psychological review · 130 richard m. shiffrin and walter schneider c/) h i r-lu 100 80 60...

Psychological ReviewJ Copyright © 1977 \^J by the American Psychological Association, Inc.

V O L U M E 8 4 N U M B E R 2 M A R C H 1 9 7 7

Controlled and Automatic Human Information Processing:II. Perceptual Learning, Automatic Attending,

and a General Theory

Richard M. ShiffrinIndiana University

Walter SchneiderUniversity of California, Berkeley

The two-process theory of detection, search, and attention presented bySchneider and Shiffrin is tested and extended in a series of experiments. Thestudies demonstrate the qualitative difference between two modes of informa-tion processing: automatic detection and controlled search. They trace thecourse of the learning of automatic detection, of categories, and of automatic-attention responses. They show the dependence of automatic detection on at-tending responses and demonstrate how such responses interrupt controlledprocessing and interfere with the focusing of attention. The learning of cat-egories is shown to improve controlled search performance. A general frame-work for human information processing is proposed; the framework emphasizesthe roles of automatic and controlled processing. The theory is compared to andcontrasted with extant models of search and attention.

I. Introduction

In Part I of this paper (Schneider &Shiffrin, 1977) we reported the results ofseveral experiments on search and attentionthat led us to formulate a theory of informa-tion processing based on two fundamentalprocessing modes: controlled and automatic.In the context of search studies, these modestook the form of controlled search and auto-matic detection. Controlled search is highlydemanding of attentional capacity, is usually

The research and theory reported here were sup-ported by PHS Grant 12717 and a GuggenheimFellowship to the first author, Grant MH23878 tothe Rockefeller University, and a Miller Fellowshipat University of California at Berkeley to the secondauthor. This report represents equal and shared con-tributions of both authors.

Requests for reprints should be sent to RichardM. Shiffrin, Department of Psychology, IndianaUniversity, Bloomington, Indiana 47401. '

serial in nature with a limited comparisonrate, is easily established, altered, and evenreversed by the subject, and is strongly de-pendent on load. Automatic detection is rela-tively well learned in long-term memory, isdemanding of attention only when a targetis presented, is parallel in nature, is difficultto alter, to ignore, or to suppress once learned,and is virtually unaffected by load.

In the present article we shall reportseveral studies to elucidate further the proper-ties of automatic and controlled processingand their interrelations, to demonstrate thequalitative difference between these processingmodes, to study the development of automaticdetection and the role of the type and natureof practice in such development, to studythe effects of categorization, and to examinethe development of automatic attending andits effects. After the presentation of thestudies we shall present a general theory ofinformation processing, with emphasis on the

127

128 RICHARD M. SHIFFRIN AND WALTER SCHNEIDER

CM -

TRIAL I

TRIAL 2 rs63

TRIAL 3

TRIAL 4

VM -

TRIAL I

TRIAL 2 CJK

TRIAL 3

TRIAL 4

Figure 1. Examples of trials in the multiple-frame search paradigm of Experiment 1, Part I. In allcases, memory-set size = 4 and frame size = 2. Four varied-mapping (VM) trials and four con-sistent-mapping (CM) trials are depicted. The memory set is presented in advance of each trial,then the fixation dot goes on for .5 sec when the subject starts the trial, and then 20 frames arepresented at a fixed time per frame. Either 0 or 1 member of the memory set is presented duringeach trial. Frame time, memory-set size, and frame size are varied across conditions.

roles of automatic and controlled processing.Our theory will then be compared and con-trasted with extant theories of search andattention.

A. Review of Paradigms and Results FromPart I

The paradigm for most of the studies ofPart I (and Part II also) is depicted inFigure 1. Four elements are presented simul-taneously in a square; and their joint presen-tation for a brief period of time is termed a

frame. Each trial consists of the presentationof 20 frames in immediate succession. Theelements presented are characters (i.e., digitsor consonants) or random dot masks. In ad-vance of each trial the subject is presentedwith several items, called the memory set, andis then required to detect any memory-setitems that appear in the subsequent frames.The frame time is kept constant across the20 frames of each trial, and the basic depen-dent variable is the psychometric functionrelating accuracy to frame time for eachcondition.

PERCEPTUAL LEARNING AND AUTOMATIC ATTENDING 129

Three basic independent variables weremanipulated in Part I/Experiment 1. Thenumber of characters in each frame (theframe size, F) was varied from 1 to 4 (butwas constant for all frames of a given trial).The number of characters presented in ad-vance of a trial (the memory-set size, M)varied from 1 to 4. The product of M and Fis termed the load. A memory-set item thatappears in a frame is called a target; anitem in a frame that is not in the memoryset is called a distractor. One half of the trialscontained one target, and one half containedno target. Finally, and most important, thenature of the training procedure across trialsand the relation of the memory-set items tothe distractors were varied. In the consistentmapping (CM) procedure, across all trials,memory-set items were never distractors (andvice versa). In addition, memory-set itemswere from one category (e.g., digits) anddistractors from another category (e.g., con-sonants). In the varied mapping (VM)procedure, memory-set items and distractorswere randomly intermixed over trials andwere all from one category (e.g., consonants).

Figure 1 gives examples of trials in boththe VM and CM conditions. Depicted arefour consecutive trials from a CM block andfour from a VM block in each of whichM = 4 and F = 2. Table 1 gives the memoryset, distractor, and target (if present) foreach trial in Figure 1. Note that memory-setitems and distractors intermix across trialsin the VM condition but do not intermix overtrials in the CM condition.

The most important results are shown inFigure 2. These results showed that the VMconditions were strongly affected by load andwere quite difficult; the CM conditions werevirtually unaffected by load and were alleasier than even the easiest VM condition.It was suggested that a controlled, serialsearch was operating in the VM conditionsand that a qualitatively different process,automatic detection, was operating in theCM conditions.

The results of Part I/Experiment 1 wereanalyzed on the basis of the accuracy of thedetection responses. Part I/Experiment 2utilized comparable conditions but presented

Table 1Examples of CM and VM Trials forFour Successive Trials

TrialMemory

setDistractor

set Target

Consistent mapping (CM)1234

1234

7481258317396S82

Varied

MJDGCJKHGMCHJLKF

KGJCMCHFLDKGFDMCMJKD

mapping (VM)

CFHKLLGDFMDLFKJCDGHM

4none

78

Dnonenone

L

only a single frame on each trial; accuracywas high and the results were analyzed onthe basis of the reaction time of the responses.The results confirmed those of Part I/Experi-ment 1, and a quantitative model was fit tothe VM results of both experiments. Thismodel assumed that controlled search wasa serial, terminating comparison process inwhich one first compared all frame itemsagainst one memory-set item before switchingto the next memory-set item. Each compari-son and each switch was assumed to requiresome time to be executed. The success of themodel in fitting the results of both experi-ments suggests that the same search mech-anisms underlie search experiments that uti-lize both accuracy and reaction time mea-sures, and suggests that the same searchmechanisms underlie performance in bothdivided-attention and search paradigms.

The vast differences between results of theCM and VM conditions provided a basis forreorganizing and classifying the results ofprevious search and detection studies. Thesestudies fell into a relatively simple organiza-tion, and many perplexing and seeminglycontradictory results became explicable.

Part I/Experiment 3 utilized a multiple-target, multiple-frame procedure. The studywas similar to Part I/Experiment 1, but thesubject was presented either zero, one, or twotargets per trial and was required to reportthe number of detected targets. The condi-


c/)HIr-

LU

100

80

60

CONSISTENTMAPPINGS

40

4,2).(4,1)

(4,4)

A" (1,4)

O)

LUO

LJ

100

80

60

VARIEDMAPPINGS

40 80 120

FRAME TIME (msec)

120200 400 600 800

FRAME TIME (msec)Figure 2. Data from Experiment 1, Part I. Hits as a function of frame time for each of the 12conditions. Three frame times were utilized for each condition. The first number in parenthesesindicates the memory-set size and the second number in parenthesis indicates the frame size.

dons of greatest interest were those in whichtwo targets per trial were presented. In thesecases the spacing between the two targetsvaried from 0 to 4 (0 spacing indicates simul-taneous presentation; spacings of 1, 2, and 4indicate the number of intervening frameplus 1). Target similarity also varied, thetwo targets being either physically identical(II) or different (NI).

The results for Part I/Experiment 3showed markedly different patterns for theVM and CM conditions. Consider targetsimilarity first. In the VM conditions de-tection of identical targets (II) was superiorto detection of different targets (NI). How-ever, in the CM conditions, the reversewas true: NI detection was superior to IIdetection. Consider target spacing next. Inthe VM conditions performance was lowestwhen the targets occurred in successive frames(spacing 1). However, in the CM conditionsperformance was lowest when the targets weresimultaneous (spacing 0). These resultshelped emphasize the qualitative differencebetween the automatic detection processingmode presumed to be utilized in the CMconditions, and the controlled search modepresumed to be utilized in the VM conditions.In certain of the studies in this paper, weshall utilize this multiple-target, multiple-

frame procedure and will infer the presenceof automatic detection or controlled searchfrom the nature of the spacing effect and thetarget-similarity effect.

B. Rationale for the Experiments

We argued in Part I that the cause of thedifference between the CM and VM resultswas the consistency of the mapping (overtrials) of the memory-set items and dis-tractors to responses. We argued that con-sistent mapping leads to the development ofautomatic detection, which enables auto-matic-attention responses to become attachedto the memory-set items. The automatic-at-tention responses enable the serial, controlledsearch to be bypassed by a parallel detectionprocess unaffected by load. However, thesehypotheses must be further tested for thefollowing reasons. First, the fact that mem-ory-set items were categorically distinctfrom the distractors was confounded with theconsistency of the mapping. These factorswill be separated in Part II/Experiments 1,2, and 3. Second, the course of developmentof the hypothesized automatic detectionprocess was not studied in Part I. The courseof learning will be traced in Part II/Experi-ments 1 and 3. Third, there was no demon-


stration in Part I that an automatic-atten-tion response is learned in CM paradigms.Such a fact will be suggested by Part II /Experiments 1 to 3 and demonstrated inPart II/Experiment 4. Finally, although thesegoals provide an initial justification for thepresent experiments, these studies will servean even more important purpose in elu-cidating the characteristics and developmentof automatic processing and in differentiatingautomatic from controlled processing.

II. The Development of Automatic Processing:Perceptual Learning

A. Perceptual Learning and Unlearning in aCM Task Using Letters Only

Experiment 1 of the present series is simplein conception. With the use of the same basicmultiple-frame paradigm used in Part I/Experiment 1 (see Figure 1), performance isexamined as a function of the amount oftraining under consistent-mapping conditions.One change is made, however: Both the dis-tractor set and the memory ensemble consistof consonants. This procedure enables us tostudy the acquisition of automaticity whenthe two sets are not already-learned cate-gories. Furthermore, it had been observedinformally that the use of digits and con-sonants (in the CM conditions of Part I/Experiment 1) led to a relatively rapidacquisition of automatic detection. It wasfelt that the use of letters only to make upthe two sets would slow down acquisition.

Values of M and F were chosen so as tomake controlled processing difficult (M = 4,F — 2 ) , and a frame time (/) of 200 msecwas chosen so as to make performance lowwhen controlled processing was being utilized.(These choices are justified by the data inFigure 2.) Thus it was expected that per-formance would be quite poor at the start oftraining, when automatic detection had notbeen learned and controlled search had to beused, but would improve markedly as auto-matic detection developed.

I . Method

The CM presentation procedure of Part I/Experi-ment 1 was utilized (see Figure 1). The frame size

was always equal to 2 and the memory-set size wasalways equal to 4. Two disjoint character sets wereused for the memory ensemble and the distractorset. One consisted of the following nine consonants:B, C, D, F, G, H, J, K, and L; the other con-sisted of the following nine consonants: Q, R, S,T, V, W, X, Y, and Z.

There were four new subjects, two of each sex,all naive to our tasks.1 Two subjects began trainingwith the consonants from the second half of thealphabet 'as the memory ensemble, and two subjectsbegan with the consonants from the first half ofthe alphabet as the memory ensemble. There werefive blocks of trials per session, each containing60 test trials. There were no practice trials. Subjectswere informed at the start of the experiment con-cerning the nature of the study and the composi-tion of the memory ensemble and the distractor set.Each trial began with the presentation of a memoryset (M = 4) selected randomly from the memoryensemble.

During the first 1,500 trials of the experimentfor each subject, the frame time was 200 msec;during the following 600 trials, 120 msec. Afterthese 2,100 trials the memory ensemble and thedistractor set were switched for each subject andthe frame time was set back to 200 msec. Thisreversal 'Condition was then run at the frame timeof 200 msec for a total of 2,400 additional trials. Thesubjects were informed of the reversal at the timeof the switch.

The subjects responded whenever a target wasdetected, or gave a negative response at the end ofthe trial. The accuracy of the response and thereaction time for hits were both recorded. Sub-jects heard a tone signifying an error after eachincorrect response.

2. Results and Discussion

The results are presented in Figure 3. Re-sults for each block, averaged across sub-jects, are graphed consecutively. Thus, thegraphed points in each interval are based on120 observations.

In the initial group of 1,500 trials, the hitrate rose from just over 50% to about 90%,while the false alarm rate dropped fromabout 12% to about 3% (and the reaction

1 Since the experiments are not reported in theirchronological order, we list here the experiments intheir original order and indicate the subjects thatwere used in each. Experiment 1 was run with newsubjects. Experiment 4 was run next) using thesubjects who took part in Experiment 3 of Part I.Experiment 2 was then run on three of the foursubjects in Experiment 4. Experiment 3 was runlast, on new subjects.


100

EC/)ot 800:=

60

HO:2<LU-J

40

20

2 o

X 850o:£

INITIAL LEARNING

• REVERSED LEARNING-200 MSEC FRAME TIME120 MSEC FRAME TIME _

*tt*^*fi*.600 1200 1800 2400

I-2 650g

550UJCC

600 1200 1800TRIAL NUMBER

2400

Figure 3. Data from Experiment 1. Initial consistent-mapping learning and reversed consistent-mapping learning for target and distractor sets taken from the first and second halves of thealphabet. Memory-set size = 4J frame size = 2, frame times are shown. Percentage of hits, percentageof false alarms, and mean reaction time for hits are graphed as a function of trial number. After2,100 trials the target and distractor sets were switched with each other.

time for hits dropped from about 770 msecto about 670 msec).

It seems plausible that the subjects adopteda controlled search strategy at the start oftraining. Some support for this hypothesis isfound in a comparison of the data from thisstudy with the data from Experiment 3 ofPart I (though different subjects were in-volved in the two studies). The hit and false

alarm probabilities in the first 60 trials ofthe present study were .56 and .13, respec-tively. The estimated probability of detectinga single target and the observed probabilityof giving a false alarm when no target waspresent were .60 and .18, respectively, in theVM condition of Experiment 3b of Part I inwhich M was equal to 4 and F was equal to 2(this condition and the first 60 trials of the


present study were both run at a frame timeof 200 msec). These two conditions shouldgive similar results if controlled search isbeing utilized in both studies; in fact, theprobabilities are quite similar.

It should be noted that the subjects all re-ported extensive, attention-demanding re-hearsal of the memory set during about thefirst 600 trials of Experiment 1, but theygradually became unaware of rehearsal orother attention-demanding controlled pro-cessing after this point. Both the objectiveevidence and subjective reports, then, sug-gest that the subjects used controlled searchat the start of Experiment 1 but graduallyshifted to automatic detection.

The hit rate and false alarm rate after1,500 trials appeared to be changing onlyslowly, but this could have resulted from aceiling effect. The frame time was thereforereduced to 120 msec for the next 600 trials.At the conclusion of this additional training,the hit rate appears to have reached about82%, while the false alarm rate dropped toabout 5%. These results may be comparedwith the CM results from Experiment 1 ofPart I (see Figure 2 ) . In that study, with/ = 120 msec, M - 4, and F = 2, the CM hitrate was about 92% and the false alarm rate(not shown) was about 5%.

Thus, the present results indicate thatsubjects were utilizing automatic detectionby the end of the initial 2,100 trials of CMtraining. First, they performed at a level notfar from that of the Part I/Experiment 1subjects in the comparable CM conditionwith the same frame time. Furthermore, thepresent study utilized a larger memory en-semble and distractor set than the VM con-ditions of Part I/Experiment 1, a fact thatcould only have increased the task difficulty.Second, from the Part I/Experiment 1 VMdata we can estimate that a frame time of600-800 msec would have been required toachieve a similar level of performance hadcontrolled search been used.

At this stage of the experiment, after 2,100trials of CM training, the subjects could beexpected to have developed a well-learnedattention response to the members of thememory ensemble. If such learning is firmly

planted in long-term memory and if the at-tention response occurs automatically, thenit should prove very difficult for the subjectto alter or unlearn his automatic response inany short period of time. To test this hy-pothesis we reversed the memory ensembleand the distractor set for each subject (andset the frame time back to 200 msec). Wehypothesized that automatic detection wouldprove impossible and that the subject wouldbe forced to revert to controlled search.

The results of the reversal were quitedramatic. The hit rate just after reversalfell to a level well below that seen at thestart of training when the subjects were com-pletely unpracticed. Very gradually there-after the hit rate recovered, so that after2,400 trials of reversal training, performancereached a level about equal to that seen after1,500 trials of original training. In summary,the original training resulted in quite strongnegative transfer, rather than positivetransfer.

Subject's verbal reports indicated that ashift back to controlled search occurred afterreversal. Initially, before reversal, all sub-jects reported rehearsing the memory setduring each trial. After the 2nd day (600trials) subjects reported that they were nolonger rehearsing and only glanced at thememory set. However, subjects reported thatafter reversal they tried various methods toperform the task and eventually ended uprehearsing the memory-set items again, thoughthis rehearsal also decreased after a week ofpostreversal practice. This pattern of re-ports could indicate a shift from controlledsearch to automatic detection during originallearning, then a shift after reversal to con-trolled search, and then finally a return toautomatic detection.

What might be the cause of the negativetransfer after reversal? If the reversal causedsubjects to revert to controlled search, thenit might have been expected that perform-ance would fall to the level seen at the startof original learning (when controlled searchwas presumably utilized). The actual resultssuggest that controlled search is hinderedwhen the distractors are items that subjectshave been previously trained to respond to as


CM targets (i.e., items that, due to previoustraining, give rise to automatic-attentionresponses). This hypothesis is verified inExperiment 4, to be reported later.2

Beyond the poor performance at the startof reversal learning, negative transfer is alsoevidenced by the extremely large amount oftraining needed to overcome the effects of orig-inal learning. After about 900 trials of originalpractice the hit rate reached about 90%.However, it took about 2,100 trials to returnto this performance level after reversal. Ap-parently the necessity to unlearn the at-tention responses to the previous memory-setitems, or the necessity to overcome somelearned inhibition to the previous distractors,or both, causes a reduction in the rate ofacquisition of automatic processing afterreversal.

It would be interesting to know whethernegative transfer would obtain in severalvariations of our paradigm. For example, inthe present study, the initial training mighthave caused the memory set to become alearned category, but probably did not causecategorization of the distractor set. If thedistractor set was a well-known category,would reversal still cause a severe perform-ance drop? Furthermore, it could be arguedthat the initial training did not continue longenough for the negative transfer to reach fullstrength. The amount of negative transfermight depend on the amount of overtrainingduring initial learning. Such a view has beenput forth for certain verbal and animal learn-ing situations (see Handler, 1962, 196S;Jung, 1965 for two views on the subject).Thus it might be asked whether performancedrops would be caused by reversal if theoriginal automatic detection responses weregreatly overlearned. These two questions areanswered by Experiment 3.

B. The Reversal of Consonants and Digitsfor Well-Practiced Subjects

Throughout the series of experiments inPart I, the memory ensemble in the CMcondition was always the same, either digitsor consonants, and no member of the CMmemory ensemble was ever a distractor, evenin the VM conditions. The VM conditions

always consisted solely of items from thedistractor set in the CM conditions. Thus asubject at the conclusion of the series ofPart I studies who had been searching, say,for consonants in digit distractors, had neverbeen given a consonant as a distractor:Whenever a consonant had appeared, it wasa target. These facts naturally suggested astudy in which the roles of consonants anddigits were reversed for these subjects.

1. Method

The procedure was identical to the multiple-frame, multiple-target paradigm of Experiment 3aof Part I (whose results were described in theIntroduction) in most respects, except that thememory ensemble and the distractor set wereswitched in both the CM and VM conditions. Thusin the CM condition the two subjects who hadbeen searching for consonants in digit distractorsnow searched for digits in consonant distractors;in the VM condition these two subjects had beensearching for digits in digits and now searched forconsonants in consonants. These contingencies werereversed for the remaining two subjects who hadpreviously been searching for digits in consonantdistractors in their CM conditions.

The frame time (/) was 60 msec for the CMconditions and 200 msec for the VM conditions.Just as in Experiment 3 a of Part I, M was equalto 2, and F was equal to 2 ; zero targets were pre-sented on one fourth of the trials, one target waspresented on one fourth of the trials and twotargets were presented on one half of the trials.When two targets were presented, they were eitheridentical (II) or nonidentical (NI), and the spac-ing between them was either 0, 1, 2, or 4. Eachsubject was given 12 blocks of 132 trials in eachof the VM and CM conditions. The VM and CMblocks alternated.

Upon 'Completion of the initial set of 24 blocks,the frame time was increased to 120 msec and anadditional 18 blocks of CM trials were run for eachsubject.

2 One might hypothesize that the basic compari-son rate of controlled search is altered and slowedafter reversal, thereby accounting for the negativetransfer. Two facts argue against this hypothesis.Experiment 2 shows that controlled search proceedsat an unchanged rate even when all items consistof targets from previous CM conditions. Experiment4 shows that even one CM target presented in ato-be-ignored spatial location greatly reduces detec-tion of a simultaneously presented target in a to-be-attended location. This result suggests that reversalperformance is harmed because attention tends tobe drawn to the distractors.


It should be noted that the subjects for this ex-periment had had roughly 20,000 trials of practicein the various CM and VM conditions in earlierstudies (see Footnote 1).


The results are depicted in Figure 4. Asimple correction for guesses and false alarmshas been carried out on the raw data, asdescribed in Part I, but this correction didnot affect the qualitative features of the re-sults. The circles in the left-hand panel givethe comparable VM prereversal data fromExperiment 3a of Part I ( /= 200 msec),while the circles in the center panel give thecomparable CM prereversal data from Ex-periment 3c of Part I (/ = ,60 msec); thetriangles in the two left-most panels give thedata from the present experiment (see Foot-note 1).

The most dramatic findings of Experiment2 are shown in the middle panel of Figure 4,which gives the CM reversal results. It isevident that performance drops off markedlyafter reversal. Estimated percentage of detec-

tion of two targets is about 80% prior toreversal and about 20% after reversal. Fur-thermore, these data represent 12 blocks oftraining over which very little, if any, re-covery was taking place. It is interesting tonote that all of the subjects predicted thatno change in their performance would occurafter the consonants and digits were reversed,and the subjects were uniformly startled andeven dismayed by the extreme difficulty ofthe reversed task.

These results resolve the confounding be-tween categorization and the mapping condi-tions. Even when there is a well-learned andwell-practiced categorical difference betweenmemory ensemble and distractor set, reversalin the CM paradigm gives rise to markedimpairment of performance. Thus, categoricaldifferences between memory ensemble anddistractor set cannot be the key factor under-lying the utilization of automatic detection.Rather it is the consistency of mapping thatunderlies the acquisition of automatic detec-tion.

Consider next the VM data given in the

REVERSAL CONDITIONS

— IUU

uLU

UJOT 8001-, llJI-Ozee 60UJ< DU

a! ;J 40o<

£1-

h ' ' ' ' '

- a A

. Nf̂ .i i i i i i

°o

"A ^^°~A

^A__^*^A— Ti-

i i i i i i

DQ

i v-:

t i i i i iw 0 ' I 2 4 0 1 2 4 0 1 2 4

SPACING SPACING SPACING

0 I 0 1 2 0 1NUMBER OF TARGETS

Figure 4. Data from Experiment 2. Probabilities of correctly detecting one target when one waspresented, or two targets when two were presented, as a function of spacing, for each condition.The data shown have been adjusted to remove effects of guessing and false alarms. The observedpercentage of "none" responses to no-target trials is also shown. Circles show the estimatedpercentage of detection prior to reversal (data from Experiment 3, Part I), while triangles indicateperformance after reversal in Experiment 2. The squares in the right-hand panel show post-reversal detection from Experiment 2 after frame time was increased to 120 msec. The opensymbols and dashed lines indicate the II conditions (identical targets); the closed circles and solidlines indicate the NI conditions (different targets). (VM = varied mapping; CM = consistentmapping; f = frame time, in msec.)


left-hand panel of Figure 4. It is evident thatthe switch from consonants to digits (orvice versa for other subjects) had almostno effect. The levels of performance remainthe same as in Experiment 3a of Part I, andthe qualitative pattern of results is still aboutthe same, showing the pattern indicative ofcontrolled search. That is, worst perform-ance occurred at spacing 1, and performancewas worst when the targets were different(NI). These effects indicate the presence ofcontrolled search. In summary, a change ofthe character set used in the VM conditionshad no appreciable effect under the presentconditions.

These results are most interesting, be-cause after reversal the memory ensembleand the distractor set both consisted ofitems that had come to elicit automatic-attention responses in previous training. Inthe reversal conditions of Experiment 1 wesaw that making the distractor set but not thememory ensemble consist of such "automatic"items resulted in considerable negative trans-fer. In the present experiment it is evidentthat performance is not worse after reversalwhen subjects have been trained to re-spond to both the targets and distractors inpreviously CM training.

These results might be expected on thebasis of the following reasoning. Just afterreversal in both Experiment 1 and in the CMconditions of Experiment 2 the subjects wereusing controlled search. In Experiment 1, afterreversal, the automatic-attention responsesto the distractor items tended to draw at-tention away from target items to which sub-jects had not previously been trained to re-spond and hence to order the controlled searchin such a way that the distractor items werecompared earlier in the search. Thus per-formance was impaired. (This hypothesis isconfirmed in Experiment 4.) However, in thepresent experiment, after reversal, all items,distractors and targets alike, gave rise toattention responses, which could be expectedto cancel each other out on the average.That is, there should be no selective bias tocompare distractors prior to targets sinceboth types of items give rise to equivalentattention responses. Thus, the controlled

search operates normally, with targets com-pared in a random order of search, and per-formance is equivalent to that seen prior tothe switch of character sets.

In short, the use in a search task of stimulithat elicit automatic-attention responses (de-veloped through previous CM training) can(a) improve performance over that seen inthe usual VM conditions if these stimuli arememory-set items but not distractors, inwhich case automatic detection will be uti-lized; (b) leave performance unchangedfrom that seen in the usual VM conditions ifthese stimuli are both memory-set items anddistractor items, in which case normal con-trolled search will be utilized; and (c) im-pair performance from that seen in the usualVM conditions if these stimuli are distractorsbut not memory-set items, in which casesubjects will utilize controlled search thatwill be reduced in effectiveness by automaticattending to distractors.

Let us now turn to the results of the finalreversed CM training blocks. We suspectedthat the subjects in the reversed CM condi-tion (the middle panel) were reverting backto the use of controlled search, but perform-ance levels were so low that it was difficultto interpret the pattern of results. Thusframe time in the CM reversal condition wasincreased to 120 msec, and 18 blocks of addi-tional CM training were run.

The results of the trials with / = 120 msecare represented by the squares in the right-hand panel of Figure 4. It should be notedfirst that the pattern of the results is in-dicative of controlled search in that the worstperformance occurs at spacing 1. However,the results do not show the target-similarityeffect that we have been led to expect (inExperiment 3 of Part I) for controlled search:Results for conditions using II target pairsare not superior to those using NI targetpairs. Furthermore, the overall level of per-formance is higher than one would expectfor the VM conditions. At a frame time of120 msec performance is higher than thatseen for the VM conditions at a frame timeof 200 msec (in the left-hand panel of Fig-ure 4).

The results can be explained simply and


elegantly by the hypothesis that the sub-jects were carrying out a controlled searchfor the category as a whole. The Experi-ment 2 reversal condition differed from thePart I/Experiment 3 VM conditions in thatthe memory-set items and the distractoritems in the present instance were cate-gorically different (numbers and letters).This categorical difference obviously did notallow the subjects to utilize automatic detec-tion after reversal. However, the categoricaldifference could very well have helped thesubjects to adopt a more efficient controlledsearch. Suppose the subject ignored thespecific members of the memory set andsearched instead for any instance of the cate-gory "numbers" (or "letters" as the casemay be). Then only one comparison wouldbe needed for each display item in eachframe, regardless of memory-set size.

Compared with the case when a categoriza-tion is unavailable, a controlled search forthe category of each input saves search timein two ways. First, there is no need for mul-tiple memory-set comparisons for each frameitem. Second, there is no need for time-consuming switches between memory-setitems. Thus performance improves consider-ably. Furthermore, the distinction betweenidentical and nonidentical multiple targets isno longer meaningful—both targets are sim-ply category members, regardless of theirphysical similarity. Thus no difference be-tween the II and NI conditions would bepredicted. The results in Figure 4 supportthese contentions.

As far as we can tell, through a block-by-block analysis of the reversal results, the sub-jects did not begin to recover any appreciabledegree of automatic detection even after 30blocks of CM reversal training. Undoubtedly,the difficulty in relearning (or unlearning)is related to the great amount of overtrain-ing with the original mapping of stimuli toresponses. Eventually, however, were thereversal experiment continued, we would ex-pect automatic detection to develop onceagain.

In summary, then, the results of Experi-ment 2 show that a switch of character setsdoes not affect VM performance, but a switch

of memory ensemble and distractor set causesa great decrement in performance in the CMconditions. The CM decrement occurs despitethe categorical difference between the mem-ory-set items and distractor items (lettersvs. numbers). It is argued that the subjects,after reversal, adopt a controlled searchstrategy in which they search for the categoryas a whole rather than search for the in-dividual members of the category.

The results of Experiments 1 and 2 couldhardly have provided a more dramatic demon-stration of the qualitative difference betweenautomatic detection and controlled search,and of the dependence of the search mecha-nism on the nature of the training procedures(i.e., the consistency of the stimulus-responsemapping).

The results of Experiments 1 and 2 alsodemonstrate the long-term nature of thelearning underlying the automatic process.Experiment 1 showed that even originallearning of an automatic response could takethousands of trials of practice to develop ifa previously known categorization did notdistinguish the memory ensemble and the dis-tractor set. Relearning after reversal waseven slower to develop. Experiment 2 showedthat relearning after reversal could be veryretarded even when the sets were categorized,as long as the original automatic detectionprocess was well overlearned.

C. The Role of Categorization in ControlledSearch and in the Development

oj Automaticity

The role played by categorization in con-trolled search and in the development ofautomatic detection is suggested by resultsof the preceding experiments, but it is animportant enough concept to be examinedin a separate experiment. Furthermore, thenature of the process by which categoriesare learned is also worth studying. Theseconsiderations underlay the design of Experi-ment 3. We decided to use new subjects andto train them in two conditions, in neitherof which were the memory and distractorsets preexperimentally categorized. The firstcondition was designed to induce controlled


100

S 80

60

40

20

M=2 M=4o • Categor icala • Mixed

f = 350

0-4 5-9 10-14 15-19 20-24 ^25 28 30-32 33-35 f

V M Training CM Tra in ingS E S S I O N S

Figure S. Data from Experiment 3. Estimated percentage of detection of both targets, as a func-tion of session number, averaged over spacing and target-similarity conditions, for each of thefour main conditions. At Session 25 training was switched to a CM procedure. At Session 30, theframe time was reduced to 160 msec. (VM = varied mapping; CM = consistent mapping; M =memory-set size; f = frame time, in msec).

search in circumstances such that a cate-gorization could not develop. The second con-dition was also designed to induce controlledsearch, but in circumstances such that acategorization of the memory sets could belearned. Both of these conditions utilized avaried-mapping procedure to prevent auto-matic detection, and both were followed bya series of CM trials, during which trialsautomatic detection developed. Thus, theCM trials traced the course of developmentof automatic detection when a categoriza-tion was, or was not, present at the start ofCM training.

1. Method

The subjects each took part in two conditionscalled mixed and categorical, run in different blocks.In each condition the memory ensemble and dis-tractor set were drawn from a total of eight con-sonants. The two eight-consonant sets did not over-lap. The sets were {GMFP, CNHD} and {RVJZ,BWTX}. The assignment of these two sets to thetwo conditions was permuted across subjects.

Both conditions used a VM procedure similar incertain respects to that of Experiment 2 of thisreport and Experiment 3 of Part I. The multiple-target, multiple-frame paradigm was used with/ — 250 msec per frame. Only 12 frames were usedon each trial, with all targets appearing on frames3 through 10. Frame size was equal to 2 in all con-

ditions, and two memory-set sizes were used indifferent blocks: M = 2 and Af = 4.

In the mixed condition, the members of thememory set were chosen randomly on each trial fromthe ensemble of eight consonants; the distractor setconsisted of four items randomly chosen fromthose consonants not used in the memory set. Thesefour .distractors were chosen randomly to fill thenontarget, nonmask, positions in the various framesof the trial. The key feature of the mixed conditionwas the fact that memory-set items and distractorswere randomly intermixed from trial to trial.

In the categorical .condition, the eight ensembleitems were divided into two sets of four that re-mained disjoint throughout the experiment for eachsubject. The sets of four are indicated by the posi-tion of the comma in the above listing of the con-sonants making up each set. Note that the two setsof four were chosen to be highly confusable witheach other in the sense that pairs of visually con-fusable letters were separated into the two sets. Ona given trial one of these sets of four was chosenrandomly to be the distractor set, and the memory-set items were then chosen randomly from theremaining set. Thus, in the categorical conditionthere were just two disjoint categories of items,one from which the memory set was drawn, andone that served as the distractor set. However, thecategory that served as the distractor set variedrandomly from trial to trial. Thus, 'the procedurestill involved a varied mapping, but the categorieswere never intermixed and could eventually becomewell learned.

To make it easier for the subjects to learn thecategories, the members of the memory set pre-


sented at the start of each trial were always pre-sented in alphabetical order (this was also done inthe mixed 'condition). In each session there werefour blocks, always run in the following order:(a) categorical condition, M = 2; (b) categoricalcondition, M = 4; (c) mixed condition, M = 2; (d)mixed condition, M = 4. There were 96 trials ineach block, 24 with no target, 24 with one target,and 6 for each of the eight combinations of spacingand target similarity when two targets werepresented.

Four new subjects, one male and three females,took part. Each ran in 24 sessions of about 1 houreach. This finished the first, varied mapping, phaseof the study. After these 24 sessions the subjectsshould have been using controlled search in bothconditions, but should have learned the categoriesin the categorical condition. Before turning to theCM procedure used in subsequent sessions, we shalldiscuss the VM results.

2. Results and Discussion of the VMConditions

The results are summarized in Figures 5and 6. Figure 5 gives the estimated prob-ability of detecting both targets when twoare present, averaged across target similarity,spacing, subjects and certain combinations ofsessions. Performance in all four main con-ditions tended to improve during VM train-ing, but is clearly stable over the last 10sessions (15-24). Initially the M = 2 con-dition is superior to the M — 4 condition inboth the categorical and mixed cases.After training, however, the pattern changesconsiderably. In the categorical conditions

100

80

o>

2 60

co*-u0>

lu•a

c<uowa>a.73a>oE

II NIo — & M = 2

Categorical, VM

i i

100

oo Oa> O

-t> Oro O

Mixed, VM

0 1 0 1 2 4i Spacing of two t a r g e t s |

Number o f t a r g e t sFigure 6. Data from Experiment 3. Estimated percentage of detection as a function of number,spacing, and similarity of targets for VM Sessions 19-23 for all conditions. (VM = varied mapping;M = memory-set size; II = identical targets; NI = nonidentical targets.)


the results of the M — 2 and M = 4 condi-tions converge (indicating category learningand the use of a categorical search strategy).In the mixed condition performance in theM = 4 condition remains much lower than inthe M = 2 condition, but performance inboth of these is worse than in either cate-gorical condition (suggesting that the pres-ence of well-known categories reduces theeffective memory-set size to 1).

To ascertain whether controlled search wasbeing utilized in these various VM conditionsand to determine the nature of that search, itis necessary to consider the spacing functionsshown in Figure 6. This figure gives thespacing functions for Sessions 19-23, whenperformance had stabilized, averaged acrosssubjects. For both M = 2 and M = 4 con-ditions, the mixed condition shows the pat-tern we have come to expect for controlledsearch, with performance worse at spacing 1and worse for different targets (NI).

Performance in the categorical conditionwas equivalent when M = 2 and M — 4.The usual performance impairment at spacing1 occurs for identical targets (II), but whenthe two targets differ (NI), performance inthe spacing 0 condition is greatly impaired,a result not seen in previous VM conditions.Furthermore, the II and NI conditions showonly a small difference except at spacing 0.

The results from the mixed condition con-form to the pattern expected for controlledsearch in three respects; When M = 4 per-formance is worse than when M — 2; thereis a performance reduction at spacing 1; andperformance is worse for NI than II con-ditions.

The results from the categorical conditionwere unexpected in some respects but areexplicable in terms of an hypothesis thatcategories were learned and utilized in con-trolled search. Let us assume that the pres-ence of a known category allows one to com-pare the category of the memory set to thecategory of any given display item in a singleoperation. Then the M = 2 conditions shouldnot differ from the M — 4 conditions, and infact both conditions should show performancelevels equivalent to those expected for amemory-set size of 1 in a normal mixed con-

dition. This reasoning explains why the cate-gorical conditions are both superior to thebest mixed condition, which uses a memory-set size of 2.

The category hypothesis, however, sug-gests that target similarity should not matter,so that the II and NI functions should beidentical. To explain the spacing 0 resultswithout abandoning the category model, wesuggest the following hypothesis. Supposethat when a subject locates a target category,he or she briefly switches to an item mode,perhaps to check that the input is truly acategory member. If the second item in theframe is identical, then it will be found in anitem-comparison mode, but if nonidentical, itwill be located only if the subject revertsquickly enough to a categorical-comparisonmode. By the next frame, the reversion to acategorical mode is complete, so the variousfunctions tend to converge. Why would sub-jects tend to switch to an item mode? Thecategories were constructed so as to be ex-tremely confusable with each other. Perhapscategory encoding is learned under these cir-cumstances but remains somewhat errorprone. Then it would be logical to check anytarget category by using an item mode.

Whatever the explanation for the details ofthe spacing function, the data as a wholemake a strong case that categories have beenlearned in the categorical condition duringVM training, and that the presence of cate-gories in a VM situation allows the subjectto adopt a simpler and more efficient form ofcontrolled search.

The argument that controlled search isoperating in these conditions would be sub-stantiated if it could be demonstrated thatperformance improves when the subjectsswitched to CM training. We turn next tothe CM procedure.

3. Method

Following the 24th session of VM training, a CMprocedure was initiated. In the categorical condi-tion, one of the categories was fixed thenceforthas the memory ensemble, and the other categorywas fixed thereafter as the distractor set. In themixed condition, a set of 4 was chosen and wasfixed thereafter as the memory ensemble—the re-maining 4 items were fixed thereafter as the dis-


tractor set. The two sets of 4 used in the mixedcondition were those indicated by the position ofthe comma in the listing of the two sets of eightconsonants: {GMFP, CNHD} and (RVJZ, BWTX}.In other words, these were the same sets that hadbeen used as categories in the categorical conditionsfor other subjects. This CM training procedure wasidentical for both conditions and followed in otherrespects the procedures used in the preceding VMconditions. Each subject was given at least 10sessions of VM training. The frame time after CMSession 4 (Session 24 overall) was reduced to 160msec and after CM Session 11 (Session 35 overall)was .reduced again to 120 msec (at the time of thiswriting this study had not been completed).

4. Results and Discussion of the CMConditions

The changes over sessions following theswitch to CM training are depicted in Figure5. Consider first the results of Sessions 25through 28, which immediately followed theswitch to CM training. Quite clearly, a veryrapid and dramatic rise in performance tookplace, and furthermore, the results of themixed and categorical conditions tend toconverge. By Session 28 the performancelevel in the mixed condition had risen to over90% and in the categorical condition hadrisen to over 95%. Since the subjects wereclearly approaching ceiling, the frame timewas then reduced to 160 msec and CM train-ing continued for six additional sessions.Note that performance was still improvingand that the mixed and categorical, and M= 2 and M = 4, conditions effectively con-verged.

It is interesting to note that during VMtraining, 20-25 sessions were necessary beforethe M = 2 and M = 4 performance levels be-came equal in the categorical condition. Thusone might tentatively conclude that categoryencoding in the present experimental contextrequired 20-25 sessions to develop. Thuscategory encoding would be unlikely to de-velop for the mixed condition in just fourCM sessions. Yet after only four sessions ofCM training, the performance level in themixed condition rose dramatically and ap-proached that of the categorical condition. Thisresult suggests that automatic detection de-veloped in the mixed condition in the absenceof a well-learned category (at least during

the first four sessions of CM training). Inother words, the learning of a category isapparently not a necessary prerequisite forthe development of automatic detection.

Figure 7 shows the spacing functions forthe six sessions (30-35) of CM training runat a 160-msec frame time. The results areaveraged over subjects and sessions. In ad-dition the M — 2 and M = 4 conditions arelumped together since they did not differ.Somewhat to our surprise the pattern is al-most identical to that seen for the categoricalcondition during the latter stages of VMtraining, except that the performance levelis much higher. The comparable pattern fromthe CM conditions of Part I/Experiment 3was quite different: Detection was aboutequal in all conditions except for a depresseddetection rate when spacing was 0 and whenthe targets were identical. At spacing 0 thepresent data clearly show performance to behigher when the targets were identical.

Disregarding the shape of the spacingfunctions, the large improvement in perform-ance when CM training commenced does sug-gest that the subjects were learning auto-matic detection. An hypothesis that reconcilesthese facts suggests that the subjects wereusing automatic detection to locate the firsttarget, and were then reverting to controlledsearch to check the accuracy of the targetdetection. Some time may be lost before thesubjects revert to automatic detection. Thusthe advantage of II at spacing 0 would bedue to the temporary use of a controlledsearch. This hypothesis is similar to thatused to explain the categorical results in theVM conditions (Figure 6). In both cases, atendency to recheck the first detected itemmay have led to an alteration in searchstrategy. The reason may be the same in bothcases: Rechecking of located targets mayhave been induced by the stimulus sets,which were chosen so that the memory setand distractor set were maximally visuallyconfusable. Thus, both automatic detectionand category encoding may have been some-what error prone, thereby requiring recheck-ing in an item mode. We are currently ex-ploring this hypothesis.

In summary, a number of conclusions can


I/) 1 \J\J

«- **c .

i- 80d 0

• S o

Es

tim

ate

de

tect

ion

-P*

cr>O

O

g

-

-II N I

a — •*

i0

° ^x ^p=Z^^ -

.^.^^^.~-^"m

Categorical ~Mixed CM, f = 160MSEC

i i i i i1 0 1 2 4

i Spac ing of two t a r g e t s t

N umber of t a r g e t sFigure 7. Data from Experiment 3. Estimated percentage of detection as a function of number,spacing, and similarity of targets for CM Sessions 30-35, averaged over memory-set size (whichhad no effect) for all conditions. (CM — consistent mapping; f = frame time; II = identical targets;NI = nonidentical targets.)

be drawn from the results of Experiment 3.First, arbitrary collections of characters canbe learned as categories. For this learning tooccur, it is necessary that the categories re-main consistently denned across trials, butnot necessary that the same category re-main the target set across trials. Second, thelearning of a category in a VM search para-digm enables the subject to adopt a moreefficient controlled search, one in which thecategory of the memory set may be com-pared in a single operation to the categoryof a display item, thus eliminating the effectsof variations in memory-set size. Third, aswitch to a CM training procedure resultsin fairly rapid acquisition of automatic de-tection, with a concomitant improvement inperformance for both the categorical and non-categorical stimuli. These conclusions sup-port those suggested by the results of Ex-periment 2 in all important respects.

D. What is a Category? A Discussion andSelective Review

1. Some Hypotheses Concerning Categories

Experiments 2 and 3 showed how searchbenefits from the learning of a categoricaldistinction between memory ensemble anddistractor set. But what is a category? Mostgenerally, we may define any object in mem-ory that refers to (stands for, consists of)any two or more objects in memory as a

category. Verbal labels are one common classof categories, but other types of categoriesmight also exist. For example, a visual symbolmight represent several objects. Of course, acategory in general need not exist in a formsimilar to any of the sensory modalities. Acompletely abstract node could represent twoor more objects in memory.

With respect to search tasks, the import-ance of categories is clear. When all of themembers of a memory set are members ofthe same category and no distractors are inthat category, then a controlled search canbypass the individual memory-set items andutilize comparisons that involve matching thesingle category against the category of eachdisplay item. For category search to be uti-lized effectively, however, the category mustbe learned well enough that each displayedelement will be encoded automatically andconsistently according to its category. Ofcourse other features, such as the element'sname and shape, will also be encoded, butonly the category feature is necessary for acategory search. We suggest that the categorycoding must be automatic, because if it werenot, a search of long-term memory wouldhave to be carried out to identify the cate-gory; the time taken for such a long-termsearch would almost certainly wipe out anygains that are due to the reduction in num-ber of comparisons allowed by the categorysearch (at least for small memory-set sizes).


Note that automatic category encoding isnot the same as automatic detection. Auto-matic detection refers to the case when astimulus gives rise to an automatic-attentionresponse that bypasses the need for a serialsearch through either memory or the display.Automatic category encoding refers to thecase when the stimulus gives rise auto-matically to a node representing the categoryof the stimulus. However, without an at-tention response attached to the stimulusnode or the category node, the presence ofthe category would have to be deducedthrough a serial search of the displayedelements.

Our studies have shown that an automaticcategory encoding of arbitrary collections ofcharacters (consonants, in the case of Experi-ment 3) can develop. The cause of the learn-ing is not yet clear. Some subjects in earlyversions of Experiment 3 never showed anyevidence of categorical search (usually sub-jects with low levels of performance). It maybe that these subjects did not notice thecategorical nature of the memory ensembleand hence could not develop a well-learnedcategory node in memory. Alternatively,these subjects could have developed an ap-propriate category node but have failed touse this node to facilitate their search.

There are several lines of evidence sug-gesting that automatic detection may developfaster if a categorization is available at thestart of consistent training. For example, itis easy to learn to search for a set of stimulidefined by a simple physical feature (seeNeisser, 1963, 1967, and studies in the nextsection). The CM results in Figure 5 show thatthe categorical condition retains some superi-ority over the mixed condition for about eightsessions (25-32) but the magnitude of thedifference is surprisingly small. It is possiblethat the high confusability between the twocategories of Experiment 3 makes the cate-gories and the automatic response difficult tolearn. If so, more training prior to the CMphase could possibly have led to a larger dif-ference in the rate of development for thetwo groups. Alternatively (or additionally)the small size of these categories may haveallowed automatic responses to the individual

members to develop so quickly that advan-tages due to a category response were mini-mized.

It is conceptually important to keep inmind the possibility that automatic responsesmight be attached to different stages of en-coding of a single stimulus. In particular,when an automatic-attention response de-velops for a category, other attention re-sponses may develop at a different rate forthe stimuli making up that category. Thus,if a categorization is available at the start ofCM training, the attention responses to thecategory might develop sooner than atten-tion responses to some or all of the individualstimuli in the category. This effect would beexpected since any one stimulus would appearonly occasionally as a target, whereas everytarget would be an instance of the category.On the other hand, in the cases in which thememory ensemble does not form a categoryat the start of CM training, there might beat least some individual stimuli that cometo elicit automatic-attention responses whilethe category is still being learned.

Once an automatic-attention or encodingresponse develops, we assume that it is nolonger under control of the subject and willoccur whenever its corresponding stimulus ispresented. This assumption will be supportedby Experiment 4. However, when a cate-gorial encoding (but not an attention re-sponse) is learned, it will not necessarily beutilized by the subject during a controlledsearch. We suppose a category response willfacilitate controlled search only if the subjectboth notices the category and also decides toalter his search to compare the categoryrather than the individual stimuli. Theseconsiderations would no longer apply if anautomatic-attention response were learned inresponse to the category; in such a case sub-ject strategy would not matter since attentionwould be directed to the category in anyevent.

2. Search Studies Using Categories

In a number of memory-search studies (inthese studies F = 1 and M varies), the mem-ory set consists of several categories, and


the distractors might be drawn from thesesame categories or from different (unknown)categories. The results of these studies alldemonstrate the use of some sort of con-trolled search, perhaps in conjunction withsimultaneous automatic detection, that isfacilitated by the presence of categories.

For example, Naus, Glucksberg, and Orn-stein (1972) and Naus (1974) used memorysets consisting of words from several cate-gories and a distractor set made up of wordsfrom these same categories. The resultsshowed linear memory-set-size functions,parallel for negative and positive displayitems. However, there was a slope reductionwhen the number of categories increased. Theauthors suggested that categories were suc-cessively searched in random order, each inserial, exhaustive fashion, until the categoryof the displayed item was reached. When thesearch of this category was completed, aresponse was initiated. Homa (1973) pre-sented a related paradigm; his results sug-gested that a serial, exhaustive search ofcategories was followed by a serial, exhaustivesearch within the category of the displayeditem. Williams (1971), who presented an-other related paradigm, also argued that aserial, exhaustive search of categories wasfollowed by a serial search within the cate-gory. Okada and Burrows (1973) used mul-tiple categories in the memory set and some-times cued the subject as to the category ofthe displayed item in advance of the trial.Their results showed that the search couldbe restricted to the cued category. Cliftonand Gutschera (1971) used memory sets con-sisting of two-digit numbers and showed thaton some trials the subjects would first com-pare the 10s digits and then would comparethe Is digit only if a 10s digit match wasfound. Lively and Sanford (1972) used amemory set from one category and a dis-tractor set consisting of some members ofthat category and other members outsidethat category. Negative items outside thecategory of the memory set showed a set-sizefunction with reduced slope.

The various findings in these memory-search studies differ in many details that willnot be discussed here. They all tend to show

the use of some sort of controlled search thatis facilitated by the presence of categories inthe memory set.3

The final studies to be considered are thosevisual search studies in which M = 1 and Fvaries, but in which the visually presenteditems fall in two or more categories, one ofwhich matches the category of the memory-

3 We shall not attempt at this time to explainwhy a particular controlled search strategy is adoptedin a particular paradigm. Procedural differencesamong the studies would make such explanationspure guesswork. Another class of studies usingcategories also deserves mention for completeness,but is not discussed in the main text because of adifficulty in ascertaining whether the effects foundin these studies were due to automatic or controlledprocessing. These are visual search studies (M = 1,F varied) that have used memory ensembles anddistractor sets from different categories. They haveshown either flat set-size functions or curvilinear set-size functions with reduced slopes (e.g., Brand,1971; Ingling, 1972; Jonides & Gleitman, 1972,1976). These studies have utilized CM designs withlow practice levels; this makes it difficult to ascertainwhether the slope reductions are due to the presenceof automatic detection, to beneficial effects of cate-gorization on controlled search, or to some combina-tion of these factors.

Note that the slope reductions for visual set-sizefunctions normally should be taken as evidence ofautomatic detection. That is, in a controlled searchone would expect a serial comparison process tobe needed to compare the memorized categoryagainst the category of each displayed item. How-ever, it was seen in our VM 'data from Part I/Experiment 2 that a reduction of slope occurredwhen M was equal to 1 and F varied, and we sug-gested that a "controlled parallel search" might beresponsible for this result. A similar argument mightbe made to explain the slope reductions in the abovevisual search studies. We do not particularly favorsuch an explanation because it does not explainwhy normal set-size functions are found in visualsearch studies that do not use categories (e.g.,Atkinson, Holmbren, & Juola, 1969).

A finding related to those from studies usingcategories in visual search is that of De Rosa andMorin (1970) in a memory search study. Theyshowed that when the memory set consisted ofconsecutive digits, then both the positive and nega-tive reaction times increased as a function of thenumerical closeness of the displayed item to theboundaries of the consecutive memory set. The CMprocedure was used with small amounts of practice,so it is again difficult to ascertain whether sucheffects were due to the action of automatic detectionor to some controlled search facilitation allowed bythe categorical character of the memory set.


set item. Smith (1962) and Green and Ander-son (1956) carried out similar studies in whicha field of colored two-digit numbers was pre-sented and the subject was asked to searchfor a particular two-digit number in a par-ticular color. The results showed that searchrate depended much more on the number ofitems in the designated color than on thetotal number of items presented, though botheffects were present. A VM procedure wasutilized in their studies so that automaticdetection could not have been used to restrictsearch to the items of a particular color.Probably a two-phase controlled search wasutilized in this situation. First, a rather fastserial search was probably carried out tolocate the next item of the designated color;then a slow comparison of that two-digitnumber was probably made to determinewhether it was the target. This explanation issupported by the fact that searching took along time (up to 20 sec), so that a 30-40 msecsearch rate for color could have accountedfor the increases in response time as totalsize increased. These increases were smallonly in comparison with the larger within-category effects.

The main conclusion to be drawn from thestudies reported in this section, based on re-peated findings, is that the presence of cate-gories in search tasks can be used to facilitate,benefit, and modify controlled search.

III. Experiments on Focused Attention

In part I we made the case that theprocesses utilized in attention experiments andthose utilized in search and detection experi-ments are often the same. In fact, in manycases it is purely arbitrary whether a givenstudy is referred to as an "attention,""search," or "detection" study. Experiment 1in Part I could be described as a divided-attention study in which attention had tobe divided among M memory-set items andF frame items during each frame. The resultsshowed tremendous deficits in dividing at-tention in the VM conditions, and virtuallyno deficit in dividing attention in the CMconditions. The results of Experiment 2 ofPart I, and the fit of the quantitative modelto both studies, showed that divided-atten-

tion deficits in these paradigms are due tothe limitations of controlled processes, inparticular, to the limited rate of short-termsearch.

Thus, all our comments and conclusionsconcerning search and detection apply equallywell to attention studies. In particular, CMtraining should lead to the development ofautomatic detection that shoulld bypassdivided-attention limitations, while VMtraining should cause controlled search thatshould severely limit the ability to divideattention.

In discussing the development of automaticdetection we have proposed that an auto-matic-attention response is learned in re-sponse to the unchanging members of thememory ensemble. The present study willtest this hypothesis. The test (Experiment4d) will entail asking the subject to ignorecertain locations and then inserting in thoselocations items that subjects had been pre-viously trained to respond to as CM targets(to see whether these targets attract atten-tion).

These studies will also answer the followingquestion: To what degree can the subjectfocus attention on a specified subset of theinputs without distraction from the remain-ing (irrelevant) inputs. Such studies areusually termed focused-attention studies, incontrast to the studies of Part I and Experi-ments 1 to 3 of the present article, studiesthat would be appropriately termed divided-attention studies,

A. Terminology

There is a problem of terminology in thestudies of this section that is best solved bythe introduction of the following definitions:

1. A foil refers to any input that appears ina to-be-ignored display location, whereas adistractor refers to a nontarget that appearsin to-be-attended display location.

2. A CM foil is a foil that has previouslybeen used in CM training as a memory-setitem.

3. A CM target foil is a CM foil that wouldhave been a target requiring a positive re-sponse if it had appeared in a to-be-attended


100

eo

40

20

F-2F-4/diagonalF-4

M ~ 0 I I 0 I 2 4 |I SPACING OF 2 TARGETS |

NUMBER OF TARGETS

Figure 8. Data from Experiment 4a. Estimated per-centage of detection as a function of number andspacing of targets in a VM procedure with frame timeequal to 200 msec. The spacing 0 points are non-identical targets, while the spacings of 1, 2, and 4 areidentical targets. (F = frame size.)

display location (although it appears in factin a to-be-ignored display location). It is amember of the current memory set.

4. A VM foil is a foil that has previouslyserved as both a target and distractor in VMconditions.

5. A VM target foil is a VM foil that wouldhave been a target requiring a positive re-sponse if it had appeared in a to-be-attendeddisplay location. It is a member of the cur-rent memory set.

6. Valid positions or characters are to-be-attended positions or characters. Invalidpositions or characters are to-be-ignored posi-tions or characters.

7. FII denotes a trial in which two identicalmemory-set items appear, one in a to-be-at-tended display position, and one in a to-be-ignored display position.

8. FNI denotes a trial in which two dif-ferent memory-set items appear, one in a to-be-attended display position, and one in ato-be-ignored position.

B. Focusing Attention in a VM Multiple-Frame Task

The first study in the present series ofexperiments is designed to show that "con-trolled search" is not a misnomer, that sub-jects can control their search to the extentthat VM foils can be ignored, and that searchcan be carried out through the valid display

positions without decrement caused by thefoils.

1. Met hod

The paradigm utilizes the multiple-target, multiple-frame VM procedure. There are three main con-ditions: (a) M = 2, F = 2; (b) M = 2, F = 4; (c)M = 2, F = 4, diagonal. Condition (c), denoted"diagonal," was designed so that one of the di-agonals of the display was always valid (to-be-attended), and the other diagonal was always in-valid (to-be-ignored). In this condition four char-acters were presented on each frame, two valid,and two invalid foils. We expected that this diagonalcondition would elicit performance similar to thatfor the M = 2, F = 2 condition, and better thanthat of the M = 2, F = 4 condition.

The four subjects in this study were the sameas those used as in Experiment 3 of Part I. Inthe diagonal condition, only the upper-left andlower-right frame positions ever contained a target.The subjects were fully instructed regarding thisfact and were instructed to ignore the invaliddiagonal. For each subject, the blocks of trialsutilizing the diagonal procedure were run after theother conditions were completed. The M = 2, F = 2condition had 14 blocks per subject; the M = 2,F = 4 condition had 11 blocks per subject; thediagonal condition had 16 blocks per subject, thefirst 4 of which were practice blocks. There were120 test trials per block, plus an additional 30trials of practice for the first block of a session and15 trials of practice for each subsequent block ina session.

The procedure used in this study differed some-what from that of Part I/Experiment 3 and fromthe other multiple-target tasks reported in thisarticle. The primary difference lay in the relationof target similarity to spacing. The spacing 0 con-dition utilized NI targets only, and the spacing 1,2, and 4 conditions utilized II targets only. Inaddition, the present experiment allowed targets toreoccur in the same frame positions if they were notin successive frames. Also, in the present study, thesubjects pushed a single response button each timethey thought they detected the target. The responseswere to be made when the targets appeared ratherthan at the trial's end.


On about 1% of the trials in each conditiona response was made before the target ap-peared, and these responses were not counted.The results are presented in Figure 8. Theresults of the F = 2 and diagonal conditionsobviously do not differ, but results of bothconditions are clearly superior to those of theF = 4 condition. These results were expectedon the supposition that the subject should


have been able to control his processing inthe diagonal condition so that comparisonswould occur only for the characters on thevalid diagonal.

The peculiar relationship of spacing totarget similarity in this study caused theshape of the spacing functions to appear todiffer from those in previous multiple-targetstudies. In fact, however, the correspondingpoints from the present study and Part I/Experiment 3 are quite close in value. It isinteresting to note the subject's commentsat the end of the experiment. They noticedan excess number of II conditions overall, butnone noticed that the II pairs were not testedat spacing 0, nor that the NI pairs were nottested at the longer spacings.

The implications of this study are straight-forward. Subjects can control their search inVM situations at least to the degree thatcomparisons can be limited to a specifieddiagonal. It is possible that characters on theinvalid diagonal are sometimes compared, butnot until the valid diagonal is searched first(if time were taken to compare foils duringthe search of the valid diagonal, then per-formance would be worse). Of course, thisstudy demonstrates only a minimum amountof subject control. In future studies it wouldbe desirable to explore this matter morethoroughly. For example, we might ask:Can search alternate between diagonals insuccessive frames? Can search order be cuedindividually for each frame?

C. Distraction Caused by "Targets"During VM Search

Experiment 4a showed no distracting effectof VM foils (on the invalid diagonal). How-ever, none of these foils was in fact identicalto any member of the memory set. Thepresent study, then, is designed to determinethe distracting effect of VM target foils: Towhat extent can members of the memory setbe ignored when they appear in invalid dis-play locations?

1. Method

The paradigm is similar in general outline tothat of the VM conditions of Part I/Experiment 3.

gHOUJHUJ0

1 —UJ

cc\-

\-UJ

o:UJ

IUU

90

80

70

en

-

*v 0 Q 0

N. /

^V o — o FII•—•FNI

i i i i-I +1 +4

NUMBER OF FRAMES FROM TARGETTO FOIL

Figure 9. Data from Experiment 4b. FII = targetfoil identical to target; FNI = target foil nonidenticalto target. Percentage of target detection as a functionof the spacing and the similarity between the targetand the target foil. Varied-mapping procedures wereused, and frame time was 200 msec.

The peculiarities of Experiment 4a were eliminated.The multiple-frame procedure was utilized withM = 2 and F = 4. The upper-left, lower-rightdiagonal was valid, and the other diagonal wasinvalid. Either zero or one targets appeared onthe valid diagonal, and an appropriate binary re-sponse was required. A VM task was utilized onthe valid diagonal, and the foils on the invaliddiagonal were chosen from the distractor set for thevalid diagonal. In addition, there was on everytrial exactly one VM target foil (i.e., a member ofthe memory set) on the invalid diagonal.

On one-third of the trials, only a VM target foilappeared, always on frames 8-13. On two-thirds ofthe trials both a target foil and target appeared;on one-half of these trials the target and targetfoil were identical (FII), and on one-half of thesetrials the target and target foil were different (FNI).Finally, the target appeared on any of frames 8-13,and the target foil appeared equally often on frames—4, —1, +1, and +4 with respect to the targetframe. That is, the spacing between the target andtarget foil was systematically varied. Thus, thetarget-foil-only trials occurred with a probabilityof J, and each of the other eight conditions occurredwith a probability of 8 X i = A. All these trials wererandomly intermixed within each block.

Each block contained 144 test trials preceded by15 practice trials. Each subject ran through a totalof 12 or 13 blocks in four sessions.


The results are shown in Figure 9. Thefigure gives the percentage of hits as a func-tion of the spacing between the target and


the target foil, when the target foil is iden-tical (FIX) and nonidentical (FNI) to thetarget. The correct rejection rate when notarget was present was 91%. The results maybe summarized briefly: The target foil hadno effect on the hit rate except when it wasdifferent from the target (FNI) and precededthe target by 1 frame, in which case detec-tion probability was decreased by .11.

In the present study a target foil reducedperformance when it preceded, but not whenit followed, the target, Suppose that this re-duction is due to the same factors that pro-duce decrements in multiple-target detectionin VM conditions (Experiment 3, Part I).Then one target probably inhibits detectionof a subsequent target, but not of an anteced-ent target. That is, the reduced detection atspacing 1 seen in the VM conditions of PartI/Experiment 3 and also seen in Figure 4and 6 of the present paper, could have beencaused by either of the two targets' reducingdetection of the other. The present resultsprovide some reason to argue that it is thefirst of two successive targets that reducesdetection of the second.

Let us assume that the decrement in per-centage of detection of two targets in VMparadigms is in fact a decrement in detectionof the second target. Then it is possible tocompare the magnitude of the decrements inthe multiple-target studies (e.g., Experi-ment 3, Part I) and the present study. Infact, the decrements are much larger in theearlier studies in which all display locationswere valid. For example, the difference indetection between spacing 1 and spacing 4in Part I/Experiment 3 was 30% in the NIcondition and 8% in the II condition; thecomparable decrements in the present studyare \\% and 0%. Thus a target foil reducesdetection of a target in the next frame, butthe reduction is much smaller than thatcaused by an actual target.

How can the present findings be reconciledwith those of Experiment 4a in which foilsdid not impair performance? The pattern offindings would be explicable if the validdiagonal were always searched first and then,whenever that search finished early, one ormore characters on the invalid diagonal were

inadvertently checked in addition. In thisevent, target foils would occasionally benoticed and might harm subsequent detectionin a fashion similar to that caused by truetarget detection. Of course, this hypothesisis speculative and other explanations areundoubtedly available.

Whatever the explanations for the detailsof the results, it is safe to summarize theresults of Experiments 4a and 4b as follows:Subjects are able to control their search invaried-mapping paradigms. The degree oftheir control is sufficient to eliminate anydistracting effect of nontarget foils and toreduce the distracting effect of target foilswell below the level caused by other targets.Thus, attention focusing is quite successful:VM foils have, at most, a small distractingeffect when they appear in to-be-ignoredlocations during controlled search.

D. The Distracting Effect of "Targets"During CM Search

Experiments 4a and 4b examined theability to focus attention during controlledsearch. We next ask: To what degree doesattention focusing affect automatic detec-tion? Our previous studies have demonstratedthat automatic detection is not affected byframe size, so there would be no point incarrying out a CM counterpart of Experiment4a in which the invalid diagonal would con-tain distractors only. We decided to carry outa counterpart to Experiment 4b to find outwhether a CM target foil would interferewith automatic target detection on the validdiagonal.

M and F were set equal to 2. During eachframe, each diagonal contained one mask andone character. One diagonal was alwaysvalid, the other invalid. On every trial a CMtarget foil appeared on the invalid diagonalin exactly one frame. A CM target appearedon the valid diagonal on two-thirds of thetrials; on one-half of these trials the targetand target foil were identical (FII) and onone-half, nonidentical (FNI). In each ofthese cases, the target foil occurred eitherfour frames before, in the same frame with,or four frames after the target (-J probability


each). There were 162 trials per block and 6blocks per each 1-hour session. Two sessionswere run with frame time equal to 60 msecand two sessions were then run with frametime equal to 30 msec.

The results are shown in Table 2. Theonly significant effect was a slight drop inperformance (about 4%) in the FII conditionat / = 60 msec with simultaneous target andfoil (0 = 4.45, p < . Q O Q l ) ; performancedropped considerably when / dropped to30 msec, but all conditions became roughlyequal.

The results of this study demonstrate thata target foil hinders detection of an identicalsimultaneous target. The magnitude of theeffect is small, but so is the magnitude ofthe decrement that occurs when two simul-taneous CM targets are presented. In factthese decrements are not much different atthe equivalent frame times (4% vs. &%).It seems reasonable to assume that the samemechanism is causing both effects.4

It is interesting to compare these findingsto those of Eriksen and Eriksen (1974). Inour terminology, they used a single-frameCM task with M — 2, F = 1, and they usedreaction time as the dependent measure.However, the distractor set and the memoryensemble were both of size 2 and were con-sistently mapped across trials, so that anequally strong but opposite tendency to auto-matically respond to the members of the twosets was probably learned. (In our CMstudies, only the memory-set items can cometo elicit automatic responses, because dis-tractors are presented on every trial. Theonly exception in our studies occurred whenF = 1 in the single-frame paradigm of Part I,and even then the distractor set was alwayslarger than the memory set.) In the Eriksenand Eriksen study, the relevant item alwaysappeared directly above the fixation point,but three invalid items were placed on eachside of the relevant item. The nature anddistance of these invalid items were varied.When the distance to the nearest item reached1 ° of visual angle, the invalid characters hadlittle effect on reaction time. At closer dis-tances, all reaction times were slowed, butflanking distractors produced the greatest

Table 2Effect of Distraction on A utomaticDetection: Experiment 4c

Variable

Session 160-msec frame time

FIIFNIFIIFNIFIIFNI


FIIFNIFIIFNIFIIFNI


FIIFNIFIIFNIFIIFNI


FIIFNIFIIFNIFIIFNI

Spacing(targetto foil)

-4-4

00

+4+4

-4-4

00

+4+4

-4-4

00

+4+4

-4-4

00

+4+4

% hits

91.994.289.495.696.195. 1

95.493.590.793.892.895.6

75.779.174.577.677.180.1

77.378.980.381.078.778.2

% correctrejections

90.0

91.2

75.2

71.5

Note. FII = trial in which two identical memory-set itemsappear, one in a to-be-attended display position, and one in ato-be-ignored display position. FNI •= trial in which two dif-ferent memory-set items appear, one in a to-be-attended posi-tion, and one in a to-be-ignored position.

slowing (since they automatically produced acompeting response); flanking charactersthat were not distractors or memory-setitems produced a moderate slowing (regard-

4 The results in Table 2 give some indicationsthat the disrupting effect may change with practice.Over the four sessions, the differences between FIIand FNI at spacing 0 was, respectively, 6.2%, 3.1%,3.1%, and .7%. If this trend actually exists, it maybe caused by the development of a new automaticresponse that restricts search to the valid diagonal.That is, position-specific information might governthe automatic-attention response; such an auto-matic process could be learned because the validdiagonal never changes over trials. It would beinteresting in future investigations to compare per-formance in this condition with performance in acondition that alternates the valid diagonal fromtrial to trial, since alternation should prevent thelong-term learning of position-specific encoding.


90

oLUr~ 80LUQ

cr

LUoOLLU0.

70

60

50

NO FOIL-

-I 0 +1

NUMBER OF FRAMES FROMTARGET TO FOIL

Figure 10. Data from Experiment 4d. Percentage oftarget detection in a varied-mapping procedure as afunction of the spacing between the target and thetarget foil, when subjects had previously been trainedto respond to the foil as a consistent-mapping target.Frame time was 200 msec.

less of their visual similarity to either set),and flanking targets, whether identical ornonidentical, produced the least slowing(since they automatically produced a com-patible response).

Both the Eriksen and Eriksen (1974)study and our study suggest that CM targetfoils can neither be excluded from processingnor prevented from affecting performance. Inaddition, however, the Eriksen and Eriksenresult suggests that the spatial configurationof the inputs can determine the magnitudeof these effects (see Footnote 4 for a discus-sion of location-specific automatic detection).

E. The Distraction of Controlled Searchby Automatic Detection

Controlled search depends on an apparentlyserial process that should easily be disturbedif an automatic-attention response occursthat draws attention to an invalid location.Such a rationale suggests a paradigm forour next study in which CM foils appear onthe invalid diagonal while a controlled searchoccurs on the valid diagonal.

This study is probably the most importantin this series (4a-4d) because it provides a

test of the hypothesis that an automatic-attention response to consistently trainedtargets is learned.

1. Method

Each frame contained 4 characters. The upper-left to lower-right diagonal was valid. Memory-setsize was 2, and the VM conditions were utilized.Search on the valid diagonal was for consonantsin consonants (or digits in digits for other sub-jects). The foils on the invalid diagonal werechosen from the distractor set used for the validdiagonal, except for the CM foil. When a CM foilappeared it was chosen from the set that had servedas the CM memory ensemble in all previous studiesfor that subject. Thus, if consonants were beingused on the valid diagonal, a CM foil would be adigit, and vice versa.

On one-sixth of the trials, neither a target nor aCM foil appeared; on one-sixth of the trials onlya target appeared; on one-sixth of the trials onlya CM foil and no target appeared; on one-half ofthe trials both a target and a CM foil appeared,with the spacing between them equally likely tobe —1, 0, +1. Each block of trials contained 144test trials. The first block of each session had ISpractice trials and the other three blocks had 5practice trials. Two sessions were run for each ofthe four subjects (the same subjects as those usedin Experiments 4a-4c.) The frame time was 200msec.


The results are shown in Figure 10. Eachpercentage is based on 7,68 observations.When no CM foil or target appeared thefalse alarm rate was 4%. When only a CMfoil appeared the fajse alarm rate was again4%. Thus, in the absence of an actual targetthe presence of a CM foil caused no increasein false alarms. Since the appearance of CMfoils was highly correlated with the ap-pearance of targets, one might have supposedthat a bias or guessing strategy would arise,such that subjects would more often guessthat targets were present if a CM foil wasseen. The present data argue against such abias (as do the data discussed below).

The hit rate when a target but no CM foilwas present was 84%. The distraction causedby a CM foil may be determined throughcomparison with this performance level.When the CM foil preceded the target by oneframe the hit rate was &2<fc, not significantly


lower than the 84% base rate. When thetarget and CM foil were presented simul-taneously, the hit rate dropped to 62%. Andwhen the CM foil followed the target by oneframe the hit rate rose to 77%, a figure stillsignificantly lower statistically than the 84%hit rate when no CM foil was present. Inshort, a CM foil provides little hindrance tothe detection of a VM target presented inthe following frame, greatly hinders detectionof a VM target presented simultaneously, andeven reduces slightly the detection of a VMtarget presented in the preceding frame.

The simplest interpretation of these re-sults is that a CM foil interrupts the ongoingcontrolled search and causes a loss of process-ing time, thereby reducing target detection.The interruption is caused by what we havetermed an automatic-attention response tothe CM foil. We have argued that subjectshave been trained to respond to CM targetswith an automatic-attention response. Evenwhen such a target appears on a to-be-ignoreddisplay diagonal, it apparently causes an at-tention response that interrupts processingalong the valid diagonal and directs attentionto the invalid diagonal. The time lost beforeattention is returned to the valid diagonaland the search is resumed causes a consider-able decrement in performance if the targetis in fact on the valid diagonal during thatframe.

Since target detection in a frame followinga CM foil is not hindered, the recovery fromthe distraction caused by a CM foil must bequite rapid; that is, recovery must take placein a time period under 200 msec. Further-more, the automatic-attention response mustoccur fairly rapidly, since a CM foil evenreduces detection of a target in the precedingframe. Perhaps the processing of one frameoccasionally overlaps the start of the next;for example, a comparison initiated but notcompleted before termination of a frame maybe completed before controlled processing ofthe next frame begins. Occasionally, a targetmight be the character undergoing compari-son when the subsequent frame begins, anda CM foil in that next frame could interruptcomparison of the target and thereby impairperformance.

The primary finding of the present studyis the demonstration that the responses toCM targets are both well learned and auto-matic. CM targets cannot be ignored, evenwhen they are known to be irrelevant andoccur in consistently invalid display loca-tions and even when subjects are instructedto ignore them.

Our results, of course, do not establishthat it is impossible to control the detectionprocesses that we have labeled "automatic."In fact, Sperling (Note 1, Note 2) has re-ported some findings that do tend to showthat at least some amount of control oversome aspects of the search process in CMmultiple-frame tasks is possible. However,the degree of control over, and the ability toignore, stimuli that subjects have beentrained to see as targets in CM situationsare clearly much less than in VM situations.Thus, for the processing mode in the CMconditions of the present tasks, we preferthe possibly slightly inaccurate, but de-scriptive, term automatic to the logicallymore accurate, but less descriptive, termsystemic, which was used in Shiffrin (197Sa)to refer to the same type of processing.

F. Limitations on the Ability toFocus Attention

The present series of studies (4a-4d) ex-amined the subjects' ability to focus atten-tion. Experiment 4a showed that attentionduring controlled search could be focused onspecified spatial locations. Experiment 4bshowed that the focusing is not complete,because certain types of stimuli in to-be-ignored locations are sometimes processedand when processed, can reduce target detec-tion. Experiment 4c showed that a CMtarget in a to-be-ignored location hindersautomatic deletion by a small amount. Ex-periment 4d showed that attention is greatlydistracted from the relevant locations whena CM target appears in a to-be-ignored loca-tion during a VM search task. This last find-ing is particularly interesting because thedistracting stimulus differs from the relevantstimuli not only in spatial location but alsoin category.


These findings are particularly importantbecause previous demonstrations of difficul-ties in focusing attention have usually utilizedincompatible responses. That is, a stimulusin response to which a subject has been trainedto emit response x is presented in a to-be-ignored location in conjunction with a stimulusrequiring a response y that is incompatiblewith x. In the Stroop Color-Word Inter-ference Test (1935), for example, the sub-ject must read names of colors that areprinted in ink colors that do not correspondto the names. Reading speed is greatly slowedin this case (see Jensen & Rohwer, 1966).Keele (1972) showed that decrements inprocessing stimuli resulted when a similarcolor-name incompatibility was present. TheEriksen and Eriksen (1974) study discussedabove also showed that response time wasdependent on the compatibility of responsesbetween stimuli to be attended to and thoseto be ignored.

Many other examples of this kind can befound in the literature, but our demonstra-tions differ in one very important respect.Namely, the distracting stimulus in our stud-ies hurts performance even though subjectshave been trained to respond to it in exactlythe same way as to the target stimulus. Bothtypes of studies demonstrate that the to-be-ignored stimulus receives processing, but thestudies utilizing incompatible responses donot demonstrate any processing interactionshort of the motor responses themselves. Onthe other hand, our studies suggest that thefocused-attention deficit results from a di-version of attention and an accompanyingloss of processing time, since there is no re-sponse incompatibility.

Our studies have shown that focused-at-tention deficits can arise due to distractionby either CM or VM items in appropriatecontexts, though probably due to rather dif-ferent mechanisms in the two cases. In manystudies in the literature that have shownfocused-attention deficits, it is not clear whichfactor is responsible. For example, in shadow-ing tasks the listener repeats aloud a desig-nated message (usually in one ear) whileother distracting messages are also presented.The early studies showed that distracting

messages barely harm shadowing performancewhen they are distinguishable by an obviousphysical characteristic (such as spatial origin;see Cherry, 1953; Cherry & Taylor, 1954).These studies are probably like our Experi-ment 4a in which one diagonal is relevantduring VM search and the other can beignored. A result related to Experiment 4bcan be found in studies by Treisman (1964a,1964c) in which items of some possiblerelevance are presented in the nonshadowedear (similar to our target foils). For example,if the distracting message is in the samevoice, but in French, a considerable distrac-tion occurs. Many studies have shown thatthe message in the nonshadowed ear is an-alyzed to some considerable depth (e.g.,Lewis, 1970; Treisman, 1960, 1964b). Inthese studies it is not clear which type ofprocessing causes the analysis, but in otherstudies it is clear that automatic processingis responsible for the analysis that occurs(Corteen & Wood, 1972; Von Wright, Ander-son, & Stenman, 1975).

Further reviews of such studies will notbe undertaken here because of the difficultyin ascertaining whether focused-attentiondeficits are caused by controlled processing ofinvalid items or by automatic processing ofinvalid items.5 If nothing else, our resultsstrongly suggest that future research onfocused attention should include controls todifferentiate deficits caused by controlled andautomatic processing.

In summary, we have seen that subjectscan divide attention almost without deficitwhen automatic detection is utilized (Part I/

6 Controlled processing might be given to oc-casional inputs in an irrelevant ear, because ear oforigin is a fairly difficult cue to utilize (comparedwith spatial location in a visual task, for example).Whether automatic processing or controlled proc-essing is used to restrict analysis to the desired ear,occasional failures of selection are to be expected,failures that will cause occasional items in the to-be-ignored ear to be given controlled processing. Inaddition, 'Controlled processing of items in the to-be-ignored ear may occur on occasion when processingof the items in the valid ear is completed and abrief interval occurs before the next valid inputarrives (an argument similar to this is used in thediscussion of Experiment 4b).


Experiments 1 and 2, CM conditions, seeFigure 2) and cannot divide attention with-out deficit when controlled search is utilized(Part I/Experiments 1 and 2, VM conditions,see Figure 2) . Focused-attention deficits arequite substantial when caused by automaticresponses to to-be-ignored items (Experi-ment 4d and Figure 10) but are quite a bitsmaller when caused by controlled processingof to-be-ignored items, since the subjects'control is usually adequate to prevent suchprocessing (Experiments 4a and 4b, andFigures 8 and 9). One implication is the pre-diction that the type of training procedures,VM or CM, will determine whether divided-or focused-attention deficits will occur.

G. The Nature of Automatic Detection

The reversal studies, Experiments 1 and 2,and the category study, Experiment 3, makeit clear that long-term learning is responsiblefor the phenomenon of automatic detectionthat is seen in the CM conditions. The lengthytraining period required for acquisition, andthe even longer training periods requiredto alter the detection process once learned,testify to the permanent nature of the learn-ing. Furthermore, the phenomenon is power-ful enough to transcend the laboratorycontext. Subjects given CM training onone group of letters as memory-set items withanother group as distractors, as in Experi-ments 1 and 3, report effects on reading out-side the laboratory. Despite the fact that allthe letters used in the experiments werecapitalized, the subjects reported that thememory-set items from the CM conditionstended to "jump out" from the page duringnormal reading. This effect was distractingenough that one subject would not attempt toread for an hour or more after an experi-mental session.

It is clear then that automatic detectionreflects a powerful long-term process. We nowask: What is it that is learned? Experiments4a-4d imply that an attention response islearned, but many details of the automaticdetection process remain to be specified.

It is our feeling that several factors con-tribute to the process that has been termedautomatic detection. First there is an auto-

matic-attention response to the features thatare encoded from an input target; this re-sponse directs attention to the representationin short-term memory of the relevant visualinput and also to the representation in mem-ory of the appropriate member of the memoryset. Second, in addition, an automatic "tar-get" response is learned that tells the subjectthat a target is among the inputs. Third, inaddition, in some situations an automaticovert motor response (such as a button press)is learned in response to a target.

The third factor, an automatically pro-duced overt motor response, is probablylimited to special situations in which thesame completely consistent response to alltargets is immediately required. Our reactiontime study (Part I, Experiment 2) may rep-resent such a situation, but our multiple-frame tasks do not. In the multiple-frametasks, no overt response is required until thetrial's end. In addition, some of the tasks inour studies require that the number oftargets be counted, rather than that a re-sponse be made to each target as it appears.Finally, the tasks in Experiment 4 occa-sionally present target foils in to-be-ignoredlocations, and the subject has little difficultysuppressing responses to such targets. Thus,we feel that automatic overt responses canbe learned but are probably not a majorcontributor to performance in most of ourtasks.

The first two factors, the occurrence ofautomatic "attention" and "target" responses,undoubtedly play a major role in our tasks.However, such responses must be followed bysome sort of controlled process or controlleddecision to generate the required overt re-sponse. A number of controlled processesmay be used. In most of our studies it wouldnot have been sufficient to simply use theoccurrence of an automatic "attention" or"target" response as a basis for a decisionto respond. For one thing, some of our studiesrequire responses to be counted; for another,some of our studies present target foils towhich responses must be withheld. Further-more, in Experiment 3 we saw that the mem-ory and distractor sets could be so confus-able that automatic responses could not be


learned in a perfectly accurate manner, sothat a switch to a controlled item mode wasnecessary to check the accuracy of automaticresponses.

For all of these reasons, we propose thatfollowing the occurrence of an automatic-attention or target response, the subject en-gages in the minimal controlled processingnecessary to satisfy the task requirements.Such controlled processes might consist ofcomparing the memory representations towhich attention has been drawn, counting thetarget instances, and checking the spatiallocation of located targets to see whetherthey are foils or valid targets.

Before concluding this discussion, it isuseful to consider briefly a few supplementaryhypotheses regarding the automatic detec-tion process. Each of these hypotheses as-sumes the occurrence of an automatic-atten-tion response.

First, consider the possibility that atten-tion is drawn automatically to the representa-tion of the relevant visual input, which isthen compared serially to the memory set.This hypothesis is easily rejected since itimplies, contrary to fact, that memory-setsize will have a large effect under CM con-ditions.

Second, consider the possibility that an at-tention response directs the subject to therelevant visual input, whose category is thencompared in one step to the category of thememory set. This hypothesis is testable inany of several ways but is difficult to rejecton the basis of present evidence. A tentativeinference from Experiment 3 suggested thatautomatic detection for items in a four-letterset developed much faster than categorylearning for that set. If so, then this categoryhypothesis could be ruled out (because theM = 2 and M = 4 CM functions convergedquickly in the mixed condition of Experiment3). Nevertheless, additional research will beneeded to test this hypothesis conclusively.

Assuming that an automatic-attention re-sponse underlies automatic detection, it seemsclear why automatic detection does not de-velop in VM situations: An attention or overtresponse that is helpful on one trial (whenthe producing stimulus is a target) will be

harmful on another (when the producingstimulus is a distractor). A subject cannotlearn both an attending and a nonattendingresponse to the same stimulus. While thisargument seems clear, an important theo-retical question remains. What is the under-lying mechanism that inhibits the learningof an automatic-attention response in theVM search situation?

Suppose that a Learning event takes placeeach time a target is found correctly and istherefore reinforced. In consistent (CM)paradigms there will be no impediment tothe learning of attention responses. In VMparadigms, the outcome is less clear. It mightbe supposed that every item, whether cur-rently a memory-set item or a distractor,comes to elicit attention responses due tothose trials on which it is a target. There arethen several possibilities: (a) Learning con-tinues until all items have acquired attentionresponses of roughly equal strength. Becausethe responses are of equal strength, they tendto cancel each other, and controlled searchmust be utilized. Later, if a switch is madeto a CM paradigm, the strength of the re-sponses to the fixed memory-set items be-comes greater than that for the distractoritems, (b) As an attention response beginsto develop it starts to occur on trials whenthe input is a distractor. Because attentionis directed to the wrong input, performancesuffers, and the response is therefore in-hibited, (c) Each time a distractor is com-pared during a controlled search, whetheror not an attention response occurs, inhibitionof any previously reinforced attention re-sponse may take place because a comparisonis carried out but not reinforced. According toexplanations (b) and (c), the inhibitioncancels any learning that would otherwiseoccur, so that attention responses to any ofthe items never develop to significant degrees.

We prefer hypotheses (b) or (c), or anysimilar proposal that implies that attentionresponses do not develop in VM paradigms.Is there any evidence to distinguish theseviews from hypothesis (a)? Only indirectevidence is available at present, but themodels are easy to test. For example, afterVM training one could introduce new items


as targets, with other new items as distrac-tors, or with the previously trained VMitems as distractors. The old VM distractorsshould severely hinder performance if theyelicit attention responses developed duringprevious training. Other similar tests couldalso be carried out, but a definitive answeris not yet available at the time of this writing.

IV. A Framework for Information Processing

This section of the paper will be organizedas follows. Section A will present an overviewof the system and an overview of the role ofcontrolled and automatic processing. SectionB will elaborate on the fundamental natureof controlled and automatic processing. Sec-tion C will present a framework for search,detection, and attention. Section D will dis-cuss how automatic and controlled processingare utilized in memory storage and retrieval.

A. An Outline of a General Theory

Memory is conceived to be a large andpermanent collection of nodes, which becomecomplexly and increasingly interassociatedand interrelated through learning. Most ofthese nodes are normally passive and inactiveand termed long-term store, or LTS, when inthe inactive state. The set of currently ac-tivated nodes is termed short-term store, orSTS. LTS is thus a permanent, passive repos-itory for information. STS is a temporarystate; information in STS is said to be lostor forgotten when it reverts from an activeto an inactive phase. Control of the informa-tion-processing system is carried out througha manipulation of the flow of informationinto and out of STS. These control processesinclude decisions of all sorts, rehearsal, cod-ing, and search of short- and long-term store.LTS contains learned sequences of informa-tion processing which may be initiated by acontrol process or by environmental or in-ternal information input, but are then ex-ecuted automatically with few demands onthe capacity of STS.

1. Long-Term Structure

The structure of the nodes making up LTSwill be treated as a very general graph with

complex interrelations among nodes. An in-dividual node may consist of a complex setof information elements, including associa-tive connections, programs for responses oractions, and directions for other types ofinformation processing. What then sets offone node from a group of nodes? One nodeis distinguishable from a group of nodes be-cause it is unitized, that is, when any of itselements are activated (i.e., placed in STS),all of them are activated. One activated nodemay of course activate another node, but itdoes not do so in all situations, only when thecontext or the state of the information-processing system is appropriate.

2. Structural Levels

It seems likely that the structure of long-term store, at least in part, is arranged inlevels (perhaps sometimes in a hierarchicaltree structure). By levels we refer to a tem-poral directionality of processing such thatcertain nodes activate other nodes but notvice versa. In sensory processing, there is atendency for increasing information reductionas successive levels are activated. Thus, avisually presented word could first be pro-cessed as a pattern of contrast regions, colors,regular variations, and so forth; then lines,angles, and other similar features could beactivated; then letters and letter names andverbal or articulatory codes; then the word'sverbal code; and finally, the meaning andsemantic correlates of the word. This se-quence is meant as an example, and we donot wish to imply that these are the relevantfeatures, that this is the only possible order-ing, or that this listing is exhaustive. Sucha sequence of feature encoding should occurautomatically to a normally skilled reader.

3. Automatic Processes

An automatic process can be defined withinthis system as a sequence of nodes that nearlyalways becomes active in response to a par-ticular input configuration, where the inputsmay be externally or internally generated andinclude the general situational context, andwhere the sequence is activated without the


necessity of active control or attention bythe subject.

An automatic sequence differs from a singlenode because it is not necessarily unitized.The same nodes may appear in differentautomatic sequences, depending on the con-text. For example, a red light might elicita braking response when the perceiver is ina car, and elicit a walking, halting, or traffic-scanning response when the perceiver is apedestrian.

Since an automatic process utilizes a rela-tively permanent set of associative connec-tions in long-term store, most automaticprocesses will require an appreciable amountof training to develop fully. Furthermore,once learned, an automatic process will bedifficult to suppress or to alter.

When an automatic sequence is initiated,its nodes are activated and hence the as-sociative information enters STS. This factdoes not, however, mean that the variouselements of the automatic process must beavailable to consciousness or recallable at alater time. The activation in STS could beextremely brief (in msec, say) and unlessattention is directed to the process when itoccurs or unless the sequence includes anautomatic attention-calling response, thenthe information may be immediately lostfrom STS, and the subject may be quite un-aware that the process took place. Evenwhen an automatic-attention response is partof the sequence, it will not necessarily affectongoing controlled processing unless thestrength of the attention response is suffi-ciently high.

4. Thresholds of Activation

However an automatic process is initiated,whether by internal or external input or by acontrol process, it is presumed that the prob-ability that the process will run to comple-tion depends on the strength of the initiatingstimulation. The simplest and most commonexamples are seen in studies of psychophysicalthresholds. If a letter, for example, is visuallypresented at a low-enough intensity or ashort-enough duration, then it may be en-coded not as a letter but as a partial collec-tion of line-like features. At lower durations

or intensities even these features will not beactivated.

Although we have described automaticprocesses as largely beyond subject control,some indirect control is possible through ma-nipulation of the activation threshold for auto-matic processes. In particular, according towhat we shall call the contextual hypothesis,the activation threshold can be lowered bythe inclusion of information in STS (at thetime of presentation of the activating input)that is associatively related to the nodesmaking up the automatic sequence.6

Note that a lowering of the threshold doesnot imply that the quality of processing isimproved. One result of threshold loweringis that the automatic process will be triggeredby inputs that normally would and shouldnot do so. For example, the feature "horse"might be incorrectly triggered by the input"house" if the threshold for "horse" islowered sufficiently.

S. Controlled Processes

A controlled process utilizes a temporarysequence of nodes activated under control of,and through attention by, the subject; thesequence is temporary in the sense that eachactivation of the sequence of nodes requiresanew the attention of the subject. Becauseactive attention by the subject is required,only one such sequence at a time may be con-trolled without interference, unless two se-quences each requires such a slow sequence ofactivations that they can be interwovenserially. Controlled processes are thereforetightly capacity-limited, but the cost of ca-pacity limitations is balanced by the benefitsderiving from the ease with which suchprocesses may be set up, altered, and appliedin novel situations for which automatic se-

6 It may also be possible to lower the thresholdby a .recent activation of the same automatic se-quence. However, this hypothesis is difficult todistinguish from the contextual one, because anactivation of a sequence is usually accomplished bythe prior, .recent presentation of information as-sociatively related to the sequence, and this relatedinformation is still likely to be in STS at the timeof test.


quences have never been learned. Controlledprocessing operations utilize short-term store,so the nature of their limitations is deter-mined at least in part by the capacity limita-tions of STS.

6. Short-Term Store (STS)

Short-term store is the labile form of thememory system and consists of the set ofconcurrently activated nodes in memory.The phenomenological feeling of conscious-ness may lie in a subset of STS, particularlyin the subset that is attended to and givencontrolled processing.7

The capacity of STS is determined sto-chastically (see Shiffrin & Cook, Note 3)so that a large amount of information maybe present (activated) at any one moment,but only a small amount of information willpersist for an appreciable time period ofseveral seconds or more. Forgetting or lossfrom STS is simply the reversion of cur-rently active information to an inactive statein LTS.

What determines the loss rate? We sup-pose that the rate of loss of any informationalelement or node in STS depends on the num-ber of similar elements simultaneously activein STS. By similarity we refer not only toformal physical similarity but also to sim-ilarity of features at comparable levels ofprocessing (e.g., the typeface of a printedword will be less likely than the word itselfto cause forgetting of a verbally encodedsecond word in memory). When a largeamount of similar information is present, theloss rate will be rapid but will slow as theamount of active information decreases. Togive an example, when a complicated visualscene is presented to a subject briefly in atachistoscope, a flood of visual informationenters STS and initiates a series of chainsof automatic processing that result in higherlevel features' also entering STS. However,most of the activated information will havedecayed and will be lost from STS in justa few hundred milliseconds after physicaloffset of the scene (see Sperling 1960); justa few of the features, perhaps at higher levels,will remain present longer than a few seconds.

STS has two somewhat distinct roles. Thefirst is the provision of a temporary store-house for information currently important tothe organism. That is, it acts as a selectivewindow on LTS to reduce the amount ofinformation for processing to manageableproportions. The second role of STS is theprovision of a work space for decision making,thinking, and control processes in general.

7. Learning: Transfer to LTS

Consider first what is meant by transferfrom STS to LTS. Transfer implies the for-mation in permanent memory of informationnot previously there. To be precise, this con-sists of the association (in a new relation-ship) of information structures already inLTS. A minimal requirement for this newassociative structure is the simultaneouspresence in STS of the separate elements tobe associated or related. That is, the variousnodes to be linked in a new relationship mustbe activated in STS. Thus, transfer to LTSdoes not imply the removal of the transferredinformation from STS, nor the placing ofnew "subunits" in LTS that do not alreadyexist in either LTS or STS. Rather, transfermeans the formation of new associations (orthe strengthening of old associations) be-tween information structures or nodes notpreviously associated (or strengthened) inLTS. Most new associative structures willinclude as a component the context in STSat the time of the transfer.

The above remarks specify the nature ofnew learning in STS but not the cause under-lying the storage mechanism. It has been

7 An alternative formulation, closer to that sug-gested by Atkinson and Shiffrin (1968), woulduse the term STS to refer to those nodes that aregiven attention and controlled processing. Otheractivated nodes would be referred to by anotherterm, such as "sensory register" in Atkinson andShiffrin's terminology. At such a general level ofdiscussion, it is doubtful that there are substantivedifferences between the two approaches—versions ofeach could be constructed to be identical to eachother. Each approach has its own heuristic ad-vantages, but a theory is better judged in terms ofits detailed assumptions and predictions than byits choice of either type of terminology.


shown that rehearsal (and coding rehearsal)are strongly implicated in learning (seeShiffrin, 197Sb, for a discussion). We preferan extension of the rehearsal hypothesis. Weassume that what is stored is what is at-tended to and given controlled processing.Rote rehearsal is just one form that attentioncan take. In fact, maintenance rehearsal mayresult in storage of low-level auditory orverbal codes not helpful for long-term recall,(but helpful for long-term recognition), whilecoding rehearsal may result in storage ofhigh-level codes useful for recall. This modelsuggests that some degree of attention orcontrolled processing is a prerequisite forstorage. Thus, incidental learning situations,in which no extended rehearsal takes place,will still result in some learning to the degreethat the input items are attended to duringpresentation.

These hypotheses that controlled processingwill underlie learning apply of course to thedevelopment of automatic processing, andin particular, to the development of auto-matic detection studied in the search anddetection paradigms earlier in this paper andin Part I. In those paradigms it was seenthat automatic detection developed only withconsistent training. Consistent training iscrucial when the learned sequences in LTScontains an internal or overt response thatwill be harmful to performance on trials thatare inconsistent. For example, an attentionresponse to an item will harm performance ontrials when that item is a distractor. Purelyinformational sequences (i.e., sequences thatdo not include responses that direct internalprocesses or overt actions) will be stored inLTS when attended to, regardless of con-sistency of training. In all cases, however,consistent training and large numbers ofrepetitions should lead to stronger automaticencodings and processes.

An implication of the hypothesis that trans-fer to LTS is engendered by controlled pro-cessing is the important concept that con-trolled processing, and hence attentionallimitations, will be involved during the ac-quisition of automatic processing. We havelargely been identifying attentional limita-tions with controlled processing (e.g., con-

trolled search) and have shown how auto-matic processing (e.g., automatic detection)can bypass attentional limitations. It shouldnot be overlooked that the initial learning ofautomatic processing may require a varietyof control processes and hence will be sub-ject to various limitations that may dis-appear when learning has progressed to ahigh level.8

8. Retrieval of Information from Short-TermStore

Several retrieval modes from STS are pos-sible. An automatic process might direct theretrieval process to just a subset of the ac-tivated information (e.g., only the letters,not the masks, are compared in our VMsearch studies).

The various controlled search strategiesthat are available are normally serial in na-ture but a variety of search orders is possible.Search order may depend upon instructions,strength of short-term traces, the nature ofcategories of the traces, the modality of theinformation, and the structure of STS. Thislast point is worth emphasizing: Since STSis embedded in LTS, it partakes of the struc-ture of LTS and is not a totally undiffer-entiated mass of information. Thus the order-ing of a search can utilize this existent struc-ture. In addition, the comparison process maybe exhaustive or terminating, and the in-formation located in one phase of the searchcan be used to redirect a later phase of thesearch.

It is the control of search order that isresponsible for many of the selective atten-tion effects that are observed. Information inlocations to which attention is first directed(i.e., the first locations searched) will ingeneral receive faster and more accurateprocessing for the obvious reasons.

8 This distinction is helpful in understanding therelation of our work to studies of attention ininfrahuman organisms: the traditional studies of at-tention in discrimination learning and blocking (e.g.see Mackintosh, 1975) are involved with thoselimitations that occur during acquisition, while anumber of newer studies are involved with limita-tions in steady state situations when learning is notpossible (e.g. see Riley and Leith, 1976).


9. Retrieval of Information from Long-TermStore

Both automatic and controlled processingare utilized in long-term retrieval. In gen-eral, an automatic associative mechanism willcause currently inactive information in LTSto become active when associatively relatedinformation is in STS. The simplest exampleis the perceptual encoding process by whichinformation concerning environmental inputsis located in LTS. As each feature is acti-vated, it can serve as a base for further acti-vation of features associatively related to itand to other previously activated features.This same principle applies to long-termretrieval of higher order information.

It should be noted well that informationonce activated automatically from LTS willthen be in STS and may have to be locatedthrough use of any of the short-term retrievalmechanisms discussed above.

How does the subject exert control overLTS retrieval? The subject can utilize re-hearsal and selective processes to raise thestrength of certain information in STS abovethe strength of other information. Then theactivated and retrieved information fromLTS will tend to be information related tothe selectively accentuated subset,

In Shiffrin's (1970) paper the details ofthe controlled LTS search were presented atlength. In summary, the controlled searchprocess is a series of cycles. On each cycle(a) the subject generates probe informationand places it in STS; (b) the probe informa-tion, along with the general contextual infor-mation presently in STS, activates associatedinformation from LTS, called the search set;(c) the subject searches the STS search set;(d) the subject decides whether the appro-priate information has been found andwhether the search is over. The process thencontinues cycling. Note that step (c) is aretrieval from STS, so that one phase of con-trolled LTS retrieval may be STS retrieval.

10. Long-Term Forgetting

The forgetting of information in LTS is,by definition in the present theory, a questionof retrieval failure. The retrieval processesdiscussed in the preceding sections are by

their nature imprecise and fallible. Whenretrieval fails, perhaps temporarily, then wesay forgetting has occurred. The causes ofsuch failure are somewhat to the side of themain interests of this paper but may be sum-marized briefly as follows. First, the probecues used to activate related information inLTS may be ineffective, either because thesubject chooses them incorrectly or becausethe general context in STS at the moment isdissimilar to that stored with the desired in-formation in LTS. Second, the process ofretrieval itself may harm additional retrievalattempts, especially if the probe cues arenot altered from one cycle of the search tothe next. Third, the search may fail due topremature termination; that is, it may faildue to a decision that further retrievals arenot worth the effort, or due to the reachingof some sort of time limit available forsearch. (See Shiffrin, 1970, 1976, for a fullerdiscussion of these matters.)

B. The Characteristics of Controlled andAutomatic Processing

In capsule form, we have covered the majorphases of the information-processing system.We shall now attempt to define and out-line the characteristics of automatic and con-trolled processing in a more precise fashion.

1. Controlled Processing

It is important to note that not all controlprocesses are available to conscious percep-tion, and not all control processes can bemanipulated through verbal instruction. It istherefore convenient to divide control pro-cesses into two classes: accessible and veiled.Accessible control processes are those likerote rehearsal or alphabetic search of LTSthat can be instituted and modified by in-struction. These are generally slow processesthat are easily perceived by the subject.Veiled control processes are those like theserial comparison of items in short-term storethat are difficult to modify through instruc-tion. They are not easy to perceive throughintrospection because they take place soquickly.

Both classes of control processes have the


following general properties: (a) They arelimited-capacity processes requiring attention.Because these limitations prevent multiplecontrol processes from occurring simulta-neously, these processes often consist of thestringing together in time of a series ofsingly controlled unitary operations, (b) Thelimitations of control processes are based onthose of STS (such as the limited-comparisonrate and the limited amount of informationthat can be maintained without loss), (c)Control processes can be adopted quickly,without extensive training, and modifiedfairly easily (though not always by verbalinstruction), (d) Control processes can beused to control the flow of information withinand between levels, and between STS andLTS. In particular, they can be utilized tocause permanent learning (i.e., transfer toLTS) both of automatic sequences and of in-formation in general, (e) Control processesshow a rapid development of asymptotic per-formance. For a given control process, ifautomatic processing does not develop (due,say, to varied-mapping procedures) and ifthe constituent elements of the process donot otherwise change due to long-term learn-ing, then performance level will stabilize veryquickly at an asymptotic value. When per-formance improves over trials, it does so be-cause the control process is changed, or be-cause automatic processing develops, or be-cause the constituent nodes that are linkedby the control process are themselves alteredby long-term learning.

Common examples of control processes in-clude maintenance or rote rehearsal, codingrehearsal, serial search, long-term memorysearch, and decisions and strategies of allkinds. These will be discussed in more detailin the sections below.

2. Automatic Processing

Automatic processes have the followingproperties, (a) They are not hindered by thecapacity limitations of STS and do not re-quire attention. Thus automatic processesoften appear to act in parallel with one an-other and sometimes appear to be indepen-dent of each other, (b) Some automaticprocesses may be initiated under subject con-

trol, but once initiated all automatic processesrun to completion automatically (thoughsome indirect control is possible throughmanipulation of the contents of STS at thetime of the inciting input), (c) They requireconsiderable training to develop and are mostdifficult to modify, once learned, (d) Theirspeed and automaticity will usually keeptheir constituent elements hidden from con-scious perception, (e) They do not directlycause new learning in LTS (though they canindirectly affect learning through forced al-location of controlled processing), (f) Per-formance levels will gradually improve overtrials as the automatic sequence is learned.In many cases asymptotic performance levelsmay not be reached for thousands of trials.

It is particularly important to specify care-fully what we mean when we say that anautomatic process does not partake of thecapacity limitations of STS. To make thisclear, let us suppose that X, Y, and Z arenodes that are automatically activated inturn by input I in the presence of generalcontext nodes C. We may depict the sequenceas follows:

C C

As each of X, Y, and Z is activated, it entersSTS and its future residence in STS will beaffected by the limitations of STS. In fact,if the nodes X, Y, and Z do not include anattention-attracting response, then it is quiteconceivable that all three nodes will decayand be lost within a few hundred milliseconds,and the subject will be quite unaware thatsuch an automatic sequence ever occurred.Nevertheless, the sequence itself is notgoverned by the limitations of STS in thesense that the probability of its being ac-tivated and the rate of its occurrence willnot be affected by other concurrent automaticand conrolled processes taking place in STS(at least if context C is present and theother concurrent processes do not also utilizenodes X, Y, or Z).

3. The Development oj Automatic Processing

The tendency for one node to activateanother will be increased if both nodes are


present simultaneously in STS and if a con-trol process and attention are directed towardthese nodes, thereby increasing their saliencein STS.

Before we can describe efficacious trainingconditions, we must distinguish between twotypes of automatic processes. One type ofautomatic sequence may be called actional,because it includes phases that direct internalprocesses (like calls for attention) or phasesthat produce overt responses (such as buttonpresses). A second type, called informational,contains no directions for actions. The dis-tinction is crucial because an informationalsequence strengthened on one trial will nothave deleterious consequences for perform-ance on other types of trials. On the otherhand an actional sequence may give rise toa response antagonistic to a response re-quired on another trial. Thus, to be useful ina given task, actional sequences requirespecial, consistent, training conditions.

To make this point clear, suppose thatnode B produces a response antagonistic tothat produced by node C, while nodes D andE produce no responses. If any of A-B, A-C,A-D, or A-E is trained alone, then thatsequence can be learned. If training on A-Dis mixed from trial to trial with training onA-E, then A will come to elicit both D and E.However, if A-B trials are mixed with A-Ctrials then learning of both will be impossible,since A cannot lead simultaneously to twoantagonistic responses. The automatic de-tection system discussed at length in thispaper is just an actional sequence, since anautomatic-attention response to one stimulusis incompatible with a simultaneous atten-tion response to another stimulus; thereforeautomatic detection requires consistent train-ing to develop. Thus, in the varied-mappingconditions of any of our studies automaticdetection could not develop and controlledprocessing had to be utilized.

4. Combinations of Automatic and ControlledProcessing

Although sensory inputs are first encodedwith the automatic processing system, withthe results being made available to con-

trolled processing, it must not be concludedthat automatic processing invariably precedescontrolled processing. In fact, automatic andcontrolled processes can proceed in parallelwith one another (as in Experiment 4d whenautomatic processing of the elements on theinvalid diagonal proceeded in parallel withcontrolled processing of the valid diagonal).Even more important, controlled processingis often used to initiate automatic processing.Particularly in complex processing situations,(such as reading), an ongoing mixture of con-trolled and automatic processing is utilized.The next steps in the serial, controlledprocessing sequence are based on the outputof automatic processes initiated earlier; thennew automatic sequences are initiated andthese run to completion in parallel with theongoing controlled processing.

5. The Benefits of Automatic and ControlledProcessing

A system based on the two basic processingmodes with the characteristics we have de-scribed has many advantages. In novel situa-tions or in situations requiring moment-to-moment decisions, controlled processing maybe adopted and used to perform accurately,though slowly. Then as the situations becomefamiliar, always requiring the same sequenceof processing operations, automatic processingwill develop, attention demands will be eased,other controlled operations can be carriedout in parallel with the automatic processing,and performance will improve. Some of theadvantages of such a system are (a) It allowsthe organism to make efficient use of alimited-capacity processing system. The de-velopment of automatic processing allows thelimited-capacity system to be cleared anddevoted to other types of processing neces-sary for new tasks, (b) It allows attentionto be directed (through automatic-attentionresponses) to important stimulation, what-ever the nature of the ongoing controlledprocessing, (c) It allows the organism toadjust to changes in the environment thatmake previously learned activity patternsuseless or harmful, (d) It allows the organismto deal with novel situations for which auto-


Long term store

sNS0RY

INPUTS

Short term storeLevel

— *=K

"T*ii i~

(^-

Level Automatic2 Encoding

v~

^f

s*

Automaticattentionresponse

4 ̂ ̂ "'— / /

"V *~ ' /

LevelN-1

*t=3—K=3—

\ ,

V. ,<!̂YAttentioiT(\ director )

LevelN

—

•̂ ,

tr

/

Control led processing

M=)

/""

he

kues

C

b

omaticponse

Dntroiledresponse

RESP0NSEPR0DUCTI0N

0UTPUTS

Figure 11. A model for automatic and controlled processing during tasks requiring detection ofcertain input stimuli. Short-term store is the activated subset of long-term store. N levels of auto-matic encoding are shown, the activated nodes being depicted within each level. The dashed arrowsgoing from higher to lower levels indicate the possibility that higher level features can sometimesinfluence the automatic processing of lower level features. The solid arrow from a node in Level 2to the attention system indicates that this node has produced an automatic-attention response, andthe large arrow from the attention system to Level 2 indicates that the attention system hasresponded. The arrow from level N to the Response Production indicates that this node has calledfor an automatic overt response, which will shortly be executed. The arrow from ControlledProcessing to the Response Production indicates the normal mode of responding in which theresponse is based on controlled comparisons and decisions. Were it not for the automatic responsesindicated, detection would have proceeded in a serial, controlled search of nodes and levels in anorder chosen by the subject.

matic sequences have not previously beenlearned, (e) It allows the organism to learnincreasingly complex modes of processing bybuilding upon automatically learned sub-systems.

C. A Framework jor Processing in Detection,Search, and Attention Tasks

The framework we have in mind is builtwithin the general theory already presented.It is illustrated in Figures 11 and 12. When aset of inputs is presented (let us supposefor convenience that the inputs are visual)then each begins to undergo automaticprocessing. The system automatically encodeseach stimulus input in a series of stages andactivates a series of features in the process.

For example, the letter "M" may first beencoded in features indicating contrast,color, and position; then curvature, con-vexity, and angles; then a visual lettercode and a verbal, acoustic-articulatorycode, then the codes "letter," "consonant,""capital;" and finally, perhaps, semanticand conceptual codes like "followed by 'N',""middle of the alphabet," and the like. (Wedo not necessarily imply that these featuresare correct or exhaustive; if, say, amplitudecomponents of the spatial-frequency analysisof the inputs prove to be relevant featuresencoded by the system, the theory we de-scribe would be unchanged in all importantrespects.) What features will become ac-tivated depends on the physical nature of thenervous system that was predetermined gen-


etically, the degree and type of prior learning,the physical characteristics of the display(like duration and contrast), and the generalcontext, both that in the environment andthat generated in STS by the subject. To theextent that the subject directs the sensoryreceptor orientation and to the extent thatinternally generated information can alter thecontext in STS, the subject will have at leastsome indirect control over automatic sensorycoding.

The automatic processing as describedabove takes place in parallel for each of theinput stimuli. The processing of each stimulusis often independent, except for lateral andtemporal interactions at early stages, calledmasking, and except for learned relationshipsbetween items that may affect processing atlater stages (if adjacent letters form a word,for example). The various features that areactivated are all placed thereby in STSwhere they reside for a short period beforebeing lost (i.e., before returning to an in-active state in LTS).

We propose that some of the features or

nodes that are automatically activated mayinitiate a response that will direct subsequentprocessing or subsequent actions. For ex-ample, an attention response might be ac-tivated by a feature; the attention responsemight direct controlled processing to the cor-responding set of features representing thatinput stimulus, so long as other competingattention responses do not occur simulta-neously. As another example, an overt re-sponse might be engendered by a particularstimulus (such as a startle response to asudden loud noise, a galvanic skin response toan aversively conditioned word, a duckingresponse to a missile thrown at the head).If the set of inputs contains a target stimulusthat gives rise to a relevant nonconflictingattention response, or to an overt response,then we say that automatic detection isoperating. Note that the attention responsecould be attached to features at any levelof processing and to more than one featureat a time. In Figure 12, as an example, at-tention responses are attached both to thefeature "8" and to the feature "number."

Inputs ^\IV h"\S I

Visualfeatures

Visualcharacter

codeCategory

code

Higherlevel

codes

Controlledprocessing

Son's age"'Modelrailroadtrack" etc.

Figure 12. Another view of the model shown in Figure 11. This figure depicts a conceivable (butabbreviated) series of feature encodings for a frame in which two characters and two masks arepresented. The 'Consistent-mapping condition is shown in which numbers are memory-set itemsand letters are distractors. The arrows skipping levels indicate that a given feature can helpactivate features at several different levels. The stimulus 8 is a consistent-mapping target, andhence both the visual and category codes produce automatic-attention responses. The attentionresponse has attracted attention to the information deriving from the input 8.


It is possible that many of the inputs willgive rise to attention or other responses andthat these responses will conflict. For ex-ample, every input might automaticallyengender an attention response; since it isimpossible to direct attention to every itemin the display at once, the various attentionresponses will conflict and cancel each other.In such cases, the subject will not be governedby the output of the automatic system andwill base subsequent actions and responseson the results of a controlled search. Thesehypotheses were supported in Experiment 2(see Figure 4, left panel for the results). Inthis study, all inputs were items that sub-ject had previously been trained to respondto with automatic-attention responses. Theresults showed that the subjects reverted tothe use of controlled search; furthermore,the controlled search was almost identicalto that seen in the usual VM conditions inwhich none of the inputs had been trained toelicit automatic-attention responses.

In cases when the input stimuli do notactivate automatic responses that govern thedetection process, then a controlled searchmust be used. In this event, the subject mustattempt to search through the set of featuresthat is activated and to use a search strategythat is as efficient as possible. For example,in our visual search studies, letter codes arenormally compared in preference to (andprior to) sets of angles and lines; a categorycode representing a set of letters is oftencompared in preference to (and prior to)individual letter codes, as shown in Experi-ment 3. Furthermore, the order of searchand the placement of decisions within thesearch are also controlled in an attempt toincrease efficiency. To give an example, com-parisons in our studies in Part I took placein an order that cycled through the framefor a given memory-set item before switchingto a new memory-set item, and a matchingdecision was made after every comparison.To give another example, matching decisionsin many memory search studies (M varying,F — 1) are withheld until all comparisonsare completed (Sternberg, 1975).

Note finally that many attention tasksutilize search or detection paradigms. When-

ever this is the case, the framework describedhere applies equally to attention tasks.Divided-attention tasks study the incrementsin performance that may occur when theload is decreased. Decreased loads are usuallyspecified through advance instructions thatcertain inputs are irrelevant, thereby de-creasing the size of the memory set, the frameset, or both.

According to our framework, experimentsthat do not show dividing attention to re-duce performance fall into two classes. Oneclass includes those studies in which auto-matic detection is operating. These are usuallystudies utilizing a CM paradigm and highdegrees of practice. Divided-attention deficitswill not be seen because the target stimuluswill be detected automatically, in parallelwith the other stimuli, and often indepen-dently of other stimuli. An exception to thisgeneral rule may apply when two (or more)targets are presented simultaneously on atrial. Even if both targets generate attentionresponses, there may be a difficulty in dis-criminating a double from a single occurrence.In such an event, the controlled comparisonsystem may have to be called into play tocount the targets. (A detection decrementmay then occur, because the second itemmay have decayed from STS by the timethe comparison of the first item is complete;see Moray, 1975, Shiffrin, 1976, and thediscussion of the 0 spacing effect in the CMcondition of Experiment 3, Part I.)

The second class of studies in which di-vided-attention deficits do not occur is thatin which controlled search is utilized but thecapacity of STS is not stressed. Examplesare seen in the series of studies by Shiffrinand his colleagues, summarized in Shiffrin's(1975a) article. In such studies either theload in STS is kept low, or a cue informingthe subject which input is relevant appearsat about the same time as the inputs.

It is, however, only in specially designedsituations that capacity limits are not stressed.In most studies requiring controlled search,the extra time required to search a largernumber of relevant inputs will impair per-formance, whether measured by reactiontime or accuracy. This was the case in the


VM conditions of the experiments reportedin this paper, and in Part I: In all thesecases increases in load reduced performance.There are two basic reasons for the perform-ance reduction in controlled search studies.First, some of the features in STS may decayand revert to LTS before the comparisonprocess reaches them. Second, new inputsmay arrive and require processing, forcingthe comparison process to switch away fromthe previous inputs. Probably only this secondfactor operated in the studies we have re-ported in this paper. For either of thesereasons, divided-attention deficits can be ex-pected in detection tasks requiring controlledsearch.

The situation is quite different, however,in focused-attention studies. In these studies,certain inputs are known to be relevant, andothers are to be ignored. If controlled searchis being utilized, there is little reason to ex-pect that any substantial deficit will becaused by the presence of to-be-ignored stim-uli, at least if it is clear to the subject well inadvance which stimuli and locations arerelevant and if there are no location con-fusions. In such cases the subjects shoulddirect the search order so that the relevantlocations are searched first; thus perform-ance deficits should not arise. Evidence sup-porting these propositions is found in Experi-ment 4a: Subjects were able to search onediagonal and ignore the other.

The tasks in which large deficits in focus-ing attention will occur are those in whichirrelevant items give rise to attention re-sponses during automatic encoding. In suchcases, attention will be attracted to the to-be-ignored item and a loss of time will occurbefore the controlled search can be redirectedto the relevant inputs. The time lost willimpair performance. Such focused-attentiondeficits were seen in the reversal results ofExperiments 1 and 2, and in the attentionresults of Experiment 4.

Thus, our framework can be applied to awide variety of search, detection, and at-tention tasks, although additional assump-tions must be made to generate precise modelsto deal with the particulars of each task. InPart I we presented a quantitative serial,

terminating search model and discussedsome alternative models. Our discussion waslimited, however, to our basic single ormultiple-frame search task with small valuesof M and F. We will now consider a fewalternative paradigms, along with specialassumptions they might require.

1. Tasks Utilizing Large Memory Sets

Suppose there are too many items in thememory set to be maintained in STS. Ofcourse, then, the memory set must be learnedin LTS in advance of the test. There arethen two basic search modes. In the first,the test item is automatically encoded andaccesses a node in memory that contains thedesired information. For example, in decidingwhether an item is any word, the input item,if a word, is encoded to the nodes represent-ing the word and the information in thosenodes is usually sufficient to classify theitem as a word; on the other hand, a non-word obeying appropriate orthographic con-straints is encoded to a lesser degree and thefailure of the automatic encoding processcould itself serve as a cue for "nonwordness."The second type of search mode would con-sist of entering successive parts of the mem-ory set from LTS into STS, using controlledsearch to examine each subset in STS. Forexample, if asked whether there is a flower,country, or first name whose fourth letter is"u" a subject might successively generatemembers of each category and serially checkthem.

The above comments make it clear thatthe case of large memory sets is not differentin principle from the cases in which thememory set may be held in STS, thoughthe models may need additional assump-tions. For example, Atkinson and Juola(1974) carried out a study in which sub-jects had a large memorized set in LTS.They proposed a hybrid model in which ajudgment based on familiarity enabled thesubject to decide on some trials without asearch that the test item was definitely in,or definitely not in, the memorized set. Ontrials when this initial familiarity judgmentwas ambiguous, then a controlled searchthrough the set was made.


2. Large Display Sets

When large numbers of items appear in thedisplay, then it may turn out that acuityvaries considerably across the display for agiven fixation point. In turn, acuity changesare likely to lead the subject to adopt astrategy in which eye movements are usedto bring subsets of the display successivelyinto a high-acuity region. The location ofeach new fixation can be governed by acontrolled search strategy or by an automaticprocess, but in either event the necessity ofobtaining good acuity will force this processto be serial across successive eye movements.Within each fixation, furthermore, eitherautomatic detection or controlled search couldbe utilized. The search studies in which sub-jects scan rows of characters for targets haveshown both modes of search. For example,the Neisser (1963) results, reviewed in PartI, showed that automatic detection developedfor subjects well trained in responding toletter sets presented in CM fashion (memory-set size had no effect), while controlled searchwas utilized for sets that had been givenonly small amounts of training. An interest-ing intermediate case occurred when the sub-ject was instructed to locate a row of char-acters that did not contain a given character.In this case, even with CM training, the taskproved quite difficult. In our view the sub-ject would have to adopt a serial processfrom row to row (rather than from fixationto fixation). Within each row automatic de-tection might be used to locate the keycharacter, but then a decision would haveto be made before the next row could beconsidered.

3. Categorical Partitioning oj Display Sets

Partitioning of the display set has usuallybeen accomplished by a categorization de-pendent on a relatively gross simple physicalfeature, such as color, shape, or size. Whendisplays are segregated into two or moregroups by such features, processing may beaffected in two ways. First, the patterningof the input can govern the nature of auto-matic processing—controlled processing maybe directed automatically to a subset of the

inputs (assuming that one particular subsetis consistently relevant). Second, the per-ceptual categorization could influence theorder and perhaps the nature of controlledsearch. To give one simple example, a dis-play arranged in two separate rows will tendto be processed one row at a time, in readingorder. A second example occurs in situationswhere controlled processing can enable oneto more quickly decide about the categoryof the input than about the memory-setmembership of the input. In these cases atwo-phase controlled search might be adopted,with the first phase locating the relevantinputs by category, and the second phasematching these inputs to the memory set.This is illustrated by Green and Anderson's(1956) and Smith's (1962) studies in whicha two-digit number of a particular color hadto be found in a display of many two-digitnumbers of differing colors.

Automatic processing based on perceptualcategorization may have played an importantrole in the studies of search and detectionreported in this paper. We have assumedthroughout that only the display positionswith characters, and not those with masks,are considered in the search. How is searchrestricted to character positions? It is con-ceivable that a preliminary controlled searchis used to identify character positions, butthe results of Experiment 4a argue againstsuch a view—knowing the relevant diagonalof a four-character frame leads to perform-ance identical to that seen when two masksand two characters appear randomly on eachframe. Thus, a preliminary controlled searchwould have to be extremely fast relative tothe time for each character comparison. Morelikely, an automatic process develops thatdirects the controlled search away from maskpositions and toward character positions.Since such a response would be consistentlytrained and reinforced, it should be learnedin a fashion similar to that for automatic-attention responses.

It should be noted that the developmentof automatic responses to direct controlledsearch is very similar to the process termedby Neisser (1967) as "pre-attentive." Neisserwas trying to explain how search could be


directed to a perceptually segregated subsetof the inputs. We are suggesting that anautomatic process might develop and directthe search in situations where the search-directing response is consistently trained (asis usually true in such studies). We do notfeel, however, that the segregation of therelevant inputs must be based on simplephysical features. With enough consistenttraining, an automatic search-directing re-sponse could develop for a set of stimulidefined by quite complex features. For ex-ample, Gould and Cam (1973) reportedresults suggesting that an automatic processwas directing search to the locations of itemsfrom a set of 10 potentially relevant lettersand away from the locations of items from aset of 8 other letters that never included atarget. (Any target was always drawn fromthe potentially relevant set of 10 letters, buton most trials distractors were chosen fromthis set of 10 as well as from the set of 8).

Although the evidence is not yet available,we raise the possibility that the converse ofthe argument in the preceding paragraphmight also hold: Even with perceptual seg-regation of the display into categories deter-mined by simple physical features, auto-matic restriction of the search to the relevantcategory might not be possible unless therelevant category is consistent across trials.

4. Attention Tasks

The models applying in attention studiesthat utilize detection or search paradigmsare just those discussed in the precedingsections. The general rule is that instructionalmanipulations affect the order of controlledsearch; increases in load cause the timeneeded for controlled search to increase; andconsistent training leads to the developmentof automatic responses that allow attentionallimitations to be bypassed.

We will say a few words, however, con-cerning attention studies that do not utilizedetection or search paradigms. One class ofstudies utilizes a memory paradigm. Theinputs are manipulated in ways similar tothose used in detection studies, but the sub-ject is asked to recall or recognize the items

at some time after presentation. One exampleis the "split-span" technique reviewed byBroadbent (19S8, 1971). Other examplesare reported by Sperling (1960) and VonWright (1970), who studied cued partialreport following tachistoscopic exposure ofcharacter displays.

In these memory paradigms, essentiallythe same underlying mechanisms apply asin the detection paradigms, though memorystorage rather than detection is the processinggoal. When controlled processing is utilized,the subject has considerable control overstorage: The items that are rehearsed orcoded are the items that will be recalled.If an input is presented that generates anautomatic-attention response, then this in-put will receive attention and tend to berecalled regardless of the nature of the con-trolled process utilized to store the otheritems. An excellent demonstration of thesepoints may be found in Kahneman (197S).

5. Threshold Detection Tasks

An input in a threshold detection task, bydefinition, is presented in such a way thatautomatic encoding on most trials will beincomplete and therefore ambiguous. That is,in Section IV.A.4 we discussed the fact theinput stimulation must be above some thresh-old level for automatic encoding to run tocompletion in an accurate fashion. In Figure12, for example, the "M," if presented nearthreshold, might cause activation of someof the features at the visual feature level, butno features at higher levels. This set ofpartial features will in general be ambiguous—several letters and numbers might be con-sistent with the activated feature set. Thus,a target will on some trials give rise to featuresets consistent with either targets or distrac-tors. Similarly, distractors will on some trialsgive rise to feature sets consistent with eithertargets or distractors. Under these conditionsit is clear even in CM situations that auto-matic-attention responses and automatic de-tection can only develop very slowly, if atall, because the internal representations oftargets and distractors will be mapped con-sistently to responses only on those few trialswhen the encoding happens to be complete.


A similar argument applies if encoding is in-accurate rather than incomplete.

Even in CM threshold tasks, therefore, thesubject will be forced to adopt controlledsearch on most of the trials, namely the trialson which encoding is incomplete. On suchtrials a serial comparison process will benecessary to match the activated featuresagainst the features of the members of thememory set. Occasionally, however, auto-matic encoding will manage to process thetarget completely. Suppose that almost al-ways, those stimuli that are completelyprocessed will be encoded correctly. Supposealso that a CM procedure is being utilized inthe task. In these special conditions, an at-tention response can be learned that mightbe utilized on those trials when completeencoding of the target happens to occur.

A model very similar to this was used byShiffrin and Geisler (1973) to fit the resultsfrom a threshold detection study by Estes(1972). In that model, stimuli automaticallyprocessed to a final, incomplete result werecalled "component detections" and stimuliautomatically processed to the complete(letter) level were called "letter recogni-tions." The model proposed a two-part con-trolled search in which the letter recognitionswere searched first (in parallel through mem-ory, but serially through the display), andthen the component detections were matchedserially, feature by feature, against the mem-bers of the memory set. In light of thepresent work, we would like to propose forconsideration an alternative but similarmodel in which the letter recognitions resultin an automatic-attention response, so thatcontrolled search need be utilized only if thetarget is not among the letter recognitions.In practice, the predictions of these twomodels would not differ greatly.

In summary, it seems clear that controlledsearch, whether at the feature or characterlevel, is bound to be the predominant de-tection mechanism in all threshold detectiontasks, but there is a possibility in CM para-digms that automatic detection can be util-ized on a subset of the trials when the targethappens to be encoded relatively completely.

Our general view of processing, in threshold

detection tasks as well as many others, canbe summarized briefly in the following man-ner: Sensory inputs are encoded auto-matically, resulting in states of evidence(consisting of feature sets) that the subjectutilizes in subsequent controlled processing.While generally accurate, this view is toosimple in several respects. First, it must notbe assumed that there is ever one momentin time at which all activated features aresimultaneously present and available to thesubject for controlled processing. Rather itwill usually be the case that some featureswill still be undergoing the process of en-coding (and hence will not yet be activated)while other features have already been acti-vated and are in the process of being for-gotten. Second, it must not be assumed thatcontrolled processing will follow automaticprocessing in time. Such an assumption is afairly accurate simplification when used inmodels for certain tasks including thresholddetection. However, it is an inherent featureof our theory that automatic and controlledprocessing can often operate in parallel, sothat controlled processing can begin whileautomatic processing is still progressing. Thispoint becomes crucial in situations in whichthe subject is required to give extremely fastresponses (including those paradigms pre-senting above-threshold stimuli), since avery fast response might have to be basedon the features available at a certain pointin time, even though better features mightbe available at a later point in time.

D. Automatic and Controlled Processing inMemory Storage and Retrieval

In this section we shall review briefly howautomatic and controlled processing might beinvolved in short-term retention, learning,and in long-term retrieval.

1. Maintenance of Information in ActiveMemory

We suggest that the control process mostevident to introspection is rehearsal. Atkinsonand Shiffrin (1968) studied many tasks inwhich subjects utilized a continually updatedrehearsal "buffer" to aid their memory sys-


tern. Rote rehearsal is of course only one ofmany control processes that cause items tohave an extended residence in STS; coding,deciding, retrieving, and the like also causea similar result, though these processes areprimarily intended to serve other purposes.Iq general, items given attention of any sortare maintained in STS, including items towhich attention is drawn by an automaticprocess.

Items from which attention has been re-moved do not necessarily leave STS quickly,however. When the load is small and newinputs are minimized, items can remain inSTS for extended periods (up to 40 sec inthe Shiffrin, 1973, study). Another exampleis seen in the overt forced-rehearsal studyreported in Atkinson and Shiffrin (1971). Along list of items was presented and subjectsrehearsed aloud a continually updated listof the three most recent items. As a result,the last three items presented were alwaysrecalled on an immediate test. However, theprobability of recall for items prior to thelast three decreased systematically as a func-tion of the spacing from last rehearsal totest. Thus an item removed from controlledprocessing still requires a period of timebefore it becomes lost from STS (i.e., be-comes inactive). This persistence in theabsence of attention might be called the auto-matic component of short-term maintenance.It is this automatic component of persistencethat was studied by Shiffrin and Cook (seeNote 3) and that determines the basic ca-pacity and persistence of STS. In brief,Shiffrin and Cook propose that the lossprocess is a stochastic mechanism whoserate increases as the amount of similar ma-terial concurrently in STS increases.

2. Coding, or Transfer to LTS

The role of control processes in long-termstorage is well established. Rote rehearsalappears to transfer low-level auditory-verbalcodes to LTS; these are not very useful forlong-term recall but can facilitate long-termrecognition. Coding rehearsal appears tofacilitate long-term retrieval of all kinds. Ingeneral, what is attended to and rehearsed iswhat is stored in LTS. (See Bjork, 1975;

Craik & Jacoby, 1975; Craik & Lockhart,1972; and Craik & Tulving, 1975, for a dis-cussion of these issues.)

The other side of the storage question con-cerns the nature of LTS transfer when at-tention is not directed to an item or sequenceof items. Studies of incidental learning tendto show that subjects learn what they attendto, not what they are told to learn (see Craik& Tulving, 1975; Hyde & Jenkins, 1969,1973; Postman, 1964; Schneider & Kintz,1967).

If controlled processing is necessary forlong-term learning then automatic processingwithout controlled processing should not leadto appreciable retention. One source of evi-dence is found by examining retention fordistractors in CM tasks. Such items pre-sumably have received almost no controlledprocessing. However, it may be assumed thatdue to prior exposures and prior learning thedistractors are given automatic encoding atleast up to the "name" level (for evidencesee Corteen & Wood, 1972; Keele, 1972;Von Wright et al., 1975, among others).Does the automatic encoding that these dis-tractors receive during the many trials of aCM task lead to any retention? Moray(1959) had subjects repeat aloud a prosepassage in one ear while a seven-word listwas repeated 35 times in the other ear. Laterrecognition for the unattended words was atthe chance level. Gordon (1968) showed thata set of four distractors in a CM search taskwas recognized near the chance level evenafter 10 days of practice. Gleitman andJonides (1976) showed that distractors weremore poorly recognized in a CM search taskthan in a VM search task (though, due tothe low levels of practice used in their study,the effect could have been due to the use ofa controlled search for categories in the VMcondition).

Evidence of a rather different sort is foundin reading tasks. If automatic processing doesnot lead to retention, then it might be ex-pected that a skilled reader who automaticallyprocesses surface-structure features and whogives controlled processing to conceptualproperties of a passage would retain littleinformation regarding the surface-structure


details. Bransford and Franks (1971) havecollected data that may be interpreted inthis fashion. Of course, if controlled process-ing is directed toward levels of analysisnormally carried out automatically, then thefeatures attended to will be remembered (seePostman & Senders, 1946).

To summarize, whether or not somenominal storage manages to be eked out inthe absence of controlled or attentive process-ing, it seems clear that any phenomenon thatwould be appropriately called "automaticstorage" is a far less important determinantof LTS learning than controlled processing.

Before leaving the general topic of learn-ing it is interesting to speculate on the de-velopment of complex information-processingskills. Since controlled processing is limited,only a small part of the memory system canbe modified at any one time. After invariantrelations are learned at one stage, the pro-cessing becomes automatic and controlledprocessing can be allocated to higher levelsof processing.

For example, the child learning to readwould first give control processing and thengive automatic processing to various units ofinformation, The sequence of automaticallylearned units might be foreground-back-ground, features, shapes, letters,, words, andmeanings of phrases or sentences. The childwould be utilizing controlled processing tolay down "stepping stones" of automaticprocessing as he moves on to more and moredifficult levels of learning. The transitionfrom controlled to automatic processing ateach stage would result in reduced discrimina-tion time, more attention to higher orderfeatures, and ignoring of irrelevant infor-mation. Gibson (1969, chap. 20) describesthese effects to be three of the major trendsin perceptual development. In short, thestaged development of skilled automatic per-formance can be interpreted as a sequenceof transitions from controlled to automaticprocessing.

3. Retrieval from LTS

Retrieval from LTS has a large automaticcomponent. Sensory inputs result in an auto-matic series of stages of encoding that ac-

tivate many features (and perhaps responses)and place them in STS. Internally generatedinputs also tend automatically to activateassociated information in LTS and to placeit in STS.

In general, the two primary controlledphases of LTS retrieval consist of the pro-cesses the subject uses to search and processthe information in STS that has just beenactivated, and the processes used to alter theprobe information from step to step of theLTS search.

It is important to note that the selectionof probe information is not entirely undersubject control. In fact, the probe informa-tion includes the general contextual informa-tion present in STS at the time of retrieval,in addition to the specific cues the subjectmanipulates to facilitate retrieval. The gen-eral context is only partly under subject con-trol, since much of it is environmentally de-termined and even the internally generatedcontext (e.g., what the subject is currently"thinking about") may be difficult to alterradically. Thus, changes in probe cues willbe under subject control to only a degree,and this fact is one of the factors underlyingretrieval failure.

This extremely brief and superficial over-view should not be allowed to give the readeran impression that distinguishing the auto-matic and controlled phases of retrieval isgenerally a simple matter. Furthermore, thecontrolled phase can be a most complex andmany-faceted process. To give just one ex-ample, Anderson and Bower (1973) considerhow subjects compare test sentences withstored sentences in long-term memory. Inour model, there are two phases to this ex-periment: (a) a retrieval of the appropriateinformational nexus from long-term memorythrough the use of probe information (in-cluding the test question) and (b) a com-parison in short-term memory of the testquestion and the retrieved search set. TheAnderson and Bower model, in the presentview, is primarily a model of the second ofthese processes—the comparison withinshort-term memory of the test sentence andthe retrieved information. See Shiffrin's(1970, 1975b, 1976) studies for a more


elaborate discussion of long-term retrievalprocesses.

V. Models of Search and Detection

In this section the theory we have pro-posed will be compared with some of theprevious models that have been proposed fordetection and search tasks.

A. The Sternberg Model: Serial, ExhaustiveSearch

The serial, exhaustive search model is pre-sented in Sternberg (1966). Extensions,further evidence, and reviews of the litera-ture are given in Sternberg (1969a, 1969b,197S). This model is meant to apply tomemory search tasks (F — 1, M varies)with small memory-set sizes. It has beenconfirmed impressively often but is limitedin scope, as it is meant to apply only to asmall range of paradigms. In our presenttheory, serial, exhaustive search is consideredto be one of the controlled search strategies,one that is very commonly adopted.

The factors that lead subjects in our situ-ation to adopt terminating, controlled searchremain uncertain at present. Possibly themuch larger search loads in our study led toa terminating strategy. One other minor dis-crepancy in results involves the Sternberg(1966) finding apparently showing fixed sets(CM) and varied sets (VM) to give equiva-lent set-size functions. In retrospect we cansee that Sternberg's subjects in the fixed setprocedure were given far too little training,and the fixed sets were varied too often forautomatic detection to have developed.

1. The Relation Between Sternberg's Modeland Our Framework

Reaction time was proposed by Sternberg(1969a, 1969b) to be the sum of motor re-sponse time and the times for four additivestages: (a) encoding (affected by stimuluslegibility), (b) serial comparison (affectedby size of the memory set), (c) binary de-cision (affected by whether a positive ornegative trial has occurred), (d) translationand response organization (affected by rela-

tive frequency of positive and negativetrials). In Figure 11, in which our moregeneral framework is depicted, these stagesproposed by Sternberg may be placed asfollows: "Encoding" is equivalent to auto-matic encoding; "serial comparison and bi-nary decision" both are part of controlledprocessing; "response organization and mo-tor response execution" are part of responseproduction. In conclusion, then, the Stern-berg model differs from the detailed quan-titative model we fit to our data in Part I,but nevertheless may be viewed as a specialcase of our general framework.

2. Problems for the Serial, ExhaustiveComparison Model and Their Resolution

The serial, exhaustive model is particu-larly important on account of the research ithas generated that give results not consistentwith the theory in its simplest form. Theseinconsistencies have led to many new models,some of which will be discussed below. Theapparent inconsistencies include the follow-ing results.

1. Reaction times depend on the serialpresentation position of the tested itemwithin the memory set (in VM procedureswith memory sets changing every trial).The most pronounced effects occur when thememory-set items are presented quickly andthe display item is presented very soon afterthe last memory-set items (Burrows &Okada, 1971; Clifton & Birenbaum, 1970;Corballis, 1967; Klatzky, Juola, & Atkin-son, 1971; Klatzky & Smith, 1972).

2. Reaction times depend on the relativefrequency of presentation of items withinthe memory set in CM procedures (see Bie-derman & Stacy; Krueger, 1970; Miller &Pachella, 1973; Shiffrin & Schneider, 1974;Theios et al., 1973).

3. Reaction times can be speeded for cueditems in VM paradigms (Klatzky & Smith,1972) or speeded for expected items in CMparadigms (Shiffrin & Schneider, 1974; butnote that the fixed sets changed every 1,60trials so that only low degrees of practicewere involved), or speeded in a VM para-digm for an item repeated during the pre-


sentation of the memory set (Baddeley &Ecob, 1973).

4. The outcomes in CM paradigms oftendiffer in fundamental ways from those inVM paradigms, especially when some simplephysical basis separates the memory and dis-tractor sets, or when the degree of trainingis high. Many such findings have been dis-cussed extensively in earlier sections of thispaper and will not be reviewed again here.

5. Subjects typically give incorrect re-sponses on 1-10% of the trials, dependingupon the condition. The serial search modelsdo not generally posit explicit mechanismsto predict errors.

Before turning to alternative models, it isuseful to consider how these phenomena aredealt with in the framework of our theory,and also to see how Sternberg explains theresults.

The various findings in CM paradigms areeasiest to explain: Automatic detection de-velops that enables the serial search to bebypassed. Much of the research in this paperwas directed toward establishing this fact.Sternberg (1975) is less specific but also sug-gests that some alternative, more efficient,search process is used in such situations.

Within varied-mapping paradigms, re-course to the hypothesis of automatic detec-tion cannot be used to explain the results.It would be parsimonious if each of theabove findings (1) to (3) proved to be theresult of a common mechanism. Shiffrin andSchneider (1974) proposed one possible mech-anism—namely, that when information con-cerning test probabilities is available to thesubject, then one item is placed in a specialstate prior to each test display (called astate of "expectancy"), and that a test ofthe expected item results in a faster responsetime than tests of nonexpected items. Stimu-lus probability, serial position, cueing, stim-ulus repetition, or differential importancecould all be expected to determine the prob-ability with which stimuli will be expected.

The effects of expectancy could possiblyact at several different stages en route to theexecution of the response. One possibility,also mentioned by Sternberg (1975), is that

the comparison process is carried out in aspeeded fashion for an expected item. An-other possibility is that the encoding processor the response production stage is speededwhen an expected item is tested. Shiffrin andSchneider (1974) attempted to carry out atest discriminating among the possible mod-els. While the general notion of expectancywas supported by the results, the study wasnot conclusive in determining which versionof the model was to be preferred.

The important point to be emphasized isthat we (and Sternberg also) suggest thatfindings (1) to (3) listed above are not in-compatible with a controlled search processthat is serial and exhaustive. We suggest thesefindings can be explained by factors that af-fect other stages of the response process thanthe comparison stage or perhaps by a factorthat affects the comparison process only oncertain trials. Many other researchers havepreferred to discard the hypothesis of a serial,exhaustive comparison process. Some of themodels proposed as alternatives will be dis-cussed below.

The final problem for the serial, exhaustivesearch model is posed by the occurrence oferrors. The explanations of controlled searchprocesses by Sternberg (and by us) for re-action time tasks do not propose an explicitmechanism by which errors may occur. Manyinvestigators have simply ignored errors aslong as the error rate rate is low, under 5%,say. This is perhaps justifiable if errors areanomalous events that do not interact withany of the other variables being studied. In-deed the stability of findings across numerousstudies that do not attempt to fix error ratesat any given value and that vary consider-ably in the observed rates (in the range 0-10%) lends some support to the view thatignoring errors will not distort the conclusionsdrawn from the reaction time data.

In principle, however, error rates cannotbe ignored. Pachella (1974) has shown thatinstructions used to change error rates byjust a few percent can cause considerablechanges in the level and pattern of reactiontimes. There are many ways in which error-producing mechanisms can be appended tocontrolled search models. Two of the most


important are as follows: First, the subjectmight terminate, or be forced to terminate,the controlled search before the search iscompleted. Then a response would have tobe made as a guess based on partial informa-tion at most, or perhaps no response at allwould be made (an omission). These typesof error resulting from early search termina-tion account for almost all errors in searchand attention tasks using above-thresholdstimuli with accuracy as a measure; in par-ticular, most of the errors in the accuracytasks reported in the present paper and inPart I are errors of the type. Second, thesearch process might be carried out incor-rectly; that is, a comparison might be car-ried out incorrectly owing to confusionsamong items, forgetting, misperception, andthe like. When error mechanisms are ap-pended to the basic search model a new ex-panded theory results, which must be testedthrough joint consideration of error data andreaction time data.

Two basic approaches may be utilized totest search theories incorporating error pre-dictions. One approach involves manipulat-ing the error rate systematically within eachcondition; the manipulation may be carriedout through instruction, deadline training,or signals-to-respond (Reed, 1973). This ap-proach has the advantage of making full useof both the error and reaction time datafrom a single experiment. It has the dis-advantage that the manipulations of errorrates may affect the nature of the controlledsearch strategy adopted by the subject. Forexample, Reed (1976) carried out a remark-able series of tests of a wide variety of modelsusing data collected in the signal-to-respondprocedure. Unfortunately, though one modelcould be singled out in preference to theothers, the basic data differed in importantrespects from those ordinarily found in thesimpler version of the same paradigm.

The alternative approach involves carryingout two separate studies involving the samesubjects. One study utilizes the usual re-action time methodology, with instructionsto keep error rates low. The second studyutilizes a paradigm in which the availablesearch time is systematically varied and the

error rates corresponding to each amount ofavailable search time are collected. The modelderived from one study can then be used topredict the results of the other. This is themethod that was adopted in Experiments 1and 2 of Part I. This approach has the ad-vantage of relating two fields of inquiry thatare normally treated separately. In thepresent paper, for example, attention tasksusing accuracy as a measure were linked tosearch tasks using reaction time as a measure.

It should be noted, by the way, that thetype of error predicted for the results of PartI/Experiment 1 is that caused by prematuretermination of controlled search whose char-acteristics were derived from the reactiontime results of Part I/Experiment 2. How-ever, the errors seen in Part I/Experiment 2were not necessarily caused by the samemechanism. In fact many, if not most, ofthem may have resulted from confusions,forgetting, or anomalous condition-indepen-dent factors.

B. The Theios Model

There are really two separate treatmentsto be discussed here: the specific micromodelused by Theios et al. (1973) to predict re-sults that would otherwise be fit by a serial,exhaustive model, and the general theoryused by Theios (1973, 197S) to describe theproduction of response times in a variety oftasks.

1. Serial, Terminating Search Through a Listof Memory Items and Distractors

In this model a list is constructed in mem-ory, a list on which appear all items thatmight be displayed for test. Each of theseitems has an associated response cue (posi-tive or negative) attached to it. The memory-set items and distractors may in general beintermingled in this list, but a variety offactors affect list ordering, and in many casesthe memory-set items may tend to appearearly in the list. During the test, the subjectserially scans the list until the test item islocated, then terminates the search and re-sponds according to the cue information


located along with that item. A group ofitems at the end of the list is assumed, how-ever, to be searched in parallel, in one step,if search has not previously terminated. Theselast items are supposed to be accessed throughLTS, while those scanned serially are in STS.(There are a variety of complications andextensions added to this model that weshall not discuss.) In at least some situations,this model can approximately predict linearset-size functions that are parallel for nega-tive and positive trials.

Various versions of the model were fittedto data collected by Theios et al. (1973).They utilized a CM procedure with a moder-ate amount of training. It seems likely, there-fore, that many of the observed effects intheir study, especially those based on differ-ences in stimulus frequency, resulted fromthe development of at least a small degree ofautomatic detection. The more frequent itemswould come to elicit automatic responsessooner, causing the observed reduction inresponse reaction times. In our terms, thesubject's strategy was probably a mixtureof automatic detection and controlled search,making simple conclusions difficult. Never-theless, even if automatic detection is notinvolved, it may be asked whether modelsof the complexity suggested by Theios (orsuggested by the comments above) are neces-sary to fit his data. Shiffrin and Schneider(1974) showed that a simple version of theexpectancy model using serial, exhaustivesearch through the positive set could providea fit to the data about as good as that pro-vided by the more complex models of Theioset al. (1973). Under these circumstances thedata do not provide strong support for theTheios model.

Aside from questions concerning the de-tails of the Theios et al. (1973) study, onecan address the more general hypothesis putforward by these investigators. This hy-pothesis suggests that most memory searchstudies are best conceived of in terms of a(rather complex) serial, terminating searchmodel. Although we have collected dataarguing for a serial, terminating searchthrough the memory set (Experiment 2, PartI), we agree with Sternberg (1975) that a

serial, terminating search is not a good can-didate as a model for the simpler para-digms in which linear, parallel set-size func-tions are collected (see Sternberg, 1975, forthe relevant arguments). There are manyadditional reasons for us to reject the hy-pothesis of Theios et al; these reasons arisefrom the results of the present paper. Bas-ically, the Theios et al. model was proposedto deal with a variety of stimulus prob-ability effects that arise in CM paradigms.We have shown in this article that indeed adifferent detection process develops in suchparadigms; however, the fully developedautomatic process and the pure controlledprocess are two different mechanisms andboth are relatively simple. In effect, Theioset al. have been led to propose a rather com-plex unitary model to try to fit data thatprobably reflect a mixture of two differentsimple search mechanisms.

2. A Hybrid Serial-Parallel Model

A much more general theory of responsetime generation in search tasks has beenpresented by Theios (1973, 1975). Thistheory has many similarities to the theorypresented in the present article. It postulatesinput and identification stages (which wewould call automatic encoding), a responsedetermination stage (which in our terms iseither a controlled search or an automaticprocess or some combination, depending onthe paradigm), and response program selec-tion and response output stages (which weterm response production). Theios's de-scriptions of the situations in which auto-matic processing should occur are quite sim-ilar to those we suggest in this paper. Hisdistinction between controlled and automaticprocessing is not made as explicit as in thepresent article, and in his view the use ofcontrolled search is tied more to stimulus-response compatibility than to the type andamount of practice (i.e., the consistency ofthe mapping). Furthermore, the details hepresents concerning controlled search pro-cesses are somewhat different and more limitedthan we have presented in the present treat-ments. On the whole, however, the Theios(1975) theory is quite compatible with the


general approach taken in this paper; infact, we suggest that interested readers readthat paper for additional evidence regardingautomatic processing in reaction time tasks.

C. Parallel Processing Models

Certain parallel processing models havebeen proposed for VM search paradigms asalternatives to the usual serial models. Atkin-son et al. (1969), Murdock (1971), andTownsend (1971) have all presented versionsof such models. The Townsend model canserve as an example. In this model the timeto complete any one comparison is distributedexponentially with a rate constant that de-pends on the number of current items under-going processing. When any item completescomparison, the rates are immediately re-adjusted.

Such models almost completely mimicserial models. The same set-size predictionsare derived as for serial models, whether ornot termination of search is assumed. Thereare some minor problems with these models.For example, the exponential assumptionimplies a particular relation between thegrowth of the mean reaction time and thegrowth of the variance of the reaction time,with set size. More important, however, thissort of parallel model is not conceptuallyvery different from the serial models—com-parisons occur, on the average, at evenlyspaced intervals of time for each of the itemsof the memory set (or the display set). Weprefer, therefore, to think of such models asmembers of a class also containing the serialmodels. As Sternberg (1966) has shown,some types of parallel models, in which eachitem is processed independently of the num-ber of alternative items, cannot predict theobserved results.

D. Direct Memory Access Models Based onTrace Strength

A number of models have included a searchmechanism that resembles in certain respectsthe automatic detection process presented inthis paper. In these models, an item presentedfor test is encoded directly to some location

in long-term memory, a location containinginformation enabling the correct response tobe given. These models are most applicableto CM paradigms, although they have some-times been applied to VM situations and al-though some have been elaborated to containserial, short-term search as an additionalprocess. Generally, in these models the tracestrength of the code in long-term memorydetermines the ease of access and the re-sponse time for a given item.

Speaking generally, it is our feeling thatsuch models capture part of the process wehave described as automatic detection. Note,however, that the two concepts are notidentical. An example will make this pointclear. Suppose we define the memory set toconsist of all words whose fourth letter isalphabetically prior to the second letter. Aword presented auditorally will presumablybe encoded automatically until the long-termmemory node is located that contains thespelling. Then a controlled process will checkthe spelling to see whether the criterion issatisfied. Such a process is not "automaticdetection." If several items are presented atonce, each would have to be checked serially.This is an example of a task designed so thatthe information found in long-term storecorresponding to a given input is sufficientfor a decision to be made correctly, withoutconsidering other members of the memory set.This property could hold for either VM orCM tasks, regardless of task-specific train-ing and regardless of whether an attention (orother task-specific) response has been learned.

The automatic detection mode is thereforesimilar to the search mode of trace-strengthmodels in that each input is encoded auto-matically, and relevant information is locatedin LTS. If the information located must beprocessed in limited, controlled fashion (sothat multiple inputs would have to be de-cided about sequentially, for example) thenwe would not consider the process to beone of automatic detection. On the otherhand, if the information located is attentiondirecting, or even response eliciting, then theprocess would be one of automatic detection.Several varieties of trace-strength modelsmay be distinguished. We consider these next.


1. Trace-Strength Discrimination Models

In these models, the information found inmemory for any tested item consists of aunidimensional value of strength (often in-terpreted as familiarity). The task is de-signed so that the response can be based onthe retrieved strength value, and it is as-sumed that criteria can be chosen so thaterrors are kept low in frequency. Examplesof such models are found in Corballis, Kirby,and Miller (1972) ; Nickerson (1972) ;Baddeley and Ecob (1973); Cavanagh(1973, 1976), and Anderson (1973).

There are many tasks in which we regarda response strategy based on trace strengthto be highly plausible, whether the tracestrength is utilized in a controlled decisionor used to initiate an automatic response.But we must ask whether such models arereasonable when applied to typical VMsearch paradigms. There is no question thatthe models are capable of generating ac-curate predictions if strengths for items invariously sized memory sets are appropri-ately adjusted. However, a model with thismuch freedom could predict anything andwould not be interesting. In practice, ofcourse, various assumptions are made togovern the possible changes in trace strengthsacross conditions. The extant models, how-ever, do not appear to us to have the sim-plicity and elegance of serial search theorywhen both are applied to VM search tasks.

The models of Corballis (1975) andCavanagh (1976) both assume that tracestrength is affected by rehearsal (called "se-quential priming" by Corballis). Both modelshave been developed primarily in response tothe findings of serial position effects. Asdiscussed earlier, such effects are not incom-patible with serial search, or even serial,exhaustive search. Thus if rehearsal affectssearch in important ways it might do sobecause it controls the trace strengths usedto respond (in the trace-strength models), orbecause it affects serial search order (in aterminating search), or because it affectsencoding or response time rather than com-parison time (in a serial, exhaustive model).Future research will be necessary to establishwhich, if any, of these possibilities is true.

2. The Atkins on-Juola Model

This model (Atkinson & Juola, 1974) wasconstructed to explain the results of searchstudies in which a large, previously learnedensemble of items forms the memory set.The model assumes that the test item isevaluated in terms of its familiarity—a valueabove a criterion leads to a positive response,a value below a lower criterion leads to anegative response, and a value between thetwo criteria leads the subject to carry out aserial search of the list. Such a model isquite compatible with our theory as appliedto a task of this kind, but there is at leastone problem requiring further research. Theestimated serial search rate is only about10 msec per item. If the serial search modewere similar to that in the usual procedurethen a rate nearer 40 msec would be ex-pected. There are several hypotheses to ex-plain the discrepancy—for example, it is pos-sible that the search is somehow restrictedto a subset of items that are similar to thetest item. A somewhat different hypothesiswould suggest that no serial search occursat all, but that some rechecking of thefamiliarity value is necessary when the firstobservation is between the criteria, and thatthe confusability of the test item, and hencethe need to recheck the test item, will de-pend on the list length.

In paradigms like that of Atkinson andJuola (1974), where trace strength or famil-iarity is used as a basis for the response(thereby sometimes bypassing a serial search),it might be asked to what degree the detec-tion is an automatic process. Certainly theencoding of the stimulus and the generationof a familiarity value is an automatic process.The decision how to respond for a givenvalue of familiarity could very well be acontrolled process initially, though with suffi-cient practice in the task, the decision andresponse initiation for extreme familiarityvalues might also become automatized.

E. Some Conclusions about Search Models

Each of the models we have considered hassome elements in common with our generaltheory. At the same time, each has been


applied to some paradigms that we thinkwould be better handled by an alternativeprocess. The direct access trace-strengththeories, for example, have a close affinitywith our automatic detection mechanismand have a natural application to CM para-digms, tasks with perceptual cues separatingthe memory and distractor sets, or tasks inwhich associations in LTS to each test itemcontain information enabling a response tobe made. Such models should not, however,be applied, in our view, to VM paradigms ofthe type that are usually studied in thelaboratory. For such tasks serial, controlledsearch seems a more attractive possibility(or at least limited controlled search of onesort or another). Much of the present articlehas been devoted to demonstrating that twoqualitatively different detection or searchmodes exist and to establishing the conditionsunder which each is utilized. Many of themodels to date have attempted, we thinkunsuccessfully, to apply a single mode ofsearch to all the search paradigms.

The second point we wish to emphasizeis that many of the prior models have beenconstructed to deal with just a few studiesor just a few results from those studies. Thehistory of investigation in this area hasdemonstrated beyond doubt that numerousmodels are capable of predicting results fromany given study. A proper evaluation ofmodels should incorporate two tests: (a)Can the model predict a wide variety ofresults in differing paradigms, and can itpredict results in both the reaction time andaccuracy domain? (b) Can the model predictresults from a single series of studies on thesame subjects, a series in which most of thecommonly examined variables are manipu-lated?

VI. Models of Selective Attention

A. The Broadbent and Treisman Models

The Broadbent (1971) model supposesthat there are two basic selective processes,one leading from the physical input to a setof internal codes that serve as the evidence,and a second leading from the internal codesto categories and responses. The selection

in Stage 1 is called "filtering" or "stimulusset," and that in State 2 is called "pigeon-holing" or "response set." There is a veryshort-term sensory-information store (lessthan 1-sec duration) called a "buffer," whichaccepts information in parallel from thephysical inputs. (This store is equivalent tothe more peripheral information in STS inour theory.) The filter selects informationfrom the buffer and sends it through alimited-capacity channel. In the Broadbent(1958) theory the filter allowed informationfrom only one source at a time to continuethrough the system. In the theory proposedin 1971, Broadbent's earlier view was modi-fied in accord with data put forth and theorysuggested by Treisman (e.g., 1960, 1969).This modified theory suggests that the filterdoes not block, but only attenuates, informa-tion from nonattended sources. The states ofevidence in the system are the output of thefilter.

While there are obvious elements of sim-ilarity between our approach and Broad-bent's, we would like to focus on the differ-ences :

1. The Role of Control Processes

In our theory, selective attention is largelydefined in terms of control processes. Selec-tivity due to structural characteristics of theprocessing system, even structure that hasbeen learned, is not considered to be an at-tentional process. In the Broadbent model,the distinction between the structural, learnedcomponents of selectivity and those undersubject control are not as clearly distin-guished. One consequence of Broadbent's ap-proach is that the type and amount of train-ing is not implicated as a particularly im-portant determinant of the presence or ab-sence of attentional deficits. Much of thedata in the present article, however, hasshown how the conditions of practice (par-ticularly the consistency of the mapping ofattention to a feature to a reinforced out-come) determine the development of auto-maticity, the bypass of controlled search, andthe elimination of attentional limitations.

The emphasis on the role of control pro-cesses in our theory also implies that we are


advocating an active rather passive view ofattentional selectivity. In our theory atten-tional selectivity is largely the result of ac-centuation of certain informational elementsthrough the use of control processing; in theBroadbent-Treisman approach, on the otherhand, inhibition and attenuation of the pro-cessing of certain inputs play the most im-portant role in selectivity.

2. Selectivity Leading to the Internal Statesof Evidence

In our theory the internal states of evi-dence are the result of automatic processingof sensory inputs, with subject control af-fecting sensory encoding only indirectly andto a small degree. On the other hand, theselectivity that results from filtering playsa large and fundamental role in Broadbent'stheory. This fact alone would not be a sourceof discrepancy between the theories if thesefiltering processes reflected only relativelypermanent, structural components of selec-tivity. However, the examples of filteringgiven by Broadbent make it clear that thisprocess is intended to reflect temporary shiftsin attentional control. Thus, in numerousstudies, filtering is said to select inputson the basis of a physical difference, likelocation. In some of these studies the samephysical cue is always assigned consistently,so that automatic detection (in our terms)could have developed, but the number oftrials is usually so small that controlled pro-cessing probably was utilized. In other ofthese types of studies the physical cue wasassigned randomly across trials, so that selec-tion must have been based on temporary con-trol processes (see Broadbent, 1971, chap.S). In either class of studies, Broadbent sug-gests filtering as the basis for selection, withattended inputs presumably resulting in abetter state of internal evidence than non-attended inputs.

We take a different view. We propose thatautomatic processing results in roughlyequivalent internal states, but that controlledprocessing must examine these internal statessequentially. Thus information on an at-tended channel is examined first, and is there-fore detected and recalled more effectively

than information on a nonattended channel.We will discuss below some studies byShiffrin and his colleagues that seem to pro-vide strong evidence in favor of this view-point.

3. Short-Term Store and Levels of Processing

In the Broadbent theory there is a short-term buffer more or less corresponding to ashort-term memory for peripheral informa-tion. In the original 1958 theory this wasthe sole basis for short-term memory, but inthe 1971 theory, a later short-term systemcalled "primary memory" was added sub-sequent to the filter and the limited-capacitychannel. Thus the new theory is in close cor-respondence with the approach of Atkinsonand Shiffrin (1968).

However, the Broadbent theory identifiesselectivity with particular levels of processing:Selectivity operates after the buffer and priorto primary memory. We, on the other hand,treat all short-term memory as a single con-tinuum consisting of the results of automaticencoding as well as inputs from LTS. ThusSTS consists of information at a wide varietyof levels of processing. In our view, selec-tivity is not restricted to any special levels.Rather, selectivity operates at whatever levelcontrolled processing is utilizing. If con-trolled processing is checking color serially,then attentional selectivity will appear at theperipheral level of hue, while if words arecompared serially to see which is a synonymof a memory item, then attentional selectivitywill appear at the central semantic level.

4. The Role of Timing

An important concept in Broadbent'stheory is that of "switching time." Broadbentinterprets the results of many studies byassuming that attention (the filter) maynot be redirected from one source to anotherbefore the lapse of some minimal time, calledswitching time. Although some of the resultsBroadbent ascribes to switching time limita-tions are due in our theory to other mech-anisms, we have no objection to argumentsthat attentional limitations are rooted inlimitations on the rate of processing opera-tions. To the contrary, we have carried this


argument much further and have suggestedthat the time utilized during controlled pro-cessing is in many tasks the immediate,proximal cause of the inability to divideattention. The results of Experiments 1 and2 in Part I, and our analysis of them, wereintended to demonstrate the role of timingby relating the accuracy of performance inattention tasks to the reaction times producedin search tasks.

B. The Shiffrin (1975a) Model

The work by Shiffrin and his colleaguessummarized by Shiffrin (1975a) may be re-garded as a preface to the present studies andpresent theory. The primary aim of theearlier studies was a demonstration that en-coding quality is virtually unaffected by at-tentional instructions. The results from aseries of experiments demonstrated that in-formation from a source (location) is pro-cessed as well when it is the only informationrequiring processing as when it arrivessimultaneously with other information alsorequiring processing. If filtering or attenua-tion were occurring between stimulus inputand the production of internal states of evi-dence, then attention would have been al-located less effectively when simultaneousinputs were used, and performance wouldhave suffered.

It could well be argued that the resultswere obtained only because attentional ca-pacity had not been exceeded. In effect, thisis our own argument. Such an argument doesnot help resuscitate the views that earlyfiltering, attenuation, or blocking of pro-cessing takes place, since the situations inwhich the results were obtained were as com-plex at the stimulus input side as most ofthe studies in which attentional deficits arefound.

It should be noted that the Shiffrin(197Sa) results are easily accommodated inBroadbent's general theory. It must be as-sumed that the various inputs are retained inthe sensory buffer long enough that they mayall be passed successfully through the limited-capacity system (in essence this is our ex-planation, also, if "sensory buffer" is replacedby "STS"). The features in the sensory buffer

must be high-level ones, however, because inthe studies masks were used to delete low-level features from storage. Thus such resultscall into question the need for two selectiveprocesses, one before and one after genera-tion of states of internal evidence. We argueinstead that only one controlled selectionprocess occurs, after automatic encoding hasproduced states of internal evidence. Evenif it should eventually be determined thatsome small degree of selection occurs duringinitial perceptual processing, the accumula-tion of data makes it clear that the magnitudeof attentional effects due to controlled pro-cessing after automatic encoding is fargreater than the magnitude of any effects dueto early selection. For further discussion seeShiffrin & Gardner (1972), Shiffrin, Craig, &Cohen (1973), Shiffrin, Gardner, & Allmeyer(1973), Shiffrin & Grantham(1974), Shiffrin,Pisoni, & Casteneda-Mendez (1974), andShiffrin, McKay, & Shaffer (1976). The hy-pothesis supported by these studies, that ofa single selection phase following automaticencoding, has been proposed by a number ofresearchers and theorists. We turn now tothese proposals.

C. The Deutsch and Deutsch Theory

As preface to the Deutsch and Deutsch(1963) model and to other similar theories,we will review some results in auditory selec-tive perception from shadowing or dichoticlistening tasks. Then we will consider theDeutsch and Deutsch treatment, objectionsto it, and our present view.

1. Experiments on Auditory Shadowing andDichotic Listening

Starting with Cherry's (1953) study, nu-merous studies have utilized a shadowingtechnique in which the subject repeats con-tinually a stream of speech (usually presentedto one ear) while other messages are pre-sented simultaneously (usually to the otherear). Moray (1959) showed that informationon the nonshadowed ear could not be recalled,recognized, or relearned with savings. How-ever, the subject's own name on the non-shadowed ear could be noticed and recalled


at the end of the experiment. The traditionalexplanations for such results hold that theinformation presented to the nonattended earis greatly attenuated, but that some informa-tion, enough to let highly salient inputs benoticed, does pass the filter. The basis forthe types of information that are noticed onthe nonshadowed ear is usually considered tobe rooted in the content of the input. Theearlier treatments supposed that simple phys-ical cues served as a basis (not only inBroadbent, 19S8, but also in Neisser, 1967).For example, two messages in the same voiceare confused, but if the messages are in aman's voice and a woman's voice, then thenonshadowed message can be ignored. Anumber of studies by Treisman (see 1969 fora review) showed that content of messagescan also serve as a basis for selection thoughusually not as effectively as physical cues.

2. The Deutsch and Deutsch Approach

Faced with selection on the basis of con-tent, Deutsch and Deutsch (1963) suggestedthat all inputs are analyzed to a relativelyhigh level and that the results of the pro-cessing are then used to select certain stimulifor further processing, for memory, and forresponse. We are in essential agreement withthis view, though we have elaborated on itconsiderably and introduced the importantdistinction between controlled search andlearned attention responses leading to auto-matic detection.

The Deutsch and Deutsch model has notyet gained general acceptance and in par-ticular has been rejected by Broadbent(1971) and Treisman and Riley (1969).Within the framework of shadowing studies,several objections have been raised to theDeutsch and Deutsch position. First, selectionwithout a physical cue remains difficult evenwith wide variations in content. Second, dif-ferences in content affect selection very littlewhen a physical cue is present. Third, selec-tion of a given word in a dichotic sequencedepends on the previously attended word andnot on the previously presented, but un-attended, word.

Our present theory, with its distinction be-tween controlled search and automatic detec-

tion, overcomes these objections. If the stim-uli presented do not cause automatic-atten-tion responses, then a controlled search situa-tion (like that of Experiment 4a) obtains,and a similar explanation holds: The subjectwill carry on attention-demanding, controlledprocessing primarily on the shadowed or at-tended message, with only minimal controlledprocessing of the nonattended information,just enough to establish which informationis to be given deeper processing. Physicalcues, when available, will make this relevancydecision easy. On the other hand, if auto-matic-attention responses have been pre-experimentally attached to stimuli, thesestimuli will be attended to and rememberedeven when they are presented on a to-be-ignored channel. This fact explains the re-call for the subject's own name, for example.Furthermore, the objection that selection isdifficult without a physical cue does not hold,since we have shown that consistent train-ing leads to the development of automatic-attention responses even without the pres-ence of a simple, consistent physical cue(though a great deal of training may benecessary—see Experiments 1 and 3).

To return to the original controversy, someof the most persuasive evidence in favor ofthe Deutsch and Deutsch position has beenthe subject's ability to divide attention with-out deficit, even in VM situations when auto-matic detection could not have been operat-ing. Such data have often been discounted byBroadbent and others on the following basis.It is argued that relatively unanalyzed in-formation is held temporarily in a sensorybuffer; thus it should not be surprising thatmultiple stimuli can pass through the filteras long as a free period exists after presenta-tion that allows the information to be passedserially from the buffer through the filter.This argument is used to explain the ap-parent ability to analyze multiple simulta-neous stimuli in a variety of tasks and isthereby used to discount much of the evi-dence supporting the Deutsch and Deutschposition. Broadbent's argument, however,does not account for the results of our multi-ple-frame studies, for in these studies therewere no blank periods to allow serial process-


ing, and yet the rate of presentation couldeven be speeded under the consistent (CM)training conditions. We do not disagree thatunder VM conditions a postpresentation freeperiod will allow simultaneous inputs to behandled serially by the attention system. Wemerely wish to point out that this explanationcannot help explain the relationship betweenour VM and CM conditions in the multiple-frame tasks. Instead, Broadbent and Treis-man would have to argue that the salience ofthe consistently trained targets is so highthat the filter plays no role in processingwhatsoever, in which case it might as well beargued that automatic processing bypassesthe selective system.

In summary, we argue that the positionof Deutsch and Deutsch, when extended inthe manner of our present theory, is bothfully defensible and in better accord with thedata than the opposing views.

D. The Norman Model

Norman, alone and with coauthors, hasbeen responsible for a number of rather dif-ferent models. We consider here the versionmost closely in accord with that of Deutschand Deutsch (Norman, 1968, 1969; see alsoLindsay & Norman, 1972). This model isbasically an extension of the Deutsch andDeutsch approach in the sense that it pro-poses a relatively complete analysis of all in-coming stimuli. As in the Deutsch andDeutsch approach, a mechanism is proposedby which certain of these complete encodingsare selected for further serial processing. Thismechanism is called "pertinence."

In the model, each set of encodings is as-signed a value of pertinence that determinesthe order of further processing. The value ofpertinence is assumed to derive from twosources: the stimulus encoding and an anal-ysis of previous inputs. The item selected forfurther analysis is that one whose pertinencevalue is highest as determined by a com-bination of sensory encoding and previousanalysis. In very general terms this model ishighly similar to the theory we have proposed.

However, the Norman model fails to makethe important and necessary distinction be-tween automatic-attention learning and con-

trolled processing. Thus extrapolating fromNorman's theory, visual search for a singlememory item, even in a VM task, should bealmost capacity-unlimited and parallel. Thesubject should, by hypothesis, raise the per-tinence value of the single item, and hencethis item should be processed first. However,all of our studies have demonstrated thatsuch pertinence changes cannot take place ona trial-by-trial basis. It is only after con-siderable CM training that an automatic-attention response develops (or, in Norman'sterms, that pertinence is raised). Thus a sub-ject may wish to find a particular stimulusbut does not in general have the attentionalcontrol to enable the stimulus to be locatedwithout a serial, limited, controlled search.

E. The Neisser Model

Neisser (1967) has proposed a generalmodel that postulates an early stage of par-allel processing not under subject controlcalled "preattentive" (similar to our sys-temic, automatic stage), and a later, controll-able, serial stage referred to by the term"focal attention." In many respects this gen-eral approach is quite similar to the presenttreatment. For example, Neisser discusses insome detail the fact that preattentive pro-cesses can control attention or responses (asin eye movements). Furthermore, it is sug-gested that practice can lead to search be-coming dependent on preattentive mecha-nisms, and hence lead detection to becomeautomatic.

Neisser's own treatment of his visual searchdata differed in some respects from our presenttreatment. For example, Neisser argued thatirrelevant targets in CM designs are not pro-cessed in as much detail as relevant targets.We argue that automatic analysis is equal,but that attention is drawn to the relevanttargets. Furthermore, there are some differ-ences in emphasis between Neisser's and ourmodels. For example, Neisser tends to em-phasize that preattentive discriminations arebased on relatively crude physical differ-ences among stimuli, whereas we feel "pre-attentive" processes are determined byconditions of training and practice, with phys-ical differences affecting the rate of develop-


ment of automaticity. However, such differ-ences should not be allowed to mask thesimilarity of the general approaches.9

F. The LaBerge Model

LaBerge (1973, 1975) presents an atten-tional model, in the context of reaction timematching tasks, that has many elements incommon with the present theory. He sup-poses, like Deutsch and Deutsch, that thereexists an automatic, learned encoding systemthat operates in parallel on input stimuli andproduces features independently of controlled,attentive processing. Depending on priorlearning, there are no limits to the depth ofsuch processing. Subsequent to such auto-matic encoding, controlled processing re-quiring attention is necessary to carry outfurther analyses of the input. The featuresresulting from automatic encoding are as-sumed to be capable of calling attention tothemselves. LaBerge proposes, and demon-strates empirically, that new automatic en-codings, that is, perceptual learning, mayoccur with practice. Finally, LaBerge em-phasizes that perceptual learning takes placemainly after high accuracy is achieved. All ofthese aspects of the system are quite similarto ours.

One additional element of the LaBergemodel holds that attention may be directedtoward a not-yet-learned feature, causing itto act temporarily as an automatically learnedunit, but only while attention is maintained.That is, a feature that would normally re-quire extended controlled processing to beencoded from automatically activated, sim-pler subunits, can, when attention is focusedon it in advance, be activated by the per-ceptual encoding process. This suggests atleast some degree of control over encoding.While we have downplayed the role of suchcontrol, we have not ruled it out. In fact,we suggested that the results from Part I/Experiment 2, when M was equal to 1, mightwell be explained by such a mechanism.

If there is a difference between our ap-proach and LeBerge's, it lies in the differentareas of applications and the completenessof the assumptions. For example, we haveelaborated on the conditions of training that

determine the ability of features to attractattention. That is, we have shown how theattention-attracting ability of features de-pends on the use of consistent-mapping train-ing procedures.

On the whole, our model and LaBerge'sagree when applied to the same data. Thus,although the two theories have been builton somewhat different data bases, they areconsistent even to the degree that there aresimilarities in the diagrammatic representa-tions (compare our Figures 11 and 12 withthe depictions of the model in LaBerge, 1973,1975).

Let us now return to models that are inbasic agreement with the Broadbent-Treis-man view, as opposed to the Deutsch andDeutsch view.

G. The Kahneman Theory

Kahneman (1973) proposes a rather ex-tensive theory and reviews the literature inconsiderable detail. We cannot begin to dojustice to his overall system in the limitedspace available and will therefore discussonly a few general features. Kahneman pro-poses a general theory of the allocation ofcapacity according to a factor called "effort."In our terms, he is concerned with the generallimits on controlled processing: How cancontrol be allocated? Can total capacityvary? What types of operations can be car-ried out together with what losses ofefficiency?

Interpreting our theory in Kahneman'sterms, one would say that we have proposeda specific time-based theory determining con-trolled processing capacity in detection tasks,and that our theory gives a set of rules or

9 One must take care to distinguish the Neisser(1967) view of attention from the Neisser (1967)view of cognitive processing in general. The generalview emphasizes the constructivist aspects of per-ception, sometimes known as "top-down" processing,in which perception is heavily influenced at allstages by the subject's expectations and generalknowledge. The essence of this general view seemsto be a bit incompatible with the notion of aninitial automatic processing system that is fixed,permanent, and largely unaffected by higher levelcontrolled processing.


strategies by which subjects reallocate theirprocessing in such tasks. Such a time-basedtheory is not specifically given by Kahneman.Furthermore, in Kahneman's theory there aremany results taken to indicate a change inallocatable capacity, a change often deter-mined by changes in effort. We feel thatsome of these findings are better explained byrecourse to the hypothesis that automaticprocessing has developed. As we have arguedin this paper, the development of automaticprocessing provides a means of improvingperformance independently of the effort putinto controlled processing. In fact, the effortrequired is usually greatly reduced by thedevelopment of automatic processing eventhough performance improves.

There is one other important specific areaof difference between Kahneman's theoryand ours. In common with a number of thetheories we have described (i.e., Broadbent),Kahneman's assumes that control of alloca-tion affects processing prior to the stage atwhich features indicating recognition are ac-tivated. We have dealt with this issue atlength and will not repeat the arguments,but we feel that control does not operate atthis point in the midst of what we call auto-matic processing (except in certain specialcases, such as instances where the inputs aredegraded or ambiguous, in which case auto-matic encoding never reaches the recognitionstage).

Reversing the emphasis, it may be askedwhat elements in Kahneman's approach arenot considered in our theory. The principalsuch element is the entire concept of effort.We have confined ourselves to tasks in whichwe assume effort is continually maintainedat the highest possible level. Of course itmust be true that performance will dropwhen effort drops to a low enough level, butour theory has not addressed this questionat all. It is natural, however, for us to assumethat changes in effort will affect controlledprocessing but not automatic processing.

H. The Norman and Rumelhart Model

The attentional features of this model(Norman & Rumelhart, 1970) derive largelyfrom Rumelhart's (1970) article. This model

is aimed to describe visual detection inthreshold tasks. It differs from many previousmodels in that it is quantitative and hencemakes explicit, testable predictions. It makesa key assumption that the number of featuresabstracted, and the rate of features ab-stracted, from a given input stimulus will beinversely related to the total number of itemssimultaneously presented. That is, attentionmust be shared (i.e., divided) among multipleinputs, even at the earliest stages of pro-cessing. The simultaneous-successive studiesby Shiffrin and Gardner (1972), describedearlier, most clearly show that this assump-tion is false. The features abstracted areof equivalent worth whether or not the in-puts are successive or simultaneous. As wehave seen in the present article, the difficultiesin dividing attention arise later, during con-trolled processing, subsequent to the en-coding of features.

I, Classification Models for AttentionalParadigms

A number of investigators have provideduseful and interesting classifications of at-tentional paradigms. One example is a classi-fication of types of attentional limitationsby Norman and Bobrow (197S). In general,Norman and Bobrow suppose that inputsmay be given limited processing for either oftwo reasons: "data-limitations" and "re-source-limitations." In our terms, data-limita-tions refer to the performance-reducing inter-actions that occur during automatic encoding;for example, a signal is heard better in quietthan in a background of white noise. Resourcelimitations refer to the kind of attentionlimitations that we would classify as theresult of controlled processing. Norman andBobrow argue that many authors have giveninsufficient thought to the source of the per-

-formance limitations observed in their tasks.They argue that many tasks can be variedparametrically (quantitatively) so that data-limitations occur under some conditions andresource limitations under other conditions.We agree fully with these arguments; ourstudies have, we think, spanned the range ofthese limitations and demonstrated thesepoints quite clearly.


Many other investigators have providedclassifications of attentional paradigms andresults. The Treisman (1969) review, forexample, is largely atheoretical and classifiesa variety of paradigms and tasks with a viewtoward indicating the types of attention re-striction that occur in each. Similarly Moray(1969a, 1969b) has classified attentionalstudies and paradigms. Unfortunately, thevariety of results and the experimenter'sability to design new tasks seem to havegrown about as fast as the growth in com-plexity of the classifications. Thus as usefulas these classifications have been, they havebeen limited to a posteriori descriptions.

J. Other Models

The brief summary of a number of theoriesgiven above by no means exhausts the field.Keele (1973) and Hochberg (1970) pre-sented models like that of Deutsch andDeutsch and generally similar to ours. Im-portant and influential reviews, with con-siderable theoretical import, have beenwritten by Egeth (1967), Moray (1969a,1969b),Lindsay (1970), and Swetsand Krist-offerson (1970). Important research has beencarried out by Eriksen and his associates(e.g., Eriksen Si Collins, 1969; Eriksen &Eriksen, 1974; Eriksen & Hoffman, 1972;Eriksen & Rohrbaugh, 1970) and by Posnerand his associates (e.g., Posner & Boies,1971; Posner & Snyder, 1975). Althoughspace limitations preclude our reviewingthis work, we feel that the models we havediscussed above represent a cross section oftypical approaches that have been put for-ward for tasks that resemble those in thispaper.

We would like to finish this discussion ofalternative approaches by considering verybriefly the relation of our theory to thosearising in the fields of animal discriminationlearning and attention (e.g., see Mackintosh,1975; Sutherland & Mackintosh, 1971). Wefeel that data and theory in the human andinfrahuman areas are in much closer accordthan previous work might lead one to believe.There has been a tendency to compare at-tentional models deriving from controlledprocessing effects in varied-mapping studies

in humans to models deriving from auto-matic-attention learning in consistent-map-ping paradigms in animals. The results in thepresent article should make it clear that sucha correspondence will fail. In fact, however,it is becoming clear that a comparison ofVM with VM, and CM with CM, studies inthe two areas reveals similar findings. Togive just one example, Riley and Leith(1976) review some paradigms in which ani-mals are forced to utilize what we term con-trolled processing; in such cases they revealcapacity limitations and attentional controlroughly similar to that found in human sub-jects (see also Mackintosh, 1975, andWagner, 1976, for discussions of relatedmatters).

K. Summary

Many of the elements constituting ourtheory may be found in previous proposals.There are certain themes, which have re-curred in earlier theories, about which wetake a particular position.

1. Control oj processing during encodingand feature abstraction. We argue that con-trol is minimal, that subject-controlled filter-ing, blocking, or attenuating does not occurduring this stage but subsequent to per-ceptual encoding.

2. The role oj crude physical differences inleading to automatic discrimination. We arguethat the conditions of training are crucial tothe development of automatic processing andautomatic detection; the nature of physicaldifferences among items is of secondary im-portance theoretically, though it may inpractice be important in determining therate of acquisition and levels of perform-ance during acquisition.

3. The limitations on, and capacity oj, con-trolled processing. We have presented aspecific, detailed model in which limitationson rate of processing determine the con-trolling system's capacity.

On the one hand, our theory may be seenas both a specification and extension of viewsproposed in various earlier models. On theother hand, our theory is grounded in a


much firmer empirical foundation (i.e., Ex-periments 1-3 of Part I, and 1-4 of Part II)than the previous proposals, encompasses agreater breadth than some of the earlierviews, and states its assumptions preciselyenough that at least in our standard paradigmand in a number of others as well, predic-tions can be made in advance of experimentalmanipulations. Perhaps most important, wehave, in Part I, taken steps toward quantita-tively linking attentional phenomena andsearch mechanisms, and have thereby tiedtogether two important fields of inquiry—attention and search—that have too oftenbeen treated separately in previous models.

VII. Summary and Conclusions

The studies reported in this article weredesigned to test and extend the ideas pre-sented in Part I. In Part I we showed thatcontrolled search is utilized in varied-mappingsearch paradigms and that automatic detec-tion is utilized in consistent-mapping searchparadigms. These findings obtained in bothmultiple-frame tasks measuring accuracy andin single-frame tasks measuring reaction time.The VM results in both types of tasks werefitted by a quantitative model assumingserial, terminating search, with comparisonscycling through the frame for each memory-set item.

The experiments in the present article weredesigned to examine the learning and unlearn-ing of automatic detection, the role of cate-gorization, and the learning of automatic at-tending. A summary of the main findings aregiven below.

Experiment 1 was carried out to study thedevelopment of automatic processing. Auto-matic detection for Letter Set 1 in a back-ground of Letter Set 2 developed over severalthousand training trials during which timeperformance improved considerably. Then thetwo sets were reversed and performance notonly deteriorated sharply but dropped belowthe level obtaining at the start of trainingwhen controlled search was being utilized.The results showed that

1. Automatic detection could develop whenthe memory and distractor sets were not pre-experimentally categorized.

2. The learning of an automatic-attentionresponse is a long-term phenomenon greatlyresistant to change (since many more trialswere needed to recover to a given perform-ance level after reversal than to reach thatlevel in initial training).

3. The presence among the distractors ofitems that automatically attract attentionharms the subject's ability to carry out acontrolled search (since postreversal per-formance dropped below initial performancelevels).

4. Eventually automatic-attention responsescan be "unlearned" and new sets of auto-matic responses learned, but only after con-siderable amounts of retraining.

Experiment 2 was carried out to test theeffects of reversal when the two sets in-volved were well-known categories. In Ex-periment 2, the subjects who had been trainedextensively to search for digits in a back-ground of consonants (or vice versa) werereversed in a fashion similar to that used inExperiment 1. The results showed

1. Performance in a VM condition afterreversal was identical to that in normal VMcontrolled search, even though all items pre-sented were from the previous CM target cate-gory. Hence, when all items have equallystrong automatic-attention responses, theseeither "cancel each other out" or can beignored, and normal controlled search takesplace.

2. A reversal of memory and distractorsets for the CM conditions reduced perform-ance almost to chance levels. This resultshows that the phenomenon of automaticdetection is not due to the presence of acategorical difference between memory anddistractor sets, since a distinction betweenletters and numbers was still available afterreversal.

3. The pattern of results after reversal in-dicated that a controlled search was beingutilized, but one that compared the categoryof each input to the memory-set categoryin a single operation. The results suggest thatcategories do improve search performancebut do so by facilitating controlled search.


Experiment 3 was designed to test directlythe role of categories in the development ofautomatic detection and the operation ofcontrolled search. It was shown that in VMconditions, category learning for arbitrary,visually confusable sets of letters eventuallytook place after 25 sessions of training. Atthis point, controlled search was still beingused, but the comparison process had switchedfrom individual items to categories as awhole (memory-set size had no effect).However, when training was then switchedto a CM procedure, performance improvedradically and automatic detection was learned.The results suggest that

1. Both categories and automatic responsescan be learned for arbitrarily constituted,visually confusable character sets.

2. The role of categories is to improve con-trolled search, and possibly to speed theacquisition of automatic detection.

As a group, Experiments 1, 2, and 3 pro-vided a convincing demonstration of thequalitatively different characteristics of con-trolled search and automatic detection.

Experiment 4 was designed to test the sub-jects' ability to focus attention in the faceof distraction by (a) neutral characters, (b)current targets in to-be-ignored display posi-tions, (c) items in to-be-ignored positionsthat subjects had been trained previously torespond to with automatic-attention re-sponses. The results showed that controlledsearch can usually be directed to locationsthat the subject desires to attend to but thatautomatic-attention responses can overwhelmthe controlled processing system and cancause attention to be allocated to positionsthat should be ignored. Thus automatic-atten-tion responses cannot be ignored and willinterrupt and redirect ongoing controlledprocessing.

The attention literature as a whole wasconsidered, and in light of our results thefollowing rules were suggested.

1. Divided-attention deficits arise fromlimitations on controlled processing. In par-ticular, detection deficits are due to thelimited rate of the serial comparison process.

2. Dividing attention is possible when thetargets have been consistently mapped duringtraining until automatic detection operates.

3. Focused-attention deficits arise when thedistracting stimuli initiate automatic-atten-tion responses.

4. Focusing attention is possible duringcontrolled processing (e.g., in VM tasks).

We shall conclude by summarizing brieflythe progress we feel has been made in organ-izing the phenomena of detection, search,attention, and perceptual learning. One ofthe greatest contributions of the work is thefact that all these areas have been tied bothempirically and theoretically to each otherand to common mechanisms. Another im-portant contribution is the delineation ofthe differences between, and the character-istics of, automatic and controlled processing.We make no claims that these experimentshave concluded the study of these areas.Quite to the contrary, one of the most im-portant results of this work is the host of newquestions that can now be phrased, and wehope, answered.

Reference Notes

1. Sperling, G. Multiple detections in a brief visualstimulus: The sharing and switching of atten-tion. Paper presented at the 16th annual meetingof the Psychonomic Society, Denver, November1975.

2. Sperling, G. Attention operating characteristicsfor visual search. Paper presented at the 9thannual Mathematical Psychology meeting, NewYork, August 1976.

3. Shiffrin, R. M., & Cook, J. R. Short-term for-getting without rehearsal or interference: Dataand theory. Paper presented at the 16th annualmeeting of the Psychonomic Society, Denver,November 1975.

References

Anderson, J. A. A theory for the recognition ofitems from short memorized lists. PsychologicalReview, 1973, 80, 417-438.

Anderson, J. R., & Bower, G. H. Human associativememory. Washington, D.C.: V. H. Winston & Sons,1973.

Atkinson, R. C., Holmbren, J. E., & Juola, J. F.Processing time as influenced by the number ofelements in a visual display. Perception & Psycho-physics, 1969, 6, 321-326.


Atkinson, R. C., & Juola, J. F. Search and decisionprocesses in recognition memory. In D. H.Krantz, R. C. Atkinson, R. D. Luce, & P. Suppes(Eds.), Contemporary developments in mathe-matical psychology (Vol. 1). San Francisco:Freeman, 1974.

Atkinson, R. C. & Shiffrin, R. M. Human memory:A proposed system and its control processes. InK. W. Spence & J. T. Spence (Eds.), The psy-chology of learning and motivation: Advances inresearch and theory (Vol. 2) . New York: AcademicPress, 1968.

Atkinson, R. C., & Shiffrin, R. M. The control ofshort-term memory. Scientific American, 1971, 225,82-90.

Baddeley, A. D., & Ecob, J. R. Reaction time andshort-term memory: Implications of repetition ef-fects for the high-speed exhaustive scan hy-pothesis. Quarterly Journal of Experimental Psy-chology, 1973, 25, 229-240.

Biederman, I., & Stacey, E. W., Jr. Stimulus prob-ability and stimulus set size in memory scanning.Journal of Experimental Psychology, 1974, 102,1100-1107.

Bjork, R. A. Short-term storage: The ordered out-put of a central processor. In F. Restle et al.,(Eds.), Cognitive theory (Vol. 1). Hillsdale, N.J.:Erlbaum Associates, 1975.

Brand, J. Classification without identification invisual search. Quarterly Journal of ExperimentalPsychology, 1971, 23, 178-186.

Bransford. J. D., & Franks, J. J. The abstraction oflinguistic ideas. Cognitive Psychology, 1971, 2,331-350.

Broadbent, D. E. Perception and communication.London: Pergamon Press, 1958.

Broadbent, D. E. Decision and stress. London:Academic Press, 1971.

Burrows, D. & Okada, R. Serial position effects inhigh-speed memory search. Perception & Psycho-physics, 1971, 10, 305-308.

Cavanagh, P. Holographic processes realizable in theneural realm: Prediction of short-term memoryperformance (Doctoral dissertation, Carnegie-Mellon University, 1972). Dissertation AbstractsInternational, 1973, 33, 3280B. (University Micro-films No. 72-234, 79)

Cavanagh, P. Holographic and trace strengthmodels of rehearsal effects in the item recogni-tion task. Memory & Cognition, 1976, 4, 186-199.

Cherry, C. Some experiments on the reception ofspeech with one and with two ears. Journalof the Acoustical Society of America, 1953, 25,975-979.

Cherry, E. C., & Taylor, W. K. Some further experi-ments upon the recognition of speech with oneand two ears. Journal of the Acoustical Societyof America, 1954, 26, 554-559.

Clifton, C., & Birenbaum, S. Effects of serial posi-tion and delay of probe in a memory scan task.Journal of Experimental Psychology, 1970, 86, 69-76.

Clifton, C., & Gutschera, K. D. Evidence for hier-

archical search processes in a recognition memorytask. Journal of Verbal Learning and Verbal Be-havior, 1971, 10, 528-541.

Corballis, M. C. Serial order in recognition andrecall. Journal of Experimental Psychology, 1967,74, 99-105.

Corballis, M. C. Access to memory: An analysis ofrecognition times. In P. M. A. Rabbit & S. Domic(Eds.), Attention and performance V. NewYork: Academic Press, 1975.

Corballis, M. C., Kirby, J., & Miller, A. Access toelements of a memorized list. Journal of Experi-mental Psychology, 1972, 94, 185-190.

Corteen, R. S., & Wood, B. Autonomic responses toshock-associated words in an unattended channel.Journal of Experimental Psychology, 1972, 94,308-313.

Craik, F. I. M., & Jacoby, L. L. A process viewof short-term retention. In F. Restle et al., (Eds.),Cognitive theory (Vol. 1). Hillsdale, N.J.: ErlbaumAssociates, 1975.

Craik, F. I. M., & Lockhart, R. S. Levels of pro-cessing: A framework for memory research. Jour-nal of Verbal Learning and Verbal Behavior, 1972,11, 671-684.

Craik, F. I. M.. & Tulving, E. Depth of processingand the retention of words in episodic memory.Journal of Experimental Psychology: General,1975, 104, 268-294.

DeRosa, D. V., & Morin, R. E. Recognition reactiontime for digits in consecutive and nonconsecutivememorized sets. Journal of Experimental Psy-chology, 1970, 83, 472-479.

Deutsch, J. A., & Deutsch, D. Attention: Sometheoretical considerations. Psychological Review,1963, 70, 80-90.

Egeth, H. Selective attention. Psychological Bulletin,1967, 67, 41-57.

Eriksen, C. W., & Collins, J. F. Temporal coursesof selective attention. Journal of ExperimentalPsychology, 1969, 80, 254-261.

Eriksen, B. A., & Eriksen, C. W. Effects of noiseletters upon identification of target in nonsearchtask. Perception & Psychophysics, 1974, 16, 143-149.

Eriksen, C. W., & Hoffman, J. E. Some character-istics of selective attention in visual perceptiondetermined by vocal reaction time. Perception &Psychophysics, 1972, 11, 169-171.

Eriksen, C. W., & Rohrbaugh, J. W. Some fac-tors determining efficiency of selective attention.American Journal of Psychology, 1970, 83, 330-342.

Estes, W. K. Interactions of signal and backgroundvariables in visual processing. Perception & Psy-chophysics, 1972, 12, 278-286.

Gibson, E. J. Principles of perceptual learning anddevelopment. New York: Prentice-Hall, 1969.

Gleitman, H., & Jonides, J. The cost of categoriza-tion in visual search: Incomplete processing oftargets and field items. Perception & Psycho-physics, 1976, 20, 281-288.

Gordon, G. T. Interaction between items in visual


search. Journal of Experimental Psychology, 1968,76, 348-3SS.

Gould, J. D., & Cam, R. Visual search, complexbackgrounds, mental counters, and eye move-ments. Perception & Psychophysics, 1973, 14,125-132.

Green, B. F., & Anderson, L. K. Color coding in avisual search task. Journal of Experimental Psy-chology, 1956, 5;, 19-24.

Hochberg, J. E. Attention, organization, and con-sciousness. In D. I. Mostofsky (Ed.), Attention:Contemporary theory and analysis. New York:Appleton-Century-Crofts, 1970.

Homa, D. Organization and long-term memorysearch. Memory & Cognition, 1973, 1, 369-379.

Hyde, T. S., & Jenkins, J. J. Differential effects ofincidental tasks on the organization of recall ofa list of highly associated words. Journal ojExperimental Psychology, 1969, 82, 472-481.

Hyde, T. S., & Jenkins, J. J. Recall for words as afunction of semantic, graphic, and syntacticorienting tasks. Journal of Verbal Learning andVerbal Behavior, 1973, 12, 471-480.

Ingling, N. W. Categorization: A mechanism forrapid information processing. Journal of Experi-mental Psychology, 1972, 94, 239-243.

Jensen, A. R., & Rohwer, W. D. The Stroop Color-Word Test: A review. Acta Psychologica, 1966, 25,36-93.

Jonides, J., & Gleitman, H. A conceptual categoryeffect in visual search: O as letter or as digit.Perception &• Psychophysics, 1972, 12, 457-460.

Jonides, J., & Gleitman, H. The benefit of cate-gorization in visual search: Target location with-out identification. Perception & Psychophysics,1976, 20, 289-298.

Jung, J. Comments on Handler's "From associationto structure." Psychological Review, 1965, 72,318-322.

Kahneman, D. Attention and effort. EnglewoodCliffs, N.J.: Prentice-Hall, 1973.

Kahneman, D. Effort, recognition, and recall inauditory attention. In P. M. A. Rabbit & S.Dornic (Eds.), Attention and performance V.New York: Academic Press, 1975.

Keele, S. W. Attention demands of memory retrieval.Journal of Experimental Psychology, 1972, 93,245-248.

Keele, S. W. Attention and human performance.Pacific Palisades, Calif.: Goodyear, 1973.

Klatzky, R. L., Juola, J. F., & Atkinson, R. C.Test stimulus representation and experimentalcontext effects in memory scanning. Journal ofExperimental Psychology, 1971, 87, 281-288.

Klatzky, R. L., & Smith, E. E. Stimulus expectancyand retrieval from short-term memory. Journal ojExperimental Psychology, 1972, 94, 101-107.

Krueger, L. E. Effect of stimulus probability ontwo-choice reaction time. Journal oj ExperimentalPsychology, 1970, 84, 377-379.

LaBerge, D. Attention and the measurement ofperceptual learning. Memory & Cognition, 1973,1, 268-276.

LaBerge, D. Acquisition of automatic processingin perceptual and associative learning. In P. M. A.Rabbit & S. Dornic (Eds.), Attention and per-formance V. New York: Academic Press, 1975.

Lewis, J. L. Semantic processing of unattendedmessages using dichotic listening. Journal of Ex-perimental Psychology, 1970, 85, 225-228.

Lindsay, P. H. Multichannel processing in perception.In D. I. Mostofsky (Ed.), Attention: Contem-porary theory and analysis. New York: Appleton-Century-Crofts, 1970.

Lindsay, P. H., & Norman, D. A. Human informa-tion processing: An introduction to psychology.New York: Academic Press, 1972.

Lively, B. L., & Sanford, B. J. The use of categoryinformation in a memory search task. Journal ofExperimental Psychology, 1972, 93, 379-385.

Mackintosh, N. J. A theory of attention: Variationsin the associability of stimuli with reinforcement.Psychological Review, 1975, 82, 276-298.

Mandler, G. From association to structure. Psycho-logical Review, 1962, 69, 415-426.

Mandler, G. Subjects do think: A reply to Jung'scomments. Psychological Review, 1965, 72, 323-326.

Miller, J. O., & Pachella, R. G. Locus of the stimulusprobability effect. Journal of Experimental Psy-chology, 1973, 101, 227-231.

Moray, N. Attention in dichotic listening. Affective«ues and the influence of instructions. QuarterlyJournal oj Experimental Psychology, 1959, 11,56-60.

Moray, N. Attention: Selective processes in visionand hearing. New York: Academic Press, 1969. (a)

Moray, N. Listening and attention. Harmondsworth,Middlesex, England: Penguin Books, 1969. (b)

Moray, N. A data base for theories of selectivelistening. In P. M. A. Rabbit & S. Dornic (Eds.),Attention and performance V. New York: Aca-demic Press, 1975.

Murdock, B. B., Jr. A parallel-processing model forscanning. Perception & Psychophysics, 1971, 10,289-291.

Naus, M. J. Memory search of categorized lists: Aconsideration of self-terminating search strategies.Journal of Experimental Psychology, 1974, 702,992-1000.

Naus, M. J., Glucksberg, S., & Ornstein, P. A.Taxonomic word categories and memory search.Cognitive Psychology, 1972, 3, 643-654.

Neisser, U. Decision time without reaction time:Experiments in visual scanning. American Journalof Psychology, 1963, 76, 376-385.

Neisser, U. Cognitive psychology. New York:Appleton-Century-Crofts, 1967.

Nickerson, R. S. Binary-classification reaction time:A review of some studies of human informationprocessing capabilities. Psychonomic MonographSupplements, 1972, 4, 275-318.

Norman, D. A. Towards a theory of memory andattention. Psychological Review, 1968, 75, 522-536.

Norman, D. A. Memory while shadowing. Quarterly


Journal of Experimental Psychology, 1969, 21,85-93.

Norman, D. A. & Bobrow, D. G. On data-limitedand resource-limited processes. Cognitive Psy-chology, 1975, 7, 44-64.

Norman, D. A., & Rumelhart, D. E. A system forperception and memory. In D. A. Norman (Ed.),Models of human memory. New York: AcademicPress, 1970.

Okada, R., & Burrows, D. Organizational factors inhigh-speed scanning. Journal of ExperimentalPsychology, 1973, 101, 77-81.

Pachella, R. G. The interpretation of reaction timein information processing research. In B. H.Kantowitz (Ed.), Human information processing:Tutorials in performance and cognition. Hillsdale,N.J.: Erlbaum Associates, 1974.

Posner, M. I., & Boies, S. J. Components of atten-tion. Psychological Review, 1971, 78, 391-408.

Posner, M. I., & Snyder, C. R. R. Facilitation andinhibition in the processing of signals. In P. M. A.Rabbit & S. Domic (Eds.), Attention and per-formance V. New York: Academic Press, 197S.

Postman, L. Short-term memory and incidentallearning. In A. W. Melton (Ed.), Categories ofhuman learning. New York: Academic Press, 1964.

Postman, L., & Senders, V. Incidental learning andgenerality of set. Journal of Experimental Psy-chology, 1946, 36, 153-165.

Reed, A. V. Speed-accuracy tradeoff in recognitionmemory. Science, 1973, 181, 574-576.

Reed, A. V. List length and the time course ofrecognition in immediate memory. Memory &Cognition, 1976, 4, 16-30.

Riley, D. A., & Leith, C. R. Multidimensional psy-chophysics and selective attention in animals.Psychological Bulletin, 1976, S3, 138-160.

Rumelhart, D. E. A multicomponent theory of theperception of briefly exposed visual displays.Journal of Mathematical Psychology, 1970, 7,191-218.

Schneider, F. W., & Kintz, B. L. An analysis of theincidental-intentional learning dichotomy. Journalof Experimental Psychology, 1967, 73, 85-90.

Schneider, W., & Shiffrin, R. M. Controlled andautomatic human information processing: I. De-tection, search, and attention. Psychological Re-view, 1977, 84, 1-66.

Shiffrin, R. M. Memory search. In D. A. Norman(Ed.), Models of human memory. New York:Academic Press, 1970.

Shiffrin, R. M. Information persistence in short-termmemory. Journal of Experimental Psychology,1973, 100, 39-49.

Shiffrin, R. M. The locus and role of attention inmemory systems. In P. M. A. Rabbit & S. Domic(Eds.), Attention and performance V. New York:Academic Press, 1975. (a)

Shiffrin, R. M. Short-term store: The basis for amemory search. In F. Restle et al. (Eds.), Cogni-tive theory (Vol. 1). Hillsdale, N.J.: ErlbaumAssociates, 1975. (b)

Shiffrin, R. M. Capacity limitations in informationprocessing, attention, and memory. In W. K.Estes (Ed.), Handbook of learning and cognitiveprocesses: Memory processes (Vol. 4). Hillsdale,N.J.: Erlbaum Associates, 1976.

Shiffrin, R. M., Craig, J. C., & Cohen, U. On thedegree of attention and capacity limitations intactile processing. Perception & Psychophysics,1973, 13, 328-336.

Shiffrin, R. M., & Gardner, G. T. Visual processingcapacity and attentional control. Journal of Ex-perimental Psychology, 1972, 93, 72-82.

Shiffrin, R. M., Gardner, G. T., & Allmeyer, D. H.On the the degreee of attention and capacity limi-tations in visual processing. Perception & Psy-chophysics, 1973, 14, 231-236.

Shiffrin, R. M., & Geisler, W. S. Visual recognitionin a theory of information processing. In R. L.Solso (Ed.), Contemporary issues in cognitivepsychology: The Loyola Symposium. Washington,D.C.: V. H. Winston & Sons, 1973.

Shiffrin, R. M., & Grantham, D. W. Can attentionbe allocated to sensory modalities? Perception &Psychophysics, 1974, IS, 460-474.

Shiffrin, R. M., McKay, D. P., & Shaffer, W. O.Attending to forty-nine spatial positions at once.Journal of Experimental Psychology: HumanPerception and Performance, 1976, 2, 14-22.

Shiffrin, R. M., Pisoni, D. B., & Castaneda-Mendez,K. Is attention shared beteen the ears? CognitivePsychology, 1974, 6, 190-215.

Shiffrin, R. M., & Schneider, W. An expectancymodel for memory search. Memory & Cognition,1974, 2, 616-628.

Smith, S. L. Color coding and visual search. Journalof Experimental Psychology, 1962, 64, 434-440.

Sperling, G. The information available in briefvisual presentations. Psychological Monographs,1960, 74(Whole No. 498).

Sternberg, S. High-speed scanning in human memory.Science, 1966, 153, 652-654.

Sternberg, S. The discovery of processing stages:Extensions of Donder's method. In W. G. Roster(Ed.), Attention and performance 11. Amsterdam:North-Holland, 1969. (a)

Sternberg, S. Memory scanning: Mental processesrevealed by reaction time experiments. AmericanScientist, 1969, 57, 421-457. (b)

Sternberg, S. Memory scanning: New findings andcurrent controversies. Quarterly Journal of Ex-perimental Psychology, 1975, 27, 1-32.

Stroop, J. R. Studies of interference in serial verbalreactions. Journal of Experimental Psychology,1935, 18, 643-662.

Sutherland, N. S., & Mackintosh, N. J. Mechanismsof animal discrimination learning. New York:Academic Press, 1971.

Swets, J. A., & Kristofferson, A. B. Attention. AnnualReview of Psychology, 1970, 21, 339-366.

Theios, J. Reaction time measurements in the studyof memory processes: Theory and data. In G. H.


Bower (Ed.), The psychology of learning andmotivation: Advances in research and theory(Vol. 7). New York: Academic Press, 1973.

Theios, J. The components of response latency insimple human information tasks. In P. M. A.Rabbit & S. Dornic (Eds.), Attention and per-formance V. New York: Academic Press, 197S.

Theios, J. et al. Memory scanning as a serial self-terminating .process. Journal of Experimental Psy-chology, 1973, 97, 323-336.

Townsend, J. T. A note on the identifiability ofparallel and serial processes. Perception & Psycho-physics, 1971, 10, 161-163.

Treisman, A. M. Contextual cues in selective listen-ing. Quarterly Journal of Experimental Psy-chology, 1960, 12, 242-248.

Treisman, A. M. Effect of irrelevant material onthe efficiency of selective listening. American Jour-nal of Psychology, 1964, 77, S33-S46. (a)

Treisman, A. M. Monitoring and storage of ir-relevant messages in selective attention. Journal ofVerbal Learning and Verbal Behavior, 1964, 3,449-459. (b)

Treisman, A. M. Verbal cues, language and mean-relevant messages in selective attention. Journal ofPsychology, 1964, 77, 206-219. (c)

Treisman, A. M. Strategies and models of selectiveattention. Psychological Review, 1969, 76, 282-299.

Treisman, A. M., & Riley, J. G. A. Is selective at-tention selective perception or selective response?A further test. Journal of Experimental Psychol-ogy, 1969, 79, 27-34.

Von Wright, J. M. On selection in visual immediatememory. In A. F. Sanders (Ed.), Attention andperformance (Vol. 3). Amsterdam: North-Holland,1970.

Von Wright, J. M., Anderson, K., & Stenman, U.Generalization of conditioned GSRs in dichoticlistening. In P. M. A. Rabbit & S. Dornic (Eds.),Attention and performance V. New York: Aca-demic Press, 197S.

Wagner, A. R. Priming in STM: An informationprocessing mechanism for self-generated or re-trieval-generated depression in performance. InT. J. Tighe & R. N. Leaton (Eds.), Habituatiom:Perspectives from child development, animal be-havior, and neurophysiology. Hillsdale, N.J.:Erlbaum Associates, 1976.

Williams, J. D. Memory ensemble selection in humaninformation processing. Journal of ExperimentalPsychology, 1971, 88, 231-238.

Received August 30, 1976 •

psychological review · 130 richard m. shiffrin and walter schneider c/) h i r-lu 100 80 60...

Documents