COGNITIVE AND NEURAL CONSTRAINTS ON HUMAN THOUGHTxunesis.org/publications/Morrison_CHTR_2005.pdf · 2005. 10. 12. · curs because these initial items have been stored in long-term

P1 : GFZ/KAB P2 : GFZ-KOD0521824176main.xml CB798B/Holyoak 0 521 82417 6 February 2 , 2005 2 :22

Par t V

COGNITIVE AND NEURALCONSTRAINTS ON HUMAN

THOUGHT

�

455


456


C H A P T E R 1 9

Thinking in Working Memory

Robert G. Morrison

Introduction

It is not an accident that this discussion ofworking memory is positioned near the cen-ter of a volume on thinking and reasoning.Central to higher-level cognitive processesis the ability to form and manipulate men-tal representations (see Doumas & Hummel,Chap. 4). Working memory is the cogni-tive construct responsible for the mainte-nance and manipulation of information andtherefore is neccessary for many of thetypes of complex thought described in thisbook. Likewise, the development and fail-ures of working memory are critical to un-derstanding thought changes with develop-ment (see Halford, Chap. 22) and aging (seeSalthouse, Chap. 24) as well as many typesof higher-level cognitive impairments (seeBachman & Cannon, Chap. 21 ). In spite ofits obvious importance for thinking and rea-soning, working memory’s role in complexthought is just beginning to be understood.In this chapter, we review several dominantmodels of working memory, viewing themfrom different methodological perspectives,

including dual-task experiments, individualdifferences, and cognitive neuroscience.

Multiple Memory Systems?

Although the idea of separate primary mem-ory is credited to William James (1 890),Waugh and Norman (1965) and Atkinsonand Shiffrin (1968) developed the ideaof distinct primary (i.e., short-term) andsecondary (i.e., long-term) memory com-ponents into defined models of the hu-man memory system. These multicompo-nent models of memory were supportedby observations from many different stud-ies during the 1950s and 1960s. Perhaps themost familiar justification for separate short-term and long-term memory systems is theserial position effect (e.g., Murdock, 1962).During list learning, the most recently stud-ied items show an advantage when testedimmediately – an advantage that goes awayquickly with a delay in test provided thatparticipants are prevented from rehearsing.This recency effect is presumably the result

457


458 the cambridge handbook of thinking and reasoning

Figure 1 9.1 . Atkinson and Shiffrin’s (1968) multicomponent memory model.

of quickly unloading short-term memory attest. In contrast, the first items in the listshow an advantage that withstands a delayperiod. This primacy effect presumably oc-curs because these initial items have beenstored in long-term memory through prac-tice. Conrad (1964) provided another im-portant finding justifying distinct systemswhen he observed that errors in short-term remembering were usually phonolog-ical whereas long-term memory was dom-inated by semantic coding. This suggestedthat rehearsal or storage systems were dif-ferent between the two types of memory.Yet another important finding was that, al-though the capacity of long-term memorywas seemingly limitless, short-term mem-ory as observed in a simple digit-span taskwas of limited capacity (Miller, 1956) –a finding confirmed using many other ex-perimental paradigms. Lastly, around thissame era, neuropsychological evidence be-gan to emerge suggesting that at least partsof the short- and long-term memory systemswere anatomically distinct. Milner’s (1966)famous amnesic patient, HM, with his long-term memory deficits but preserved short-term digit span, and Shallice and Warring-ton’s (1970) patient, KF, with his intactlong-term memory but grossly impaireddigit span, presented a double dissociationfavoring at least partially distinct short- andlong-term memory systems. Atkinson andShiffrin’s (1968) memory model was typicalof models from the late 1960s with distinctsensory, short-term, and long-term memorystores (Figure 19.1 ). Short-term memory wasviewed as a short-term buffer for informa-tion that was maintained by active rehearsal.It was also believed to be the mechanism by

which information was stored in long-termmemory.

A Multi-component WorkingMemory Model

While exploring the issues described in theprevious section, Baddeley and Hitch (1974)proposed a model that expanded short-termmemory into the modern concept of workingmemory – a term that has been used in severaldifferent contexts in psychology.1 Baddeley(1986) defined working memory as “a systemfor the temporary holding and manipula-tion of information during the performanceof a range of cognitive tasks such as com-prehension, learning, and reasoning” (Ref. 3 ,p. 34). In a recent description of his working-memory model, Baddeley (2000) proposeda four-component model (Figure 19.2), in-cluding the phonological loop, the visuospa-tial sketchpad, the central executive, and themodel’s most recent addition, the episodic

Figure 1 9.2 . Baddeley’s (2000) four-componentworking memory model.


thinking in working memory 459

buffer. This model has primarily been con-ceptualized based on results from behav-ioral dual-task paradigms and neuropsychol-ogy. For instance, using behavioral methods,Baddeley and Hitch (1974) reasoned thatthey could identify the separable elementsof working memory by looking for task in-terference. If you assume the various compo-nents of working memory are capacity lim-ited, then if the simultaneous performanceof a secondary task degrades performanceof a primary task, these two tasks musttap a common limited resource – partic-ularly if there exists another primary taskthat is unaffected by performance of the sec-ondary task and is affected by a differentsecondary task that does not affect the firstprimary task. Likewise, neuropsychologicalevidence such as the existence of patientswith selectively disabled verbal (e.g., patientKF, Shallice & Warrington, 1970) and visual(e.g., de Renzi & Nichelli, 1975) digit spansuggested that verbal and visual working-memory systems are somewhat separableas well.

Using this type of methodology, Baddeleyhas suggested that the phonological loop andvisuospatial sketchpad are modality-specificslave systems that are responsible for main-taining information over short periods oftime. The phonological loop is responsiblefor the maintenance and rehearsal of infor-mation that can be coded verbally (e.g., thedigits in a digit-span task). It is phonemi-cally sensitive (e.g., Ted and Fred are harderto remember than Ted and Bob), and its ca-pacity is approximately equal to the amountof information that can be subvocally cy-cled in approximately 2 seconds. Baddeley(1986) argues that these two characteris-tics of verbal working memory are best ex-plained by two components: (1 ) a phonologi-cal store that holds all of the information thatis currently active and is sensitive tophonemic interference effects and (2) anarticulatory loop that is used to refresh theinformation via a process of time-limitedsubvocal cycling. The articulatory loopis specifically disrupted by the commonphonological loop secondary task, articula-tory suppression (i.e., repeating a word or

H*yes yes

no no

no no

yes yes

yes yesyes yes

Figure 1 9.3 . The Brooks (1968) letter task.Participants are to image a block letter and thendecide whether each corner of the letter is anoutside edge.

number vocally). Thus, verbal span is con-strained by both the amount of informationto be maintained and the time that it takesto rehearse it. In contrast to the phonologi-cal loop, the visuospatial sketchpad has beenmore difficult to describe. In a dual-task ex-periment, Baddeley (1986) asked subjects tosimultaneously perform a pursuit rotor task(i.e., track a spot of light that followed acircular path with a stylus) while perform-ing either a verbal or spatial memory taskpreviously developed by Brooks (1968; Fig-ure 19.3). The verbal task required subjectsto remember a sentence (e.g., “A bird in handis not in the bush”) and scan through eachword deciding whether it was a noun or not.The correct pattern of output for this exam-ple would be: no, yes, no, yes, no, no, no, no,

yes. In the visual memory task, participantsare first shown a block letter with one cornermarked with an asterisk (Figure 19.3). Theyare then asked to imagine the letter and, be-ginning at the marked corner, judge whethereach corner is an outside corner or not. Thus,in both the verbal and visual memory tasks,participants are required to hold a modality-specific object in memory and inspect it, an-swering yes or no to questions about theirinspection. Baddeley found that the visualmemory task, but not the verbal memorytask, seriously degraded pursuit rotor track-ing performance.

Logie (1995) has argued for a visual sim-ilarity effect analogous to the phonemicsimilarity effect used to support the phono-logical store. Participants were visually



presented strings of upper- and lowercaseletters (e.g., “KcPs” or “gBrQ”). Letters werechosen based on the similarity of their lowerand uppercase characters. Thus Kk, Cc, Pp,Ss were visually similar while Gg, Bb, Rr, Qqwere visually dissimilar. To discourage use ofthe phonological loop to perform the task,participants performed simultaneous articu-latory suppression. After a retention period,participants had to write down the lettersequence in correct order and case. Logiefound that participants made significantlymore errors when the letter cases were vi-sually similar. This finding suggests the ex-istence of a visual store analogous to thephonological store in the phonological loop.It is possible that a visual rehearsal loop anal-ogous to the articulatory loop exists; how-ever, to date evidence is limited to introspec-tive accounts of mnemonics. What is clear isthat both visual and spatial qualities of stim-uli can be stored in the short term; how-ever, the independence of systems responsi-ble for visual and spatial memory is the topicof much debate (see Logie, 1995).

The third component of Baddeley’s work-ing memory model, the central executive,was initially a catch-all for the working-memory-processes necessary for certain cog-nitive abilities that did not fit cleanlyinto the phonological loop or visuospatialsketchpad. This category included manyof the cognitive abilities discussed in thisbook, including reasoning, problem solv-ing, and language. For instance, Shallice andWarrington’s (1970) patient KF had a dras-tically degraded verbal span (i.e., two let-ters) with relatively intact language compre-hension. Believing that both of these abili-ties required working memory, Baddeley andHitch (1974) reasoned that verbal span andlanguage comprehension must use separateworking-memory modules. To test this hy-pothesis, they devised a short-term mem-ory load task that balanced maintenance loadand time (Baddeley & Hitch, 1974). For in-stance, a low-load condition might requireparticipants to remember three numbers,outputting them every 2 seconds, while ahigh-load condition might require partici-pants to remember six numbers, outputting

them every 4 seconds. Participants per-formed this secondary task while simulta-neously performing a primary task involv-ing auditory language comprehension. Theyfound that language comprehension onlysuffered at high concurrent memory loadand not under lower memory load condi-tions. At low memory load, participants hadsufficient resources to carry out the compre-hension task; however, at high memory loadthere were insufficient resources for lan-guage comprehension. Adding the results ofthis study to many other similar experimentsand the neuropsychological evidence frompatients like KF, Baddeley and Hitch postu-lated that comprehension and digit span uti-lizes separate modules of working memorythat taps a common resource pool.

Given the amorphous nature of the cog-nitive tasks for which the central executivewas necessary, Baddeley (1986) initially em-braced Norman and Shallice’s (1980; 1986)concept of a Supervisory Attentional Sys-tem as a model for the central executive.Norman and Shallice suggested that mostwell-learned cognitive functions operate viaschemata, or sets of actions that run auto-matically. Although many schemata may beshared by most individuals (e.g., driving acar, dialing a telephone, composing a simplesentence, etc.), additional schemata may beacquired through the development of spe-cific expertise (e.g., writing lines of com-puter code, swinging a golf club, etc.). Atmany times during an ordinary day, we mustperform more than one of these schemataconcurrently (e.g., talking while driving).Norman and Shallice suggest that when wemust perform multiple schemata, their co-ordination or prioritization is accomplishedvia the semi-automatic Contention Schedulerand the strategically controlled SupervisoryAttentional System. The Contention Sched-uler uses priorities and environmental cues(e.g., a car quickly pulls in front of me),whereas the Supervisory Attentional Systemtends to follow larger goals (e.g., convinc-ing my wife that I’m a good driver). Thus,when the car rapidly pulled in front of me,I pressed the brake on the car and then pro-ceeded to tell my wife how attentive I am



while on the road. One important charac-teristic of the Supervisory Attentional Sys-tem as a model of the central executive wasthat it was sensitive to capacity limits. Ac-cording to Norman and Shallice, capacitylimits constrain thinking and action during(1 ) complex cognitive processes such as rea-soning or decision making; (2) novel tasksthat have not developed schemata; (3) life-threatening or single, difficult tasks; and (4)functions that require the suppression ofhabitual responses.

Baddeley (1986) suggested that the Su-pervisory Attentional System provided auseful framework for understanding randomgeneration, a task frequently associated withthe central executive. In random genera-tion, a participant is asked to generate aseries of random responses from a prede-termined list (e.g., integers from 0 to 9,for instance: 1 ,8,4 ,6,0,7,6, 8,4 ,5 ,6,1 ,2). Re-sponse patterns from this task usually exhibittwo characteristics: (1 ) certain responses ap-pear at much lower frequencies than oth-ers (e.g, 3 or 9 did not appear whereas1 ,4 ,6, and 8 appeared repeatedly) and (2)stereotyped responses (e.g., 4 ,5 ,6 or 1 ,2)are much more common than other equallylikely two- or three-number sequences (Bad-deley, 1966). Baddeley suggested that thehigher-order goal of randomness is at oddswith the dominant schemata for the pro-duction of numbers (i.e., counting). Thus,random generation potently requires the ser-vices of the Supervisory Attentional Sys-tem to override or inhibit the dominantschemata. When random number genera-tion is performed with another working-memory–intensive task, the resources avail-able to the Supervisory Attentional System(i.e., central executive) are in even more de-mand and responses become more stereo-typed (Baddeley et al., 1998).

Although the Supervisory AttentionalSystem describes an important ability thatunderlies complex cognitive processes suchas language comprehension and problemsolving, it fails to offer a tenable accountof how, short of a homunculus, this direc-tion would occur. Acknowledging this prob-lem, Baddeley’s current model of the central

executive fractionates the central executivein the hope that by understanding preciselywhat the central executive does we mightlearn how it does it. Baddeley (1996) sug-gested four arguably distinct central exec-utive functions: “(1 ) the capacity to coor-dinate performance on two separate tasks,(2) the capacity to switch retrieval strategiesas reflected in random generation, (3) thecapacity to attend selectively to one stimu-lus and inhibit the disrupting effect of oth-ers, and (4) the capacity to hold and ma-nipulate information in long-term memory,as reflected in measures of working memoryspan” (Ref. 4 , p. 5). Thus, Baddeley arguedthat the central executive is important fortask switching, inhibition of internal repre-sentations or prepotent responses, and theactivation of information in long-term mem-ory during an activity that requires the activemanipulation of material. In comparison tothe slave systems, relatively little attentionhas been paid to the central executive utiliz-ing dual-task methodologies.

The last and most recently added compo-nent of Baddeley’s working-memory modelis the episodic buffer. One problem encoun-tered by a modal working-memory model isthe need for integration. How can a com-plex problem requiring the integration ofinformation across modalities be solved ifall the information is being held in sepa-rate distinct buffers? This binding problem,whether it is binding information withina modality or across modalities, is one ofthe central challenges for a working-memorysystem capable of high-level cognition (seeDoumas & Hummel, Chap. 4). To addressthis issue, Baddeley (2000) has proposeda third type of buffer that uses a multidi-mensional code. Thus, this buffer can main-tain information from several modalities thathas been bound together by the central ex-ecutive. Fuster, Bodner, and Kroger (2000)have found evidence of the existence of neu-rons in prefrontal cortex that seem to be re-sponsible for this type of function. Anotherimportant function of the episodic bufferis serving as a scratchpad for the develop-ment of new mental representations duringcomplex problem solving. There are many



examples of situations requiring the func-tions ascribed to the episodic buffer, but themethods for studying such a resource utiliz-ing the task-interference paradigm are stillunder development.

Embedded-ProcessesWorking-Memory Model

Although Baddeley’s multi-componentworking-memory model has dominatedthe field for much of the past thirty years,there are alternative conceptions of work-ing memory. Cowan (1988, 1995) hasproposed a model that tightly integratesshort- and long-term memory systemswith attention. In his Embedded-Processesworking-memory model (Figure 19.4),Cowan defines working memory as the setof cognitive processes that keep mentalrepresentations in an easily accessible state.Within this system, information can eitherbe within the focus of attention, whichCowan believes is capacity limited, orin active memory, which Cowan suggestsis time limited. The focus of attention issimilar to James’s (1 890) concept of primarymemory and is equated to the informationthat is currently in conscious awareness. Incontrast, active memory, a concept similarto Hebb’s (1949) cell assemblies or Ericssonand Kintsch’s (1995) long-term workingmemory, refers to information that hashigher activation either from recently beingin the focus of attention or through sometype of automatic activation (e.g., priming).In the Embedded-Processes model, a centralexecutive, somewhat similar to Normanand Shallice’s (1980, 1986) SupervisoryAttention System, is responsible for bring-ing information into the focus of attentionwhile an automatic recruitment of attentionmechanism can bring information intoactive memory without previously havingbeen in the focus of attention.

A critical distinction between Cowan’sEmbedded-Processes model and Baddeley’smulti-component model is how the twomodels deal with the topic of maintenance of

information. As previously discussed, Bad-deley hypothesizes modality-specific buffersfor the short-term storage of informationthat coordinate with the Episodic Buffer,which is responsible for storing integrated in-formation. In contrast, Cowan suggests thatinformation is maintained in working mem-ory simply by activating its representationsin long-term memory via short-term – spe-cific neurons in the prefrontal or parietal cor-tices. This latter view suggests that informa-tion from different modalities will behavedifferently to the extent that they are codeddifferently in long-term memory, a viewsomewhat at odds with findings of phonolog-ical errors in short-term memory tasks andsemantic errors in long-term memory tasks.Cowan counters this objection by notingthat different codes are used in the storage ofinformation in long-term memory and, de-pending on the nature of the task, differentcodes are likely to be more important. Like-wise, Baddeley has argued that short-termand long-term memory systems are distinctbased on neuropsychological evidence sug-gesting that short-term and long-term sys-tems can be dissociated and therefore mustbe distinct systems. This argument, how-ever, relies to some extent on the belief thatthe individual short- and long-term systemsare anatomically unitary, an assumption thatseems unlikely given recent evidence fromcognitive neuroscience. Fuster has argued,based on results from single-cell recordingin nonhuman primates, that neurons in pre-frontal cortex are responsible for maintain-ing information in working memory (Fuster& Alexander 1971 ); however, disrupting cir-cuits between this area and more poste-rior or inferior regions associated with long-term storage of information can also resultin working-memory deficits (Fuster, 1997).Recent evidence from electrophysiology inhumans seems to confirm that areas in pre-frontal cortex and areas associated with long-term storage of information are temporallycoactive during working-memory tasks (seeRuchkin et al., 2003 , for a review).

A second important distinction be-tween Baddeley’s multi-componentworking-memory model and Cowan’s



Figure 1 9.4. Simplified diagram of Cowan’s (1988) Embedded-Processes Model.

Embedded-Processs model is modalityspecificity. Specifically, Baddeley has pro-posed independent modules within workingmemory for maintaining information fromdifferent modalities (e.g., visual or verbal).In contrast, Cowan suggests only a domain-general central executive that, in turn, canactivate networks for various modalities ofinformation stored in long-term memory.Baddeley also proposes a domain-generalcentral executive, so the main distinctionbetween the models is whether informationto be maintained in working memory isloaded into domain-specific buffers orwhether it is simply activated in long-termmemory. From our earlier discussion, thereseems to be no doubt that it is easier tomaintain a certain quantity of informationacross several modalities than to maintainthe same amount of information within justa single modality. Although this observationdoes not necessitate independent buffers, itdoes suggest that capacity limitations maybe somewhat domain-specific.

Reasoning and Working Memory:Using the Task-Interference Paradigm

Although the task-interference paradigmhas been very useful in exploring working

memory slave systems, relatively little hasbeen done using this technique to studyhigh-level cognition or the central executive.Central to high-level cognitive processes isthe ability to form and manipulate men-tal representations. Review of the functionsof the central executive in either Baddeleyor Cowan’s models suggests that the cen-tral executive should be critical for thinkingand reasoning – a hypothesis that has beenconfirmed in several studies. In their semi-nal work on working memory Baddeley andHitch (1974) asked participants to performa reasoning task in which they read a sim-ple sentence containing information aboutthe order of two abstract terms (i.e., A andB). Their task was to judge whether a let-ter sequence presented after the sentencereflected the order of the terms in the state-ment. For instance, a TRUE statement wouldbe “A not preceded by B” followed by AB(Ref. 7, p. 50). Baddeley and Hitch varied thestatements with respect to statement voicing(i.e., active or passive), negation, and verbtype (i.e., precedes or follows). They foundthat low concurrent memory loads (i.e., oneto two items to remember) had no effect onreasoning accuracy or response time; how-ever, high concurrent memory load (i.e., sixitems to remember) had a reliable effect onresponse time. Depending on the empha-sis of the instructions used, they found that



the decrement in performance was eitherin the reasoning task or the memory task.There was no statistical interaction betweenconcurrent memory and the reasoning taskdifficulty.

Several other researchers have investi-gated how working memory is important fordeductive reasoning. Gilhooly et al. (1993),utilizing methods similar to Baddeley andHitch, asked participants to perform ver-bal syllogisms (Evans, Chap. 8, for a de-scription of syllogistic reasoning) of varyinglevels of complexity. In a first experiment,participants either viewed the premises ofthe syllogisms visually, all at once, or heardthe premises read one at a time. Gilhoolyet al. hypothesized that verbal presentationwould result in a higher working-memoryload because participants would have tomaintain the content of the premises beforethey were able to solve the problem. Theyfound this result: Participants made more er-rors in the verbal condition than in the vi-sual condition. An error analysis indicatedthat the errors made were the result of notremembering the premises correctly, not er-rors made in the process of integration of in-formation between premises. In a second ex-periment, they had participants perform thesyllogism task visually while performing oneof three different secondary tasks. Theyfound that only random number generationinterfered with performance of syllogisms.Gilhooly et al. concluded that the centralexecutive is critical for relational reason-ing and the phonological loop (as interferedwith by articulatory suppression) may beinvolved to a lesser extent. They also con-cluded that the visuospatial sketchpad, asinterfered with by spatial tapping (i.e., tap-ping a fixed pattern with the fingers), was notimportant for performing verbal syllogismsand thus argued against models of reason-ing that are at least in principle dependenton involvement of visual working memory(e.g., Kirby & Kosslyn, 1992 ; Johnson-Laird,1983). In a similar study, Toms, Morris, andWard (1993) found no evidence that a vari-ety of secondary tasks loading on either thephonological loop or visuospatial sketchpad

had any effect on either reasoning accuracyor latency. Of the secondary tasks they used,only a high concurrent memory load (i.e., sixdigits) affected reasoning performance, andthis effect appeared to be limited to difficultsyllogisms.

Klauer, Stegmaier, and Meiser (1997) hadparticipants perform syllogisms and spatialreasoning tasks that involved transitive in-ference (see Halford, Chap. 22 , for a de-scription of transitive inference tasks). Thespatial reasoning problems varied in com-plexity from simple transitive inference (e.g.,“The circle is to the right of the trian-gle. The square is to the left of the tri-angle.” See Ref. 44 , p. 1 3) to more com-plicated transitive inference problems thatrequired greater degrees of relational inte-gration. Klauer et al. had participants per-form a visual tracking task (i.e., followone object on a screen filled with distrac-tor objects) while listening to the premisesof the reasoning problems. They foundthat this visuospatial secondary task inter-fered with spatial reasoning but had littleeffect on syllogism performance. In anotherexperiment, Klauer et al. presented syllo-gisms or spatial reasoning problems eitherauditorally (as in the previous experiment)or visually on a computer screen. Whileperforming these primary tasks, participantsperformed random generation either ver-bally or spatially, by pressing keys in a ran-dom pattern. They found that both forms ofrandom generation affected both syllogismand spatial reasoning performance; however,spatial random generation caused somewhatless interference than verbal random gener-ation – a finding consistent with Baddeley etal.’s (1998) extensive study of random gen-eration. In their final experiment, Klauer etal. found that articulatory suppression (i.e.,counting repeatedly from 1 to 5) had a mildeffect on syllogism and spatial reasoning la-tencies. Overall, Klauer et al. found evidencefor involvement of the central executive (asinterfered with by random generation) andsomewhat less interference by slave systemtasks consistent with the modality of the rea-soning task.



Unlike the examples of deductive andspatial reasoning we discussed previously,analogical reasoning frequently requires theextensive retrieval of semantic informationin addition to the relational processing char-acteristic of all types of reasoning (seeHolyoak, Chap. 6, for a detailed discussionof analogical reasoning). Waltz et al. (2000)had participants perform an analogical rea-soning task while performing one of severalsecondary tasks. In the analogical reasoningtask (adapted from Markman & Gentner,1993), participants studied pairs of picturesof scenes with multiple objects (see Figure6.3 in Holyoak, Chap. 6). For instance, oneproblem showed a boy trying to walk a dogin one picture while the companion pictureshowed a dog failing to be restrained by aleash tied to a tree. Participants were asked tostudy each picture and pick one object in thesecond picture that “goes with” a target ob-ject in the first picture. In the example prob-lem in Figure 6.3 , the man in the first pictureis a featural match to the boy in the secondpicture while using an analogy the boy is arelational match to the tree in the secondpicture. Participants were simply asked to se-lect one object; they were thus free to com-plete the task based on either featural sim-ilarity or make an analogical mapping andinference, answering based on relational sim-ilarity. Waltz et al. found that participantswho maintained a concurrent memory loador performed verbal random number gener-ation or articulatory suppression (i.e., sayingthe word “the” once each second) gave fewerrelational responses than a control group notperforming a dual task. In a recent extensionwith this task, my lab replicated Waltz et al.’sarticulatory suppression finding (i.e., sayingthe English nonword “zorn” once each sec-ond) and also found a similar effect for a vi-suospatial working-memory dual task (man-ually tapping a simple spatial pattern).

In the previous studies, the extent of in-terference with the analogy task was similarfor both central executive (concurrent mem-ory load and verbal random number gen-eration) and slave system (articulatory sup-pression and spatial tapping) dual tasks. One

explanation of these results is that analogicalreasoning is more resource demanding thanthe deductive and spatial reasoning taskspreviously discussed, and thus even the slavesystem tasks cause significant interference.Another possibility is that analogical reason-ing places greater demands specifically onthe modality-specific slave systems of work-ing memory than other forms of relationalreasoning. To investigate this issue, Morri-son, Holyoak, and Truong (2001 ) had par-ticipants perform either a verbal or visualanalogy task, while performing articulatorysuppression (i.e., saying the nonword “zorn”once a second), spatial tapping (i.e., touch-ing one of four red dots each second in apredetermined pattern), or verbal randomnumber generation. In the verbal analogytask, participants verified verbal analogies,such as BLACK:WHITE::NOISY:QUIET,answering either TRUE or FALSE via a floorpedal. In the visual analogy task, participantsperformed Sternberg’s (1977) People Piecesanalogy task. In this task, participants ver-ify whether the relational pattern of charac-teristics between two cartoon characters isthe same or different than between a sec-ond pair of characters. Morrison, Holyoak,and Truong found that, for verbal analogies,articulatory suppression and verbal randomnumber generation resulted in an increase inanalogy error rate, whereas only verbal ran-dom number generation increased analogyresponse time for correct responses. Spatialtapping had no reliable effect on verbal anal-ogy performance. In contrast, for visual anal-ogy, both spatial tapping and verbal randomnumber generation resulted in more analogyerrors, whereas only random generation in-creased analogy response time. Articulatorysuppression had no reliable effect on visualanalogy performance. Thus, there seems tobe a modality-specific role for working mem-ory in analogical reasoning.

In summary, all of the reasoning tasksdescribed in the previous section are inter-fered with by dual tasks considered to tapthe central executive (e.g., random numbergeneration or concurrent memory load).The deductive reasoning tasks reported



require the manipulation and the alignmentof premises that are provided in the prob-lem. In addition to these operations, analog-ical reasoning may require the reasoner toretrieve information from semantic memory(e.g., the relations that bind the terms in theanalogy) and then map the resulting rela-tional statements (and in some cases makean inference that requires retrieving a termthat completes the analogy).

To evaluate the extent that working-memory resources are necessary for seman-tic memory retrieval and relational binding,my lab went on to examine the compo-nent processes in working memory is neces-sary for analogical reasoning. We wonderedwhether working memory is necessary forthe simple process of relational binding oronly becomes necessary when multiple rela-tions need to be maintained and comparedduring the analogical mapping process. Toaddress this question, we used the stim-uli from the verbal analogy task but simplyasked participants to verify relational state-ments instead of comparing two of them asin the analogy task. Thus, participants wouldrespond TRUE to a statement like “black isthe opposite of white” and FALSE to thestatement “noisy is the opposite of nois-ier.” As in the verbal analogy task, articula-tory suppression and verbal random numbergeneration affected performance with spa-tial tapping also having a smaller, but reli-able effect. Thus, relational binding, not justmaintenance and mapping, require use ofthe working-memory system, including themodality-specific slave systems.

Individual Differences inWorking Memory

An alternative to Baddeley’s dual-task me-thodology uses individual differences tostudy working memory. Daneman and Car-penter (1980) first used this approach toinvestigate how working memory was in-volved in language comprehension. They de-veloped a reading span task that requiredsubjects to read several sentences and then

later recall the last word of each sentence inthe correct order. The participant’s span istypically defined as the maximum-sized trialwith perfect performance. This measure cor-related relatively well with individuals’ read-ing comprehension ability. Unlike a simpleshort-term memory-span task, the working-memory–span task required the subjects todo a more complex task while also remem-bering a list of items. In this way, the spantask is believed to tap both the mainte-nance (slave system) and manipulation (cen-tral executive and episodic buffer) aspectsof working memory. Other span tasks havebeen developed to vary the nature of thetask that participants perform and whatthey maintain. For example, Turner and En-gle (1989) asked participants to solve sim-ple arithmetic problems and then remem-ber a word presented at the end of eachproblem. In the n-back task (Figure 19.5 ;Smith & Jonides, 1997 for a complete de-scription), the manipulation task is changedto having to continuously update the setof items. Using this approach, researchershave found working-memory capacity to bean important predictor of performance in abroad range of higher cognitive tasks, includ-ing reading comprehension (Daneman &Carpenter, 1980), language comprehension(Just & Carpenter, 1992), following direc-tions (Engle, Carullo, & Collins, 1991 ), rea-soning (Carpenter, Just, & Shell, 1990; Kyllo-nen & Christal, 1990), and memory retrieval(Conway & Engel, 1994).

Researchers using working-memory-spanmeasures typically measure participants’working-memory span using one or moremeasures and then use this to predict per-formance on another task. A high correlationsuggests that working memory is an impor-tant target for the task. More sophisticatedstudies collect a variety of other measures ofinformation processing ability (e.g., process-ing speed or short-term memory span) anduse either multiple regression or structuralequation modeling to determine whetherthese various abilities are separable withrespect to the target task. Engle and hiscollaborators (Engle, Kane, & Tuholski,1999; Kane & Engle, 2003b; see also



Figure 1 9.5 . The n-back task. Participants see a stream of letters, numbers, orsymbols and have to continuously answer whether the current item was thesame as the item presented “n-back” in the stream. This task requiresmaintenance of the current in-set item and continuous updating of this set – anability considered to be manipulation of the set.

Salthouse, Chap. 24) have used this ap-proach to argue that, although working-memory-span and short-term-memory-spantasks share much variance, it is working-memory capacity that best predicts highercognitive performance as measured by taskssuch as the Ravens Progressive Matrices (seeFigure 19.6).

Kane and Engle believe that the abilitymeasured by a working-memory-span taskonce simple maintenance is stripped away isbest described as controlled attention. Theyhave argued that working-memory capacityis a good predictor of task performance intasks that (a) require maintenance of taskgoals, (b) require scheduling competing ac-tions or responses, (c) involve response com-petition or (d) involve inhibiting informa-tion irrelevant to the task (Engle, Kane, &Tuholski, 1999). This list is very similar tothe functions that Baddeley (1996) attribu-tes to the central executive. Obviously, theseare the types of cognitive processes that

are omnipresent in high-level cognition.They are also the types of cognitive abil-ities necessary to perform traditional testsof fluid or analytical intelligence such asthe Ravens Progressive Matricies (1938),leading researchers to hypothesize thatworking-memory capacity is the criticalfactor that determines analytical intelli-gence (see Kane & Engle, 2003a; Sternberg,Chap. 31 ).

The Where, What, and Howof Working Memory and Thought

So far, we have suggested that there areat least two important aspects of workingmemory for human thinking – a modality-specific maintenance function that is ca-pable of preserving information over shortperiods of time and a manipulation or at-tentional control function that is capable



Figure 1 9.6. Structural Equation Model of the relationship of working memory and short-termmemory and their role in analytic problem solving and intelligence. From Engle, Kane, and Tuholski(1999).

of activating, operating, and updating thisinformation during conscious thought. Re-cently, cognitive neuroscientists have de-voted much effort to answering the questionof where in the brain these working mem-ory mechanisms operate. This topic is be-yond the scope of this chapter [see Goel,Chap. 20, for a more detailed treatment ofthe cognitive neuroscience of problem solv-ing and Chein, Ravizza, and Fiez (2003) fora recent appraisal of the ability of Baddeleyand Cowan’s models to account for recentneuroimaging findings]; however, we knowthat at least several areas of the prefrontaland parietal corticies are critical for thesefunctions. Although these areas may be spe-cific to working memory, there is mount-ing evidence from both electrophysiologyand functional magnetic resonance imaging(fMRI) that working memory is the resultof activation of networks involving manybrain regions.2 A more interesting questionthan where, is how working memory op-erates thinking. Unfortunately, much lessattention has been given to this question;however, several of the computational ap-proaches outlined in this book begin to ad-dress this topic.3

It is the belief of many of the authors inthis volume that high-level cognition is in-trinsically relational in nature, a position longargued by many scientists (see Fodor andPylyshyn, 1988; Spearman, 1923). In thisaccount, one critical function for workingmemory to accomplish is the flexible bindingof information stored in long-term memory.Working memory must also be able to nestrelations to allow more complex knowledgestructures to be used. Halford (Chap. 22) hasreferred to this factor as relational complexity.As the relational complexity of a particularproblem increases, so do the demands placedon working memory. Goals are a particu-lar subclass of relations that are especiallyimportant in deductive reasoning (see Goel,Chap. 20). Maintaining the complex goal hi-erarchies (high relational complexity) nec-essary for solving complex problems such asthose encountered in chess or in tasks suchas the Ravens Progressive Matrices or theTower of Hanoi makes great demands onthe working memory system (see Lovett &Anderson, Chap. 1 7; Carpenter, Just, &Shell, 1990; Newman et al., 2003). Mostwork directed at understanding how thebrain implements working memory has



focused on relatively simple tasks in whichprocessing of relations is minimal.

The ways in which the brain’s distributedarchitecture is used to process problemsthat require relation flexibility and relationalcomplexity have just begun to be explored(see Christoff & Gabrieli, 2000; Christoff etal., 2001 ; Morrison et al., 2004 ; Prabhakaranet al., 2000; Waltz et al., 1999). Hum-mel and Holyoak’s (1997, 2003 ; see alsoDoumas & Hummel, Chap. 4) LISA modelsolves the binding problem created by theneed for the flexible use of information ina distributed architecture. The LISA modeldynamically binds roles to their fillers inworking memory by temporal synchrony offiring. This allows the distributed informa-tion in long-term memory to be flexiblybound in different relations and for the sys-tem to appreciate that the various entitiescan serve different functions in different re-lations and relational hierarchies. It is possi-ble that one role of the prefrontal cortex is tocontrol this synchrony process by firing thedistributed network of neurons representingthe actual fillers in long-term memory (seeDoumas & Hummel Chap. 4 , and Morrisonet al., 2004 , for a more detailed account ofthis approach). Although no direct evidenceexists for synchrony of binding in high-levelrelational systems, several studies in animals(e.g., Gray et al., 1989) and in humans (e.g.,Muller et al., 1996; Ruchkin et al., 2003)suggest that synchrony may be an importantmechanism for other cognitive processes im-plemented in the brain. This type of sys-tem is also consistent with Baddeley’s (2000)concept of an episodic buffer that binds in-formation together in working memory.

Implicit in a working-memory system ca-pable of handling relations is not only theability to precisely activate information inlong-term memory but also the ability todeactivate or inhibit it. Consider the simpleanalogy problem:

BLACK:WHITE::NOISY: ? (1 ) QUIET(2) NOISIER

If the semantic association between NOISYand NOISIER is stronger than that between

NOISY and QUIET, the correct relationalresponse, QUIET, may initially be less activebecause of spreading activation in memorythan the distractor item, NOISIER. Thus,during reasoning, it may be necessary to in-hibit information that is highly related butinconsistent with the current goal (Morri-son et al., 2004). This function of work-ing memory has also been ascribed to theprefrontal cortex (see Kane & Engle, 2003band 2003a; Miller & Cohen, 2001 ; and Shi-mamura, 2000, for reviews). Many complexexecutive tasks associated with frontal lobefunctioning (e.g., Tower of Hanoi or Lon-don, Analogical Reasoning, Wisconsin CardSorting) have important inhibitory compo-nents [Miyake et al., 2000; Morrison et al.,2004 ; Viskontas et al. (in press), and Welsh,Satterlee-Cartmell, & Stine, 1999]. Shima-mura (2000) suggested that the role of pre-frontal cortex is to filter information dynam-ically – a process that requires the use ofboth activation and inhibition to keep infor-mation in working memory relevant to thecurrent goal. Miller and Cohen (2001 ) ar-gued that “the ability to select a weaker, task-relevant response (or source of information)in the face of competition from an otherwisestronger, but task-irrelevant one [is one ofthe most] fundamental aspects of cognitivecontrol and goal-directed behavior” (Ref. 48,p. 1 70) and is a property of prefrontal cortex.More generally, many researchers believedthat inhibition is an important mechanismfor complex cognition (see Dagenbach &Carr, 1994 ; Dempster & Brainerd, 1995 ; andKane & Engle, 2003a, for reviews) and thatchanges in inhibitory control may explainimportant developmental trends (Bjorklund& Harnishfeger, 1990; Hasher & Zacks, 1988;Diamond, 1990) and individual differences(Dempster, 1991 ; Kane & Engle, 2003a,2003b) in complex cognition.

Conclusions and Future Directions

Working memory is a set of central processesthat makes conscious thought possible. Itflexibly provides for the maintenance and



manipulation of information through bothactivation and inhibition of information re-trieved from long-term memory and newlyaccessed from perception. Relations are crit-ical to thought and the working-memory sys-tem therefore must provide for the flexiblebinding of information. It also allows theproblem solver to maintain goals that allowsuccessful navigation of single problems butalso allows for integration of various parts oflarger problems. Working-memory capacityis limited, and this is an important individ-ual difference that affects and perhaps evendetermines analytic intelligence. We knowthat working memory is critically dependenton prefrontal cortex functioning, but likelyinvolves the successful activation and inhi-bition of large networks in the brain. Main-tainence of information in working mem-ory tends to be somewhat modality specific;however, attentional resources typically as-cribed to a central executive tend to bemore modality independent and allow forthe connection of information from differ-ent modalities.

The future of working memory researchresides in better understanding how theseprocesses operate in the brain. Computa-tional approaches allow researchers to makeprecise statements about functional pro-cesses necessary for a working-memory sys-tem to perform thinking and can provideuseful predictions for evaluation with cogni-tive neuroscience methods. Whereas mucheffort has been placed on understandingwhere working memory resides in the cor-tex, much less attention has focused on howit functions. Understanding the neural pro-cesses underlying working memory will al-most certainly require tight integration ofmethods that provide good spatial localiza-tion (e.g., fMRI) and good temporal informa-tion (e.g., electrophysiology) in the brain.

Acknowledgments

Preparation of this chapter was supported bygrants from the Office of Naval Research,Xunesis (www.xunesis.org), and National

Service Research Award (MH-064244) fromthe National Institute of Mental Health. Ap-preciation is due to Marsha Lovett and KeithHolyoak for comments on an earlier draft ofthis chapter.

Notes

1 . The term “working memory” was originallyused to describe rat behavior during radial armmaze learning [see Olton (1979) for a de-scription of this literature). It was also usedby Newell and Simon (1972)] to describe thecomponent of their computational models thatholds productions – that is, operations thatthe model must perform (see also Lovett andAnderson, Chap. 1 7).

2 . Fuster (1997) has long argued for this approachto working memory based on electrophysiolog-ical and cortical cooling data from nonhumanprimates. In Fuster’s model, neurons in pre-frontal cortex drive neurons in more posteriorbrain regions that code for the information tobe activated in long-term memory. This per-spective is also consistent with Cowan’s (1988)Embedded-Processes model. See also Chein,Ravizza, and Fiez, 2003 .

3 . Both ACT (Lovett and Anderson, Chap. 1 7)and LISA (Doumas and Hummel, Chap. 4)provide accounts of how working memory maybe involved in higher-level cognition. Thesetheories and computational implementationsprovide excellent starting points for investi-gating how the brain actually accomplisheshigh-level thought. An excellent edited vol-ume by Miyake and Shah (1999) reviews manyof the traditional computational perspectiveson working memory.

References

Atkinson, R. C. & Shiffrin, R. M. (1968). Hu-man memory: A proposed system and its con-trol processes. In K. W. Spence, & J. T. Spence(Eds.), The Psychology of Learning and Motiva-tion Vol. 2 (pp. 89–195). New York: AcademicPress.

Baddeley, A. D. (1966). The capacity for gener-ating information by randomization. QuarterlyJournal of Experimental Psychology, 1 8, 1 19–29.



Baddeley, A. D. (1986). Working Memory. Ox-ford, UK: Oxford University Press.

Baddeley, A. D. (1996). Exploring the central ex-ecutive. Quarterly Journal of Experimental Psy-chology, 49A, 5–28.

Baddeley, A. D. (2000). The episodic buffer: Anew component of working memory? Trends inCognitive Sciences, 4 , 417–423 .

Baddeley, A., Emslie, H., Kolodny, J., & Dun-can, J. (1998). Random generation and the ex-ecutive control of working memory. QuarterlyJournal of Experimental Psychology, 51 A, 819–852 .

Baddeley, A., & Hitch, G. J. (1974). Workingmemory. In G. H. Bower (Ed.), The Psychologyof Learning and Motivation Vol. 8 (pp. 47–89).New York: Academic Press.

Bjorklund, D. F., & Harnishfeger, K. K.(1990). The resources construct in cognitivedevelopment: Diverse sources of evidence anda theory of inefficient inhibition. Developmen-tal Review, 1 0, 48–71 .

Brooks, L. R. (1968). Spatial and verbal compo-nents of the act of recall. Canadian Journal ofPsychology, 2 2 , 349–368.

Carpenter, P. A., Just, M. A., & Shell, P. (1990).What one intelligence test measures: A theo-retical account of the processing in the ravenprogressive matrices test. Psychological Review,97, 404–431 .

Cattell Culture Fair Test (Institute for Personalityand Ability Testing, 1973). Champaign, IL.

Chein, J. M., Ravizza, S. M., & Fiez, J. A.(2003). Using neuroimaging to evaluate mod-els of working memory and their implicationsfor language processing. Journal of Neurolin-guistics, 1 6, 315–339.

Christoff, K., & Gabrieli, J. D. E. (2000). Thefrontopolar cortex and human cognition: Evi-dence for a rostrocaudal hierarchical organiza-tion within the human prefrontal cortex. Psy-chobiology, 2 8, 168–1 86.

Christoff, K., Prabhakaran, V., Dorfan, J., Zhao,Z., Kroger, J. K., Holyoak, K. J. et al. (2001 ).Rostrolateral prefrontal cortex involvement inrelational integration during reasoning. Neu-roimage, 1 4 , 1 1 36–1 1 49.

Conrad, R. (1964). Acoustic confusion in imme-diate memory. British Journal of Psychology, 55 ,75–84 .

Conway, A. R. A., & Engle, R. W. (1994).Working memory and retrieval: A resource-

dependent inhibition model. Journal of Experi-mental Psychology: General, 1 2 3 , 354–373 .

Cowan, N. (1988). Evolving conceptions of mem-ory storage, selective attention, and their mu-tual constraints within the human informationprocessing system. Psychological Bulletin, 1 04 ,163–191 .

Cowan, N. (1995). Attention and Memory: An In-tegrated Framework. New York: Oxford Univer-sity Press.

Cowan, N. (1999). An embedded-processesmodel of working memory. In Akira Miyake &Priti Shad (Eds.) Models of Working Memory:Mechanisms of Active Maintenance and Exec-utive Control (pp. 62–101 ). Cambridge, UK:Cambridge University Press.

Dagenbach, D. E., & Carr, T. H. (1994). InhibitoryProcesses in Attention, Memory, and Language.San Diego, CA: Academic Press.

Daneman, M., & Carpenter, P. A. (1980). Individ-ual differences in working memory and read-ing. Journal of Verbal Learning and Verbal Be-havior, 1 9, 450–466.

Dempster, F. N. (1991 ). Inhibitory processes:A neglected dimension of intelligence. Intelli-gence, 1 5 , 1 57–1 73 .

Dempster, F. N., & Brainerd, C. J. (Eds.). (1995).Interference and inhibition in cognition. SanDiego, CA: Academic Press.

de Renzi, E., & Nichelli, P. (1975). Verbal andnon-verbal short-term memory impairmentfollowing hemispheric damage. Cortex, 1 1 ,341–354 .

Diamond, A. (1990). The development and neu-ral bases of memory functions as indexed bythe A-not-B and delayed response tasks in hu-man infants and infant monkeys. In DiamondA. (Ed.), The Development and Neural Bases ofHigher Cognitive Functions (pp. 267–317). NewYork: New York Academy of Sciences.

Engle, R. W., Carullo, J. J., & Collins, K. W.(1991 ). Individual differences in working mem-ory for comprehension and following direc-tions. Journal of Educational Research, 84 , 253–262 .

Engle, R. W., Kane, M. J., & Tuholski, S. W.(1999). Working memory and controlled atten-tion. In Akira Miyake & Priti Shad (Eds.), Mod-els of Working Memory: Mechanisms of ActiveMaintenance and Executive Control (pp. 102–1 34). Cambridge, UK: Cambridge UniversityPress.



Ericsson, K. A. & Kintsch, W. (1995). Long-termworking memory. Psychological Review, 1 02 ,21 1–245 .

Fodor, J. A., & Pylyshyn, Z. W. (1988). Connec-tionism and cognitive architecture: A criticalanalysis. In S. Pinker & J. Mehler (Eds.), Con-nections and Symbols (pp. 3–71 ). Cambridge,MA: MIT Press.

Fuster, J. M. & Alexander, G. E. (1971 ). Neuronactivity related to short-term memory. Science,1 73 , 652–654 .

Fuster, J. M. (1997). The Prefrontal Cortex:Anatomy, Physiology, and Neuropsychology ofthe Frontal Lobe. (3 rd ed.). Philadelphia:Lippincott-Raven.

Fuster, J. M., Bodner, M., & Kroger, J. K. (2000).Cross-modal and cross-temporal association inneurons of frontal cortex. Nature, 405 , 347–351 .

Gilhooly, K. J., Logie, R. H., Wetherick, N. E., &Wynn, V. (1993). Working memory and strate-gies in syllogistic-reasoning tasks. Memory andCognition, 2 1 , 1 1 5–1 24 .

Gray, C. M., Koenig, P., Engel, A. K., & Singer,W. (1989). Oscillatory responses in cat visualcortex exhibit inter-columnar synchronizationwhich reflects global stimulus properties. Na-ture, 338: 334–337.

Hasher, L., & Zacks, R. T. (1988). Working mem-ory, comprehension, and aging: A review and anew view. In G. Bower (Ed.), The Psychology ofLearning and Motivation: Advances in Researchand Theory (Vol. 22 , pp. 193–225). San Diego,CA: Academic Press.

Hebb, D. O. (1949). The Organization of Behavior.New York: Wiley.

Hummel, J. E., & Holyoak, K. J. (1997). Dis-tributed representations of structure: A theoryof analogical access and mapping. PsychologicalReview, 1 04 , 427–466.

Hummel, J. E., & Holyoak, K. J. (2003). Asymbolic-connectionist theory of relational in-ference and generalization. Psychological Re-view, 1 1 0, 220–264 .

James, W. (1 890). Principles of Psychology. NewYork: Henry Holt & Co.

Johnson-Laird, P. N. (1983). Mental Mod-els. Cambridge, UK: Cambridge UniversityPress.

Just, M. A., & Carpenter, P. A. (1992). A capac-ity theory of comprehension: Individual differ-ences in working memory. Psychological Review,99, 1 22–1 49.

Kane, M. J., & Engle, R. W. (2003a). The roleof prefrontal cortex in working-memory ca-pacity, executive attention, and general fluidintelligence: An individual-differences per-spective. Psychonomic Bulletin and Review, 9,637–671 .

Kane, M. J., & Engle, R. W. (2003b). Working-memory capacity and the control of attention:The contributions of goal neglect, responsecompetition, and task set to Stroop interfer-ence. Journal of Experimental Psychology: Gen-eral, 1 32 , 47–70.

Kirby, K. N., & Kosslyn, S. M. (1992). Thinkingvisually. In G. W. Humphreys (Ed.), Under-standing Vision: An Interdisciplinary Perspective.(pp. 71–86). Oxford, UK: Blackwell.

Klauer, K. C., Stegmaier, R., & Meiser, T. (1997).Working memory involvement in propositionaland spatial reasoning. Thinking and Reasoning,3 , 9–47.

Kyllonen, P. C., & Christal, R. E. (1990). Reason-ing ability is (little more than) working- mem-ory capacity?! Intelligence, 1 4 , 389–433 .

Logie, R. H. (1995). Visuo-spatial Working Mem-ory. Hillsdale, NJ: Erlbaum.

Markman, A. B., & Gentner, D. (1993). Struc-tural alignment during similarity comparisons.Cognitive Psychology, 2 5 , 431–467.

Miller, E. K., & Cohen, J. D. (2001 ). An integra-tive theory of prefrontal cortex function. An-nual Review of Neuroscience, 2 4 , 167–202 .

Miller, G. A. (1956). The magical number seven,plus or minus two: Some limits on our capac-ity for processing information. Psychological Re-view, 63 , 81–97.

Milner, B. (1966). Amnesia following operationon the temporal lobes. In C. W. M. Whitty &O. L. Zangwill (Eds.), Amnesia (pp. 109–1 33).London: Butterworths.

Miyake, A., Friedman, N. P., Emerson, M. J.,Witzki, A. H., & Howerter, A. (2000). Theunity and diversity of executive functions andtheir contributions to complex ‘frontal lobe’tasks: A latent variable analysis. Cognitive Psy-chology, 41 , 49–100.

Miyake, A. & Shah, P. (1999). Models of workingmemory: Mechanisms of active maintenance andexecutive control. New York: Cambridge Uni-versity Press.

Morrison, R. G., Holyoak, K. J., & Truong,B. (2001 ). Working memory modularity inanalogical reasoning. Proceedings of the Twen-tyThird Annual Conference of the Cognitive



Science Society (pp. 663–668). Mahwah, NJ:Erlbaum.

Morrison, R. G., Krawczyk, D. C., Holyoak, K. J.,Hummel, J. E., Chow, T. W., Miller, B. L.,et al. (2004). A neurocomputational modelof analogical reasoning and its breakdown inFrontotemporal Lobar Degeneration. Journal ofCognitive Neuroscience, 1 6, 1–1 1 .

Muller, M. M., Bosch, J., Elbert, T., Kreiter, A.,Sosa, M. V., Sosa, P. V., et al. (1996). Visu-ally induced gamma-band responses in humanelectroencephalographic activity: A link to an-imal studies. Experimental Brain Research, 1 1 2 ,96–102 .

Murdock, B. B., Jr. (1962). The retention of in-dividual items. Journal of Experimental Psychol-ogy, 64 , 482–488.

Newman, S. D., Carpenter, P. A., Varma, S. &Just, M. A. (2003). Frontal and parietal par-ticipation in problem solving in the Tower ofLondon: fMRI and computational modeling ofplanning and high-level perception. Neuropsy-chologia, 41 , 2003 .

Newell, A., & Simon, H. A. (1972). Human prob-lem solving. Englewood Cliffs, NJ: Prentice-Hall.

Norman, D. A., & Shallice, T. (1980). Atten-tion to action: Willed and automatic control ofbehaviour. University of California, San DiegoCHIP Report 99.

Norman, D. A., & Shallice, T. (1986). Attentionto action: Willed and automatic control of be-havior. In G. E. Schwartz & D. Shapiro (Eds.),Consciousness and Self-Regulation. New York:Plenum Press.

Olton, D. S. (1979). Mazes, maps and memory.American Psychologist, 34 , 583–596.

Prabhakaran, V., Narayanan, K., Zhao, Z., &Gabrieli, J. D. E. (2000). Integration of di-verse information in working memory withinthe frontal lobe. Nature Neuroscience, 3(1 ), 85–90.

Raven, J. C. (1938). Progressive Matrices: A Per-ceptual Test of Intelligence, Individual Form.London: Lewis.

Ruchkin, D. S., Grafman, J., Cameron, K., &Berndt, R. (2003). Working memory retention

systems: A state of activated long-term mem-ory. Behavioral and Brain Sciences, 2 6, 709–777.

Shallice, T., & Warrington, E. K. (1970). Inde-pendent functioning of verbal memory store:A neuropsychological study. Quarterly Journalof Experimental Psychology, 2 2 , 261–273 .

Shimamura, A. P. (2000). The role of the pre-frontal cortex in dynamic filtering. Psychobiol-ogy, 2 8, 207–218.

Smith, E. E., & Jonides, J. (1997). Workingmemory: A view from neuroimaging. CognitivePsychology, 33 , 5–42 .

Spearman, C. (1923). The Nature of Intelligenceand the Principles of Cognition. London, UK:MacMillian.

Sternberg, R. J. (1977). Intelligence, InformationProcessing, and Analogical Reasoning: The Com-ponential Analysis of Human Abilities. Hillsdale,NJ: Erlbaum.

Toms, M., Morris, N., & Ward, D. (1993). Work-ing memory and conditional reasoning. Quar-terly Journal of Experimental Psychology: HumanExperimental Psychology, 46, 679–699.

Turner, M. L. & Engle, R. W. (1989). Is work-ing memory capacity task dependent? Journalof Memory and Language, 2 8, 1 27–1 54 .

Viskontas, I. V., Morrison, R. G., Holyoak, K. J.,Hummel, J. E., & Knowlton, B. J. (in press).Relational integration, inhibition and analog-ical reasoning in older adults. Psychology andAging.

Waltz, J. A., Knowlton, B. J., Holyoak, K. J.,Boone, K. B., Mishkin, F. S., de Menezes San-tos, M., et al. (1999). A system for relationalreasoning in human prefrontal cortex. Psycho-logical Science, 1 0, 1 19–1 25 .

Waltz, J. A., Lau, A., Grewal, S. K., & Holyoak,K. J. (2000). The role of working memory inanalogical mapping. Memory and Cognition, 2 8,1 205–1 21 2 .

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

Welsh, M. C., Satterlee-Cartmell, T., & Stine, M.(1999). Towers of hanoi and london: Contribu-tion of working memory and inhibition to per-formance. Brain and Cognition, 41 , 231–242 .


474

Documents

COGNITIVE AND NEURAL CONSTRAINTS ON HUMAN THOUGHTxunesis.org/publications/Morrison_CHTR_2005.pdf · 2005. 10. 12. · curs because these initial items have been stored in long-term