Science, 1966, 153, 307-311.

Sternberg, S. The discovery of processing stages: Exten-sions of Donder's method. Acta Psychologica, 1969,50,276-315. (a)

Sternberg, S. Memory scanning: Mental processes re-vealed by reaction time experiments. American Sci-entist, 1969, 57, 421-457. (b).

Thorndike, E. L., & Lorge, I. The teachers' word-bookof 30,000 words. New York: Columbia University,1944.

Tulving, E. Episodic and semantic memory. In E. Tulving& W. Donaldson (Eds.), Organization of memory. NewYork: Academic Press, 1972.

Tulving, E., Schacter, D. L., & Stark, H, A. Priming effectsin word-fragment completion are independent of rec-ognition memory. Journal of Experimental Psychology:Learning, Memory, and Cognition, 1982, 8, 336-342.

Underwood, B. J. False recognition produced by implicitverbal responses. Journal of Experimental Psychology,1965, 70, 122-129.

Wallace, W. P. Incidental learning: The influence of as-sociative similarity and formal similarity in producingfalse recognition. Journal of Verbal Learning and Ver-bal Behavior, 1968, 7, 50-54.

Winnick, W. A., & Daniel, S. A. Two; kinds of responsepriming in tachistoscopic recognition. Journal of Ex-perimental Psychology, 1970, 84, 74-81.

Received September 29, 1982Revision received February 10, 1983 •

Third Edition of the Publication Manual

APA has just published the third edition of the Publication Manual, This newedition replaces the 1974 second edition of the Manual. The new Manual updatesAPA policies and procedures and incorporates changes in editorial style and practicesince 1974. It amplifies and refines some parts of the second edition, reorganizesother parts, and presents new material. (See the March issue of the American Psy-chologist for more on the third edition.)

AH manuscripts to be published in the 1984 volumes of APA's journals will becopy edited according to the third edition of the Manual. Therefore, manuscriptsbeing prepared now should be prepared according to the third edition. Beginningin 1984, submitted manuscripts that depart significantly from third edition style willbe returned to authors for correction.

The third edition of the Publication Manual is available for $12 for members ofAPA and $15 for nonmembers. Orders of $25 or less must be prepaid. A charge of$1.50 per order is required for shipping and handling. To order the third edition,write to the Order Department, APA, 1400 N. Uhle Street, Arlington, VA 22201.

journal of experimental psychology: general · the repetition effect. experiment 2 used separate...

Documents