VOL. 1972 Psychological Bulletin - Northwestern Universitygroups.psych.northwestern.edu/medin/documents/... · VOL. 77, No. 5 MAY 1972 Psychological Bulletin ... consin General Test

VOL. 77, No. 5 MAY 1972

Psycho log ica l B u l l e t i n

ROLE OF REINFORCEMENT IN DISCRIMINATIONLEARNING SET IN MONKEYS1

DOUGLAS L. MEDIN2

Rockefeller University

Following a review of empirical research on the role of reinforcement inlearning-set formation, the major theoretical explanations of learning-setformation in monkeys are analyzed. Studies showing that a reward can func-tion to decrease as well as increase the probability of choosing an object castdoubt upon theories based on an automatic strengthening function of reward.Hypothesis or strategy selection theories avoid this problem by assuminghypotheses, rather than responses, are subject to reinforcement principles, buthypothesis theories are at best incomplete in their treatment of retention. Atheory which assumes that learning-set formation results from between-prob-lem stimulus generalization of feedback from expected rewards is consistentboth with retention studies and with experiments on the function of rewardin learning set, suggesting that learning-set formation need not be considereda complex abstractive process.

When monkeys are given a series of two-choice object discriminations, their perform-ance on new problems gets better and better.They start out making 50% errors on thesecond trial of new problems, but after beingtested on a few hundred problems make only10% or fewer errors on Trial 2 of new prob-lems. The monkeys, so to speak, "learn howto learn." This improvement is not attributa-ble to the problems having a common solution,since attributes correlated with reward in oneproblem may not be correlated with rewardon a subsequent problem. Nor is this im-provement simply a matter of the monkeysadjusting to the experimental situation—sub-jects typically receive extensive pretrainingdisplacing objects and picking up rewardsbefore learning-set (LS) training begins.Given that these uninteresting explanations ofLS can be rejected, a major theoretical prob-

research was supported by Grants GM16735and MH16100 from the United States Public HealthService. W. K. Estes and D. Robbins provided valu-able comments on earlier drafts of this paper.

2 Requests for reprints should be sent to DouglasL. Medin, Rockefeller University, New York,New York 10021.

lem is to specify the source of this improve-ment in learning ability.

This study examines the major theoreticalexplanations of discrimination LS after a re-view of the principal findings on the role ofreward in LS in monkeys. The review borrowsheavily from and supplements the excellentearlier summaries by Harlow (1959), Reese(1964), and Miles (1965).

REVIEW OF LITERATURE

Basic Learning-Set Procedures and Data

In a typical LS experiment, subjects aregiven a series of simultaneous discriminationproblems having the following proceduralcharacteristics: (a) a pretraining involvingsingle objects being displaced for rewards; (b)a small, fixed number of trials on each prob-lem; (c) a different pair of stimuli for eachproblem; (d) a noncorrection procedure; (e)a reward for every correct response; (/) an in-tertrial interval of 10-20 seconds. The basicmeasure of LS performance is improvement inwithin-problem learning as a function of num-ber of problems given. Subjects cannot im-prove their performance on Trial 1 across

3051972 by the American Psychological Association, Inc.

306 DOUGLAS L. MEDIN

problems as it functions solely as an informa-tion trial,

Harlow (1949) performed the first experi-ment specifically designed to study between-problem transfer in monkeys.3 In this classicLS study, monkeys were tested in 344 inde-pendent discrimination problems in a Wis-consin General Test apparatus. On the initialproblems, learning improved gradually acrosstrials, and subjects averaged 52% correctresponses on Trial 2 of the first 8 problems.On the last set of discriminations, virtuallyall the learning took place between the firstand second trials of a problem—subjectsaveraged over 90% correct responses on Trial2 for the last 112 problems. This apparentchange from gradual or trial-and-error learn-ing to immediate solution of new problemsHarlow labeled "learning set formation."

Harlow (19SO) suggested that the responsepatterns of the monkeys might provide aninformative supplement to proportion correctas a dependent variable. He showed that er-rors made during LS were not random butrepresented systematic response tendenciesthat were inappropriate to the solution of theproblems. The four principal error factorsidentified were stimulus perseveration, differ-ential cue, position preference, and responseshift. A brief description of these errors fol-lows:

1. Stimulus perseveration. Stimulus per-severation errors consist of repetitive choicesof the incorrect stimulus object. These errorsare attributed to innate or learned preferencesfor, or avoidance of, particular stimulus ob-jects. Stimulus perseveration errors, as mea-sured by runs of consecutive errors, decreaseas the LS progresses.

2. Differential cue. If the correct object oc-cupies the left side of the apparatus on Trial1, there is some ambiguity as to whether re-sponses to the left or to the object itself arebeing reinforced. Differential cue errors aremeasured by the excess of errors on the trialwhen the correct object first shifts positionover the errors made on comparable trials

8 The term "monkeys" refers to rhesus monkeyssince approximately 90% of learning-set studies inprimates have used rhesus monkeys. Departuresfrom the use of rhesus are noted in the text.

when the correct object has not yet shiftedpositions. The number of errors on shift trialsdivided by the number of errors on corres-ponding nonshift trials is called the compara-ble-trial error ratio. When this ratio is equalto 1, presumably no differential cue errorsoccurred. Harlow (1950) also reported thatdifferential cue errors decreased across trials,and the comparable-trial error ratio ap-proached 1.0. However, Davis, McDowell,and Thorson (1953) found that while the ab-solute number of differential cue errors de-creased across trials, the comparable-trialerror ratio, if anything, increased.

3. Position preference. Position preferencesare consistent responses to the left or rightfoodwell, regardless of the position of thecorrect object. Abordo and Rumbaugh (1965)found that squirrel monkeys given trainingsuch that the correct object switched posi-tions after each correct response performedbetter under conventional LS procedures thansubjects given conventional training through-out. In general, however, position preferencesare only a minor source of errors in monkeys'LS formation.

4. Response shift. Response shift errors areerrors of responding to the incorrect objectafter the correct object has been displaced onprevious trials. This is normally measured asan excess of errors following a correct re-warded trial over the errors following an in-correct trial. Harlow suggested that this wasattributable to the monkeys' tendency toexplore the test situation, that is, the unchosenalternative.

Learning-Set Performance after Trial 1Reward and Nonreward

On logical grounds reward and nonrewardon Trial 1 of a problem should be equallyinformative—that is, if the object chosen wasrewarded, the subject should continue to re-spond to it; while if the object chosen wasnonrewarded, he should avoid that object onfuture trials and choose the other (correct)object. However, initial reward and nonre-ward do not have equivalent effects on per-formance within a problem nor is their effectconstant across problems.

The following generalization can be made:Early in training there are more correct re-

DISCRIMINATION LEARNING SET 307

sponses following a correct Trial 1 responsethan following a Trial 1 error; but later in LStraining the opposite holds—there are morecorrect responses following an error than fol-lowing a correct Trial 1 response (e.g., War-ren, 1966; see also Reese, 1964, for otherwork). An exception may be for difficult two-dimensional pattern discriminations, in whichstrong stimulus preferences appear to occur,in which correct first-trial responses aremore efficient in reducing errors throughoutLS training (e.g., Leary, 1958a; McConnell &Schuck, 1962).

If only a single object is presented on Trial1 and either rewarded or nonrewarded beforethe two-choice trials begin, a different pictureemerges. This procedure has the advantage ofbreaking the correlation between stimuluspreferences and Trial 1 choices. With single-stimulus training one finds that throughoutLS training, Trial 1 nonreward uniformly re-duces errors more than does Trial 1 reward(Boyer [1965],* 1966; Fletcher & Cross,1964; see Reese, 1964, for earlier- work;Schwartzbaum & Poulas, 1965).

Two possible confoundings present in stud-ies employing single-stimulus pretraining ob-viate drawing any firm conclusions as to theeffectiveness of reward and nonreward. First,there is the possibility that adapting animalsto the test situation by having them displacesingle objects for rewards prior to LS trainingteaches the monkeys not to attend to a singlypresented rewarded object. Harlow (1959)cited an unpublished study by Schrier andHarlow in which the difference in performancebetween initial reward and nonreward (favor-ing nonreward) was five times greater thanthe usual 10% difference, implying thatmonkeys learned virtually nothing followingan initial rewarded response. The only differ-ence in procedure was that the Trial 1 objectwas placed over the center foodwell whichotherwise was used solely for adapting ani-mals to displace objects. This study suggeststhat processes related to attention are prob-ably producing effects during single-stimulus

4W. N. Boyer. Discrimination performance inthree species of monkeys as a function of Trial 1reward contingency, intertrial interval, and prior testexperience. Unpublished doctoral dissertation, Okla-homa State University, 196S.

pretraining which make the relative effective-ness of reward and nonreward difficult toevaluate using this procedure. A secondshortcoming of the single-stimulus pretrain-ing paradigm is that responses to novel ob-jects have differential effects on performancefollowing rewarded and nonrewarded firsttrials. Any tendency to respond to novelobjects would increase the estimate of amountlearned from a nonrewarded first trial anddecrease the estimate of the amount learnedfrom a rewarded trial. As the following sec-tion of this study indicates, novelty andfamiliarity can be distinctive cues for mon-keys.

Effects of Novelty and Familiarity

Test-experienced monkeys appear to retaina tendency to approach novel stimuli, butunder conditions favorable to generalizationfrom the positive to the negative stimulus(e.g., the negative stimulus has not beenchosen and nonrewarded), the tendency toapproach familiar, negative stimuli becomesstronger than the tendency to approach novelobjects (Behar, 1962a, 1962b; Leary, 1956;see also Reese, 1964).

Monkeys can solve two-trial LS problemswhen the correct solution is to choose a newlypresented Trial 2 object and to avoid eitherthe correct or incorrect Trial 1 object (Brown,Overall, & Blodgett, 1959). Monkeys can alsolearn to approach or avoid specific recurringstimuli from previous problems (Gentry,Overall, & Brown, 1958) even on Trial 1 ofnew problems (Riopelle, Chronholm, & Addi-son, 1962).

Cross, Fletcher, and Harlow (1963) showedthat the cue of familiarity can be establishedfrom home-cage experience with objects. Forone group (positive) the home-cage objectswere designated as correct in the test situa-tion, for another group (negative) the ob-jects were always incorrect in the experiment,and for the third group (mixed group) thehome-cage stimuli were arbitrarily designatedas correct and incorrect. A control group re-ceived no home-cage experience with the ob-jects. The positive group and the negativegroup performed significantly better thanchance on Trial 1 of the problems involvingone home-cage stimulus, showing that they

308 DOUGLAS L. MED1N

were able to use familiarity as a cue. Thenegative group performed better than thepositive group, suggesting that subjects had atendency to approach the novel objects. Themixed group did not differ from control sub-jects on Trial 1 of new problems and wereactually inferior to them on Trials 2-12 (seealso Shell & Riopelle, 19S8).

In summary, novelty and familiarity canbe and are effective cues for monkeys and, assuch, complicate inferences concerning theeffectiveness of reward and nonreward. In thefollowing section, we see that experiencing adiversity of problems is important in LS for-mation.

Problem Diversity

Schusterman (1962, 1964) demonstratedthat chimpanzees given repeated reversal ofa single discrimination problem show immedi-ate efficient Trial 2 performance whenswitched to LS training. Schusterman's apesseemed to transfer a generalized strategy ofwin-stay, lose-shift with respect to objects.

Monkeys in contrast to apes may require amore diverse selection of stimuli to form anLS. Riopelle (19S3) reported that giving2,000 training trials on six object discrimina-tions followed by conventional LS training re-sulted in LS performance similar to that of agroup of monkeys that had not received priortraining. Treichler (1966) trained monkeysfor 840 trials on two discrimination problemsand found that an immediate win-stay, lose-shift strategy did not result. The modest trans-fer observed in his experiment may be at-tributable to specific interproblem stimulusgeneralization.

A study by Riopelle (1955a) suggests thatthe minimum number of different stimulineeded for LS formation may be quite small.Five naive monkeys who had learned 10 pre-liminary discriminations were trained to acriterion of S trials in a row correct (or amaximum of SO trials) on two-choice problemsconsisting of the various combinations of fourdifferent objects. They received 18 scrambledrepetitions of each of the 12 possible combi-nations of the four stimuli, or a total of 216problems, before being shifted to conventionalLS. Since all the combinations of the fourobjects were used, subjects received consider-

able reversal learning experience. The mon-keys showed immediate excellent LS perform-ance on the conventional six-trial problems.

On the other hand, Riopelle and Moon(1968) found that diversity of problems en-hances the LS formation. Two groups of fourstumptail monkeys were given 10 six-trialproblems each day of training consisting of 7familiar (repeated) and 3 new problems. Thepredictable reward group had the same objectsdesignated as correct for the repeated prob-lems, and the unpredictable reward group hadreward randomly assigned to objects for therepeated problems. For a control group all 10problems each day were new. After this dif-ferential pretraining, all three groups weregiven 50 six-trial problems with all new stim-uli. The predictable reward group quicklyreached perfect performance on the first trialof recurrent problems. On transfer to LS prob-lems that employed all new stimuli, the con-trol group performed significantly better thaneither of the other groups having received lessdiverse stimuli. This suggests experiencing avariety of stimuli is important for the estab-lishment of LS in monkeys (see also the sec-tion on transfer of learning sets).

Retention of Discriminations

Most theoretical descriptions of LS leadone to expect relatively poor retention of indi-vidual discriminations. According to Harlow:By the time the monkey has run 232 discriminationsand followed these by 112 discriminations and re-versals, he does not possess 344 or 456 specific hab-its, bonds, connections, or associations. We doubt ifour monkeys at this time could respond with muchmore than chance efficiency on the first trial of anyseries of previously learned problems. But themonkey does have a generalized ability to learn anydiscrimination problem or any discrimination re-versal problem with the greatest of ease [Harlow,1949, p. 63].

1. Intertrial interval. Relatively short-termretention of object discriminations can be as-sessed by studies varying intertrial interval.For intertrial intervals between 10 and 150seconds, performance typically declines al-though only modestly (Boyer, 1966; Fletcher& Cross, 1964; Harlow, 1959; Harlow & War-ren, 1952; Kruper, Patton, & Koskoff, 1961;Riopelle & Churukian, 1958). For example, inthe Riopelle and Churukian study there was


about a 5% decrease in performance as theintertrial interval increased from 10 to 60seconds.

2. Concurrent versus consecutive problemsequences. Typically, a series of problems ispresented to a monkey such that he receivesall the predetermined trials on one problembefore being given the next problem. In theconcurrent procedure the problems are pre-sented in a paired-associate fashion so thatTrial 1 of each problem appears before Trial2 on any problem is given. The concurrentprocedure may involve both interference fromother pairs in the list and increased forgettingfrom longer interpresentation intervals. Asone might expect, consecutive stimulus presen-tation results in better performance than con-current (Darby & Riopelle, 1955), and con-current list performance decreases with in-creases in list length at least for squirrel mon-keys (King & Goodman, 1966) and chimpan-zees (Hayes, Thompson, & Hayes, 1953). Theconcurrent presentation procedure on theother hand is not especially difficult for rhesusmonkeys (e.g., Leary, 1958a, 1962), and re-tention of a well-learned list is excellent at a24-hour retention interval (Sledjeski &French, 1968).

3. Long-term retention. Monkeys are ableto retain not only object discrimination LS(e.g., Braun, Patton, & Barnes, 1952) butalso specific object discriminations. Monkeyshave been found to retain discriminationsvirtually perfectly for at least 24 hours afteras few as six acquisition trials per problem(Riopelle & Moon, 1968; Riopelle et al.,1962). Mason, Blazek, and Harlow (1956)reported above-chance first-trial performanceby young monkeys on a series of 90 six-trialobject discriminations which were uninten-tionally repeated after a 1-month interval.Strong (1959) gave four naive monkeys ex-tensive training on 72 pairs of stimulus ob-jects and found extremely good performance(90% correct) at retention intervals between30 and 210 days. Zimmermann (1969) re-ported about a, 15% retention loss in monkeysfor each cycle of 100 discrimination problems,which were repeated every 20 days. By thesixth cycle, Trial 1 retention was 83% correct.A 3-month retention interval before the lastcycle resulted in a 20% memory loss.

In a series of studies of retention in LS-experienced monkeys, Bessemer6 found effi-cient performance on object discriminationsover a 24-hour retention interval but that,strikingly, the retention loss appeared to beconfined to those problems on which the firsttrial response during training had been in-correct and nonrewarded. Since subjects pre-sumably responded to their preferred stimu-lus on Trial 1, stimulus preferences can bepinpointed as playing a significant role inretention; when the preferred stimulus iscorrect, retention performance is better. Evenif the associative information were lost, stimu-lus preferences would insure good retentionperformance when the preferred stimulus iscorrect. In an experiment that eliminatedstimulus preferences by using single-stimuluspresentations, the differential retention effectdisappeared. Bessemer's finding that stimuluspreferences were strongly influencing retentionis of considerable theoretical importance sincethe manifestation of stimulus preferences inthese LS-experienced subjects would perhapsbe unexpected.

4. Transfer suppression and inter-probleminterference. Riopelle (1953) proposed that inthe course of LS, monkeys suppress transferof inappropriate problem solutions and treatsuccessive problems increasingly indepen-dently. Eventually, according to the transfersuppression theory, responses to a particularproblem do not transfer to succeeding prob-lems. To assess this idea, Riopelle tested naivemonkeys on six problems each day for 63 days,with the sixth problem being a reversal ofeither the first or fourth problem of that day.Initially the subjects made 80% errors onthe first trial of the reversed problems, buteventually their performance reached 60%errors on the first trial of reversals and wasnot different from that on nonreversal prob-lems on Trials 2-6.

However, these results do not imply thatLS is accompanied by increased forgettingabout specific object discriminations. AsStollnitz and Schrier (1968) pointed out, LSformation could not have been based on the

5 D. W. Bessemer. Retention of object discrimina-tions by learning set experienced monkeys. Unpub-lished doctoral dissertation, University of Wisconsin,1966.


development of stimulus independence, sincethe monkeys in Riopelle's study had largelyformed their LS before they started to sup-press transfer. Schrier and Stollnitz testedLS-experienced monkeys for transfer suppres-sion by replicating Riopelle's (19S3) pro-cedure. Their monkeys averaged only 17%correct responses on Trial 1 of the reversals.Transfer suppression had not resulted fromtheir prior LS training. A second experimentemployed LS-experienced stumptail monkeysusing Riopelle's procedure for a more ex-tensive testing period. Trial 1 performance onreversals was initially poor and, surprisingly,did not improve with practice. A replication ofSchrier (1969) employed both rhesus andstumptail monkeys, and again stumptailsfailed to improve on Trial 1 of reversed prob-lems over a 13-week practice period. Therhesus monkeys did improve (showed transfersuppression), and one subject performed sig-nificantly above chance on Trial 1 of thereversals. This above-chance performance iscontrary to the transfer suppression proposi-tion of problem independence. The pattern ofresults supports the contention that improve-ment in performance on Trial 1 of reversalsthat are interspersed during LS training rep-resents learning to reverse when familiarstimuli reappear rather than any forgettingphenomenon such as implied by transfer sup-pression theory. There is independent evi-dence that monkeys can learn to reverse aprevious choice after an arbitrary signal(Riopelle & Copeland, 1954).

Transfer of Learning Sets

Learning sets may facilitate or hinder per-formance on different types of stimuli orproblems. Wilson and Wilson (1962) reporteda small but reliable transfer between visualand tactual LS (see von Wright, 1970, for areview of cross-modal transfer). Harlow andWarren (1952) reported positive transfer be-tween pattern-discrimination LS and object-discrimination LS. Takemura (1960) testedtwo Japanese and one Formosan monkey onform-discrimination LS after they had re-ceived 168 color-discrimination LS problems.Performance on the form-discrimination prob-lems started at chance on Trial 2, but theform LS developed faster than the color-dis-

crimination LS. King (1966) found thatconcept-formation training (i.e., choose red,regardless of form) transferred to object-dis-crimination LS better than training involvingrewarded presentations of single objects. Rum-baugh, Sammons, Prim, and Phillips (1965)reported that giving squirrel monkeys pre-training for 3,000 trials with either a singlestimulus or with 500 different stimuli undera 50% reward schedule retarded LS forma-tion. Interposing a task where monkeys' per-formance does not rise much above chance(such as double alternation) also may disruptobject-discrimination LS (Rumbaugh & Prim,1964; Warren & Sinha, 1959).

Particular theoretical significance attachesto transfer between object-discrimination LSand reversal LS. In terms of abstract rules forproblem solution, both procedures have thesolution win-stay, lose-shift (with respect toobjects), and on some grounds one mightexpect perfect transfer between them. Chim-panzees show immediate direct transfer fromrepeated reversals involving just a few pairsof stimuli to object LS (Schusterman, 1962,1964). Repeated reversal learning also facili-tated the LS formation of Philippine andstumptail monkeys but to a much smallerextent (Schrier, 1966). Even repeated re-versal of a position discrimination may facili-tate object-discrimination LS in monkeys(Warren, 1966).

If the nonreward on the first trial of re-versal comes to act as a cue or sign for re-versal, then object-discrimination LS mightfacilitate reversal LS only to the extent thatefficient prereversal performance makes thenonreward on the first reversal trial a dis-tinctive cue. The results of a number of stud-ies with a variety of primate species suggestthat object-discrimination LS facilitates re-versal LS but that the transfer is by nomeans complete or perfect (Cross & Brown,1965; Harlow, 1944, 1950, 1959; Meyer,1951; Rumbaugh &Ensminger, 1964; Schrier,1966). In other words, transfer between ob-ject and reversal LS does not consist simply ofemploying a win-stay, lose-shift rule.

Information Value oj Rewards

1. Partial reinforcement. Behar (196ilb)gave naive and sophisticated monkeys a


choice between receiving two units of reward50% of the time or one reward 100% of thetime. Naive subjects showed no preferenceswhile the test-experienced monkeys chose theconsistently rewarded object suggesting thatnonreward had a greater effect on the sophisti-cated than on the naive monkeys.

Bowman (1963) tested LS-experiencedmonkeys under conditions where correct re-sponses were rewarded with various probabili-ties. The within-subject design used rewardpercentages of 100%, 75%, 50%, and 25%.Discrimination performance was efficient onlyfor reward percentages of 100% and 75%.For another condition, one of two colors wasrevealed when the objects were displaced.These color markers followed the positioningof the correct and incorrect objects and thuscould serve as secondary reinforcers. The par-ticular colors used changed from problem toproblem, so the color information could beuseful only after the first rewarded trial.Under this procedure, performance on the25% and 50% conditions improved signifi-cantly.

2. Ambiguous problems. Consider threestimuli, A, B, and C, that are presented inpairs consisting either of A and B, with Acorrect, or B and C, with B correct. To solvethe problem, B must be chosen when pairedwith C and must be avoided when paired withA. Performance on AB trials versus BC trialsmay serve to estimate the relative effects ofreward and nonreward. The data reveal thatBC trials typically result in fewer errors thanAB trials for objects, while the reverse ap-pears to hold true for two-dimensional stimuli(Bernstein, 1961; Fletcher, Grogg, & Garske,1968; Leary, 1958b; Thompson, 1954). Un-fortunately, inferences concerning the effec-tiveness of reward and nonreward dependcrucially upon assumptions concerning theabsence or presence of within-problem general-ization so that no strong conclusions havebeen drawn.

3. Information acquired in one trial. Stoll-nitz (1965) summarized a number of studiesshowing that when the stimuli are separatedfrom the site of the monkey's response byonly a minimal distance, discrimination per-formance is severely retarded. It would followfrom this that monkeys might learn only

about the object chosen in a discriminationproblem. Lockhart, Parks, and Davenport(1963) used test-experienced pigtail monkeysto directly test this proposition. For one groupof animals, the object not chosen on the firsttrial was replaced for Trials 2-6 of six-trialproblems, while for the other group the dis-placed object was replaced for Trials 2-6. Thereward conditions were dictated by and con-sistent with the first trial outcome. The sub-jects having the unchosen object replacedwere correct on 85% of their Trial 2 choices,while the subjects having their chosen objectreplaced performed at chance level, regardlessof whether their first trial was correct or in-correct.

Learning is not solely restricted to the ob-ject chosen. Brown and Carr (1958) placedthe objects for the next problem 6 inches be-hind the objects being used for the currentproblem. The object which was to be correcton the following problem was always behindthe object which was currently correct. Theirmonkeys showed significantly better-than-chance performance on the first trial of thenew problems, indicating that they hadlearned something about the incidental cues(see also Davis, 1965 and Zeis [1964],° forother incidental learning studies).

There is some evidence that with extendedpractice, animals can learn something aboutthe unchosen object. Bowman and Takemura(1966) also used the procedure of replacingeither the chosen or the unchosen Trial 1 ob-ject which could be rewarded or nonrewarded.For one group the recurring object was alwayscorrect, while for the other the recurring ob-ject was incorrect. When the chosen objectwas brought forward, both groups were ableto master the problems. The group havingthe unchosen object brought forth as correctlearned much more slowly but eventually didmaster the problem. One might argue thatsubjects were using familiarity as a cue, sincethe recurring object was always correct. Evenso, this would indicate that familiarity withobjects can be acquired without directly dis-placing, them.

6 S. M. E. Zeis. The use of peripheral cues inlearning set formation by rhesus monkeys. Unpub-lished doctoral dissertation, Catholic University ofAmerica, 1964.


The monkeys having the unchosen objectbrought forward as incorrect immediatelyreached a level of 70% correct on Trial 2 anddid not appear to improve their performance.Note that one need not invoke any associativeprocess to explain this result if stimulus pref-erences are assumed to occur. Variables lead-ing to not choosing the object on the firsttrial may also operate on Trial 2 to produce70% performance even in the absence oflearning.

Fletcher and his associates (Fletcher, 1966;Fletcher & Takemura, 1965; Fletcher et al.,1968) have employed a prompting procedurewhere a cue (i.e., the prompt) is attached toeither the correct (positive prompt) or in-correct (negative prompt) object. Groups aregiven pretraining with either a positive or anegative prompt to teach the monkeys thesignificance of the prompt. As a result, fewerrors occur on prompted trials during themain experiments. Learning is assessed bytrials in which the prompts are removed; and,in general, negative prompting results in bet-ter performance than positive prompting. Re-placing the correct object with a new objecton transfer tests after negative promptingtrials shows that monkeys apparently learnabout the incorrect object on the basis of itsprevious pairings with the negative prompt,even though they have never chosen it.

4. Control by -positive and negative cue.French, Birnbaum, Levine, and Pinsker(1965) tested monkeys under the followingexperimental procedures: (a) both the correctand incorrect objects were constant; (b) thecorrect object remained the same, but a newincorrect stimulus was constantly introduced;(c) the correct object remained the same,and new incorrect stimuli were introduced;(d) both new correct and incorrect objectswere introduced for each problem. The firstcondition yielded the best performance, thenext two conditions had intermediate effects,and performance was least efficient under thefourth condition. In a somewhat related ex-periment Ettlinger (1960) found for both tac-tile and visual discriminations that separat-ing the correct cue by 2 inches from the re-sponse site was more disruptive than separat-ing the incorrect object from the normal re-sponse sites.

By using an automated apparatus and two-dimensional stimuli, Sheridan, Horel, andMeyer (1962) were able to allow eithercorrect, incorrect, both, or neither stimuli topersist for certain time intervals after ananimal's choice response. Persistence of thecorrect stimulus aided performance more thanpersistence of the incorrect stimulus, both ofwhich were better than either of the other twoconditions.

5. Rewards as cues. Apart from anystrengthening function, rewards can serve ascues. A suitable paradigm to show this in-volves two-trial problems where the correctresponse on Trial 2 is determined by the Trial1 outcome. That is, one can give problemshaving the solution win-stay, win-shift, lose-stay, or lose-shift with respect to objects.The combination of the first and fourth typeof problem is simply the usual object-dis-crimination LS, but the second and thirdtype of problem require the subject to aban-don responses to the just-rewarded cue or topersist in responding to a cue despite nonre-ward, respectively. The four types of prob-lems are typically administered in a between-subjects design where only one of the fourproblems is presented to a given group ofmonkeys. For example, if a group is tested onwin-shift, its Trial 1 response will always berewarded, and on Trial 2 subjects must selectthe previously unchosen object. The resultsof a number of studies (Brown, McDowell, &Gaylord, 1965; McDowell & Brown, 1963a,1963b, 1963c; McDowell & Brown, 1965a,1965b, 1965c, 1965d, 1965e; McDowell, Gay-lord, & Brown, 1965a, 1965b; see Reese,1964) show that monkeys can learn all foursolution types. Win-stay is by far the mostdifficult of the four solutions to learn—a re-sult puzzling from the point of view thatrewards directly strengthen responses. In afew experiments (Brown et al., 1965; Mc-Dowell et al., 1965b), win-stay was notlearned at all. Perhaps responses to noveltyand overall reward probability are related tothis result, but there is at present no clearexplanation. Within-subjects comparisons ofcompatible problem types (e.g., win-stay com-bined with lose-stay) would control overallreward probabilities and might elucidate mat-ters. Finally, Behar (1961a) has shown that


monkeys can form an object-alternation (win-shift, lose-stay) LS.

Riopelle, in a series of studies (Riopelle,IQSSb; Riopelle & Francisco, 19SS; Riopelle,Francisco, & Ades, 1954), has used a marbleas a Trial 1 signal. Six-trial problems weregiven with no food, food, or a marble on Trial1 signaling the chosen or unchosen object tobe correct and rewarded with food in Trials2-6. Under these conditions, where win-shiftand lose-stay were not rewarded on Trials 3-6of a problem, win-stay and lose-shift wereperformed better than win-shift and lose-stay(see also Schwartzbaum & Poulas, 1965). Aninteresting feature of Riopelle's data is thatgroups trained on marble-stay, if anything,perform better than groups trained on food-stay. It may be that food cues are less effi-cient than marble cues since the food maydistract the animals' attention from thechosen object. These experiments show thatas a Trial 1 cue (reinforcer), a marble canfunction as effectively as food reward, andthey support emphases on the cue value ofreward.

ANALYSIS AND REVIEW OF THEORIES

This section reviews several theories thathave addressed themselves to the question ofwhat is learned during LS formation.

Modified Hull-Spence Theory

Reese (1964) suggested a modified versionof Hull-Spence theory that might accountfor LS formation. It correctly predicts thatearly in LS training a success will reducefuture errors more than an error, while laterin training an error will reduce future errorsmore than a success. Reese also drew theimplications that the stimulus preferencesmust be overcome before the efficient LS per-formance can occur, and retention should begreater early in training than later. Doubt iscast on the former proposition by Bessemer's(see Footnote 5) finding that stimulus pref-erences persist in LS-sophisticated animalsand on the latter by the efficient retention ofspecific discriminations and the correspondingfailure to obtain transfer suppression.

More fundamentally, one might questionthe assumption of a direct strengthening func-tion of rewards. We have seen that monkeys

can learn problems requiring the animal toavoid the object just chosen and rewarded(i.e., win-shift) and to approach an objectthat had just been followed by nonreward(i.e., lose-stay) (e.g., McDowell et al, 1965a)and that a marble is as reinforcing as a raisin(e.g., Riopelle et al., 1954).

Hypothesis Theories

The previously mentioned difficulty withReese's modification of Hull-Spence theory isresolved by hypothesis theories. They viewLS formation as a more abstract processin which hypotheses rather than single re-sponses are regarded as subject to principlesof reinforcement. Strategies or hypothesesrather than single responses may incorporateTrial 1 outcomes as cues, and thus a win-shifthypothesis is just as possible as a win-stay hy-pothesis.

1. Harlow's error-factor theory. Harlow'serror-factor theory is a uniprocess theory, andonly one basic learning process, inhibition, isassumed to occur. It is assumed that the cor-rect response is immediately available butmust compete with many inappropriate re-sponses in the learning situation. Formationof a LS, according to the theory, occurs whenthe monkey has eliminated these error factorsor inappropriate response tendencies.

Error-factor theory is not fully developedsince specific assumptions and postulates havenot yet been presented. For example, whiletransfer suppression would seem to be con-sistent with error-factor theory, it is not ob-vious that lack of suppression constitutes evi-dence against the theory. The various errorfactors do not completely disappear duringLS formation (e.g., Bessemer, see Footnote 5;Davis et al., 1953), but the implications ofthis are unclear because the proportion ofresponses controlled by a given error factorpresently cannot be determined, and conse-quently the strengths of the various errorfactors cannot be compared.

2. Levine's hypothesis behavior model.Levine's model (Levine, 1959, 1965) differsfrom Harlow's in several fundamental ways.The model includes both error-producing andreward-producing response patterns, hencethe term hypothesis rather than error factor.Formation of LS occurs by the strengthening


of the correct hypothesis by 100% reinforce-ment and the gradual extinction of other hy-potheses because of 50% reinforcement. Allresponse patterns are measured, and fromthis the proportion of the time a specifichypothesis is used can be estimated. The de-tailed hypotheses are specific enough so thatby some algebraic manipulations Levine canderive the proportion of time each of thesehypotheses was used in a set of data. Themodel was tested (Levine, 19S9) by observingsome proportions of response sequence pat-terns and then predicting the proportion ofoccurrence of other response patterns. In gen-eral the agreement between predicted and ob-served proportions of responses is quite good.Interestingly, with monkeys, nonzero estimatesof hypotheses seem to occur only for positionpreference, stimulus preference, the correctsolution,7 and random or residual responding.At present, the model says nothing about re-tention of specific information except thathypotheses, and therefore memory concerningstimulus properties, persist for at least threetrials.

3. Restle's mathematical model. Restle's(19S8) explanation of LS formation empha-sizes the cue function of rewards and centersimportantly on an assumption of abstractcognition. The theory assumes three classes ofcues are available during LS formation. Thereare Type a cues, which are relevant and com-mon to all problems of the experiment; Typeb cues, which are relevant within any oneproblem but which are not valid across prob-lems ; and Type c cues, which are not valid atany time. The Type a cues are abstract, andLS formation is seen to involve learning touse Type a cues and to ignore or adapt outinvalid cues (Type c) and those valid withinindividual problems (Type b).

Restle cast his theory in mathematical formand discovered parameter values which al-lowed him to accurately describe both intra-and interproblem learning curves of data fromprevious LS studies.

7 Bowman (Bowman, 1963; Bowman & Takemura,1965) has suggested separating win-stay and lose-shift from each other since the two components maybe learned separately. Bowman's analysis is similarenough to Levine's that it is not given separatetreatment here.

Since Type b cues get adapted, it seemscrucial for the theory to predict transfer sup-pression, and, unfortunately for the theory,transfer suppression does not occur.

A question applying to all three modelsconcerns the interplay between abstract cuesor rules and specific stimulus properties. Hy-pothesis models principally have been con-cerned only with describing or representingdifferences in performance between initial andterminal states of learning with little empha-sis on transitional functions. Little attentionhas been directed to ways in which learningprocesses studied in other contexts operate inLS formation. The escape from the assumptionof an automatic strengthening function ofreward appears to have been at the expenseof a loss of contact with such fundamentalareas of learning theory as stimulus generali-zation and retention. The model discussed inthe next section ties LS formation moreclosely to specific stimulus properties and isconsistent with efficient retention of specificdiscriminations.

Feedback Theory of Learning-Set Formation

The reader is referred to a companion pa-per 8 by this author for a full treatment ofthe feedback theory of LS formation. Thistheory is an elaboration of a scanning model(Estes, 1966) which has already been suc-cessfully applied to reward magnitude learn-ing by monkeys (Meyer, LoPopolo, & Singh,1966). The main feature of the scanningmodel is that on a given trial, the subjectscans the available cues, generates a feedback(covert prediction of reward value) for eachcue, and makes the response which he pre-dicts will yield the highest feedback.

Choices are controlled by stimulus proper-ties of cues (i.e., their salience) and by ex-pected feedback from previous rewards. Re-wards affect choices by the facilitatory or in-hibitory feedback their anticipation causes,but they do not directly affect learning. Asapplied to LS, the model assumes that re-wards and nonrewards increase or decreasethe expected feedback or anticipated rewardvalue for the cue to which the animal re-

8 D. L. Medin. A feedback model for discrimina-tion learning set in monkeys. Unpublished manu-script.


sponded.9 An individual problem is solved byan active selection of valid cues rather thanan adaption of irrelevant cues. For object-discrimination LS, the model assumes themonkey responds to one of four cues: (a) oneobject, (b) the other object, (c) the left po-sition, and (d) the right position.

The basis for LS improvement is assumedto be interproblem stimulus generalization offeedback from expected rewards. This idearequires some explanation, since the sugges-tion that LS is mediated by specific between-problem stimulus generalization was rathercasually dismissed in the introduction to thisreview. According to the model, transfer offeedback from anticipated rewards does notmysteriously become associated solely withthe object to be correct on future problems.Positive feedback from rewards associatedwith similar cues on previous problems istransferred by stimulus generalization to boththe future correct and incorrect objects. Eventhough no differential advantage accrues tothe object to be correct on new problems, thescanning or cue-selection process functionsmore efficiently when initial feedback fromanticipated rewards is high. Given high initialexpectation of rewards, the monkey is able tofocus on the correct cue almost immediately,and the discrimination is solved quickly.When the feedback associated with the correctand incorrect stimuli is low, even large differ-ences in feedback for the correct and incor-rect cues do not lead to efficient performance.One reason for this is that both cues wouldbe at a disadvantage in competing for atten-tion with other cues in the situation. When thefeedback associated with both cues is high,the scanning process functions to narrow at-tention down to these two cues. Though thefeedback associated with the incorrect objectmay be large, the feedback associated withthe correct cue need only obtain a small ad-vantage in order to control responding vir-tually perfectly. High initial feedbacks, inaddition to placing certain cues at a competi-tive advantage, allow the scanning process to

8 However, in general, reward does not automati-cally strengthen responses since a decrease in feed-back might follow reward after the monkeys havebeen trained that nonreward follows reward as ona win-shift problem.

perform more efficiently. This ability to effi-ciently focus on, or so to speak, home in on,cues with high feedback is a property of manystochastic learning models.

Most of the detailed predictions of themodel were drawn from a computer simula-tion. The model is able to generate typicalbetween- and within-problem LS functions. Itcorrectly predicts that early in LS training acorrect response reduces future errors morethan an error, while late in LS an error re-duces future errors more than a correct re-sponse. In addition, data produced by themodel showed a pattern of error factors quitesimilar to that of real monkeys (Davis et al.,1953).

Specific criticisms of the feedback modelawait its future development and testing. Ifthe feedback model for discrimination LSwere to receive consistent support, it wouldhave major implications for what monkeyslearn during LS training. Historically, LS for-mation has been considered to be a complexabstractive process involving a conceptualunderstanding on the part of the monkeys.The feedback model does not assume any suchabstract cognition and suggests that specificstimulus properties play a major role in LSformation. The feedback model suggests thatthe monkey does not so much learn how tolearn as learn what to expect.

REFERENCES

ABORDO, E. J., & RUMBAUGH, D. M. Response con-tingent learning set training in the squirrel monkey.Psychological Reports, 1965, 16, 797-802.

BEHAR, I. Analysis of object-alternation learning inrhesus monkeys. Journal of Comparative andPhysiological Psychology, 1961, 54, 539-542. (a)

BEHAR, I. Learned avoidance of nonreward. Psycho-logical Reports, 1961, 9, 43-52. (b)

BEHAR, I. Evaluation of cues in learning set forma-tion in mangabeys. Psychological Reports, 1962,11, 479-485. (a)

BEHAR, I. Evaluation of the significance of positiveand negative cue in discrimination learning. Jour-nal of Comparative and Physiological Psychology,1962, 55, 502-504. (b)

BERNSTEIN, I. L. Presentation factors in discrimina-tion learning by chimpanzees. Psychological Re-ports, 1961, 8, 151-153.

BOWMAN, R. E. Discrimination learning set per-formance under intermittent and secondary rein-forcement. Journal of Comparative and Physio-logical Psychology, 1963, 56, 429-434,


BOWMAN, R. E., & TAKEMURA, K. Hypothesis be-havior of monkeys as inferred from intermit-tently rewarded and other modified learning sets.Proceedings of the 73rd annual convention of theAmerican Psychological Association Convention1965. Washington, D. C.: APA, 1965.

BOWMAN, R. E., & TAKEMURA, K. The response shiftphenomenon in discrimination learning set. Psy-chonomic Science, 1966, 4, 31-32.

BOYER, W. N. Discrimination performance in threespecies of monkeys as a function of Trial 1 re-ward contingency, intertrial interval, and priortest experience. Psychonomic Science, 1966, 5, 27-28.

BRAUN, H. W., PATTON, R. A., & BARNES, H. VV.Effects of electroshock convulsions upon thelearning performance of monkeys: I. Object-qual-ity discrimination learning. Journal of Comparativeand Physiological Psychology, 1952, 45, 231-238.

BROWN, W. L., & CARR, R. M. The learning of inci-dental cues by rhesus monkeys. Journal of Com-parative and Physiological Psychology, 1958, 51,459-460.

BROWN, W. L., & MCDOWELL, A. A. Response shiftlearning set in monkeys. Journal of Comparativeand Physiological Psychology, 1963, 56, 335-336.

BROWN, W. L., MCDOWELL, A. A., & GAYLORD, H. A.Two-trial learning-set formations by baboons andby stump-tailed macaques. Journal of Compara-tive and Physiological Psychology, 1965, 60, 288-289.

BROWN, W. L., OVERALL, J. E., & BLODGETT, H. C.Novelty learning sets in rhesus monkeys. Journalof Comparative and Physiological Psychology,1959, 52, 330-332.

BROWN, W. L., OVERALL, J. E., & GENTRY, G. V.Conceptual discrimination in rhesus monkeys.Journal of Comparative and Physiological Psychol-ogy, 1958, 51, 701-705.

CROSS, H. A., & BROWN, L. T. Discrimination re-versal learning in squirrel monkeys as a functionof number of acquisition trials and prereversal ex-perience. Journal of Comparative and Physio-logical Psychology, 1965, 59, 429-431.

CROSS, H. A., FLETCHER, H. J., & HARLOW, H. F.Effects of prior experience with test stimuli onlearning-set performance of monkeys. Journal ofComparative and Physiological Psychology, 1963,56, 204-207.

DARBY, C. L., & RIOPELLE, A. J. Differential problemsequences and the formation of learning sets.Journal of Psychology, 1955, 39, 105-108.

DAVIS, R. T. The radiation syndrome. In A. M.Schrier, H. F. Harlow, & F. Stollnitz (Eds.), Be-havior of nonhuman primates: Modern researchtrends. New York: Academic Press, 1965.

DAVIS, R. T., MCDOWELL, A. A., & THORSON, N.Four-trial object-quality discrimination learningby monkeys. Proceedings of the South DakotaAcademy of Science, 1953, 32, 132-142.

RSTES, W. K. Transfer of verbal discriminationsbased on differential reward magnitudes. Journalof Experimental Psychology, 1966, 72, 276-283.

ETTLINGER, G. Discrimination learning theory: Exci-tation vs. inhibitory tendencies in monkeys.Quarterly Journal of Experimental Psychology,1960, 12, 41-44.

FLETCHER, H. J. Prompted vs trial-and-error 3-trialobject discrimination learning by monkeys. Psycho-nomic Science, 1966, 4, 111-112.

FLETCHER, H. J., & CROSS, H. A. Effects of trial 1reward contingency, intertrial interval, and experi-ence on intraproblem discrimination performanceof monkeys. Journal of Comparative and Physio-logical Psychology, 1964, 57, 318-320.

FLETCHER, H. J., GROOG, T. M., & GARSKE, J. P.Ambiguous cue problem performance of children,retardates, and monkeys. Journal of Comparativeand Physiological Psychology, 1968, 66, 477-482.

FLETCHER, H. J., & TAKEMURA, K. Prompted vstrial and error concept discrimination learning inmonkeys. Perceptual and Motor Skills, 1965, 20,737-743.

FRENCH, G. M., BIRNBAUM, I., LEVINE, R., & PINS-KER, H. Discrimination choice in normal and pre-frontal rhesus monkeys. Journal of Comparativeand Physiological Psychology, 1965, 59, 225-230.

GENTRY, G. V., OVERALL, J. E., & BROWN, W. L.Discrimination after pretraining on componentstimuli. Journal of Comparative and PhysiologicalPsychology, 1958, 51, 464-466.

HARLOW, H. F. Studies in discrimination learning bymonkeys: I. The learning of discrimination seriesand the reversal of discrimination series. Journalof General Psychology, 1944, 30, 3-12.

HARLOW, H. F. The formation of learning sets. Psy-chological Review, 1949, 56, 51-65.

HARLOW, H. F. Analysis of discrimination learningby monkeys. Journal of Experimental Psychology,1950, 40, 26-39.

HARLOW, H. F. Learning set and error factor theory.In S. Koch (Ed.), Psychology: A study of a sci-ence. Vol. 2. New York: McGraw-Hill, 1959.

HARLOW, H. F., & WARREN, J. M. Formation andtransfer of discrimination learning sets. Journal ofComparative and Physiological Psychology, 1952,45, 482-489.

HAYES, K. J., THOMPSON, R., & HAYES, C. Concur-rent discrimination learning in chimpanzees.Journal of Comparative and Physiological Psy-chology, 1953, 46, 105-107.

KING, J. E. Transfer relationships between learningset and concept formation in rhesus monkeys.Journal of Comparative and Physiological Psy-chology, 1966, 61, 416-420.

KING, J. E., & GOODMAN, R. Successive and concur-rent discrimination by rock squirrels and squirrelmonkeys. Perceptual and Motor Skills, 1966, 23,703-710.

KRUPER, D. C., PATTON, R. F., & KOSKOFF, Y. D.Delayed object-quality discriminations in hemi-cerebrectomized monkeys. Journal of Compara-tive and Physiological Psychology, 1961, 54, 619-624.


LEARY, R. W. The rewarded, the unrewarded, thechosen and the unchosen. Psychological Reports,1956, 2, 91-97.

LEARY, R. W. Analysis of serial-discrimination learn-ing by monkeys. Journal of Comparative andPhysiological Psychology, 1958, 51, 82-86. (a)

LEARY, R. W. The learning of ambiguous cue-prob-lems by monkeys. American Journal of Psychology,1958, 71, 718-724. (b)

LEARY, R. W. The temporal factor in reward andnonreward of monkeys. Journal of ExperimentalPsychology, 1958, 56, 294-296. (c)

LEARY, R. W. 'Spontaneous reversal' in serial dis-crimination reversal learning of monkeys. CanadianJournal of Psychology, 1962, 16, 228-233.

LEVINE, M. A model of hypothesis behavior in dis-crimination learning set. Psychological Review,1959, 66, 353-366.

LEVINE, M. Hypothesis behavior. In A. M. Schrier,H. F. Harlow, & F. Stollnitz (Eds.), Behavior ofnonhuman primates: Modern research trends.New York: Academic Press, 1965.

LOCKHART, J. M., PARKS, T. E., & DAVENPORT, J. W.Information acquired in one trial by learning-setexperienced monkeys. Journal of Comparative andPhysiological Psychology, 1963, 56, 1035-1037.

MASON, W. A., BLAZEK, N. C., & HARLOW, H. F.Learning capacities of the infant rhesus monkey.Journal of Comparative and Physiological Psy-chology, 1956, 49, 449-453.

MCCONNELL, D. G., & SCHUCK, J. R. An analysis ofprimate responses to colored filmstrips stimuli.Journal of Comparative and Physiological Psychol-ogy, 1962, 55, 505-510.

McDowEix, A. A., & BROWN, W. L. The learningmechanism in response shift learning set. Journalof Comparative and Physiological Psychology,1963, 56, 572-574. (a)

MCDOWELL, A. A., & BROWN, W. L. Learning mech-anism in response perse veration learning sets.Journal o,f Comparative and Physiological Psy-chology, 1963, 56, 1032-1034. (b)

McDowEix, A. A., & BROWN, W. L. Sex and radia-tion as factors in peripheral cue discriminationlearning. Journal of Genetic Psychology, 1963, 102,261-265. (c)

MCDOWELL, A. A., & BROWN, W. L. Further evidencefor excitatory properties of nonrewarded cues.Journal of Genetic Psychology, 1965, 106, 137-140.(a)

MCDOWELL, A. A., & BROWN, W. L. Learning mech-anisms in response shift learning set to nonre-warded cues. Journal of Genetic Psychology, 1965,106, 173-176. (b)

MCDOWELL, A. A., & BROWN, W. L. Perseverationlearning set formation to nonrewarded cues bynormal and previously irradiated rhesus monkeys.Journal of Genetic Psychology, 1965, 107, 309-311.(c)

MCDOWELL, A. A., & BROWN, W. L. Response per-severation under constant stimulus-position condi-tions by normal and previously irradiated female

rhesus monkeys. Journal of Genetic Psychology,1965, 106, 81-84. (d)

MCDOWELL, A. A., & BROWN, W. L. Response shiftlearning-set formation by normal and previouslyirradiated female rhesus monkeys. Journal of Gen-eral Psychology, 1965, 72, 339-342. (e)

MCDOWELL, A. A., GAYLORD, H. A., & BROWN, W. L.Learning set formation by naive rhesus monkeys.Journal of Genetic Psychology, 1965, 106, 253-257.(a)

MCDOWELL, A. A., GAYLORU, H. A., & BROWN, W. L.Perseveration learning set formation by monkeyswith previous discrimination training. Journal ofGenetic Psychology, 1965, 106, 345-347. (b)

MEYER, D. R. Intraproblem-interproblem relation-ships in learning by monkeys. Journal o.f Compara-tive and Physiological Psychology, 1951, 44, 162-167.

MEYER, D. R., LoPoPOLo, M. H., & SINGH, D.Learning and transfer in the monkey as a functionof differentiated levels of incentive. Journal ofExperimental Psychology, 1966, 72, 284-286.

MILES, R. C. Discrimination-learning sets. In A. M.Schrier, H. F. Harlow, Si F. Stollnitz (Eds.), Be-havior of nonhuman primates: Modern researchtrends. New York: Academic Press, 1965.

REESE, H. W. Discrimination learning set in rhesusmonkeys. Psychological Bulletin, 1964, 61, 321-340.

RESTLE, F. Toward a quantitative description oflearning set data. Psychological Review, 1958, 65,77-91.

RIOPELLE, A. J. Transfer suppression and learningsets. Journal of Comparative and PhysiologicalPsychology, 1953, 46, 108-114.

RIOPELLE, A. J. Learning sets from minimum stim-uli. Journal of Experimental Psychology, 1955, 49,28-32. (a)

RIOPELLE, A. J. Rewards, preferences and learningsets. Psychological Reports, 1955, 1, 167-173. (b)

RIOPELLE, A. J., CHRONHOLM, J, N., & ADDISON, R.G. Stimulus familiarity and multiple discrimina-tion learning. Journal of Comparative and Physio-logical Psychology, 1962, 55, 274-278.

RIOPELLE, A. J., & CHURUKIAN, G. A. The effect ofvarying the intertrial interval in discriminationlearning by normal and brain-operated monkeys.Journal of Comparative and Physiological Psy-chology, 1958, 51, 119-125.

RIOPELLE, A. J., & COPELAND, E. L. Discriminationreversal learning to a sign. Journal of ExperimentalPsychology, 1954, 48, 143-145.

RIOPELLE, A. J., & FRANCISCO, E. W. Discriminationlearning performance under different first-trialprocedures. Journal of Comparative and Physio-logical Psychology, 1955, 48, 143-145.

RIOPELLE, A. J., FRANCISCO, E. W., & ADES, H. E.Differential first-trial procedures and discrimina-tion learning performance. Journal of Comparativeand Physiological Psychology, 1954, 47, 293-297.

RIOPELLE, A. J., & MOON, W. H. Problem diversityand familiarity in multiple discrimination learn-


ing by monkeys. Animal Behaviour, 1968, 16, 74-78.

RUMBAUGH, D. M., & ENSMINGER, L. W. Discrimi-nation reversal training with single and multiplestimulus pairs in the squirrel monkey. Journal ofComparative and Physiological Psychology, 1964,57, 304-306.

RUMBAUGH, D. M., & PRIM, M. M. Temporary in-terference of insolvable discrimination reversaltraining upon learning set in the squirrel monkey.Journal of Comparative and Physiological Psy-chology, 1964, 57, 302-304.

RUMBAUGH, D. M., SAMMONS, M. E., PRIM, M. M.,& PHILLIPS, S. Learning set in squirrel monkeys asaffected by pretraining with differentially rewardedsingle objects. Perceptual and Motor Skills, 1965,21, 63-70.

SCHRIER, A. M. Transfer by macaque monkeys be-tween learning set and repeated reversal tasks.Perceptual and Motor Skills, 1966, 23, 787-792.

SCHRIER, A. M. Learning set without transfer sup-pression: A replication and extension. PsychonomicScience, 1969, 16, 263-265.

SCHUSTERMAN, R. J. Transfer effects of successivediscrimination reversal training in chimpanzees.Science, 1962, 137, 422-423.

SCHUSTERMAN, R. J. Successive discrimination-re-versal training and multiple discrimination train-ing in one-trial learning by chimpanzees. Journalof Comparative and Physiological Psychology,1964, 58, 153-156.

SCHWARTZBAUM, J. S., & POULAS, D. A. Discrimina-tion behavior after amygdalectomy in monkeys:Learning set and discrimination reversals. Journalof Comparative and Physiological Psychology,1965, 60, 320-328.

SHELL, W. F., & RIOPELLE, A. J. Progressive dis-crimination learning in platyrrhine monkeys.Journal of Comparative and Physiological Psy-chology, 1958, 51, 467-470.

SHERIDAN, C. L., HOREL, J. A., & MEYER, D. R. Ef-fects of response-induced stimulus change onprimate discrimination learning. Journal of Com-parative and Physiological Psychology, 1962, 55,511-S14.

SLEDJESKI, M., & FRENCH, G. M. Multiple transferparadigms following discrimination problem solu-tion in monkeys. Journal of Comparative andPhysiological Psychology, 1968, 66, 228-231.

STOLLNITZ, F. Spatial variables, observing responses,and discrimination learning sets. Psychological Re-view, 1965, 72, 247-261.

STOLLNITZ, F., & SCHRIER, A. M. Learning set with-out transfer suppression. Journal of Comparativeand Physiological Psychology, 1968, 66, 780-783.

STRONG, P. N., JR. Memory for object discriminationin the rhesus monkey. Journal of Comparativeand Physiological Psychology, 1959, 52, 333-335.

TAKEMURA, K. Saru ni okeri benbets u gakushu nokamae no jigenkav icho [Inter-categorical transferof discrimination learning sets in monkeys.] An-nual Animal Psychology, 1960, 10, 55-63.

THOMPSON, R. Approach vs. avoidance in an am-biguous-cue discrimination problem in chimpanzees.Journal of Comparative and Physiological Psy-chology, 1954, 47, 133-135.

TREICHLER, F. R. Transfer effects in monkey dis-crimination learning after extensive two-problemtraining. Psychonomic Science, 1966, 5, 201-202.

VON WRIGHT, J. M. Cross-modal transfer and sensoryequivalence—A review. Scandinavian Journal ofPsychology, 1970, 11, 21-30.

WARREN, J. M. Reversal learning and the formationof learning sets by cats and by rhesus monkeys.Journal o.j Comparative and Physiological Psychol-ogy, 1966, 61, 421-428.

WARREN, J. M., & SINI-IA, M. M. Interaction betweenlearning sets in monkeys. Journal of Genetic Psy-chology, 1959, 95, 19-25.

WILSON, M., & WILSON, W. A., JR. Intersensory fa-cilitation of learning sets in normal and brain-operated monkeys. Journal of Comparative andPhysiological Psychology, 1962, 55, 931-934.

ZIMMERMANN, R. R. Effects of age, experience, andmalnourishment on object retention in learningset. Perceptual and Motor Skills, 1969, 28, 867-876.

(Received August 4, 1970)

Documents

VOL. 1972 Psychological Bulletin - Northwestern Universitygroups.psych.northwestern.edu/medin/documents/... · VOL. 77, No. 5 MAY 1972 Psychological Bulletin ... consin General Test