ArticlePDF Available

The tripartite model of representation

September 2002
Philosophical Psychology 15(3):239-270

Authors:

UNSW Sydney

Robert Cummins [(1996) Representations, targets and attitudes, Cambridge, MA: Bradford/MIT, p. 1] has characterized the vexed problem of mental representation as "the topic in the philosophy of mind for some time now." This remark is something of an understatement. The same topic was central to the famous controversy between Nicolas Malebranche and Antoine Arnauld in the 17th century and remained central to the entire philosophical tradition of "ideas" in the writings of Locke, Berkeley, Hume, Reid and Kant. However, the scholarly, exegetical literature has almost no overlap with that of contemporary cognitive science. I show that the recurrence of certain deep perplexities about the mind is a systematic and pervasive pattern arising not only throughout history, but also in a number of independent domains today such as debates over visual imagery, symbolic systems and others. Such historical and contemporary convergences suggest that the fundamental issues cannot arise essentially from the theoretical guise they take in any particular case.

Content uploaded by Peter Slezak

Content may be subject to copyright.

PHILOSOPHICAL PSYCHOLOGY

VOL

2002

The tripartite model of representation

ETER

LEZAK

BSTRACT

Robert Cummins [(1996) Representations, targets and attitudes, Cambridge, MA:

Bradford/MIT, p. 1] has characterized the vexed problem of mental representation as “the topic in

the philosophy of mind for some time now.” This remark is something of an understatement. The

same topic was central to the famous controversy between Nicolas Malebranche and Antoine Arnauld

in the 17th century and remained central to the entire philosophical tradition of “ideas” in the

writings of Locke, Berkeley, Hume, Reid and Kant. However, the scholarly, exegetical literature has

almost no overlap with that of contemporary cognitive science. I show that the recurrence of certain

deep perplexities about the mind is a systematic and pervasive pattern arising not only th roughout

history, but also in a number of independent domains today such as debates over visual imagery,

symbolic systems and others. Such historical and contemporary convergences suggest that the

fundamental issues cannot arise essentially from the theoretical guise they take in any particular case.

… if men had been born blind philosophy would be more perfect, because

it would lack many false assumptions that have been taken from th e sense

of sight. (Galileo Galilei, 1610)

Mental representation: “the topic for some time now”

Robert Cummins (1996, p. 1) has recently characterized the vexed problem of

mental representation as “the topic in the philosophy of mind for some time now.”

However, this remark is something of an understatement. In fact, the same topic was

central to the famous controversy between Antoine Arnauld and Nicolas Male-

branche in the 17th century, and also central to the entire philosophical tradition of

“ideas” in the writings of Locke, Berkeley, Hume, Reid and Kant. This pattern of

recurrence is a striking fact. However, the cognitive s cience lit erature has almost no

overlap with that of the history of early modern philosophy. This mutual neglect is

remarkable in view of the intimate connection of their c oncerns. I am c oncerned

here to reveal something of the rich and mutually illuminating connections between

these disjoint literatures. In principle, such mutual illumination can make a valuable

and perhaps novel contribution both to contemporary cognitive science and also to

the scholarship of early modern philosophy. The possibility of mutual bene!t is even

Peter Slezak, Program in Cognitive Science, School of History & Philoso phy of Science, University

of New South Wales, Sydney, NSW 2052, Australia, email: p.slezak@unsw.edu.au

ISSN 0951-5089/print/ISSN 1465-394X/online/0 2/030239–32

2002 Taylor & Francis Ltd

DOI: 10.1080/0951508021 000006085

240

SLEZAK

more evident when we notice that the parallels extend beyond merely having

common concerns. That is, not only the problems of the 17th century, but the same

solutions are being rehearsed today at the forefront of research in cognitive science.

Descartes’ de´ja` vu: Edelman’s Traite´ de l’homme

A preliminary indication of the modern relevance of early philosophy may be seen

in Edelman’s (1998) work on perception. Despite its concern with the latest theories

of perception, the central problem is stated in terms identical with that of the entire

tradition of writers on “ideas” since the 17th century. Edelman writes: “Advanced

perceptual systems are faced with the problem of securing a principled (ideally,

veridical) relationship be tween the world and its internal r epresentation.” Edelman’s

bold new solution “is a call for the representation of similarity instead of representa-

tion by similarity.” However, this might have been taken verbatim from Descartes’s

Traite´ de l’h omme (1662/1972) or Dioptrics (1637/1985) where he said “the problem

is to know simply how [images] can enable the soul to have sensory perceptions of

all the various qualities of th e objects to which they correspond—not to know how

they can resemble these objects” (Descartes, 1637/1985, pp. 1, 165). In the same

vein as Edelman, Meyering (1997) points out that, despite its advocates today

(Wright, 1993), resemblanc e cannot be analyzed without circularity. As we will see,

this issue arises inescapably as part of a deeper problem concerning the nature of

representation.

The tripartite schema

In a recent article, Bechtel (1998, p. 299) states the essentials of a modern theory of

representation: “There are … three interrelated components in a representational

story: what is represented, the representation, and the user of the representation.”

Z: System Using Y

Y: Representation

X: Thing Represented

Bechtel’s schema articulates a tripartite conception of ideas as representatives

intervening between the mind an d the world. As we will see, among the problematic

assumptions, Bechtel’s diagram (modi!ed here) and discussion crucially fail to

distinguish internal and externa l representations (see Abell & Currie, 1999). Impor-

tantly, Bechtel’s conception in this regard is not idiosyncr atic, but accurately re"ects

an almost universal conception in cognitive science (D ennett, 1978a; Lloyd, forth-

coming; Newell, 1986, p. 33; Rumelhart & Norman, 1983). As w e will note

presently, the same tripartite conception in the case of the pictorial t heory of images

inherently involves the same assimilation of internal and external representations,

and thereby encourages the illegitimate postulate of a user or external observer—the

notorious h omunculus. I will suggest that the same tacit assimilation of external and

internal representations is at the heart of Searle’s (1980) “refutation” of symbolic AI

and also leads to the doctrine that we think “in” language (Carruthers, 1996; Slezak,

2002). The assimilation just noted in Bechtel will also be seen in the seemingly

unrelated problem of consciousness and the mind–body p roblem (Place, 1956). The

THE TR IPARTITE MODEL OF REPRESENTATION

241

tripartite scheme appears obvious and innocuous enough, though it has been

remarkably fraught with dif!culties. Indeed, the inescapability and ubiquity of this

picture in one form or another is apparent from the fact that Bechtel’s diagram is a

variant of the scheme which we s ee throughout the long history of the subject.

“The vision of all things in God”

Thus, for example, nothing could seem more remote from mo dern theories in

cognitive science today than Malebranche’s (1712/1997) 17th century doctrine of

“the vision of all things in God ”—the theory that ideas are objects of our perception

that exist in God’s mind. On the contrary, however, despite the theological trap-

pings, it is instructive to recognize the profound af!nity of Malebranche’s views with

those at the very forefront of theorizing today in psycholog y and ar ti!cial intelli-

gence: Malebranche’s theory is just Bechtel’s tripartite model (Nadler, 1992), and

the modern problem of representation is h ow to avoid the notorious dif!culties

clearly articulated by his critic Arnauld (1683/1990).

Although these parallels need to be defended with detailed argument and

exegesis, it is signi!cant that Fodor has occasionally made the af!nities explicit.

Recently he suggests that his own Representational Theory of the Mind (RTM) may

be understood on the model of th e classical empiricist conception:

Just for the purposes of building intuitions, think of m ental representations

on the model of what Empiricist philosophers sometimes called “Ideas”.

That is, think of them as mental particulars endowed with causal powers

and susceptible of semantic evaluation. (Fodor, 1998 , p. 7)

In this light, it is hardly surprising that modern problems might be simply the

reinvention of old problems in a new guise. Fodor endorses the classical conception

of ideas though he rejects a conception of representation by means of resembling

images. He says “To a !rst approximation, … the idea that there are mental

representations is the idea that there are Ideas minus the idea that Ideas are image s”

(1998, p. 8). Despite such disclaimers, we will see that there is particular irony in the

fact that the problem for images may be, at a deeper level, the problem for Fodor’s

RTM as well.

Independently of Fodor’s explici t allusion, his conception of a representational

theory of mind has always been evocative of traditiona l acco unts. Thus, it may be

just a fac¸on de parler, but Fodor’s (1978) analysis of propositional att itudes has

consistently been expressed in terms of “relations between organisms and internal

representations” which are “sentence-like entities” (1978, p. 198), that is, “formulae

in an Internal Representational System” (1978, p. 194) and whose intentional

contents refer to things in the world. Fodor earlier explained this idiom by the same

analogy with traditional theories:

This is, quite generally, the way that representational theories of the mind

work. So in classical versions, thinking of John (construed opaquely) is a

242

SLEZAK

relation to an “idea”—viz., to an internal representation of John. (1978,

p. 200)

Fodor speaks of internal representations as the “immediate objects” of beliefs,

thereby revealing the close similarity of his theory with the classical Lockean theory

of ideas as the “immediate objects” o f perception. This conception of internal

representations as being in a relation to a person is an explicit tripartite scheme

which Fodor takes to specify “a priori” conditions on propositional attitudes. This

may be, at best, an awkward locution and, at worst, encouraging a notoriously

problematic theory.

It is important to acknowledge that Fodor and Bechtel, like most theorists, are

fully aware of the fat al problem lurking here in principle. However, awareness of the

problem in principle does not necessarily preclude falling victim in practice. For

example, we will see that proponents of pictorial imagery have been repeatedly

charged with committing the homunculus error. Notwithstanding their advocates’

protestations of innocence [1] and full awareness of the hazards, there are grounds

for seeing pictorial theories as problematic in the traditional m anner. The charge is

that the representational format cannot be made to work without tacitly invoking the

very abilities it is supposed to explain (Pylyshyn, 1973, 1978, 1981, in press; Slezak,

1992, 1995, in press). The error is not con!ned to imagery and is made unwittingly

by failing to notice that the accessing mechanisms cannot perform their function on

their own in view of the particular properties ascribed to the representation. As

Bechtel (1998 , p. 299) no tes, a process which uses a representation as “stand in”

must be coordinated with the format of the representation. However, the nature of

the format may be such as to require a user which is not merely a process in an

innocuous sense. Speci!cally, taking internal representations to be too closely

modeled on our external representational artifacts clearly risks requiring the “user”

to share our relevant pe rceptual and cognitive abilities, thereby begging the question

in the traditional manner. We will see that the assimilation of internal and external

representations in just this way is frequently made as an explicit doctrine. Adverting

to the virtues of computational models which ultimately discharge their homunculi

and pay back their loan on intelligence (Dennett, 1978b) is not suf!cient as a plea

of innocence to these charges (Kosslyn et al., 1979). As Rorty (1979, p. 235) has put

it, there is no advance in replac ing the lit tle ma n in the head by a little machine in

the head. In particular, I will argue later that t he common appeal to an internal

symbolic language analogous to a formal sys tem appears to be guilty of the same

charge. The dispute, then, is about whether the theoretical models succeed in

avoiding the well-known dif!culty desp ite their authors’ int entions.

Thus, although rejecting the charge of being “ontologically promiscuous”

(1978, p. 179), Fodor’s locution may be symptomatic of the deep dif!culties which

pervade the problem of representation. Signi!cantly, Fodor says that his conception

corresponds precisely with the view that psychologists have independently arrived at.

To the extent that Fodor is correct in this observation, not only philosophers have

been prey to the deeply compelling mistakes of theorizing about the mind.

It is no accident that Gibson’s “ecological” approach, like the closely related

THE TR IPARTITE MODEL OF REPRESENTATION

243

“situated cognition,” are theories of direct realism which have been proposed as

alternatives to the representationalism of computational theories. This is merely one

form in which the Malebranche–Arnauld debate is being rehearsed today. This

celebrated debate is described by Nadler (1989) as a debate between an “o bject

theory” of ide as and an “act theory,” respectively. He explains

… the object theory of ideas involves a commitment to a representationalist

or indirect realist theory of perception, such as Malebranche (and, on the

traditional reading, Lo cke) put forth. An act theory of ideas, on th e other

hand, forms the c ore of Ar nauld’s perceptual direct realism. If ideas are

representational mental acts [rather than entitie s], then they can put the

mind in direct cognitive contact with the world—no intervening proxy, no

tertiu m quid, gets in the way. (1989, p. 6)

As Nadler (1989, p. 6) points out, Malebranche’s “vision in God” is a

“theo logization of cognition” according to which the contents of our own thoughts

are dependent upon their divine source in the mind of God. However, although

Malebranche’s theological and epistemological concerns are woven together, the

threads may be separated and his doctrine of ideas identi!ed as the familiar,

compelling and widely held theory until the present t ime. Although there is room for

scholarly dispute [2], most commentators share a reading of Malebranche according

to which ideas are intermediaries or proxie s representing external objects and

intervening between the mind and the world. This same “representative theory of

perception” has been more familiar as John Locke’s “veil of ideas” in the tradition

referred to as the “way of ideas.” On this view, ideas are internal mental objects of

some kind toward which the mind’ s operations are directed. Nadler ec hoes Bechtel,

describing Malebranche’s theory as assuming that there are three elements in the

normal perception or knowledge of the world (Nadler, 1989, p. 81). As Arnauld

explained in his critique, Malebranche “regards this representation as bein g act ually

distinct from our mind as well as from the object” (1683/1990, p. 63). A crucial and

frequently quoted passage from Malebranche himself explains:

Thus, it does not see them by themselves, and our mind’s immediate object

when it sees the sun, for example, is not the sun, but something that is

intimately joined to our soul, and this is w hat I call an idea. Thus, by the

word idea, I mean here nothing other than the immediate object, or the

objec t closest to the mind, when it perceives something, ie., that which

affects and modi!es t he mind with th e perception it has of an object.

(1712/1997, p. 217)

Situated cognition: the “canonical” cottage cheese case

Signi!cantly, John Yolton has expressed a hope that from the study of early thinkers

“we may be able to understand how we can have representation (cognitivity) and

realism too” (1996, p. x). This is, of course, a comment on the perennial problem

244

SLEZAK

posed by the tripartite scheme. Yolton’s remarks on earlier thinkers is apt to describe

the central problem of theories today:

The pivotal concept for the accounts of perceptual acquaintance in the

seventeenth and eighteenth centuries is that of objects present to the mind.

Depending on how that concept was interpreted, those accounts moved

between an indirec tness of knowledge (because only a representative, proxy

objec t can be present to the mind) and a strong direct realism where the

objec t known was, in some way, itself present to or in the mind. (1984,

p. 6)

Recent proponents of “situated cognition” have been complainin g of exactly the

same indirect, mediated conception in computational theories of cognition, recog-

nizing that these embody essentially the Locke–Malebranche’s scheme of represen-

tations intervening between mind and world. For example, Greeno (1989)

unknowingl y echoes Arnauld:

I am pe rsuaded … that in normal activity in physical and social settings, we

are connected directly with the environment , rather than connected in-

directly through cognitiv e representations.

… An individual in ordinary c ircumstances is considered as interacting

with the structures of situations directly, rather than constructing represen-

tations and interacting with the representations. (1989, p. 290)

Greeno cites the Weight Watcher who had studied calculus but nevertheless answers

a question about a daily allotment of cottage cheese by means of a simple, directly

physical, op eration dividing up a portion of cheese, rather than by any symbolic

computation such as a multiplication o n fractions. Ergo, reasoning is not symbolic

but “situated.” The Weight Watcher case is supposed to illustrate the th esis that the

person’s actions are somehow unmediated by mental representations [3]. The cause

for Greeno’s concern is the modern version of Locke’s view: “It is evident that the

mind knows not things immediately, but only by the intervention of the ideas it has

of them” (Locke, 1690, Book IV, Chapter IV).

Deux Carte´siens: plus c¸a change, plus c’est la meˆ me chose

It is amusing to notice ho w Malebranc he’s attempt to articulate this picture is

echoed today by Fodor. Malebranche wrote:

I think everyo ne agrees that we do not perceive objects external to us by

themselves. We see the sun, the stars and an in!nity of objects external to

us; and it is not likely that the soul should leave the body to stroll about the

heavens, as it were, in order to behold all these o bjects. (1712/1997,

p. 217)

Fodor writes in the same vein:

It is, to repeat, puzzling how thought could mediate between behavior and

THE TR IPARTITE MODEL OF REPRESENTATION

245

the world … The trouble isn’t—anyhow, it isn’t solely—thinking that

thoughts are somehow immaterial. It’s rather that thoughts need to be

in more places than seem s possible if they’re to do the job that

they’re assigned to. They have to be, as it were, “out there” so that

things in the world can interact with them, but they als o have to be,

as it were, “in here” so that they can proximally cause behavior.

… it’s hard to see how anything could be both. (Fodor, 1994a,

p. 83)

Malebranche and Arnauld are not chosen for mention here at random or merely

in retrospect for t heir current interest. As Gaukroger (1990) has noted,

Malebranche’s Search after truth was “the most in"uential philosophical treatise

of the second half of the seventeenth century, eclipsed only at the end o f that

century b y Locke’s Essay” (1690/1964, p. 1).

In particular, Malebranche’s doctrines were at the ce nter of a famous contro-

vers y with Antoine Arnauld whose treatise On true and false ideas (1683) was a

reply to Malebranche. Indeed, this debate was not only a major e´ve´nement

intellectuel of its time, as Moreau (1999) has recently described it, but one

whose echoes may be heard throughout the subsequent centuries of specu-

lation about the mind. Moreau’s (1999) recent book-length study in French

is perhaps the !rst devoted to the dispute as such, and attests to its importance

as an intellectual cause ce´le`bre in the 17th century [4]. N adler writes that, follow-

ing the !rst round w ith Arnauld’s critique of Malebranche,

For the next decade, until Arnauld’s death in 1694, these two men engaged

in a public debate that attracted the attention of intellectual circles

throughout Europe. Sides were taken in articles, reviews and letters in the

foremost journals of the day, and the issues were debated by others as hotly

as they were by the primary combatants themselves. … it remains … one of

the most interesting episodes in seventeenth-century intellectual history.

(1989, p. 2)

Nadler adds that the debate is indispensable for understanding the central

philosophical issues of the period, and this is especially true in relation to the

work of Descartes. As t he title of Moreau’s (1999) book indicates, Malebranche

and Arnauld were, despite their differences, !rst and foremost Deux Carte´siens.

Arnauld insisted that his conceptions were faithful to those of Descartes and,

as Nadler notes, “Arnauld would remain committed to la pense´e carte´sienne for

the rest of his life” (1989, p. 34). Though differing over the doctrine of ideas

and perceptual acquaintance, both accepted the fundamental principles of

Descartes’s philosophy (Nadler, 1989, p. 59). Arnauld’s view takes on special

interest today since h is crit ique of Malebranche constitutes a way out of

the analogous problematic conceptions of modern cognitive science.

246

SLEZAK

Precursors: pointless exercise ?

propos of historical re"ections, with some justice, Stephen Gaukroger (1996) in his

landmark intellectual biography of Descartes has described as a “pointless exercise”

the efforts to show the extent to which Descartes, for example, was a precursor of

modern cognitive science. However, in some cases we may discern so mething mor e

than fortuitous, independent reinvention. There is a more interesting kind of

recurrence wh ich deserves attention because it is a manifestation of deeper, and

therefore more illuminating, causes—a chronic malaise whose recurrence is symp-

tomatic of deep pathology.

Noting anticipations o f c urrent theories is likely to be re vealing in both direc-

tions: precursors of cognitive science provide an independent, extensive source of

insight into contemporary issues and, conversely, are t hemselves elucidated in novel

ways unavailable to traditional scholarship. (For preliminary steps in this direction,

see Yolton, 1984, 1996, 2000; Slezak, 1999, 2000.) Thus, beyond merely noting the

parallels, I would like to offer some preliminary diagnosis of the malaise and its

etiology along the lines of Arnauld’s defense of Descartes’ view against Male-

branche.

Tables & chairs: bumping into things

From Yolton’s statement of earlier concerns, we can see their relevance to contem-

porary issues:

From the scholastics’ intelligible species, through th e Cartesian’s objective

reality, to Berkeley’s and Hume’s talk of ideas as the very things them-

selves, we see writers on perception striving for some way to sa y that we

perceive physical ob jects. … One of the ways in which some of the writers

tried to preserve the accuracy, if not the directness, of perceptual awareness

was by talking of a conformity or agreement between ideas and objects;

otherwise they said ideas represent objects. (1996, pp. 1–2)

This is, of course, just the modern problem of intentionality or “psychosemantics”

which Cummins describes as just that of saying “in some illuminating way, what it

is for something in the mind to represent something” (1996, p. 1). Despite the

seeming simplicity o f the phenomenon, the burgeoning liter ature attests to the fact

that there is a c onsensus, at least, on Fodor’s judgment that “of the semanticity of

mental representations w e have, as things now stand, no adequate account” (1985b,

p. 28). Typic ally, Stalnaker, too, says, “There is lit tle agreement about how to do

semantics, o r even about the questions that de!ne the subject of semantics” (1991,

p. 229). Likewise, B.C. Smith confesses, “It should be admitted that how this all

works—how symbols ‘reach out and touch someone’—remains an almost total

mystery” (1987, p. 215).

In a report on the state and prospects of interdisciplinary cognitive science,

Fodor (1985a) joked that philosophers are notorious for having been prey to absurd,

eccentric worries such as the “fear that there is something fundamentally unsound

THE TR IPARTITE MODEL OF REPRESENTATION

247

about tables and chairs.” Nevertheless, he optimistically contrasted such “mere”

philosophical worries with those that occasionally turn out to be “real,” as with the

representational character of cognition. Triumphantly, Fodor points to the fact that

today, unlike other proprietary concerns, this problem is no longer just a philoso-

phers’ preoccupation because its solution has become of general importance as a

precondition of progress in several disciplines of cognitiv e science. However, there

is an acute unintended irony in Fodor’s contrast, because the problem of represen-

tation at th e forefront of cognitive science today is, in fact, identical with the

philosophical anxiety about tables and chairs. In various more or less independent

domains, cognitive scientists have simply rediscovere d the very same sterile conun-

drums which have kept philosophers busy since Descartes.

We see a revealing clue to this commonality in Jackendoff’s (1992, p. 161)

question which is a reductio ad absurdum of contemporary symbolic, computational

theories: In view of the “internalism” and “narrow” syntactic character of computa-

tional symbols, Jackendoff asks facetiously: “Why, if our understanding has no direct

access to the real world, aren’t we always bumping into things?” Jackendoff’s satire

is evocative of Samuel Johnson’s famous response to Berkeley’s “ingenious

sophistry”: “I refute it thus,” he said, that is, by kicking a stone. In both cases,

appealing to bumping into things, the responses bring into relief the way in which

classical and modern theories entail a disc onnection of the mind and the world. The

suggestive parallel between Jackendoff and Johnson is no accident. Jackendoff

captures precisely the paradox charged against Locke and also Malebranche, who

Nadler (1992, p. 7) say s “is often portrayed by his critics as enclosing the mind in

a ‘palace of ideas,’ forever cut off from any kind of cognitive or perceptual contact

with the material world.” Of course, Berkeley’s idealism is just the worry about the

reality of tables and c hairs, and Berkeley’s reaction to Locke’s “ideas” is analogous

to Fodor’s reaction to Simon’s symbols—“m ethodological solipsism.” Seemingly

isolating thought in a realm of its own, the representations intervene between mind

and world—two items whose systematic connections with each other become

mysterious. The traditional problem, rediscovered in cognitive science, is how to

make sense of the relation between these three elements—mind, representation and

world—seemingly essent ial to any model of cognition.

The “philosophick topick” of ideas

In his recent book, Yolton (1996, p. 43) mentions the anonymous author of a

pamphlet written in 1705 titled Philosophick essay concerning ideas who says, “There

is hardly any Topick we shall meet with that the Learned have differed more about

than that of Ideas.” It is a remarkable fact that little has changed in this regard

concerning the “Topick” in dispute, the underlying reasons for the problem and th e

solutions adopted [5]. Although the ter minology of ef"uvia, essences, modes and

substances has been replaced by information processing jargon, the essential issues

are unchanged. Thus, Palmer’s (1978) article on “Fundamental aspects of cognitive

representation” says “Anyone who has attempted to read the literature related to

cognitive representation quickly becomes confused—and w ith good reason. The

248

SLEZAK

!eld is obtuse, poorly de!ned, and embarrassingly disorganized.” After enumerating

a dozen distinct conceptions Palmer adds “These are not c haracteristics of a

scienti!c !eld with a deep understanding of its problem, much less its solution”

(1978, p. 259). The situation does not appear to have improved in the two decades

since Palmer wrote. It is no accide nt that Palmer’s lament and his litany echo

Yolton’s anonymous author because th e theoretical disarray, like the doctrines

themselves, are not unrelated. Suf! cient evidence of this is the fact that the 18th

century auth or’s analysis of the problem and its causes remains appropriate today.

… in considering the Mind, some men do not suf!ciently abstract their

Thoughts from Matter, but make use of such Terms as can prope rly relate

to Matter only, and apply them to the Mind in the same Sense as t hey are

spoken of Matter, such as Images and Signatures, Marks, and Impressions,

Characters and Notes of Things, and Seeds of Thoughts and Knowledge.

(quoted in Yolton, 1956/1993, p. 96)

Translated into current terminology, this is an insightful diagnosis of the latest

disputes concerning representation in cognitive science today. It is, in fact, a

re-statement of Arnauld’s orthodox Cartesian view which insists that mental repre-

sentations cannot be properly characterized in terms taken too directly from those

apt for our e xternal, material representations—the problem of “original” versus

“derived” intentionality.

“Malebranchean Theatre”?

Dennett’s (1991) reference to a “Cartesian Theater” has given wide currency to this

term and thereby served to draw attention to the supposed provenance of a

conception which is, indeed, at the heart of philosophical puzzles about the mind

and consciousness . Indeed, the related mistakes of the “Theater” and the homun-

culus are at the heart of much theorizin g in cognitive science. However, fully

acknowledging the value of Dennett’s analysis, it remains t hat his terminology, at

least, perpetuates an historical solecism. Conceding that Dennett was not concerned

with exegetical, scholarly niceties, it remains important to correct a serious error of

misattribution. The “Theater” in question is more appropriately ascribed to Male-

branche than to Descartes. Although “Malebranchean Theater” does not have the

same pleasing sonority, th ere is good philosophical reason to correct the usage

besides mere historical pedantry. It is important to recognize that a commitment to

the picture of an inner perso n observing a scene on the stage of consciousness is

independent of, and does not follow directly from, dualism. Dennet t recognizes this

in his talk of “Cartesian materialism” (1991, p. 107) which he says is “the view yo u

arrive at when you discard Descartes’ dualism but fail to disc ard the imagery of a

central (but material) Theater where ‘it all comes together’.” However, Dennett

seems to blame Descartes for holding this “Theater” conception together with, or

directly as a consequence of, his dualism. Nevertheless, contrary to Dennett’s

implication , while undeniably a Cartesian dualist, Descartes was emphatically no t a

Cartesian materialist as well. That is, he was not guilty of the Theater fallacy in this

THE TR IPARTITE MODEL OF REPRESENTATION

249

sense. On the contrary, despite positing the anatomica l convergence of nerve

!laments in the pinea l gland, Descartes did not subscribe to the picture of an

observer in the problematic “Theater” because he explicitly argued against positing

concomitant representations of a kind which would require the notorious homun-

culus. Descartes’ “ghost in the machine” is not this observer, but a posit based on

entirely different, independent considerations—namely, the “Cogito” argument and

the lim itation s of machines (Discourse V). The rationale for Descartes’ immaterial

soul is quite different and independent of the “Theater” conception which he

explicitly repudiates in his Dioptrics (1637) and Trea tise of ma n (1662).

Quite apart from the evidence of Descartes’ own texts, ample support for these

ascriptions is found in Arnauld’s writings which articulate an act-theory as an

alternative to Malebranche’s representative “veil of ideas” (see Nadler, 1989, pp. 34,

118, 126, fo otnote 36). Arnauld saw this “direct perception” view as faithful to

Descartes, and Descartes says in correspondence that Arnauld “has entered further

than anyone else into the sense of what I have written” (AT III, 331). Thus,

Arnauld’s Cartesian position adopts precisely Dennett’s stance against the

“Malebr anchean Theater.”

Reinventions: synchronic and diachronic

If Malebranche and Arnauld anticipate d contemporary concerns about representa-

tion in cognitive sc ience, then it is clear that the current theoretical problem has

nothing to do with t he theoretical framework of symbolic, computational approaches

as universally assumed. It is particularly signi!cant, then, that the recurrences of

interest here are found not only throughout history but in seemingly unrelated

domains of cognitive science today. This recurrence of essentially the same dispute

in widely varying contexts con!rms that the underlying problem does not arise

essentially from the special features o f any one of them. Given a seductive mistake

concerning representation as such, multiple, seemingly independent, reinventions

are just what we would expect to !nd. I will presently suggest that we may discern

the same underlying problem at the heart of notorious disputes such as the “Imagery

Debate,” Searle’s Chinese Room conundrum, the thinking-in-languag e debate and

a numbe r of others which have been prominent and recalcitrant.

No representations?

The “cognitive revolution” of the 1960s was characterized by a renewed recognition

of the indispensability of internal representations following their repudiation by

Skinnerian behaviorism. There is considerable irony in recent approaches which

appear to reject internal representations once again (Brooks, 1991; Clark & Tor ibio,

1994; Freeman & Skarda, 1990; Greeno, 1989; van Gelder, 1998). Notwithstandin g

Eliasmith’s (1996) claim, these views are not plausibly s een as a return to behav-

iorism since, strictly speaking, they do not reject internal representations at all (see

Markman & Dietrich , 2000). Nevertheless, these approaches and their rhetoric are

symptoms of the profound dif!culties posed by the phenomena. Particularly in view

250

SLEZAK

of the revolutionary hype associated with the latest fashions, it is sobering to notice

that Arnauld’s (1683/1990) critique of Malebranche exactly pre!gures these recent

attacks on representational theories. It is no coincidence that Arnauld’s treatise On

true and false ideas is concerned to repudiate what he describes as “imaginary

representations,” saying, “I can, I believe, show the falsity o f the hypothesis of

representations” (1683/1990, p. 77) for “one must not make use of alleged entities of

which we have no clear and distinct idea in order to explain the effects of nature,

whether corporeal or spiritual” (1683/1990, p. 65).

Illusions & misrepresentation: “curious and melancholy fact”

In seeking t o understand the per sistence and recalcitrance of the problems of

intentionality, it is instructive to examin e one facet of the issue which reveals the

seductiveness of the mistake. The problem of misrepresentation h as arisen fo r causal

or co-variation th eories of intentional content (Dretske, 1986; Fodor, 1994a) since

these theories seem to be unable to capture t he way a mismatch might arise between

a representation and the world. If a mentalese token “mouse” might be caused not

only by mic e but also by shrews, then the symbol must ipso facto mean “shrew” and

cannot be in error. It seems not to have been noticed that this modern philosophical

problem of mis representation is a var iant of the well-known classical “Argument

from Illusion” (Reynolds, 2000) which was employed in support of Locke’s “ideas”

and A.J. Ayer’s (1940) sense-data as the immediate objects of per ception. The

parallel s hould not be surprising s ince, after all, an illusion in the relevant sense (that

is, an hallucination) is precisely a misrepresentation. The problem of misrepresen-

tation, then, appears to be one of the loose threads which may be pulled to unravel

the rest of the tangled ball (see Slezak, forthcoming).

Responding to Ayer (1940), Austin (1962, p. 61) remarked on the “curious”

and “melancholy fact” that Ayer’s position on sense-data echoes that of Berkeley. It

is an even more melancholy fact today that Fodor’s “real” problems of representa-

tion also echo Berkeley. Q uestions of veridicality for Locke’s ideas and Ayer’s

sense-data arose from precisely the same assumptions as Fodor’s—namely, the

assumption of being able to compare representations and the world.

The earlier Fodorian passage from Malebranche is followed by a paragraph that

explicitly articulates the “Argument from Illusion”:

It should be c arefully noted that for the mind to perceive an object, it is

absolutely necessary for the idea of that object to be actually present to

it—and about this there can be no doubt; bu t there need not be any

external thing like t hat idea. For it often happens that we perceive things

that do not exist, and that even have never existed—thus our min d often

has real ideas of th ings that have never existed. When, for example, a m an

imagines a golden mountain, it is absolutely necessary that the idea of this

mountain really be present to his mind. When a madman or someone

asleep or in a high fever sees some animal before his eyes, it is certain that

what he sees is not nothing, and that therefore the idea of this animal really

THE TR IPARTITE MODEL OF REPRESENTATION

251

does exist, though the golden mountain and the animal have never existed.

(1712, p. 217)

Illusions in this sense are cases in which the correspondence between represen-

tations and world fails—misrepresentations of exactly the sort relevant to the

contemporary puzzle for symbolic, computational accounts of c ognition. In the

modern case, as posed by Fodor (1994a) and Dretske (1986), the problem is, given

causation between these elements, how to explain the possibility of illusion; in the

classical case the problem is, given illusion, how to explain causation. The modern

problem of misrepresentation arises because causal or correlational theories don’t

appear to permit a distinction between true and false representations. If a dog causes

a representation of “cat” in mentalese, on the causal account it must ipso facto count

as meaning “dog” and is, therefore, not a mistaken representation of cat. Con-

versely, the classica l Argument from Illusion, starts from the other end, as it were.

Beginning with the distinction between true and false representations, the Argument

recognizes that these cannot both be correlated with an external reality, and

concludes that in both veridical and non-veridical cases there must be some other

object of direct perception, the “idea” or sense-datum (see Reynolds , 2000). In view

of these analogies, therefore, I suggest it is no coincidence that Fodor’s (1980)

“meth odological solipsism” is st rongly evocative of a Berkeleyan idealism.

The Malebranche–Locke argument for representative ideas recognizes that

illusions cannot be caused in the usual way by external objects—essentially Fodor’s

puzzle expressed in reverse: Fodor argues that, if ideas are caused by external

objects, we can’t have illusions. The parallels h ere appear to be more than super!cial

or terminological [6]. My claim is not that the problem of misrepresentation and the

Argument from Illusion are directed towards the same ends, but only that they arise

from an identical conceptual scheme and are mirror-images of one another: The

classical argument asse rts: if there are illusions, then there is no direct connection or

correlation with the external world (i.e. there must be intermediate objects of

perception); conversely, Fodor’s argument asserts: If there is a direct connection

(i.e. c ausal correlation) with the external world, then there can be no illusions.

These are eq uivalent co ntrapositives: if we take “I”

illusion, “C”

correlation,

then the Malebranche–Locke proposition is [I

® ,

C] and the Fodor–Dretske

propositio n is [C

® ,

I].

In passing, we may note that a degree of confusio n ha s been intro duced in these

discussions by the failure to distinguish crucially different kinds of “illusion.” An

illusion in the sense rele vant to the argument concerning ideas, sense-data or

representations is, strictly speaking , hallucination. Ho wever, certain other phenom-

ena commonly referred to as “illusions” in this context such as mirages or bent sticks

in water are not illusory at all in an important sense. These are veridical perceptions

of the light patterns entering the eye unlike cognitive errors such as the Mu¨ ller–Lyer

illusion. Richard Gregory (1997), for example, has explicitly assimilated these

phenomena , but no theory of cognitive processes could explain the “illusion” in the

case of mirages and seemingly bent sticks due to refracted light. Gregory’s mistake

in this regard is interesting and perhaps no mere mistake . Assuming that our

252

SLEZAK

knowledge of the actual conditions in the world must be use d in characterizing

mental representations is precisely the seductive error which I am concerned to

expose in its various guises. A stick appearing bent in water is a case of “the world

gone wrong” in just the sense of this felicitous phrase used by Fodor, as we will see

presently,

Truth conditions as explanatory?

In both the case of misrepresentation and t hat of illusion the puzzle arises from a

commitment to the tripartite conception in w hich representations intervene between

the mind and the world and are somehow correlated with it. In particular, the

questions of veridicality for Locke’s ideas arose from the impossibility of any

comparison between representations and the world, except from the perspective of

an independent outside observer. As Berkeley recognized, the very distinction

between true and false ideas cannot be made without comparing representations and

the world. Of course, this perspective is unavailable to the mind itself. Correspond-

ingly, an explanatory theory cannot make tacit appeal to such a perspective without

committing the homunculus error. This means that the veridicality or otherwise of

mental representation does not serve an explanatory role and is, therefore, not a

legitimate part of a theory o f m ind. In Berkeley’s idealist response to this problem

we can see the precursor to F odor’s problem arising from a commitment to truth

conditions for mental representations. Securing the veridical connection between

representations and the world through causation simply binds them in such a way as

to preclude error and thus causation functions for Fodor in the way that a mysteri-

ou s correspondence worked for Locke .

Of course the problem of explaining error and that of explaining truth are two

sides of the same coin. Accordingly , t he puzzl e of misrepresentation is symptomatic

of fundamental problems in the conception of mental representations as semanti-

cally evaluable. Fodor is emphatic about the centrality of truth preservation for the

computational RTM . Regarding the fact that mental processes tend to preserve

semantic properties like truth Fodor says

This is, in my view, the most important fact we know about minds; no

doubt it’s why God bothered to give us any. A psychology that can’t make

sense of such facts as that mental processes are typically truth preserving is

ipso facto dead in the water. (Fodor, 1994a, p. 9)

Fodor’s dilemma arises from the fact that content doesn’t ap pear to supervene on

mental processes and, therefore, “semantics isn’t part of psychology” (Fodo r,

1994a, p. 38).

My point, then, is of course not that solipsism is true; it’s just that truth

reference and the rest of the semantic notions aren’t psychological cate-

gories. (1980, p. 253)

It seems that we can’t do psychology with the semantic notions, but we can’t do

THE TR IPARTITE MODEL OF REPRESENTATION

253

psychology without them either. This formulation of Fodor’s dilemma is reminiscent

of a remark by Dennett in a quite different context in which he explained:

… psychology without homunculi is impossible. But psychology with ho-

muncul i is doomed to circ ularity or in!nite regress, so psychology is

impossible. (Dennett, 1978b, p. 123)

My suggestion is that Fodor’s and Dennett’s dilemmas appear to be the same

because at root the puzzle of semantics is a version of the homunculus problem. Just

as Dennett (1978a, p. 122) pointed out that nothing is intrinsically a representation

of anything but only for someone who is the interpreter , so nothing is a misrepresen-

tation for the same reason. That is, Fodor’s current problem of misrepresentation

might be accounted for by noting that it arises from the demand for tacitly adopting

the stance of external interpreter: The very problem itself cannot be cohere ntly

formulated except in terms of judgments which are not part of the explanatory

enterprise. The veridicality of representations is not a property which can play any

role in the functioning of representations or the explanation of them. Like the

picture on a jigsaw puzzle , the meaning of representations conceived as semantically

evaluable in this way is for our own bene!t and not intrinsic to the arrangements of

interlocking components. The sense in which a mental representation does its work

is not one which requires judgment of its truth-value since this is only possible from

the point of view of an observer, the theorist, for whom the representation is

construed as an external symbol.

The very concern with misrepresentation arises from tacitly adopting a ques-

tionable assumption endorsed by Davidson (1975) that having a belief requires also

having the concept of belief, including the concept of error. Davidson says “someone

cannot have a belief unless he understands the possibility of being mistaken, and this

requires grasping the contrast between truth and error—true and false belief” (1975,

p. 22). However, it seems that animals might have beliefs even if they are unable to

know that they have th em and re"ect on their truth-value. A cat can surely be

correct in thinking that a mouse is in a certain hole without having the concepts of

belief and truth.

The judgment of truth or error in a belief must be distinguished from merely

having a belie f which is true or false. We as theorists may judge true and false beliefs

(just as we may judge pictorial resemblance) since these are meta-linguistic or

second-order beliefs, but truth and error are not intrinsic properties of representa-

tion s as such, only to the judgments made about them. Cat psychology must be

possible without invoking cat epistemology.

It is not a big leap from misrepresentation and illusion to notice that images are

a species of the same genus. Imagery involves illusory or non-veridical experiences

of exactly the sort required for the classical argument for sense-data. The proverbial

Pink Elephant of inebriated apprehension is a visual image par excellence, not

relevantly diff erent from Malebranche’s golden mountain or subjects’ imaginings in

the celebrated experiments of Shepard and Metzler (1971) and K osslyn (1994). As

we will see presently, of course, if my conjectured parallel is warranted, it is perhaps

no surprise that the image ry debate has been among the most persistently intractable

254

SLEZAK

disputes in cognitive science also arising from the theorist doing the work of the

theory.

Twin Earth

Putnam’s (1975) Twin Earth puzzles, too, seem to be an unnoticed variant on the

problem of misrepresentation we sa w earlier. In the familiar scenario, instead of my

Twin Earth double, we may substitute myself after having been unknowingly

transported to Twin Earth. There, like my twin in the original story, I will refer to

XYZ as “water.” However, on this variation of t he original scenario my term

“water” now fails to refer correctly rather than being a correct reference with a term

having a different meaning. Since my twin and I are identic al, the two scenarios

must also be indistinguishable. That is, the problem of “wide” and “narrow”

meaning is just the p roblem o f misrep resentation in another guise.

Instead o f th inking of Twin Earth, then, we may imagine alternatively that on

this earth, Go d might have switched all H

O to XYZ without my knowledge. Instead

of taking the original Twin Earth story as showing that my twin must mean

something other than “water,” we may equally conclude that my use of the term is

simply in error when the worlds have been surreptitiously switched. The Twin Earth

scenario is, indeed, simply another wa y of tellin g Dretske’s (1986) story of the

magnetic micro-organisms which are fooled into “thinking” that up is down. Or, in

a different case, as Fodor (in Millikan, 1991, p. 161) has put it, “it’s not the frog but

the world that has gone wrong when a frog snaps at a bee-bee.” Undoubtedly, if the

world is suf!ciently perverse, or it is contrived to alter things in certain ways, our

concepts may accidentally “fail” to refer in the usual manner. It is not clear why such

possibilities should be of interest to a theory of representation for their description

depends on knowledge from a “God’s Eye” p erspective available to the theorist.

Whether the liquid substance is really XYZ or H

O is known only to the external

omniscience of the theorist and has no explanatory role in a theory of representation.

In this sense, the philosophical concern with misrepresentation is analogous to the

spurious assimilation of m irages and seemingl y bent sticks to genuine cognit ive

illusions, as noted earlier. In both cases, the actua l truth about the world is invoked

irrelevantly to explain cognition.

Philosophers as three-year-olds?

Ironically, the mistak e I am indicating is not unknown in cognitive science: In the

cases of interest here, philosophers are like the three-year-olds an d autistics in the

much-discussed “false belief task” o f Wimmer and Perner (1983) (see Carruthers &

Smith, 1996; Davies & Stone, 1995a,b ). Like three-year-olds, philos ophers fail to

discount what they know to be the truth about the world in their “theory of mind.”

The surreptitious switching of XYZ for H

O, bee-bees for " ies or magnetic “up ” for

“down” ar e ways of making “the world go wrong” precisely analogous to switching

the candy while the child is looking in the “false belief” paradigm. Knowing how the

world really is, philosophers truly ascribe false be lief, just as the three-year-olds

THE TR IPARTITE MODEL OF REPRESENTATION

255

falsely ascribe true belief. I n both cases, belief at tributions are independent of any

facts about the believer, depending instead on irrelevant ext ernal facts about the

world. In these cases, the believer’s state of mind can remain !xed and yet the beliefs

can be made to change from true to false by manipulating the world. The child, like

the philosopher naively takes this po ssibility to be relevant to a “theory of mind” in

ascribing mental representations.

Justi!ed true belief?

In case the foregoing analogy may be thought far-fetched or merely whimsical, it is

perhaps worth noting en passant that yet another notorious philosophical puzzle may

be seen to be merely a version of the same problem. Gettier (1963) paradoxes may

be seen as a species of misrepresentation in which the world conspires to make a

propositio n true for reasons which are entirely independent of a person’s grounds for

believing it. In these cases the problem can only be described because the theorist

knows the truth about circumstances which make a belie f accidentally true, even

though the actual circumstances are irrelevant to the agent’s own reasons for

believing the proposition. The Gettier cases are structurally identical with those of

misrepresentation and Twin Earth because the truth or falsity of the mental

representation (i.e. the state of the world) is varied independently of the agent’s

belief-!xing mechanisms. Such considerations in all cases should be irrelevan t to the

problem of understanding mental representation. The moral of the Gettier cases,

like th at of misrepresentation, is that the only sensible, and perhaps the only

possible, theory of knowledge is one that invokes justi!cations and not truth from a

“God’s Eye” perspective [7]. Any adequate, or even complete, account of a person’s

psychology would have to invoke only the relation of beliefs to available evidence

and not their actual, ultimate truth-value. The world can mislead us in various ways,

giving us good reasons for things that may be false, bad reasons for things that may

be true and good reasons for things that may be true fo r other reasons. None of this

should occasion philosophical anxieties for those interested in psychology. Does the

speedometer of a bicycle misrepresent when the bike is ridden on rollers and not

moving? Once again, it is the w orld that h as gone wrong, known to us as external

observers. However, psychology has no obligation to explain why the world may go

wrong. Thus, conceivably, one might contrive things so that Cabernet Sauvignon

replaced the usual liquid in someone’s veins. However, such a p ossibility is of no

more theoretical concern for medical sc ience than Dretske’s (1986) disoriented

microbes are of interest to cognitive science.

Idea-objects

Bechtel’s (1998, p. 299) re-statement of the tripartite model in Malebranche’s terms

makes explicit the widely he ld assumptions which are the potential source of the

dif!culties in understanding representation [8]. In particular, Bechtel’s assimilation

of internal and external representations is acknowledged where he lists the sorts of

“high-level” representations which have been postulated by cognitive scientists.

256

SLEZAK

These include “concepts that might designate objects in the world or linguistic

symbols, !gures and diagrams which we can use in reasoning and problem solving”

(1998, p. 305). Bechtel suggests that if cognition does require such higher-level

representations, “the most plausible analysis is that such rep resentations ar e built

upon these low-level representations and perhaps inherit their content from them”

(Bechtel, 1998, p. 306). However, the dif!culty is that the distinction between

!gures or diagrams, on the one hand, and representations operating in the frog’s

retina, for example, is not simply a matter of higher and lower “levels” in an

unproblematic sense. Linguistic symbols, !gures and diagrams which we use in

reasoning and problem solving, as Bechtel says, are obviously “used” in a sense

which is different and precisely inappropriate for internal mental representations.

“Higher” concepts of this kind could no t inherit their content from low-level

concepts mentioned because t he difference here is not one of level, but of a kind

which precisely de!nes the distinction between original and derived intentionality.

In this regar d Bechte l’s account accurately re"ects the assumption s built in to

the foundational notion of symbolic computation, as Allan Newell explains:

The idea is that there is a class of systems which manipulate symbols, and

the de!nition of these systems is what’s behind the programs in AI. The

argument is very simple. We see humans using symbols all the time. They

use symbol systems like books, they use !sh as a symbol for Christianity,

so there is a whole range of symbolic activity, and that clearly appears to be

essential to the exercise of mind. (1986, p. 33)

This passage is striking for the explicitness with which Newell assimilates internal

mental representations with our external communicative symbols. The assimilation

of representations of radically different kinds appears, then, to be among the

foundational assumptions of cognitive science. It was self-consciously articulated in

Dennett’s (1978a) review of Fodor’s (1975) important work The language of thought

which was the philosophical manifesto for the classical symbolic approach to

cognition:

What is needed is nothing less than a completely general t heory of repre-

sentation, with which we can explain how words, thoughts, thinkers,

pictures, computers, animals, sentences, mechanisms, states, functions ,

nerve impulses, an d formal models (inter alia) can be said to represent one

thing or another. (1978, p. 91)

The hoped-for uni!cation is to be achieved by showing that these seemingly

heterogeneous items are all, in f act, variants of a common, underlying scheme.

Dennett makes this explicit, explaining:

It will not do to divide and conquer here—by saying that these various

things do not represent in the same sense. Of course that is true, but what

is important is that there is something that binds them all together, and we

need a theory that can unify the variety. (1978a, p. 91) .

The pictorial account of imagery is perhaps the clearest example of taking our

THE TR IPARTITE MODEL OF REPRESENTATION

257

external artifacts as the model for internal representations. More generally, the

dif!culties arise from an equivocation on the notion of “understanding” w hich can

mean interpreting a meaningful representation as intelligible, or explaining it as in

science. We will see presently that this con"ation is evidently at th e heart of Searle’s

(1980) Chinese Room scenario, for Searle asks whet her he, as homunculus in the

system, can unde rstand the symbols. This criterion should be irrelevan t to the

question of whether a system ha s “original” intentionality, but Searle’s mistake is

no t his alone. That is, the Chinese Room scenario accurately captures the orthodox

assumptions of the Simon–Newell “physical symbol system hypothesis.” Searle’s

argument is, therefore, best understood, not as a refutation of “strong” AI, but as a

reductio ad absurdum of the widely held assumptions on which AI and cognitive

science are based. I will suggest that this standard formalist or logicist view

conceived on the model of an uninterpreted logical calculus is problematic as an

account of the way an intelligent system is related to the external world by

depending on the intentions of the external user who supplies the interpretation of

the meaningless formal symbols (see Birnb aum, 1991; Rosenschein, 1985). Embrac-

ing this conception, Nilsson (1987, 1991), like Newell above, explicitly invokes our

external symbolic artifacts such as books to defend his view against the

“proc eduralist” position that representations are to be used by the system itself

rather than understood or ascribed meaning by the designer. Undoubtedly recogniz-

ing the problems with invoking an intelligent user or understander, most theorists

would not knowingly embrace such an account, but their intentions may be

inconsistent with the actual properties of their model by virtue of assimilating

external and internal representations [9]. It is not dif!cult to !nd leading theorists

explicitly endorsing this assimilation. Thus, Rumelhart and Norman (1983) wrote:

We de!ne a symbol to be an arbitrary entity that stands for or represents

something else. By “entity” we mean anything that can be manipulated and

examined … Humans also use external devices as symbols, such as the

symbols of writing and printing, electronic displays or speech waves.

(p. 78)

In a recent statement for an encyclopaedi a entry on rep resentation Dan L loyd

explains:

Humans are representing animals, and we have built a world crammed

with representations of many kinds. Consider, for examp le, the number

and variety of pictorial representation: paintings, photographs, moving

pictures, line drawings, caricatures, diagrams, icons, charts, graphs, and

Maps. Add the variety of linguistic representations in signs, titles, texts of

all kinds, and especially spoken words and sentences … Human life, in

short, is largely a cycle of making and interpreting representations. (Lloyd,

forthcoming)

By contrast, Block (1986) recognizes that “The representation on the page must be

read or heard to be understood, but not so for the representation in the brain”

(p. 83). However, despite making this distinction, Block’s discus sion appears to

258

SLEZAK

lapse into the characteristic error. Block asks “what it is to grasp or understand

meaning? ” (p. 82). Of course, we don’t grasp or understand the meaning of our own

mental representations, we just have them. In Dennett’s (1978b) felicitous phrase,

the representations must understand themselves. Arnauld’s words appear to directly

address theorists today:

To say that our ideas and our perceptions (taking these to be the same

thing) represent to us the things that we conceive and that they are their

images, is t o say something completely different from saying that pictures

represent their originals and are the images of them, or that spoken or

written words are the images of our thoughts. For in the case of ideas we

mean that the things we conceive are objectively in our mind and in our

thought. And this way of being objectively in the mind is so peculiar to the

mind and to thought, since it is what speci!cally gives them their nature,

that one seeks in vain anything similar outside the min d and thought. As

I have already remarked, what has thrown the question of ideas into

confusion is the att empt to explain the way in which objects are repre-

sented by ideas by analogy with corporeal things, but there can be no real

comparison between bodies and minds on this question. (Arnauld, 1683/

1990, p. 66)

Symbols & Searle: the meaning of meaning

I have been suggesting that a crucial equivocation on distinct meanings of

“meaning” has led to the postulation of symbols having me aning in an observer-rel-

ative sense in which a representat ion is necessarily apprehended and understood by

someone. Cummins (1996) clearly points to this mistake of construing int ernal

representations as if they may function through being understood. The question of

meaning of mental representations is regularly confused between whether represen-

tations are intelligible and whether they are explainable. Searle (1980) trades directly

on this confusion by asking whether an intelligent understander can interpret the

symbols which are the sub strate of thought. But however we might explain inten-

tionality, it cannot depend on whether anyone can “understand” the goings-on in a

machine or a head in the sense of apprehending them. The only sense in which these

goings-on are to be understood is the quite different sense o f scienti!c explanation .

It is no accident that the same pernicious equivocation has bedeviled longstanding

disputes in the social sciences b etween subjectivist advocates of verstehen as a

method and “positivist” advocates of erkla¨ ren (Slezak, 1990; Winch, 1957). How-

ever, understanding qua participant is not the same as understanding qua scientist

[10]. Undeniably, Searle’s (1980 ) Chinese Room demonstrates that computational

symbols are meaningless in the former sense, but this is no more problematic than

the meaninglessness, in this sense, of action potentials or synapti c activations.

Predictably, the problem may be seen arisin g in debates over connectionist systems

where, in this case, it is the alleged absence of “explicit” symbols in dispute.

THE TR IPARTITE MODEL OF REPRESENTATION

259

However, the criterion is taken to be whether symbols may be “directly read off” or

“immediately grasped” (Ramsey et al., 1991). The obvious question is: by whom?

Searle’s conundrum is remarkably evoked by Glanvill’s response in 1661 to

Descartes’ own version of a coding or information processing theory of perception.

“But how is it, and by what Art doth the soul read that such an image or stroke in

matter … signi !es such an object? Did we learn such an Alphabet in our Embryo-

state?” (quoted in Yolton, 1984, p. 28). Echoing Searle, Glanvill suggests that the

“motio ns of the ! laments of nerves” learn the quality of objects by analogy with the

wa y in which a person learns to understand a language, for otherwise “the soul

would be like an infant who hears sounds or sees lips move but has no understanding

of what the sounds or movements signify, or like an illiterate pe rson who sees letters

but “knows not w hat they mean” (1984, p. 28). It is signi!cant that, unknowingly,

Yolton also evokes Searle in the Chinese Room when discussing Locke’s conception

in terms of a “perspective box” or camera obscura. Yolton asks “Was there some

temptation to think of our awareness being like the face at the perspective box

scanning the images on the wall of the box?” (1984, p. 127).

Arnauld’s act-theory of direct realism

Arnauld proposed a “direct perception” account against Malebranche’s indirect,

object-mediated theory. For Arnauld, combating the tripartite view means that ideas

are not dis tinct entities but just those very activities of the mind which are essentially

representative p er se.

Since God desired our mind should know bodies, and that bodies should

be known by ou r mind, it was undoubtedly simpler for him to render our

mind capable of knowing bodies immediately, that is, without representa-

tive entities distinct from perceptions … and bodies capable of being

known immediately by our mind, rather than leaving the soul powerless to

see them otherwise than by means of certain representative entities. (VFI,

222–3, quoted in Nadler, 1989, p. 97)

Arnauld insists that enclosing the mind in a “palace of ideas” as Berkeley was

to do is an ab surd conclusion to draw from the representative t heory. Speci!cally,

as we have seen, Arnauld diagnoses the absurdity as due to a mistake o r false analogy

between “being present to the mind” in the sense of having ideas, thinking or

perceiving, on the one hand, and being present to the eyes in seeing. Of course, this

is the assimilation of internal and external representations we have seen. That is,

“seeing” with the mind or sou l is confused wit h seeing with the eyes or body.

Arnauld argues that philosophers have t ried to explain how we think or p erceive with

the mind—mental vision or la vue spirituelle—by analogy with optical vision or true

seeing [11] with the eye—la vue corporelle. This is the same insight expressed in the

epigraph from Galileo. As Nadler points out, Arnauld insists that the problem with

this analogy is that it rests on false assumptions and “One must not base one’s

reasoning about the mental act of perception or observations on, or beliefs about,

the physiological processes which constitute bodily seeing. ”

260

SLEZAK

Nadler (1989, p. 93) construes Arnauld’s distinction between mental

seeing and bodily vision as arising from an adherence to a strict Cartesian

dualism. However, Nadler appears to be making a similar mistake to the

on e we have already noted by Dennett in his diagnosis of the “Cartesian

Theater.” Arnauld, like Descartes, avoids the problems inherent in represent-

ative ideas qua intervening, apprehended e ntities because he has a better

theory of perceptual an d intellectual activity—namely, in terms o f mental

processes which are themselves inherently referential . This is to avoid the

necessity of any central observing homunculus and the Theater by avoiding

a conception of representative entities which require an intelligent perceiver

to contemplate them. Contrary to Nadler, this conception is entirely indepen-

dent of dualism “which rules out any such analogies” between mental and

corporeal vision. Arnauld’s view is more subtle. The analogy is ruled out not

because of any dualism but becaus e Arnauld conceives mental activity as

itself essentially representative and thereby dispenses with ideas as surrogat e

objects to be observed by the mind’ s eye: “… I do not see any need for this

alleged ‘representative entity’ in order to know any object, be it present or absent”

(VFI, 221, quoted in Nadler, 1989, p. 96).

Not much of a revolution?

By contrast with the usual hype, Chomsky recently expressed skepticism regard-

ing th e radical novelty of the so-called Cognitive Revolution saying “it wasn’t all

that muc h of a revolution in my opinion” (1996, p. 1). Chomsky suggests that

the same convergence of disciplinary interests had taken place in the 17th century

in what he calls “ ‘the !rst cognitive revolution,’ perhaps the only real one” (p. 1).

Chomsky (1966) began his Cartesian linguistics by quoting Whitehead, who said that

the recent history of intellectual life may be accurately described as “living upon

the accumulated capital o f ideas provided … by the genius of the seventeenth

century.” Chomsky was concerned to show that a return to classical concerns and

appreciation of their parallels with contemporary developments is valuable in help-

ing to advance the study of language [12]. I hav e been suggesting that, with some

interesting differences, Chomsky’s point ho lds for recent speculation about the min d

outside of linguistics as well. Indeed, Nadler (1992, p. 73) notes th at it is both

“strange an d no t a little embarrassing” that the Malebranche–Arnauld debate should

remain largely ignored or misunderstood by philosophers and historians alike. The

same goes for cognitive scientists. Thus, Chomsky’s example may be usefully

followed in relation to these different issues such as the hotly contested represent-

ational theories of the mind. However, the differences f rom the case of linguistics are

revealing. Though equally neglected in cognitive science, not only the good ideas of

the 17th century are being rediscovered: We are not only reinventing the same

theories and reliving the same debates, but also rehearsin g the same notorious

mistakes.

THE TR IPARTITE MODEL OF REPRESENTATION

261

Imagery: the pictorial theory

The “I magery Debate” is perhaps the most remarkable modern duplication of 17th

century cont roversies. In this re-enactment, among the dramatis personae Pylyshyn

plays Arnauld against Kosslyn’s Malebranche. Kosslyn (1994) c laims to have

resolved the debate in favor of his “pictorial” theory, but there remain grounds for

skepticism. The mise-en-sce`ne is faithful even to the extent of the acrimony of the

disputes. More importantly, the central error identi!ed by Arnauld of ascribing

corporeal properties to mental ones is exactly the one charged by Pylyshyn (1973,

1978, 1981, in press) against Kosslyn.

Despite its computational and neuroscience trappings, Kosslyn’s (1994) picto-

rial account of image ry takes mental images to represent by virtue of a relation of

resemblance to their objects and by virtue of actually having spatial properties which

they r epresent. Furthermore, “depictive” representations in a “visual buffer” are

taken to have the spec i!c function of permitting a re-inspection of images by the

higher visual apparatus. There is said to be an “equivalence” between imagery and

perception according to which the “higher” cognitive processing apparatus for visual

perception is simply applied to an alternative input other than the retina—namely,

the visual buffer. Thus, on the pictorial account, a mental image is conceived to be

a “surrogate percept” (Pinker & Finke, 1980). In this way, an image may be

“repro cessed as if it were perceptual input … thereby accomplishing the purposes of

imagery that paralle l those of perception” (Kosslyn 1987, p. 155). Of course this is

an implicit endorsement of the tripartite conception of mental representation in

imagery.

Dennett (1991) takes this pictorial theory of imagery to be a paradigm example

of the “Theater” misconception and, not surprisingly, t his “quasi-perceptual” model

has been repeatedly charged with the error of importing an “homunculus.” The

charge is vigorously rejected on the grounds that “the theory is realized in a

computer program” (Kosslyn et al., 1979, p. 574), but undischarged homunculi can

lurk in computational models just as easily as in traditional discursive theories (see

Slezak, 1995, in press).

Thus, Kosslyn et al. (1989) offer a diagram of the visual imagery system which

is a profusion of interconnected boxes and arrows. The box labeled “visual buffer”

contains another box labeled “attention window” which is lef t unexplained. This

box is, in fact, the observer in the “theater” which is the source of the traditional

problem. The elaborate diagram is reducible to the same tripartite schema we have

seen in Malebranche. Signi!cantly, following D escartes, Arnauld explicitly p ointed

to the seductive error of taking pictures as an appropriate model of mental represen-

tation (Arnauld, 1683/1990, p. 67), and he cites the camera obscura as an erroneous

model for imagery.

In a revealing misunderst anding, Kosslyn (1980, p. 30) has charged alternative

propositional or “tac it knowledge” theories (Pylyshyn, 1973) as being “no imagery”

accounts, but denying pictorial images is not to deny imagery per se. Rather, in an

Arnauldian spirit, the denial of pictorial format for representations is actually to

deny the problematic homunculus an d its pseudo-explanation. In effect, as we see,

262

SLEZAK

the debate between Malebranche and Arnauld is being replayed throughout the

history of speculation o f the mind and contemporary cognitive science. Thus, one

need not go back as far as the 17th and 18th centuries to discover the same

concerns. F.C. Bartlett’s (1932) theory of schemata in his book Remembering, was a

reaction to theories of “!xed, lifeless and fragmentary traces” or images which are

merely “reduplicative,” capable only of being re-excited. Reminiscent of Arnauld’s

rejection of “super"uous entities” and his view “that idea and perception are the same

thing,” Bartlett wishes to substitute a cognitive process for objects which are

pictorial or “reduplicative traces” (1932, p. 215) [13].

Phenomenological fallacy

Kosslyn claims to have clinched the debate about imagery by app ealing to the

!ndings of neurophysiology and neuroanatomy [14]. Topographically organized

regions of cortex or “retinotopic mapping” are said to “support depictive represen-

tations,” that is, pictures in some sense. Thus, for example, a monkey may be given

a visual stimulus like a dartboard to look at. If the brain tissue is treated in a certain

way, it can be shown to have a likeness of the dartboard “etched” on the cortex. The

result was anticipated and p erfectly understood by one psychologist 30 years before:

At some point the organism must do more than create duplicates … The

need for something beyond and quite different from copying is not widely

understood. Suppose someone were to coat the occipital lobes of the brain

with a special photographic emulsion which, when developed, yielded a

reasonable copy o f a current visual stimulus In many quarters this woul d be

regarded as a triumph in the physiology of vision. Yet nothing could be

more disastrous … (Skinner, 1963, p. 285)

Skinner was acutely sensitive to the source of homunculi pseudo-explanations even

if his behaviorist remedy is no longer attractive.

Kosslyn’s TV screen metaphor reveals the link between seemingly unrelated

problems in cognitiv e science. For example, in the classic statement of materialism,

U.T. Place (1956) argued that the implausibility and rejection of materialism as a

solution to the mind–body problem is based on the qualitative features of subjective

experience. Although these features have recently been supposed to constitute the

“hard” problem of consciousness (Chalmers, 1996), Place suggested that they are

the source of the “p henomenological fallacy.” Anticipating Dennett (1991), Place

wrote, this is “the mistake of supposing that when the subject describes his experi-

ence, how things look, sound, smell, taste, or feel to him, he is describing the literal

properties of objects and events on a particular s ort of internal cinema or television

screen.”

Thinking in language

On the face of it, the persistent doctrine that we think “in” language is not obviously

connected with the others we have considered such as pictorial imagery. Neverthe-

THE TR IPARTITE MODEL OF REPRESENTATION

263

less, these two theories are variat ions on the same theme. Symptomatic is the fact

that both depend on a deep intuitive, introsp ective appeal. Just as w e seem to be

looking at pictures when w e imagine visually, so we appear to talk to ourselves when

we think. Indeed, Carruthers (1996), who seeks to revive what he acknowledges to

be an unfashionable doctrine, explicitly bases his argument against Fodor’s Lan-

guage of thought on such evidence of introspection. This is the evidence that we

sometimes !nd ourselves in a silent monologue, talking to ourselves in our natural

language, sotto voce, as it were.

However, in a neglected article, Ryle (1968) suggested that the very idea that

we might think “in” language is unintelligible, and the undeniable experience of

talking to ourselves cannot supp ort any claim about the vehicles of thought. It is

signi!cant that Ryle mentions en passant among the equally problematical cases, that

in which we claim to see things in our “mind’s eye”—taken to involve mental

pictures of some kind. Ryle’s compariso n and his warning is unwittingly con!rmed

by Carruthers (1996, 1998), who exp licitly invokes Kosslyn’s pictorial account of

imagery as support for his own analogous theory. In doing so, however, Carruthers

only brings into relief the notorious dif!culties o f his own model which relies on a

representational format—sentences of natural language—which is, like pictures,

paradigmatically the kind requiring an external intelligent observer (see Slezak,

2002).

Still more processing: triadic or dyadic?

As we have seen, the traditional dif!culty, rediscovered in various forms today, arises

from the problematic three-part relation between world, ideas and consciousness.

Cognition without these basic features seems inconceivable, and yet they le ad to

seemingly intractable dif!culties. In its essentials, the Malebranche–Locke account

may be captured in the following schematic diagram:

External World

Ideas

Consciousness

The commonality between this schema and modern ones is clearly revealed in a

diagram of Ulric Neisser (1976 ). In its essentials the diagram may be represented as

follows:

Storage Storage St o rage

¯! ¯! ¯!

External

Retinal

Processing

M ore processing

Still m o re processing

Consciousness

W orld image

Neisser’s schema obviously abstracts from the details of any speci!c account, but

despite its mildly whimsical character, it purports to be a serious generic sketch of

information processing theories. For example, the welter of boxes and arrows in

Kosslyn’s diagram reduces to Neisser’s picture which has the virtue of not disguising

its essential commitment to the problematic third element “consciousness.”

One early attempt to avoid the dif!culties inherent in this account is in the

reactions of Locke’s critic John Sergeant:

264

SLEZAK

He never found a satisfactory answer to the question of the nature of ideas,

but he was convinced that t hey functioned to deny the mind direct access

to things by restricting it to some kind of third entity. Cognition was thus

made to consist in a triadic relation involving the knowing mind, the object

or referent, and the ideas by means of which the mind came to know

things. Sergeant wished to reduce the process of knowing to a dyadic

relation consisting only of the knowing mind and the object known.

(quoted in Yolton, 1956/1993, p. 103)

No ideas?

However, given a p roblematic three-part relation, there are only a few ways to ge t

a dyadic relation. First, eliminating the middle term, ideas or representations,

permits two construals. On one view, which Yolton describes as “wildly impossible”

but which has nonetheless been explicitly held, an object is itself somehow literally

present to, or in, the mind. Thi s is t he view we saw parodied by Malebranche saying

that the soul does not stroll in the heavens among the stars.

External World

Consciousness

No external world?

Of course, a dyadic relation can also be obtained by dropping one of the other relata

instead of the middle one. Thus, we might eliminate the external world to get

Berkeley’ s idealism.

Ideas

Consciousness

Berkeley’ s strategy has not been popular recently among cognitive scientists, though

it is precisely the paradox of Fodor’s “methodological solipsism.”

No consciousness?

Of course, we have another choice besides getting rid of the world or getting rid of

the representations. We can get rid of consciousness! Securing a direct connection

between cognition and the world can also be achieved by dispensing with the agent

[15].

External World

Ideas

Despite its counter-intuitiveness, in some respects this option is preferable to the

others. Of c ourse, the agent as an element in models of perception is the locus of the

potential homunculus. The paradoxical rejection of consciousness is equivalent to

the !rst of these three options, being the same as rejecting representations when

these are conceived in ce rtain ways. Therefore, expressed less paradoxically, the

Gibsonian or “situated” case against representations may be best understood, not as

an outright rejection of an internal mental medium as such, but rather as the

THE TR IPARTITE MODEL OF REPRESENTATION

265

Arnauldian rejection o f a certain particular conception of representations which are

explanatorily question-begging.

Conclusion

Thirty years ago in his book Psychological explanation, Fodor (1968 , p. vii) remarked:

“I think many philosophers secretly harbor the view that there is something dee ply

(i.e. conceptually) wrong with psychology, but that a philosopher with a little

training in the techniques of linguistic analysis and a free afternoon could straighten

it out.” Today , the suspicion of deep conceptual problems at the heart of cognitive

science is perhaps more clearly seen to be justi!ed, although Fodor’s joke was

intended to re"ect as much upon philosophy as upon psychology. Notoriously, deep

conceptual problems at the heart of philosophy have been no more dispelled than

those in psychology. I have been suggesting that by adopting a broader perspective

we may see why the sorry fortunes of the two disciplines have been inextricably

linked.

Acknowledgements

This paper has bene!ted greatly by the most helpful comments and criticisms of the

editor, W. Bechtel, and two anonymous referees of Philosophical Psychology. Versions

of the material have be en presented at the Department of Philosophy, University of

Sydney, the Sixteenth Annual Me eting of the J apanese Cognitive Science Society,

Tokyo, August 1999, the Annual Meeting of the Australasian Association for

History, Philosophy & Social Studies of Science, Melbourne, June 2001, and the

Twenty Third Annual Conference of the Cognitive Science Society, Edinburgh,

August 2001. I am particularly grateful for the comments of Stephen Gaukroger,

Ron Giere, Pat Langley, Robert Nola, Zenon Pylyshyn and John Sutton.

Notes

[1] Kosslyn writes with evident annoyance: “Once and for all, the ‘homunculus problem’ is simply

not a problem. We thought this would be obviou s given that the theory is realized in a computer

program, but it seems necessary to address this complaint again ” (Kosslyn et al., 1979, p. 574).

[2] The scholarly, exegetical niceties need not concern us here. However, Nadler (1992, p. 8) points

out that th e standard reading presented here is mistaken and even a caricature, though it has been

almost universally held among commentators including Arnauld, Locke, Leibniz, Berkeley and

Reid.

[3] The “situated” critique confuses conscious calculation with the “sub-personal” use of symbols

and their purely causal, functional role in cognition (see Slezak, 1999). As Pasnau says: “Surely,

any modern direct-realist theory of perception will allow causal interme diaries between object and

percipient: no one would dream of denying the title of direct realism to a theory of percepti on

merely because it tole rates causal intermediaries” (Pasnau, 1997, p. 300).

[4] Above all, Moreau expresses irritation at the view which has been widely held, especially among

“nos contemporains anglo-saxons,” that this affair is at base nothing other than one of opposing

temperaments—“Arnauld-la-teigne ” against “Malebranche-le-grognon”—that is, Arnauld-the-nuis-

266

SLEZAK

ance against Malebranche-the-grouch (1999, p. 16). In fact, numerous great !gures of the time

became embroiled in the debate, such as Leibniz who followed it closely (see Nadler, 1989, p. 5).

[5] In his earlier work on Locke and the way of ideas, Yolton (1956/1993) gives a longer passage from

the same author which is worth repeating here as testimony to the curious persistence of the

puzzle of ideas, albeit in different guises: “… like Men blundering in the dark, they feel after them

to !nd them; some catch at them under one Appearance, some under another; some make them

to be Material, others Spiritual; some will have them to be Ef!uvia, from the Bodies they

Represent, others Totally Distinct Essences; some hold them to be Modes, others Substances; some

assert them All to be Innate; others None: So that one would think there must needs be a very great

Intricacy in that which has given Rise, not only to such a Variety but also such a Contradiction of

Opinions” (quoted in Yolton, 1956/1993, p. 96).

[6] I am grateful to an anonymous referee for criticisms which have helped me to try clarifying this

argument.

[7] This moral seems to extend in an obvious way to the broader concern with knowledge in

philosophy of science. Although it cannot be pursued here, another deep parallel may be seen in

the more or less distinct literatures of perceptual realism and scienti!c realism. Here too the

problem of truth and error arises for realism in the form of the so-called “pessimistic meta-induc-

tion” from history—analogous to the Argume nt from Illusion. It is no accident that Mach’s

instrumentalism was harshly criticized in Lenin’s (1927) Materialism and empirio-criticism as

betraying science by deserting to a Berkeleyan idealism. See also Popper (1963a,b) on Berkeley

as precursor to Mach and Einstein. See also Zahar (1981).

[8] Among the few explicit attempts to articulate the relevant distinctions here, see Cummins (1996,

p. 87) on the distinction between “meaning” and “meaningfor”—the latter described by Cum-

mins as “a three-place relation between a representation, a concept, and a cognitive system.” See

discussion on p. 130.

[9] Conversely, it may be that researchers’ actual theories or programs are exemplary and free of

problems, but only their “meta-theoretical” analyses suffer from the fatal "aws. I am grateful to

Pat Langley and Ron Giere for this point.

[10] Chomsk y (1962) has drawn attention to the way in which traditional grammars produce an

illusion of explanatory completeness while, in fact, they have serious limitations fro m a scienti!c,

explanatory point of view. The apparent success

The tripartite model of representation

Abstract