Extended practice with a single-character classification task 1

JOHN ROSS2

UNIVERSITY OF WESTERN AUSTRALIA

Table 1Checklists for Each S

Size Size 2 Size 4 Size 8

M.E. L. V V8 V8F6 V8F62NPBJ.H.H. 4 4 7 47FJ 47FJ8Z56P. B.P. B B5 B5SU B5SUEDFAD. L.S. X XA XARF .XARFV8E3

Extended practice was given on asingle-character classification task withfour nested checklists of Sizes I, 2, 4, and8 defining positive sets. Reaction timereduces with practice, but is linear with thelogarithm of checklist size at all stages oftraining, and transfers to characters of largesize and different case.

When people search through long stringsof characters, looking for targets defmedby a memorized checklist, the rate atwhich they can search seems to beindependent of the size of the checklist.They are just as quick working from a listof ten as from a list of one, at least afterextended practice (Neisser, Novick, &Lazar, 1963; Shurtleff & Marsetta, 1968).The results suggest the simultaneouscomparison of stimuli against all items onthe memorized checklist. They suggest, in aword, parallel processing in memory.

In stubborn, and puzzling, contrast,choice reaction time to charactersdisplayed singly increases with the size ofthe checklist both when the checklist variesfrom trial to trial, and when it is keptconstant for many trials or sessions.Sternberg (1966, 1967) consistently fmdsreaction time to increase linearly withchecklist length, and normally to beindependent of serial position on thechecklist. Targets high up on the checklisttake just as long to react to as targets listedlower down. Sternberg's conclusion is thatdecision as to whether an item displayed ison the checklist or not depends on a searchthrough the memorized checklist that isboth serial and exhaustive, which meanSthat the whole list has to be inspected,item by item, even after a match hasoccurred early. Nickerson (1966) andBriggs and Blalta (1969) both confmnSternberg's fmdings for the case of a singlecharacter display, and extend them byshowing that when stimulus strings of upto four are presented, reaction time stillincreases linearly, or roughly so, withchecklist size. So though the task is movedin the direction of the search task discussedin the previous paragraph, by introducingstrings of stimuli, the results refuse tobudge.

The purpose of the experiment reportedin this paper was to determine the effectsof extended practice on a single-characterclassification task, since one difference,though by no means the only one, between

the studies that seem to point in oppositedirections, parallel and serial processing,respectively, is in amount of practicetypically given. On early trials search time,like reaction time in classification, doesshow an increase with memory load, butthe effect diminishes and finally disappearswith trials.

SUBJECTSFour senior students at the University of

Western Australia, three male and onefemale, served as Ss.

APPARATUSStimuli for the main part of the

experiment were made from upper-casecharacters from the Letraset SheetNo. 690, photographed and back projectedfrom negative slide transparencies by anautomatic slide projector for an exposuretime of 40 msec. As projected stimulicharacters were 15/16-in. high and 7/8-in.wide.

S viewed the projection screen from adistance of 48 in., eyes level with thecenter of a rectangle, 43/4 x 5 1/4 in.,marked with black insulation tape as afixation aid. Stimuli appeared in the centerof the rectangle. S wore a pair of PioneerModel SE·P2 earphones, through which atone to mark apparatus noise wasdelivered, and his index fingers rested onseparate response buttons. A downwardforce of 70 g closed a contact, and stoppeda timer that started simultaneously withthe onset of stimulus exposure. Theluminance of the characters as projectedwas 180 mL, and the luminance of thescreen 0.5·1 mL.

DESIGN AND PROCEDUREChecklists for each S are shown in

Table 1. It should be noted that thechecklist of eight items always includesthose on the checklist of Size 4, and so ondown. In each condition S is required tomake a positive response if a checklisteditem appears, and a negative response ifnot. The negative sets for each S containthe 24 alphabetic characters and digits

remammg after removing items on thechecklist of Size 8, the alphabeticcharacters I and 0, and the digits I and O.

There are four conditions, each definedby checklist size. The negative set remainedthe same for all conditions. The probabilityof occurrence of both positive and negativestimuli was also constant at 1/2 overconditions. The probability of any ~iven

negative set member was constant at 1/48in all conditions, while the probability ofany given positive set member was 1/2 forchecklist Size 1, 1/4 for Size 2, 1/8 forSize 4, and 1/16 for Size 8.

Each S first learned the four checklistsassigned to him, and was given pretrainingwith them on a card-sorting task. Trials onthe main tasks were run for 20 sessions, asfar as possible on consecutive days for eachS. Each session consisted of eight blocks oftrials, two blocks per condition, the orderof conditions being determined by arandomizing procedure. Each blockconsisted of 20 trials, so that each Shad 40trials per condition at each of the 20sessions. The checklist for the block wasnominated just before the trials began. Thesequence within each trial was: (1) maskingtone for 1/2 sec; (2) test stimulus for40 msec; (3) response; (4) feedback lightgiving knowledge of results. There was anintertrial interval of approximately 3 sec.

Three extra sessions were run at the endof the main sequence. In Session 21 thesize of the stimuli was doubled byincreasing the distance of the projectorfrom the screen. In Sessions 2 and 23lower-ease stimuli from Letraset SheetNo.691 were substituted for theupper-case stimuli used in all previoussessions, and projected at the size used inSession 21. Procedures in the three extrasessions were otherwise identical withthose in the main test ~equence.

RESULTSFigure I shows that choice reaction

time, averaged over all conditions and forresponse types, positive and negative, dropssharply over the first few sessions and

276 Copyright 1970, Psychonomic Journals, Inc., Austin, Texas Perception & Psychophysics, 1970, Vol. 8 (SA)

bOO

DAY I

,",-504-3+50·0 lOGzS

600

650

Fig. I. Mean reaction time, averaged overSs and checklists for each of the 20 mainsessions, and for the three extra sessions inwhich display size was changed(Session 21) and the case changed fromupper to lower (Sessions 21 and 2).

main test sequence. A second session withlower case stimuli shows improvement(Session 23). The differences betweenSessions 21 and 22 and between Sessions22 and 23 are both highly significant.

DISCUSSIONExtended training with four checklists

substantially improves choice reactiontimes in a task requiring the classificationof a single character as belonging to achecklist or not, and reduces but does notremove the effect of checklist size. There isnothing in the results to suggest thatreaction time would ever becomeindependent of checklist size, as scanningrate appears to be in the studies of Neisser,Novick, and Lazar (1963) and Shurtleffand Marsetta (1968). Training of heroicproportions, like that given by Mowbrayand Rhoades (l959), who gave 40,000trials, might do the trick, but clearly moreis needed than is sufficient with scanning.

While the results therefore argue againstthe possibility that classification comes todepend on parallel processing afterextended practice, they also poseconsiderable difficulties for a serial andexhaustive model, since reaction timeincreases linearly with the logarithm of

I

II

I 0

I

I

I

~I I

I 0 I

I I

roughly linearly with the logarithm ofchecklist size. Training has little effect onerror rate, the effect being a slight but notsignificant increase in error rate oversessions. Table 2 gives intercept values forthe straight lines fitted to plots of reactiontime against logarithm of checklist size forSessions I, 2, 3, 10, and 20. It can be seenthat both slope and intercept reducesystematically, the sharp reduction in bothcases being on the early sessions.

Performance on the three additionalsessions is shown in Fig. 1 and, in moredetail, in Table 3. It can be seen thatincreasing the size of the stimuli(Session 21) sharply improves reactiontime, and consistently so, for all checklists.The effect is highly significant. Changingfrom upper to lower case at the larger size(Session 22) brings reaction time back tonear the level reached at the end of the

g 550

'"~

z..~ 450

....2:;:z 500<>;:u.....""

continues to drop slowly over all sessions.Positive and negative reaction times bothimproved similarly. the interaction ofresponse type with sessions being small.The response to positive stimuli wasconsistently faster than the response tonegative stimuli for all checklist sizes, andthe two curves nearly parallel at all stagesof training. The average difference in favorof positive responses was about 40 ",~ec.

Figure 2 shows choice reactjl)l1 timeaveraged for positive and negativeresponses on Session 1 and Session 20,plotted against the logarithm of checklistsize. In can be seen that the increase inreaction time is roughly linear withincrease in logarithm of checklist size, notwith checklist size itself, and that trainingsubstantially reduces the slope of the lineas well as its intercept. Figure 2 also showsthat the proportion of errors increases

Table 3Mean Reaction Time in Msec for Each Checklist on the Last Main Session and the

Three Extra Sessions, Where S Equals Checklist Size

5SO

40010

iS :i~

3SO l--L.L-__..J..1.__-L.L-__J.l......J 0 ~2 4

SIZE Of CHECK LiST(LOGARITHMIC COOIIOINlTESl

415.422.3

20

2 In.2nd Session

Session 23

10

420.425.4

Lower Case

2 In.1st Session

Session 22

3

Day

438.326.7

442.236.7

2

2 In.Stimuli

Session 21

Upper Case

504.350.0

1 In.Stimuli

Session 20

InterceptSlope

Table 2Slope and Intercept Parameters (Msec) for Best Fitting Straight Lines to Plots of Reaction Time

Against Logarithm of Size of Checklist on Each of Five Different Sessions(Plots for Sessions I and 20 are Shown in Fig. 2)

Condition

S = 1S = 2S = 3S = 4

Mean

413.7439.6462.4480.0

449.0

386.0417.7440.2450.8

423.7

40I.l438.4456.1479.4

443.8

393.4421.6455.5460.0

432.6

Fig. 2. Reaction time for each checkliston Session 1 and Session 20, and the meanerror for each checklist over all sessions,and best fitting logarithmic functions,where S equals size.

Perception & Psychophysics, 1970, Vol. 8 (SA) 277

checklist size. It is possible that the resultis partly due to a higher error rate forlonger checklists, with Ss trading accuracyfor speed, but error rates are notsubstantially different in this study fromthose reported for the comparableconditions by Sternberg (1967). It mightbe noted in passing that a checklist ofSize 8 is needed to show the logarithmicrelationship. The evidence would beambiguous, and similar to Sternberg's(1967) without the largest checklist.

A model in which choice is based on aset of features characterizing each checklistas a whole, and distinguishing it from thenegative set might more easily explain thelogarithmic relationship, since eachadditional feature doubles the possiblestimuli if redundancy is at a minimum. Noattempt will be made to develop any suchmodel in detail, but three points may bemen tioned that should govern thedevelopment of any detailed model. Thefirst is that reaction time is typical of thechecklist, not of individual stimuli in it. Ananalysis of reaction times for individualstimuli within the checklist of Size 8showed no difference between them,despite the fact that the first of them is the

checklist of Size I and as such has thebenefit of more practice and shows a fasterreaction time under Size 1 conditions.Similarly, the first two and four are thechecklists of Size 2 and Size 4,respectively. The second point is thatreaction time for negative responses isslower than for positive responses by about40 msec (cf. Briggs & Blaha, 1969) for allchecklist sizes, suggesting that a negativeresponse is made after a failure to decide tomake a positive response.

The last and most intriguing point is thatwhatever is learned seems to transfer notonly to stimuli of different size(Session 21) but to stimuli of different sizeand form (Sessions 22 and 23).Interpretation must be cautious, since thestudy did not include controls, bu t it seemsunlikely that reaction time to large upperand lower case stimuli would be as fastwithout training as it is after training onsmall upper-case stimuli.

If reaction time is relatively indifferentto the size and case of characters, it wouldseem that the decision process inclassification does not, or at least need not,rely on information about the physicalform of the stimulus.

REFERENCESBRIGGS, G. E., & BLAHA, J. Memory retrieva'

aIld central comparison times in informatiorprocessing. Journal of ExperimentaPsychology, 1969, 79, 395402.

MOWBRAY, G. H., & RHOADES, M. U. On thtreduction of choice-reaction times wittpractice. Quarterly Journal of ExperimentaiPsychology, 1959, 11,16-23.

NEISSER, U., NOVICK, R., & LAZAR, R.Searching for ten targets simultaneously.Perceptual & Motor Skills. 1963, 17,955-961.

SHURTLEFF, D. A., & MARSETTA, M. Y.Visual search In a letter-cancelling taskreexamined. Journal of ExperimentalPsychology, 1968,77.19-23.

STERNBERG, S. High-speed scanning in humanmemory. Science. 1966, 153,652-654.

STERNBERG, S. Two operations in characterrecognition: Some evidence from reaction-timeexperiments. Perception & Psychophysics,1967,2,45-53.

NOTES1. This report is based on an Honors thesis

investigation carried out at the University ofWestern Australia by J. S. Warnock, under thesupervision of the author.

2. Address: Department of PSYChology,University of Western Australia, Nedlands,Western Australia 6009

(Accepted for publication January 19, 1970.)

278 Perception & Psychophysics, 1970, Vol. 8 (SA)