112-125 stimulus similarity and retroactive interference and facilitation in monkey short-term...

Upload: oana-cirjan

Post on 03-Apr-2018

215 views

Category:

Documents


0 download

TRANSCRIPT

  • 7/29/2019 112-125 Stimulus Similarity and Retroactive Interference and Facilitation in Monkey Short-term Memory.

    1/14

    J o u r n a l of Experimental Psychology:A n i m a l Behavior Processes1980, Vol . 6, No. 2, 112-125

    Stimulus Similarity and Retroactive Interferenceand Facilitation in Monkey Short-Term MemoryDouglas L. Medin T homa s J . ReynoldsUniversity of Illinois at Urbana-Champaign Rockefeller University

    John K . ParkinsonUniversity of Illinois at Urbana-ChampaignFour experiments examined th e effects of s im i la r i ty be tween th e sample an dan interpolated st im ulu s in a modified delayed-matching-to-sample ( D M T S )parad igm. The basis t r ia l sequence was as f o l l o w s : (a) a sample w as pre-sented, and the response to it was either rewarded or n o n r e w a r d e d , (b) aninterpolated stimulus was presented, and the response to it was either re-warded or nonrewarded , and (c) after a delay interval a choice test w as givenbetween the ini t ia l sample and a n ew st im ulu s, w ith the sam ple being cor-rect if it had been i n i t i a l ly rewarded and incor rec t if it had been in i t i a l ly non-rewarded. When th e sample and the interpolated stimu lus w ere associated w ithopposite outcomes (sample rew ard ed, interpolated st imu lus not rew ard ed orsample not rewarded, interpolated st imulus r e w a r d e d ) , ret roact ive interferencew as observed to increase as the similarity of the two st imul i increased. Whenthe two st imuli were e i ther both rewarded or both norewarded, ret roact ivefaci l i tat ion was observed so long as the sample and interpolated s t imul i weresimilar along either th e relevant or an irrelevant dimension. Finally, testprobes in volv ing the in te rpo la ted s t im ulu s and a n ew st im ulu s revea led asym-metr i ca l interact ions such that th e in te rpo la ted s t imulus influenced p e r f o r m -ance on the init ial stimulus more than the in i t ia l st imulus altered performanceon th e interpolated st imulus. Taken as a whole , th e results a re inconsis ten twi th the idea tha t alternat ive s t imul i are encod ed ind epend ent ly of each other .A t best, independence m ay hold on ly fo r cases in w h i c h th e s t imul i are nots imi lar to each other.

    From th e ear l ies t beginnings o f w h a t is f e re nce i n h u m a n m e m o r y (e.g. , W i c ke l g r e n ,no w called classical interference theory (Me- 1965, 1966).Geogh, 1932, 1952; Mel ton & I r w i n , 1940; O n e might , the re fore , have expected tha tU n d e r w o o d , 1948, 1957), s t imulus simi la r i ty the recent up sur ge of in terest in an ima lw a s a s s ume d to play a n i m p o r t a n t r o l e in m e m o r y (e.g. , D'Amato, 1973; Honig &ret roac t ive interference (RI) . T he t ransfe r J a m e s , 1971; M e d i n , R ob e r t s , & Davi s ,an d re t roac t i on sur face o f Osgood (1953, p. 1976) would have i n c l ude d a c o r r e s p ond i ng532) , fo r e xa mp l e , p r o v i d e d b o t h a s u m m a r y e m p h a s i s o n st imulus s imi la r i ty effects,of research o n similar i ty effects a n d a f r a m e - R a t h e r t h a n focus ing o n s imi la r i ty re l a t i onsw o r k fo r l a ter research o n re t roac t ive in t e r - in RI, h o w e v e r , theoret ica l treatments havea s s ume d tha t in te r fe r ing s t imul i p r i m a r i l y

    T T a l t e r p e r f o r ma nc e by reducing or eliminatingThis research was supported by U.S. Publ ic . . . . , , . . ,Heal th Service Grants M H 2S134 a n d MH 32489 rehearsal of the to-be-remembered mforma-an d b y a U.S. Public Heal th Service postdoctoral tion in short-term memory ( S T M ) (e.g.,fellowship MN 07SS1 to the second author. Roberts & Grant, 1978; Wagner, Rudy, &Requests fo r repr ints should be sent to Douglas , , , , . , i n m \ T M i uL. Medin, Psychology Department , Univers i ty of Whitlow, 1973). This emphasis is by noIllinois, Champaign, Illinois 61820. m e a n s a r b i t r a r y ; fo r e xa mp l e , Grant an d

    Copyright 1980 by the American Psychological Associat ion, Inc. 0097-7403/80/0602-0112$00.75112

  • 7/29/2019 112-125 Stimulus Similarity and Retroactive Interference and Facilitation in Monkey Short-term Memory.

    2/14

    S T I M U L U S S I M I L A R I T Y A N D S T M 113Roberts (1976) tested pigeons in a clelayed-matching-to-sample paradigm (DMTS) withinterpolated s t imu l i in order to examine theeffects of similarity, cue salience, familiarity,s t i m u l u s complexity, an d degree o f illumina-tion. A small set of stimuli consisting o f f o u rcolors a n d f o u r f o r m s w a s used, a n d withina day only two s t imu l i appeared as the sam-ple ; for example, on color test days red andgreen served as sample and test stimuli,and the interpolated stimulus could be eitherred or yellow (high similarity) or a diagonalor cross pattern (low s i m i l a r i t y ) . Grant andRoberts reported that degree o f illuminationwas the only significant source of retroac-tive i n t e r f e r e n c e ; similarity ha d virtually n ooverall effect .

    Although Grant and Roberts f o u n d thatsimilarity was not a prominent factor control-ling retroactive interference in their pigeons,interactions between particular sample andi n t e rpo la t ed s t imu l i suggested that similarityindeed played an important role. Specifically,very strong RI was observed when the sam-ple was green and the interpolated stimuluswas yellow or when the sample was red andthe interpolated stimulus was blue. In con-trast, little RI (or even modest retroactivef ac i l i t a t i on ) w a s f o u n d when green w a s fol-lowed by a blue or when red was f o l l owed bya yellow.

    How might one interpret the Grant andRoberts results on similarity and RI ? Itseems that one could either emphasize therelation between the interpolated stimulusand the initial stimulus or the relation be-tween the interpolated stimulus and thechoice test. Consider the latter possibility.If a blue stimulus is more similar to greenthan red and if a yellow stimulus is moresimilar to red than green, then the patternof interactions observed by Roberts andGrant would apparently fol low f rom prin-ciples of stimulus generalization. Accordingto this interpretation, one could argue thatthe initial sample and interpolated stimuluswere encoded independently, did not in-fluence each other, and indeed that the samepattern of results would have occurred ifthe initial sample stimulus was eliminatedand only the blue or yellow stimulus had

    been presented prior to the choice test be-tween red and green. This interpretation isconsistent with the Roberts and Grant tracestrength competition model which assumesthat the alternative stimuli are independent.The other main possibility is that the ap-pearance of a sample stimulus and a lateri n t e rpo la t ed stimulus cannot be treated asindependent events and that the particulari n t e r a c t i o n can produce patterns o f bothretroactive interference and retroactive fa -cilitation that cannot be explained by appeal-ing to relations between an interpolateds t i m u l u s and the choice test. The theoreticaland experimental contrasts can best be seenw i t h a concrete example, to which we nowturn.

    Experiment 1The design of Experiment 1, shown sche-

    matically in Figure 1, serves to bring outsome of the theoretical contrasts of interest.S t i m u l i for each trial sequence were drawnf r om a large pool of stimulus objects. There-fo re , although Figure 1 accurately depictsthe logical relations among the alternativetrial sequences, the particular stimuli em-b o d y i n g these conditions varied f r om trialto trial. Characteristics of interpolated stim-uli differentiated the conditions. The fivemain conditions consisted of (a) no inter-polated stimulus (control), (b) an interpo-lated stimulus identical to the sample (A-A),(c) an interpolated stimulus matching thesample along the dimension that is relevanto n the choice test but di f fe r ing f rom thesample and test stimuli in either color orfo rm (A-C), (d) an interpolated stimulusd i f fe r ing f rom the sample and test stimuli inboth color and f o rm (A-D), and (e) an in-terpolated stimulus identical to the test stim-ulus which had not appeared as the sample(A-B). Sample and interpolated stimuliwithin one trial were either both rewarded( + ) or both nonrewarded ( ) . The cor-rect choice on the test was consistent withthe reward status of the sample; that is, ifthe sample was rewarded, it was correct onthe choice test, but if the sample was notrewarded, it was incorrect on the choice test.The task required attention to both the color

  • 7/29/2019 112-125 Stimulus Similarity and Retroactive Interference and Facilitation in Monkey Short-term Memory.

    3/14

    114 D . M E D I N , T . R E Y N O L D S , A N D J . P A R K I N S O Nand fo rm of the sample since on half of thet r ia ls color was relevant and on half formw as relevant.The main compar i son of theoretical in -terest is Condi t ion A -A versus Condi t ionA-C. Since on A - C tr ials th e interpolateds t i m u l u s matches the sample on the relevantdimens ions an d does n o t match th e a l te rna-tive on either color or f o r m , there should bemore general izat ion f rom the in terpolatedst im ulus to the sample than to the al terna t ivestimulus. Given that the same outcome isalways associated with th e sample and in-terpolated s t imulus , performance should befacil i tated, relat ive to the contro l condi t ion .The A -A tr ia ls con tain this sam e source offacilitation, but, in addition, the interpolateds t imulus matches both stimuli along the ir-relevant d imension. Whether matching alonga constant , i r relevant d imension should im -pai r o r improve per fo rmance is a theory-specific question, but most theories predicti mp a i r men t .We distingu ish betw een two classes ofgeneral ization theories. In one class arethose formulations that assume that overal lgeneral ization is a simple, additive funct ionof the general ization f rom i nd iv idua l com-ponent d imensions (e.g., Spence's 1936 dis-cr iminat ion learning theory) . In contras t ,other theories assume that general izationalong each dimension is not independent ofdifferences along other dimensions, and thusthat com ponent dim ensions interact (e.g.,Me din , 1975, 1976; Medin & Schaffer,1978; Spiker, 1963, 1970).To begin with , we focus on predictionsderived f rom analysis of the relation betweeninterpolated stimuli and choice objects . Ad-ditive generalization theories lead to theprediction that A -C trials should produceper fo rmance at least as good as that associ-ated wi th A -A trials. O n A - A a n d A - Ctrials there wil l be the same difference ingeneral ization tendency to the correct and in-correct choice st im ulus a r ising f rom th e rele-vant d imension, and the only question con-cerns the contribution of general ization alongth e i r relevant d imension. Whe n th e sampleand interpolated stimu li both a re rewarded ,some response tendency should be added to

    each choice s t imulus ; when the sample andinterpolated st imul i both are not rewarded,some response tendency should be subtractedf rom each choice stimulus. To observe effectsof these condi t ions on per fo rmance , w e needto specify some choice ru le mapping thesegeneral izat ion tendencies onto choice re-sponses. Choice rules that ignore constant,i r re levant cues (see Flagg & Me di n, 1973,fo r a r e v i e w ) wi l l , of course, predict thatCondi t ion A -C wil l produce per fo rmanceequivalent to Condi t ion A-A, s ince th e samedifference in general izat ion wil l be presentin each case. Choice rules that take absolutestrengths into account, such as rules employ-in g a threshold concept (e.g., Spence, 1956,pp. 204 206) ,present a s o mew ha t m o r e co m-plicated picture. If performance is a m o n o -tonically increasing or decreasing funct ionof absolute strength with the difference instrength between choices held constant, thenthe relative ease of Cond i t i ons A-A and A-Cwill interect wi th r ewar d an d n o n r ewar d .If A-A yields better performance than A-Cwhen both sample and interpolated s t imul iare rewarded, then the order ing should bereversed when th e st imul i a re not r ewar d ed .If p er f o r man ce first increases with absolutestrength and for higher values then de-creases, an y difference between A-A an dA-C tr ia ls should in teract wi th reward andn o n r ewar d . The only case in which A -Acould be un i f o r m l y better than A-C wouldbe if p er f o r man ce first decreased with in -creases in absolute strength a nd later in -creased with absolute strength a nd A - Ctr ials produced strengths that fell near thatm i n i m u m point . To ou r knowledge, n o cur -rent theory employs such a choice rule, andwe tentat ively conclude that addi t ive gener-al ization theor ies imply that A -A tr ials wil ln o t yield better p erform anc e than A -C tr ials .Predict ions of interactive general izationtheories that focus on the relation betweeninterpola ted stim uli and c hoice test objects(e.g., Medin, 1975 ; Medin & Schaffer, 1978)depend on the specific choice ru le employed.In these form ula tions, generalization alongeach s t imulus d imension is not independentof differences along other dimensions . Ac-co rd ing to interactive theories, generaliza-

  • 7/29/2019 112-125 Stimulus Similarity and Retroactive Interference and Facilitation in Monkey Short-term Memory.

    4/14

    S T I M U L U S S I M I L A R I T Y A N D S T M 115tion along th e relevant dimension in the A-Ccondi t ion will be reduced by the differencesbetween the interpolated stimulus and the teststimuli along the i r relevant d imension. If aratio rule is used (as in Medin & Schaffer,1978), this reduction in general ization wil laffect th e correct and incorrect choice s t imul iequally, and there wil l be no difference pre-dicted between A-A and A-C trials . Ifchoices are a function of differences ratherthan rat ios (as in Medin , 1975), then Con-di t ion A-A should yield better performancethan Con di t ion A-C. It should be noted, how-ever, that Medin's (1975) theory specificallyrejects the idea that two alternative stimuliappear ing in the same experimental context(such as the sample an d interpolated stim-u l i ) a re encoded independently. For them om e nt , we propose that accounts of retro-active interference based solely on the re-lation between th e interpolated stimulus a ndthe choice test st imul i generally predict thatA -C trials will produce per fo rmance com-parable with that seen on A-A trials.In contrast, theories that focus on therelation between the sample and interpolatedstimuli predict that A -A trials will be asso-ciated w i th better perform anc e than A -Ctrials. Our rationale is based on an analysisof stimulus processing in relation to rein-statement o r retrieval cues (e.g., Spear,1973, 1976; Wagner et al., 1973). Supposethat when th e sample is presented and asso-ciated with some outcome, the stimulusand outcome are rehearsed together in short-te rm memory . Further assume that th elonger th e stimulus-outcome pair is re-hearsed th e better will be p e rf o r man ce . N o r -mal ly , posttrial events, such as the loweringof the Wisconsin General Test Apparatus(WGTA) screens, tend to disrupt rehearsal(see Motiff , Dekock, & Davis, 1969), bu twh e n th e interpolated stimulus is presented,i t may act as a retrieval cue for the samples t imulus and i ts associated outcome (rewardo r n o n r e w a r d ) and result in additional re-hearsal of the stimulus-outcome pair . Themore similar an interpolated stimulus is toth e in i t ia l s t imulus , th e m o r e effective th einterpolated stimulus would be in reacti-vat ing the initial stimu lus-outcom e pair . A

    red triangle should be a better retrieval cu efo r a red tr iangle than a green tr iangle, andtherefore Condition A -A should yield betterper formance than Cond i t i on A-C.A brief s u m m a r y is in order . If retroactivein terference f rom an in terpolated s t imulusresults solely f rom general ization betweenth e interpolated s t imulus and the choice tests t imul i , then Condi t ion A -C should be at

    least as good as Condition A-A. If retroac-tive interference also reflects co n t r i b u t i o n sf rom th e relation between sample an d inter-polated s t imul i , than A -A should be betterthan A-C.Of secondary interest a re compar i sons be-tween A-D and Contro l t r ia l sequences . Onemight expect A-D trials to yield lower per-fo rmance than control tr ials, to the extentthat pos ts t imu lus processing of the sampleis disrupted by the appearance of an i rrele-van t , interpolated s t imulus . However , the ef-fect might well be small in that performancein this paradigm presumably reflects con-t r ibu t ions from both short- and long-termmemo r y , w i th th e mixture increas inglybiased toward long-term memory as theretention interval lengthens. In Exper iment1 the shortest retention interval was 21 sec(12 sec between sample and interpolateds t imu lus , plus the 3 sec that it takes on theaverage fo r monkeys to respond to the in-terpolated s t im ulu s , p lus the m inim um 6-secdelay in terval between th e interpolated stim-u lus and the choice test). Under these cir-cumstances p er f o r man ce may be pr imar i lybased on long-term memory.

    Fina l ly , the A-B trial sequences were in-cluded as a fo l l ow-up to ear l ier exper iments(R eyno lds & Med in, 1979) aimed at as-sessing the conditions under which monkeyscould learn to choose the first of two samplest imul i if both were given on the choice test.Reynolds and Medin d id not observed abovechance per fo rmance on this condition aftermore than 600 trials . The A-B test in thepresent study w a s more demand ing , in thatReynolds and Medin used a procedure inwhich initial sample stimulus was alwayscorrect on choice tests w herea s in the presentstudy, the sample stimulus could be correcto r i nco r rec t on the choice test, depending

  • 7/29/2019 112-125 Stimulus Similarity and Retroactive Interference and Facilitation in Monkey Short-term Memory.

    5/14

    116 D . M E D I N , T. R E Y N O L D S , AND J. P A R K I N S O NF O R M R E L E V A N T

    SAMPLECO N T R O L A-21,30,48sec.--

    TEST

    A - A - - A B

    A - C A B

    A-D-ABA-B--AB

    CONTROL

    A-A-AB

    A-C-ABA- D - - AB

    A-B-AB

    k-IZ sec. A-6.l5,33sec.-L--l2sec. A-6,15,33!: t^--I2sec.k--l2seo.[!}6,l5,33sec.V& lR

    TESTC O L O R R E L E V A N TSAMPLE

    A-2l.30,488ec.-hr - IZsec. Ar6,l5,33sec.-^tx- I 2 8e c . rb,-- I2sec.(

    - -I2 s eo . ^Figure 1. Schemat ic i l lust ra t ion of the design ofE x p e r i m e n t 1. (A sam ple s t im ulus w a s fo l lowedby ei ther no interpola ted st imulus [Cont rol ] or byinterpola ted st imul i tha t var ied in t he i r s im i l a r i t yto the sample and a l ternat ive test s t i m u l u s . )on its reward status when it appeared asth e sample .Method

    Subjects. The subjects were f ou r pigta i ledm o n k e y s (Macaco nemestrina), tw o rhesus m onkeys(Macaco mulatto), and one cynomolgus monkey(Macaco fascicularis). A ll were 5-8-yr-old j ung le-born females with previous discr iminat ion andDMTS experience. They were m a in t a ined o n a12: 12 hr light/dark cycle initiated at 0600 hourswere fe d labora tory chow da i ly af ter test ing .Apparatus and materials. The m onkeys weretested in a darkened room whose outside soundswere masked by white noise. The WGTA w a sl ighted by two 15-W f luorescent bulbs, and thethree food wells on the gray f o r m b o a r d w e r espaced IS cm center to center . The s t i mu l i were 10geometric shapes cu t f rom .61-cm plywood ( s t a r ,a r row, semic i rc le , t r iangle, inver ted u , inver ted v,x, hourglass, house, bar) in each of the 10colors (red, green, yellow, blue, black, white,orange, purple, brown, tan). T he objects wereequated su bject ively for s ize by the exper im enterssuch tha t they appeared to be of equal overall area,and each was presented f la t on the food t r a y (theobjects were not mounted on a base) in a s ingleor ien t a t ion th roughout testing.

    Procedure. The exper imenta l design (shownschemat i ca l ly in F i g u r e 1) consisted of the fac to r i a lcombina t i on of S t r ial types X 2 r e levant d i m e n -sions (color or f o r m ) X 2 r ewa rd cond i t i ons ( sam -ples rewarded or no t r e w a r d e d ) X 3 delay inter-va ls (21, 30, 48 sec) for a total of 60 t r i a lsequences.On cont rol t r ia ls , a s ingle objec t was presentedover th e center food wel l , and the m o n k e y dis-placed i t for e i ther a ra is in rew ard ( + ) or no re-w a r d ( ) . F o l l o w i n g a delay, a choice test w a sgiven with the sample and a new object coveringth e side food wells . If the sample ha d been re-w a r d e d , i t was cor rec t an d r e w a r d e d on the tes t ;i f i t had not been r ewa rded , the new s t i m u l u s w ascor rec t a n d r e w a r d e d .On inter ference t r ia ls , th e sam ple ob jec t w aspresented over th e center food wel l , and the m o n k e ydisplaced it for either a r a i s in r ewa rd ( + ) or nor e w a r d ( ) . Fol lowing a 12-sec i n t e r s t im ulusin terval , an i n t e rpo la t ed s t im ulus w a s given, be-ing assoc ia ted w i th the same rew ard /no - rew ardcond i t i on as the ini t ia l sample, that is, the twopresen ta t i ons were ei ther both rewarded or bothn o n r e w a r d e d . Af t e r a delay, a choice test w a sgiven with the f i r s t sample and a new st imuluscovering the side food wells. The correct, rewardedobject on the test w a s consistent wi th th e r e w a r dsta tus of the in i t ia l sam ple . Each s t i m u l u s a ndtest presentat ion ended with the retract ion of thefo rm board and the l o w e r i n g of the opaque sc reenof the WGTA.The f ou r inter ference condi t ions shown in Figure1 were dist inguished by the s imi la r i ty rela t ionamong the samples, interpola ted st imul i , and testobjects : in Condi t ion A-A, i t was ident ica l to thesam ple ; i n A-C, it differed on the irrelevant di-mens ion o n l y ; and in A-D, i t d i ffe red on both there l evan t an d ir relevant dimensions. Fina l ly , th ein t e rpo la t ed s t im ulus in the A-B condi t ion dif -fered f r om the first sample on the re levant d i-mension only, but unlike other trial sequences,th e interpola ted st imulus appeared a long with thesamp le on the choice test.These S basic t r ial types ( con t ro l , 4 i n t e r f e r -ences) X 2 r ewa rd cond i t i ons X 3 delays werepresented with either color o r f o r m as the rele-vant dimension. When color was relevant , the f i r s tsamp le and the two choice objects were a l l thesame f o r m ; when fo rm w as relevant , they werea ll the same color .The m on keys w ere tested S days/w eek, 30 trials/day , for 48 days. In each session the tr ials w erepresented in one of eight quasi- random orders as-sembled such that half of the t r ia ls had colorrelevant a n d hal f , f o r m re levant , and the sampleswere rewarded and nonrewarded equally often.Fur ther , each of the 60 tr ial sequences appearedonce every tw o sessions, and the posi t ion of thecor rec t objec t o n a test w as r andom ized bu t bal-anced w i t h i n a day,subject to the cons t r a in t thatth e correct object occupy th e same posit ion nomore than four t r ia ls in a row. The 72 objects for

  • 7/29/2019 112-125 Stimulus Similarity and Retroactive Interference and Facilitation in Monkey Short-term Memory.

    6/14

    STIMULUS SIMILARITY AND STM 117each session were randomly drawn without replace-me n t f ro m the 100-object stimulus pool , with a llthe monkeys receiving the same stimuli on anygiven day. T he i n te r t r i a l in te rva l w a s a constant15 sec, correct responses were rewarded with ara i s in , a n d a noncorrect ion procedure w as employed.Results

    The m ain co ndi t ions were assoc ia ted w i thsubstantial differences in performance, butcolor-versus-form tests and rewarded sam-ples versus nonrewarded samples d id nota l te r pe r form anc e . R esu l t s for the m a in c on-d i t ions of interest are shown in Figure 2.Both A-A an d A-C tr ials produced facilita-t ion re la tive to control (no interpolated st im-u lus ) , and consistent with the idea that in-terpolated st imuli affect processing of samplestimuli, Condition A - A w a s associated withbet te r pe r formance than Condi t ion A-C.The A-D condit ion w as clearly n o worse an dperhaps slightly better than the control con-di t ion . The differences interacted somewhatwi th delay; the crucia l A-A, A-C differencebeing more apparent at the longer delays.Finally, the A-B condition produced per-f o r ma nc e that w as con sistent ly below chance(48%, 43%, 40% correct at delay intervalsof 21, 30, 48 sec).Since a preliminary analysis of 8-day prac-tice blocks indicated that there were no sig-nificant practice effects, the data were col-lapsed across days and a Delays X RewardCondit ion X Trial Types X Color VersusForm-relevant analysis of variance w as con-ducted. Data f rom A -B trials were excludedsince this condition did not bear directly onthe interpretations of RI and st imulus gen-era l izat ion under considerat ion. The overallanalysis of errors indicated that the effectsof delays, F(2, 12) = 6.18, MSe = 7.90,p < .05, trial types, F(3, 18) = 22.08, MSe= 4.63, p < .001, and the interaction of De-lays X Trial Types, F(6, 36) = 2.40, MSe 3.64, p < .05, were significant. No othereffect or interaction w as statistically sig-nif icant.

    Individual compar isons employing a N e w -man -Keu l s test a nd the .05 co nfid enc e levelindicated that Condition A - A w a s betterthan Conditions A-C, A-D, and Control andthat Condit ion A-C in turn yie lded bet ter

    .90-

    oUJ

    oo.80-

    .70-

    21 30SAMPLE-TEST INTERVAL

    48(SEC.)

    Figure 2. Proportion correct as a function of sam-ple test i n te rva l fo r the main cond i t ions in Experi-m e n t 1. (See Fi gu re 1 for iden t i f i ca t ion of g r o u p s . )performance than A-D an d Control se -quences we r e no t rel iably dif ferent . Detailedanalyses of the Condi t ions X Delay inter-act ion revealed th e fo l lowing pattern o f d i f -ferences : A t the 21 -sec d elay, Co nd itions A -Aand A -C w ere rel i ab ly be tter than A -D andC o n t r o l ; at the 30-sec delay, Condition A -Aw as better than th e other three cond i t i ons ;at the 48-sec delay, A -A perfo rm an ce w asbetter than A-D an d Control performance.Discussion

    A f u l l r epe t i t ion of the sample s t imulusw as m o r e effective than repet i t ion o f j u s tthe r e levant s t im ulu s d im ens ion . Th is r esu l tis consistent with models that assume thatth e in i t i a l sample s t imulus and an in te r -polated st imulus are not independent. Spe-cifically, the interpolated st imulus may ac tas a retr ieval cue for the sample representa-t ion and thereby a l low addit ional processingof the sample s t imulu s .A n interpolated stimulus different in bothcolor an d form f rom th e choice st imulineither facilitated no r impaired performancecompared with control t r ia ls . The persis-tently poor performance of A-B tr ial se-quences indicates that the monkeys were un-able to base their performance on the first oftw o sample st imuli when both appeared onthe test. This result is mildly surprisingsince some investigators have argued thatDMTS involves a tempora l d isc r imina t ion

  • 7/29/2019 112-125 Stimulus Similarity and Retroactive Interference and Facilitation in Monkey Short-term Memory.

    7/14

    118 D . M E D I N , T . R E Y N O L D S , A N D J . P A R K I N S O NSAMPLE

    CONTROL A 15,30 sec . -TEST

    DIFFERENT-*^

    Figure 3. Schematic illustration of the design ofExpe r imen t 2. (The in terpolated st imu lus a lw aysmatched the sample and the al te rnat ive s t im ulus incolor . )(e.g., D ' A m a t o , 1973), an d o n that basisone would expect that either th e second orthe first of two sample st imuli could controlperformance.One possible object ion to the main con-trast between the A-A and the A-C condi-t ions is that an exact repetit ion of the samplest imulus may have some special (perhapsconf igura t iona l ) properties. The next ex-per iment extended th e generality of the find-ings of Ex perim ent 1 to the case in w hichth e interpolated st imulus w as never a repeti-tion of the in i t ia l sample .

    E xpe r im e n t 2The key cont r as t in Exper iment 2 was be-tween the cont ro l condi t ion and a condi t ionin wh ic h th e interpolated st imulus matchedthe sample on the irrelevant dimension. If

    the sample was a red tr iangle, for example,and the choice test was to be between the redt r iang le and a red square (i .e. , color ir rele-v a n t ) , then th e interpolated stimulus mightbe a red circle (see Figure 3). On the aver-age, th e interpolated st imulus should gen-eralize equal ly to the two choice st imuli ,since th e interpolated st imulus matches th echoice s t im u l i in color and is not more s imi-lar to one choice object than the other inf o r m .If th e interpolated st imulus inf luences per-f o r ma nc e solely in te rms of general izat ion tothe test stim ul i, interpola ted stim uli eitherwil l not help or may impair performance.The f u l l design of Exper iment 2 is sche-

    matized in Figure 3. Control t r ia ls have noin terpola ted s t imulus and prov ide a base l ineto evaluate the other tr ial types. The in te r -pola ted st imulus on the other t r ia l sequencesdiffers f rom th e choice st imuli in fo rm bu tmatches them in color . O n Same tr ials th esample and in terpola ted st im uli are associ-ated with the same outcome (both rewardedo r both not r e w a r d e d ) , a nd on Differentt r ia ls the sample and interpolated st imulusare associated with different outcomes (onerewarded , one nonrewarded). As in Experi-m en t 1, the correct choice on the test w asdetermined by the reward sta tus of thesample. In a dd it ion , both the interval be-tween the sample and interpolated stimulusand the delay between th e i n i t i a l sample an dth e test were varied.A n y outcome associated with th e inter-polated stimulus should generalize equallyto the sample and alternative-choice tests t imu lus . By the same logic developed inExper iment 1, addit ive general izat ion mod-el s based on extant choice rules predict eithertha t the interpolated st imulus wil l have n oeffect o r that its effect wil l interact withw hether or not the interpolated st im ulu s isrewarded and will thereby produce facili-ta t ion fo r one ou t c om e an d im pa i r m e n t fo rth e other (a t this level o f generality on ec a nno t specify which outcome will producewhich effect, only that reward an d n o n r e -ward should produce oppos i te effects) . In -te rac t ive general izat ion models, whetherbased on ratios or differences, predict thesame lack of effect so long as one considersonly genera l iza t ion f rom th e interpolateds t i mu l us to the test s t imuli .If an interpolated st imulus is v iewed asa retr ieval cue, it could provide the oppor-tuni ty for addi t iona l process ing of the i n i -t ial sample and i ts associated outcome. OnSame tr ials this should only be beneficial,but on Diffe ren t t r ia ls th e memory representa-t ion for the sample s t imulus may be modif iedby th e (d i f f e ren t ) outcome paired with theinterpolated stimulus. This should inter-fere wi th pe r f o r m a nc e . This inter ferencem ay possibly derive f rom th e sample an dinterpolated st imulus being rehearsed si-mul taneosu ly in shor t- term memory. Tech-

  • 7/29/2019 112-125 Stimulus Similarity and Retroactive Interference and Facilitation in Monkey Short-term Memory.

    8/14

    S T I M U L U S S I M I L A R I T Y A N D S T M 119nically, this state of af fa i r s might arise eitheri f th e interpolated stimulus reactivated therepresentation of the sample s t imu lus or ifthe sample stim ulu s w as stil l being rehearsedat the t im e the in terpolated s t im ulus w aspresented. We prefer the f o rm e r possibilitybecause it leads more directly to the idea thatsimilar i ty between the in terpolated s t imulusand the sample is a cri tical variable al teringper formance .According to the above l ine of thinking,w e should observe RI o n Different t r ia ls butretroactive faci l i tation on Same trials . Davisan d Fitts (1976) studied forgetting inmonkeys , using a direct analogue of theSame versus Different contrast shown inFigure 3 . They reported that RI was sub-stantial ly greater in the Different condi t ionthan in the Same condi t ion . Since thei r ex -per iments had a somewhat d i f ferent pur-pose, they used stim uli having no specials imi lar i ty relat ions (magazine pictures) a n dd id not include a contro l condi t ion havingno in terpolated s t imulus .Method

    Subjects. The subjects were the f our pigtailed,two rhesus, and the one cynom olgus m onkeys usedin Experiment 1. A pprox im ate ly 2 mo and a moveto a new laboratory intervened between Experi-ment 1 and Exper imen t 2 during which t ime th emonkeys participated in DMTS studies using junkobjects. In these experiments proactive interferencew as studied as a function of intertrial interval andinterstimulus interval, with some samples rewardedand some not rewarded.Apparatus and materials. The monkeys wereagain tested in a WGTA wi th 252 commonly usedand manufactured ( junk) objects that were eachpainted 1 of 14 different colors. Both the WGTAand th e form board were black rather than gray.The st imul i were a lways the same color within atrial, but they differed in color between trials.The st imulus pool w as exhausted, and the st imul iwere randomly re-sorted every 3 days.Procedure. The experimental condi t ions com-prised a factorial design with certain missing cellsusing the variables of trial type, in terst imulus in -terval , reward , and delay interval. Trial sequences

    were of three main types as indicated in Figure 3:Control , Same, o r Different. Control trials in -volved a single sample presentation that was re-warded or nonrewarded, fo l lowed by a de lay of15 or 30 sec, and a choice test between the sampleand a new st imulus. On Same trials either 6 or12 sec after the sample was responded to , an in-

    terpolated st imulus matching the sample in colorbut not form was presented and associated with thesame reward outcome as the sample. When the in-terst imulus interval was 6 sec, delays of either 6or 21 sec preceded the choice test so that thesample-to-test interval matched that of Controltr ia ls . For the 12-sec in terstim ulus in terval , a 15-sec delay interval was used to match the longercontrol trial delay. Different t r i a l s mi r ro red Samet r i a l s exactly except that the outcome associatedwi th the in terpolated st imulus was a lways the op-posite of that associated with the sample on thatt r i a l . The correct object on the choice test w asconsistent with the outcome associated with thesample, that is, if the sample w as rewarded, i t wascorrect on the choice test, and if it was not re-warded, i t was incorrect on , the choice test.The co m bination of trial types, sam ple rew ardstatus, interst imulus interval, and delay intervalproduced 16 un ique presentation sequences, whichwere presented twice each session in one of 6quasi- random orders. The position of the correctobject on a test was randomized, but balancedwithin a day, subject to the constraint that thecorrect object occupy the same position no morethan f our t r ia ls in a row. Testing continued 32trials each day, 5 days a week, for 4 wk. The in-tertrial interval was a constant 15 sec, correct re -sponses were rewarded with a ra is in, and a non-correct ion procedure was employed.Results

    The main resul ts a re shown in Figure 4 .Compared wi th th e contro l condi t ion , in -terpolation of a stimulus sharing the valuealong th e constant , i r relevant d imension hada d ramat i c effect. When reward condi t ionsfo r the sample and interpolated stimulus dif-fered (Different t r ia ls) , retroact ive inter-ference w as strong (performance on Differ-en t tr ials w as significantly above chancebu t very poor ) ; w hen reward cond i t ions fo rth e sample and interpolated stimulusmatched, retroactive facilitation was ob-served, especially at the longer delay inter-val. This pattern of results is exactly thatpredicted by the idea that the initial samplepresentat ion and the in terpolated s t imulusdo not act as independent events. The re-sults cannot be explained in terms of gen-eral izat ion f rom the interpolated stimulus toth e choice s t imul i . The interval betweenth e sample and the interpolated stimulus hada negligible effect, and forgetting with delayinterval is apparent only o n contro l trials.Statistical tests confirmed this pattern of

  • 7/29/2019 112-125 Stimulus Similarity and Retroactive Interference and Facilitation in Monkey Short-term Memory.

    9/14

    120 D . M E D I N , T . R E Y N O L D S , A N D J . P A R K I N S O N

    i-oLU . 7 0 -

    |.50+a.

    ControlDifferent-12* Different-6

    15 30SAMPLE-TEST INTERVAL (SEC.)

    Figure 4. Proportion correct as a funct ion ofsample-test in terval for the main condit ions inExper iment 2 .

    retroactive interference and facilitation. Be-cause the experiment was not a completefactorial design, two separate analyses ofvariance were performed. The first excludedthe interpolated stimulus trials with a 12-secinterstimulus interval and used the factorsof trial type, delay interval, and sample re-ward versus nonreward. This analysis indi-cated that th e effect o f trial type w as signif i -cant, F(2, 12) = 30.21, MSe = 16.5, p

  • 7/29/2019 112-125 Stimulus Similarity and Retroactive Interference and Facilitation in Monkey Short-term Memory.

    10/14

    S T I M U L U S S I M I L A R I T Y A N D S T M 121on Different trials, relative to low stimulussimilar ity. For the high similar ity condition,th e interpolated st imulus differed from th esample st imulus in both color and form, asin the A-D condi t ion in Exper iment 1. Thestimuli for the choice test were a lways of thesame color an d di f ferent form. Generaliza-t ion models consider ing only general izat ionf rom the interpolated stimulus to the choicestimuli predict that neither variable shouldhave much effect, since generalization fromth e interpolated st imulus to the two choicest imul i should be approxim ately equal.

    MethodSubjects and apparatus. The four pigta i led andone cynomolgus monkeys used in the first two ex-periments again served as subjects. Experiment 3immedia tely followed Exper iment 2 and employedthe same appara tus .Stimuli. A pool of 288 junk objec ts tha t wereeach painted 1 of 14 di fferent colors w as used.Each day's testing required 96 objects, and whenthe stimulus pool was exhausted, the st imul i wererandomly re-sorted and used again.Procedure. The experimental design consistedo f a 2 X 2 X 2 X 2 factor ia l with trial types ( Sam eversus Different) , reward status (sample rewardedversus nonrewa rded) , de l ay (6 , 21 sec) , and s imi -lari ty (high versus l o w ) of the interpolated stim-u lus to the sample as fac tors . On each trial thesample w as presented and r e w a r d e d or nonre-warded, 6 sec la ter an interpola ted st imulus waspresented and rewarded or nonrewarded, and aftera delay of 6 or 21 sec a choice w as given betweenthe sample and a new stimulus of the same colorbut a different form. If the sample had been re-warded, i t was correct on the test ; i f i t had notbeen rewarded, the other stimulus w as correct. OnSame t r ia ls the sample and interpola ted st imuluswere associated with the same outcome (both re-w a r d e d ) , a n d o n Different trials they were associ-ated with contrasting outcomes. The interpolatedstimulus was always di fferent in form f rom th eother objects used within th e trial . On high-simi-la r i ty t r i a l s th e interpolated stimulus was the samecolor as the other objects, and on low-simi la r i tytrials th e interpola ted st imulus differed in bothcolor and fo rm .Each of these 16 dist inc t trial sequences appearedtwice each day in one of s ix quasi -random orders

    fo r 20 days. The position of the correct object ona test was rand om ized, ba lanced wi thin a day, andsubject to the constra int tha t th e correc t objec toccupy the same position no more than four trialsin a r o w . The intert r ia l interva l was a constant 15sec, correct responses were rewarded with a ra is in,and a noncorrec t ion procedure was employed.

    Table 1Proportion Correct on Same an d DifferentTrials as a Function of the Similarity of theInterpolated Stimulus to the Other Stimuli

    Trial typeSimi la r i ty Same Different

    HighLo w .16.2 0 .4 2.39

    ResultsOverall, subjects averaged 18% erro rs onSame tr ia ls an d 36% e r ro r s o n Differenttr ials. A s Table 1 shows, h igh similar i ty fa -c i l i ta ted performance on Same t r ia ls and im-pa i red per form anc e on Different trials. Thisis jus t th e interaction predicted on the basisof th e idea that the in i t ia l sample and theinterpola ted st im ulu s are no t processed in-dependent ly .A n analysis of variance with the factors ofsimilar i ty, delay, t r ia l type, and reward re-vealed that the m a in effect of tr ial type

    (Same versus D if fe ren t ) was s ign i f ican t ,F(l, 14) =84.0, MSe = 20.0, p < .01, asw as the theoret ica l ly impor tant Tria l TypeX Sim i la r i ty in ter ac t ion , F(l, 4 ) = 9.2, M S?= 3.27, p < .05. The effect of delay ap-proached signif icance, F(l, 4 ) =4.66, M S f= 3.68, .05 < p < .10. N o other effect or in-teraction w a s signif icant.Discussion

    The results provide yet another demon-st ra t ion that sample and interpolated st imulido not ac t independent ly in inf luencing d e-layed matching performance. Monkeys madetwice as m a n y erro rs when th e sample an dinterpolated st im ulu s w ere associated w ithdifferent ra ther than matching reward out-comes. Further, these effects depended on thesimilarity of the interpolated and samples t imul i . When the sim ilar i ty w as h igh, per-f o r ma nc e w as better on Same tr ials andworse o n Different t r ia ls than when simi-la r i ty wa s l ow . This interact ion betweens imi lar i ty a n d tr ia l type w a s produced whenthe interpolated st imulus had no more for-m al simi la r i ty to the sample stimulus than

  • 7/29/2019 112-125 Stimulus Similarity and Retroactive Interference and Facilitation in Monkey Short-term Memory.

    11/14

    122 D . M E D I N , T . R E Y N O L D S , A N D J . P A R K I N S O Nto the new s t i mu l u s for the choice test. Theidea that the inf luence of an in terpolateds t im u lus can be described solely on the basisof general ization to the choice stimuli can-no t accoun t fo r this pattern of results . A l-though the in teract ion between s imi lar i tyan d tr ia l type w as qui te smal l , i t was statis-t ical ly rel iable, a n d w e have subsequentlyreplicated this effect as par t of a largers tudy, using a different g r o u p of monkeys( M e d i n , Note 1).

    Exper iment 4A major unresolved quest ion is whether

    the interactions between th e sample and theinterpolated s t imulus a re und i rec t iona l .T h a t is to say, if the in terpolated s t imuluswas pai red wi th a new object on the choicetest, would interactions between th e process-ing of the in i t ia l sample and the interpolateds t im u lus produce proact ive in terference andfaci l i tat ion s imi lar to the retroact ive effectsoccurring when the first sample is probed ona choice test ? Predictions f rom the idea thatthe interpolated stimulus acts as a retrievalcue leading to access an d fu r ther processingof the sample stimulus depend on fu r therdeta i ls concerning just what is reactivatedand rehearsed (e.g., just the representation ofth e in i t ia l s t imulus o r both th e representat ionof th e i n i t i a l s t imu lus and its associated out-come) . In the one study bear ing o n this is-sue, Davis an d Fitts (1976) f o u n d that th ereward s ta tus of the second of two st imul iaffected per fo rmance on the f irst st imulussubstantial ly, but they noted little if any in-fluence of the reward status of the first st im-ulus on performance based on the second.The Davis an d Fitts study used unrelatedst imul i , and it is possible that proactive in-f luences might be greater for more s imi larst imuli . Exper iment 4 tested this possibility.Method

    Subjects an d apparatus. The four pigtailed andone cynomolgus monkeys from the earlier experi-ments again served as subjects. E xperiment 4 im-media tely followed Experiment 3 and used thesame apparatus.Stimuli. Th e pool of 288 junk objects eachpainted 1 of 14 different colors was again em-ployed. Each day's testing required 96 objects, and

    when the s t imulus pool w as exhausted, th e s t imul iwere randomly re-sorted and used again.Procedure, The experimental design involvedth e fac to rs of trial types (Same versus Different) ,probe type (f i rs t sam ple tested versu s second sam-pl e tested) , reward sta tus (probe rewarded assample versus probe not rewarded as sample) , anddelay (short versus long) . On each trial the firstsample was presented and rewarded or nonre-warded, 6 sec later a second sample was presentedand r e w a rd e d or nonrewarded , and after a fur therdelay of 6 or 15 sec, either the first or the seconds t imu lus appeared, along with a new s t imu lus o nthe choice test. If the sample had been rewarded,it was correct on the test; if it had not been re-warded, the other st imulus was correct . On Sametria ls the first and second sample st imul i were as-sociated with the same outcome, and on Differenttrials the two samples were associated with con-t rast ing outcomes. The two sample s t imul i andthe new st imu lus on a given trial were always ofthe same color, but color could vary between trials.Each of these 16 distinct presentation sequencesappeared in one of six quasi- random orders twiceeach day. Testing continued 5 days a week for 18days. The posi t ion of the correct object on a testwas randomized, balanced wi th in a day, and sub-ject to the constraint that the correct object occupythe same position no more than four trials in arow. The in tertr ia l in terval was a constant IS sec,correct responses were rewarded with either ara is in or a smal l piece of apple, and a noncorrec-t ion procedure was employed.Results

    Performance as a function of trial type,probe type, and delay is shown in Table 2.The interference produced by D i f f e r e n t out-comes is obvious, being more clearly evi-denced when the first sample was probedthan when the second sample was probed. Inother words, the interaction between the twosample stimuli was asymmetrical, with thefirst of two sample stimuli being influencedby the reward status of the second stimulusmore than the second was influenced by thefirst.

    A 2 x 2 X ' 2 x 2 analysis of variance wasconducted with the factors of trial type,probe type, reward status, and delays.1 The

    1Technically, Delay is not a proper factoria lvariable because the total delay between the to-be-probed sample and the choice test depended onwhether the first or the second sample was probed.It is more convenient, however, to treat delay asa factor ia l variable than to compute separate analy-ses of variance, and the main experimental resultsdo not h inge on this practice.

  • 7/29/2019 112-125 Stimulus Similarity and Retroactive Interference and Facilitation in Monkey Short-term Memory.

    12/14

    S T I M U L U S S I M I L A R I T Y A N D S T M 123mai n effects of probe type, F(l, 4 ) = 11.42,MSe = 18.50, p < .05, and tr ial type, F(l, 4)= 43.33, MSe = 10.44, p < .01, were signifi-cant , wi th tests using th e second sample pro-ducing better performance than tests withthe first sample, and Same trials being betterthan Different tr ials . The interaction of probetype with tr ial type, F(l, 4) = 35.58, MSe= 5.94, p < .01, and the three-way interac-tion invo lving tr ial type, probe type, and de-lay, F(l, 4) = 10.21, MSe = 6.70, p < .05,were also reliable. Ne w ma n- K e u l s tests re-vealed that th e difference between Same andDifferent outcomes w as significant when th esecond sample w as probed as wel l as w h e nthe first sample w as probed (p < .05). Thethree-way interact ion in Table 2 appears toarise from th e fact that performance droppedonly sl ightly with delay on Same or Differenttr ials when th e second s t imulus w as probedbu t dropped considerably on Same trials andactual ly improved with delay on Differentt r ia ls when the first s t imulus w as probed.Discussion

    Perfo rm an ce on the second of two samplesw as inf luenced by whether or not its ou tcomematched that of the first sample, which indi-cates that the interactions between samplestimuli produced both proactive an d retro-active effects. The interaction of the samplest imul i , however , was asymmetr ical , wi thretroactive effects being substantial ly largerthan the proactive effects. The idea that thesecond stimulus acts as a retrieval cue to ac-cess and provide an oppor tun i ty fo r fu r therrehearsal of the first s t imulus is generallycompatible with these results. To handle th easymmetrical interference effects, one couldoffer the assum ption that w hen m em ory forthe ini tial sample stimulus is reactivated, theoutcome associated with i t may also be reac-tivated an d that this outcome has some(sm all) proba bil i ty of becom ing associatedwi th the second sample. Another possibilityis that these modest proactive effects mightarise on t r ia ls when animals remember thatthe two sample s t imul i were associated withdifferent outcomes but have forgotten whichoutcome w as associated with which st imulus.The results of Experiment 3 imply that such

    Table 2Proportion Correct for the Two Probe Typesan d th e Same Versus Contrasting SampleOutcomes

    DelayFirst sampleprobed

    Same DifferentSecond sampleprobed

    Same DifferentShortLong .86.77 .55.63 .82.81 .7 9.7 5

    an explanat ion might need to take in to ac-coun t the s imila r i ty of the in i t ia l s t imulusto the second (or in terpolated) s t imulus .General Discussion

    These exper im ents show a co nsistent, clearpattern of interact ions between sample andinterpolated st imul i . A s might be expected,when th e sample an d interpolated s t imulusa re both rewarded o r both no t rewarded ,facilitation is observed if both s t imu l i havethe same value along th e relevant dimension(Ex p er i men t 1, Condi t i ons A -A and A-Cversus Contro l ) . Less obvious is the findingthat this facil i tat ion is greater if the sampleand in terpolated s t imulus match on a co n -stant , i r relevant d imension as well (Exper i -m e n t 1, Condi t i on A -A versus A-C). Per-haps most surprising of all is the finding thatretroact ive facil i tat ion an d retroact ive in ter-ference ( i f the sample and in terpolated s t im -u lus a re associated with contrasting out-comes) are observed when two s t imul i shareonly a constant , i r relevant d imension an ddiffer along the relevant d imension (Exper i -men t s 2 and 3).The present results a re inconsis tent wi thth e idea that when two stimuli are presentedconsecutively as samples, they are encodedindependently and that any patterns of in-terference or faci l i tat ion can be described int e r ms of a s imple, ad di t ive funct ion of gen-eral izat ion along component s t imulus d i-mensions . The first experiment provided evi-dence that componen t s t imu lus d imens ionscanno t be treated as additive, a result con-sistent with a large body of results in thedom a in of discr im inat ion learning (e.g. ,

  • 7/29/2019 112-125 Stimulus Similarity and Retroactive Interference and Facilitation in Monkey Short-term Memory.

    13/14

    124 D . MEDIN, T. REYNOLDS, AND J. PARKINSONSpiker, 1963, 1970; Meclin, 1975, 1976).E xpe r im e n t s 2 and 3 showed that th e differ-ence between performance when the sampleand interpolated st imulus have the same out-come and per formance when they havecontrast ing outcomes is greater when th etwo st imuli match each other o n a constanti r re levant dimension than when they do not .Taken as a whole, the results of these ex -periments suggest that the under ly ing mech-an isms produc ing nonindependence of s t im-ulus events are closely tied to retr ieval andrehearsal processes. The idea that an in te r -polated st im ulu s acts as a retr ieval cue thatm ay lead to further processing of the in i t ia ls t imulus is consistent with the results of Ex-per im ent 4 w hich showed that whether ornot the reward status of the f irst and secondst imulus matched inf luenced performancem ore o n test probes in volving the fi rst s t im u-lus than o n test probes involving th e seconds t imu lus .A l t ho u g h the t ies to the discr iminat ionl e a rn ing l i terature give some assurance thatth e present analysis will have some gener-ality, i t is worth considering whether thepresent results have implications for DMTSprocedures in which sample presentat io ns areno t rewarded and reward s are available onlyon choice tests (e.g., D ' A m a t o , 1973) .The idea that a s t im u lus may act as aretrieval cue is not restricted to si tuat ions inwhich sample presentations are associatedwith r e i n f o r c e m e n t . On e might , the re fore ,expect to observe retroactive facilitation sim-i lar to that reported here in other situations.To opt imize the l ikel ihood of detect ing suchan effect, it might be necessary to use m o r ethan a few alternative samples in order toreduce between-tr ia l inter ference. Anotherpossibly re levant considerat ion is that in ourexperiments using a WGTA, each st imulusan d test presentation ends with retractiono f the fo rm board and l owe r ing of the opa quescreen which should tend to in te r fe re wi thshor t- term memory processes. This "inter-rupt ion" may set the stage for interpolateds t imul i to be more necessary and effectiveas retr ieval cues. If the in i t ia l sample pre-sentation was always being actively rehearsedwhen an interpolated st imulus was presented,

    o n e can imagine that only retroact ive inter-ference, ra ther than fac i l i ta t ion, wo uld beobserved. In any event , th e present experi-ments suggest that accounts o f a n im a l m e m -o ry m ight benef it f rom reconsiderat ion of theo ld idea that stimulus similar ity plays animpor tant role in inter ference (and faci l i ta-t i o n ) . The idea that the a l ternat ive st im uliappearing on a trial can be t reated as inde-penden t m ay hold only for the case in wh ic hth e s t imu l i a re to ta l ly d iss im i la r .

    Reference N o t e1. Medin , D. L. Stimulus similarity an d interactions

    in monkey short-term memory. Unpublishedm a n u s c r i p t , 1978.References

    D ' A m a t o , M. R. Delayed matching an d shor t - te rmm e m o r y in monkeys . In G . H. Bower (Ed.),The psychology of learning an d motivation (Vol .7). New Y o r k : A cad e mi c Press, 1973.Davis, R . T., & Fitts, S. S. Memory and cod ingprocesses in d iscr iminat ion learn ing . In D. L.Medin, W. A. Roberts, & R. T. Davis (Eds.),Processes of animal memory. Hil l sda le , N.J.:E r l b a u m , 1976.Flagg, S. F., & Medin , D. L. Consant i r re levantcues and s t imulus generalization in monkeys.Journal of Co mparative and Physiological Psy-chology, 1973, 5, 339-345.Grant , D. S., & Rober ts , W. A. Sources of retro-active inhibit ion in pigeon short-term m e m o r y .Journal of Experimental Psychology: AnimalBehavior Processes. 1976, 2, 1-16.H o n i g , W. K., & J ame s , P. H. R. ( E d s . ) . Animalmemory. New Y o r k ; Academic Press, 1971.McGeogh, J . A . Forgetting and the l aw of disuse.Psychological Review, 1932, 39, 352-370.McGeogh, J . A. The psychology of human learn-ing. N ew York: Longmans, Green, 1952.Medin , D. L. A theory of contex t in d i s c r i mi n a t i o nlearn ing . In G. H. Bo w e r (Ed.), The psychol-og y of learning and motivation (Vol . 9). NewYork: A cad e mi c Press, 1975.Medin , D. L. A n i m a l m o d e l s an d me mo ry mo d e l s .In D. L. Medin , W . A . Roberts , & R. T . Davis(Eds.), Processes of animal memory. Hil l sda le ,N. J. : E r l b a u m , 1976.Medin , D. L., Roberts , W . A ., & Davis, R. T .(Eds.). Processes of animal memory. Hil l sda le ,N . J.: E r l b a u m , 1976.Medin , D. L., & Schaffer, M. M. Contex t theoryof c lass i f i ca t ion learn ing . Psychological Revieiv,1978, 85, 207-238.Melton , A . W., & Irw in , J . M. The i n f l u e n ce ofdegree of i n te rpo la ted learn ing on re t roact ive

  • 7/29/2019 112-125 Stimulus Similarity and Retroactive Interference and Facilitation in Monkey Short-term Memory.

    14/14

    STIMULUS S I M I L A R I T Y A N D S T M 12 5inh ib i t ion and overt t ransfer of specific responses.American Journal of Psychology, 1940, 53 , 173-203.Motiff , J. P. , Dekock, A. R., & Davis, R. T. Con-cealment of st imul i dur i ng de l ay in the delayed-response problem. Perceptual and Motor Skills,1969, 29, 788-790.Osgood, C. E. Method and theory in experimentalpsychology. New Yo rk : Oxford U niversi ty P ress ,1953.Reynolds, T. J., & Medin, D. L. Strength versustemporal order informat ion in delayed matchingto sample per fo rmance by monkeys . AnimalLearning and Behavior, 1979, 7, 294-300.Roberts, W. A ., & Gran t, D. S. A n ana lysis ofl ight- induced retroactive inhibition in pigeonshor t - te rm memory . Journal of ExperimentalPsychology: Animal Behavior Processes, 1978,4, 219-236.Spear, N. E. Retr ieva l of m emory in anim als .Psychological Review, 1973, 80, 163-194.Spear , N. E . Retrieval of memor ies . In W. K.Estes ( E d . ) , Handbook of learning and cognitiveprocesses (V o l . 4). Hil lsda le , NJ.: Er lbaum,1976.Spence, K. W. The nature of di sc r im ina t ion lea rn-ing in animals. Psychological Review, 1936, 43,427-449.Spence, K. W. Behavior theory an d conditioning,N ew Haven, Conn. : Yale Universi ty Press, 1956.

    Spiker, C. C. The hypothesis of s t imulus i n te rac -t ion and an explanat ion of s t imu lus compounding .In L. P. Lipsitt & C. C. Spiker ( E d s . ) , Ad-vances in child development and behavior (Vol .1) . Ne w Y o rk : A cad em ic Press, 1963.Spiker, C. C. An extension of Hull-Spence dis-c r imina t ion l ea rn ing theory . Psychological R e-view, 1970, 77, 496-515.Underwood, B. J . Retroac t ive and proac t ive in-hibition after f ive and for ty-eight hours . Journal

    o f Experimental Psychology, 1948, 38, 29-38.U n d e r w o o d , B. J. Interference and forgetting.Psychological Review, 1957, 64 , 49-60.Wagner, A . R., Ru d y , J . W., & Whi t l ow, J . W.Rehearsa l in anim al con di tioning. Journal of Ex-perimental Psychology, 1973, 97 , 407-426.Wickelgren, W. A. Acoust ic s imi la r i ty and ret ro-ac t ive interference in short- term memory. Jour-nal of Verbal Learning and Verbal Behavior,1965, 4, 53-61.Wickelgren, W. A . Phonemic s im i la r i ty and inter-ference in shor t - te rm memory fo r single letters.Journal o f Experimental Psychology, 1966, 71,396-404.Winer, B. J. Statistical principles in experimentaldesign. New York: McGraw-Hi l l , 1962.

    Received March 12, 1979Revision received A u g u s t 31. 1979