monkey memory: same i different concept learning, serial ...different by moving the response lever...

18
Journal of Experimental Psychology: Animal Behavior Processes 1984, Vol. 10, No. 4, 513-529 Copyright 1984 by the American Psychological Association, Inc. Monkey Memory: SameI Different Concept Learning, Serial Probe Acquisition, and Probe Delay Effects Anthony A. Wright and Hector C. Santiago University of Texas Health Science Center at Houston Graduate School of Biomedical Sciences Stephen F. Sands University of Texas at El Paso Three rhesus monkeys were trained and tested in a same/different task with six successive sets of 70 item pairs to an 88% accuracy on each set. Their poor initial transfer performance (55% correct) with novel stimuli improved dramatically to 85% correct following daily item changes in the training stimuli. They acquired a serial-probe-recognition (SPR) task with variable (1-6) item list lengths. This SPR acquisition, although gradual, was more rapid for the monkeys than for pigeons similarly trained. Testing with a fixed list length of four items at different delays between the last list item and the probe test item revealed changes in the serial-position function: a recency effect (last items remembered well) for 0-s delay, recency and primacy effects (first and last list items remembered well) for 1-, 2-, and 10-s delays, and only a primacy effect for the longest 30-s delay. These results are compared with similar ones from pigeons and are discussed in relation to theories of memory processing. The primacy effect, good memory of the first list items, has been recently demonstrated for two monkey species (Roberts & Kraemer, 1981; Sands & Wright, 1980a, 1980b) and a chimpanzee (Buchanan, Gill, & Braggio, 1981). The rinding of animal primacy effects, as well as the more common recency effects (remembering the last list items well) is im- portant because it demonstrates the continuity between human and nonhuman primate memory processing. Prior to the discovery of a nonhuman primate primacy effect, serial- position functions for monkeys revealed only recency effects (Davis & Fitts, 1976; Devine & Jones, 1975;Gaffan, 1977). Recency effects The research was supported by National Institute of Mental Health Grant MH 35202 to Anthony A. Wright. Preparation of this article was supported in part by Fogarty International Fellowship (1 F06 TW00827) to Anthony A. Wright while he was on leave in New Zealand. We are indebted to David Floyd for his assistance with some of the experiments and we thank D. F. Kendrick, R. G. Cook, P. J. Urcuioli, and D. C. McCarthy for their comments and encouragement during the re- search and preparation of the article. Requests for reprints should be sent to Anthony A. Wright, University of Texas Health Science Center, Sensory Sciences Center, 6420 Lamar-Fleming Avenue, Texas Medical Center, Houston, Texas 77030. typically show a general monotonic decay as the retention interval is lengthened in much the same way that memory for single items shows monotonic decay with retention inter- val (Cox & D'Amato, 1982; D'Amato, 1973; Medin, Reynolds, & Parkinson, 1980; Over- man & Doty, 1980). Indeed, any memory performance based upon a single underlying mechanism would be expected to show a monotonic decrease with retention interval. The U-shaped serial-position function may be different. A dual process is likely operating. It has been suggested that two storage mech- anisms are related by a rehearsal process (e.g., Atkinson & Shiffrin, 1968; Waugh & Norman, 1965). Others argued that two dif- ferent interference mechanisms combine to produce the U-shaped serial-position function (e.g., Postman, Stark, & Fraser, 1968; Postman & Underwood, 1973). Recall experiments with human subjects have shown the primacy effect to be invariant over a wide variation in the retention interval, whereas the recency effect rapidly dissipates as retention interval is lengthened (e.g., Glan- zer & Cunitz, 1966; Postman & Phillips, 1965). These different time courses for the primacy and recency effects argue for a dual- process, serial-position function. 513

Upload: others

Post on 14-Oct-2020

1 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Monkey Memory: Same I Different Concept Learning, Serial ...different by moving the response lever to the right or left, respectively. Correct responses were followed by a tone (500

Journal of Experimental Psychology:Animal Behavior Processes1984, Vol. 10, No. 4, 513-529

Copyright 1984 by theAmerican Psychological Association, Inc.

Monkey Memory: Same I Different Concept Learning, SerialProbe Acquisition, and Probe Delay Effects

Anthony A. Wright andHector C. Santiago

University of Texas Health ScienceCenter at Houston

Graduate School of Biomedical Sciences

Stephen F. SandsUniversity of Texas at El Paso

Three rhesus monkeys were trained and tested in a same/different task with sixsuccessive sets of 70 item pairs to an 88% accuracy on each set. Their poor initialtransfer performance (55% correct) with novel stimuli improved dramatically to85% correct following daily item changes in the training stimuli. They acquireda serial-probe-recognition (SPR) task with variable (1-6) item list lengths. ThisSPR acquisition, although gradual, was more rapid for the monkeys than forpigeons similarly trained. Testing with a fixed list length of four items at differentdelays between the last list item and the probe test item revealed changes in theserial-position function: a recency effect (last items remembered well) for 0-sdelay, recency and primacy effects (first and last list items remembered well) for1-, 2-, and 10-s delays, and only a primacy effect for the longest 30-s delay. Theseresults are compared with similar ones from pigeons and are discussed in relationto theories of memory processing.

The primacy effect, good memory of thefirst list items, has been recently demonstratedfor two monkey species (Roberts & Kraemer,1981; Sands & Wright, 1980a, 1980b) and achimpanzee (Buchanan, Gill, & Braggio,1981). The rinding of animal primacy effects,as well as the more common recency effects(remembering the last list items well) is im-portant because it demonstrates the continuitybetween human and nonhuman primatememory processing. Prior to the discovery ofa nonhuman primate primacy effect, serial-position functions for monkeys revealed onlyrecency effects (Davis & Fitts, 1976; Devine& Jones, 1975;Gaffan, 1977). Recency effects

The research was supported by National Institute ofMental Health Grant MH 35202 to Anthony A. Wright.Preparation of this article was supported in part byFogarty International Fellowship (1 F06 TW00827) toAnthony A. Wright while he was on leave in NewZealand. We are indebted to David Floyd for his assistancewith some of the experiments and we thank D. F.Kendrick, R. G. Cook, P. J. Urcuioli, and D. C. McCarthyfor their comments and encouragement during the re-search and preparation of the article.

Requests for reprints should be sent to Anthony A.Wright, University of Texas Health Science Center, SensorySciences Center, 6420 Lamar-Fleming Avenue, TexasMedical Center, Houston, Texas 77030.

typically show a general monotonic decay asthe retention interval is lengthened in muchthe same way that memory for single itemsshows monotonic decay with retention inter-val (Cox & D'Amato, 1982; D'Amato, 1973;Medin, Reynolds, & Parkinson, 1980; Over-man & Doty, 1980). Indeed, any memoryperformance based upon a single underlyingmechanism would be expected to show amonotonic decrease with retention interval.The U-shaped serial-position function maybe different. A dual process is likely operating.It has been suggested that two storage mech-anisms are related by a rehearsal process(e.g., Atkinson & Shiffrin, 1968; Waugh &Norman, 1965). Others argued that two dif-ferent interference mechanisms combine toproduce the U-shaped serial-position function(e.g., Postman, Stark, & Fraser, 1968; Postman& Underwood, 1973).

Recall experiments with human subjectshave shown the primacy effect to be invariantover a wide variation in the retention interval,whereas the recency effect rapidly dissipatesas retention interval is lengthened (e.g., Glan-zer & Cunitz, 1966; Postman & Phillips,1965). These different time courses for theprimacy and recency effects argue for a dual-process, serial-position function.

513

Page 2: Monkey Memory: Same I Different Concept Learning, Serial ...different by moving the response lever to the right or left, respectively. Correct responses were followed by a tone (500

514 A. WRIGHT, H. SANTIAGO, AND S. SANDS

One purpose of the experiments reportedin this article was to determine whether ornot dual processes are indeed responsible forthe monkey's U-shaped serial-position func-tion and whether or not the monkey too hasdifferent time courses for its dual processes.Varying the retention interval (called probedelay) in a serial-probe-recognition (SPR)task is a better method than varying retentioninterval in a recall task. It would, of course,be impossible to conduct a free-recall exper-iment with monkeys even if this were themore desirable procedure. In the SPR task,a list of items is presented and followed (aftera probe delay) by a single test (probe) item.The subject then indicates whether or not theprobe was in the list. In a recall task, on theother hand, the subject has a free-recall periodwhere as many items as possible are recalled.Thus, more time has elapsed when the lastitems are recalled (effectively longer retentioninterval) than when the first items were re-called. This time difference due to recallorder may be particularly important at shortretention intervals where it may be a substan-tial fraction of the retention interval itself.The SPR task takes more time to generate aserial-position function, but the retention in-terval is precisely controlled because only oneitem is tested with each trial.

The monkeys used in this study had notpreviously participated in memory experi-ments, and in order to test their serial-positionfunctions at different probe delays, they hadto be trained in several steps to perform theserial-probe-recognition task. The first exper-iment reports training in a same/differenttask which is the first step. In addition,conditions are reported that promote same/different concept learning, which by the na-ture of being a relational strategy sharescommon properties with the strategy requiredto perform the SPR task.

Experiment 1: Same/Different Acquisition

The first step in training the monkeys toperform a complicated task like the SPR taskis to make it as simple as possible. In thiscase, it means using only one list item andpresenting it simultaneously with the probeitem so that they overlap in time. The mon-keys can look back and forth between the

two before making their choice response.This is a samel different task: If the items areidentical a same response is correct, and ifthey are not identical a different response iscorrect. Only a small change is necessary tomake the same/different task an SPR task. Ifthe two items do not overlap in time, thefirst item becomes the list item, an SPR taskwith one list item. It is then a simple matterto display more than one list item; if theprobe item matches any one of the list items,a same response is correct; otherwise, a dif-ferent response is correct. Monkeys trainedto use a same/different concept may be morelikely to learn the SPR task rapidly than theywould without this concept.

One of the purposes of the first experimentwas to test for development of any same/different concept by testing the degree towhich they would transfer performance froma learned set of items to a completely newset of items. Regarding a same/different con-cept, the subjects make their choice responsebased upon the relation of the probe item totheir memory of the list item. This is a formof memory scanning (cf., Sternberg, 1966)whereby the list item is stored during itspresentation, retrieved when the probe itemis presented, and compared with the probeitem. If the memorial representation matchesthe probe item, then they are judged to bethe same; otherwise, they are judged to bedifferent. The alternative to memory scanning,a relational response, is an item-specific re-sponse. The correct response to individualitems of the pair are memorized. We felt thatif we could encourage relational responding,as demonstrated by good transfer in thesame/different task, then the subjects mightmore rapidly acquire the subsequent SPRtask than if they were item-specific respondingin the same/different task. These were thepurposes of Experiment 1; the first phaseinvolves same/different training with six dif-ferent sets of items, and the last phase involvestransfer with additional training interveningbetween the transfer tests.

Method

Subjects

The subjects were 3 fourtyear-old experimentally naiverhesus monkeys (Macaco mulatto), Joe, Linus, and Max.

Page 3: Monkey Memory: Same I Different Concept Learning, Serial ...different by moving the response lever to the right or left, respectively. Correct responses were followed by a tone (500

MONKEY MEMORY 515

The monkeys obtained their entire daily ration of foodand liquid as reinforcers in the experimental chamber.They were reinforced with a 5-cc squirt of Tang orangejuice or a 1-gm Noyes banana pellet for correct responses;the particular reinforcer was determined on a randombasis for correct responses. This deprivation schedule wasmaintained during the 5 working days. On Friday after-noons, they were given free access to Purina monkey labchow and water. On Saturdays, they were given 500 ccof water and 20 Purina monkey chow pellets. On Sundays,they were deprived of food and water in order to getthem ready for the next week's experimental trainingand testing.

Apparatus

The monkeys worked the experimental sessions in astandard primate chair with movement somewhat re-stricted by a collar. The monkeys viewed two rectangular(12 cm X 9 cm) rear-projection screens arranged verticallyand separated 16 cm center to center. A Carousel slideprojector (Kodak 760H) was positioned 61 cm behindeach screen. The images were viewed at a distance ofapproximately 61 cm, and subtended visual angles of 12degrees vertically and 20 degrees horizontally. Solenoid-operated shutters, constructed "in house," controlledpresentation times of the slide items. A three-position(left, right, and down) lever was placed within easy reachof each monkey's right hand. A Cromemco Z-2D micro-computer controlled the experiment, and collected andanalyzed the data.

Same/Different Procedure

Trials began by the simultaneous projection of twoslides, one on each of the two screens. On same trials,identical slides were displayed on top and bottom screens;on different trials, nonidentical slides were displayed. Themonkeys' task was to classify the slides as either same ordifferent by moving the response lever to the right or left,respectively. Correct responses were followed by a tone(500 Hz), a reinforcer (either a banana pellet or a 5-ccsquirt of Tang orange juice), and a 2-s intertrial interval.Incorrect responses were followed by illumination of thehouse light for 10 s and no reinforcer. Each sessionconsisted of 70 trials; 35 were same trials and 35 weredifferent trials, intermixed with one another in a pseu-dorandom sequence. A correction procedure was usedwhereby a trial was repeated until the correct responsewas made. This helped remove any response biases thatmight have developed. However, all the data presented inthis article included only the first-trial performances, andthe correction procedure was used only during this initialset of six acquisitions. Generally, three 70-trial sessionswere conducted daily; the same 70-trial session wasrepeated until the monkeys performed at 88% correct orbetter on two sessions (not necessarily consecutive). Theywere then trained on the second 70-trial session and soon until they had mastered all 6 sessions.

The stimuli were 210 distinctly different 35 mm colorslides of fruit, flowers, animals, people, and other naturaland man-made objects. All items were used in the firsttwo 70-trial sets of items. For the third and fourth sets,these items were reorganized, and in some cases items

Table 1Organization and Reorganization of Color-SlideStimuli Into Same/Different Trial Types

Slide numbers

Acquisitions

1 and 23 and 45 and 6

1-70

SameDifferentDifferent

71-140

DifferentDifferentSame

141-210

DifferentSameDifferent

that were used on same trials were used on differenttrials and vice versa. For the fifth and the sixth acquisitions,these 210 different items were reorganized again. Themethod of organization and reorganization of the itemsfor the six different acquisitions is shown in Table 1. Asan example, notice the group of items designated bynumbers 1 through 70. Half of them were shown withtheir identical copies as same trials in the first acquisitionand the other half in the second acquisition. Pairs ofitems from this group were used in the third and fourthacquisitions as different trials. Similarly, the group ofitems identified by the numbers 141 through 210, whichinitially were used on different trials in the first andsecond acquisitions, were used (along with their identicalcopies) on same trials in the third and fourth acquisitions.The reason for this organizational scheme was to breakdown any item-specific associations that the monkeysmight attach to individual items; it was hoped that theywould learn to attend exclusively to relations betweenitems of each pair. Good transfer to novel items wouldbe evidence that the monkeys were using a relationalstrategy.

Transfer Procedure

The procedures used for transfer testing and the trainingbetween transfer tests were similar in many respects tothe previously described acquisition procedures. However,there were three differences: The items were presentedsuccessively rather than simultaneously, no correctionprocedure was used, and the monkeys initiated eachtrial. Trials began with a faintly audible clicker (5 Hz).Each monkey pressed down on the three-position lever,which displayed the first item in the upper screen. Thefirst item was displayed for 1 s, and 1 s after its removala second item appeared on the lower screen. This seconditem remained in view until the monkey made a choiceresponse (lever movement to the right = same response;lever movement to the left = different response) or until10 s elapsed, whichever occurred first. An incorrectresponse or no response during the choice responseperiod (abort) resulted in a 10-s time-out period withthe house light illuminated.

Baseline trials used in these transfer tests were thoseof the sixth and final acquisition. Transfer tests wereconducted on five consecutive sessions; each session wascomposed of 50 baseline trials plus 20 transfer trials.Twenty of the original 70 baseline trials were removedfrom the session, and 20 pairs of novel stimuli weresubstituted; the particular substitutions varied amongtest sessions. These novel test stimuli were shown only

Page 4: Monkey Memory: Same I Different Concept Learning, Serial ...different by moving the response lever to the right or left, respectively. Correct responses were followed by a tone (500

516 A. WRIGHT, H. SANTIAGO, AND S. SANDS

once (by definition of novelty), and correct choice re-sponses were reinforced. Previously, we had discoveredthat when transfer test trials were conducted in extinction,the monkeys learned that responses to novel stimuli werenot rewarded (Wright, Santiago, Urcuioli, & Sands, 1984).This discrimination was learned even with partial rein-forcement for correct responses on training trials. Theextinction test, consequently, proved to be an invalid testof the subjects' transferability.

The two transfer tests were separated by training. Theamount of training varied somewhat from monkey tomonkey: Joe received 4 days, Max 8 days, and Linus 12days. Four new sets of 70 trials composed of novel itemswere used in this training. Each day a particular set wasrepeated three times, and then the next day a differentset was used. If training progressed beyond 4 days, thenit was repeated on the four sets of stimuli in the sameorder. The training order was Sets A, B, C, D for Joeand Linus and Sets D, A, C, B for Max.

Results

Same/Different Acquisition

Individual acquisition results for 2 monkeysare shown in Figure 1. One monkey, Max,participated in only three out of six acquisi-tions. Consequently, its acquisition data, al-though similar to the other monkeys' data,

are not presented for this phase of the exper-iment. The main result shown in Figure 1 isthat the first acquisition took over twice aslong as subsequent acquisitions. Notice thatthe first acquisition for Linus was plotted interms of 560 trials per block; Linus's firstacquisition took approximately four timeslonger than the others. Another result is thatthe even-number acquisitions began at about70% correct, whereas the odd-number acqui-sitions (except for Joe on the fifth acquisition)began at about 50% correct (chance perfor-mance). An examination of Table 1 may helpto explain this difference in acquisition. Onthe second acquisition, all the slides werecompletely novel to the monkeys and as suchrepresented a transfer test. Their first-sessionperformance on the second acquisition (notshown |in Figure 1 because it is an average oftwo sejssions) was 72% correct. This wasmoderately good transfer performance. Thethird acquisition began at chance perfor-mance, which was not completely surprisingbecause a substantial number of previouslyseen items were rearranged into new combi-

OLUOCtroo

LUoQCLU0.

100

80

60

40

100

80

60

40

LINUS

(360 TRIALS PER SLOCK)

JOE

ACQUISITIONS1 •—« 4 o—a2 0—'0 5 in— A3 o—o 6 1—*

4 8 12 16 20 24 28 32 36

BLOCKS OF 140 TRIALSFigure 1. Six successive acquisitions for two monkeys with each set containing 70 pairs of color-slidestimuli. (See Table 1 for an explanation of the composition of trials using 210 different pictures.)

Page 5: Monkey Memory: Same I Different Concept Learning, Serial ...different by moving the response lever to the right or left, respectively. Correct responses were followed by a tone (500

MONKEY MEMORY 517

JOE MAX LINUS

100

80

60

40

20

T m BASELINE

. .-

1

1S,s•

»

J

i

jJ

B

«

«

*

1

f |T

A

1

TIi

QS TRANSFER T

i

1

t

I

o111DCCCOOI-

lilODCUlQ.

1 2 1 2 1 2

TRANSFER TESTS SEPARATED BY TRAINING

Figure 2. Transfer test results with novel pictures (hatched histograms) and baseline performance withfamiliar trials and pictures (open histograms). (Range markers are performance ranges over five sessionsfor each transfer test, each session contained 50 baseline trials and 20 transfer trials. Transfer tests [1 and2] for each monkey were separated by training [4-12 days] where there were daily changes among foursets of training stimuli.)

nations; some had been used to make upsame trials and were now used on differenttrials and vice versa. During the first sessionof the third acquisition, performance on trialswhere the role had changed (from same todifferent or from different to same) averaged31% correct, whereas on trials where therewas no role change (for slide numbers 71through 140) performance was perfect (100%correct). None of the items in the thirdacquisition was used in the fourth acquisition.Like the third acquisition, the fourth acqui-sition was made up of items from both ofthe first two acquisitions. The same argumentsused to explain the initial performance onthe third acquisition might be expected toapply to the fourth acquisition. First-sessionperformance on the fourth acquisition, how-ever, revealed that the monkeys performed at67% correct on items where the role hadchanged and 67% correct on items where therole had not changed. This trend in the first-session performance may represent the begin-ning of concept development by these mon-keys. Indeed, their first-session performancenever again fell below chance performance asit did on the third acquisition for items wherethere was a role change. On the fifth andsixth acquisitions the monkeys' performanceto items with a role change was 63% correct

in both cases, again better than chance per-formance. However, 63% was not very goodtransfer and was even a slight decrease intransfer performance, rather than an im-provement which might be expected if themonkeys had continued to develop the same/different concept.

Transfer

Transfer results from the first transfer testare shown in the left-hand pair of histogramsfor each monkey in Figure 2. Performanceon the 100 transfer trials is shown by thehatched histograms, and performance on the250 baseline trials is showed by the unfilledhistograms. There was no significant transferon the first test for Joe and Max, t(5) = 1.03,p > .1 ns; t(5) = 2.3, p > .05 ns, respectively,and although Linus's transfer was significantlydifferent from chance, t(4) = 3.58, p < .05,it was less than chance performance.1 Alltransfer performances were significantly dif-ferent (p < .05) from their baselines, F[l,8) = 8.2, F(l, 8) = 123.8, F(l, 8) = 23.2 forJoe, Linus, and Max, respectively.

1 These are significance tests for a single mean becausethe performance is tested against a fixed value—50% orchance performance (see Hays, 1963, p. 311).

Page 6: Monkey Memory: Same I Different Concept Learning, Serial ...different by moving the response lever to the right or left, respectively. Correct responses were followed by a tone (500

518 A. WRIGHT, H. SANTIAGO, AND S. SANDS

Prior to the second transfer test, the mon-keys were given training on four new sets ofitems. Their individual session performanceis shown in Table 2. On the three dailysessions of the third set of items, Joe showeda dramatic improvement in performance,possibly an example of insight learning.This improved performance was maintainedthrough the fourth set of items and was stablethereafter. As shown in Figure 2, performanceon the second transfer test for Joe improveddramatically. The other two monkeys weretrained somewhat more than Joe on the foursets of items, and their performance is shownin Table 2. Their transfer performance im-proved also. Transfer by all three monkeysdid not differ from their baseline perfor-mances, F(\, 8) = 2.4, p > .20; F(\, 8) =0.0, p > .75; F(\, 8) = 0.5, p > .75 for Joe,Linus, and Max, respectively, and was signif-icantly above chance, *(4) = 11.4, p < .002;t(4) = 11.6, p < .002; t(4) = 9.8, p < .002for Joe, Linus, and Max, respectively.

Discussion

Previous studies of acquisition and transferof the same/different concept showed thatgreat apes (chimpanzees and orangutans) canacquire and correctly apply this concept(King, 1973; Robinson, 1955, 1960). Theexperiment of this article showed that rhesusmonkeys can also learn and correctly apply

Table 2Percent Correct Performance on Slide SetsSeparating Transfer Tests

Slideset

ABCD

ABCD

ABCD

Linus

55, 60, 4965,' 68, 7172, 76, 7551,59,49

57, 72, 7281,71,7470, 87, 8454, 61, 64

77, 78, 7580, 74, 8075, 77, 8874,74,71

Monkey

Max

55, 54, 6864,68,7177, 71, 7261,65,55

83, 78, 8075, 83, 8684, 86, 9470, 81, 88

Joe

66, 71, 7764, 75, 6584, 87, 9686, 89, 94

OUJococoo

"l-UJo£

o 0o> O

o> 0-ik 0

to 0

. _.

iT11

n B*, E TR

3Af

f

= LJS

NEPER

MONKEY PIGEONAVERAGE AVERAGE

Figure 3. Comparison of monkey and pigeon transferson their Anal transfer test. (Range markers specify thetotal range of performance for three monkeys or twopigeons.)

the same/different concept. Indeed, the 85%correct transfer by the monkeys of this ex-periment was even better than that shown bythe apes. The apes' transfer performance wasless than 80% correct. In one experiment byKing (1973), their transfer performance wasless than 70% correct. The apes' transferperformance also was typically less than theirtraining item performance. By contrast, ourmonkeys showed no deficit in transfer per-formance relative to training item perfor-mance. Thus, with regard to a. same/differentconcept, monkeys were not shown to beinferior to apes in whether or "not they couldlearn the concept or in the degree to whichthey could correctly apply it.

A companion experiment to the one pre-sented here with monkeys was conductedwith pigeons (Santiago & Wright, 1984). Pi-geons were trained and tested with identicalitems and with procedures virtually identicalto those used with monkeys. Pigeons, however,did not transfer as well as monkeys, F(\,23) = 44.6, p < .001 for individual sessionsfor each species as replications. Figure 3shows average pigeon transfer and averagemonkey transfer for comparison. Althoughpigeons did not transfer as well as monkeys,they did show transfer nonetheless, r(9) =13.8, p < .002 for individual sessions asreplications. To our knowledge this is thebest reported transfer by pigeons. The pigeons'baseline performance was less than the mon-keys' although at one time both species had

Page 7: Monkey Memory: Same I Different Concept Learning, Serial ...different by moving the response lever to the right or left, respectively. Correct responses were followed by a tone (500

MONKEY MEMORY 519

performed about equally well on these base-line trials. Pigeons had met the 88% perfor-mance criterion and had maintained betterthan 80% correct for several baseline sessionsbefore transfer. The decrement in the pigeons'baseline performance indicates that the trans-fer test itself may have been disruptive. Pi-geons are generally more disrupted by pro-cedural changes than are monkeys. Possiblycontinued training with daily item changesinterspersed with occasional novel items mighteventually produce pigeon transfer as goodas monkey transfer.

The monkeys' transfer performance was85% correct and as good as their baselineperformance. This means that they had fullydeveloped the same/different concept. Themonkeys could deal with new items in thesame way they dealt with old items. Decisionswere made on the basis of the relation betweenthe items of each pair. Because they hadnever before seen any of the items, they hadno opportunity to memorize the correct re-sponses to individual items. This was notalways the case, as shown by their poor initialtransfer. Apparently, repeating the same trialsover and over again promoted developmentof item-specific strategies to the detriment ofrelational ones.

Premack (1983) raised the question ofwhether transfer, such as reported in thisarticle, is actually based upon a concept orwhether it is based upon the animal makinga familiarity judgment, choosing "the itemthat it has seen before or seen most recently"(p. 356). Premack did not present evidencethat any transfer results were based uponfamiliarity judgments or that they were notbased upon a same/different concept. Indeed,he did not even identify what the propersame/different task strategy should be in orderto qualify as a concept, what the familiaritystrategy actually is in this task, or how theydiffer. For purposes of this discussion, a same/different concept is denned as one where (a)the subject compares the second item (or itsmemorial representation) with its memory ofthe first item, (b) a similarity criterion is usedto make the response decision, and (c) thesubject responds same if this comparisonmeets the criterion and different if does not.

Good transfer performance, and hence ev-idence of good concept development, neces-

sitates that the subject: (a) use tie relationalstrategy as outlined above, (b) use a suffi-ciently strict criterion for judging the pair ofitems, and (c) adequately learn the properidentification responses of samet.nd different.

If subjects showing good st me/differenttransfer could be separated into those thatuse same I different concepts and those thatuse familiarity judgments, then these twogroups would probably be distinguished bydifferences in their criterion (b sibove) ratherthan by any basic strategy differences (aabove). Such a distinction would be one ofdegree (quantitative), not kind (qualitative).Premack (1983), on the other hand, seemedto argue for a basic strategy difference, aqualitative difference.

The common use of the tern familiarityimplies that the comparison extends backwardin time, possibly to all previous items seenduring the session. Perhaps if the comparisonwere restricted to the single previous item,then this strategy would be closei to a conceptfor those who distinguish betwjen conceptsand familiarity judgments. Thfe basic rela-tional strategy would remain the same inboth cases. Support for this argument comesfrom a monkey memory-scanning experiment(Sands & Wright, 1982). The procedure (inExperiment 2) had been changed by preview-ing (in prior lists) some of the probe itemson different trials. Thus, for ths monkey tobe correct the relational judgment had to belimited to list items of the current trial. Thememory-scanning results revealed that thiswas indeed what the monkey did. The monkeyquickly adapted to this procedu re change, anunlikely adjustment if the morkey had hadto learn a completely different strategy.

Another common use of the term famil-iarity judgment is that it is a snap judgment,one that is imprecise or superficial. For ex-ample, a probe item might be judged (on thebasis of a snap, familiarity judgment) to bethe same as the previous item, if both con-

sky notwith-ildings, and

tained large expanses of bluestanding differences in land,so on in the lower portions of the pictures.Training may have laid the foundation forsuch a strategy. If all pictures had previouslybeen very different from each onher, then thissnap-judgment strategy should work well.Training establishes how similar the two items

Page 8: Monkey Memory: Same I Different Concept Learning, Serial ...different by moving the response lever to the right or left, respectively. Correct responses were followed by a tone (500

520 A. WRIGHT, H. SANTIAGO, AND S. SANDS

(images) must be in order to be correctlyjudged identical. Once again this is a differ-ence in criterion, not a basic strategy differ-ence. Tests of this argument could be con-ducted. Lures (probe items on different trials)could be selected to be more similar to theirrespective preceding item. The subjects wouldthen have to make detailed comparisons inorder to be correct, not just gross or snapjudgments. Properly selected pairs of testitems, tested before and after this procedurechange, would reveal changes in the degreeof comparison and changes in similarity cri-terion.

Thus, if subjects were trained to comparethe test item (or its memorial representation)only to the previous item, make a point-by-point comparison of these memorial repre-sentations, and respond same only when theywere identical in all respects, would not theybe said to be responding on the basis of asame/different concept? Even the most skep-tical concept theorist should agree. All ofthese can be accomplished by restricting thejudgment criterion of what is commonlythought of as a familiarity judgment. It doesnot involve a different strategy.

Finally, the strategy used in the SPR taskcan also be argued to be based upon a same/different concept. The task is similar to asame/different task, except that there is awhole list of items preceding the test itemrather than a single item. The subject retrievesthe list items one at a time from memory(cf., Sternberg, 1966) and compares each withthe test item (or memory of the test item). Ifany list item matches the test item in allaspects, then the subject responds same. Ifnone matches, the subject responds different.(Again, the requirement that the list itemmatch the test item "in all aspects" is acriterion adopted by the subject and imposedby the experimenter through the selection ofthe stimuli and the training conditions.) Thissame behavior and strategy is performedwhether the items are new (transfer test), old,or just in new combinations. We regard suchbehavior to be evidence for, and based upon,a same/different concept. Performing thesame/different task on the basis of the sameconcept used in the SPR task (same/differentconcept) was possibly an important step inthe monkeys' acquisition of the SPR task.

The monkeys' acquisition of the SPR task ispresented in the next experiment.

Experiment 2: Serial-Probe-Recognition Acquisition

Experiment 2 reports the acquisition ofSPR performance. The monkeys began SPRtraining immediately after demonstrating ex-cellent transfer in Experiment 1. They couldthus be expected to perform well on one-item, list-length trials, but how would theydo with longer list lengths? By presentingthem with list lengths that varied from oneto six items, how much training would berequired before they learned all list lengths?These were some of the questions exploredin Experiment 2.

Method

Subjects and Apparatus

The monkeys were the same three monkeys thatparticipated in the previous experiment. The apparatusused was identical to that described in the first experiment.

Procedure

Trials began by the monkey pressing down on thethree-position lever in the presence of a clicker (5 Hz)signal. A downward press of the lever initiated presentationof the list items which could include 1, 2, 3, 4, 5, or 6items. Trials of different list lengths were intermixed ina pseudorandom order within the session. Each list itemwas presented for 1 s, with a 1-s delay between items.Following the last list item, there was a 1-s delay beforepresenting the probe item. The probe item was presentedon the lower screen and remained in view until themonkey made a choice response. Movements 6f the leverto the right (single-choice response) were correct if theprobe item matched any list item. Movements to the leftwere correct if it matched none of them. Correct responseswere rewarded with either a 5 cc squirt of orange juiceor a banana pellet, the type of reinforcer was determinedrandomly. Incorrect responses initiated a 10-s time-outwith the house light turned on. A 3-s intertrial intervalfollowed reinforcement for correct responses. The nexttrial began after the intertrial interval or the time-outperiod.

There were 216 unique trials used in this experimentconstructed from 864 uniquely different items (108 sametrials and 108 different trials). The Kodak Carouselprojectors that were used held only 140 items, so thetrials were distributed into six different pairs of Carouseltrays (one tray held the list items and the other held theprobe items). The monkeys generally performed in all216 trials daily. The items were selected from a 3,000-item color-slide pool; they were novel items and had notbeen previously seen by the monkeys. Care was taken initem selection and composition of the trials so that items

Page 9: Monkey Memory: Same I Different Concept Learning, Serial ...different by moving the response lever to the right or left, respectively. Correct responses were followed by a tone (500

MONKEY MEMORY 521

12 20 28 36 44 52 60

BLOCKS OF 72 TRIALS

66 76 64 92-"50

100

Figure 4. Individual acquisition functions for 3 monkeys on a serial-probe recognition task where the listlengths varied from one to six items.

were easily discriminable to the experimenters (and prob-ably to the monkeys as well).

Results

The monkeys' acquisition of the variable-list-length task is shown in Figure 4. Thethree monkeys acquired the variable-list-length SPR task at about the same rate. Theygradually acquired the task beginning at about60% correct and improved their performanceto about 85% correct. Strangely, none showedany acquisition differences among the differentlist lengths. Acquisition differences were testedover the first 21 sessions (blocked by 3 ses-sions), where differences in acquisition mightbe expected to be greatest; there were nodifferences, F(5, 36) = 0.8, p > .5 ns; F(5,36) = 3.0, p > .05 ns; F(5, 36) = 0.6, p > .5ns for Joe, Linus, and Max, respectively.Average performance over the first 21 sessionswas: 74.5%, 71.0%, 69.7%, 65.9%, 64.5%,68.8% correct for one- through six-item listlengths, respectively. At no other time duringtheir acquisition did the subjects show list-length performance differences either. Priorto this, they showed 85% transfer to one-itemlist lengths. Thus, we thought it likely thatthey might immediately perform well withone-item list lengths (equivalent to a same/different task), rapidly acquire the two-itemlist length trials, and more gradually acquirethe others, in order of list length. This wasnot so. There must have been some levelingeffect. All list-length performances began

somewhat above chance performance andremained at about the same level throughoutacquisition.

Discussion

It is not clear to us why the shorter listlengths were not more rapidly acquired thanthe longer ones. We had expected, from thesame/different transfer results, about 85%correct transfer to the one-item list lengths.Pigeons too had been disrupted in their SPRperformance when the list length varied(Wright et al., 1984). Pigeons performed muchbetter when list length was fixed within blocksof trials than when it varied. Pigeons andmonkeys generally are disrupted by the samesorts of things, only the disruption is usuallymore pronounced for pigeons than for mon-keys. Pigeons were trained on the same vari-able list-length task with the same items asthe monkeys. Pigeons, however, showed nosigns of acquiring the task. Figure 5 showsthe average monkey and average pigeon SPRperformance on the one- to six-item variable-,list-length task. Monkey acquisition, althoughslow, began well above pigeon performanceand showed a gradual increase. The pigeons'performance remained at chance levelthroughout the experiment. Pigeons appar-ently are more disrupted by changing listlength than are monkeys.

The mechanism of this disruption is notclear, but human subjects can have theirmemory processing disrupted when the post-

Page 10: Monkey Memory: Same I Different Concept Learning, Serial ...different by moving the response lever to the right or left, respectively. Correct responses were followed by a tone (500

522 A. WRIGHT, H. SANTIAGO, AND S. SANDS

100oUJa:acOo

iuoDCuia.

80

60

MONKEYS

PIGEONS

12 20 28 36 44 62 60

BLOCKS OF 72 TRIALS

68 76 84 92 100

Figure 5. Mean acquisition functions for 3 monkeys and 3 pigeons on a serial-probe-recognition taskwhere the list lengths varied from one to six items.

stimulus interval varies or when the exposureduration itself varies (Proctor, 1983). Proctorhypothesized that the human subjects wouldnot rehearse when they were uncertain aboutonset and offset of the stimulus items. Finally,human subjects do not typically reveal pri>-macy effects when the list length varies(Sternberg, 1966). Possibly this too is a dis-ruption of some memory process. Thus, thereare a number of variables that probablyinfluence SPR performance. We mentionedthree of them: list length, interstimulus inter-val, and item viewing time. All three arelikely to have different effects depending uponwhether they are blocked or variable. In thenext experiment we deal with a fourth vari-able—probe delay. This is the time intervalbetween the last list item and the probe (test)item. In this experiment we used a blockeddesign to investigate the effects of probedelay.

Experiment 3: Serial Position Curvesas a Function of Probe Delay

This experiment was conducted after themonkeys had acquired the SPR task withvariable list lengths of one to six items. Theytransferred well to fixed three-item list-lengthSPR task (average of the first five sessions forthe three monkeys was 80% correct), andlater to a four-item list-length SPR task (five-session average of 74% correct). Their main-tained good performance in the SPR taskallowed the retention interval to be increasedwithout the risk of a floor effect. The effectof a variable such as probe delay (retentioninterval) can be measured only if performanceis high enough to leave room for decrements

to be visible. Our purpose was to obtainserial-position functions with four-item listlengths arid see how the form of the serial-position function changed with probe delay.We were encouraged in this exploration byour previous results showing a U-shapedmonkey serial-position function (Sands &Wright, 1980a, 1980b), indications in thehuman memory literature that the serial-position function results from dual processes,and the previous finding (Santiago & Wright,1984) that the pigeons' primacy and recencyeffects change in separate ways with probe-delay changes.

Method

Subjects

The subjects were 2 monkeys, Joe and Linus, from theprevious experiment. They were tested in this experimentimmediately after they had acquired the SPR task withthe variable list lengths (1-6 items) and had shown goodperformance with fixed list lengths of three and fouritems.

Apparatus and Procedure

The apparatus was the same as that used in theprevious experiments. The procedure was similar exceptthat the list length was fixed at four list items for eachtrial. The monkeys were trained and tested with two 20-trial sessions, each composed of 90 novel pictures whichthe monkeys had not previously seen. Each of six probedelays (0, 1, 2, 10, 20, and 30 s) was tested on four 20-trial sessions. The order of probe delay testing was: 2,10, 0, 20, 30, 2, 1, 10, 30, 20, 2, 30, 10, 2, 0, 1, 20, 0,1, 30, 10, 1, 20, and 0 s.

Results

Figure 6 shows the results for the 2 mon-keys tested at different probe delays. The

Page 11: Monkey Memory: Same I Different Concept Learning, Serial ...different by moving the response lever to the right or left, respectively. Correct responses were followed by a tone (500

MONKEY MEMORY

• -SAME A-DIFFERENT

523

HI

100

80

E 60

OU 40

zLLJO

£ 80

0.60

40

JOE

1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4

SERIAL POSITION0 SEC 1 SEC 2 SEC 10 SEC 20 SEC 30 SEC

PROBE DELAY

Figure 6. Serial position functions for six different probe delays between the last list item (Serial Position4) and the probe item for 2 monkeys. (Performance on different trials where the probe item did not matchany list item is shown by triangles.)

serial-position functions are from perfor-mance on same trials. Triangles are fromperformance on different trials (where theprobe item did not match any of thelist items). The 2 monkeys revealed thesame trends in their serial-position functionchanges: The shortest probe delay (0 s) showedperformance generally increasing from thefirst to the fourth item, revealing a recencyeffect and no primacy effect. The next threeprobe delays (1,2, and 10 s) show prominentprimacy effects as well as recency effects,revealing a U-shaped serial-position function.The 20-s probe delay shows the beginningsof a change from a U shape to monotonicdecreasing shape. The longest probe delay(30 s) showed a monotonic decreasing serial-position function, revealing a prominent pri-macy effect and no recency effect.

These serial-position function changes weretested by a series of weighted contrast tests(Keppel, 1973, p. 94) and are shown in Table3. At 0-s probe delay, Serial Position (SP) 4was significantly greater than SP1 for bothmonkeys. In addition, Joe's SP1 was signifi-cantly less than its other three SPs, andLinus's SP4 was significantly greater than itsother SPs.1 At 1-s probe delay, SP2 was sig-nificantly less than other SPs for both mon-keys, demonstrating the U shape of the SP

function. Linus also showed this same resultat 2- and 10-s probe delays. Joe's troughmoved from SP2 to SP3 at 2- and 10-s probedelays, and SP3 was significantly less thanother SPs. At 20 s the SP functions are tran-sitional. For both monkeys at 20 s, SP4 wassignificantly less than SP1 demonstrating thedownward trend, but they both show inver-sions at SP3, which are probably the remainsof the U shape. At 30-s probe delay, SP4 wassignificantly less than all other serial positions,demonstrating the decreasing trend of thefunctions.

In addition, the absolute levels of primacyand recency effects were compared. Changesin SP1 were tested with a series of contrasttests (Keppel, 1973). SP1 at 0-s probe delaywas significantly less (p < .05) than SP1 atall other probe delays except 10 s. Among allother probe delays there were no significantdifferences for SP1. These tests support theSP1 trends shown in Figure 6. There was nohint of a primacy effect at 0-s probe delay.There was a prominent primacy effect at 1-sprobe delay, and this primacy effect remainedfairly stable out to the longest probe delay of30 s. Contrast tests were also performed onSP4. There were no significant differencesamong the SP4 comparisons for 0-, 1-, 2-,and 10-s probe delays. However, SP4s at 20-

Page 12: Monkey Memory: Same I Different Concept Learning, Serial ...different by moving the response lever to the right or left, respectively. Correct responses were followed by a tone (500

524 A. WRIGHT, H. SANTIAGO, AND S. SANDS

Table 3F Values for Contrast Tests Between All Pairs of the Four Serial-Position Performances

Probe delay/monkey

O sJoeLinus

1 sJoeLinus

2sJoeLinus

10sJoeLinus

20sJoeLinus

30sJoeLinus

Serial position"

4 vs. 3

0.016.2**

2.128.8**

248.1**1.4

115.6**2.6

19,4**25.8**

48.0**19.9**

4 vs. 2

5.244.0**

79.5**141.7**

0.036.0**

3.622.0**

3.80.0

200.9**7.2*

4 vs. 1

130.1**20.5**

0.90.2

0.90.0

0.01.4

64.5**46.1**

200.9**48.3**

3 vs. 2

5.26.8

55.6**42.7**

248.1**23.0**

78.3**9.6*

6.025.8**

52.5**3.2

3 vs. 1

130.1**0.3

0.324.2**

279.1**1.4

115.6**0.2

64.5**2.9

52.5**6.2**

2 vs. 1

83.2**4.5

63.5**131.2**

0.936.0**

3.612.4*

36.8**46.1**

0.018.3**

Overall serial-position effect

59.0**15.4**

33.6**61.5**

129.5**16.3**

52.8**8.0*

23.9**24.5**

92.5**17.2**

Note. Degrees of freedom for pairwise serial position tests and overall serial-position effect are 1,9 and 3, 9, respectively.* 4 = last list item.* p< .05. ** p < .01.

and 30-s probe delays were significantly dif-ferent (p<.01) from SP4s at the shorterprobe delays. Performances for SP4 at 20 and30 s did not differ significantly from eachother. These validate the trends in the recencyeffects (SP4) shown in Figure 6. The recencyeffect was high and stable for 0-, 1-, 2-, and10-s probe delays, but dissipated at 20- and30-s probe delays. Finally, the absolute levelsof the primacy and recency effects were notdifferent. The mean SP1 performances forboth monkeys at 1,2, 10, 20, and 30 s werepooled, because these values were shown tobe similar. They were compared with pooledSP4 values for 0, 1,2, and 10-s probe delaysbecause these values also had been shown tobe similar. There was no significant difference,F(l, 16) = 0.5, p > .50. This means that theprimacy effect rose to the same level as therecency effect. In both cases the asymptoticlevel was very high: a mean of 96% for theprimacy effect and 98% for the recency effect.Likewise, the low performance on SP1 wasnot significantly different from the low per-formance on SP4. Pooled SP4 mean perfor-mance at 0-s probe delay was compared with

that for SP4 at 20 and 30 s, F(\, 4) = 2.1,p > .20. The primacy and recency effectscovered a very broad range because they hadlow mean values of 70% and 60%, respectively.

Discussion

Only recently have primacy effects beenshown for animals (Buchanan et al., 1981;Kesner & Novak, 1982; Roberts & Kraemer,1981; Sands & Wright, 1980a, 1980b; San-tiago & Wright, 1984). Figure 5 shows thatonly intermediate probe delays will producethe characteristic U-shaped function, revealingboth primacy and recency effects. If the probedelay is too short, a primacy effect will notbe found. If it is too long, a recency effectwill not be found. Although pigeons show thesame changes in their serial-position functionsas do monkeys, a long probe delay (10 s) forpigeons is only a moderate probe delay formonkeys. Thus, for each individual speciesthe particular probe delay to reveal bothprimacy and recency effects may have to beempirically determined.

Probe delay had a powerful effect on the

Page 13: Monkey Memory: Same I Different Concept Learning, Serial ...different by moving the response lever to the right or left, respectively. Correct responses were followed by a tone (500

MONKEY MEMORY 525

serial-position function. It was somewhat sur-prising to find that initial increases in probedelay produced increases, rather than de-creases, in performance; the primacy effect,totally absent at 0-s delay, emerged with theinitial increases in probe delay. As probedelay was further increased, the recency effectbegan to dissipate, eventually leaving onlythe primacy effect intact. These dynamicserial-position function changes argue for dualmemory processes. They are, at the sametime, evidence against single-process expla-nations of the U-shaped serial-position func-tion such as a network of associations andinhibitions (Hull, 1935; Lepley, 1934), orend-point distinctiveness (Bower, 1971; Mur-dock, 1960). The dynamic serial-positionfunction changes show that two processes areoperating quite independent of one another.The primacy effect emerges before there isany apparent change in the recency effect.The primacy effect is quite stable thereafter,and the recency effect begins to dissipate withfurther delay increases.

Dual memory processes have been pro-posed previously. The primacy portion andthe medial portion of the serial-position func-tion have been argued to represent a long-term memory store (LTS), whereas the re-cency portion has been argued to representa short-term memory store (STS; see, forexample, Glanzer, 1972). The recency portionmay actually be a combination of LTS andSTS. Techniques have been developed to par-cel out the separate contributions of LTS andSTS to the various portions of the serial-position function. This separation is madepossible by comparing and analyzing resultsfrom immediate recall and delayed (withdistractor task) recall. Delayed recall is freeof any STS contribution (so it is argued) andrepresents pure LTS memory. The STS con-tribution is calculated by subtracting the de-layed-recall, serial-position function (pureLTS) from the immediate-recall function(combination of LTS and STS). This subtrac-tion is possible as long as delayed recall isequal to or less than immediate recall at allpositions of the serial-position function. Oth-erwise, the result will be negative at thoseserial positions, an impossibility from thestandpoint of memory. However, this is pre-cisely the finding that we have shown in this

article for the initial increases in probe delay.Immediate recognition of the first list itemsis poor. After about 1-s delay, performancehas substantially risen revealing a primacyeffect. This emergence of a primacy effectwith delay is a problem for all theories thatcalculate STS in this manner (e.g., Glanzer,1972; Raymond, 1969; Waugh & Norman,1965).

The literature on human memory does notafford many comparisons to the probe-delayexperiments in this article. Human retentioninterval (the analogous variable to probe de-lay) studies have largely been confined topaired-associate procedures or free-recallprocedures. Paired-associate procedures donot yield serial-position functions. Recallprocedures do not provide adequate controlof the effective delay time; there is a longerdelay for items that are recalled last than forthose that are recalled first. One study (Jahnke& Erlick, 1968), however, did use an SPRprocedure (delay is precisely controlled inSPR), and the probe delay was varied. Theproblem with this study was that delay wasnot varied over a sufficiently wide range.Only three intermediate delays were used: 4,8, and 12 s. All the resulting serial-positionfunctions were U shaped. Tests were notperformed with the delay values (0 s and >30s), which might have produced the moreinteresting shaped serial-position functions,monotonic increasing, and monotonic de-creasing.

Several studies have used free-recall pro-cedures and have shown dissipation of therecency effect with probe delay (Gardiner,Thompson, & Maskarinec, 1974; Glanzer &Cunitz, 1966; Postman & Phillips, 1965;Roediger & Crowder, 1975).2 These studiesused delays of up to 30 s. Glanzer and Cunitz(1966) and Postman and Phillips (1965) usedseveral delay values, whereas the other re-searchers used only 0-s and 30-s delay values.The results were consistent from these fourstudies. As delay was increased, the recency

2 Related to this is the negative recency effect (e.g.,Craik, 1970), where there is an immediate recall followedlater by another recall. The typical finding is that therecency effect present in the immediate recall is absentin the later recall, and terminal item performance is lessthan medial item performance.

Page 14: Monkey Memory: Same I Different Concept Learning, Serial ...different by moving the response lever to the right or left, respectively. Correct responses were followed by a tone (500

526 A. WRIGHT, H. SANTIAGO, AND S. SANDS

effect dissipated. The degree to which it dis-sipated varied somewhat among the studies.Gardiner et al. (1974) and Roediger andCrowder (1975) showed an actual negative-recency effect with auditory presentation ofthe items. That is, performance continued tofall throughout the list with the terminal-item performance being less than the medial-item performance. Gardiner et al. (1974) didnot show a negative-recency effect with visualpresentations. Terminal-item performance didnot fall below medial-item performance. Per-haps they would have found a negative-re-cency effect with visual items had they usedeven longer probe delays. We showed a neg-ative-recency effect for monkeys with visualitems at 30-s delay, whereas only 10 s wasrequired to reveal one for pigeons. Humansmight require a delay considerably longerthan monkeys to reveal their negative-recencyeffect with visual stimuli. The probe-delayexperiments with pigeons and monkeys dem-onstrated that the negative-recency effect wasnot a separate phenomenon; it formed oneend point on a continuum of effects thatprobe delay had on the serial-position func-tion.

None of the four studies of humans foundmonotonic increasing functions at 0-s delayas we have shown for pigeons and monkeys.The reason for their failure to find monotonicincreasing functions at 0-s probe delay maybe due to the nature of the free-recall pro-cedure itself. The free-recall procedure allowsa block of time (e.g., 60 s) for the subject towrite down all of the list items that areremembered. Items recalled at the end of therecall period are delayed more than thoserecalled at the beginning of the recall period;"the S's own responses act as a delay task"(Glanzer, 1972, p. 163). The probe-delayresults with pigeons and monkeys show thatsuppression of the primacy effect lasts onlyabout 1 s. Even with humans the transitionto a U-shaped serial-position function shouldtake place within the first few seconds. Thusrthe finding of a monotpnic-increasing functioncould have easily been lost during the com-paratively long recall period. If the first listitems are not immediately recallable, perhapsthey will be after some of the later items arerecalled. Analysis of output gives support tothis argument: "The terminal items of a

series of unrelated words have the highestprobability of recall and also are usuallyrecalled first" (Postman & Phillips, 1965,p. 133).

One recall study used a procedure thatcontrolled the time from the end of the listto the item recalled (Ellis & Hope, 1968,Experiment IV). They presented a singleprobe (former list item) following the list.The subjects had to identify the list position(by pressing the proper button) of the item.Their performance at a 0-s probe delayshowed no primacy effect and a strong recencyeffect, a result similar to the ones we showedfor pigeons and monkeys. At 10-s probedelay, the serial-position function revealedprimacy as well as recency effects, the familiarU shape. The recency effect had started todissipate slightly relative to the 0-s delaycondition. Thus, this study demonstratessimilar trends to those we showed for pigeonsand monkeys. It indicates that if the intervalfrom end-of-list to test is precisely controlled,then humans too will show an initial mono-tonic increasing function with a recency andno primacy effect, transition to a U-shapedfunction with longer delays, and finally thebeginning of dissipation of the recency effectwith further delay increases. Discussing thetransition from the monotonic increasingfunction to the U-shaped one, Ellis and'Hopesaid that the emergence of the primacy effectresulted "perhaps by providing opportunityto rehearse during the delay interval" (p.617). There seems to be an inconsistency insuch an argument. If the subject has theability to retrieve the first items in order torehearse them during the delay interval, thenwhy can this same retrieval not serve as thebasis for the correct response when testedimmediately? This finding of emergence ofthe primacy effect with increasing probe delayis a problem for all theories that use rehearsalto explain the primacy effect (Atkinson &Shiffrin, 1968; Raffel, 1936; Waugh & Nor-man, 1965; Welch & Burnett, 1924).

Inadequacy of rehearsal as a theory toexplain emergence of the primacy effect leadsto the question of what is an adequate theoryto explain it. There is one theory that seemsadequate in this regard. It is interferencetheory, specifically the mechanism of release-from-retroactive-interference, also known as

Page 15: Monkey Memory: Same I Different Concept Learning, Serial ...different by moving the response lever to the right or left, respectively. Correct responses were followed by a tone (500

MONKEY MEMORY 527

spontaneous recovery. Retroactive interference(RI) is interference from memory items,which follow the items being tested. InitiallyRI Js large, but it dissipates comparativelyrapidly with time (Abra, 1969; Briggs, 1954;Ceraso & Henderson, 1966; Forrester, 1970;Kamman & Melton, 1967; Koppenaal, 1963;Martin & MacKay, 1970; Melton & Irwin,1940; Postman et al., 1968; Postman, Stark,& Henschel, 1969; Shulman & Martin, 1970;Silverstein, 1967; Slamecka, 1966; Under-wood, 1948a, 1948b). Large initial RI ac-counts for the lack of a primacy effect whenthe initial items are tested immediately. Thelater list items interfere (retroactively) withmemory of the first ones. Rapid dissipationof RI allows their memory to be recovered,resulting in emergence of the primacy effect.Turning now to dissipation of the recencyeffect with probe delay, this too can be ac-counted for by interference theory. Proactiveinterference (PI) is interference from memoryitems that precede the items being tested."PI is virtually absent at first but increasessteadily with time" (Postman & Phillips, 1965,p. 123). A number of studies have shownthat PI is absent initially and slowly grows,and its growth is slower than RI dissipation(Keppel & Underwood, 1962, Postman, 1962;Postman et al., 1968). Initial absence of PIand its slow growth accounts for the recencyeffect being present well after the primacyeffect has emerged. The eventual increase inPI accounts for the eventual dissipation ofthe recency effect.

These different RI and PI effects and theirseparate time courses have been worked outwith paired-associate procedures. It is unfor-tunate that some interference theorists didnot use an SPR procedure and investigatethe effects on the serial-position function.The SPR procedure seems ideally suited andvery sensitive to the different interferenceeffects with which the previously mentionedtheorists have been concerned. The timecourses, however, are somewhat different,being considerably faster in the SPR proce-dure than in the paired-associate procedure.In the SPR procedure, RI dissipated, revealingthe primacy effect in about 1 s. By contrast,in the paired-associate procedure, first-listrecall (index of release-from-RI) increasedover 30 min in some studies (Forrester, 1970;

Kamman & Melton, 1967; Martin & MacKay,1970; Postman et al., 1968; Postman et al.,1969; Shulman & Martin, 1970) and as muchas 24 hr or more in others (Abra, 1969;Briggs, 1954; Ceraso & Henderson, 1966;Silverstein, 1967; Underwood, 1948a, 1948b).The possibility that a neurophysiological con-solidation mechanism (cf., Hebb, 1949) isrelevant to these results has been previouslyexcluded because of the slow time coursedemonstrated by these paired-associate pro-cedures and because an influential study(McGeoch, 1942) showed that the same RIeffect could be obtained when the second list(interfering list) was presented 6 weeks afterthe first one. Any first list consolidation wouldhave certainly run its course long before 6weeks! The much shorter time course dem-onstrated here with the SPR procedure onceagain opens up the possibility that consoli-dation does play a role; the time values areof a proper order of magnitude for neuro-physiological consolidation.

Concluding RemarkThese changes in the serial-position func-

tion with probe delay for both monkeys andpigeons demonstrate that these animals haveprimacy and recency effects just as humansdo. Their similar dynamic changes in serial-position functions indicate that underlyingmemory processes are similar if not identicalin these two species. This is an importantstep in establishing continuity of species interms of their cognitive ability, which mayeventually lead to establishing that animalsprocess information and think in ways basi-cally similar to humans.

References

Abra, J. C. (1969). List-1 unlearning and recovery as afunction of the point of interpolated learning. Journalof Verbal Learning and Verbal Behavior, 8, 494-500.

Atkinson, R. C., & Shiffrin, R. M. (1968). Humanmemory: A proposed system and its control processes.In K. W. Spence & J. T. Spence (Eds.), The psychologyof learning and motivation (Vol. 2, pp. 89-105). NewYork: Academic Press.

Bower, G. H. (1971). Adaptation-level coding of stimuliand serial position effects. In M. H. Appley (Ed.),Adaptation-level theory (pp. 175-201). New York: Ac-ademic Press.

Briggs, G. E. (1954). Acquisition, extinction, and recoveryfunctions in retroactive inhibition. Journal of Experi-mental Psychology, 47, 285-293.

Page 16: Monkey Memory: Same I Different Concept Learning, Serial ...different by moving the response lever to the right or left, respectively. Correct responses were followed by a tone (500

528 A. WRIGHT, H. SANTIAGO, AND S. SANDS

Buchanan, J. P., Gill, T. V., Braggio, J. T. (1981). Serialposition and clustering effects in a chimpanzee's "freerecall." Memory and Cognition, 9, 651-660.

Ceraso, J., & Henderson, A. (1966). Unavailability andassociative loss in RI and PI:, Second try. Journal ofExperimental Psychology, 72, 314-316.

Cox, J. K., & D'Amato, M. R. (1982). Matching tocompound samples by monkeys (cebus apella); Sharedattention or generalization decrement? Journal of Ex-perimental Psychology: Animal Behavior Processes, 8,209-225.

Craik, F. I. M. (1970). The fate of primary items in freerecall. Journal of Verbal Learning and Verbal Behavior,9, 143-148.

D'Amato, M. R. (1973). Delayed matching and short-term memory in monkeys, In G. H. Bower (Ed.), Thepsychology of learning and motivation (Vol. 7, pp.227-269). New York: Academic Press.

Davis, R. T., & Fitts, S. S. (1976). Memory and codingprocessing in discrimination learning. In D. L. Medin,W. A. Roberts, & R. T. Davis (Eds.), Processes ofanimal memory (pp. 167-180). Hillsdale, NJ: Erlbaum.

Devine, J. V., & Jones, L. C. (1975). Matching-to-successive samples: A multiples-unit memory task withrhesus monkeys. Behavior Research Methods and In-strumentation, 1, 438-440.

Ellis, N. R., & Hope, R. (1968). Memory processes andthe serial position curve. Journal of ExperimentalPsychology, 77, 613-619.

Forrester, W. E. (1970). Retroactive inhibition and spon-taneous recovery in the A-B, D-C paradigm. Journalof Verbal Learning and Verbal Behavior, 9, 525-528.

Gaffan, D. (1977). Recognition memory after short re-tention intervals in fornix-transected monkeys. Quar-terly Journal of Experimental Psychology, 29, 577-588.

Gardiner, J. M., Thompson, C. P., & Maskarinec, A. S.(1974). Negative recency in initial free recall. Journalof Experimental Psychology, 103, 71-78.

Glanzer, M. (1972). Storage mechanisms in recall. InG. H. Bower (Ed.), The psychology of learning andmotivation (Vol. 5, pp. 129-193). New York: AcademicPress.

Glanzer, M., & Cunitz, A. R. (1966). Accuracy of per-ceptual recall: An analysis of organization. Journal ofVerbal Learning and Verbal Behavior, 5, 351-360.

Hays, W. L. (1963). Statistics for psychologists. NewYork: Holt, Rinehart & Winston.

Hebb, D. O. (1949). The organization of behavior: Aneuropsychological theory. New York: Wiley.

Hull, C. L. (1935). The conflicting psychologies of learn-ing—A way out. Psychological Review, 42, 491-516.

Jahnke, J. C., & Erlick, D. E. (1968). Delayed recognitionand the serial organization of short-term memory.Journal of Experimental Psychology, 77, 641-647.

Kamman, R., & Melton, A, W. (1967). Absolute recovery,of first-list responses from unlearning during 26 minutesfilled with an easy or difficult information processingtask. Proceedings of 75th Annual Convention of theAmerican Psychological Association, 2, 63-64.

Keppel, G. (1973). Design and analysis, EnglewoodCliffs, NJ: Prentice-Hall.

Keppel, G., & Underwood, B. J. (1962). Proactive inhi-bition in short-term retention of single items. Journalof Verbal Learning and Verbal Behavior, 1, 153-161.

Kesner, R. P., & Novak, J. M. (1982). Serial positioncurve in rats: Role of the dorsal hippocampus. Science,218, 173-175.

King, J. E. (1973). Learning and generalization of a two-dimensional sameness-difference concept by chimpan-zees and orangutans. Journal of Comparative andPhysiological Psychology, 84, 140-148.

Koppenaal, R. J. (1963). Time changes in the strengthsof A-B, A-C lists: Spontaneous recovery. 'Journal ofVerbal Learning and Verbal Behavior, 2, 310-319.

Lepley, W. M. (1934). Serial reactions considered asconditioned reactions. Psychological Monographs, 46,205.

Martin, E., & MacKay, S. A. (1970). A test of the list-differentiation hypothesis. American Journal of Psy-chology, 83, 311-321.

McGeoch, J. A. (1942). The psychology of human learning,New \brk: Longmans, Green.

Medin, D. L., Reynolds, T. J., & Parkinson, J. K. (1980).Stimulus similarity and retroactive interference andfacilitation in monkeys' short-term memory. Journalof Experimental Psychology: Animal Behavior Pro-cesses, 6, 112-125.

Melton, A. W, & Irwin, J. (1940). The influence ofdegree of interpolated learning on retroactive inhibitionand the overt transfer of specific responses. AmericanJournal of Psychology, 53, 173-203.

Murdock, B. B., Jr. (1960). The distinctiveness of stimuli.Psychological Review, 67, 16-31.

Overman, W. H., & Doty, R. W. (1980). Prolonged visualmemory in macaques and man. Neuroscience, 5, 1825-1831.

Postman, L. (1962). Transfer of training as a function ofexperimental paradigm and degree of first-list learning.Journal of Verbal Learning and Verbal Behavior, 1,109-118.

Postman, L., & Phillips, L. (1965). Short-term temporalchanges in free-recall. Quarterly .Journal of Experi-mental Psychology, 17, 132-138.

Postman, L., Stark, K., & Fraser, J. (1968). Temporalchanges in interference. Journal of Verbal Learningand Verbal Behavior, 7t 672-694.

Postman, L., Stark, K., & Henschel, D. (1969). Conditionsof recovery after unlearning. Journal of ExperimentalPsychology, 82 (1, Pt. 2).

Postman, L., & Underwood, B. J. (1973). Critical issuesin interference theory. Memory and Cognition, 1, 19-40.

Premack, D. (1983). Animal cognition. Annual Reviewof Psychology, 34, 351-362.

Proctor, R. W. (1983). Recognition memory for picturesas a function of poststimulus interval: An empiricalclarification of existing literature. Journal of Experi-mental Psychology: Learning, Memory, and Cognition,9, 256-262.

Raffel, W. A. (1936). Two determinants of the effects ofprimacy. American Journal of Psychology, 48, 654-657.

Raymond, B. (1969). Short-term and long-term storagein free recall. Journal of Verbal Learning and VerbalBehavior, 8, 567-574.

Roberts, W. A., & Kraemer, P. J. (1981). Recognitionmemory for lists of visual stimuli in monkeys andhumans. Animal Learning and Behavior, 9, 587-594.

Robinson, J. S. (1955). The sameness-difference discrim-

Page 17: Monkey Memory: Same I Different Concept Learning, Serial ...different by moving the response lever to the right or left, respectively. Correct responses were followed by a tone (500

MONKEY MEMORY 529

ination problem in chimpanzees. Journal of Compar-ative and Physiological Psychology, 48, 195-213.

Robinson, J. S. (1960). The conceptual basis of thechimpanzee's performance on the sameness-differencediscrimination problem. Journal of Comparative andPhysiological Psychology, 53, 368-370.

Roediger, H. L., Ill, and Crowder, R. G. (1975). Spacingof lists in free recall. Journal of Verbal Learning andVerbal Behavior, 14, 590-602.

Sands, S. F., & Wright, A. A. (1980a). Primate memory:Retention of serial list items by a rhesus monkey.Science, 209, 938-939.

Sands, S. E, & Wright, A. A. (1980b). Serial proberecognition performance by a rhesus monkey and ahuman with 10- and 20-item lists. Journal of Experi-mental Psychology: Animal Behavior Processes, 6,386-396.

Sands, S. E, & Wright, A. A. (1982). Human and monkeypictorial memory scanning. Science, 216, 1333-1334.

Santiago, H. C., & Wright, A. A. (1984). Pigeon memory:Same/different concept learning, serial probe recogni-tion acquisition, and probe delay effects in the serialposition function. Journal of Experimental Psychology:Animal Behavior Processes, 10, 498-512.

Shulman, H. G., & Martin, E. (1970). Effects of response-set similarity on unlearning and spontaneous recovery.Journal of Experimental Psychology, 86, 230-235.

Silverstein, A. (1967). Unlearning, spontaneous recovery,

and the partial-reinforcement effect in paired-associatelearning. Journal of Experimental Psychology, 73, 15-21.

Slamecka, N. J. (1966). A search for spontaneous recoveryof verbal associations. Journal of Verbal Learning andVerbal Behavior, 5, 205-207.

Sternberg, S. (1966). High-speed scanning in humanmemory. Science, 153, 652-654.

Underwood, B. J. (1948a). Retroactive and proactiveinhibition after five and forty-eight hours. Journal ofExperimental Psychology, 38,29-38.

Underwood, B. J. (1948b). "Spontaneous recovery" ofverbal associations. Journal of Experimental Psychology,38, 429-439.

Waugh, N. C.,. & Norman, D. A. (1965). Primacymemory. Psychological Review, 72, 89-104.

Welch, G. B., & Burnett, C. T. (1924). Is primacy afactor in association-formation? American Journal ofPsychology, 35, 396-401.

Wright, A. A., Santiago, H. C., Urcuioli, P. J., & Sands,S. E (1984). Monkey and pigeon acquisition of same/different concept using pictorial stimuli. In M. L.Commons, R. J. Herrnstein, & A. R. Wagner (Eds.),Quantitative analyses of behavior (Vol. 4, pp. 295-317). Cambridge, MA: Ballinger.

Received October 21, 1983Revision received May 21, 1984 •

Page 18: Monkey Memory: Same I Different Concept Learning, Serial ...different by moving the response lever to the right or left, respectively. Correct responses were followed by a tone (500

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.