New Perspectives on Concepts (Grazer Philosophische Studien: Internationale Zeitschrift Fur Analytische Philosophie)

New Perspectives On Concepts

Grazer Philosophische Studien International Journal for Analytic Philosophy

fOuNded by rudolf haller edited by Johannes l. brandl Marian david Maria e. reicher leopold stubenberg

VOL 81 - 2010

Amsterdam - New York, NY 2010

New Perspectives On Concepts

Edited by

Julia Langkau and Christian Nimtz

Die Herausgabe der GPS erfolgt mit Unterstützung des Instituts für Philosophie der Universität Graz, der Forschungsstelle für Österreichische Philosophie, Graz, und wird von folgenden Institutionen gefördert: Bundesministerium für Bildung, Wissenschaft und Kultur, Wien Abteilung für Wissenschaft und Forschung des Amtes der Steiermärkischen Landesregierung, Graz Kulturreferat der Stadt Graz

The paper on which this book is printed meets the requirements of “ISO 9706:1994, Information and documentation - Paper for documents Requirements for permanence”. layout: thomas binder, graz ISBN: 978-90-420-3018-3 E-book ISBN: 978-90-420-3019-0 ISSN: 0165-9227 E-ISSN: 1875-6735 © Editions Rodopi B.V., Amsterdam - New York, NY 2010 Printed in The Netherlands

TABLE OF CONTENTS

Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

v

Christian NIMTZ & Julia LANGKAU: Concepts in Philosophy— A Rough Geography. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1

Robert BRANDOM: Conceptual Content and Discursive Practice . . .

13

José Luis BERMÚDEZ: Two Arguments for the Language-Dependence of Thought . . . . . . . . . . . . . . . . . . . . . . . . . . . .

37

Martine NIDA-RÜMELIN: Thinking without Language. A Phenomenological Argument for its Possibility and Existence . . . . . . . .

55

Hannes RAKOCZY: From Thought to Language to Thought: Towards a Dialectical Picture of the Development of Thinking and Speaking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

77

Anthony KENNY: Concepts, Brains, and Behaviour . . . . . . . . . .

105

Hans-Johann GLOCK: Concepts, Abilities, and Propositions . . . . .

115

Sebastian RÖDL: The Self-Conscious Power of Sensory Knowledge . . .

135

Katia SAPORITI: In Search of Concepts . . . . . . . . . . . . . . . . . .

153

Frank JACKSON: Conceptual Analysis for Representationalists . . . . .

173

Christian NIMTZ: Philosophical Thought Experiments as Exercises in Conceptual Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

189

Finn SPICER: Kripke and the Neo-Descriptivist . . . . . . . . . . . . . . .

215

Mark TEXTOR: Frege on Conceptual and Propositional Analysis . . .

235

ACKNOWLEDGMENTS The contributions to his volume have grown out of the invited papers to three international workshops on concepts held at Zurich University (UZH): Concepts—No Language, No Thought? (March 2007), Concepts and Abilities (May 2008), and Concepts and Their Analysis (September 2008). All three workshops have been organized and for the most part funded by the then newly established chair in theoretical philosophy held by Hanjo Glock. We would like to thank Prof. Glock for his encouragement and for making it all possible, the Swiss National Foundation for the generous funding it provided for the third of our workshops, Prof. Hampe of the ETH Zurich and the Competence Centre “History of Knowledge” for their hospitality, and everyone else involved for making the workshops the success we feel they have been. Julia LANGKAU and Christian NIMTZ St Andrews / Erlangen, December 2009

Grazer Philosophische Studien 81 (2010), 1–11.

CONCEPTS IN PHILOSOPHY—A ROUGH GEOGRAPHY Christian NIMTZ & Julia LANGKAU University of Erlangen-Nürnberg / University of St Andrews Much recent work on concepts has been inspired by, and is developed within, the bounds of the representational theory of the mind often taken for granted by philosophers of mind, cognitive scientists, and psychologists alike (see e.g. Margolis and Laurence 1999; Stich and Warfield 1994). The contributions to this volume take a more encompassing perspective on the issue of concepts. Rather than modelling details of our representational architecture in line with the dominant paradigm, they explore three traditional issues concerning concepts. Inquiring into how language and the mind are interrelated, Brandom, Bermúdez, Nida-Rümelin, and Racokzy debate Is mastery of a language necessary for thought? Pondering whether drawing on concepts to explain our thinking requires us to adopt the representational paradigm of the mind in the first place, Kenny, Glock, and Saporiti are concerned with the question Do concepts reduce to abilities? Finally, in order to assess the prospects for philosophical reliance on conceptual analysis, Jackson, Nimtz, Spicer, and Textor discuss Is the analysis of concepts a viable means to ascertain truths from the proverbial armchair? Needless to say, there is no consensus to be had on either issue. This introductory essay explores the backdrop to the debate our authors engage in. We will provide a rough geography of key ideas and issues shaping the overall debate on concepts within contemporary philosophy. We will proceed in two steps. In a first step, we will present and discuss

key ideas shaping the assumptions and expectations concerning concepts philosophers harbour (§ 1). Taken together, they form what we want to think of as the conventional picture of concepts. In a second step, we will focus on recent developments and divisive fundamental issues that have brought about tectonic changes in the philosophical views on concepts (§ 2). These explain why the conventional picture has basically gone out of fashion and why the philosophical debate on concepts in general appears heterogeneous, and feels fragmented. 1. Concepts in philosophy: some traditional ideas In line with well-known instructive attempts to characterize the nature and role of concepts (see Fodor 1998, ch. 1, Margolis and Laurence 1999b, and especially Burge 1993), we present a list of popular ideas concerning concepts. Although items from our list are often presented as platitudes, we hold that it neither catalogues mere truisms, nor states a general consensus. The list rather registers key ideas as to what concept are, and what concepts do, that figure prominently in the contemporary debate. Apparently, few contemporary philosophers subscribe to the conventional picture of concepts in its entirety. As we shall see in § 2, many philosophers embrace fundamental views about thought, language, and content which lead them to re-interpret or reject core ideas of the conventional picture. Talk of ‘concepts’ looms large in the philosophy of mind and the philosophy of language, and it plays an important role in epistemology. These disciplines, however, focus on different aspects when debating concepts. Philosophers of mind primarily invoke concepts to account for the distinct features of our thinking. Here it is widely held that (1) Concepts are sub-components of thoughts.

On this assumption (see e.g. Rey 1998), Alfie’s belief that cats are more dangerous than groundhogs and his belief that Joe’s only pet is a cat share a sub-component—viz. the concept CAT. This puts Alfie in a position to infer that Joe’s only pet is more dangerous than a groundhog by mere logic, and it allows him to infer that groundhogs are not the most dangerous animals by analytical inference. (The latter requires the popular though contentious assumption that GROUNDHOG is a complex concept containing, as it were, the concept ANIMAL.) It also explains

2

why Alfie’s ability to think the thoughts mentioned puts him in a position to grasp systematically related thoughts such as the thought that groundhogs are more dangerous than cats. It finally explains why Alfie can entertain thoughts he has never thought before, and why acquiring a single new concept allows a thinker to entertain a whole new range of thoughts. In sum, the idea that there are concepts as characterized by (1) explains why thoughts typically come related, why thoughts appear to be structured, and why our thinking is systematic, creative, and expansive. It also provides a first characterization of what concept possession amounts to: a thinker possesses a concept C if and only if she can think thoughts containing C. In addition to taking concepts to be sub-components of thoughts, philosophers of mind standardly embrace an idea about what concepts contribute to the thoughts they are sub-components of. They think that (2) Concepts are representational and determine intensions. Although there is substantial debate on how we are to understand (2) precisely, it seems safe to say that if Alfie possesses the concept CAT, he can think of objects in a specific way, viz. as cats rather than, say, as groundhogs. A concept hence constitutes a particular way of thinking of things. Any such way of thinking determines, for any possible world, an extension of items—objects, properties, relations, events—thought about. But since ways of thinking are finer individuated than the items thought about, one may think of the same items in different ways. Taking concepts to be representational thus allows one to account for the opacity of intentional states as well as the differences in intentional actions resulting from intentional states involving the same extensions. This explains why thinking that cats are furry and thinking that the only animals Joe actually likes are furry amount to different things, even though the only animals Joe actually likes are cats. It also explains why Alfie gets up to bang on his neighbour’s door on learning that Joe is mistreating his cat, but doesn’t do so on learning that Joe is mistreating his only pet, even though Joe’s only pet is a cat. The representational nature of concepts provides a second characterization of what concept-possession amounts to: a thinker possesses a concept C if and only if she can think of items in the particular way constitutive of C. Being representational, concepts determine intensions. But to determine an intension is to mark off specific items—traditionally captured as 3

those ‘falling under’ the concept—from those that do not. Taking concepts to be representational thus fits nicely with holding that (3) Concepts guide categorization. Taking concepts to guide categorization explains why it is so useful to possess them (see Carey 2009, Murphy 2002). It also allows us to see the point of enriching our conceptual repertoire. Suppose that although Alfie has had concepts such as CAT, BIRD and GROUNDHOG for some time, he only recently acquired the concept FUR. He can now distinguish in thought furry creatures such as cats and groundhogs from non-furry creatures such as birds. This allows him to attune his actions to specific features of his surroundings cutting across the categories he used to work with. For instance, he can now decide to shun all creatures with fur. This close connection between concepts and categorization provides a third characterization of what concept-possession amounts to: a thinker possesses a concept C if and only if she can distinguish in her thought between items falling under C and those that do not. Once we turn to the philosophy of language, the focus changes from accounting for distinct features of thought to accounting for manifest properties of meaningful expressions. A common idea (see Hanfling 2000, ch. 4) within the theory of meaning is that (4) Concepts are the meanings of (general) sub-sentential expressions. There is no consensus as to which sub-sentential expressions express concepts. It is often held that general terms such as ‘cat’ or ‘stockbroker’ do, whereas singular terms such as ‘Zurich’ do not. Concentrating on the former, (4) assures us that we can sum up the semantic properties of the English expression ‘cat’ by noting that it expresses the same concept as the French expression ‘chat’, viz. the concept CAT. The fact that ‘cat’ expresses CAT explains why ‘cat’ applies to cats (they are the items falling under the concept CAT), why ‘cat’ is often understood to analytically entail ‘animal’ (CAT is often understood to contain, and hence to analytically entail, ANIMAL), and what ‘cat’ contributes to the meaning of sentences it occurs in (it contributes the concept CAT). Theorists embracing (4) invoke concepts to explain what expressions mean. Concepts moreover figure prominently in accounts of understanding. It is habitually surmised that 4

(5) Concepts are pivotal to the understanding of language. On (5), Alfie understands the term ‘creature’ just in case he knows that ‘creature’ expresses the concept CREATURE. This in turn requires Alfie to master that concept. Understanding the general terms of a simple language thus requires mastering specific concepts. This fits the common assumption that proficiency in the employment of general terms is a criterion for the mastery of the relevant concepts: if Alfie is proficient with the English term ‘cat’, we can rest assured that he masters the concept CAT. By the same token, acquiring new general terms brings with it the acquisition of the concepts they express. According to (5), then, enriching our vocabulary of general terms amounts to broadening the range of concepts we possess. A thinker can extend the range of her thoughts by learning how to master new general terms. For example, becoming proficient with the legal term ‘plaintiff’ or the astronomical term ‘meteor’ is a way to acquire the concepts PLAINTIFF or METEOR, respectively. It is often assumed that this is not a piecemeal matter. Just as understanding our term ‘cat’ is said to require understanding ‘animal’ and ‘living’ as well, the respective concepts are held to be acquired as a bundle rather than one by one. More importantly still, it appears that one can acquire sophisticated concepts only via linguistic proficiency. If this is true, then the range of our thoughts is bounded by the language we speak, which might well lead one to wonder whether having a language might not be necessary for thought in the first place. This suggests a controversial idea as to what concept-possession might involve: a thinker possesses a concept C only if she understands sentences comprising a term expressing C. Turning finally to epistemology, we find that concepts figure in two ways in epistemic endeavours. If we are to trust (1) to (5), concepts are representational items guiding categorization in thought and language. But describing and explaining features of the world requires that we represent these features in language and thought in the first place. Hence, any theorizing about the world will rely on concepts as means of representation. But concepts have also been considered suitable objects of analysis. Although some think of it as dated, the view that doing philosophy involves analysing concepts is still held and defended (see Jackson 1998). Its champions obviously rely on the idea that (6) Thinkers possessing a concept can ascertain its constituent structure, or at least its intension, by mere reflection. 5

Advocates of conceptual analysis often presume that concepts such as PERSON or JUSTICE have constituent structure, and thus are inherently complex. But if conceptual analysis merely aims at specifying a term or concept’s intension (see Jackson, this volume), there is no need to assume this. What advocates of conceptual analysis need to presuppose is that mastery of a concept puts a thinker in a position to establish by mere reflection how or at least what the concept represents. 2. Concepts in philosophy—divisive fundamental issues Combining ideas (1) to (6) yields what we think of as the conventional picture of concepts (compare Burge 1993). On this picture, concepts are sub-components of thoughts as well as meanings of terms. Equating concepts with meanings locates concepts and thoughts on the level of contents. Understood thus, concepts such as CAT or JUSTICE are sub-components of thoughts by being sub-components of thought contents. They are mental representations in the sense of being part of what is thought. This amounts to what we think of as the content-view of concepts (see e.g. Evans 1982 or Glock, this volume). The claim that thoughts and their parts are contents may seem tautological. But a major tradition in contemporary theorizing about concepts disagrees. Many philosophers and cognitive scientists employ talk of ‘concepts’ and ‘thoughts’ to refer to mental representations in the sense of structured mental particulars, often described to be words and sentences of a language of thought, that are the bearers or vehicles of contents, rather than the contents themselves. That is to say, they embrace what we call the vehicle-view of concepts (see e.g. Fodor 1998; 2008 or Margolis and Laurence 1999b). The antagonism between the content-view and the vehicle-view has brought about a systematic ambiguity in the terms ‘concept’ and ‘thought’ as used in philosophy. As Margolis and Laurence (2006, § 1.4) rightly note, proponents of the two views do not just squabble about terminology. Their divergent terminologies are rather a symptom of a substantial and deep disagreement about the right approach to the mind that manifests itself in the respective stances advocates of the two views take on a divisive fundamental issue. We call it the issue of the level of explanation. Should we account for intentional mental states primarily in terms of contents of thoughts, or in terms of vehicles of thinking? 6

Traditionally, differences in belief states—or more generally: of intentional states involving the same kind of propositional attitude—have been explained in terms of differences in the contents involved. Suppose that Alfie believes that groundhogs are endearing, and he believes that woodchucks are endearing. An explanation of the relevant differences in terms of contents ascribes to Alfie two beliefs with different contents—a content containing the concept WOODCHUCK and a content containing the concept GROUNDHOG. Commitment to a language of thought affords a different explanation. Its advocates can hold that the differences between Alfies groundhog-belief and his woodchuck-belief results from a difference in the vehicles of his thought: Alfie employs two different mental symbols to represent the same kind of animal. From here it is only a small step to identifying the concepts WOODCHUCK and GROUNDHOG with mental symbols bearing contents, rather than with the contents they bear. By the same token, advocates of a language of thought commonly identify a person’s thoughts with sentence-like mental symbols bearing full propositional contents, rather than with the full propositional contents these symbols bear. The highlighted difference in explanatory approach yields two points of disagreement between advocates of the content-view and champions of the vehicle-view. First, there is disagreement about the explanatory reach of a theory of concepts. Champions of the content-view are free to embrace the conventional picture of concepts in its entirety. They even are likely to regard all of (1) to (6) as platitudes defining the topic of concepts in the first place. By consequence, advocates of the content-view are likely to expect a theory of concepts to explain manifest properties of both thought and language. Advocates of the vehicle-view will also embrace (1) to (3), reading ‘concept’ and ‘thought’ as designating content-bearing mental vehicles (see e.g. Fodor 1998, Margolis and Laurence 1999b, 5–8). But since (4) straightforwardly identifies concepts with meanings and hence with contents, they cannot accept this idea. In consequence, advocates of the vehicle-view are likely to have less far-reaching expectations of a theory of concepts. Save for a friendly nod towards a Gricean programme of explaining meaningful speech in terms of intentional states, they will typically confine their explanatory aspirations to thought. Secondly, there is disagreement about the idea of a medium of mental representation (see e.g. Stich and Warfield 1994). Advocates of the vehicleview hold that thinking consists in the manipulation of re-combinable mental particulars having syntactic structure and bearing semantic con7

tents. Some of those who disagree grant that there is a medium of mental representation, but think that proponents of the vehicle-view have got its structure wrong. For example, they might hold that mental representation is map-like and hence holistic, rather than language-like and hence discrete (see Braddon-Mitchell and Jackson 2006, 177–184). Others dismiss the idea of a medium of thinking wholesale. Those who do so are prone to explain thinking in terms of abilities rather than in terms of contentbearing vehicles (see e.g. Evans 1982: § 4.3 who follows the tradition of Geach 1957). By the same token, critics of the idea that there is a medium of thinking are prone to account not only for our cognitive feats, but also for linguistic proficiencies we describe in terms of ‘concepts’ by appeal to abilities (see Kenny and Glock, this volume). There is a second divisive fundamental issue shaping much of the current debate on concepts. The issue we have in mind is that of individualism vs. externalism. Should we hold that mental and linguistic contents are determined by intrinsic properties (as the individualist claims), or that they are (at least in many cases) determined by those substances and kinds in the thinker’s environment that she is suitably related to (as the externalist thinks)? Both individualists and externalists actually hold rather more nuanced positions, but this sketch will do for our purposes (see Lau 2008, Mendola 2009, Rowlands 2003 and Segal 2000). Their disagreement is of importance to the topic of concepts. A theorist drawn to externalism is likely to hold specific views about the coarseness of contents. She will also most likely hold specific views about a thinker’s epistemic access to the contents of her thoughts. These views are prone to affect her ideas on the nature and role of concepts, and they are likely to influence her association with the content-view or the vehicle-view of concepts, respectively. Let us begin with the issue of a thinker’s epistemic access to thoughtcontents. An individualist need not grant that a thinker can typically ascertain the constituent structure or at least the intensions of the complex concepts she employs. Still, none of her individualist commitments bars her from doing so. By contrast, the key tenets of externalism appear to commit externalists to holding that a thinker cannot establish the intensions of those of her complex concepts whose contents are determined externally. Suppose that the contents of Alfie’s thoughts he expresses with ‘woodchuck’sentences—his ‘woodchuck’-thoughts, for brevity—are determined by the 8

fact that Alfie is suitably related to the woodchucks rather than the hoary marmots in his environment. Had the animals he is suitably related to been hoary marmots, Alfie’s ‘woodchuck’-thoughts would have had hoarymarmot-contents. On these externalist premises, it is very hard to see how mere reflection could possibly put Alfie in a position to determine the contents of his ‘woodchuck’-thoughts in any informative way. Turning to the issue of coarseness, the property of being a woodchuck just is the property of being a groundhog. So, ‘Alfie believes that Joe owns a woodchuck’ and ‘Alfie believes that Joe owns a groundhog’ ascribe contents to Alfie in which the same property is predicated of the same individual. This could bring one to equate both contents with one and the same Russellian proposition <Joe, owning a woodchuck>. However, the sentences ‘Joe owns a woodchuck’ and ‘Joe owns a groundhog’ differ in cognitive significance. For one can accept the one as true and reject the other as false. This might lead one to equate the first content ascribed to Alfie with the Fregean thought [that Joe owns a woodchuck] and the second content ascribed to Alfie with the different Fregean thought [that Joe owns a groundhog]. One expects externalists to opt for the Russellian view. Suppose you embrace the externalist idea that the contents of our ‘groundhog’- and ‘woodchuck’-thoughts and of our terms ‘groundhog’ and ‘woodchuck’ are fixed by the kinds of creatures we interact with. But there is just one kind of creature out there. Consequently, you seem bound to agree that ‘groundhog’ and ‘woodchuck’ have the same content. This generalizes: given that externalists hold that contents as a rule do not cut finer than the kinds or properties interacted with, they are committed to a Russellian view of contents. By the same token, one expects individualists to harbour sympathies for Fregean thoughts. Since individualists hold that contents are determined by properties intrinsic to the thinker, rather than by the items interacted with, they might well agree that contents are more finely individuated than properties. Advocates of the vehicle-view can honour individualist convictions (see Block 1986). Still, it is more important to note that a commitment to externalism will quite naturally bring a commitment to the vehicle-view in its wake. Just like everyone else, externalists typically agree that concepts constitute particular ways of thinking (as is claimed in (2) above). But ways of thinking are more finely individuated than kinds or properties, for we can think of the same kind or property in different ways. The concepts WOODCHUCK and GROUNDHOG are cases in point. The externalist 9

associates the same content with both concepts. Hence, he needs to look beyond contents to account for their difference. Embracing the idea of a structured medium of mental representation promises a solution, for it allows the externalist to trace the acknowledged differences in thinking to differences in the vehicles involved. Since on this view it is the vehicle that makes the difference between groundhog-thoughts and woodchuckthoughts, it again seems natural to identify concepts with vehicles rather than with contents. As has become clear, anyone committed to externalism will most likely deny the feasibility of conceptual analysis (at least for all of those terms or concepts whose contents she deems to be externally determined), embrace a Russellian understanding of content, and side with the vehicle-view of concepts. Hence, a commitment to externalism is indeed prone to affect a thinker’s ideas on the nature and role of concepts. So both of the two fundamental divisive issues we have identified—the issue of the level of explanation and the issue of individualism vs. externalism—affect if not shape the contemporary debate on concepts. After all, it is the noted ambiguity in ‘concept’ and ‘thought’ that makes the philosophical discussion on concepts appear heterogeneous, and it is the dissolution of the conventional picture of concepts along the lines of the two issues discussed that explains why the debate feels fragmented.

REFERENCES Block, Ned. 1986. “Advertisement for a Semantics for Psychology”. In: Stephen Stich and Ted Warfield 1994, 81–141. Braddon-Mitchell, David and Frank Jackson. 2006. The Philosophy of Mind and Cognition. An Introduction. Oxford: Blackwell. Burge, Tyler. 1993. “Concepts, Definitions, and Meanings”. Metaphilosophy 24, 309–325. Carey, Susan. 2009. The Origin of Concepts. Oxford: Oxford University Press. Evans, Gareth. 1982. The Varieties of Reference. Oxford: Clarendon. Fodor, Jerry L. 1998. Concepts. Where Cognitive Science Went Wrong. Oxford: Clarendon Press. — 2008. LOT 2: The Language of Thought Revisited. Oxford: Oxford University Press. Geach, Peter. 1957. Mental Acts. London: Routledge & Kegan Paul.

10

Glock, Hans-Johann. 2010. “Concepts, Abilities, and Propositions”. In this volume, 115–134. Hanfling, Oswald. 2000. Philosophy and Ordinary Language. The Bent and Genius of our Tongue. London: Routledge. Jackson, Frank. 1998. From Metaphysics to Ethics. A Defense of Conceptual Analysis. Oxford: Oxford University Press. — 2010. “Conceptual Analysis for Representationalists”. In this volume, 175–190. Kenny, Anthony. 2010. “Concepts, Brains, and Behaviour”. In this volume, 105–113. Lau, Joe. 2008. “Externalism About Mental Content”. In: The Stanford Encyclopedia of Philosophy. Margolis, Eric, and Stephen Laurence (eds).1999. Concepts. Core Readings. Cambridge (Mass.): MIT Press. — 1999b. “Concepts and Cognitive Science”. In: Eric Margolis and Stephen Laurence 1999, 3–81. — 2006. “Concepts”. In: The Stanford Encyclopedia of Philosophy. Mendola, Joseph. 2009. Anti-Externalism. Oxford: Oxford University Press. Murphy, Gregory L. 2002. The Big Book of Concepts. Cambridge (Mass.): MIT Press. Nimtz, Christian. 2010. “Philosophical Thought Experiments as Exercises in Conceptual Analysis”. In this volume, 189–214. Rey, Georges. 1998. “Concepts”. In: Edward Craig (ed). 1998. The Routledge Encyclopedia of Philosophy. London: Routledge. Rowlands, Mark. 2003. Externalism: Putting Mind and World Back Together Again. Chesham: Acumen. Segal, Gabriel. 2000. A Slim Book about Narrow Content. Cambridge (Mass.): MIT Press. Stich, Stephen and Ted Warfield (eds.). 1994. Mental Representations. Oxford: Blackwell.

11

Grazer Philosophische Studien 81 (2010), 13–35.

CONCEPTUAL CONTENT AND DISCURSIVE PRACTICE Robert BRANDOM University of Pittsburgh Summary This paper discusses the integrated approach to the semantics and pragmatics of language developed in my Making It Explicit (Brandom 1994). The core claim is that there are six consequential relations among commitments and entitlements that are sufficient for a practice exhibiting them to qualify as discursive, that is, as a practice of giving and asking for reasons, hence as one conferring genuinely conceptual content on the expressions, performances, and statuses that have scorekeeping significances in those practices. I divide the six consequential relations into two groups, the fundamental-semantic and the social-pragmatic, and I characterise the complex interactions between them. The bold and potentially falsifiable overall claim is that any practice that exhibits this full six-fold structure will be interpretable in a broadly Davidsonian sense: roughly, mappable onto ours in a way that makes conversation with us possible.

1. Two approaches to the philosophy of language For the past hundred years or more the philosophy of language has been inscribed in a space delineated by two polar approaches. On the one hand, a structural-semantic approach whose avatars include Frege, Carnap, and Tarski focuses primarily on the way the contents expressed by complex or compound expressions1 depend on those of simpler ones. On the other 1. Not the same thing. Compound expressions actually contain the simpler expressions from which they are constructed. Complex expressions, paradigmatically complex predicates (the sort of thing quantifiers attach to), are formed by assimilating compound expressions into equivalence classes of substitutional variants. There is no simple predicate—the sort of thing used to build compounds such as ‘admired and wrote about’—that ‘Rousseau admired Rousseau,’ and ‘Kant admired Kant’ share with each other that they do not also with ‘Kant admired Rousseau.’ But the first two share a complex predicate ‘D admired D’ that they do not share with ‘Kant admired Rousseau.’ Complex expressions are not parts of compound ones, but patterns compound expressions exhibit.

hand, a pragmatic-anthropological approach characteristic of the American pragmatists, the early Heidegger, and the later Wittgenstein directs its attention in the first instance to the natural history of language as a social practice. The former addresses meaning, the latter use. Where the first is concerned with what is said, the second is concerned with what one is doing in saying it. Where one asks what it is that one knows (or believes) when one knows that p, the other asks what one must know how to do in order to count as being in a state that exhibits, or producing a performance that expresses, such a content. Thought of this way, I take it that the two enterprises should be thought of as complementary rather than competing. For semantics, the theory of the contents expressed by using various sorts of vocabulary, and pragmatics, the theory of the practices of using those locutions, can each be pursued in ways that are at most provisionally independent of one another—subject always to the proviso that neither sort of theory can count as adequate unless compatible with an at least acceptable version of the other. Nonetheless, the Fregean and the Wittgensteinean traditions have not as a matter of historical fact had a lot to do with one another. With some honorable exceptions2, philosophers of language in the model-theoretic tradition of formal semantics do not concern themselves with the issue of what it is linguistic practitioners must be able to do in order to associate semantic interpretants of the favored kind—typically construed in rather abstract ways—with the simple expressions of the language. (One excuse for that failure—to my eyes, one that is particularly lame—appeals to the division of intellectual labor: the philosophy of mind is to make up the difference, by offering an account of the thoughts and intentions that accompany speech. But surely the generic challenge of semantics encompasses accounting for the contents both of thought and of talk.) Conversely, pragmatists and neo-pragmatists, whether of classical, early Heideggerian, or Wittgensteinean lineage, have proven not so much unsuccessful at as simply uninterested in extracting from their investigations of the practices of using linguistic expressions detailed semantic lessons of the sort that abound in, say, possible worlds semantics (concerning for instance, the contents expressed by modal or conditional locutions). Of course the disjunction between semantic and pragmatic concerns is not total. Two of the most discussed approaches in the philosophy of language 2. I’m thinking of efforts such as that of David Lewis, in his ‘Languages and Language’ (Lewis 1975) and Robert Stalnaker, in Inquiry (Stalnaker 1984).

14

of our day are those of Davidson and Dummett—philosophers each distinguished precisely by the fundamental and distinctive ways in which they bring semantic and pragmatic concerns to bear on one another. (Outside of the Anglophone tradition, one should in this connection esteem as well the tradition that grew out of Lorenzen’s constructivism.) In Making It Explicit, I expound an approach to semantics and pragmatics that seeks to do better justice to the insights of the two polar traditions. On the semantic side, it pursues an idea of Frege’s: that specifically conceptual content is to be distinguished by its relevance to the proprieties of inference associated with an expression.3 The conceptual contents of states, performances, or expressions, should be identified with their inferential roles, that is, their roles in reasoning. To be rational is to be sensitive to the normative force of reasons, to tell in practice what is a reason for what, to be able to distinguish good reasons from bad ones. So on this line of thought there is a deep and intimate connection between the concepts concept and rationality, and hence between semantics and reason. According to this semantic rationalism, the relation is a reciprocal sense-dependence: one cannot grasp the concept of conceptual content except insofar as one grasps the concept of reasoning, and vice versa. 2. Three models of rationality How one thinks about rationality is then of the first importance for this approach to semantics. The account developed in Making It Explicit is best understood against the background of three more familiar models of rationality: the logical, instrumental, and interpretational. On one picture, to be rational is to be logical. Being sensitive to the force of reasons is a matter of practically distinguishing logically good 3. ‘There are two ways in which the content of two judgments may differ; it may, or it may not, be the case that all inferences that can be drawn from the first judgment when combined with certain other ones can always also be drawn from the second when combined with the same other judgments. The two propositions ‘the Greeks defeated the Persians at Plataea’ and ‘the Persians were defeated by the Greeks at Plataea’ differ in the former way; even if a slight difference of sense is discernible, the agreement in sense is preponderant. Now I call that part of the content that is the same in both the conceptual content. Only this has significance for our symbolic language [Begriffsschrift] (…). In my formalized language [Begriffsschrift] (…) only that part of judgments which affects the possible inferences is taken into consideration. Whatever is needed for a correct [‘richtig’, usually misleadingly translated as ‘valid’] inference is fully expressed; what is not needed is (…) not.’ Frege (1879), section 3.

15

arguments from those that are not logically good. For a set of claims to serve as a good reason for another claim is for there to be a logically valid argument relating them to that claim as premises to conclusion. Nonlogical facts and the meanings of nonlogical vocabulary contribute to reasoning only by providing premises for logically valid inferences. The program of assimilating all good reasoning to this model has been immensely influential and productive in the philosophical tradition. It took its modern form when Frege vastly increased the expressive power of logic by giving us formal control over the inferential significance of quantificationally complex properties. The success this idiom was shown to have in codifying mathematical reasoning—by Frege himself, by Hilbert, and by Russell and Whitehead—was a major impetus for logical empiricism, whose central project was to extend the logical model of reasoning to include empirical science. Just when it looked as though the limits of this enterprise had been reached, technical advances in the logical expression of modalities gave the undertaking new life. The logical model of reasoning is most at home close to its origins: in codifying theoretical inference, the way beliefs can provide reasons for other beliefs. The instrumental model of reasoning begins with practical inference—in particular, the way desires or preferences, together with beliefs, can provide reasons for action. It identifies rationality with intelligence, in the sense of a generalized capacity for getting what one wants: the reason of Odysseus, rather than of Aristotle. What one has reason to do, on this model, is what provides a means to an endorsed end. Means-end reasoning is formally codified in rational choice theory, in both its decision-theoretic and game-theoretic species. Dutch book arguments show that utility (the measure of preference) will be maximized by practical reasoners who assign probabilities to compound beliefs in ways that satisfy the axioms of classical probability theory. And the laws of classical logic can be deduced as special cases from those axioms. So the instrumental model of rationality has some claim to subsume the logical one as a special case. One thing to notice about these two models of rationality is that they both treat (nonlogically) contentful beliefs and desires as inputs. Given a set of beliefs, and perhaps desires, they purport to tell us which connections among them are rational: which constellations of them provide genuine reasons for which others. They accordingly presuppose that the contents of those psychological states can be made intelligible independently and in advance of considering rational connections among them. The idea that one can first fix the meaning or content of premises and conclusions, and 16

only then worry about inferential relations among them, is characteristic of traditional and twentieth century empiricism. This implicit semantic commitment is questioned, however, by the rationalist tradition in semantics, which sees issues of what is a reason for what as essential to the identity and individuation of the conceptual contents that stand in those inferential relations. The logical and instrumental models of reasons are also (and not coincidentally) alike in their formality. Each sees rationality as being a matter of the structure of reasoning, rather than its content. The substantial content of the beliefs and desires that provide the premises for candidate theoretical and practical inferences are wholly irrelevant to the rationality of the conclusions drawn from them. All that matters for the correctness of the inference is that they have the form of deductively valid inferences or maximization of expected utility given those premises. The premises themselves are beyond criticism by these models of rationality, unless and insofar as they themselves were acquired as conclusions of prior inferences, which are assessable in virtue of their form—and then only relative to the prior (only similarly criticizable) commitments that provide their premises. A model of rationality that is not in this way purely formal is the translational-interpretational model, most fully developed by Davidson. According to this view, to say that some behavior by others is rational is roughly to say that it can be mapped onto our linguistic behavior in ways that make it possible for us to converse with them—at least to draw inferences from their claims, to use them as premises in our own reasoning. The idea is to use our own practical know how, our ability to distinguish reasons from nonreasons and to tell what follows from what, to assess the theoretical rationality of others. They are rational insofar as their noises (and other behavior, described in nonintentional terms) can be mapped onto ours so as to make them make sense by our standards: to exhibit them as believers in the true and seekers after the good by our own lights. Rationality, then, is by definition what we’ve got, and interpretability by us is its definition and measure. Rationality is not on this view a formal matter at all. For the unintelligibility or wackiness of the substantive, nonlogical beliefs and desires we take our interpretive targets to be evincing in their behavior, both linguistic and nonlinguistic, is every bit as relevant to assessments of their rationality as the connections between them we discern or take them to espouse. We have to be able to count the others as agreeing with us in the contents of and (so) connections among enough of their beliefs and desires to form 17

a background against which local disagreements can be made intelligible, if we are to find them interpretable, that is, rational—for what they have to show up as beliefs and desires—at all. Rationality as interpretability can also claim to subsume or incorporate both the logical and the instrumental models of rationality. For the first, the explicit form of a Davidsonian interpretation includes a recursive truth theory for the idiom being interpreted, including novel sentential compounds that have never actually been used. So identifying expressions functioning as logical vocabulary can provide a formal framework within which the rest of the interpretive process can take place. Being logical creatures is on this view a necessary condition of being rational ones, even though there is a lot more to rationality than just that. For the second, making the behavior of the interpreted creatures intelligible requires attributing sample bits of practical reasoning. And Davidson takes it that those will have the form of what he calls ‘complete reasons’: constellations of beliefs and desires that rationalize the behavior according to the instrumental model. Unless one can interpret the target behavior as for the most part instrumentally rational, one cannot interpret it at all. Finally, the interpretive model does not take the rational connections among psychological states or the sentences that express them to be irrelevant to the contents they are taken to evince. On the contrary, what makes something have or express the content it does is what makes it interpretable in one way rather than another. And that is a matter of its connections to other things, the role it plays in the overall rational behavioral economy of the one being interpreted. What makes it right to map another’s noise onto this sentence of mine, and so to attribute to it the content expressed by that sentence in my mouth, is just that its relations to other noises sufficiently mirrors the relations my sentence stands in to other sentences of mine: what is evidence for and against it, and what it is evidence for and against, as well as what environing stimuli call forth my endorsement of it and what role it plays in practical reasoning leading to nonlinguistic action. Those consequential relations are of the essence of interpretability, and so of rationality on this model. 3. The strategy of making it explicit I think the interpretational model of rationality is correct as a criterion of rationality, but that it needs to be supplemented in order to yield an account 18

of rationality. For a set of practices to be interpretable in this sense means that there is a mapping of it onto our discursive practices that preserves the goodness of a whole variety of inferences: theoretical, practical, and reliability inferences (about which more later). Insisting on that is not yet saying what it is for the articulation of our practices to count as genuinely inferential articulation, and so as capable of conferring conceptual content on the states, performances, and expressions that are so articulated and related. Making It Explicit aims to offer a direct account of a structure such that any set of practices exhibiting that structure thereby counts as discursive, and hence as conferring conceptual content on things playing suitable roles in those practices. In keeping with its semantic rationalist inspiration, the basic role in question is that of premise and conclusion in inferences. If so, then the primary form of conceptual content is propositional content, the sort expressed by declarative sentences. This is the thought that led Kant to insist that the judgment is the smallest unit of awareness or experience, and Frege to privilege truth values as the most basic sort of semantic intepretant. The rationalist answer to the sometimes vexed question of how to understand ‘the unity of the proposition’ is an inferential one: to be propositionally contentful is to play a distinctive role in a practice of giving and asking for reasons, namely to both be able to serve as and to stand in need of—be offered as and provided with—reasons. Playing that dual inferential role is what distinguishes expressions for propositions from expressions of other sorts of content. What, then, are the minimal conditions on being a practice of giving and asking for reasons? It is at this point that the basic pragmatic idea of Making It Explicit enters the scene. Discursive practices are those involving the adoption and attribution of two sorts of normative status, commitments and entitlements. When the practical consequential relations among those statuses have the right structure—a matter of the practical consequences of adopting or attributing one for the adoption or attribution of another— they count as inferentially articulated. Shifting attention from semantic concern with contents expressed to pragmatic concern with the acts of expressing brings to the fore the performance-kind asserting or judging. For it is assertings (publicly) or judgings (privately) that can have the pragmatic significance both of being liable to demands for reasons and of serving to respond to such demands. What one is doing in making an assertion or a judgment is undertaking a special kind of commitment. (Indeed, as I read him, Kant’s reason for taking judgments to be the minimal form of awareness is that they are the smallest unit for which one can be responsible.) It 19

is characteristic of that kind of commitment that what one is committed to by undertaking an assertional commitment includes other such commitments—presystematically, those that follow from it inferentially. Apart from such consequences for other commitments, performing an assertion or judgment would be idle, of no pragmatic significance whatever, a wheel that engaged with no further mechanism. And to say that assertings and judgings are things liable to demands for reasons is to say that part of what one makes oneself responsible for by undertaking such a commitment is, under appropriate circumstances, to vindicate one’s entitlement to it, by undertaking other commitments that serve as reasons for it, i.e. from which it follows. The basic thought, then, is that for a practice to count as specifically discursive, that is as conferring conceptual content—which according to the thesis of semantic rationalism means to be a practice of giving and asking for reasons—it must involve at least two sorts of normative status, commitments and entitlements, standing in consequential relations to one another. That thought motivates a picture of discursive practice as a kind of deontic scorekeeping. To engage in a discursive practice is to keep track of the commitments and entitlements of other practitioners. The pragmatic significance of a speech act is the difference it makes to what the performer and his audience are committed and entitled to. One’s grasp or understanding of the significance of such a performance is the difference one takes it to make to everyone’s commitments and entitlements to commitments. So the significance for an interlocutor of a speech act or other performance in a deontic context consisting of the commitments and entitlements of the various participants is a matter of the way it mandates the updating of that context, yielding a subsequent context or ‘score’. And the content of the expression uttered can be identified roughly with the update function, which specifies for each possible deontic context of utterance what the scorekeeping significance of producing that expression (paradigmatically, with assertional force) in that context would be. The connection between the normative scorekeeping pragmatics and the inferentialist semantics is secured by the idea that the consequential scorekeeping relations among expression-repeatables needed to compute the significance updates can be generated by broadly inferential relations among those expression-repeatables.4 The theory propounded in Making 4. In Making it Explicit there are two fundamental sorts of tokening-repeatables: symmetric lexical-syntactic cotypicality equivalence classes and asymmetric anaphoric substitution-inferential inheritance tree structures. See Chapter Seven.

20

It Explicit is that there are six consequential relations among commitments and entitlements that are sufficient for a practice exhibiting them to qualify as discursive, that is, as a practice of giving and asking for reasons, hence as conferring inferentially articulated, thus genuinely conceptual content on the expressions, performances, and statuses that have scorekeeping significances in those practices. The bold and potentially falsifiable overall claim of the whole work is that any practices that exhibit this full six-fold structure will be interpretable in a broadly Davidsonian sense: roughly, mappable onto ours in a way that makes conversation with us possible. The six sorts of deontically definable relations can be put into two groups: fundamental-semantic and social-pragmatic. 4. Three fundamental semantic consequence relations The three fundamental semantic relations are commitment-preserving inferences, entitlement-preserving inferences, and incompatibility entailments. Commitment-preserving inferential relations among the contents expressed by two sentence-repeatables5 p and q are imputed by adopting in practice a consequential scorekeeping regularity: that anyone committed to (what is expressed by asserting) p is committed to (what is expressed by asserting) q. The concept of commitment-preserving inference is a generalization of that of deductive inference, from the case of logically good to the case of materially good inferences. Thus any discursive scorekeeper who in practice takes it that as a matter of fact all the yachts in the harbor are sloops will take it that one who claims the John B. is in the harbor is thereby committed, whether she knows it or not, to the John B. being a sloop. Entitlement-preserving inferential relations among the contents expressed by two sentence-repeatables p and q are imputed by adopting in practice a corresponding consequential scorekeeping regularity: that anyone entitled to a commitment to (what is expressed by asserting) p is thereby prima facie (more about that later) entitled to a commitment to (what is expressed by asserting) q. The concept of entitlement-preserving inference is a generalization of that of inductive inference. Thus anyone who endorses the sailor’s homily ‘Red sky at night, sailor’s delight; red sky in morning, sailor take warning,’ will take it that being entitled to the claim that the sky is red at sunset provides a reason for, and in that sense 5. See footnote 4.

21

entitles one (other things being equal) to the claim that the weather will be fine tomorrow. It does not yet commit one to that claim, for one may also believe that if the barometer falls in the evening, there will be stormy weather the next day, and that the barometer is falling at sunset. Unlike good deductive inferences from true premises, the conclusions of good inductive inferences from true premises may contradict one another. The third fundamental semantic relation involves the interaction of commitments and entitlements. What is expressed by p and q may be said to be incompatible just in case commitment to one precludes entitlement to the other. Thus ‘I will work all day tomorrow in my office,’ and ‘I will drive you to the airport tomorrow afternoon,’ are incompatible in this sense. It is not that I cannot commit myself to both. But if I do, I do not count as entitled to either commitment, for the other one in this sense rules it out. Incompatibilities of content generate entailments among them. Thus being a mammal entails being a vertebrate, in that everything incompatible with being a vertebrate (for instance, being a squid) is incompatible with being a mammal, but perhaps not vice versa. (For instance, being a reptile is incompatible with being a mammal, but not with being a vertebrate.) Incompatibility entailments are modally robust: if my favorite pet (an octopus, we may suppose) were a mammal, she would be a vertebrate. Here is how these semantic relations among contents help determine the pragmatic significances of (paradigmatically assertional) speech acts involving expressions with those contents. When S asserts p, scorekeepers first add to the repertoire of commitments attributed to S commitments to all those claims q such that (according to the scorekeeper) there is a commitment-preserving inference from p to q. (This will obviously include p.) Next, one adds to the repertoire of entitlements attributed to S all the q’s such that (according to the scorekeeper) there is an entitlement-preserving inference from p to q. Finally, one subtracts from the repertoire of entitlements attributed to S all the claims q such that there is some r incompatible with q to which S is (according to the scorekeeper) committed. This last feature is the primary locus of the important difference between the sort of (social) normative functionalism about semantic content put forward in Making It Explicit and more familiar (individual) causal functionalisms. For accounts of content in terms of causal roles have difficulty making sense of the possibility of inconsistent, or more generally, incompatible beliefs. One cannot, after all, simultaneously be in states individuated in part by the fact that they cause incompatible behaviors. Such theories sometimes, implausibly, simply deny the possibility of having inconsistent 22

beliefs. (Possible comparison: the denial by rational choice theorists that it could be so much as intelligible to interpret behavior by attributing cyclical preferences.) Otherwise, causal-functional theories typically make ad hoc adjustments, either by partitioning believers with inconsistent beliefs into competing consistent subsystems, or to merely statistical causal influences. By contrast, no corresponding surds threaten the normative functionalist conception of agents with incompatible commitments. The possibility of undertaking (and hence of attributing) incompatible doxastic commitments is straightforwardly intelligible in the same way that undertaking incompatible practical commitments is. Just as I can, if I am sufficiently thoughtless, injudicious, or just unlucky, make incompatible promises concerning what I will do, and so how things are to be—at the cost, to be sure, of being unable to fulfill them all—so I can make incompatible commitments concerning how things are. The corresponding cost is that I cannot then be entitled to any of those commitments. But that fact does not by itself in any way undercut my status as nonetheless committed. One of the principal ways in which I can non-culpably acquire incompatible commitments is when some of the inferential consequences of prior commitments to which I am entitled collide with the current deliverances of my non-inferential reporting capacities. Consider a microbiologist who has concluded, on the basis of symptoms and exposure, that the rod-shaped bacteria recovered from a diseased tissue is E. coli, a Gram-negative bacillus. If she then finds that it turns purple when exposed to crystal violet stain, and hence is Gram-positive, she has acquired incompatible beliefs, one as the result of inference, and the other non-inferentially, by inspection of the slide. Response to the stain indicates B. cereus, but the presence of that bacillus is, we may suppose, incompatible with the observed symptoms and demonstrated exposure. Semantics dictates only that she cannot be entitled either to the claim that the agent is E. coli or to the claim that it rather B. cereus. It does not tell her what to do to repair this situation of incompatible commitments—for instance, whether to suspect that the symptoms were misdescribed, the exposure misreported, or the stain misapplied. The details of that sort of updating are a cognitive, not merely a semantic matter.6 6. And this is true even if the incompatibility of commitments is more diffuse. For material incompatibilities can be like formal logical ones, in that one may have a set of n commitments, any n-1 of which are compatible, but which jointly are incompatible—as for instance is true on the logical side of {p, poq, qor, ™r}. Semantically, anyone with all these commitments is entitled to none of them. But that semantic fact only settles it that one ought to do something

23

5. Three basic social pragmatic structures This example indicates a crucial mechanism whereby features of the causal order can come to be reflected in the conceptual order. The essential empirical dimension of conceptual content depends not only on the fundamental semantic notion of incompatibility (definable, as we have seen, in terms of the still more basic normative status concepts commitment and entitlement), but also on the distinctive semantic contribution of noninferential reports. That topic brings us to the second triad of deontically definable content-constituting relations: the social-pragmatic ones. The three basic social pragmatic structures are the empirical, the practical, and the testimonial. The fundamental semantic relations were all strictly inferential, and reflecting practically what Sellars calls ‘language-language moves’ (Sellars 1954). The social pragmatic relations widen the practical focus. The empirical dimension articulates the contribution to conceptual content made by language-entry moves, in which a knower responds to a (typically, but not always) non-linguistic situation by acknowledging an inferentially significant doxastic commitment. The practical dimension articulates the contribution to conceptual content made by language-exit moves, in which a knower responds to the acknowledgment of an inferentially significant practical commitment by altering a (typically, but not always) non-linguistic situation. The first encompasses non-inferentially elicited perceptual reports, and the second non-inferentially efficacious intentional agency. The testimonial dimension articulates the contribution to conceptual content made by the possibility of interpersonal noninferential inheritance of commitments and entitlements. Our capacity to find out how things are in the world around us depends on our ability to practically distinguish different kinds of environing states of affairs by responding differentially to those stimuli. In the simplest cases, there is nothing distinctively cognitive or discursive about such abilities. Chunks of iron rust in wet environments and not in dry ones, branches of a given diameter break with sufficiently heavy loads and not with lighter ones, motion detectors turn on the porch light if large things move across their field of sensitivity, but not if small ones do, and so on. By reliably responding differentially to different kinds of stimuli, such systems can be understood as classifying their environments as being of different kinds. But to repair the situation, undertake some revision of those commitments. It does not offer advice about which of the many ways of doing that might be best.

24

this is not yet conceptual classification, and so not yet conceptual awareness. The boundary from mere sensitivity, irritability, or even sentience, to true sapience is crossed when the response that is reliably differentially keyed to stimuli of a certain kind is the application of a concept—the use of a word that plays an appropriate role in inferential relations of the three sorts already discussed. The difference between a parrot trained to respond by uttering the noise ‘That’s red!’ to the presence of visible red things, on the one hand, and a genuine reporter of red things, on the other, is that the reporter’s utterance is, as the mere responder’s is not, the undertaking of an inferentially articulated commitment: the making of a move in a game of giving and asking for reasons, the endorsement of a content that can both serve as a reason for endorsing others and is liable to demands for vindication of the reporter’s entitlement to a commitment with that content. Empiricists have always been right to stress that apart from the capacity reliably to respond differentially to stimuli in perception, not only empirical knowledge, but even empirical conceptual content would be unintelligible. And rationalists have always been right to stress that apart from the role a performance, state, or expression plays in reasoning, its conceptual contentfulness is unintelligible. As Kant taught us, they err when they succumb to their characteristic complementary temptations to treat one or the other of these necessary conditions of content, and so cognition, as sufficient. The causal dimension of reliable responsiveness and the inferential dimension of rational endorsement are equally essential to empirical conceptual contentfulness. The two aspects of empirical concepts are to a certain extent intelligible independently of one another. We’ve just seen that reliable differential responsive dispositions can be discerned in non-discursive systems. And the furthest reaches of mathematics—abstract algebra or pure set theory, for instance—give us a grip on inferential relations that in principle need not be anchored in perceptual or practical interactions with concrete things. Nonetheless, the two aspects—causally reliable responsiveness and inferentially articulated commitment—should not be understood as simply bolted together to yield non-inferential reporting capacities. For they interact intimately. The basic idea is that at least some inferential relations must be brought to track causal ones, so that reliable responsiveness is incorporated in the inferential articulation of conceptual contents. To see how this works, think about the semantics in terms of the pragmatics (inferential relations in terms of scorekeeping activities). What must one do in order to be taking or treating another interlocutor as a reliable 25

reporter, of, say, red things? To take someone to be a reliable reporter is taking it that he is likely to be right, that his observational claim is likely to be true. But taking a claim to be true is just endorsing it oneself. That is, to treat the other as reliable is to take it that his non-inferentially elicited response to something as red—his endorsement of the observation report—provides a reason for endorsing that claim oneself. To do that is to endorse an inference of a distinctive kind, a reliability inference. That is an inference from an attributed commitment to a corresponding acknowledged commitment. More carefully put, for S to take S’ to be a reliable reporter of red things is for S to keep deontic score in such a way that whenever S takes S’ to be responsively committed to the claim that something is red, everyone else, S included, is thereby counted as prima facie entitled to that same claim. The entitlement in question is only prima facie, since any entitlement is defeasible by concomitant commitment to incompatible claims. And for the same reason, S need not always endorse the claim attributed to S’ in order to treat S’ as reliable, since in some cases S may have collateral commitments incompatible with that one. Thus I may take you to be a generally reliable reporter of red things and still not endorse your report of a particular object as red, if I have information you lack about the non-standard lighting conditions under which you are viewing it. For your authority to be genuine, it need not be indefeasible. The reliability of the causal process by which a reporter responds to the fact of visible red things by endorsing the claim that there are red things there is then reflected in conceptual, that is to say, inferential relations by the consequential scorekeeping relation between attributing commitments to the reporter and acknowledging corresponding entitlements for the reliability-attributing scorekeeper and (according to that same scorekeeper) also for others. The reliability inference is an inference, but unlike committive, permissive, and incompatibility entailments, it is an essentially socially articulated inference. For it relates commitments attributed to a reporter to entitlements acknowledged by the scorekeeper, and attributed not only to the reporter, but also to others in the reporter’s audience. This social articulation of the reliability inference shows up the authority of reliable observers as a species of a more general kind of interpersonal entitlement-inheritance: testimony. The dimension of testimonial authority is an essential element of the deontic scorekeeping understanding of the pragmatic significance of the fundamental (because pathognomic for the characterization of a practice as specifically discursive) speech act of assertion. The basic thought is that in asserting that p, one is doing two 26

things. On the side of authority, one is licensing others to reassert one’s claim. On the side of responsibility, one is committing oneself to justify the claim if suitably challenged. Both of these should be understood in terms of the inheritance of entitlement to the commitment one has undertaken. A scorekeeper who acknowledges the authority of one’s assertion will count those in the audience as having inherited from it prima facie entitlement (defeasible, as always, by concomitant commitment to incompatible contents) to the content one has endorsed. And if one is challenged, paradigmatically by the assertion of an incompatible claim to which the challenger is at least prima facie entitled, one is then obliged, in the eyes of the scorekeeper, to vindicate one’s entitlement to the commitment undertaken in the original assertion, paradigmatically by undertaking other commitments that stand to it as premise to conclusion in an at least entitlement-preserving inferential relation. What if one fails to fulfill that justificatory responsibility in response to a suitably qualified challenge? The cost of that failure is simply the loss of the authority one implicitly claimed by issuing the assertion in the first place. For a scorekeeper who takes it that one has not shown oneself to be entitled to the commitment in the first place thereby takes it that one does not have a heritable entitlement to bequeath to others. On the other hand, while that re-assertion license is in effect, either because it has not been challenged or because any actual challenges have been satisfactorily responded to (all in the eyes of the scorekeeper to whom we are attending), the force of the authorizing is that those who by its means inherit prima facie entitlement to the claim can, if they are challenged, vindicate their entitlement by deferral to the original assertor, in lieu of having to produce independent justifications. Thus on this picture the interpersonal, intracontent inference-licensing authority of an assertion, and the intrapersonal, intercontent inferential-justificatory responsibility one undertakes by implicitly asserting that authority are co-ordinate and interdependent: two sides of one coin. We saw that the authority of reliable observers is presumptive, but defeasible. So too with the more general category of testimonial authority. The picture is that assertions by competent speakers have a default authority. Their normative power to license or entitle others to reassert the claim, to use it as a premise in their own inferences, is considered to be in force until and unless their assertion is challenged by an incompatible claim to which the challenger has at least an equivalent prima facie entitlement. (That would be lacking if the challenger were also committed to claims 27

incompatible to the challenging one, for instance.) For granting that default prima facie entitlement to assertions just on the basis of the fact that the commitment has been undertaken is what it is for a scorekeeper to treat the assertor as a competent deployer of the concepts in question— whether they are observation concepts or not. These observations about the contribution of pragmatics to semantics in the system under discussion make it possible to show something about the relation between the various sorts of fundamental inferential relations that is not otherwise obvious. In particular, it is possible to show that if one claim incompatibility entails another, then there is also a good commitment-preserving inference from the first to the second. Suppose that p incompatibility entails q. This means that everything incompatible with q is incompatible with p, and suppose that S is committed to p. The question is whether S thereby counts as committed to q. We can start by asking what it means in scorekeeping terms to be committed to q. We’ve seen that that status is specified in terms of the two dimensions of authority and responsibility it involves: First, undertaking a commitment to a claim involves authorizing or licensing others to assert it (undertake commitment to it), which in practice means authorizing them to defer to the original assertor the responsibility to justify or demonstrate entitlement to the claim if appropriately challenged. Second, (and presupposed by the first), undertaking commitment to a claim involves the responsibility to justify or demonstrate entitlement to it if it is appropriately challenged. From the point of view of our current concerns, the important thing to notice about this articulation of the pragmatic scorekeeping significance of doxastic commitments is the crucial role played in it by appeal to entitlements. It is this, ultimately, that connects incompatibility entailments with commitment-preserving inference (and indeed, the latter with entitlementpreserving inference). With this reminder on board, consider whether commitment to p involves commitment to q wherever p incompatibility entails q. Think first of the dimension of responsibility: If one is committed to justify p when it is appropriately challenged, is one thereby committed to justify q when it is appropriately challenged? An appropriate challenge is the (at least prima facie justified) assertion of something incompatible. But since by hypothesis p incompatibility entails q, everything incompatible with q is incompatible with p. So every challenge to q is a challenge to p. So one must be able to respond to all these challenges in order to respond to all the challenges to p. That is, in order to be able to justify p, one must be 28

able to justify q. So on the side of responsibility, commitment to p involves commitment to q. What about the dimension of authority? There too, in authorizing others to assert p, one has thereby authorized them to assert q. For the cash value of that authorization is authorization to defer justificatory responsibility, in response to warranted challenges. And again, by hypothesis any warranted challenge to q is a warranted challenge to p. So in taking on the responsibility to answer for p, one has thereby also taken on the authority to answer for (respond to challenges to) q. Thus along both the dimension of authority and that of responsibility, commitment to p involves commitment to q, if p incompatibility entails q. We can see in much the same way why if there is a commitmentpreserving inference from p to q, there is also a (prima facie) entitlementpreserving inference from p to q. For if everyone who is committed to p is thereby committed to q, then on the side of authority, in authorizing others to assert p, I am thereby authorizing them to assert q. And to say that is to say that they can inherit entitlement to q from my entitlement to p. And on the side of responsibility, in undertaking the responsibility to justify or otherwise vindicate my entitlement to p, I am thereby undertaking the responsibility to justify, or otherwise vindicate my entitlement to q. So I cannot be entitled to p unless I am also entitled to q, which is to say that there is a good inference from p to q preserving prima facie entitlements. So the social pragmatics of testimonial entitlement inheritance settles it that the three flavors of fundamental semantic inference are strictly ordered in strength. If p incompatibility entails q then there is a good commitmentpreserving inference from p to q. And if there is a good commitment-preserving inference from p to q, then there is a good entitlement-preserving inference from p to q. This is another crucial way in which the normative pragmatic account of what one is doing in committing oneself to a claim is inextricably bound up with the inferential semantic account of the conceptual contents to which one thereby becomes committed. 6. Practical inference The third and final sort of social pragmatic inferential relation is that of practical inference. By that I mean inferences whose conclusions include practical commitments: commitments to do something. These are commitments to make some claim true, rather than commitments whereby 29

one takes some claim to be true. In the normative pragmatics of Making It Explicit, the concept of doxastic commitment, picking out the kind of thing expressed by assertions, does much of the explanatory work for which the concept of the intentional state kind belief is called into service by more traditional philosophies of mind. In much the same way, the concept of practical commitment does much of the explanatory work for which the concept of the intentional state kind intention is called into service by those traditional theories. The idea is that being a competent agent is being able reliably to respond to acknowledging at least the most basic sort of practical commitment by bringing about a state of affairs of the sort specified by its content: making true a claim with that propositional content. This capacity for reliable language-exits is to be understood by strict analogy to the capacity for reliable language-entries. We can be trained reliably to respond to visible red things by acknowledging a commitment to their being red, and because of the way the world is and the way we are wired up we cannot be so trained non-inferentially to report without the aid of instruments the presence of things that emit radio waves or that once belonged to the emperor. Just so we can be trained reliably to respond to a commitment to raise our arms by making it true that our arms go up, and because of the way the world is and the way we are wired up cannot be so trained non-inferentially to bring it about without the aid of instruments that we emit radio waves or that the Moon turns a different face toward the Earth. As in the theoretical case of observation, attributing practical reliability to an agent with respect to a range of possible performances is a matter of the scorekeeper adopting a certain sort of inferential commitment. In both cases, the inference is one in which a commitment attributed to another is taken as reason for endorsing a claim oneself. In this case, my taking you to be reliable arm-raiser is taking it that your undertaking a practical commitment to raise your arm gives me good reason (entitles me) to commit myself to the claim that your arm will go up. Practical inferential relations then can be thought of as governing transitions (commitment or entitlement inheritance) from doxastic to practical commitments, that is, from the commitments acknowledged in assertions to commitments to do something, made explicit by saying something like ‘I shall M.’ Practical reasoning of this sort is what is rehearsed in deliberation, and attributed in assessments of what others had good reason to do. We make each other’s behavior rationally intelligible by attributing and assessing sample bits of practical reasoning that would rationalize what 30

they do, even in the cases where we do not take it that the behavior in question was the result of explicit deliberation, that is, causally resulted by detachment of a practical conclusion from such a process of practical reasoning. Seen from this angle, expressions of preference or desire show up as codifying commitment to the propriety of patterns of practical inference. Thus S’s preference or desire to stay dry is a commitment to the propriety (here, in the sense of entitlement-preservation) of inferences of the form: Only doing A will keep me dry. ? I shall do A. Attributing a preference or desire is attributing commitment to such a pattern of practical inferences. The explicit claim ‘I prefer to stay dry,’ stands to such implicit commitments to patterns of practical inference in the same expressive relation in which the conditional p oq stands to a commitment to the propriety of theoretical inferences from p to q (with different flavors of inference corresponding to different conditionals). In both cases, it is a mistake to confuse the statements that make inference licenses explicit with premises required for the inference to be licit in the first place—for reasons Lewis Carroll has made familiar in ‘Achilles and the Tortoise.’ Further, preferences and desires are only one sort of practical inference license. For in general, this is the expressive role distinctive of normative vocabulary as such. Thus a statement of the obligations associated with some institutional status, such as ‘Civil servants are obliged to treat the public with respect,’ licenses inferences of the form: Doing A would not be treating the public with respect. ? I shall not do A. This institutional pattern of practical inference differs from the preference pattern in that the latter is binding only on those who endorse the preference in question, while the latter is binding on anyone who occupies the status in question, i.e. on civil servants—regardless of their desires. Another pattern of practical reasoning is codified by normative claims that are not 31

conditioned upon occupation of an institutional status. Thus ‘It is wrong to (one ought not) cause pain to no purpose,’ licenses inferences of the form: Doing A would cause pain to no purpose. ?I shall not do A. Endorsing the unconditional normative claim is committing oneself to the bindingness of this form of practical inference for anyone, regardless of their preferences or institutional status. On the inferentialist picture, all of these ‘oughts’—the instrumental, the institutional, and the unconditional—are in the most basic sense rational oughts. For they codify commitments to patterns of practical reasoning. From this point of view, the Humean, who insists on assimilating all practical reasoning to the first, or instrumental model, on pain of a verdict of practical irrationality, and the Kantian, who insists on assimilating all practical reasoning to the third, or unconditional model, on pain of a verdict of practical irrationality in the form of heteronomy, are alike in pursuing Procrustean explanatory strategies. The real questions concern the justification of normative commitments of these various forms: the circumstances under which one or another should be endorsed, and what considerations speak for resolving incompatibilities among such commitments in one way rather than another. Some vocabulary plays a distinctive role in explicitly marking a commitment as the result of a language-entry move: ‘I see that the light is red.’ Similarly, some vocabulary plays a distinctive role in explicitly marking a commitment as the origin of a language-exit move: ‘I shall raise my arm.’ Some vocabulary has an immediate observational role: ‘red’, ‘square’, and ‘cat’, but not (usually) ‘emits radio waves’ or ‘owned by the emperor’. Some vocabulary has an immediate practical role: ‘painful’, ‘cruel’, and ‘desirable’, but not (usually) ‘rapid’ or ‘distant’. But the conceptual contents of a great deal of vocabulary that does not have this sort of special status are nonetheless affected by their inferential links, typically through seriously multipremise inferences, to vocabulary that does. So the content even of concepts at some inferential distance from reasons provided by observation and reasons for action nonetheless has an empirical and practical aspect, in virtue of those inferential connections.

32

7. Discursive practice I have now sketched six kinds of consequential scorekeeping relations among commitments and entitlements. The three fundamental semantic ones are consequential commitment, consequential (prima facie) entitlement, and incompatibility entailment. We’ve seen that these are strictly ordered by strength: if p incompatibility entails q, then anyone who is committed to p is committed thereby to q, and if anyone who is committed to p is committed thereby to q, then anyone who is provisionally (that is, apart from incompatible collateral concomitant commitments) entitled to a commitment to p is provisionally entitled to a commitment to q. The three social pragmatic consequential relations are language-entries through observation, language-exits through intentional action, and the testimonial structure of authority and responsibility that defines the basic pragmatic significance of assertional speech acts—thereby defining the concepts both of propositional content, on the one hand, and declarative sentence on the other. Each of these is to be understood in scorekeeping terms by appeal to reliability-inferential commitments on the part of scorekeepers. Inferences of this sort are essentially social or interpersonal in that they relate normative statuses attributed to one individual to those attributed to others or undertaken by the scorekeeper herself. Inferences of this sort are also normatively intermodal in that they relate commitments to entitlements. The core of the theory of discursive practice and conceptual content put forward in Making It Explicit is that these six consequential relations among commitments and entitlements are sufficient for a practice exhibiting them to qualify as discursive, that is, as a practice of giving and asking for reasons, hence as conferring inferentially articulated, and so genuinely conceptual content on the expressions, performances, and statuses that have the right kind of scorekeeping significances in those practices. This theory is developed to begin with from the point of view of someone looking at a set of social practices from the outside, and asking the question: what must be true of those practices, what structure must they be taken to exhibit, so that understanding them that way is implicitly taking them to be discursive practices? By the end of the book, however, this sort of external interpretive stance—what one must do, how one must treat an alien community in order thereby to count as taking them to be making assertions and inferences—is seen to be equivalent to an internal scorekeeping stance within a discursive community. That is, one must adopt toward 33

the practitioners the same sort of attitude one both takes them to adopt towards each other and adopts towards one’s own discursive fellows. One must keep deontic score on them, attributing commitments and entitlements that both have authority for one’s own, and to which one’s own commitments and entitlements are responsible. This collapse of levels can be thought of indifferently as the interpreter entering into the community to whom discursive practices are being attributed and as the interpreter treating those practitioners as members of her own discursive community. And that is to say that the interpreter, by becoming a scorekeeper, enters into a dialogical relation with the community being interpreted, mapping their utterances onto her own in a way whose adequacy is to be assessed by the fluency of conversation it enables. In short, the stance in question is a translational-interpretive stance that evidently belongs in a box with the orthodox Davidsonian variety. The overall claim of Making It Explicit then, is that the six broadly inferential structures outlined in the body of this paper articulate the fine structure of rationality in the Davidsonian interpretational sense. More specifically, the claim is that exhibiting the six-fold structure relating discursive scorekeeping practices to conceptual contents construed as inferential roles is necessary and sufficient for being interpretable, and hence rational, in the Davidsonian sense. That structure purports to be the structure of discursive practice as such—the structure distinctive of practices that deserve to count as practices of giving and asking for reasons, and hence (according to the semantic rationalism that is at the core of the theory) as conferring genuinely conceptual content. If it is correct, that is the structure that communities of extraterrestrials or of digital computers, however unlike us they may be in other respects, must exhibit if they are to qualify as potential interlocutors—as sapient knowers and agents, endorsers of claims and aims, makers and takers of reasons, seekers and speakers of truths.

REFERENCES Brandom, Robert. 1994. Making It Explicit. Cambridge (Mass.): Harvard University Press. — 2000. Articulating Reasons: An Introduction to Inferentialism. Cambridge (Mass.): Harvard University Press.

34

Frege, Gottlob. 1879. Begriffsschrift, eine der arithmetischen nachgebildete Formelsprache des reinen Denkens. Halle: Louis Nebert. (Translated as Conceptual Notation and Related Articles (1972). Ed. and trans. by Terrell W. Bynum. London: Oxford University Press.) Lewis, David. 1975. “Languages and Language”. In: Keith Gunderson (ed.). 1975. Language, Mind, and Knowledge. Minneapolis: University of Minnesota Press. Sellars, Wilfried. 1971. “Some Reflections on Language Games”. (Reprinted in his Science, Perception and Reality (1971). London: Routledge & Kegan Paul, 321–58.) Stalnaker, Robert. 1984. Inquiry. Cambridge (Mass.): Bradford Books.

35

Grazer Philosophische Studien 81 (2010), 37–54.

TWO ARGUMENTS FOR THE LANGUAGE-DEPENDENCE OF THOUGHT José Luis BERMÚDEZ Washington University in St Louis Summary This paper contrasts two ways of arguing that conceptual thought is languagedependent. One argument (due to Brandom) claims that conceptual content is fixed by inferential role, which is in turn fixed by inferential practices. The second argument (due to Bermúdez) claims that intentional ascent (thinking about thoughts) requires semantic ascent (thinking about words), because sentences are the only vehicles that reflect the canonical structure of a thought. The two lines of argument differ in how they view the relation between the internal structure of a thought and its inferential role. The paper argues that Brandom is mistaken in viewing inferential role as explanatorily prior to internal structure, but that seeing inferential role as fixed by internal structure is compatible both with the conclusion of his argument and with most of its premises.

1. Concepts and language: the blindspot The relation between concepts and language can be explored in a number of different ways. For example, cognitive scientists, and philosophers influenced by cognitive science, frequently debate whether concepts are fundamentally linguistic or imagistic. This is best understood as a debate about the representational format of a certain type of subpersonal information-processing—the information-processing involved in making certain types of basic perceptual discriminations and categorizations. There is a powerful view in the psychology of perception, dating back at least as far as R. L. Gregory’s The Intelligent Eye, to the effect that perceptual categorization is a fundamentally theoretical process—a matter of using information arriving at the sensory periphery to formulate and test hypotheses about the lay-out of the distal environment (Gregory 1970). According to the representational theory of mind, as developed by Jerry Fodor and others,

these hypotheses must be formulated in a linguaform representational medium (Fodor 1975). According to Fodor, the very possibility of perceptual categorization depends upon there being a language of thought. This is denied by proponents of the imagistic view (e.g. Prinz 2002). As is well known, many of the arguments for and against the language of thought hypothesis hinge upon whether, and if so in what sense, mental representations are structured.1 The issue here is whether subpersonal information-processing requires a linguaform vehicle in order to explain the possibility of structure-sensitive information-processing. Since the type of perceptual categorization under discussion is plainly performed by non-linguistic creatures, the presence or absence of a linguaform representational medium has nothing to do with participation in a public language. For this reason alone, one might reasonably think that this debate is orthogonal to more mainstream philosophical debates about whether concept mastery has an essentially linguistic dimension. This impression should be reinforced when one remembers that concept mastery, in the sense that many philosophers discuss it, is a much richer intellectual achievement than perceptual discrimination and categorization. Possessing even the simplest type of perceptual concept, such as the concept red, involves much more than perceptual sensitivity to the relevant similarities between objects. Although concept possession often requires perceptual sensitivity, it invariably goes far beyond it. Many philosophers have stressed that concept mastery is an inferential ability. Possessing a concept involves inferential sensitivity—sensitivity to the validity of inferential transitions between judgments involving that concept. One possible conclusion here would be that these two debates are completely independent of each other. So one might think that there is, on the one hand, a debate about the subpersonal vehicles of cognition and, on the other, a debate about the type of personal-level inferential abilities involved in deploying concepts. These two debates are independent of each other, and (on this view) it is only unfortunate accidents of vocabulary and homonymity that can lead the incautious to think that they share some common ground. Something like this is held by Robert Brandom (1994, 2000), John McDowell (1994), and others. In one respect it is a version of what I have elsewhere termed the picture of the autonomous mind (Bermúdez 2005). Despite its appeal I do not think that this picture of the relation between subpersonal and personal-level psychology is ultimately 1. See Bermúdez (2005), ch. 9 for discussion of Fodor’s arguments.

38

sustainable. I do not want to discuss that broader question in this paper, however.2 My topic instead is a blindspot that can arise when one takes seriously the very real differences between questions about subpersonal vehicles and questions about personal-level concept possession. The blind spot comes when one thinks that issues of structure and the vehicle of thought arise purely at the subpersonal level, so that they need not feature in discussions of concept possession and conceptual inference. I’d like to bring out one version of this blindspot by contrasting two ways of arguing for the language-dependence of conceptual thought. The first is due to Robert Brandom and developed both in Making it Explicit (Brandom 1994) and Articulating Reasons (Brandom 2000). Brandom arrives at the conclusion that conceptual thought is essentially linguistic from the twin premises, first, that the propositional content of a belief or any other propositional attitude is fixed by what he calls its inferential articulation and, second, that inferential articulation is fixed by commitments and entitlements that arise in discursive practices of giving and asking for reasons. The second line of thought is one that I put forward in my book Thinking without Words (Bermúdez 2003). There I argued that a certain kind of thinking, namely thinking that involves what I called intentional ascent (or thinking about thinking), is essentially language-dependent. The language-dependence of this type of second-order thought is not a discursive matter. It has to do with language’s role as a vehicle for thought, rather than with the social and communicative dimension of language. However, unlike the line of thought canvassed by Fodor, what is required is a public language, not a linguaform subpersonal representational medium. There are some important differences between the scope of the two arguments. Brandom quite explicitly draws the conclusion that only languageusers can be believers. Here is a representative passage: Claiming and believing are two sides of one coin—not in the sense that every belief must be asserted, or that every assertion must express a belief, but in the sense that neither the activity of believing nor the activity of asserting can be made sense of independently of the other, and that their conceptual contents are essentially, and not just accidentally, capable of being the contents indifferently of both claims and beliefs. (Brandom 1994, 150)

In contrast I want to allow for the possibility of nonlinguistic believers, since one of the aims of Thinking without Words is to provide a philo2. I have discussed it elsewhere. See Bermúdez (1995), for example.

39

sophical framework and foundation for explanatory practices in cognitive ethology, developmental psychology, and other scientific disciplines that treat nonlinguistic creatures as genuine thinkers. I dispute Brandom’s claim that propositional attitude states cannot be contentful simply in virtue of the roles that they play in practical reasoning. Nonetheless, the differences may not be as acute as it initially seems, since we are both agreed that a certain type of essentially reflective thinking is necessarily linguistic. The issue for this paper is our respective ways of getting to that conclusion. 2. Argument 1: Brandom on the primacy of the discursive Brandom’s fundamental premise is that conceptual content is fixed by inferential role, which he terms inferential articulation. The content that a belief has is a function of the reasons that a thinker has for judging it to be true and the rational relations in which it stands to further beliefs and to behavior. This premise is not, of course, unique to Brandom’s inferentialism. It would be denied, I think, only by those who hold that causal relations between physical realizers of belief states can completely supplant inferential relations between the contents of those states. We can put this extreme position to one side and concede that inferential role fixes conceptual content. Much more controversial is how Brandom develops the basic idea of inferential articulation. This is where the argument for language-dependence comes into the picture. Brandom’s key claim is that propositional attitudes have inferentially articulated contents only to the extent that those attitudes are linguistically expressed. This is so because inferential articulation is essentially performative. It is a matter of the commitments and entitlements that particular judgments and beliefs bring with them. According to Brandom, a belief ’s conceptual content is a function of three things: (a) the evidence on which it is based (b) the inferences to which it commits the believer (c) the inferences that it entitles the believer to make Again, this is a familiar idea. But Brandom gives it a very distinctive twist. He sees conceptual content as fundamentally dynamic. It is not given by abstract logical relations between propositions in logical space. Instead, we 40

are dealing with commitments and entitlements that arise and are exercised in the context of making claims, justifying our own claims, and evaluating the claims made by others. The social and linguistic context is essential to fixing inferential articulation. Conceptual content is determined by what happens in the discursive space of communication and dialog. The contrast between logical space and discursive space is very important for Brandom’s larger project, since he is trying to develop an alternative to the standard model-theoretic semantics that we have inherited from Frege via Tarski (although, as it happens, Brandom thinks that the seeds of his own approach are to be found in Frege’s early writings3). Brandom is emphatic that an account of inferential articulation can be given independently of an account of what we might think of as logical articulation, where the logical articulation of a propositional attitude content is given by the formally valid inferences that can be made as a function of its logical structure and/or truth-value. In fact, it is a cardinal tenet of his logical expressivism that logical articulation is to be derived from inferential articulation.4 There are two independent claims here. One is about the relation between formally valid inferences and materially valid inferences, where a logically valid inference is a substitution instance of a logically valid conditional and a materially valid inference is one that preserves truth as a function of the semantics of the relevant propositions. The claim here is that materially valid inferences are more fundamental than formally valid inferences. In particular, formally valid inferences, whether codified in the propositional or predicate calculuses, are explicit versions of the materially valid inferences that fix the inferential content of conceptual contents.5 The second claim has to do with the logical structure of a conceptual content. The logical structure of a sentence is a function of its subsentential articulation and, according to Brandom, subsentential structure is less basic than, and to be derived in terms of, the inferential articulation of the sentence.6 Let us call the first thesis the primacy of material validity and the second thesis the primacy of the propositional. Here are representative passages to illustrate each claim. 3. 4. 5. 6.

See for example Brandom (1994), 94–97. See, for example, Brandom (2000), ch. 1. Here Brandom is following Sellars 1953. See ch. 6 of Brandom (1994) and ch. 4 of Brandom (2000).

41

Primacy of material validity Logical vocabulary has the expressive role of making explicit, in the form of logically compound assertible sentential contents, the implicit material commitments in virtue of which logically atomic sentences have the contents that they do. (Brandom 2000, 153)

Primacy of the propositional An inferentialist pragmatism is committed to a top-down order of semantic explanation. It must give pride of place to propositional contents, for it is expressions with that sort of content that can play the basic inferential roles of premise and conclusion. (Brandom 2000, 40)

These two theses are not directly implicated in Brandom’s argument for the language-dependence of conceptual thought. The key to that argument is the thesis that inferential articulation cannot be understood apart from social and communicative practices of making and assessing claims. Against the widespread thesis that a content’s inferential articulation is fixed by its role in deliberation and practical reasoning, Brandom argues that deliberation and practical reasoning can only be understood in a social and communicative context. Let us call this thesis the primacy of the discursive. Here is a representative passage. Primacy of the discursive The explanation of behavior according to the model of rational agency depends upon treating intentional states as having propositional contents, which involve objective truth conditions. But there is no pattern of moves a single individual might make that would qualify that individual’s states as inferentially articulated in this sense. The inferential practice (including practical reasoning) that confers contents of this kind comprises not only first-person reasoning, but also third-person attributions and assessments of it, and both aspects are essential to it. Deliberation is the internalization of the interpersonal, communicative practice of giving reasons to, and asking reasons of, others, just as judgment is the internalization of a public process of assertion. Inferring cannot be understood apart from asserting. (Brandom 1994, 158)

We can follow Brandom in referring to the practice of giving and asking for reasons as discursive scorekeeping. The argument in this passage for the language-dependence of conceptual thought is essentially that without discursive scorekeeping propositional contents cannot have objective 42

truth-conditions. His reasons for claiming this are, roughly, that in order to conceive of a state of affairs as objective (and so to be able to have propositional contents with objective truth conditions) a thinker must be able to keep track of how that state of affairs might appear from different perspectives, which in turn requires being able to keep track of how that state of affairs might be represented by other people. We keep track of how states of affairs are represented by other people by monitoring the different commitments and entitlements that those people incur when they express those representations. This yields the desired result on the plausible hypothesis that no propositional content without objective truth conditions can be inferentially articulated. One problem with this line of argument is that it leaves Brandom exposed to objections of an obvious kind—namely, that what he claims to be a necessary condition upon being in states with objective truth conditions is not really necessary at all. Even if we grant that being in states with objective truth conditions requires being able to understand how different thinkers might be able to triangulate on a particular state of affairs from different perspectives, how can Brandom be so sure that this can only be done from within the practice of discursive scorekeeping? Certainly, some commentators have expressed their doubts (e.g. Gibbard 1996). There is a very basic methodological difficulty here. Brandom’s position commits him to certain claims about the cognitive capacities of prelinguistic infants and nonhuman animals. Those claims must be empirically adequate, and it is natural to think that their empirical adequacy cannot be determined purely on conceptual grounds.7 But nor, on the other hand, can their empirical adequacy simply be read off the data. There is a plethora of data from experiments that set out to test the cognitive abilities of infants and nonhuman animals. Brandom thinks that he has good reasons for denying that the data should be interpreted in any way that confers genuine intentionality upon the nonlinguistic. Others think that the data offer good reasons for rejecting Brandom’s arguments. Quite plainly, any appeal to empirical data will end in an impasse. The data need to be interpreted and the principles that might govern such interpretation are precisely what are under discussion and at stake. All this suggests that theorists convinced of the language-dependence of conceptual thought would do well to think about other, perhaps less contentious, routes to that conclusion. But can we argue for the language7. See Gibbard (1996), 706f.

43

dependence of conceptual content without proceeding via discursive scorekeeping? This brings us back to the possibility mentioned earlier of arguing for the language-dependence of conceptual content from considerations of structure and vehicle. I will in the next section sketch out a version of this argument, which leads to a conclusion that I think Brandom would find very congenial. However, as we shall see, it is not a line of argument that Brandom can exploit, because of the particular way in which he interprets the principle of the primacy of the propositional. This is the blindspot that I mentioned earlier. I will end the paper by diagnosing the blindspot. 3. Argument 2: Conceptual content and sentential structure As recent discussions of the possibility of nonconceptual content have made clear, there is no agreed upon formulation of what conceptual content is and what it is to possess a concept. For the purposes of this paper I propose to take concept possession, and conceptual thought more generally, to be a fundamentally reflective matter. That is to say, conceptual thinkers have to be capable of reflecting upon their own thoughts. They have to have a reflective grasp of the inferential relations between thoughts, of the evidential relations between perceptions and thoughts, and of the justificatory relations in which thoughts stand to action. This characterization of the realm of the conceptual is intended to be very much in the spirit of Brandom’s own work (while leaving open the possibility of ways of thinking and forms of inference that are not reflective, and hence that will count as nonconceptual). Whether it captures how the word ‘concept’ is used more generally within philosophy is quite another matter, but not (I think) an important one. It is quite plain that there is an important epistemic natural kind here, whatever label one attaches to it. Someone unconvinced that conceptual thought is essentially reflective in this manner can take the following argument to apply only to the higher-order aspects of conceptual thought. If concept possession is understood in this fundamentally reflective way, then it involves higher-order thoughts, where these are thoughts that take other thoughts as their objects. Quine once described semantic ascent as ‘the shift from talking in certain terms to talking about them’ (Quine 1960, 271). By analogy we can characterize intentional ascent as the shift from thinking in certain ways to thinking about those ways of thinking. My argument for the language-dependence of conceptual thought is, in 44

effect, that intentional ascent requires a form of semantic ascent. We can only think about thoughts through thinking about words.8 Since conceptual thought is fundamentally reflective, this yields the desired conclusion that conceptual content is language-dependent. Intentional ascent requires semantic ascent because we have to be able to represent thoughts if we are to think about them. This is where the vehicles of thought enter the picture, but in a very different way from how they enter in the standard arguments for the language of thought hypothesis with which we began. It is quite plain that all sorts of things go on below the threshold of consciousness when we think (and it may well be, as Fodor and others have suggested, that thinking involves manipulating sentences in a subpersonal language of thought, for example). There is a sense in which these subpersonal events are the vehicles of thought. But they are not what we think about when we engage in reflective thinking about our own thoughts. There is a difference between thinking about thoughts and thinking about the machinery of thinking. So, since we have to be able to represent thoughts in order to think about them, the question is: What are the personal-level vehicles of thought that enable us to think about them? There are, I think, only two possibilities. On the one hand, representation might be secured symbolically through the complex symbols of a natural language. A thought would be represented, therefore, through its linguistic expression and would appear as a potential object of thought qua linguistic entity. On the other hand, representation might be secured in an analog manner, through some kind of pictorial model. On this conception of the vehicles of thought, which we find developed in different ways in mental models theory in the psychology of reasoning (Craik 1943, Johnson-Laird 1983), and in the conception of mental maps put forward by Braddon-Mitchell and Jackson (1996), the vehicle of a thought is a pictorial representation of the state of affairs being thought about. The argument in favor of public language sentences and against pictorial models rests upon considerations of structure and inferential role. I am assuming that thoughts are individuated at least in part by their inferential role. What makes a given thought the thought that p is partly a matter of the inferential relations in which it stands to other thoughts. Some of these relations are entailment relations (the thoughts that entail p and the 8. This is not quite semantic ascent in Quine’s sense, as we can think about words without talking about them.

45

thoughts that p entails), but they also include evidential relations (the thoughts whose holding true would be good evidence for thinking that p holds true, and the thoughts that would be judged more likely to be true if p were true). Any thinker capable of thinking a higher-order thought directed at a particular thought must have some grasp of the individuation conditions of the target thought. He must have some grasp of what it is that he is thinking about. This requirement is not in any sense peculiar to higher-order thoughts. It is just an application of the very general requirement that to think about anything one must have some sort of ‘cognitive access’ that enables one to pick that thing out. One must be able to individuate it in thought.9 When one is thinking about things in the world, the requirement applies to whatever it is that one is thinking about. In order to think about an object in the distal environment I must have some sort of epistemic access to that object—epistemic access that will enable me to understand, for example, how the world would have to be for that thought to be true. I do not need to be able to establish whether or not the thought is true, but I certainly need to have some understanding of what would count as evidence for its truth. Similar things apply when I am thinking about a thought. I must have some sort of epistemic access to the thought that is the object of my higher-order thought, because this is what enables me to understand how the world would have to be for my higher-order thought to be true. The requirements upon higher-order thoughts are no different in source or import from the requirements upon lower-order thoughts. In both cases they stem from the requirement that one have epistemic access to what it is that one is thinking about. What is different in the two cases is simply the object of thought. In one case it is a state of affairs in the world. In the other it is a thought about that state of affairs. Nor is the inferential role of a thought different according to whether one is thinking it or thinking about it. If I think that a is F then my thought is partly individuated by its inferential role. But there is no need for me to grasp that inferential role in order to think that thought. I must reason in accordance with it, but my grasp of it can remain entirely implicit and practical. When I think a higher-order thought that has this first-order thought as its object, however, my grasp of the inferential role of the first-order thought has to 9. See Evans (1982) for a sustained working out of this idea, together with a diagnosis (particularly in ch. 3) of the conflations that have led some theorists to abandon it.

46

be explicit (unlike the inferential role of the second-order thought, which can remain implicit, unless that second-order thought becomes the object of a third-order thought—and so on). It follows that a higher-order thinker must have some sort of grasp of the entailment and evidential relations in which the target thought stands. Of these entailment and evidential relations at least some are a function of the structure of the thought. In order to understand the inferential role of a thought we need to be able to view it as made up of distinguishable components that can feature in further thoughts and, moreover, we need to be able to view it as made up from those components in a way that determines its semantic value (thereby capturing the difference between the true thought Zurich is located in Switzerland and the false thought Switzerland is located in Zurich). We may say, therefore, that the structure of the thought must be perspicuous in the consciously accessible representation that is the target of the higher-order thought. The final step in the argument is that the structure of a thought cannot be perspicuous in the right sort of way in thoughts that are represented in a pictorial manner. The qualification is important, since pictorial representation in mental maps and mental models does depend upon a notion of structural isomorphism between the models/maps and what they represent. The relations holding between elements of the mental model/map can be mapped onto the relations holding between objects in the represented state of affairs. This comes across very clearly in the following passage from Braddon-Mitchell and Jackson. There is no natural way of dividing a map at its truth-assessable representational joints. Each part of a map contributes to the representational content of the whole map, in the sense that had that part of the map been different, the representational content of the whole would have been different. Change the bit of the map of the United States between New York and Boston, and you change systematically what the map says. This is part of what makes it true that the map is structured. However, there is no preferred way of dividing the map into basic representational units. There are many jigsaw puzzles you might make out of the map, but no single one would have a claim to have pieces that were all and only the most basic units (Braddon-Mitchell and Jackson 1996, 171).

We might gloss this as follows. Pictorial representations do not have a canonical structure. Their structure can be analyzed in many different ways (corresponding to the jigsaw puzzles that one can construct from it), but 47

none of these can properly be described as giving the structure of the state of affairs. Yet, in order to understand the inferential role of a thought one does need to understand the canonical structure of that thought (what is often termed its logical form). This canonical structure is perspicuous, although not always perfectly perspicuous, when thoughts are expressed in public language sentences. It is because of this that higher-order thought is language-dependent. Only public language sentences can make the canonical structure of a thought available to thinkers in a way that allows them to grasp its inferential role. The conclusion of the argument, then, is that thinking about thinking is only available to language-using creatures. Since we are understanding conceptual thought to be fundamentally reflective, this yields the language-dependence of conceptual thought. 4. Brandom on the inferential role of sentences The argument canvassed in the previous section reaches a general conclusion about the relation between conceptual thought and language that is similar in spirit to Brandom’s. But it is not a line of argument that Brandom is likely to endorse. He would reject it, I think, because of the particular way in which he thinks about the principle of the primacy of the propositional. My argument rests upon a way of explaining the inferential role of thoughts and sentences that he would think inverts the true order of explanation and is ultimately incompatible with his version of pragmatism. The argument in the previous section rests upon the claim that the inferential role of a thought is to be explained in terms of its structure. It is because we understand the structure of a thought through understanding the structure of the sentence that expresses it that we can only think about thoughts through thinking about language. Brandom, in contrast, is emphatic that, while the inferential role of thoughts can only be understood through the inferential role of sentences, the inferential role of sentences is fixed at the sentential rather than the subsentential level. Accordingly, a grasp of subsentential structure is not required for a grasp of the inferential role of the sentence, and hence of the corresponding thought. In fact, Brandom holds that our understanding of subsentential structure is subordinate to, and derived from, our understanding of the inferential role of sentences. 48

We have already seen one passage illustrating Brandom’s principle of the primacy of the propositional. Here is another that is even clearer about how he sees the order of explanation: The sort of conceptual content that is expressed by whole declarative sentences is prior in the order of explanation to the sort of content that is expressed by subsentential expressions such as singular terms and predicates. (Brandom 2000, 12f.)

The order of explanation is in this direction, he argues, because of how he thinks that conceptual content is fixed. The conceptual content of a sentence is fixed by the discursive practices within which it features. It is fixed by the linguistic performances to which it commits a speaker who utters it; by the linguistic performances that it entitles its utterer to make; and by the linguistic performances that it rules out.10 There remains an unargued step, however. One might agree with Brandom that the conceptual content of a sentence is fixed by what follows from endorsing it and what should lead one to endorse it while still thinking that these facts should not be taken as primitives. Why are these relations not themselves to be explained in terms of the sentence’s subsentential structure? Surely, one might think (and many have thought), a sentence’s entitlements and commitments are dictated by how it is built up. But Brandom cannot agree, given his views about the explanatory priority of the sentential over the subsentential. At this point we need to turn back to Brandom’s distinction between logical articulation and inferential articulation. As we saw earlier this is closely connected to the distinction between material validity and formal validity. For Brandom formal logic makes explicit inferences that are materially correct (i.e. that preserve truth as a function of the propositional contents of the relevant sentences, rather than of their logical structure). Formal validity is to be explained in terms of material validity, and there is no prospect of explaining material validity in terms of formal validity. This is because we need the notion of material validity to explain what counts as formal validity. An argument is formally valid just if it is materially valid in virtue of its logical structure. In order to identify an inference as formally valid we need, first, to know that it is materially valid and, second, to identify which elements of it are to count as specifically logical (see Brandom 1994, 104f. for more details). 10. See Brandom (1994), ch. 3 for the full story of how semantics is to be given in terms of discursive commitment.

49

Given that he takes material validity to be fundamental in this way, it is not surprising that Brandom takes the inferential role of a sentence to be a matter of material inference rather than formal inference. Once again, though, this view is perfectly compatible with the thesis that the entitlements, commitments, and incompatibilities that come with a particular sentence are a function of how that sentence is built up. But there is an additional component to Brandom’s position, which does rule out this widely held thesis. This additional component is the claim that a sentence’s material inferential properties are conferred upon it by discursive social practices. When a linguistic community engages in practices of giving and asking for reasons this should be seen as a process of practically assessing inferences as correct or incorrect. This process of practical assessment has the effect of ‘instituting material inferential properties that confer propositional conceptual content on their states and performances’ (Brandom 2000, 136). So, the material inferential properties of sentences are derived from what linguistic communities do with those sentences, rather than from their internal, compositional structure. In the last analysis, this is why Brandom has to hold that considerations of vehicle and structure should not be central to our thinking about conceptual content, and why he would not welcome the argument that I have proposed. Of course, there is a place in Brandom’s thinking for subsentential structure and for what he calls representational (as opposed to propositional) content. Representational content comes to the fore when we engage in de re ascriptions of propositional attitudes, when we try to make sense of the important distinction between what a speaker is committed to, and what that speaker is entitled to. We need representational content and the subsentential articulation that it brings with it when we need to think about other people and what they are saying and thinking (see Brandom 1994, ch. 8 and 2000, ch. 5). We do not, however, need it either to think or to think about thinking. 5. Inferential articulation and logical articulation Let me end by stating why I think that Brandom is mistaken to put considerations of vehicle and structure to one side in thinking about conceptual content. As I mentioned earlier, there is a blindspot here. The source of the blindspot is, I think, the particular way in which Brandom formulates the contrast between material validity and logical validity. Let me explain. 50

The view from which Brandom wants to distance himself holds that inferential articulation is logical articulation. On this view, which we can call the logical articulation view, the inferential properties of a sentence fall into two groups. The first group is given by the inferential relations into which it can enter as a function of its logical structure. So, to take a very simple example, if we are entitled to assert a compound sentence whose principal connective is the conjunction operator, the rule of conjunction elimination confers on us warrant to assert both of its conjuncts. These inferential properties have nothing to do with the sentence’s nonlogical elements. We can shift and change all of these without altering them. But of course, even the logical articulation view has to accept that certain inferential properties are content-specific. We can certainly infer ‘Socrates is mortal’ from ‘Socrates is a man’. But all such inferences are enthymematic. The missing premise is a universal quantification about the mortality of men. It is hard not to agree with Brandom that the enthymematic view of content-specific inferences is hopeless. The universal quantification about the mortality of men is a codification of our inferential practices, not a license for them. The licensing works in the opposite direction, in a way that exploits conceptual connections between the concept man and the concept mortality. And Brandom may well be right that these conceptual connections are ultimately fixed by the discursive practices of speech communities. He may well even be right that this extends to the first type of inferential properties, those that exploit formal structure. After all, we can view these formal inferences as underwritten by the meanings of the logical constants and it can plausibly be argued that the formal rules governing the behavior of logical constants are codifications of common patterns among inferences that we intuitively take to be valid (where ‘we’ refers to the members of our speech community).11 The question, though, is why all this should be incompatible with the thesis that sentences stand in inferential relations in virtue of their structure. 11. A view along these lines has been proposed by Jonathan Lowe (see, e.g., his 2002). Lowe quotes a witty passage from Locke’s Essay: ‘God has not been so sparing to Men to make them barely two-legged Creatures, and left it to Aristotle to make them Rational. … He has given them a Mind that can reason without being instructed in the Methods of Syllogizing: The Understanding is not taught to reason by these Rules; it has a native Faculty to perceive the Coherence, or Incoherence of its Ideas, and can range them right, without any such perplexing Repetitions’. I imagine that Brandom would find the spirit of this passage congenial, if not the appeal to native inferential faculties.

51

It is perfectly coherent to hold both that material validity is more fundamental than formal validity and that certain inferences are materially valid for structural reasons. A sentence can have inferential properties in virtue of its structure without those inferential properties having to be formal. Here is an obvious example. The inference from the sentence ‘Socrates is mortal’ to the sentence ‘Something is mortal’ is materially valid. Anyone who understands both sentences can see that the first cannot be true without the second being true. And I am sure that Brandom is correct that this should not be understood in terms of tacitly applying a rule of existential generalization. Nonetheless, the material validity of the inference does seem quite clearly to depend upon the structure of the sentence ‘Socrates is mortal’. It is because ‘Socrates’ is a proper name that ‘completes’ the function-expression ‘– is mortal’ that we can infer that something is mortal from the sentence ‘Socrates is mortal’. It is because we see that the sentence has this structure that we see that the inference cannot fail to preserve truth. And, one might continue, it is because we become sensitive to structural similarities between truth-preserving inferences of this type that we come to formulate some version of the rule of existential generalization. We can hold that inferences are materially valid in virtue of their structure without collapsing material validity into formal validity. If Brandom were to take this point on board then it would be open to him to adopt the type of argument that I sketched out in section 3. Grasping a thought’s inferential properties would have to involve grasping its structure. This would open the door to the argument from intentional ascent to semantic ascent. I don’t think that by doing so he would be in any sense compromising either the principle of the primacy of material validity or (a plausible version of ) the principle of the primacy of the propositional. And he would gain the benefit of having a new way of arguing for the principle of the primacy of the discursive—one that avoids the methodological difficulties identified in the second section. He would certainly have to make some adjustments to his formulations of the principle of the primacy of the propositional (and, for that matter, to his interpretation of Frege12), but one might well already have independent reasons for thinking that he needs to make those adjustments. All those reasons will, in the last analysis, stem from the impossibility of explaining how one 12. Brandom is taking a very controversial stand on a live debate in the interpretation of Frege. For further details and references see Bermúdez (2001).

52

could possibly understand a sentence, or grasp a thought, except in virtue of understanding how it is built up from subsentential components, or the corresponding elements at the level of thought. Brandom does not, of course, think that it is impossible to give such an explanation. But nor, on the other hand, does he offer one, or explain how to go about providing it. His conviction that there must be such an explanation is, I conjecture, what explains his version of what I am calling the blindspot, namely, the view that questions of vehicle and structure are orthogonal to questions about personal-level inferential abilities.13

REFERENCES Bermúdez, José Luis. 1995. “Nonconceptual content: From perceptual experience to subpersonal computational states”. Mind & Language 10(4), 333–369. — 2001. “Frege on thoughts and their structure”. Philosophiegeschichte und Logische Analyse 4, 87–105. — 2003. Thinking without Words. New York: Oxford University Press. — 2005. Philosophy of Psychology: A Contemporary Introduction. London: Routledge. Brandom, Robert. 1994. Making It Explicit. Cambridge (Mass.): Harvard University Press. — 2000. Articulating Reasons: An Introduction to Inferentialism. Cambridge (Mass.): Harvard University Press. Evans, Gareth. 1982. The Varieties of Reference. Oxford: Oxford University Press. Fodor, Jerry. 1975. The Language of Thought. Cambridge (Mass.): Harvard University Press. Gibbard, Allan. 1996. “Thought, norms, and discursive practice: Commentary on Robert Brandom ‘Making It Explicit’”. Philosophy and Phenomenological Research 56(3), 699–717. Gregory, Richard L. 1970. The Intelligent Eye. London: Weidenfeld and Nicholson. Lowe, E. J. 2002. “The rational and the real: Some doubts about the program of ‘rational analysis’”. In: José Luis Bermúdez and Alan Millar (eds.). Reason and Nature. Oxford: Oxford University Press. 13. This paper was written for a workshop entitled ‘No Language; No Concepts?’ at the University of Zurich on 17–18 March 2007. I am grateful to Hans-Johann Glock and the other organizers for making this event possible; and to participants at the workshop for their comments and questions.

53

McDowell, John. 1994. Mind and World. Cambridge (Mass.): Harvard University Press. Prinz, Jesse. 2002. Furnishing the Mind: Concepts and their Perceptual Basis. Cambridge (Mass.): MIT Press. Sellars, Wilfrid. 1953. “Inference and meaning”. Mind 62, 313–358.

54

Grazer Philosophische Studien 81 (2010), 55–75.

THINKING WITHOUT LANGUAGE. A PHENOMENOLOGICAL ARGUMENT FOR ITS POSSIBILITY AND EXISTENCE Martine NIDA-RÜMELIN University of Fribourg

Summary The view is defended that the mere lack of language in a creature does not justify doubts about its capacity for genuine and complex thinking. Thinking is understood as a mental occurrent activity that belongs to phenomenal consciousness. Specific kinds of thinking are characterized by active or passive attending to the contents present to the subject, by the thinking being goal-directed, guided by standards of rationality or other standards of adequacy, and finally by being a case of critical reflection upon one’s own thinking. It is argued that none of these properties of thinking introduce the necessity that the thinking subject has a language except for, probably, the last one. There is reason to believe that the capacity to critically reflect upon one’s own thought requires internal verbalization of the thoughts being criticized. The view that emerges is that we might share larger parts of our cognitive phenomenally conscious life with non-linguistic creatures than is commonly assumed.

1. Introduction1 Is it possible to think without language? The question is ambiguous. It might be read like this: (Q1) Is it possible for a creature without a language to think? (Can non-linguistic creatures think?) While lying awake on the floor, a dog’s changing facial expression and its eye movements may make it appear obvious that something is going on 1. I would like to thank an anonymous referee for his or her very helpful critical remarks that motivated a number of modifications.

‘in its mind’. The dog seems to think. But does it think and if so, in what sense? Or imagine a small child not yet capable of using a language. The child runs towards an object, suddenly encounters an obstacle, and stops. She looks around as if searching for alternative approaches. She hesitates, then runs around the table avoiding the obstacle, and finally reaches the desired object. Does the child think? In both cases the answer is, in my view, quite clearly positive. Non-linguistic creatures (small children and higher animals) are capable of thinking. In what follows, I will motivate this view. I will use ‘thinking’ in that context to designate occurrent mental processes. In thinking, certain contents are phenomenally present to the thinking subject. I take phenomenal presence of contents as a substantial common feature of all cases of occurrent thinking. I presuppose in what follows that there is something it is like to think, and that in thinking, certain contents are present to the thinking subject. The question of whether or not it is possible to think without language has a second, quite different reading: (Q2) Is it possible to think without thereby using a language? In many cases we use language in our thinking even though we do not speak aloud. We then internally verbalize our thoughts. In verbalized thinking, thoughts are developed using specific sentences. One might wonder what exactly internal verbalization consists in and how its specific phenomenology may be adequately described. One might suspect that internal verbalization has something to do with auditory imagination. According to a simple proposal, internal verbalization is nothing but imagining hearing sentences with understanding. But a little reflection shows that this is not right. One may imagine hearing a sentence with understanding and yet not thereby use the sentence to internally express one’s thought. Internal use of language in thought is different from imagining someone talking, and it is different from imagining hearing a voice where the author is underdetermined in imagination. Also, internal verbalization is not non-verbal thinking plus auditory imagination of an utterance of sentences. Rather, the thought is developed in verbalized thinking by ‘internally producing sentences’. Maybe thinking is just like speaking without making noises. But what is it to speak without making noises? One might be tempted at this point to come back to the idea that thinking has something to do with auditory imagination, and one might propose that thinking in a verbalized way is imagining oneself 56

speaking. But this description does not seem to capture the phenomenology of verbalized thought either. It is one thing to imagine oneself saying something and it is quite another to use the sentences at issue to ‘silently express’ and thereby develop one’s thought.2 To give an adequate description (if this is at all possible) of what it is like to think in an internally verbalized way is not a trivial task. However, each of us is familiar with the phenomenon and thereby has—for present purposes—a sufficiently clear understanding of what is meant by ‘internal verbalization’ (or ‘internal use of language in thought’ or ‘internally verbalized thought’). A point of clarification: ‘internally verbalized thinking’ is meant to refer to a kind of thinking characterized by its specific phenomenology. There is something it is like to think internally using language. The way it is like to think internally using language is different from the way it is like to think a similar content without internally using language. Some may doubt the existence of thinking without internal verbalization. But it seems to me that a bit of phenomenology clearly suggests that a great part of our thinking is non-verbalized. Examples will be discussed below. As is obvious from what precedes, I will defend a positive answer to the second question as well. There are many cases of thinking, I claim, where no internal use of language is involved.3 A negative answer to the second question would of course require a negative answer to the first question. If thinking was in general impossible without internal verbalization, then obviously no non-linguistic creature could think. But a negative answer to the first question does not imply a negative answer to the second. One might defend the view that non-linguistic creatures are incapable of thinking although linguistic creatures can have non-verbalized and even non-verbalizable thoughts. According to a view of this kind, the capacity to think non-verbally is in both cases based on the capacity to use a language. The view defended in the present paper is different: some cases of non-verbalized human thought require the possession of concepts that cannot be acquired by a non-linguistic creature. But there are also kinds of non-verbalized human thinking that do not require linguistically based concepts in that sense. There is no reason to exclude the possibility that those kinds of thinking occur in non-human, non-linguistic beings. 2. See Siewert (1998), ch. 8, for detailed and rich reflection upon the phenomenology of thinking. 3. ‘Internally verbalized thought’, ‘thinking using language’ and similar locutions will be used synonymously in what follows.

57

2. Thinking in the broadest sense Sometimes the question of whether a given individual is capable of having thoughts is equated with the question of whether the individual has beliefs, desires and other propositional attitudes (compare for instance Ludwig 2004). This is not the issue I will enquire. My interest is rather in thinking as a conscious occurrent activity. When you lie on your bed awake, leaving yourself to whatever comes up ‘in your mind’, then you are thinking in the broad sense I am concerned with. You may be just thinking without any goal, not searching for a solution to any problem, nor trying to form a justified opinion, nor deliberating about what to do. You may just be lying there, letting your thoughts wander. In this situation memories may briefly pop up, you may attend to them and stay for a while with the remembered period or situation, you may actively or just passively change the topic. You may anticipate in imagination some part of the rest of the day, you may imagine a conversation with a friend, you may try to remember a tune you heard in a concert last night; suddenly the thought may pop up that you should go out and buy some milk for the next day. In the next moment you might be taken by fantasies about standing at the beach, feeling the wind and listening to the waves, it may occur to you that you haven’t been at the beach for a long time. All these ‘contents’ may come up very briefly and may be quite fragmented, and they may fade away before you begin to attend to them more closely. You may be more or less actively involved. You may actively try to remember more clearly a certain detail but you can also passively leave yourself to whatever occurs in your ‘stream of consciousness’. The example illustrates a very broad sense of thinking. It is a sense of thinking that includes occurrent memories, imagination, non-verbal as well as verbal reflection about a theoretical issue, deliberation, and many other mental processes. All these processes have substantial features in common. In all these cases, ‘contents’ in a broad sense to be explained in a moment surface in consciousness. It seems safe to assume—on the basis of empirical knowledge as well as phenomenological observation—that these contents surface in consciousness as a causal consequence of some unconscious happenings in the brain that must be described as a kind of information processing. The subject often has indirect causal influence on the sequence of contents that occur. When you try, for instance, to remember a detail about your childhood, you will enhance the probability that memories surrounding the detail will occur. When you try to get clear 58

about a philosophical problem, you will enhance the probability that some relevant insight will occur in your mind. By directing one’s attention to a given content (a theoretical idea, an imagined melody, a possible future event), one may cause it to remain present in mind. But we may not succeed in directing our thought in the way we like. Voluntary control is limited in all cases of thinking in the broad sense. A substantial common feature of thinking in the broad sense is that contents following each other are phenomenally present to the thinking subject. The way these contents are phenomenally given is of course different in imagination, in memory or in abstract thought.4 Thinking in the sense at issue requires an experiencing subject. Thinking in that sense is a kind of experiencing. It is a case of phenomenal consciousness. Information processing in a complex machine without phenomenal consciousness is not a case of thinking in that sense. Subconscious processes that lead to the popping up of certain contents are not themselves a case of thinking. It is a necessary feature of thinking in the sense at issue that the relevant contents are phenomenally present to the thinking subject.5 One might suspect that using the term ‘thinking’ in the way here proposed, i.e., to include daydreaming, a-rational associative thinking, imagination and the like, is to broaden its sense in an unnecessary, arbitrary and artificial way. And one might think that—once the term is used in this very wide sense—the result that thinking without language is possible becomes an almost trivial and in any case philosophically uninteresting thesis. But there is, I claim, good theoretical motivation to do so. All other kinds of thinking and in particular rational reflective thinking are sub-species characterized by additional conditions that may or may not obtain in a specific moment of a given individual’s mental life. The basic and quite amazing natural phenomenon is the phenomenon of thinking in the broadest sense: being phenomenally aware of contents that occur in one’s mind relatively independently of what is going on in one’s environment and in a way that is not—like perceptual experience—simply triggered by inputs from the external world. A more specific kind of thinking is characterized 4. The above description does not exclude that some forms of thinking, e.g. rational deliberation, involve genuine activities of the thinker which are not causally determined by neurological mechanisms. 5. One might also take the following terminological decision: thinking is the whole complex process that involves phenomenally present contents and unconscious information processing as well. As long as it is required that the subject is consciously presented with a part of the relevant contents, this definition seems equally acceptable to me.

59

by being attentive to the contents that occur in one’s mind. Still more specific cases of thinking are specified adding the further constraint that the activity of thinking is goal-directed. A special case of goal-directed thinking is rational thought. A still more specific case is given when the thinker can make her own thinking the object of critical evaluation. Each additional condition (attending, pursuing a goal, being guided by principles of rationality, critical reflection) is a major achievement of mental evolution in nature. But still, even the most specific and complex kind of thinking is just a sub-species of the fundamental phenomenon of being phenomenally aware of contents that occur ‘in one’s mind’ on the basis of internal information processing. 3. Contents I have been using the term ‘content’ in a non-technical and very wide sense. A content of thought in the sense at issue is something we may attend to and something that may ‘pop up in our mind’. A content is something we can be phenomenally aware of. A content in the relevant sense can be present to the subject in an episode of thinking. Examples of content in this sense are of very different kinds: a philosophical issue, a vague insight about life in general, a possible future action, the visual ‘picture’ of some scene, a melody, a scene experienced in the past, a vague idea of something one dreamt of, etc. All these are ‘contents’ we may be consciously presented with in a given moment, and that may in this sense be the object of our thought. In each case there is something it is like to think the specific content under the specific mode at issue.6 For present purposes there is no need to clarify the ontological status of contents. For some cases one might argue that the content is nothing but the real thing the thought is about. But there is also reason to think that contents have to be individuated by the way they are given to the subject. Also, in some cases there is a content, but no thing the thought is about. The use of the term ‘content’ should not be thought of as carrying with it any theoretical commitment. No theoretical account of contents is intended to be presupposed. Nor is it assumed that all phenomena here 6. For an elaborate and interesting argument to the effect that ‘non-iconic’ thought (thought without sensory imagery) is a case of phenomenal consciousness see Siewert (1998), ch. 8. Siewert argues in detail for the view that phenomenal character determines intentional content in terms of veridicality conditions. I do not need to presuppose the latter claim for my present purposes.

60

classified as ‘content’ in a broad sense are of a homogenous nature and can be accounted for by some unique theoretical approach. The contents that pop up in our mind do not come in a clearly individuated manner. There is in general no answer to the bizarre question: how many contents have you been aware of between 8.00 and 8.05 this morning? Also, whether or not a given content is phenomenally present to the mind is not an all-or-nothing matter. It appears phenomenologically more adequate to say that awareness of contents in one’s thinking may fade in and out without a clear-cut limit between being present and being absent. I do not wish to exclude the possibility that some contents of thinking in the broad sense may be captured in terms of veridicality or truth-conditions. But I wish to allow for contents that cannot be captured in this way. One and the same content can be an object of thinking in the broad sense and also the content of a different mental process that is not a case of thinking. A melody, for instance, can be the content of an auditory perception that is not a thinking, but also the content of thinking when a composer considers different ways to continue a composition. 4. Attending to contents In most cases of thinking in the broad sense, active and passive elements are intertwined in a complex manner. A part of a scene dreamt of the preceding night sometimes surfaces to consciousness during the day (passive element). The scene may capture one’s attention (passively) and its interesting details may motivate the subject to actively direct its attention towards some particular aspect. The activity of the subject in directing its intention may trigger further parts of the dream to surface. A philosopher may suddenly be struck by a new thought about freedom and determination (passive element) and—impressed by the apparent quality of the thought—she may then seriously consider (active element) the new idea. By considering the idea she will cause the occurrence of further ideas about the issue. This complex interconnection of passive and active elements is a common feature of cases of a-rational thinking without any goal and of the most sophisticated cases of thinking like the reasoning of a scientist. In both cases attending is substantially involved. Attention may be actively directed towards a certain content and will thereby actively influence what comes up later, or it may be passively caught by 61

some content. Active direction of attention does not require deliberation or conscious decision. We may actively direct our attention without taking a decision to do so and, of course, we do not normally choose—while thinking —between different alternative contents we might attend to. In some cases, however, a conscious decision about what to attend to may occur, for instance when a mathematician decides to enquire about one potential proof rather than another. What we attend to in thinking is only partially subject to voluntary control. As anyone knows who ever woke up in the middle of the night and was incapable to fall asleep again for a while, our attention may be attracted to and captured by certain contents of thought against our will. You may ask how one can claim these things without empirical investigation. I take it that these observations are quite obvious results of a bit of phenomenological reflection. Any reader, I am confident, will be able to find the passive and active elements I mentioned and the complex interrelation between them in his or her own experience of thinking. We have an intuitive grasp of the difference between active and passive. I moreover assume that these terms are capable of picking out two different kinds of mental events in a successful manner. However, to explain the nature of the difference between active and passive cases of e.g. attending is a substantial issue that will not be investigated in the present paper.7 Attending to contents is typical for most cases of thinking. It may therefore be tempting to characterize thinking in the broadest sense by the following description: the subject attends to contents that are phenomenally present where these contents occur as a result of inner information processing, which is quite independent of external stimuli. But attending is not a necessary feature of the general natural kind of mental processes one should have in mind. Contents may be phenomenally present and may develop and follow each other just like in other, more specific cases of thinking, even when they remain ‘in the background of the mind’ because, for instance, the subject’s attention is completely caught by something else, e.g., by observing the traffic while driving a car. It seems arbitrary to exclude cases of phenomenal presence of contents that ‘flow in the background’ from genuine thinking in the broadest sense.8 7. M. Nida-Rümelin, 2006. 8. I here assume without argument that in general as well as in the case of thinking phenomenal presence of a content does not require attending to the content. The general claim is illustrated by the example of a headache that one may not attend to and that one may not be consciously aware of while being involved in intense working, but that one later remembers as

62

5. Goal-directed thinking Much work in philosophy is devoted to issues about the rationality of the formation of belief (theoretical rationality) and the rationality of choices between alternative ways to act (practical rationality). In large part, the formation of belief happens without conscious awareness. We find ourselves being convinced of many things we have never consciously thought about. Analogously, many of our practical choices are taken without previous conscious deliberation. Unconscious formation of belief and unconscious processes leading to a practical choice are not cases of thinking in the present sense. But it does happen—of course—that we engage in conscious thinking with the goal of forming a reasonable opinion. The goal may be quite specific, as when, for instance, in a situation with insufficient information, we reason about whether a friend will arrive at the station at 10.05 or rather at 11.05. We may be engaged in estimating the time it takes to leave the plane, to go through passport control, to pick up the suitcases and to reach the railway station. In this case we will be involved in imaginative activities and calculation with the specific goal of forming a well-founded opinion about the most likely time of arrival. In other cases, the goal of thinking is much less specific, quite general and somehow vague, as it is for instance in the case of a philosopher who reflects upon phenomenal consciousness with a view to developing a better understanding, but without thereby considering any specific argument or thesis. Similarly, the goal of taking a decision about taking an umbrella can guide our thoughts when looking out of the window. We may look at the sky and estimate the likelihood of rain. Simultaneously the thought might occur that we will probably lose the nice umbrella in case the day will be dry. As both examples may illustrate, the goal of thinking will have implications for what we attend to in thinking, and it will have consequences for what will pop up—involuntarily—in our mind. The goal of thinking can be more or less present to the thinker. The thinker may have taken a conscious decision to reflect upon a certain question and may be aware of that goal during the whole process of thinking. But it also happens that we find ourselves thinking about something, for instance about a nasty remark made by a friend, trying to understand her motivation without ever having decided to do so; and we may even be quite having been present all along. The example of unattended but phenomenally present thought while driving might be supported in a similar way: after arriving, the driver might remember what has been present in the background of his mind.

63

unaware of the fact that this is precisely what we are doing. Emotional involvement leads to cases of this kind. In cases of goal-directed thinking, the goal guides what we think. But that guidance by a goal does not in general consist in conscious and voluntary control by a subject trying to reach a certain result. One’s thinking may be guided by a goal even in cases where there is no simultaneous awareness of the goal that guides our thinking. The goal that guides someone’s thinking need not presently motivate the subject to exercise any voluntary control over her thoughts either. The goals of forming a reasonable epistemic attitude or of taking a justified decision are only special cases of goals that we may have in our thinking. Other examples of goals of thinking can, for instance, be found in creative activities. Suppose a writer has taken the decision to locate his new novel in the city of Naples. He may in his imagination wander through the noisy small streets near the harbour, imagining the typical melodies and rhythms of Neapolitan voices and imagining the mixture of the wonderful smell of a pizza margherita with the horrible smell of rubbish lying around in a corner. These imaginations may be accompanied by verbalized or non-verbalized thoughts about Naples’ specific political and social situation and the sad role of the Camorra. The goal that guides thinking in that case is not to make up one’s mind about a question or to take a decision. Although the goal is thus of a different kind, the way the goal guides thinking is quite similar to the way in which, e.g., the goal of forming an opinion guides one’s thought. The goal is relevant to what the writer will attend to and it influences the direction his thoughts will take. Given the goal of developing a story, different contents will surface from those that would have surfaced if he were simply daydreaming about Naples.9 6. Rational thinking Goal-directed thinking can be guided by principles of rationality. This normally is the case when the goal of thinking is the formation of a belief or a choice between practical alternatives; at least it should be. To be guided by rationality is to be guided by normative principles of adequacy that may or may not be consciously present or consciously accessible to 9. This is of course an empirical assumption about a causal dependence that is supported by every day life experience and is likely to be confirmable by empirical investigation.

64

the thinker. Can we characterize those kinds of thinking that should be guided by principles of rationality by the goals the thinking is directed towards? One might think that the two goals already mentioned (formation of belief, practical choice) are, taken together, a plausible candidate. But the resulting kind of thinking is too narrow. One at least needs to add all other epistemic attitudes one might take towards a propositional content (doubting, withholding opinion, accepting, attributing a probability, etc.). A similar generalization might be necessary for the practical case. But still, there may be further kinds of thinking that are subject to normative principles of rationality. Thinking directed at the formation of rational emotion (if there is such a thing) might be a further candidate. Some goals of thinking require that any thinking directed towards that goal conforms to rationality, but other goals of thinking do not have that property. Let us call the former goals ‘rationality-specific goals’. Is thinking that is directed towards rationality-specific goals the only kind of thinking guided by normative or evaluative standards or standards of adequacy? It does not seem so. Consider the musical thinking of a composer while composing. She thinks about different possible ways of continuing the piece, she attends to how they fit into the structure of the whole piece, in imaging two simultaneous lines she attends to the resulting series of intervals, etc. Her musical thinking is directed towards the goal of creating a good piece of music. It is plausible to assume that there are evaluative principles that guide her thinking. Other cases of thinking that are subject to principles of ‘adequacy’ though not to principles of rationality may be found in imaginative thinking about a practical problem. Consider the game designed for small children where objects of different shapes must be put into wholes of corresponding shapes. Children playing the game acquire imaginative capacities that make it possible for them to judge just by thinking whether and how a given object fits into a given whole. This is a case of goal-directed thinking that is guided by a-rational standards of adequacy. What is it for thoughts to be guided by standards or principles? It is not necessary that the thinking subject consciously tries to follow the standards. It is not necessary that the subject has some consciously accessible conception of the standards that guide her thinking. To develop a positive account of what it is to be guided by standards in one’s occurrent thinking is, however, beyond the scope of this paper.

65

7. Reflective thinking Critical reflective thought involves thinking about one’s own thinking, as it is often present in philosophical reflection. A philosopher not only reasons about a topic but also reasons about his or her own reasoning about the topic. Having arrived at a conclusion by a particular chain of thoughts, she may make that reasoning the object of critical evaluation: what were my premises? Is each of them well-founded? What are the logical principles applied in my deduction? Are they valid? Reflective thought sometimes occurs in daily life. When a person finds out to her surprise that things are quite different from the way she took them to be, then she may be interested in understanding where exactly she made a mistake. This may motivate reflection upon how she arrived at her opinion in the first place. It should be noted that not every case of self-criticism is the result of reflective thinking. Counter-evidence may motivate a subject to correct herself without making her own thinking the object of her thought. You may, for instance, consciously judge that it is raining outside on the basis of hearing noises and then see when looking out of the window that this is not the case. You will take your judgement back without any thought directed upon your former thinking. So self-correction does not require reflection. But reflection can lead to a particularly interesting kind of self-criticism. Reflection can reveal to the subject why and how she went wrong, and can thereby help to avoid similar mistakes in the future. What are the kinds of thought that can be the object of critical reflection? Any episode of goal-directed rational thinking can become the object of critical reflection. Can other episodes of thinking become the object of critical reflection? What is at issue here is not just reflection in the sense of having a second-order thought that has a first-order thought as its object, and it is not just making one’s own thinking the object of one’s awareness. What is at issue is critical reflection that involves evaluating one’s own thoughts in retrospect. Critical reflection is likely to be restricted to goaldirected thinking that is guided by standards of adequacy. Where thinking does not have a goal, there does not seem to be much point in critical reflection. And where thinking is not guided by standards of adequacy there does not seem to be a basis for criticism. Is there any goal-directed a-rational thinking guided by principles of adequacy that can be subject to critical reflection? Maybe the musical example can motivate a positive response. Maybe it is possible to reflect upon one’s own musical thoughts, thereby evaluating their quality in a critical manner. But a problem seems 66

to be that it is not clear what constitutes the difference between just having musical thoughts and thereby critically evaluating their contents, and making the musical thoughts themselves the object of evaluation. Maybe the case of fitting objects into a whole on the basis of imaginative thinking can provide a better example. A subject who realizes that she made a mistake may rehearse the steps that led to the mistaken result, searching for the point where she went wrong. But again a similar doubt may arise: what is the difference between going through the same thoughts once again and critically reflecting about them? I am not sure what the difference consists in. One might be led to the following hypothesis: non-verbal thoughts cannot be critically reflected upon without first verbalizing their content. Without verbalization, what comes closest to critical reflection does not amount to thinking about a thought but collapses into reconsidering the content. If this is correct, then critical reflection in the rational as well as in the a-rational case requires verbalization. 8. Thinking and language Let me come back to the second question asked at the very beginning. Is it possible to think without internal verbalization? For some cases of thinking in the broad sense the answer is quite obviously positive. We can surely rehearse memories, anticipate future action, imagine walking through a town and critically consider a possible continuation of a piece of music without verbalizing those memories, anticipations, imaginations and musical judgements. Consider the question more closely for imagining walking through a town. Surely an imagination of this kind may be accompanied and controlled by internally verbalizing sentences like ‘now I turn left, now I look through a window and see people eating some pizza’, etc. But these sentences do not verbalize the contents of the imagination. Furthermore, verbalizing sentences to guide one’s imagination is an effective technique, but not a necessary feature of imagination. So, certain kinds of thinking in the broadest sense do not require internal verbalization. More specific cases of thinking are characterized by the property that the thinker actively or passively attends to the contents of thought. Once it is accepted that we can think in the broadest sense without internal verbalization, there does not seem to be any reason to assume that passive or active direction of attention to the contents of thought require internal verbalization. Our attention can be captured by 67

some detail of a scene that occurs ‘in our mind’ and we may actively attend to that detail without internal verbalization. Do things change when we add the further condition of being directed by a goal? Does all goal-directed thinking require internal verbalization? Counterexamples to a positive answer are easy to find: imaginative thinking with the goal of solving a practical problem, like the one of fitting an object of a certain shape through a whole. We can think about and evaluate different options of moving the object without verbalizing the different possibilities. Highly complex practical thinking as the one that may occur while repairing a complicated machine surely is possible without internal verbalization, although internally verbalizing comments may help to control the process. The case of the writer who thinks about Naples with the goal of developing a story is a further example where no internal verbalization needs to be involved. The composer, of course, does not need to and cannot possibly verbalize her musical thinking. So adding the property of being goal-directed does not make thinking an activity that requires the use of language. What about thinking that is directed towards rationality-specific goals? Does all thinking of this kind require verbalization? The reader is invited to investigate the issue through phenomenological reflection. It seems quite clear to me that great parts of thinking with the goal of getting a better understanding, e.g., of a philosophical problem take place without internal verbalization. One may consider a way to go, e.g., in an argument, without internally using words; one may come to see some interdependence between different problems without thereby having any specific verbalization of the connection in mind. One may even come to see a possible solution without yet being able to capture the insight in words. The observation that no internal verbalization is required is even more obvious, it seems to me, in the case of mathematical thinking. Once one has acquired the relevant concepts, it is easy to see, e.g., that two different bases of one and the same vector space have the same number of elements. Seeing this can be the result of non-verbal mathematical reflection. It is difficult to describe the phenomenology of mathematical thinking; it is similar to visual imagination without being visual. Many cases of mathematical conscious thinking are, it seems to me, imaginative, non-visual and non-verbalized. So, there are clear cases of goal-directed a-rational and rational thinking that do not require internal verbalization. But certainly some kinds of non-verbal thinking require having a language in order to acquire the 68

relevant concepts in the first place. Plausibly it is at least psychologically impossible and maybe even in principle impossible to acquire the concepts of ‘vector space’ and of a ‘basis of a vector space’ without mastering some mathematical language. But if non-verbal mathematical thinking is possible, then mathematical concepts can be used in non-verbal thinking after their language-based acquisition. Similar remarks apply to the case of non-verbal philosophical thinking. It is hard to imagine how nonverbal philosophical thinking could occur without previous language-based acquisition of the relevant concepts. But there does not seem good reason to assume that all non-verbal thinking is in that way parasitic upon verbal thinking. Non-verbal thinking requires having a language only if the concepts involved must be acquired using a language. This does not seem to be true for all cases of non-verbal goal-directed rational thinking. Take the example of an adult person who looks from above on a labyrinth and tries to find the way out. Her conscious thinking if verbalized might take something like the following form: ‘(…) At this point I can either turn left or turn right. If I turn right, there are three open possibilities. The first possibility branches into two further ways but both are blocked in the next step. Same for the second; same for the third. So I must not turn right at this point.’ Verbalization of this kind may help some people to find the way but it is surely not necessary for thinking the contents involved. One may think those thoughts without using words. More importantly, the concepts involved in thinking about the way out plausibly do not require language-based concept acquisition. The subject needs to have some concept of turning left, turning right, of a way being blocked, and she needs to understand the goal of getting out of a labyrinth by following a certain path. There is no reason to believe that the acquisition of these concepts requires having a language. Those concepts are rather formed, I suppose, on the basis of acting. Also, the subject needs to be able to draw simple logical conclusions like ‘I can either go right or left, I cannot go right, so I have to go left’. But in order to be able to draw conclusions of this general form, one need not have concepts of the logical operators involved, nor does one have to be in any sense aware of the logical structure of the deduction. It is sufficient that the subject’s deductive intuition accords with the logical principle at issue. There is no good reason to suppose that non-linguistic creatures must be incapable to intuit a logical consequence when confronted nonverbally with certain premises. The present example illustrates the claim. A non-linguistic creature may see and thus realize non-verbally that there 69

are two possible ways to continue at a given location. Visually following both paths, it may realize non-verbally that both paths are blocked. It may immediately occur to it as obvious—again in a non-verbal manner —that in order to exit the labyrinth, the other path must be taken. No explicit deductive reasoning needs to be present to the mind. No languagebased concepts appear to be required. In order to see the plausibility of this point it may help to realize that even in our own case, immediate non-verbal recognition of consequences given certain new information where no awareness of logical structure plays a role in what is present ‘to the mind’ is quite common in daily life and in theoretical thinking. So, the labyrinth example supports and illustrates the following claim: there are cases of goal-directed rational thinking that do not require having a language. If this is correct, then there is no reason to exclude small, nonlinguistic children and higher, non-linguistic animals from being rational goal-directed thinkers. There are many other cases of goal-directed thinking where no languagebased concepts appear to be involved. Musical thinking with the goal of developing a melody is one example. There is no absurdity in the supposition that an intelligent bird without language could in principle be engaged in conscious thought about a tune to be developed. Complex practical problems like those to be solved in playing with the so-called magic cube require thinking, but do not require the use of language-based concepts. Rational thinking in the context of social interaction need not require the possession of language-based concepts either. There is no reason to exclude the possibility of the following situation: a non-linguistic being detects a place with food. It occurs to it non-verbally that there is the danger that other animals of its group will find the place too. The idea pops up ‘in its mind’—non-verbally, of course—that producing a sign for danger will cause the others to flee. It therefore produces signs for danger.10 Whether this is the correct description of what is going on in a given non-linguistic creature is an empirical issue. The point here is this: the mere fact that the creature does not have a language does not render an explanation by reference to conscious thinking of that kind implausible, since there is no reason to suppose that the concepts involved in these thoughts can only be acquired in the presence of language. Experience of interaction with others might well be sufficient. 10. Deceptive behaviour in chimpanzees is described in Hirata 2006. For further examples that invite the hypothesis that conscious thinking is at work are discussed in Glock 2000 who refers to work by Rakoczy. See e.g. Rakoczy, this volume; Byrne 1995.

70

Let us call non-verbal thinking in the weak sense those cases of thinking that do not require internal verbalization; and let us call non-verbal thinking in the strong sense those cases of thinking that do not require having a language. Some cases of non-verbal thinking in the strong sense can be classified as goal-directed and some cases of non-verbal thinking in the strong sense can even be classified as rational. Some cases of nonverbal thinking in the strong sense are equally complex and sophisticated as standard cases of rational goal-directed and verbalized thinking. The latter point is quite clearly illustrated by the example of musical thought. So far we have arrived at the following result. None of the additional properties of thinking so far considered (attending; goal-directedness; being guided by standards of adequacy; being guided by standards of rationality) lead by themselves to the requirement of language possession. In each of the kinds characterized by one or more of these properties, we find possible cases of non-verbal thinking in the weak and in the strong sense. But there is still a further condition left: rational reflection. Does rational reflective thought require internal verbalization, and if not, does it at least require having a language? Let me address the first question first. Is it possible to critically reason about one’s own reasoning without having a language? Take the example of philosophical reasoning. Suppose you arrive at a certain conclusion—say that two views are incompatible—by nonverbal thinking. You think of the first view—without internally verbalizing the view in a specific way—and you think of the second view—without internally verbalizing it—and after a while you have the clear intuition that the two views are incompatible. But then you have doubts about your conclusion. You can go back to thinking about the first and the second view in a similar way and test your intuition once again. But this does not seem to be a case of reflection; it is rather a case of repeating a former instance of thinking in order to test its reliability. So we should take an example where a person non-verbally draws an explicit conclusion from premises that are somehow non-verbally present to the mind. I tend to think that there are cases where we non-verbally draw theoretical conclusions. But can we critically reflect upon our own deductive thinking in a non-verbal way? In the case of doubt we may of course engage in going through the non-verbal reasoning once again. But can we critically check the validity of the conclusion or reflect upon whether the premises are well-founded? It seems clear that verbalization will facilitate critical reflection. We are surely more sensitive to possible objections (both to the validity of conclusions as well as to the acceptability of premises), 71

once we have found a satisfying verbalization of the argument we wish to critically evaluate. So, internal verbalization surely is of great help in critical reflection. But is it necessary? Could there be a being that reflects critically about its own thoughts without internal verbalization? And could it be that we do so occasionally ourselves? The issue is difficult to judge by phenomenological reflection. Certainly specific kinds of critical questions about one’s own reasoning cannot be answered without the choice of some specific formulation of one’s argument. It is hard to imagine, for instance, how one could critically assess the formal validity of an argument without couching it somehow into an explicitly formulated deduction. Also, it is surely a psychological fact about humans that they need verbalization (internal or even external) in order to make substantive progress in their critical reflective thought. But maybe there are also general and principled obstacles to critical reflection without language. There are two distinct questions to be answered. (a) Are the second order thoughts involved in critical reflection upon ones own thinking necessarily verbalized? (b) Does critical reflection require verbalization of the first-order thoughts that are the object of reflection? A positive answer to (b) can be motivated by the following suspicion. It might be impossible to make a first-order thought the object of one’s critical reflection without first verbalizing the first-order thought (see the considerations at the end of section 7).11 9. Final remarks No philosophy, no science, no major intellectual human achievement could have developed without the use of language in the exchange of ideas and in mutual rational control in theoretical and practical reasoning. Individual rational thinking and critical reflection is to a large extent impossible without verbalization, and thus without the possession of language. Nonetheless, it seems clear to me that the role of language in thought has been overestimated in many ways in contemporary philosophy. Many have denied the possibility of genuine thinking in non-linguistic creatures and it is common to believe that genuine thought requires external or internal verbalization. Even those who accept the possibility of thought without verbalization and the possibility of thought in non-linguistic creatures, 11. A different argument for the claim that critical reflection requires the possession of language is developed in detail in Bermúdez 2003.

72

tend to think that (a) the lack of language dramatically restricts the possible complexity of thought and (b) that intelligent behaviour in non-linguistic creatures needs to be explained by a type of thinking quite different from ours. Given the picture developed in the present paper, there is reason to doubt both assumptions. As some examples mentioned earlier may illustrate, considerable complexity of thinking may be present in cases where no verbalization and no possession of language is required. The example of musical thinking and the example of thinking about a complex labyrinth can illustrate the point. It is an open question—I take it—whether there could be a species on some other planet that has not developed a language but has developed—in non-linguistic collaboration—a complexity and sophistication in their thinking that is comparable to ours. To illustrate the idea one might think of a species of whales in the far future that has not developed language, but a highly elaborated culture of singing that resembles classical music in its beauty and complexity. It would be inappropriate to deny that these creatures would be genuinely thinking when composing their songs in much the same way a human composer does (though without using a notational system). There does not seem to be any good reason to claim that the required complexity of musical understanding and emotional sophistication could not possibly develop in the absence of language. Harmonic, melodic and rhythmic understanding does not require the possession of any concept that could only be acquired in verbal exchange. The specific kind of emotional experience involved in music does not seem to require the possession of concepts that are linguistically expressible or require for their possession linguistic understanding.12 Other possible examples of highly complex non-linguistic thought might be taken from the practical realm. One might well imagine a non-linguistic yet highly intelligent species that has developed complex mechanical skills and complex capacities to think about mechanical problems without thereby using a language. The focus in philosophy on verbalized thought may be responsible for a common distorted picture of the way humans think in daily life. Nonverbalized thought is a common and quite normal phenomenon when we are confronted with a practical problem. Suppose a person at the hotel reception gives you a new kind of key, and when you try to open your door, you don’t understand what you have to do. You consider different 12. An anonymous referee objected that highly complex music is a system of signs and so a language. This claim can certainly be defended but it does not undermine the point. Music is not a language in the ordinary sense here at issue.

73

alternatives and try to at least partially understand the mechanism in order to succeed. You may verbalize some of your thoughts, and you may do so in order to help your concentration and your memory about what you have already tried. But you need not do so, and in many cases we solve problems of this practical kind without internally verbalizing our reasoning. The solution of practical problems of this kind can be based on thinking that neither requires verbalisation nor the possession of a language. Now it is well-known that some animals are capable of solving quite difficult practical tasks (compare for impressing examples Rakoczy, this volume). Do we have reason to doubt that they solve these problems on the basis of genuine thinking? I don’t think so. The lack of language does not justify the doubt. Since our reasoning in similar situations is non-verbal in the broad and in the narrow sense, the lack of language in ravens and monkeys does not provide any reason to deny that their intelligent behaviour is due to genuine thinking too. (This is not to deny that their strategies might be quite different and that their phenomenology of thought might differ from ours.) Furthermore, the capacity for thinking about practical problems of this kind in an effective way does not seem to depend on language skills. If this is so, then we should even be open to the possibility that some non-human creatures are more successful thinkers about specific kinds of problems than humans (compare the results for monkeys and the result for humans in Rakoczy, this volume). These considerations may cause a certain welcome and adequate Gestalt switch in the way we conceive of ourselves and in the way we think of ourselves in relation to the rest of the animal kingdom. If it is correct that large parts of our own thinking are non-verbal in both senses, then we probably share much more of our cognitive life with other animals than is usually assumed. To think about how to put a key into a hole being a monkey of a certain species and to think about the same problem being a human might be phenomenally quite similar. Maybe even to think about how to get an object out of a bottle being a raven and to think about the same problem being a human is phenomenally quite similar, too. In the absence of counter-evidence, we should rather assume fundamental similarities between our non-verbal thinking and non-verbal thinking in other creatures. If this is correct, then interpreting animals as not only endowed with phenomenal consciousness, but also as genuine thinkers is not a sign of an uncritical, romantic and naive attitude. Rather, it is in accordance with what we should take as the ‘default position’. We may thus arrive at a view of ourselves that puts us much closer to those who do 74

not have a language. One might take this to be a reason to lower the status of the human being. But I think that the opposite reaction is adequate. The switch in our self-conception that makes the way in which we are just animals more salient, should rather motivate more respect and admiration for those conscious beings on our planet that do and have to think without external or internal verbalization.

REFERENCES Bermúdez, José Luis. 2003. Thought without Words. Oxford: Oxford University Press. Byrne, Robert. 1995. The Thinking Ape. Oxford: Oxford University Press. Glock, Hans-Johann. 2000. “Animals, Thoughts and Concepts”. Synthèse 123, 35–64. Hirata, Satoshi. 2006. “Tactical Deception and Understanding of Others in Chimpanzees.” In: Tetsuro Matsuzawa, Masaki Tomonaga, Masayuki Tanaka (ed.). Cognitive Development in Chimpanzees. Springer: Tokyo, 225–276. Ludwig, Kirk. 2004. “Rationality, Language and the Principle of Charity”. In: Alfred Mele, Piers Rawling (eds.). The Oxford Handbook of Rationality. Oxford: Oxford University Press, 343–362. Nida-Rümelin, Martine. 2006. “Doings and Subject Causation”. In: Albert Newen, Verena Hoffmann, Michael Esfeld. Mental Causation, Externalism, and Self-Knowledge. Special Issue of Erkenntnis 67(2), 147–372. Siewert, Charles. 1998. The Significance of Consciousness. Princeton: Princeton University Press. Rakoczy, Hannes. 2010. “From thought to language to thought: towards a dialectical picture of the development of thinking and speaking’”. In this volume, 77–103.

75

Grazer Philosophische Studien 81 (2010), 77–103.

FROM THOUGHT TO LANGUAGE TO THOUGHT: TOWARDS A DIALECTICAL PICTURE OF THE DEVELOPMENT OF THINKING AND SPEAKING Hannes RAKOCZY University of Göttingen In both the evolution of thought in the history of mankind, and the evolution of thought in an individual, there is a stage at which there is no thought followed by a subsequent stage at which there is thought. To describe the emergence of thought would be to describe the process which leads from the first to the second of these stages. What we lack is a satisfactory vocabulary for describing the intermediate steps (…). That means there is a perhaps insuperable problem in giving a full description of the emergence of thought. I am thankful that I am not in the field of developmental psychology! (Davidson 1999, 11)

Summary Lingualism claims there is no thought without language. At the other end of the theoretical spectrum, strong nativist ‘Language of Thought’ theories hold that public language is inessential to private thought. For an adequate empirical description of the ontogeny of thought and language, however, we need an intermediate position recognizing the dialectical interplay between pre-linguistic thought, language acquisition and the development of full-fledged linguistic reason. In this article recent findings from developmental and comparative psychology are reviewed that highlight the need for such a dialectical picture. These findings concern basic cognitive capacities common to pre-linguistic infants and many non-linguistic animals, specific social-cognitive abilities in human infants that enable language acquisition, and the influence of language on the subsequent cognitive development.

1. Introduction Developmental and comparative psychologists try to earn their living by struggling to describe the emergence of thought and language in evolu-

tion and ontogeny. The project is to characterize the development from non-minds via half-formed minds to proper minds. Philosophy can be quite obstructive to this project—when it takes the form of either of two radical positions. Lingualism denies the intelligibility of any kind of thought whatsoever in the absence of language. There just are no halfformed minds, according to this position. There are purely differential responses on the one hand, to be described physically, and full-fledged linguistic thoughts on the other hand, to be described rationally, and nothing worthy of the label ‘cognitive’ in between (e.g., Brandom 1994; Davidson 1982, 1999). With no middle-ground between the differential responses of rusting iron to humidity here, and the full-blown discursive rationality of adults there, describing the emergence of thought amounts to unravelling the one big mystery how language, and with it thought, emerges out of mindless responses. Language of thought (LOT) theories (Fodor 1975), in contrast, do not see any substantial role for natural language to play in thinking. According to this position, all thought is manipulation of internal private symbols in an innately specified lingua mentis. Natural languages are mere transcriptions of private thoughts, and so acquiring a language in ontogeny does not categorically affect cognitive abilities. In terms of phylogeny, humans share the language of thought with many other species. What makes them cognitively unique is just that their LOT vocabulary happens to be bigger, and that they happen to have a language module that allows for the public exchange of ideas. Both radical positions are deeply unsatisfying for the purposes of describing the development of thought. Language of thought theories essentially neglect the role, both conceptual and empirical, of discursive practices in the shaping of full-fledged adult rationality. Lingualism, on the other hand, ‘over-intellectualizes the mind’ (Hurley 2001), rejecting all talk of non-linguistic animal and infant rationality as mere metaphor, and thus flies in the face of perfectly fine common sense and scientific explanatory practices. Developmental and comparative psychology are just elaborated extensions of our common sense folk psychology; as, arguably, is all psychology. And while folk psychology’s notion of thinking, as lingualists stress, is probably in a Sellarsian sense modelled on language (in particular, on thinking aloud), this does not mean that thinking and speaking are conceived of as basically identical. Folk psychology seems clearly committed to a deep cognitive realism regarding different animals as well as different stages of humans, to a continuum of cognitive abili78

ties that implies the falsity of the lingualists’ drastic dichotomy—eliminativism or fictionalism regarding non-linguistic creatures, but realism regarding speaking creatures. Common sense surely does not have the last word here, but it does have the first, and deviating from this commitment, I surmise, puts the burden of argument on the eliminativist. Not surprisingly, none of the arguments for lingualism presented so far, such as the argument from the holism of the mental, or the argument from the premise that belief requires the concept of belief, which requires language (Davidson 1982), has been met with much enthusiasm in developmental psychology, or common sense (see, e.g., Bermúdez 2003; Glock 2000). In a similar vein, from a psychological point of view, the ability to speak a language itself must not be taken as a primitive, but stands in need of explanation. Such an explanatory project might not sit well with philosophical accounts picturing language acquisition as a mere conditioning process. (Think, for example, of Wittgenstein when he talks about ‘Abrichten’, or of Davidson and Sellars in their more behaviouristicsounding passages.) No question, parrots can be conditioned to parrot. But parroting is a far cry from speaking; and from the empirical point of view of comparative psychology, one of the most impressive findings in recent decades has been how pathetically even human-raised chimpanzees perform in language learning (e.g., Rivas 2005). From a psychological perspective, then, what we need for describing the emergence and development of thought is a dialectical picture that—much in the spirit of Vygotsky (1934/1986)—meets the following desiderata: (a) It does justice to the possibility and existence of non-linguistic cognition in both animals and pre-linguistic humans. (b) It acknowledges specific, uniquely human cognitive abilities as part of the basis of language acquisition, while (c) at the same time recognizing the ways speaking enables, shapes and transforms thinking of very special kinds. In what follows, I will review recent empirical findings from developmental and comparative research to highlight and illustrate the building blocks from which such a picture could be pieced together. Different ways of thought with and without language should be distinguished for this purpose. First of all, while the philosophical disputes naturally centre around the question whether and to what degree language is constitutive of thinking, the psychological side of it, upon which I will focus, is more concerned with the less demanding question whether and to what degree language is empirically necessary for certain kinds of thought 79

and behaviour. Let me distinguish three ways we might talk about thought without language: (i) (ii)

Thinking without speaking any language at all. Thinking about a domain D without (dispositionally) having the requisite explicit vocabulary to talk about D. (iii) Thinking about a domain D without thinking aloud or (occurrently) speaking subliminally with the requisite vocabulary about D. (iii) is an interesting topic both phenomenologically (see, e.g., Dennett 1991, on language as a Joycean consciousness machine) and in cognitive psychology dealing with the role of (subliminal) speaking in onlinethinking (see, e.g., Spelke 2003). (i) pertains to animals and pre-verbal infants, (ii) to older children who do speak, but whose explicit linguistic abilities in a given domain lag behind their non-linguistic cognitive abilities. Different theories regarding thought and language emphasize or deny the possibility of (i) or (ii). For example, Davidson’s lingualism is officially not concerned with (ii), but only with (i): ‘Language, that is, communication with others, is thus essential to propositional talk. This is not because it is necessary to have the word to express a thought (for it is not); it is because the ground of the sense of objectivity is intersubjectivity.’ (1994, 234) In this paper, I will mainly be concerned with (i) in the case of animals and very young children, and with (ii) in the case of older children who do have some command of language, but a very rudimentary one, lagging behind their systematicity in nonverbal action. In the following, I will first review findings of common cognitive abilities in non-linguistic animals and pre-verbal infants that are best described as simple forms of thinking without words. The second part deals with specifically human cognitive foundations of the emergence of language and culture. These foundations are to be found in uniquely human social cognitive abilities of understanding one another as intentional beings and of entering into shared or collective ‘we’ intentionality. Once children have acquired a language and participate in cultural practices, however, individual cognitive abilities get radically re-shaped and transformed; this will be the topic of the third part. Finally, in the last section I will try to explore the implications of these findings for a comprehensive picture of the dialectical developmental of thought and language.

80

2. The roots of non-linguistic cognition in animals and infants Piaget (1952) set the scene for work on the roots of non-linguistic thought in infancy. His and much subsequent research focused on the early development of intentionality in perception and goal-directed action. In these areas, recent comparative psychology has found striking parallels in the cognitive developments of pre-verbal infants and other animal species: both infants and other animals, notably primates, perceive and segment the world around them with a (proto-conceptual1) grasp of such basic categories as objects, numerical identity, relations, space, and themselves; they use this grasp in simple forms of instrumental reasoning and inferences, and they act goal-directedly and solve problems in flexible, insightful and creative ways, often involving the use of tools.2 To illustrate these common cognitive abilities, let me focus on two philosophically relevant domains: the most basic forms of objective thought, namely object cognition; and the most basic form of practical rationality, namely goal-directed action. 2.1 Object cognition All thinking requires a minimal notion of objectivity: an at least implicit distinction between things perceived and the perceiver, and some sensitivity to the fact that things persist unperceived. The positing of objects, or reification, can be seen as the most basic form of objective thought. Following philosophical tradition (Strawson 1959), let us distinguish three kinds of object perception and cognition:3 (a) mere feature placing, (b) spatio-temporal tracking of physical bodies, and (c) full-blown sortal object individuation. Feature placing, on a wide reading (‘Lo! Redness!’, ‘Raining’), is involved in all kinds of sensory receptivity and does not necessarily involve any kind of positing of persisting individuals. Spatio-temporal tracking of physical bodies, in contrast, is the most basic form of reification. Here the world is segmented into non-overlapping, bounded bodies that move cohesively and continuously through space and 1. I follow much philosophical and psychological writing in calling such pre-linguistic cognitive abilities ‘proto-conceptual’. In the final section I will briefly turn to the question what that might actually mean. 2. For a comprehensive review, see Tomasello & Call (1997). 3. For an excellent review of philosophical and psychological approaches to these kinds of object cognition see Clark (2004).

81

time.4 Though this does not yet supply full individuation under sortals, it does supply some primitive criteria for persistence and countability (Xu 1997). Piaget was the first to study this kind of object tracking, dubbed ‘object permanence’, in infancy. In manual search behaviour, the simplest form of object permanence appears around 8 months of age. While younger infants would not care about objects that disappear behind an occluder (‘out of sight, out of mind’), from this age on, infants actively search behind the occluder and try to remove it—which is plausibly interpreted as an awareness that the object must still be there.5 While this form of object permanence is still limited in some respects, at around 18 months infants master full object permanence: they now track physical bodies even when their movements are not directly perceivable. In the classic so-called ‘invisible displacement task’, the infant sees an object disappear in a small container A, which is then subsequently moved into the bigger containers B and C, and finally taken out again. Younger infants tend to randomly search at one location and then give up, while infants from 18 months systematically search all three containers. This is standardly interpreted in the following way: infants track the physical body, even when its displacement is not directly visible, making use of a primitive sense of necessity and reasoning from a negated disjunct: ‘It is in either A, B, or C. If it is not in A, then it must be in B or C’.6 Research with non-human animals has revealed that many species, including cats, dogs, and monkeys, reach the stage of simple object permanence appearing in human ontogeny around 8 months. The great apes, furthermore, reach the stage of full object permanence mastered by children at 18 months.7 Merely tracking physical bodies, however, falls short of providing a full-fledged schema for individuating objects in the proper sense (see Wiggins 1997). Mere spatio-temporal tracking of bodies does not allow a distinction between an A (say, a piece of clay) disappearing and then a B 4. These have been called ‘Spelke objects’ in developmental psychology, and ‘proto-objects’ in research on visual attention (see, e.g., Pylyshyn 2000). 5. Subsequent research with so-called habituation measures has revealed even much earlier sensitivity to object permanence; see, e.g., Baillargeon (1987). 6. Additional evidence for such an interpretation comes from a recent study showing that children (some years older, though) increase their speed of searching as more alternatives become excluded, i.e. when moving from the first to the second to the third container (Watson et al., 2001). For more work on reasoning from negated disjuncts in apes, see Call (2004). For a philosophical treatment of this form of reasoning in non-linguistic creatures, see Glock (2000). 7. For a review, see Tomasello & Call (1997).

82

(say, a statue made of clay) appearing, and the same A disappearing and re-appearing. Sortal concepts are needed for such distinctions regarding numerical identity (‘Is it the same A?’) and full-fledged object individuation. In the case of language users we have clear criteria for mastery of sortal concepts, namely use of count nouns and identity statements. But what about pre-verbal analogues to such conceptual abilities? Recent studies have once more used looking time and search behaviour as an index of such pre-verbal analogues. When young infants see two objects simultaneously disappear in a box and then only find one, they continue searching. This just requires object permanence and can be done based on mere spatio-temporal criteria. In a crucial variation supposed to tap proto-sortal individuation, however, infants saw an A disappear in a box, then a B appear from the box and disappear again, and then only found one object. Purely spatio-temporal information is of no help here, as it does not specify criteria of identity and distinctness. Younger infants, though capable of object permanence and mastering the spatio-temporal version, fail the proto-sortal version (Xu & Carey 1996; van de Walle et al. 2000).8 Only from around 10-12 months do infants begin to master the latter, and such mastery has been found to be correlated with language comprehension (Xu & Carey 1996). These findings lend prima facie support to the idea, popular in philosophy (e.g., Quine 1974), that reification comes with language and is therefore uniquely human (Xu 2002). However, subsequent research with non-human primates has documented analogous cognitive abilities in nonlinguistic creatures. Both monkeys and apes perform in exactly the same way as human one-year-olds in spatio-temporal and proto-sortal individuation tasks (Mendes, Rakoczy & Call 2008; Phillips & Santos 2007). In sum, pre-linguistic infants and non-linguistic animals share with mature thinkers the basic roots of objective thought—the ability to track and individuate objects persisting ‘out there’ independently of being perceived. They use such abilities in systematic and flexible ways in simple forms of reasoning, e.g., in reasoning from negated disjuncts. 8. It might be objected that this task does not require (proto-)sortal object individuation but just sophisticated feature-placing. Infants see A-features and B-features, but then only find A-features and thus look for the missing B-features. Empirically, this does not seem plausible. Infants in such tasks do not take into account just any property differences between objects. In particular, they have been found to disregard property differences within, but not across kinds, which indirectly suggest that they track the object as of a certain kind and not just as bundles of properties (e.g., Xu, Carey & Quint 2005).

83

Another prerequisite for objective thought is some rudimentary awareness of oneself as an object in space (Strawson 1959). Again we have clear criteria in speaking creatures such as the use of personal pronouns etc., but what could count as a pre-/non-verbal indicator of some such rudimentary awareness in non-linguistic animals? The non-linguistic task standardly employed in developmental and comparative psychology is the so-called ‘mirror rouge’ task (Gallup 1970). A mark of rouge is surreptitiously applied to the infant’s/animal’s forehead (infants are distracted, animals often narcotised), and then the subject is placed in front of a mirror. Touching one’s own face to remove the mark is interpreted as an indicator of some rudimentary awareness of oneself as an object in space, the ‘Me’ in the sense of James and Mead. While younger infants and most other species, including monkeys consistently fail the task and treat the mirror image like a conspecific, infants from around 18 months begin to master this task, and great apes have been shown to succeed. (See Tomasello & Call 1997, for an overview.) 2.2 Goal-directed action and practical rationality Infants and many animals do not just sensorily register features. They also perceive, track and reason about individuals, themselves among them. What is more, infants and many animals do not just behave in response to stimulation, such as iron rusts in response to humidity, but act intentionally. Again, while we have clear criteria for intentional action in speaking creatures, think of: expressions of intentions; non-inferential descriptions of what one is doing; explicit practical inferences; statements about success and failure etc. How do we know this in creatures without language? How do we distinguish mere silent behaviour from silent action? Common sense folk psychology supplies us with an explanatory practice that has a set of rather clear indicators for silent action, indicators that developmental and comparative psychology as sophisticated common sense have inherited. Here, flexibility and generativity, sensitivity to means-ends structures, hierarchical and sequential behaviour organization, persistence and signs of (non-) fulfilment as a function of goal-attainment are the most prominent ones. We don’t want to say that iron rusts intentionally, or that sweating as such is an intentional action, and we do want to say that annoying a detested neighbour by making his car rust, or going to the sauna for a sweat can be perfectly fine instances of intentional action, because the latter behaviours can have, and the former lack, just these characteristics. 84

Piaget (1952) pioneered the study of the development of intentional action in preverbal infants. From around 8 months, at the time object permanence begins to emerge, infants’ behaviour for the first time displays instrumental structure. Infants begin to engage in sequences of action that bear some means-ends relations such as, e.g., removing an obstacle in order to reach for an object, persistently pursuing their goals, varying the means if necessary, and indicating fulfilment (happiness) or non-fulfilment (frustration). From this age on, infants’ instrumental actions begin to incorporate the use of tools in increasingly sophisticated and flexible ways (e.g., Willatts 1985). Infants early in their second year also grasp the basic intentional structure of others’ instrumental acts, as indicated in their imitation (see also section 3). When in a recent study infants observed another person perform an instrumental act with a bizarre means (e.g., switching on a light with the head), they saw this only as a means to an end when the other person could not have done otherwise (e.g., because her hands were occupied); they themselves subsequently switched on the light with their hands. In contrast, when the model could have done otherwise, infants saw the means as an end in itself, and faithfully imitated it (Gergely et al. 2002). Turning to non-human animals, the flexible and insightful use of tools for instrumental purposes in chimpanzees, both in the wild and in experimental settings, is well-known. For example, chimpanzees in the wild flexibly and systematically use stones as nutcrackers in quite sophisticated ways (Boesch & Boesch 1990), and they use different kinds of sticks adaptively for fishing different kinds of insects (Goodall 1986). Best known from experimental settings is probably the problem-solving of Köhler’s (1926) apes who—in creative and in insightful ways—used different kinds of novel tools to obtain food from beyond their reach. In addition to familiar documentations of such problem-solving in chimps, more recent experimental research has found that many primate species that have not been reported to use tools in the wild, readily do so in creative and novel ways in experimental settings (Tomasello & Call 1997). Some apes have recently been reported to even store tools systematically for future use (Mulcahy & Call 2006). And in some problem-solving tasks, apes clearly outperform human children up until the age of 6 years (Mendes, Hanus & Call 2007). Also, the systematic use of tools seems not confined to primates. Crows, for example, have been found to actively cast hook-shaped tools for subsequent usage to retrieve food from a test-tube (e.g., Weir, Chappell & Kacelnik 2002). 85

To summarize the findings from object cognition and instrumental action: we see analogous abilities in pre-linguistic infants and non-linguistic animals that common sense and academic psychology don’t hesitate to describe as simple forms of thought. ‘The infant sees two objects disappear in a box, finds only one and thus knows there must be another one’, or ‘The ape wants to eat the nut, knows that using a stone helps, and therefore uses the stone’ are natural ways of describing the phenomena in question. In contrast to ‘The flowers are looking forward to the spring’, they are meant non-metaphorically. Granted, there do lurk serious conceptual challenges regarding the ascription of contentful attitudes to creatures in the absence of full-blown linguistic concepts, in particular regarding the determinacy of content. Does the ape intend to crack the nut, or the brown object, or the thing with the nut-appearance, or undetached nut-parts, etc.? But to common sense folk psychology, the radically holistic manoeuvre to conclude from the fact that there are many things the ape does not know about nuts—say, that they’re a product of biological processes, and no prime numbers—that the ape cannot intend to crack nuts, seems more like a reductio ad absurdum of radical holism than a serious problem for apes. 3. From thought to language: social cognition and the acquisition of culture As we have seen, both infants and other animals engage in rudimentary forms of objective thought and practical rationality. Why, then, don’t other animals begin to talk? Given the remarkable commonalities and continuities in the development of non-linguistic thought in infants and many animals, why do only infants begin to speak a language and grow into a full-blown culture? In particular, why don’t even chimpanzees raised in human environments with extensive training in sign language acquire anything close to human linguistic and cultural abilities? Recent research strongly suggests that it is not so much the lack of the right kind of speech apparatus or proto-grammatical abilities9 that explains why the brutes remain brute, but rather missing pragmatic, socio-cognitive background abilities. Infants and other animals seem to cognitively relate to the physical world around them in basically the same ways. But socio-cognitively, 9. Contrary to a recent claim by Hauser, Chomsky & Fitch 2002, these might even be present in birds (Gentner et al., 2006).

86

non-human animals don’t relate to each other in the right kind of way to develop language and culture. 3.1 Understanding persons From a broadly Gricean point of view, language has to build upon some basic folk psychology, upon some grasp of conspecifics as persons—as rational agents and potential cooperative and communicative partners. In human ontogeny, as much recent research has found, the first crucial milestone in the development of a folk psychology occurs around one year of age and lays the foundation for language acquisition. (See Tomasello 1999; Tomasello et al. 2005, for an overview.) After having interacted dyadically with either objects or persons for the first months of their life, infants now begin to enter into triadic interactions with other people oriented towards objects. They follow the gaze and the pointing gestures of others to external objects, and they begin to use the pointing gesture in varied ways themselves. They use others’ emotional responses towards ambiguous situations to guide their own actions (‘social referencing’), and they imitate others’ acts on objects and join into collaborative actions.10 These kinds of behaviour, which co-emerge in systematic and synchronic fashion and which reliably precede and predict language acquisition (e.g., Carpenter et al. 1998a), have been interpreted in the following way. What is dawning here is a simple folk psychology in the sense that children begin to perceive one another as persons with perceptual access to the world and engaged in intentional rational action. Making use of Searle’s (1983) taxonomy of intentional phenomena, one could say that just as intention is the biologically and ontogenetically primary attitude with world-to-mind direction of fit, and perception the corresponding primary attitude with mind-to-world direction of fit, so understanding perception and action are the primary forms of folk psychology, or second-order intentionality. The main evidence for understanding intentional action comes from imitation: straight imitation, but more convincingly imitation in cases of mis-match between the mere behaviour seen and what was intended; imitation of unfulfilled attempts (Meltzoff 1995), differential imitation of accidents versus intentional acts (Carpenter et al. 1998b), and what has been dubbed ‘rational imitation’ (Gergely et al. 2002, see above). The 10. This bears interesting relations to Davidson’s notion of triangulation, a discussion of which, however, goes beyond the scope of the present paper (see Brink 2004; Eilan, 2005; Roessler 2005).

87

main evidence for understanding perception comes from children’s gaze and point following from around age 1, and, somewhat later from around 24 months; from engagement in what has been called ‘level 1 perspective taking’ (Flavell et al. 1981; Moll & Tomasello 2006), i.e. the ability to understand that different people may perceive different things (‘I see something you do not see’). Why do we, folk and professional psychologists alike, class these abilities as cognitive? As in the case of simple individual intentionality, the main reason is that the kinds of behaviour that are the basis for ascribing these abilities to infants manifest a differential, systematic and generative structure that is reminiscent of the inferential structure of full-blown linguistic thought. (I will return to this point in the final section.) While it has long been thought that no other species, not even great apes, follow humans in the development of this simple folk psychology, some recent evidence led to qualifications of this bland proposal. It has been found that Chimpanzees are in fact capable of a simple understanding of intentional action; for example, they systematically distinguish between unfulfilled acts where the actor is unwilling from those where the actor is unable (Call et al. 2004). They also are capable of understanding perception in the form of level 1 perspective taking, in that they take into account what a conspecific has and has not seen in a food competition task (Hare et al. 2000).11 Simple forms of folk psychology thus no longer seem to be uniquely human—infants and chimpanzees share basic abilities of interpreting conspecifics as individual agents. 3.2 Collective ‘we’-intentionality Understanding each other as individual agents is surely necessary for entering into a linguistic practice, for one can get a grasp on reference only with a rudimentary notion of other speakers’ perceptual perspectives (e.g., Quine 1990) and with a rudimentary understanding of what interlocutors are up to. But to participate in linguistic and other cultural practices, it is not enough to understand others as individual agents. What is needed beyond such simple individual second-order intentionality is the ability 11. These empirical data fit nicely with a more conceptually derived claim by José Bermúdez (2003) that in the absence of language and the recursive embedding it allows in propositional attitude ascriptions, folk psychology is confined to understanding non-epistemic perception and goal-directed action.

88

to engage in shared or collective ‘we’-intentionality: ‘The biologically primitive sense of the other person as a candidate for shared intentionality is a necessary condition of all collective behavior’ (Searle 1990, 415, my italics). What humans develop early in ontogeny and what other species lack is precisely this propensity for collective intentionality (Rakoczy & Tomasello 2007; Tomasello & Rakoczy 2003; Tomasello et al. 2005). We are facing collective intentionality when two or more subjects share an intentional ‘we’ attitude that is not straightforwardly reducible to individual intentional attitudes.12 When you and I meet and agree to take a walk together, to take an example from Margaret Gilbert (1990), we form and then pursue the joint we-intention ‘We walk together’ that is not reducible to the sum of my individual intention ‘I walk’ plus your analogous one, not even when supplemented with our mutually knowing about these intentions. This irreducibility of collective intentionality becomes even more obvious in the case of more wide-ranging social affairs. ‘We play/one plays chess like this’ is clearly no sum of ‘I play it like this’, ‘You play it like this’ and ‘She plays it like this’. Specific normative dimensions go along with collective intentionality. In simple cooperative actions, the partners bind themselves to acting jointly and are thus committed to the pursuit of a joint goal. And in the case of more wide-ranging social affairs, the way ‘one does it’ quite obviously fixes a framework of right and wrong moves. In human ontogeny, simple collective intentionality develops from the second year in the domains of cooperative actions and pre-linguistic communication. Children from the age of one and a half begin to engage in collaborative games with complementary roles and turn-taking structure, and in collaborative instrumental activities with clearly differentiated roles (e.g., Warneken, Chen & Tomasello 2006). In the course of such collaborative acts, they communicate pre-linguistically in appropriate ways, e.g., by pointing to the required place for the partner. When the collaboration threatens to break down, they re-engage the partner and assign him his role (again by pointing; Warneken et al. 2006.13) Children 12. For the central works in recent analytical philosophy on this, see Bratman (1992); Gilbert (1990); Searle (1990), (1995), (2005), Tuomela & Miller (1988). For an overview, see Tollefsen (2004). 13. While human-raised chimpanzees in this study did show some social coordination in instrumental problems that needed two individuals for the solution, they did not engage in such communication and re-engagement behaviour. More generally, many researchers have argued that prima facie truly cooperative behaviour in chimpanzees, in particular social hunting, in fact are just sophisticated social coordination: one individual starts hunting at a certain place, then

89

this age, but not chimpanzees, also seem to have a simple understanding of complementary roles in joint activities, as indicated in their spontaneous role-reversal imitation.14 Communication itself is of course a cooperative activity characterized by collective intentionality. Using pointing and other gestures, even prelinguistic infants make proto-declarative communicative acts that are not just instrumental for attaining some individual end; for example, they point out information that others need, e.g., about the location of a lost object (Liszkowski et al. 2006). In contrast, chimpanzees do not spontaneously point, and the ones who learn to do so in human environments only ever use it proto-imperatively for instrumental purposes (Rivas 2005; Tomasello et al. 2005). Infants’ rudimentary ‘sense of the other as candidate for shared intentionality’ enables participation in these forms of joint cooperative and communicative activities which in turn function as a foundation and scaffold for the acquisition of language (Bruner 1983; Tomasello 2003). Young children’s participation in shared intentional activities is not confined to simple cooperation. Even pre-verbally they begin to partake in activities with collective assignment of status functions to objects and thus with proto-institutional structure, in particular in the domain of playing games. From around 18 months, children begin to engage in games of social pretend play—mostly organized topic-wise around such mundane things as pretending to eat, drink etc. (For example, pretending that a wooden block is an apple, pretending to peel, cut and eat it.) Only human children do so. There are a few anecdotes of pretence-like behaviour in some human-raised animals (see Mitchell (2002) for an overview), but these are difficult to interpret, and it is in general quite clear that no other species habitually engages in pretend play as we know it.15 Games with established rules are among the paradigmatic examples of activities with constitutive rules and status function assignment (‘this piece of wood counts as a queen in chess’, ‘moving it thus and thus counts as attacking’; Searle 1969, 1995), and games of pretence can be seen as local, ad-hoc analogues of such established games. The two levels of the next individual starts hunting, but cannot take the same place, then the third individual has to take yet a different place etc.. See, e.g., Tomasello & Call (1997); Tomasello et al. (2005). 14. For children, see Carpenter et al. (2005); for chimpanzees, see Tomasello & Carpenter (2005). 15. For an excellent review of precursors to pretend play in great apes, see Gomez & MartinAndrade (2005).

90

fact and fiction mirror the two levels of institutional phenomena generally—institutional fact put on top of brute facts in line with the schema ‘X counts as a Y in context C’. ‘This wooden block counts as an ‘apple’ in our pretence game’ is on a par with ‘This piece of wood counts as a queen in chess’ or ‘This slip of paper counts as money in our currency area’ (Walton 1990). These assignments of fictional function bring with them a normative structure of the joint pretence activity. ‘X counts as Y in context C’ means that in C, X ought to be treated as a Y. Once declared an apple, the block in the block/apple pretence game ought to be treated accordingly in the game. Some pretence acts are inferentially licensed in the game, others are not. Pretending to peel the block/apple, pretending to eat it or to bake a cake with it are licensed; pretending to drive it or pretending to fax it are not (see Walton 1990). Children from the age of 2 do in fact seem to grasp this normative structure created through joint pretence stipulations, as is indicated by their inferentially appropriate responses to others’ pretence acts. When an experimenter pretended to pour tea into a cup, for example, children pretended to drink from the cup. When the experimenter pretended to spill tea on the table, children pretended to clean the table (Harris & Kavanaugh 1993; Rakoczy & Tomasello 2006; Rakoczy et al. 2004). What is more, the children systematically distinguish such pretence acts from superficially analogous behaviours with different intentional structure. When an experimenter pretended to pour from a full but closed container into a cup, they themselves—inferentially appropriately—pretended to drink from the cup. However, when the experimenter made the same pouring movements with the same kind of container, but marked them as frustrated attempts, they—again inferentially appropriately—completed the failed attempt by opening the container and really pouring (Rakoczy & Tomasello 2006; Rakoczy et al. 2004). Interestingly, though children this age have acquired simple forms of language, their competence in action revealed in these studies by far exceeds their explicit verbal competence. Only some years later can children reliably tell explicitly whether the experimenter had pretended, or had actually tried to pour (Rakoczy et al. 2006). Not only do children themselves act inferentially appropriately in joint pretence games. They also indicate an awareness of the normative structure of such games more directly in their responses to others’ mistakes. When in the context of a shared pretence game, a third party entered and then 91

confused the pretence identities (status functions) of the objects, children intervened by protesting, criticizing and teaching (Rakoczy 2008).16 In summary, pre-verbal children and apes share not only abilities of non-linguistic intentionality, they also share some rudimentary folk psychology, or second-order intentionality. Human infants, however, do not just interpret each other as individual agents, but develop a ‘sense of the other person as a candidate for shared intentionality’ (Searle 1990, 415) in the second year, and thus enter into collective we-intentionality in the domains of collaboration and pre-verbal communication. Form a comparative point of view, even the pre-verbal communication of human infants is already markedly different from the communication of all other species in that it involves shared intentionality in the form of a basic Gricean cooperative structure. And from an ontogenetic point of view, we know from longitudinal studies (e.g., Brooks & Meltzoff 2007, Carpenter et al. 1998a) that these forms of pre-verbal communication lay the foundation for the acquisition of language—language being a continuation of communication with other means. Children in their second year also begin to enter into collective intentional practices with proto-institutional structure, in particular, joint games, around the same time they begin to acquire language. Here their cognitive abilities, as indicated by their systematic actions, by far exceed explicit verbal competence. 4. From language to thought: the development of linguistic cognition The acquisition of a natural language has specific cognitive prerequisites: social cognition and, particularly, shared intentionality, the latter being a prime candidate that makes humans, and only humans, ready for becoming linguistic, or so I would like to argue. But language acquisition does of course have its cognitive consequences, too. From a philosophical point of view, the central question is how language constitutes new forms of thinking. In the following, I will deal with the much narrower, empirical psychological question of how language in fact changes and shapes cognition. Of special interest here are areas where there is some pre-verbal cognitive competence—that is, where some thinking 16. For similar results in the domain of simple non-pretence rule games, see Rakoczy et al. (2008).

92

without words is possible in the first place, in contrast to areas such as, say, theoretical physics where we do not even know what thinking preverbally could mean. Here, much recent research has focussed on the areas of numerical and spatial cognition. Regarding numerical cognition, many habituation studies have shown that pre-verbal infants and other primates share two simple abilities of representing numerosity—a so-called ‘subitizing’ system for simultaneously tracking very small numbers (< 4) of objects (this, again, is just more complex object permanence) and an approximate analogue magnitude system for roughly estimating the sizes of assemblies of individuals.17 But these systems are narrowly confined in their application, and of course a far cry from anything approaching mathematical cognition proper. One prominent hypothesis recently pursued in this context is that language, and in particular the acquisition of the counting routine, is the medium that transforms the limited cognitive abilities residing in these primitive systems into truly general mathematical competence (Carey 2001; Spelke 2003). Regarding spatial cognition, quite similarly, pre-verbal infants and other animal species share some basic though domain-limited abilities for spatial cognition (e.g., Wang & Spelke 2002). Here the acquisition of spatial language has been found to dramatically increase the flexibility and relational complexity of spatial thought (Hermer-Vazquez, Spelke & Kasnelson 1999; Spelke 2003; see also Carruthers 2002, for a general account along such lines).18 However, the one example that most clearly illustrates the need for a dialectical picture of the development of thought and language is the development of increasingly sophisticated forms of folk psychology. As we have seen, pre-verbal infants, and to some degree apes, develop some rudimentary folk psychology and interpret one another as individual intentional agents capable of perception and action. Some such rudimentary folk psychology is part of the foundation of language acquisition. But clearly, such rudimentary folk psychology is quite rudimentary indeed. Above all, it does not yet involve a full-fledged notion of propositional attitudes organized around belief as the most central one. Consequently, an explicit distinction between appearances and reality is still missing. Though 17. For an overview, see, e.g., Carey (2001); Feigenson, Dehaene & Spelke (2004). 18. Here I am dealing only with the, somewhat simplified, question of how language (singular) shapes cognition. Cross-linguistic research documents impressively how different languages (plural) differentially shape spatial cognition, e.g., through the acquisition of egocentric versus allocentric frames of reference (Haun et al., 2006).

93

it does allow for some rational action explanation based on perception and intention, it does not yet permit reconstructing others’ practical reasoning in the form of practical syllogisms based on premises in the form of belief and pro-attitude (Perner 1991). More sophisticated folk psychological abilities have been found to emerge ontogenetically around the age of four. Children this age begin to predict and rationally explain actions based on false beliefs (Wimmer & Perner 1983; for an overview, see Wellman et al. 2001) and to explicitly master the appearance-reality distinction (Flavell et al. 1983). Further developments in reflexive self-consciousness such as growing self-control and autobiographical memory go along with these developments (Perner 1991). Much recent research has documented the language-dependence of this developmental step.19 First, numerous correlation studies have found that 4-year olds’ folk psychology is highly correlated with language input and competence; here different studies have stressed different aspects of language, among them general linguistic competence, pragmatic competence and specific tensed ‘that’-complementation (deVilliers & deVilliers 2000). Still, pure correlations are difficult to interpret. A second line of research has more convincingly used a training methodology and found that training with specific linguistic material, mostly complemented propositional attitude discourse, significantly boosts false belief and related understanding (Hale & Tager-Flusberg 2003; Lohmann & Tomasello 2003). Finally, research with deaf populations has documented that while native signing deaf children develop false belief reasoning in parallel to hearing children, orally educated deaf children with delayed linguistic competence reveal a dramatic delay, sometimes of several years, in their development of folk psychology (deVilliers 2005).20 In sum, while simple forms of folk psychology—understanding conspecifics as perceiving and acting subjects—seems to be a common non-/ pre-linguistic cognitive heritage of human infants and other apes, sophisticated folk psychology, centred around the notion of belief, arises much later in human ontogeny, and is essentially dependent on language.21 19. For an overview, see Astington & Baird (2005). 20. So, Davidson might be partly right: you do need language to have the concept of a belief. But this does not mean that you need the concept of belief to have beliefs, as he claims. 21. For an elaborated philosophical argument that full-fledged belief-desire folk psychology (but not the more rudimentary infant folk psychology) requires a natural language as a metarepresentational medium, see Bermúdez (2003). For a similar approach trying to reconcile

94

5. Conclusion The recent findings in developmental and comparative psychology reviewed so far illustrate the need for a dialectical picture of the development of language and thought, a picture meeting the following three desiderata: (a) doing justice to the common pre-linguistic cognitive abilities in humans and other animals; (b) acknowledging species-specific human (social) cognitive abilities as part of the foundation of language acquisition, and (c) recognizing the ways speaking a language shapes and transforms thinking. (c) is relatively uncontroversial nowadays, except, perhaps, amongst hard-nosed nativist LOT theorists. Interestingly, after having been dead for some time, a moderate Whorfianism currently seems to be experiencing a renaissance in psychology, yielding a newly booming field of research on the role of language and languages in shaping cognition (see, e.g., Gentner & Goldin-Meadow 2003). Even many broadly Fodorian modularity theorists now assign a relatively substantial role to language in cognition. The basic idea here is that language functions as a domain-general glue to bind together encapsulated information from separate modules (e.g., Carey 2001; Carruthers 2002; Spelke 2003). Similarly, from a connectionist perspective, Andy Clark and Annett Karmiloff-Smith have argued that initial know-how represented in local networks gets re-described into domain-general know-that, inter alia through the acquisition of language (Clark & Karmiloff-Smith 1993; Karmiloff-Smith 1992; see also Dennett 1993). More controversial and challenging is the question of how we should describe the common pre-linguistic abilities in infants and other animals. Do they deserve the title ‘cognitive’? Radical lingualists say ‘no’. Only fullfledged participation in discursive practices counts as cognitive, descriptions of infant and animal rationality are discounted as metaphorical. What infants and animals do is no more than complex discrimination, on a par with iron’s responses to humidity. Thought and language come together as one big package. LOT theorists, in contrast, say ‘yes’. Infants and animals think just the same way as adults do, limited only in the scope and expressability of their thoughts. The truth probably lies in an intermediate position. On the one hand, the abilities of animals and infants in the domains of object tracking, philosophical approaches (in particular, Sellars) with psychological ones (in particular, Vygotsky) see Garfield et al. (2001).

95

problem-solving and social understanding—to name just a few reviewed here—clearly go beyond dumb discrimination, and even beyond Dummett’s (1993) ‘proto-thoughts’ of superimpositions of spatial images on spatial perceptions. In contrast to complex but largely pre-programmed and inflexible behaviours such as a spider’s nest-weaving, they deserve the title ‘cognitive’ because they are flexible, systematic, general, and compositionally as well as (proto-)inferentially structured in ways analogous to the structure of linguistic cognition (see Clark & Karmiloff-Smith 1992; Bermúdez 2003; Hurley 2003). Lingualist functionalism views thoughts as defined through their position in the systematic network of languageentry moves, language-language moves and language-exit moves (Brandom 1994; Sellars 1974). Analogously, folk and academic psychologists alike ascribe non-linguistic thoughts in line with their position in a systematic network of reliable responses to states of affairs, (proto-)inferential relations among each other, and practical reasoning relations to action. In contrast to the lingualists’ dichotomy—rusting iron, animals and infants fall on the discrimination side, adult speakers on the reason side—what this suggests is a gradation of cognitive abilities along phylogenetic and ontogenetic lines. ‘There is certainly a continuum of possibilities between proto-conceptual and fully conceptual behaviour (and in the case of some of the higher primates, it may be that the line is blurry)’ (Putnam 1999, 161). One of the pressing conceptual questions, though, is how to spell out exactly the non-linguistic analogues of full-blown conceptual and inferential abilities. In the absence of language as a formal vehicle, what model of drawing inferences should we look for other than relating sentences as premises and conclusions? Should non-linguistic (proto-)inferences be modelled on causal reasoning (Bermúdez 2003)? Or is there a viable truthfunctional notion of material inference applicable at the non-linguistic level? The following questions are just as difficult: in which ways exactly does non-linguistic thought—granted that there is such a thing—fall short of being full-fledged thought as we find it in speaking creatures? And how exactly does language re-shape cognition? Clearly, lack of determinacy and restrictions of scope set apart non-linguistic cognition from full-fledged thought. Malcolm’s and Wittgenstein’s dogs can think something like ‘The cat went up this tree’ and ‘My master is coming’, though they can’t think ‘The cat went up the biggest oak tree in this country’ and ‘My master might come the day after tomorrow’. But why exactly is this? Where is the line to be drawn between what can and what cannot be thought without 96

language22? Regarding language’s role in re-shaping thought, Putnam has noted that ‘one must not make the mistake of supposing that language is merely a ‘code’ that we use to transcribe thoughts we could perfectly well have without the code’ (1999, 48). But what exactly beyond such transcribing does language do? All in all, the problems we are facing in describing the emergence of thought are certainly deep ones, as Davidson noted, but in contrast to his categorical pessimism, they might not be insuperable. We know, at least in principle, what kind of conceptual and theoretical solutions we are in need of. To date, however, the fact that in talking about non-linguistic cognition we haven’t got beyond using ‘proto-’ so much—meaning not much more than ‘somehow but not fully like the real thing, but I don’t know exactly how and why’—highlights the challenges we are still facing. It is still ‘somewhat surprising how little we know about thought’s dependence on language’ (Schiffer 1994, 593).

REFERENCES Baillargeon, Renée. 1987. “Young infant’s reasoning about the physical and spatial properties of a hidden object”. Cognitive Development 2, 375–392. Bermúdez, José. 2003. Thinking without Words. Oxford: Oxford University Press. Boesch, Christophe and Hedwige Boesch. 1990. “Tool use and tool making in wild chimpanzees”. Folia Primatologica 54, 86–99. Brandom, Robert. 1994. Making it Explicit. Cambridge (Mass.): Cambridge University Press. Bratman, Michael. 1992. “Shared cooperative activity”. The Philosophical Review 101(2), 327–341. Brinck, Igar. 2004. “Triangulation and Radical Interpretation”. Dialectica 58(2), 179–206. Brooks, Rechele and Andrew Meltzoff. 2008. “Infant gaze following and pointing predict accelerated vocabulary growth through two years of age: A longitudinal, growth curve modeling study”. Journal of Child Language 35, 207–222. Bruner, Jerome. 1983. Child’s Talk. Oxford: Oxford University Press. Call, Joseph. 2004. “Inferences about the location of food in the great apes (Pan paniscus, Pan troglodytes, Gorilla gorilla and Pongo pygmaeus)”. Journal of Comparative Psychology 118(2), 232–241. 22. For an initial and incomplete attempt at listing scope restrictions of non-linguistic thought, see Searle (1994).

97

Call, Joseph, Brian Hare, Malinda Carpenter and Michael Tomasello. 2004. “‘Unwilling’ versus ‘unable’: Chimpanzees’ understanding of human intentional action”. Developmental Science 7(4), 488–498. Carey, Susan. 2001. “Evolutionary and ontogenetic foundations of arithmetic”. Mind and Language 16, 37–55. Carpenter, Malinda, Katherine Nagell and Michael Tomasello. 1998a. “Social cognition, joint attention, and communicative competence from 9 to 15 months of age.” Monographs of the Society for Research in Child Development, 63(4), 176. Carpenter, Malinda, Nameera Akhtar and Michael Tomasello. 1998b. “Fourteenthrough 18-month-old infants differentially imitate intentional and accidental actions”. Infant Behavior & Development 21(2), 315–330. Carpenter, Malinda, Michael Tomasello and Tricia Striano. 2005. “Role reversal imitation and language in typically developing infants and children with autism”. Infancy 8(3), 253–278. Carruthers, Peter. 2002. “The cognitive functions of language”. Behavioral and Brain Sciences 25(6), 657–726. Clark, Andy. 2004. “Feature-placing and proto-objects”. Philosophical Psychology 17(4), 443–469. Davidson, Donald. 1982. “Rational Animals”. Dialectica 36, 317–327. — 1994. “Davidson, Donald.” In: Samuel Guttenplan (ed.). A Companion to the Philosophy of Mind. Oxford: Blackwell. — 1999. “The Emergence of Thought”. Erkenntnis 51, 7–17. Dennett, Daniel. 1991. Consciousness Explained. Boston (Mass.): Little, Brown & Co. — 1993. “Learning and Labelling”. Mind and Language 8(4), 540–548. deVilliers, Jill G. and Peter A. deVilliers. 2000. “Linguistic determinism and the understanding of false beliefs.” In: Peter Mitchell and Kevin J. Riggs (eds.). Children’s reasoning and the mind. Hove, England: Psychology Press/Taylor & Francis (UK), 191–228. deVilliers, Peter A. 2005. “The Role of Language in Theory-of-Mind Development: What Deaf Children Tell Us.” In: Janet W. Astington and Jodie Baird (eds). Why Language Matters for Theory of Mind. Astington, New York: Oxford University Press, 266–297. Dummett, Michael. 1993. The Origins of Analytical Philosophy. London: Duckworth. Eilan, Naomi. 2005. “Joint attention, communication, and mind”. In: Naomi Eilan, Christoph Hoerl, Teresa McCormack and Johannes Roessler (eds.). Joint attention: Communincation and other mind. Oxford: Clarendon Press, 1–33. Feigenson, Lisa, Stanislas Dehaene and Elizabeth Spelke. 2004. “Core systems of number”. Trends in Cognitive Sciences 8(7), 307–314.

98

Flavell, John H., Barbara Everett, Karen Croft and Eleonor R. Flavell. 1981. “Young children’s knowledge about visual perception: Further evidence for the Level 1-Level 2 distinction”. Developmental Psychology 17(1), 99–103. Flavell, John H., Eleonor R. Flavell and Frances L. Green. 1983. “Development of the appearance-reality distinction.” Cognitive Psychology 15(1), 95–120. Fodor, Jerry A. 1975. The Language of Thought. New York: Crowell. Gallup, Gordon G. 1970. “Chimpanzees: Self-recognition”. Science 167, 417– 421. Garfield, Jay L., Candice Peterson and Tricia Perry. 2001. “Social cognition, language acquisition and the development of the theory of mind”. Mind and Language 16(5), 494–541. Gentner, Timothy, Kimberley Fenn, Daniel Margoliash and Howard Nusbaum. 2006. “Recursive syntatctic pattern learning by songbirds”. Nature 440, 1204– 1207. Gergely, György, Harold Bekkering and Ildikó Király. 2002. “Rational imitation of goal-directed actions”. Nature 415, 755. Gilbert, Margaret. 1990. “Walking Together: A Paradigmatic Social Phenomenon”. Midwest Studies in Philosophy 15, 1–14. Glock, Hans-Johann. 2000. “Animals, Thoughts and Concepts”. Synthese 123, 35–64. Gomez, Juan Carlos and Beatriz Martin-Andrade. 2005. “Fantasy play in animals”. In: Anthony Pellegrini and Peter K. Smith (eds.). The nature of play: Great apes and humans. New York: Guilford, 139–171. Goodall, Jane. 1986. The chimpanzees of Gombe. Patterns of behavior. Cambridge (Mass.): Belknap Press of Harvard University Press. Haight, Wendy and Peggy Jo Miller. 1992. “The development of everyday pretend play: A longitudinal study of mothers’ participation”. Merrill Palmer Quarterly 38(3), 331–349. Hale, Courtney M. and Helen Tager-Flusberg. 2003. “The influence of language on theory of mind: A training study”. Developmental Science 6(3), 346–359. Hare, Brian, Joseph Call, B. Agnetta and Michael Tomasello. 2000. “Chimpanzees know what conspecifics do and do not see”. Animal Behaviour 59(4), 771–785. Harris, Paul L. and Robert D. Kavanaugh. 1993. “Young children’s understanding of pretense”. Monographs of the Society for Research in Child Development 58(1)[231]), v–92. Haun, Daniel, Christian Rapold, Joseph Call, Gabriele Janzen and Stephen Levinson. 2006. “Cognitive cladistics and cultural override in Hominid spatial cognition”. Proceedings of the National Academy of Sciences of the USA 103, 17568–17573 Hauser, Marc, Noam Chomsky and Tecumseh Fitch. 2002. “The faculty of language: What is it, who has it, and how did it evolve?” Science 298, 1569–1579.

99

Hermer-Vazquez, Linda, Elizabeth Spelke and Alla S. Knatsnelson. 1999. “Sources of flexibility in human cognition: Dual-task studies of space and language”. Cognitive Psychology 39(1), 3–36. Hurley, Susan. 2001. “Overintellectualizing the mind”. Philosophy and Phenomenological Research LXII(2), 423–431. — 2003. “Animal Action in the Space of Reason”. Mind and Language 18(3), 231–257. Köhler, Wolfgang. 1926. The mentality of apes. London: Routledge & Kegan Paul. Liszkowski, Ulf, Malinda Carpenter, Anne Henning, Tricia Striano and Michael Tomasello. 2004. “Twelve-month-olds point to share attention and interest”. Developmental Science 7(3), 297–307. Lohmann, Heidemarie and Michael Tomasello. 2003 “The Role of Language in the Development of False Belief Understanding: A Training Study”. Child Development 74, 1130–1144. Meltzoff, Andrew. 1995. “Understanding the intentions of others: re-enactment of intended acts by 18-month-old children”. Developmental Psychology 31, 838–850. Mendes, Natacha, Daniel Hanus and Joseph Call. 2007. “Raising the level: Orangutans use water as a tool”. Biology Letters 3, 453–455. Mendes, Natacha, Hannes Rakoczy and Joseph Call (2008). “Ape metaphysics: Object individuation without language”. Cognition 106(2), 730–749. Mitchell, Robert (ed.). 2002. Pretending and imagination in animals and children. Cambridge: Cambridge University Press. Moll, Henrike and Michael Tomasello. 2006. “Level I perspective-taking at 24 months of age”. British Journal of Developmental Psychology 24(3), 603–613. Mulcahy, Nicholas and Joseph Call. 2006. “Apes save tools for future use”. Science 312(5776), 1038–1040. Perner, Joseph. 1991. Understanding the Representational Mind. Cambridge (Mass.): MIT Press. Phillips, Webb and Laurie Santos. 2007. “Evidence of kind representations in the absence of language: Experiments with rhesus monkeys (Macaca mulatta)”. Cognition 102(3), 455–463. Piaget, Jean. 1952. The Origins of Intelligence. New York: Basic Books. Putnam, Hilary. 1999. The Threefold Cord: Mind, Body, and World. New York: Columbia University Press. Pylyshyn, Zenon. 2000. “Situating vision in the world”. Trends in Cognitive Science 4(5), 197–207. Quine, Willard V. O. 1990. Pursuit of Truth. Cambridge (Mass.): Harvard University Press. Rakoczy, Hannes. 2006. “Pretend play and the development of collective intentionality”. Cognitive Systems Research 7, 113–127.

100

Rakoczy, Hannes. 2007. “Play, games, and the development of collective intentionality”. In: Chuck Kalish and Mark Sabbagh (eds.). Conventionality in cognitive development: How children acquire representations in language, thought and action. New Directions in Child and Adolescent Development, No. 115. San Francisco: Jossey-Bass, 53–67. — 2008. “Taking fiction seriously: Young children understand the normative structure of joint pretend games”. Developmental Psychology 44(4), 1195–1201. Rakoczy, Hannes and Michael Tomasello. 2006. “Two-year-olds grasp the intentional structure of pretense acts”. Developmental Science 9(6), 558–565. — 2007. “The ontogeny of social ontology: Steps to shared intentionality and status functions”. In: Savas L Tsohatzidis (ed). Intentional Acts and Institutional Facts: Essays on John Searle’s Social Ontology. Berlin: Springer Verlag, 113–137. Rakoczy, Hannes, Michael Tomasello and Tricia Striano. 2004. “Young children know that trying is not pretending—a test of the “behaving-as-if ” construal of children’s early concept of “pretense””. Developmental Psychology 40(3), 388–399. — 2006. “The role of experience and discourse in children’s developing understanding of pretend play actions”. British Journal of Developmental Psychology 24(2), 305–335. Rakoczy, Hannes, Felix Warneken and Michael Tomasello. 2008. “The sources of normativity: Young children’s awareness of the normative structure of games”. Developmental Psychology 44(3), 875–881. Rivas, Esteban. 2005. “Recent use of signs by chimpanzees (Pan troglodytes) in interactions with humans”. Journal of Comparative Psychology 119(4), 404–417. Roessler, Johannes. 2005. “Joint attention and the problem of other minds”. In: Naomi Eilan, Christoph Hoerl, Teresa McCormack and Johannes Roessler (eds.). Joint attention: Communincation and other minds. Oxford: Clarendon Press, 230–259. Schiffer, Stephen. 1994. “Thought and Language”. In: Simon Guttenplan (ed.). A Companion to the Philosophy of Mind. Oxford: Blackwell. Searle, John. 1969. Speech Acts: An Essay on the Philosophy of Language. Cambridge (Mass.): Cambridge University Press. — 1994. “Animal Minds”. Midwest Studies in Philosophy, 14, 206–219. — 1983. Intentionality: An Essay in the Philosophy of Mind. Cambridge: Cambridge University Press. — 1990. “Collective Intentions and Actions”. In: Philip Cohen, Jerry Morgan and Martha Pollack (eds.). Intentions in Communication. Cambridge (Mass.): MIT Press, 401–415. — 1995. The Construction of Social Reality. New York: Free Press. — 2005. “What is an Institution”. Journal of Institutional Economics 1(1), 1–22.

101

Sellars, Wilfrid. 1974. “Meaning as Functional Classification”. Synthese 27, 417–437. Spelke, Elizabeth. 2003. “What makes us smart?” In: Dedre Gentner and Susan Goldin-Meadow (eds.). Language in mind. Advances in the study of language and thought. Cambridge (Mass.): MIT Press, 277–311. Strawson, Peter F. 1959. Individuals: An Essay in Descriptive Metaphysics. London: Methuen. Tollefsen, Deborah. 2004. “Collective intentionality”. The Internet Encyclopedia of Philosophy. Retrieved 12/22/2005 from http://www.iep.utm.edu. Tomasello, Michael. 1999. The cultural origins of human cognition. Cambridge (Mass.): Harvard University Press. — 2003. Constructing a language: A usage-based theory of language acquisition. Cambridge (Mass.): Harvard University Press. Tomasello, Michael and Joseph Call. 1997. Primate cognition. New York: Oxford University Press. Tomasello, Michael and Malinda Carpenter. 2005. “The emergence of social cognition in three young chimpanzees”. Monographs of the Society for Research in Child Development 70(1), 1–136. Tomasello, Michael, Malinda Carpenter, Joseph Call, Tanya Behne and Henrike Moll. 2005. “Understanding and sharing intentions: The origins of cultural cognition”. Behavioral and Brain Sciences, 28, 675–691. Tomasello, Michael and Hannes Rakoczy. 2003. “What makes human cognition unique? From individual to shared to collective intentionality”. Mind and Language 18(2), 121–147. Tuomela, Raimo and Kaarlo Miller. 1988. “We-intentions”. Philosophical Studies 53, 367–389. Van de Walle, Gretchen A., Susan Carey and Meredith Prevor. 2000. “Bases for object individuation in infancy: Evidence from manual search”. Journal of Cognition and Development 1(3), 249–280. Vygotsky, Lev. 1934/1986. Thought and Language. Cambridge (Mass.): MIT Press. Walton, Kendall L. 1990. Mimesis as Make-Believe. Cambridge (Mass.): Harvard University Press. Wang, Ranxio F. and Elizabeth Spelke. 2002. “Human spatial representation: Insights from animals”. Trends in Cognitive Sciences 6(9), 376–382. Warneken, Felix, Frances Chen and Michael Tomasello. 2006. “Cooperative activities in young children and chimpanzees”. Child Development 77(3), 640–663. Watson, J., G. Gergely, V. Csanyi, J. Topal, M. Gasci and Z. Sarkozi. 2001. “Distinguishing logic from association in the solution of an invisible displacement task by children (Homo sapiens) and dogs (Canis familiaris): Using negation of disjunction”. Journal of Comparative Psychology 115(3), 219–226.

102

Weir, Alex and Alex Kacelnik. 2002. “Shaping of hooks in New Caledonian crows”. Science 297, 981. Wellman, Henry, David Cross and Julanne Watson. 2001. “Meta-analysis of theory-of-mind development: The truth about false belief ”. Child Development 72(3), 655–684. Wiggins, David. 1997. “Sortal concepts: A reply to Xu”. Mind and Language 12(3-4), 413–421. Willatts, Peter. 1985. “Adjustment of means-ends coordination and the representation of spatial relations in the production of search errors by infants”. British Journal of Developmental Psychology 3(3), 259–272. Xu, Fei. 1997. “From Lot’s wife to a pillar of sand: Evidence that physical object is a sortal concept”. Mind and Language 12(3/4), 365–392. — 2002. “The role of language in acquiring object kind concepts in infancy”. Cognition 85(3), 223–250. Xu, Fei and Allison Baker. 2005. “Object individuation in 10-month-old infants using a simplified manual search method”. Journal of Cognition and Development 6(3), 307–323. Xu, Fei and Susan Carey. 1996. “Infants’ metaphysics: The case of numerical identity”. Cognitive Psychology 30(2), 111–153.

103

Grazer Philosophische Studien 81 (2010), 105–113.

CONCEPTS, BRAINS, AND BEHAVIOUR Anthony KENNY St. John’s College, Oxford Summary Concepts are best understood as a particular kind of human ability: a person who has mastered the use of a word for F in some language possesses the concept of F. Abilities are individuated by their possessors and their exercises, though they are not to be identified with either. Typically abilities are associated with vehicles, that is to say underlying actualities which account for their exercises. The mind is the human ability to form concepts, and its principal vehicle is the brain; but the mind should not be identified with the brain any more than it should be identified with the behaviour in which its concepts are expressed.

1. The mind as a capacity: concepts as abilities It is common nowadays to reject a Cartesian notion of the mind. Few are prepared to declare themselves dualists: to maintain that in addition to material substances there are independent mental substances that are only contingently related to material bodies. But many different theories may lie behind an official rejection of the dualism of Descartes. We have to ask, what does a non-Cartesian account of the mind look like? If the mind is not a substance, what kind of thing is it? The answer is that the mind is a capacity. (That, as Wittgenstein would say, is a grammatical remark). I call it a capacity rather than an ability because it is a comprehensive ability to acquire abilities. One might be tempted to call it a second order ability, but that would be to underestimate the possible stratification of abilities. There can be abilities to acquire abilities to acquire abilities … without limit. The mind is the capacity to acquire linguistic and symbolic skills, which are themselves abilities of a certain kind. Language and symbolic skills such as arithmetic are themselves generalised abilities to acquire more specific abilities—as one does when one first masters long division or learns the meaning of a new word. We may use ‘concept’ as a term for the specific

abilities that are particular exercises of the universal capacity that is the mind. A sufficient, but not a necessary condition for a person to possess the concept of F is that she shall have mastered the use of a word for ‘F’ in some language. Abilities and capacities are individuated by their possessors and their exercises, but they are distinct from both. The possessor of an ability is what has the ability: in the case of the mind, the human being whose mind it is. It is I who have a mind, know English, possess certain concepts and am exercising these abilities in writing this paper. The thing, the substance, that knows English is not my mind, but I myself. (Myself, not ‘my self ’ whatever that might be.) My ability to speak English is a different individual item from your ability or President Obama’s ability: it is not just that we may differ in articulateness or fluency, but that there are three different items with different scopes, items which may vary, or cease to exist, independently of each other. That is the sense in which abilities are individuated by their possessors: but of course to say that there are in the world n abilities to speak English is only a fancy way of saying that there are n English speakers in the world. Abilities are individuated not only by their possessors but by their exercises. Which ability we are talking about, on any occasion, is explained by specifying what would count as an exercise of that ability. The medicinal capacities of aspirin, for example, are exercised in episodes of pain-killing. Similarly, the speaking, understanding, writing or reading of English counts as an exercise of the knowledge of English. An ability is a more or less enduring state, while any particular exercise of an ability will be a datable event or process. Aspirins possess their capacity while bottled up in the medicine cabinet, but only begin to exercise it after being swallowed. The human mind lasts a lifetime, while its exercises are our fleeing thoughts and projects and the other variegated items of our mental lives. 2. Capacities and their vehicles A capacity is clearly something distinct from its possessor and from its exercise, but it must also be distinguished from a third thing, which we may call its vehicle. Consider the capacity of whisky to intoxicate. The vehicle of this capacity to intoxicate is the alcohol that the whisky contains: it is the ingredient in virtue of which the whisky has the power to intoxicate. The vehicle of a power need not be a substantial ingredient like alcohol 106

which can be physically separated from the possessor of the power, though it is in such cases that the distinction between a power and its vehicle is most obvious (one cannot, for example, weigh the power of whisky to intoxicate as one can weigh the alcohol it contains). Take the less exciting power which my wedding ring has of fitting on my finger. It has that power in virtue of having the size and shape it has, and size and shape are not modal, relational, potential properties in the same way as being able to fit on my finger is. They are not the power, but the vehicle of the power. I first introduced the concept of vehicle many years ago in a book called Will, Freedom, and Power. Having explained it as above, I went on to say The connection between a power and its vehicle may be a necessary or contingent one. It is a contingent matter, discovered by experiment, that alcohol is the vehicle of intoxication; but it is a conceptual truth that a round peg has the power to fit into a round hole. (Kenny 1975, 10)

It did not take long to discover that this was an error. It is indeed a conceptual truth that a round peg has the power to fit into a round hole; but I should have realised that from that very fact it followed that roundness was not the vehicle of the power to fit into a round whole. For the whole point of distinguishing between powers and their vehicles was to separate conceptual features of powers from the empirically necessary conditions of their exercise. When I returned to the topic in The Metaphysics of Mind I wrote The connection between a capacity and its exercises is a more intimate one than the connection between a capacity and its vehicle. In the case of the mind the connection between the capacity and its exercise is a conceptual one: one could not understand what the mind was if one did not understand what kinds of thing constitute the exercise of mental capacity. The connection between capacity and vehicle, on the other hand, is a contingent one, discoverable by empirical science. (Kenny 1989, 74)

A vehicle is something concrete, something that can be located and measured. An ability, on the other hand, has neither length nor breadth nor location. This does not mean that an ability is something ghostly: my front-door key’s ability to open my front door is not a concrete object, but it is not a spirit either.

107

3. Reductionism and transcendentalism In every age philosophers have been tempted to blur the distinctions between powers and their exercises and their vehicles. There is a perennial tendency to reduce potentialities to actualities. There are two kinds of reductionism: one reduces abilities to their exercises, the other reduces abilities to their vehicles. Hume, who said that the distinction between a power and its exercise was frivolous, was an exercise-reductionist. Descartes, who identified the powers of bodies with their geometrical properties, was a vehicle-reductionist. While it is important to distinguish between powers and their vehicles and exercises it is important also not to err on the other side. A power or capacity must not be thought of as something in its own right, for instance as a flimsy actuality or insubstantial vehicle. The difference between a power on the one hand, and its exercise or vehicle on the other is a category difference, not a difference like that between solid and shadow. One can err on the anti-reductionist side by hypostatising powers, by thinking, say, of a power as something which, while remaining numerically the same, might pass from one person to another—in the way in which in the Andersen fairy tale the witch takes the old wife’s gift of the gab and gives it to the water-butt. One can make a similar ‘transcendentalist’ error by thinking of a particular ability as the kind of thing somebody might have two of, asking for instance how many senses of humour Oliver Cromwell had. 4. Actualities and potentialities The notion of a vehicle, as I have explained it, is an extension of Aristotle’s exploration of the differences between actuality and potentiality. The vehicle of a power is the abiding actuality in virtue of which a substance possesses a potentiality which finds expression in transitory exercises. This underlying actuality may be an ingredient, or a property, or a structure. The identification of the actuality provides an explanation of the possession of the power or active potentiality. Molière, in Le Malade Imaginaire, mocked at physicians who explained that opium put people to sleep because it had dormitive power. The statement was philosophically correct, but medically vacuous. The connection between a power and its exercise is not a causal one like that linking 108

smoking and cancer. The job of the physicians was to look for the vehicle of the power, not to hypostatise the power itself, treating it as a substance in its own right. The vehicle was identified when the soporific ingredient of opium was discovered to be morphine. The following question may be raised. Pure morphine puts people to sleep just as opium does, so surely their soporific powers have the same vehicle. It would seem to follow that in the case of morphine the agent and the vehicle are one and the same thing. This would in effect reproduce the quack pharmacology at a lower level—only, now it is the morphine rather than the opium that has the dormitive power. Such an objection rests on a misunderstanding of the Aristotelian framework which I introduced, tacitly, at the very beginning of this paper. Actualities and potentialities are, for Aristotle, stratified. Something that is a potentiality with respect to a particular exercise may itself be an actuality of a more primitive potentiality. Aristotle’s favourite example is the one I have constantly employed in this paper: knowledge of a language. This is a potentiality, exercised in the actual, episodic, use of the language. But it is itself an actuality, a skill acquired on the basis of the underlying human, species-specific, capacity for language learning. So too with each individual concept in a person’s repertoire: it is a very specific skill, an actuality acquired at a certain stage of the subject’s development; but it is also a potentiality, exercised in the application of the concept on particular occasions. The Aristotelian system of stratification can be applied to the vehicles of powers. Morphine is the vehicle of opium’s dormitive power because it is the abiding ingredient that explains opium’s ability to put people to sleep. But it is not its own vehicle precisely because ‘morphine always contains morphine’, while true, provides no explanation of its soporific qualities. To discover the vehicle of these powers, one has to move to a different level, to morphine’s molecular structure and its relationship to the anatomy of the animal brain. 5. Mind and behaviour Philosophical errors about capacities show up particularly vividly when they occur in the philosophy of mind. Cartesian dualism commits the transcendentalist error: it treats a capacity as if it were a substance. But nowadays reductionist errors are more common. Applied in this area,

109

exercise-reductionism is behaviourism: the attempt to identify mind with its particular exercises in behaviour. Applied in this area, vehicle-reductionism becomes materialism: the reduction of our mental capacities to the parts and structures of our bodies in virtue of which we possess them, and in particular to our brains. The identification of mind and brain is a category mistake, because the brain is a material object and the mind is a capacity. The mind is the capacity to acquire intellectual abilities. The possessor of human mental capacities is neither the mind nor the brain but the human being. ‘Only of a living human being’ Wittgenstein wrote ‘and what resembles (behaves like) a living human being can one say: it has sensations; it sees; is blind; hears; is deaf; is conscious or unconscious’ (PI, § 281). This does not mean that Wittgenstein is a behaviourist: he is not identifying experience with behaviour, or even with dispositions to behave. The point is that what experiences one can have depends on how one can behave. Only someone who can play chess can feel the desire to castle; only a being that can discriminate between light and darkness can have visual experiences. The relation between experiences of certain kinds, and the capacity to behave in certain ways, is not a merely contingent connection. Wittgenstein made a distinction between two kinds of evidence that we may have for the obtaining of states of affairs, namely symptoms and criteria. Where the connection between a certain kind of evidence and the conclusion drawn from it is a matter of empirical discovery, through theory and induction, the evidence may be called a symptom of the state of affairs; where the relation between evidence and conclusion is not something discovered by empirical investigation, but is something that must be grasped by anyone who possesses the concept of the state of affairs in question, then the evidence is not a mere symptom, but a criterion of the event in question. Such is the relation between scratching and itching, between crying and pain. Exploiting the notion of criterion enabled Wittgenstein to steer between the Scylla of dualism and the Charybdis of behaviourism. He agreed with dualists that particular mental events could occur without accompanying bodily behaviour; on the other hand he agreed with behaviourists that the possibility of ascribing mental acts to people depends on such acts having, in general, a behavioural expression.

110

6. Mind and brain If it is wrong to identify the mind with behaviour, it is even more mistaken, according to Wittgenstein, to identify the mind with the brain. Such materialism is in fact a grosser philosophical error than behaviourism because the connection between mind and behaviour is a more intimate one than that between mind and brain. The relation between mind and behaviour is a criterial one, something prior to experience; the connection between mind and brain is a contingent one, discoverable by empirical science. Any discovery of links between mind and brain must take as its starting point the everyday concepts we use in describing the mind, concepts which are grafted on to behavioural criteria. Developments in the philosophy of mind since Wittgenstein have shown that it is possible to combine the errors of materialism with those of dualism. One of the standard dualist misunderstandings of the nature of the mind is the picture of mind’s relation to body as that between a little person or homunculus on the one hand, and a tool or instrument or machine on the other. This misunderstanding is compounded if we assign the role of homunculus to the brain, identifying it, as materialists do, with the mind so conceived. What is wrong with the homunculus fallacy? In itself there is nothing misguided in speaking of images in the brain, if one means patterns in the brain, observable to a neurophysiologist, that can be mapped onto features of the visible environment. What is misleading is to take these mappings as representations, to regard them as visible to the mind, and to say that seeing consists in the mind’s perception of these images. The misleading aspect is that such an account pretends to explain seeing, but the explanation reproduces exactly the puzzling features it was supposed to explain. Sometimes cognitive scientists write as if the relation between mind and brain was that the mind made inferences from events in the brain and nervous system. This is a form of the homunculus fallacy, and it was explicitly rejected by Wittgenstein. An event leaves a trace in the memory: one sometimes imagines this as if it consisted in the event’s having left a tract, an impression, a consequence in the nervous system. As if one could say: the nerves too have a memory. But then when someone remembered an event, he would have to infer it from this impression, this trace. Whatever the event does leave behind in the organism, it isn’t the memory. (RPP I, § 220)

111

When something is recorded on a tape, the alteration in the tape is not a memory, and when the tape is played it is not remembering. Not all cognitive scientists go so far as explicitly to identify the mind with the brain, but all seek a parallelism between mental and physical events. Wittgenstein rejected this. ‘Nothing seems more possible to me’ he wrote ‘than that people some day will come to the definite opinion that there is no copy in either the physiological or the nervous systems which corresponds to a particular thought, or a particular idea, or memory’. (LW 504) The history of the mind is not a history of events in the way that the history of the body is. A thought does not have temporal parts as the utterance of a sentence does. Of course, there are such things as mental events and processes—hitting on an idea, or reciting a nursery rhyme in the imagination—but such events and processes are what they are because of the abilities from which they issue and which provide their background. In conclusion, we may pose a fundamental question. Is it the case that every mental ability does have a physical vehicle? Wittgenstein called this in question. No supposition seems to me more natural than that there is no process in the brain correlated with associating or with thinking; so that it would be impossible to read off thought processes from brain processes… It is perfectly possible that certain psychological phenomena cannot be investigated physiologically, because physiologically nothing corresponds to them. I saw this man years ago: now I have seen him again, I recognize him, I remember his name. And why does there have to be a cause of this remembering in my nervous system? (…) Why should there no be a psychological regularity to which no physiological regularity corresponds? (Z, §§ 608ff., cf. RPP I, §§ 903–6)

So Wittgenstein is willing to countenance a case of a causality between psychological phenomena unmediated physiologically. He is anxious to say that this does not amount to the ‘admission of the existence of a soul alongside the body, a ghostly, spiritual nature.’ (RPP I, § 906) The entity that does the associating, thinking, and remembering is not a spiritual substance, a la Plato and Descartes, but a corporeal human being. But this passage seems to envisage the possibility of an Aristotelian soul or entelechy, which operates with no material vehicle, a formal and final cause to which there corresponds no mechanistic efficient cause. I must confess that, like many of Wittgenstein’s critics, I find this suggestion flies in the face of my understanding of causality. But who am I to challenge

112

Wittgenstein’s immediately following remark ‘If this upsets our concept of causality then it is high time it was upset’? (Z § 610).

REFERENCES Kenny, Anthony. 1975. Will, Freedom, and Power. Oxford: Blackwell. — 1989. The Metaphysics of Mind. Oxford: Oxford University Press. Wittgenstein, Ludwig. [Z]. Zettel. G. E. M Anscombe and G. H. von Wright (eds.), Oxford: Blackwell. 1967. — [RPP I]. Bemerkungen über die Philosophie der Psychologie I, Remarks on the Foundations of Psychology, I. G. H. von Wright and Heikki Nyman (eds.), Oxford: Basil Blackwell. 1980. — [RPP II]. Bemerkungen über die Philosophie der Psychologie II, Remarks on the Foundations of Psychology, II. G. H. von Wright and Heikki Nyman (eds.), Oxford: Blackwell. 1980. — [PI]. Philosophische Untersuchungen, Philosophical Investigations, 2nd ed. Oxford: Blackwell. 1958. — [LW]. Last Writings on the Philosophy of Psychology. G. H. von Wright and Heikki Nyman (eds.), Oxford: Blackwell. 1982.

113

Grazer Philosophische Studien 81 (2010), 115–134.

CONCEPTS, ABILITIES, AND PROPOSITIONS Hans-Johann GLOCK Zurich University Summary This article investigates whether the concept of a concept can be given a fairly uniform explanation through a ‘cognitivist’ account, one that accepts that concepts exist independently of individual subjects, yet nonetheless invokes mental achievements and capacities. I consider various variants of such an account, which identify a concept, respectively, with a certain kind of abilitiy, rule and way of thinking. All of them are confronted with what I call the ‘proposition problem’, namely that unlike these explananda concepts are standardly regarded as components of propositions. The paper ends by suggesting that this problem can be resolved by recognizing the different ways in which concepts can be involved in judgements or propositions, and by undermining the building-block model of concepts as abstract parts of abstract wholes.

1. The project My aim is to elucidate the ordinary use or uses of the term ‘concept’, and those of its equivalents and cognates. This ordinary use includes everyday uses. As Ryle pointed out, however, the ordinary use which is important to philosophy also includes the established uses of terms in specialized forms of discourse, including technical uses (1971, 301–4). In the case of ‘concept’, these special disciplines include the history of ideas, psychology, logic and philosophy. These days many philosophers vehemently reject an interest in ordinary language. Unfortunately, few of them provide arguments of any kind. A laudable exception to this rule is Christopher Peacocke. He rejects an investigation of the ordinary use of ‘concept’ on the grounds that it is ‘something of a term of art’ which lacks ‘a unique sense that is theoretically important’. His evidence is a Woody Allen movie in which a character from the entertainment industry says: ‘Right now, it’s only a notion, but I think I can get money to make it into a concept, and later

turn it into an idea’ (Peacocke, 1992, 1f.). In my view, by contrast, the quote is funny precisely because it presupposes a distinction between concept, notion and idea that we do not recognize in everyday speech, and which we suspect to be obscure even to the speaker himself. What we should concede is that the everyday use of ‘concept’ involves different strands, and this is even more obvious for its use in specialized disciplines. Furthermore, some of these uses may not just be distinct but downright incompatible. That would be all the more reason, however, to disentangle these different strands. Furthermore, it is rare even for technical uses of ‘concept’ to be purely stipulative. For these uses are part of philosophical or psychological theories that purport to explain cognitive and semantic phenomena that are commonly described with the help of the concept of a concept (concerning, e.g., cognitive development or synonymy). This makes it imperative to chart the similarities and differences between novel and established senses of the term, since we need to clarify whether the novel theories even address the phenomena they aspire to explain.1 Like some related terms with a philosophical provenance—notably ‘idea’—but unlike others—notably ‘universal’—‘concept’ is widely employed in everyday parlance. In many everyday contexts, moreover, ‘concept’ means roughly: general term with a meaning. This definition also fits the role of ‘concept’ in logic in so far as concepts have an extension and are components of sentences (see below). Yet it does not capture crucial uses in philosophy and psychology, where concepts are supposed to cut across different natural languages. In these contexts, ‘concept’ is more closely aligned with ‘idea’ than with ‘word’ or ‘term’. To be sure, a venerable tradition reaching from Occam to Fodor would allow us to extend the definition to these uses, since it postulates a mental language that is shared by all creatures capable of conceptual thought. Elsewhere I have argued, however, that the idea of a universal language of thought is incoherent and that in their semantic capacity concepts are not themselves linguistic symbols, but things represented or signified by such symbols (Glock, 2006, 45–8; 2009, 23–9). I shall not repeat this criticism here; instead I simply set aside concepts qua general terms for current purposes. At the same 1. But I shall ignore here to what extent the authors to which I attribute certain definitions need to pay heed to ordinary use for their own theoretical purposes. Accordingly, pointing out that such a definition fails to capture ordinary use simply serves to advance my search for a definition that does, and is not necessarily meant as a criticism of the author to whom the unsuccessful definition is attributed.

116

time, in pursuing the idea of concepts that transcend languages, I shall concentrate on concepts expressed by general terms, which are sometimes called ‘predicative concepts’. It is relatively uncontroversial that such concepts are involved when rational creatures entertain thoughts like (1) Dogs bark. The nature of this involvement remains controversial, however. In the history of philosophy, one can distinguish three fundamental approaches to concepts. According to subjectivist conceptions, concepts are mental phenomena, particular entities or goings-on in the mind or in the head of individuals. According to objectivist conceptions, concepts exist independently of human minds, as self-subsistent abstract entities. Finally, there is an intermediate position, which may be termed cognitivist.2 It agrees with objectivism in denying that concepts are mental particulars, while at the same time maintaining, with subjectivism, that they have an ineliminable mental or cognitive dimension. One version of cognitivism is intersubjectivism. It holds that concepts exist independently of individual rational subjects, but insists that they are constituted by intersubjective linguistic practices. Another version brackets the question of existence, yet holds that what concepts are—their essence, if you wish—can be explained only by reference to the operations and capacities of rational subjects. It is this less committal idea that I seek to explore in this article. In sections 2–4 I discuss and ultimately reject the popular cognitivist proposal that concepts are simply identical with a certain kind of ability. Sections 5–6 consider three other cognitivist proposals, namely that concepts are identical, respectively, with tools, techniques or rules, or ways of thinking. Section 7 turns to a challenge confronting all four cognitivist proposals, which I call the proposition problem: unlike abilities, tools, techniques or rules and ways of thinking, concepts appear to be constituents of propositions. In section 8 I argue that the ultimate solution to this problem is to treat both propositions and concepts as logical constructions from conceptual abilities. I end by indicating how concepts could be ways of thinking and yet occur in propositions, and thereby suggest how seemingly incompatible features of the ordinary use of ‘concept’ might be reconciled. 2. In previous publications I have used the label ‘pragmatism’, reluctantly following the lead of Fodor. But ‘cognitivism’ avoids inapposite connotations. It is also superior to talk of ‘epistemic conceptions’, since not all conceptual judgement amounts to knowledge.

117

2. Concepts and abilities The most popular and perhaps the most natural version of cognitivism identifies concepts with abilities (See also Kenny, this volume). Thus, in response to the question ‘Are concepts entities or are they dispositions?’ Price states in no uncertain terms: ‘a concept is not an entity (…) but a disposition or capacity’ (1953, 320, 348). In the same vein Geach pronounces that concepts ‘are capacities exercised in acts of judgement’ (1957, 7, see also 13). This proposal respects several features of established use. First, properties are objective, something possessed by things of all kinds. By contrast, concepts are something possessed by rational subjects capable of classifying things according to their properties. This is simply a crucial aspect of the cognitive dimension of concepts stressed by cognitivism.3 Secondly, the identification of concepts and capacities does not fall foul of the constraint that concepts must be shareable.4 As Geach points out, it does not entail that ‘it is improper to speak of two people as “having the same concept”’, since different individuals can possess the same mental capacities (1957, 14). Thirdly, concepts and abilities alike can be acquired, applied and lost, and some of them may be innate. Finally, to possess a concept is to possess a certain kind of mental ability, capacity or disposition. In what follows, I refrain from deciding which of these types of potentiality is the most appropriate general category (see Glock 2010b, sct. 5). Barring that issue, identifying concept-possession with an ability, capacity or disposition of some kind is inevitable.5 Concepts are involved not just in occurrent thoughts or beliefs, but also in longstanding or dispositional beliefs. Consequently, the possession of concepts must be at least as stable as the possession of dispositional beliefs. Put in Aristotelian terms, concept-possession must be a potentiality of some kind, since it combines two features. On the one hand, it is enduring rather 3. The identification of concepts with properties is one of the most popular versions of objectivism. For a more elaborate critique of it see Glock 2010b. 4. By contrast to those versions of subjectivism which treat concepts as mental particulars. See Glock 2009. 5. That concept-possession is an ability of some kind is accepted, willy-nilly, even by Fodor, who purports to contradict cognitivism (a.k.a. ‘pragmatism’) on this issue (2003, 19). The real bone of contention between representationalists like Fodor and those cognitivists that can properly be called ‘concept pragmatists’ in a wide sense—e.g. Wittgenstein, Ryle, Travis and Brandom—concerns the question of whether concept-possession is simply the ability to represent the property of being F or Fs as Fs, or whether it should be explained as the ability to classify things into those which are Fs and those which aren’t, or to draw inferences from thoughts about Fs.

118

than episodic. On the other hand, it is something which manifests itself in certain episodes, notably of overt or silent classification or inference. 3. The individuation of concepts and abilities Its prima facie plausibility notwithstanding, the identification of concepts with abilities faces an objection concerning principles of individuation. Thus subjectivists like Fodor have alleged that concepts are more finely individuated than abilities. For instance, ‘creature with a kidney’ and ‘creature with a heart’ apply to all and only the same things, but they express different concepts. Furthermore, ‘equilateral triangle’ and ‘equiangular triangle’ apply necessarily to the same things, yet they still express different concepts. In current jargon, concepts are not just ‘intensional’ but ‘hyperintensional’. Now, an ability is individuated by reference to its exercise. But, Fodor maintains, the same sorting and inferential performances can manifest the possession of different concepts. Confining ourselves to the ability to sort or discriminate, sorting equilateral triangles from all other figures is also sorting equiangular triangles from all other figures (2003, 25f., 143–6). It seems to follow that concepts cannot be individuated by the exercise of an ability, and hence that they cannot be individuated by reference to abilities. In effect, Fodor’s objection runs as follows: P1: Abilities are individuated by their exercise (ability to ) = ability to < iff )ing =
119

thereby the possession of different concepts. And it is obvious that one can have one of these abilities or concepts without having the other. Indeed, most children actually learn how to measure lengths before learning how to measure angles. But individuation also poses another challenge to identifying concepts with abilities. Many cognitivists grant that there is no precise way of individuating abilities. Thus Travis (2000) grants that linking concepts to abilities may not be much help in individuating concepts, since it is not clear how abilities are to be counted. That concession needs to be put in perspective, however. Like Travis, Geach (1957, 15) accepts that it is absurd to ask how many abilities are exercised in a judgement. Yet he also insists, rightly, that we can still distinguish between such abilities. More generally, one must distinguish between the possibility of enumerating and the possibility of individuating entities of a particular kind (see Strawson 1997, ch. 1; Glock 2003, 47–52). And this general lesson applies equally to abilities. Still a problem remains. It is prima facie plausible to hold that we are able not only to distinguish the concept of a dog from that of barking, but also to specify that precisely two concepts are involved in judging that (1). So concepts and abilities seem to come apart on the issue of enumerability. This verdict can be contested, however, on the grounds that it does not compare like with like. The claim that the number of concepts involved in (1) is determinate is only even remotely plausible if we confine ourselves to predicative concepts (otherwise we have to add at least one quantitative concept that corresponds to the plural in English; alternatively, if we analyse (1) with the help of Fregean logic, we need to add the logical concepts of universal quantification and of material implication). But the very same consideration applies to abilities. It is just as plausible to insist that precisely two predicative abilities are involved in judging that (1)—namely that of thinking about dogs and that of thinking about things that bark—as it is to maintain that precisely two predicative concepts are involved in (1). Accordingly, individuation is no obstacle to the idea that concepts are a kind of ability. 4. Differences between concepts and abilities Other weighty objections remain, however. The established use of ‘concept’ differs from that of ‘ability’ in several respects. 120

First, one thing we do with concepts is to define or explain them. But to define or explain a concept is not to define or explain a capacity. Normally, to explain an ability is to explain its causal preconditions (causal explanation), whereas to explain a concept is to explain its content (semantic explanation). Furthermore, even when we define an ability (i.e. explain its content), we specify what it is an ability to do (this is the reason why abilities are individuated through their exercise, as mentioned in sct. 3). By contrast, to explain a concept is to specify the conditions that an object must satisfy to fall under it. Secondly, and relatedly, concepts can be instantiated or satisfied by things; conversely, things instantiate, satisfy or fall under concepts. These things cannot be said of abilities, or at least not in the same sense. Thirdly, and once more relatedly, concepts have an extension (the set of objects which fall under them) and an intension (the features which qualify objects for falling under them); yet this cannot be said of abilities. Insofar as the ability linked to possessing the concept F has an extension, it is not the range of things that are F, but either the range of subjects that possess F, or the range of situations in which these possessors can apply or withhold F. Fourthly, a concept can occur in a proposition or statement, but an ability cannot. Of course, abilities can occur in propositions in the sense of being mentioned in them, as in (2) The ability to lie convincingly is a great asset in business. But it seems that concepts occur in propositions in yet another and more pervasive way, not just as topics or referents, something the proposition is about, but as components. The concept of being sweet occurs in the proposition that (3) Sugar is sweet even though no ability occurs in it. 5. Tools, techniques and rules At this point it behoves us to return to the issue of concept possession, since it provides the strongest argument in favour of the identification 121

(I) To possess a concept = to possess an ability. But one cannot simply apply the general principle (II) having x = having y Ÿ x = y to this case. For it remains an open question whether (I) cannot be glossed as: (Ic) S has the concept F œ S has the ability of operating with F. Of course, someone who identifies concepts with abilities will resist that paraphrase and insist that the ability with which possessing the concept F is to be identified must be explained without mention of the concept F, an entity with which the subject operates. But it is an option that she has not ruled out. We need to consider the following alternative to identifying concepts with abilities. If having a concept is an ability, it is an ability to operate with concepts. In that case, however, the concept itself cannot be identical with the ability. Rather, it is something employed in the exercise of that ability. A cognitivist conception which picks up this cue is the popular idea that concepts are a kind of cognitive or linguistic tool. Concepts are things employed in the exercise of conceptual abilities, just as tools are things employed in the exercise of manual (technical) abilities. Unfortunately, it is far from clear what kind of tool concepts might be. Worse still, the analogy is misleading to begin with. The idea that concepts are akin to tools in that they are objects (concrete, mental or abstract) with which we operate in conceptual thought amounts to a reification. There is a difference between the possession of a tool and the possession of the ability to employ the tool—as I keep discovering to my cost when trying to operate with our electric drill. This distinction cannot be drawn in the case of concepts. To possess a concept is ipso facto to possess the ability to use the concept. A third cognitivist account promises to heed that point. It maintains that a concept is not an object, properly speaking, but a technique. Thus Wittgenstein maintained that ‘a concept is a technique of using a word’, or ‘the technique of our use of an expression: as it were, the railway network that we have built for it’ (LPP 50 and 2000, MS 163: 56v).6 To master or 6. For Wittgenstein’s approach to concepts see Glock 2010c; concerning the question whether concept-possession requires linguistic capacities see Glock 2010a.

122

possess a technique is to master or possess an ability. Yet techniques are not themselves abilities, but something which the possessor of an ability uses in exercising the ability. There is a difference, for instance, between the ability to skin a rabbit and the various techniques one might employ to this end. Wittgenstein regarded concepts as linguistic techniques. But his idea can be given a Kantian twist, in order to avoid the potentially problematic implication that concepts are the prerogative of linguistic creatures. One can tie concepts in the first instance to thought or understanding rather than language. Concepts are techniques not just for using words, but for mental operations or mental acts which may or may not be expressed in language. The capacity for such mental operations may presuppose possession of language, yet it can be exercised by a subject that does not engage in either overt or silent speech at the time. But what kind of mental operation? A plausible answer is that conceptual thought revolves around classification and inference. Accordingly, the proposal currently under consideration is this: a concept is not identical with the capacity to classify or infer, but only with the technique employed by someone who exercises the ability to classify or infer. Next, the term ‘technique’ needs to be made more specific, in line with both Kant and Wittgenstein. In so far as conceptual thought involves a technique, it is a technique of operating according to a rule or principle. Concepts, the proposal now runs, are rules or principles of classification and/or inference. Even this modified proposal is threatened by category mismatches. It does not seem that to define a concept is to define a principle or rule. Rather, the principle or rule features in the definition. On the other hand, perhaps this is just a vagary of the current use of ‘definition’ in English, without further conceptual import. There is no linguistic infelicity in maintaining that to explain a concept is to explain a principle or rule for performing certain mental or linguistic operations. Another qualm would be that principles can be true or false, whereas concepts cannot. Prima facie, at least, rules escape this difficulty, in so far as they are expressed by sentences in the imperative rather than indicative mood. Note, however, that this exemption does not even hold for all regulative rules, not to mention constitutive rules. Thus a regulative rule of the form (4) x ought to ) in condition C can be prefixed by ‘it is true that …’ (See Glock 2005). 123

Furthermore, the question arises what form these principles or rules should take. Here we seem to be facing a dilemma. One option is that these rules are standards for the employment of concepts. They might, for instance, take the form of the rules Bennett extracts from Kant (Bennett, 1966, 145): (5) You may apply concept F to x iff x is … In that case the concept F itself is not identical with the rule. It is rather a kind of mental predicate the use of which is governed by the rule. A second option is that the rule specifies another activity, e.g. (6) You may treat x in way W iff x is … In that case the danger is that we are stuck with two unpalatable options. One is that W is a place-holder for practical activities which may presuppose concept-possession, but which someone who has mastered the concept need not engage in; the other is that W is a place-holder for conceptualization or classification, which would render the account unexplanatory. 6. Ways of thinking Ultimately, any cognitivist account of concepts must show that this dilemma is apparent rather than real. Or so I shall argue in the next section. First, however, I briefly want to consider a version of cognitivism that prima facie avoids this challenge. It is the Neo-Fregean proposal—epitomized by Peacocke and Künne—that concepts are senses or ‘modes of presentation’. Unfortunately, the latter is no more than a catch-phrase, and one Frege himself never elaborated, least of all with respect to concepts, which he regarded as referents rather than senses of predicates. But we can put some flesh on it by treating concepts as ways of thinking about objects, though not in the adverbial sense of thinking about them hard or longingly. More specifically, concepts are ways of thinking about or conceiving of objects as possessing certain properties, without themselves being properties. To render that suggestion viable, we need to avoid literal interpretations of the Fregean idea that a sense is a mode of presenting a referent. Strictly speaking, there can be no way of presenting a referent unless there is a referent or extension. In the case of concepts, this would rule out unin124

stantiated concepts, which is absurd. On my construal, therefore, ways of thinking about objects are directed not just at those objects which possess the relevant properties, but at all objects of which the relevant properties can be predicated either truly or falsely. To put it differently, a concept is a way of thinking of objects from a suitable range as possessing or lacking certain properties. This reintroduces the idea of classification. Finally, we can operationalize the idea of ways of thinking, thereby further spelling out its cognitive dimension. The concept expressed by a predicate is determined by the features to which the subject refers in deciding whether a given objects falls under the concept, or would decide, if the question arose.7 7. The proposition problem Unsurprisingly, given their intellectual roots, the Neo-Fregeans emphasize the role of concepts in logic, or, more generally, in inferences, whether these be formally or materially valid. Concepts are, among other things, components of propositions. Nonetheless they have the same difficulty here as other cognitivist positions. At least prima facie, like abilities, principles or rules but unlike concepts, ways of thinking do not occur in all propositions, but only in those which are about them, i.e. explicitly mention or refer to them. How can concepts be both ways of thinking and components of propositions? There are two ways of responding to this challenge, which are associated, respectively, with the figureheads of Strawson and Wittgenstein. I shall argue that in combination, these responses promise to resolve the ‘proposition problem’, which would otherwise seem intractable. If Strawson is to be trusted, universals like properties can enter a proposition not just in the direct sense that the sentence expressing the proposition contains words or phrases referring to the property of being F, but also in the less direct sense that the sentence contains words or phrases signifying it (see Strawson 1959, Part II). By a similar token, that sentence contains general terms expressing the concept F, even though it does not refer to it. Finally one can extend this courtesy to any otherwise plausible explanans of ‘concept’. Sticking to the Neo-Fregean proposal, this would mean that the predicate in (3) expresses a way of thinking about, or of, substances, namely as possessing the property of being sweet. 7. For a more elaborate discussion of Neo-Fregeanism see Glock 2011.

125

Crucial to this construal is that standard propositions, and by implication our common or garden thinking, are not about concepts. That insight seems to go back at least as far as Aquinas, on Kenny’s interpretation in any case. Ideas (species) are ‘not what is thought of (id quod intelligitur) but that by which thinking takes place (id quo intelligitur)’ (Kenny, 1980, 71). It is also accepted by Price, who identifies concepts with capacities: ‘The concept is not before the mind as an object of inspection. It is at work in the mind, but not as one inspectable content among others … It shows itself not as a detectable item of mental furniture, but rather as a guiding force, determining the direction which the series of presented particulars [mental images or words] takes …’ (1953, 342). And it is congenial to the Neo-Fregeanism propounded by Künne. In the spirit of Strawson, Künne distinguishes between application, connotation and expression: the general term ‘dog’ applies to all and only dogs, connotes the property of being a dog, and expresses the concept of being a dog (2005, 254 and fn. 31, 263; see also his 2003, 4). Some such distinction is prerequisite for capturing the different semantic properties or dimensions of general terms. Nonetheless the Strawson–Künne solution to the proposition problem immediately faces two challenges. First, can’t the courtesy of being allowed to enter into a proposition indirectly be extended from ways of thinking to all otherwise plausible candidates for being concepts, notably abilities or rules? Secondly, why should one accept that any of these candidates feature in all propositions, however indirectly? The answer to the first question is straightforward in so far as we stick to the relationship between concepts and general terms. It is perfectly commonplace to speak of words as expressing concepts. And there is no violent infelicity in speaking of general terms as expressing ways of thinking. The same goes for rules of classification and/or inference. Perhaps one of these notions—‘way of thinking’ or ‘rule’—comes closer to capturing the ordinary meaning of ‘concept’, yet it is not on account of the possibility of being expressed by general terms. By contrast, it is at best misleading to speak of general terms as expressing an ability. Conceptual abilities are possessed by cognitive subjects, and they are expressed—in the sense of being manifested—by the mental and linguistic activities—notably the judgements and inferences—of such subjects. And we might say that those activities manifest concepts indirectly, keeping this relation apart from the expression of concepts by general terms. The notion of a conceptual 126

ability points to the subject of conceptual thought and to the activity (in a suitably loose sense of the term) of conceptual thinking. By contrast, it is out of place when it comes to the content of conceptual thought, which is precisely what the idea of concepts as components of propositions points to. What we still need is a way of reconciling the mental or cognitive dimension of concepts with the objective dimension suggested by their occurrence in propositions. 8. Propositions and concepts as logical constructions This takes us straight to the second challenge. The Strawson–Künne response at best removes an obstacle to claiming that concepts can be ways of thinking and yet appear in propositions. But what positive reasons do we have for accepting that ways of thinking appear in propositions, let alone as components of propositions? The answer, I submit, is that both propositions and their components are logical constructions out of the practices and abilities of concept-exercising creatures. In this article I can only go some way towards justifying that answer. A first step consists in scrutinizing the idiom of concepts as ‘components’ of propositions in a Wittgensteinian spirit. Perhaps that idiom is misleading and we need to unearth its underlying function. What is the rationale for speaking about concepts and propositions and for parsing propositions into concepts? These notions are helpful in accounting for facts like the following: first, different people can think the same thing—that is, share the same thoughts; secondly, they can entertain thoughts which, though different, stand in logical relations to each other; thirdly, they can do both of these things without sharing a language. When a monoglot Anglophone A and a monoglot Germanophone B both believe that (7) Cats are animals then they share a thought. Similarly, if A believes that (8) Cats are mammals and B believes that

127

(9) Cats are vertebrates then what B believes follows from what A believes. These facts are easily explained in terms of propositions and concepts. A and B can both believe that (7) because they have both mastered the concepts that occur in (7), irrespective of the fact that they express them through different words (e.g. ‘animal’ vs. ‘Tier’). What A believes entails what B believes because of the relations that obtain between the concepts that occur in (8) and (9). All of which is in no way undermined by the fact that A and B would express these propositions and concepts through different words (e.g. ‘animal’ vs. ‘Tier’). Now, the most straightforward explanation of these features appears to be ‘the building-block model’ of propositions and concepts.8 According to this model, what a subject believes (the content of A’s belief ) is a proposition or thought, a complex (abstract) object of which concepts are the components; thus the thought that dogs bark is a complex abstract object of which the concepts DOG and BARK are abstract parts. By a similar token, what a subject has (A’s state of believing) is a mental process of accepting the whole proposition, and thinking one of the component concepts is a stage in this process; thus to believe that dogs bark, A must first think DOG and then BARK. In summary, if A believes that p, then she stands in a relation of grasping and accepting to an abstract entity, a proposition, of which concepts are (equally abstract) components. It follows that one cannot grasp or accept the whole proposition without having or grasping its constituent concepts. Its popularity notwithstanding, however, the building-block model is problematic. There are both empirical and conceptual qualms about the idea that entertaining the part of a thought correlates with a definite stage of a more protracted mental or neuro-physiological process—the entertaining of the whole thought. Even if these could be waved, we would only be dealing with stages of thinking a thought, not with stages of thoughts. As regards the latter, the building-block model transposes the part/whole relation from the spatial and temporal sphere to a sphere—that of abstract entities—to which, ex hypothesis, neither spatial nor temporal notions apply. What seems to give sense to talk of parts and wholes in the case of propositions or thoughts is the fact that the linguistic expressions 8. As far as I can tell, the dismissive phrase ‘building-block theory’ goes back to Davidson. See his 1984, 4, 220.

128

of thoughts—namely sentences—have components—namely words (see Kenny 1989, 126f.). What is said or thought has genuine components to the extent to which its linguistic expression has components (which may, for instance, be explained when A is called upon to state and explain what she believes). Following Quine, many philosophers regard propositions as dubious entities. They are not just abstract objects, but intensional, and hence, allegedly, lack criteria of identity. Such philosophers often replace propositions by sentences as the objects of propositional arguments. I am more inclined to challenge an assumption which the orthodox view shares with its nominalist-cum-extensionalist critics, namely that intentional verbs signify relations to either abstract or concrete objects. The idea of propositional attitudes is problematic not just on account of ‘propositional’ but also on account of ‘attitudes’. For the idea that belief is a relation between a subject and an entity amounts to a reification. Admittedly, noun-clauses like ‘that the cat went up the oak tree’ or ‘what Carl believes’ are grammatically speaking the objects of beliefs. But they are intentional rather than object-accusatives (White 1972). (10) Clare Short believes Tony Blair entails that there is an object x such that Short believes x. In (10) the psychological verb expresses a genuine relation, since here two relata must exist, one to believe, and one to be believed. By contrast, (11) Short believes that Iraq possesses weapons of mass destruction does not entail that there is an object x such that Short believes x. Nothing in reality need correspond to the noun-phrase of (11), since the relevant state of affairs need not exist or obtain. A defender of the building-block model might dig his heels in and insist that something must exist, namely a propositional content which is a real object, though an abstract one. But this ‘something’ is a grammatical projection from that-clauses rather than a genuine object.9 Brentano was right to insist that to believe is to believe something. (11) entails that 9. Pace Quine, the term ‘something’ is wider than ‘object’. ‘Something’ is syntactically transcategorial: it can quantify into the positions of singular term, predicate, and sentence. Only in the first case is it equivalent to ‘object’. For the complex relations between these expressions, as well as ‘exists’, ‘there is’ and ‘real’, see Glock 2003, 52–63.

129

there is something Short believes. Yet in the first instance this simply means that Short cannot believe anything unless there is an intelligible answer to the question ‘What does Short believe?’ Furthermore, the wh-clause ‘what Short believes’, like ‘what Short weighs’, incorporates an interrogative rather than a relative pronoun. Thus ‘Prescott knows the person Short believes’ and ‘The person Short believes is Blair’ together entail ‘Prescott knows Blair’. Yet ‘Prescott knows what Claire Short believes’ and ‘What Short believes is that Iraq possesses weapons of mass destruction’ do not entail ‘Prescott knows that Iraq possesses weapons of mass destruction’, if only because one cannot know a falsehood. Similarly, ‘Prescott knows what Short weighs’ and ‘Short weighs 70 kg’ do not entail ‘Prescott knows 70kg’, since that sentence is ungrammatical. Neither ‘what Short weighs’ nor ‘what Short believes’ signify an object to which Short is related. By the same token, believing that p is no more a genuine relation to an object than weighing n kilograms. It might be objected that there are pertinent contexts in which ‘what Short believes’ incorporates a relative pronoun. In conjunction with (11) (12) Prescott believes what Short believes entails (13) Prescott believes that Iraq possesses weapons of mass destruction. But (12) is not underwritten by our knowledge that Short and Prescott are related in the same way to an entity beyond space and time, whatever that might mean. It is underwritten by the fact that both share certain properties regarding a particular question, namely the question of whether Iraq possesses weapons of mass destruction. Even in this context, ‘what Short believes’ is an interrogative clause in a less direct sense, since its sense derives from the way in which Short would or could respond to a certain question, or react in certain situations, e.g. when it comes to approving the attack on Iraq in Parliament. That different people A and B can think the same thought or hold the same belief does not mean that there is an abstract object to which they severally stand in the relation of thinking, believing, saying, etc. It just means that both A and B believe that snow is white; that is to say, what they both believe can be expressed by the same declarative sentence. If A and B are to disagree, what A says or asserts must be what B denies. But 130

this does not commit one to the existence of self-subsistent entities beyond space and time, but only to the conceptual truism that if B denies what A asserts, and A asserts that p, then B denies that p. The building-block model also goes astray in assuming that the alleged objects to which subjects of belief are related are propositions. Many intentional verbs cannot be characterised as expressing a relation either to a proposition or to a sentence. It makes no sense to expect, fear or hope a sentence or proposition, at least not the same sense as to expect, fear or hope that p. And given that what I can expect is what you can believe, this difficulty may be contagious. That is to say, it may show that even though it makes sense to believe the proposition that p, believing that p is not the same as believing the proposition that p. One might respond that in its philosophical usage, ‘proposition’ is a term of art which is exempted from the vagaries of certain intentional verbs in English that rule out locutions like (14) A fears/expects/hopes the proposition that p. But this invites the challenge to explain what precisely that technical term means. And because of the illicitness of (14) that challenge cannot be met by stipulating that propositions are simply what we believe, expect, hope, etc. The denial that what we believe is always a proposition seems to imply, however, that in cases in which we do believe the proposition that p, we have two beliefs, the belief that p and the belief in the proposition that p. But this objection can be fended off as follows. To say that A believes the proposition that p is not to ascribe to her a belief in addition to her belief that p. Rather, it is to place her belief that p in a certain context. Believing that p is simply a matter of believing something to be so, whereas believing the proposition that p is a matter of believing something to be true. In the case of simply believing, the focus is on how things are or might be, in the case of believing a proposition, on how they have or might be stated or believed to be (see Rundle 2001). 9. Conclusion The measurement analogy does not just pinpoint a weakness in the building-block model of belief, it also promises to furnish the basis for a cogni131

tivist alternative. (On the measurement analogy see Beckermann 1996.) When we ascribe a weight to a person, we do not ascribe to them a genuine relation to an abstract object. Rather, we ascribe to the person a relation to other material objects, for instance that of being in balance with a cube which contains 60 litres of water. Mutatis mutandis for the case of belief. In ascribing a belief to a person, we ultimately describe and explain their actual or possible behaviour. We place the subject not in a relation to a genuine object, but in the context of a system of describing and explaining the subject’s behaviour and behavioural capacities. In the final analysis, talk about propositions and concepts is to be elucidated in terms of what subjects think or say, or, more accurately still, could think or say. Although propositions are not themselves linguistic entities, they are akin to what Prior called logical constructions from linguistic phenomena, namely from the that-clauses by which we report and refer to what people say or think (Prior 1971, ch. 2). The criteria of identity for propositions make essential references to linguistic acts (sayings or utterances). There are propositions no one has ever uttered or thought of. But what distinguishes two such propositions is evident from the declarative sentences which express them. Although our criteria of identity for propositions are not the same as our criteria of identity for sentences, we can only identify the former because we can identify the latter. Although there are different linguistic expressions for the most important truth discovered by Newton and the most important truth discovered by Einstein, what distinguishes these two truths is evident from their expressions—‘F = ma’ and ‘E = mc2’. At this juncture a satisfactory solution to the proposition problem requires a detailed logical construction of talk about both propositions and concepts out of talk about the abilities of rational subjects. Although I do not know of any entirely convincing execution of this programme, there are several noteworthy attempts;10 and I myself have contributed to the project by blocking possible objections (1997; 2006, 53–7). Assuming the feasibility of the cognitivist explanation, let us return to the proposition problem from the perspective of the measurement analogy. In what sense can ways of thinking occur in propositions? The answer is, very roughly: in the sense that S can only think that a is F if S has the capacity to think about objects as being F. Propositions are what is or can be said or thought. Concepts are ways in which subjects do or could conceive of properties. 10. In addition to the aforementioned measurement analogy and to Prior’s own account, see, e.g., Sellars 1963; von Savigny 1983; Brandom 1998; Dolby 2007.

132

To talk of propositions and concepts is not just a façon de parler, on this view, and propositions and concepts are not just ‘make-believe entities’ (to use what is indeed a currently fashionable façon de parler). Rather, they are logical constructions in a non-reductive sense. It may prove impossible to paraphrase concepts away. We may need to refer to them in order to describe the practices of creatures with highly evolved cognitive and/or linguistic abilities. At the same time, the existence and nature of concepts, as well as their individuation, becomes unmysterious once their role within that practice and its description is understood. It is only possible to state what propositions and concepts are in terms which implicitly refer to what people say or do; and we identify propositions and concepts by grouping or classifying actual or potential token-expressions according to what they say or mean. On this basis we may at least hope to reconcile two apparently incompatible features of the ordinary use of ‘concept’, the cognitive dimension and the appearance in propositions.11

REFERENCES Beckermann, Ansgar. 1996. “Is there a problem about intentionality?” Erkenntnis 51, 1–23. Bennett, Jonathan. 1966. Kant’s Analytic. Cambridge: Cambridge University Press. Brandom, Robert. 1998. Making it Explicit. Cambridge (Mass.): Harvard University Press. Crane, Tim. 2004. “Something Else, Surely”. Times Literary Supplement 07.05.04, 4. Dolby, David. 2007. Propositions, Substitution and Generality. PhD thesis. University of Reading. Fodor, Jerry. 2003. Hume Variations. Oxford: Clarendon. Geach, Peter T. 1957. Mental Acts. London: Routledge & Kegan Paul. Glock, Hans-Johann. 1997. “Truth without People?” Philosophy 72, 85–104. — 2003. Quine and Davidson on Language, Thought and Reality. Cambridge: Cambridge University Press. — 2005. “The Normativity of Meaning Made Easy”. In: Christian Nimtz and Ansgar Beckermann (eds.). Philosophy and Science: Proceedings of GAP 5. Paderborn: Mentis, 219–241. 11. I wish to thank David Dolby and the editors for helpful comments on a first draft. This material has also profited from discussions in Bielefeld and Berlin, for which I am very grateful.

133

Glock, Hans-Johann. 2006. “Concepts: Representations or Abilities?” In: Ezio Di Nucci and Connor McHugh (eds.). Content, Consciousness, and Perception. Newcastle: Cambridge Scholars Press, 37–61. — 2009. “Concepts: where Subjectivism goes wrong”. Philosophy 84, 5–29. — 2010a. “Can Animals Judge?” Dialectica, forthcoming. — 2010b. “Concepts; between the subjective and the objective”. In: John Cottingham and Peter M.S. Hacker (eds.). Mind, Method and Morality: Essays in Honour of Anthony Kenny. Oxford: Oxford University Press. — 2010c. “Wittgenstein on Concepts”. In: Arif M. Ahmed (ed.).Critical Guide to Wittgenstein’s “Philosophical Investigations”. Cambridge: Cambridge University Press. — 2011. “Concepts and Ways of Thinking”. Grazer Philosophische Studien. Kenny, Anthony. 1980. Aquinas. New York: Oxford University Press. — 1989. The Metaphysics of Mind. Oxford: Clarendon. — 2010. “Concepts, Brains, and Behaviour”. In this volume, 105–113. Künne, Wolfgang. 2003. Conceptions of Truth. Oxford: Oxford University Press. — 2005. “Properties in Abundance”. In: Peter F. Strawson and Arindam Chakrabarti (eds.). Universals, Concepts and Qualities. Aldershot: Ashgate, 249–300. Peacocke, Christopher. 1992. A Study of Concepts. Cambridge, MA: MIT Press. Price, Henry H. 1953. Thinking and Experience. London: Hutchinson. Prior, Arthur. 1971. Objects of Thought. Oxford: Clarendon Press. Rundle, Bede. 2001. “Objects and attitudes”. Language and Communication 21, 185–198. Ryle, Gilbert. 1971. Collected Papers Vol. II. London: Hutchinson. von Savigny, Eike. 1983. Zum Begriff der Sprache: Konvention, Bedeutung, Zeichen. Stuttgart: Reclam. Sellars, Wilfrid. 1963. “Abstract Entities”. Review of Metaphysics 16, 627–671. Strawson, Peter F. 1959. Individuals: an Essay in Descriptive Metaphysics. London: Methuen. Travis, Charles. 2000. Unshadowed Thought: Representation in Thought and Language. Cambridge, (Mass.): Harvard University Press. White, Alan. 1972. “What we believe”. In: Nicholas Rescher (ed.). Studies in the Philosophy of Mind: American Philosophical Quarterly Monograph Series 6. Oxford: Blackwell, 69–84. Wittgenstein, Ludwig. [LPP]. Wittgenstein’s Lectures on Philosophical Psychology 1946-47. Peter Geach (ed.). Hassocks: Harvester Press 1988. — [MS]. Wittgenstein’s Nachlass. The Bergen Electronic Edition. Oxford: Oxford University Press 2000. (References according to manuscript number and page number.)

134

Grazer Philosophische Studien 81 (2010), 135–151.

THE SELF-CONSCIOUS POWER OF SENSORY KNOWLEDGE Sebastian RÖDL University of Basel Summary The essay develops a disjunctive account of perception, showing that it needs to be renamed ‘self-conscious power account’. For it is by reference to a selfconscious power of sensory knowledge that, on the one hand, the unity of perception and illusion and, on the other hand, the priority of perception over illusion, specifically, its priority in knowledge, is understood. The concept of a self-conscious power thus transpires as lying at the basis of a sound epistemology and response to sceptical arguments such as the argument from illusion.

1. A form of explaining judgment In many an epistemological treatise, one finds the following argument. Suppose I think p, and someone asks me, Why do you think this? Because q, I say. My answer attracts the same question. Why think q?, I will be asked. And unless I can answer this question, I will not have answered the original question. For, if there is no reason for thinking q, then q affords no reason for thinking p. After a few familiar moves, it transpires that I am in principle unable to say why I think what I think. (See, e.g., BonJour 1985, 17–25, and Williams 2001, 58–68.) I may explain that I think one thing because I think another. Here, my account gives something of the same kind as that for which it accounts, hence something that requires the same kind of account. But this does not seem to be the only way in which I may answer the question. I can say, for example, that I think something because I perceive or have perceived it. Here the question does not re-apply to my answer. One cannot ask again, Why think this?, for I did not offer something I think. I said I perceived what I, therefore, think. So it seems there are two forms of answering the question, Why think this? One introduces a new content—q as the ground of p—to which I bear the same relation as I do to the old one: I think p,

and I think q. The other does not introduce a new content, but re-describes my relation to the content in question: I not only think, but I perceive p.1 But is not this distinction spurious? You say you think something because you perceive it is so. But how do you know you perceive this? It has happened to you, has it not, that you thought you perceived something while in truth you did not. So when you say, I perceive, you give something you think. You think you perceive. But why do you think you perceive? We have made no progress. This line of reasoning denies that the two ways of saying why differ in form. It maintains that ‘I think p because I perceive it’ is of the form ‘I think p because I think q’, ‘I perceive p’ taking the place of ‘q’. My answer is, really, ‘I think p because I think I perceive it’. I introduce a new content. It is a special content, to be sure—it includes the first person, a cognitive verb, and the original content—but a new content nevertheless, to which I bear the same relation as I do to the old one: it is something I think. Hence, it is proper to ask me why I think it. If ‘I perceive it’, said in response to the question, Why think p?, gave a content of which I could be asked why I think it, it would be a pointless answer. For ‘I perceive p’ contains ‘p’. Thus I cannot know I perceive p without knowing p, which appears to entail that I cannot know p on the basis of knowing I perceive p. However, ‘I perceive it’, said in response to the question, Why think it?, does not seem to be pointless. It must be wrong to assimilate it to an answer that gives something else that I think. Perhaps this is too hasty. Does not John McDowell teach that we must not assume that, I quote, ‘a basis for a judgment must be something on which we have a firmer cognitive purchase than we do on the judgment itself ’? (McDowell 1982, 385). Let it be that I do not know I perceive p independently of knowing p. We ought not to conclude that this prevents me from explaining why I think p by the fact that I perceive it. However, suppose I say, I think p because I think p š q (p and q being logically independent propositions). What is wrong with this? It is true: I do not have a firmer cognitive purchase on p š q than I do on p. But McDowell said that was fine. Now, I quoted McDowell out of context. He would not approve of this application of his principle. The context makes it clear that his point is that, in my explanation ‘I think p because I perceive it’, ‘I perceive p’ does not bear the form ‘p š q’. For, this is a 1. Compare the general introduction of this difference in Andrea Kern’s Quellen des Wissens, Kern (2006), 68.

136

way of putting the disjunctive account of experience, which McDowell is expounding in the essay from which the quote is drawn.2 We must consider this account in order to comprehend McDowell’s remark on the cognitive purchase we enjoy on a judgment and its ground respectively. The disjunctive account responds to the argument from illusion. Let us recapitulate this argument. There are two possible cases: in the one, I perceive how things are. This, the case of perception, is the good case, in which, so it seems, the deliverances of my senses equip me with knowledge. In the other case, I think I perceive something, while in fact I do not. (It may be that there is nothing there to be perceived, that I misperceive what is there, or that, although things are as I think I perceive them to be, I am not related to them in the way necessary for perceiving them.) This, the case of illusion, is the bad case, in which I wrongly take myself to acquire knowledge through the senses. Now the argument from illusion is so called because it claims to show that, since illusion is possible, perception can never be a basis of knowledge. We can represent the argument as conjoining three premises. (1) It belongs to the bad case, the case of illusion, that I cannot, in this case, tell my predicament apart from that of the good case. In the bad case, I take my case to be good.3 (2) Something on the basis of which I know something must be something I know.4 (3) As I cannot distinguish the bad case from the good case (this is the first premise),5 what distinguishes the good case from the bad one is nothing of which I am aware. Therefore, given the second premise, it is nothing on the basis of which I could know anything. The basis of my knowledge can only be an experience that is the common element of the good case (the case of perception) and the bad case (the case of illusion). From this follows the conclusion: since an experience that is common to perception and illusion can be present while things are otherwise than I experience 2. McDowell discusses, not experience, or knowledge based on perception, but knowledge of other subjects based on knowledge of what they do and how they act. However, as far as our interest goes, these are perfectly analogous. 3. If I did not, I would not be deceived, and the case would not be bad. As Martin Heidegger puts it in his Der Ursprung des Kunstwerks: “Das Verbergen verbirgt sich.” (Heidegger (1950), 40.) 4. This principle may be rejected. I suspect the main motive for rejecting it is the conviction that it leads to scepticism. (This motive is evident in early statements of the rejection. See, e.g., Dretske (1971), or Alston (1980). If the argument from illusion is unsound and an account of perceptual knowledge can be developed that respects the principle, this motive falls away. 5. Looking forward to the end of this essay, we notice that this is imprecise. (We need the imprecision to state the (fallacious) argument.) The first premise is not that I cannot distinguish the good case from the bad case. Rather, it is that, in the bad case, I cannot distinguish the bad case from the good case.

137

them as being, it provides no ground on which I could know that things are that way. Hence, contrary to what we first supposed, even when I perceive p, the deliverances of my senses do not put me in a position to know p. The disjunctive account seeks to bring out a fallacy in this argument. It maintains that someone who perceives p and someone who wrongly thinks she perceives this (and thus is under an illusion) do not share a psychic state in common. Or, more precisely, if there is a state that is common to both, it does not exhaust the basis on which someone who perceives p is thereby in a position to know p.6 An act of perceiving is not an experience satisfying a further condition (the condition, say, of being caused in the right way by the fact that it represents).7 ‘I perceive p’ is not composed of a statement about a state of my psyche—reporting my experience—and an independent statement about further conditions which that state satisfies and which contain the fact that p. Now we can return to McDowell. Defending the idea that perception can be a ground of knowledge, he says that my cognitive purchase on the ground on which I rest a judgment need not be firmer than my cognitive purchase on that very judgment. This seemed to entail that one may judge p on the basis of one’s judgment p š q. However, McDowell makes his remark in the context of defending a disjunctive account. It is the very point of this account that ‘I perceive p’ does not bear the form ‘p š q’: it is not composed of a statement of fact, ‘p’, and an independent statement about my psyche, ‘q’. Hence, when McDowell says that I need not have a firmer cognitive purchase on the basis of my judgment than I do on that judgment, he does not mean to show that ‘I think p because I perceive it’ may be of the form ‘I think p because I think q’ and yet be a valid justification. On the contrary, he means to insist on the formal difference of ‘I think p because I perceive it’ from ‘I think p because I think q’. We are considering ways of saying why one thinks such and such, where this is giving a ground for thinking it such that, ideally, thinking it on this ground is knowing it. Now—this was the second premise in our reconstruction of the argument from illusion—a ground on which I know something is something I know, and know as ground. Hence, if there is a 6. Whether this is more precise depends on whether it is possible to identify an act of perception (except per accidens) independently of the way in which it provides, or fails to provide, its subject with knowledge. We shall find reason to deny this. 7. The point of the disjunctive account is that there is no way of specifying the condition without using the concept of perception. Specifically, the right way of causing cannot be specified without relying on that concept.

138

distinct form of ground for thinking p that I give saying ‘I perceive p’, then there is a distinct form of knowing, a form of knowing this ground. ‘I think p because I perceive it’ differs in form from ‘I think p because I think q’ in that I cannot be asked ‘Why do you think you perceive it?’, only if I know that I perceive something in a different way from that in which I know what I know through perception. When we return to the passage quoted from McDowell, we see that it points us to this distinct form of knowledge. The passage draws a conclusion from this claim: I think we should understand criteria to be, in the first instance, ways of telling how things are […]; and we should take it that knowledge that a criterion is satisfied […] would be an exercise of the very capacity we speak of when we say that one can tell, on the basis of such-and-such criteria, whether things are as the claim would represent them as being. (McDowell 1982, 385)

We can rephrase this as follows. I do not have a firmer cognitive purchase on the fact that I perceive p than I have on the fact that p by virtue of perceiving it, because the basis on which I know that I perceive p is the same as the basis on which I know p, knowing it through perception. If this is right, then I know that I perceive something in a different manner from that in which I know that which I perceive. I know p by perceiving p. Now, if I know on the same basis that I perceive p, then I know that I perceive p, again, by perceiving p. I know that I perceive p, not by perceiving the thing that I know (I do not perceive that I perceive). I know it by the thing that I know: by perceiving p I know that I perceive p. The thing that I know contains my knowledge of it. We began with an argument that presupposes that any answer to the question, Why think this?, re-attracts this same question. We naïvely objected that we sometimes say, ‘I think this because I perceive it’, thinking of this as laying the question to rest. But we were told that, when I say ‘I perceive p’, I announce that I think I perceive it, so that it is proper to ask me why I think this. We now see that this line of thought presupposes that the way in which I know that I perceive something does not differ from the way in which I know that which I perceive. And this is wrong. I know that I perceive, not by perceiving the thing that I know, but by the very thing that I know. This explains why we cannot repeat the question and ask, How do you know you perceive p? I already said how: I perceive p. This is how I know that I perceive p. The disjunctive account of experience seeks to hold on to the difference in form of ‘I think p because I perceive it’ from ‘I think p because I think 139

q’. Therefore it must explain this form, and explain it in a way that reveals it to express knowledge of a special kind: knowledge of an act contained in this very act. In what follows I shall expound the disjunctive account in a way that satisfies this requirement. I shall claim that the form of ‘I think p because I perceive it’ is this: it explains why I think p not by referring my judgment to something else that I think, but by re-describing it as an act of a power, a power to know by means of the senses.8 This explains how I know that I perceive. For, as the power to know by means of the senses is self-conscious, knowing something through perceiving it is the same as knowing that one knows it in this way. 2. Power Let me quote McDowell again: I think we should understand criteria to be, in the first instance, ways of telling how things are […]; and we should take it that knowledge that a criterion is satisfied […] would be an exercise of the very capacity we speak of when we say that one can tell, on the basis of such-and-such criteria, whether things are as the claim would represent them as being.

I put it in this way: the basis on which I know that I perceive p is the same as the basis on which I know p, knowing it through perception. McDowell says more: he identifies this basis with a capacity, or, as I shall prefer to say, a power. Although McDowell does not emphasize it, the concept of a power of knowledge is the basis of the disjunctive account of experience. The disjunctive account holds that she who perceives p and she who is under the illusion of perceiving this do not share a psychic state in common, which then would be compatible with its not being the case that p. This first statement of the account is insufficient because it is merely negative. It says how she who perceives p and she who wrongly thinks she perceives are not joined together: not by a common psychic state. It will not do to leave it at that. For it is not the case that the good case and the 8. St Thomas explains that knowledge of truth, which is an act of the intellect, is distinguished from the manner in which truth is in the merely sensory representations of an animal by the fact that the act of knowing the truth refers itself to (is conscious of itself as springing from) the power of which it is the act, for only in this way can the act of knowledge contain an understanding of its relation to its object. (De Veritate, Quaestio 1, Articulus nonus.) Our reflections will show that St Thomas’s teaching is sound.

140

bad case have no connection. We need a positive account of their unity. We said there is no common psychic state—an experience as of p—to which, in the good case, something is added, namely, inter alia, the fact that things are as the subject experiences them to be. If the good case is not like the bad case with something added, then perhaps the bad case is like the good case with something subtracted. Then the good case is prior. We shall see that this is true. The good case is prior; it is prior in being, in concept, and in knowledge. If we consider only acts of perceiving, we see no priority of the good case over the bad case. In order to recognize the priority, we must relate the acts to their source, a power, which, in man, the rational animal, is a power of sensory knowledge. As the concept of a power and specifically the concept of a self-conscious power is the concept on which the disjunctive account of experience rests, the name ‘disjunctive account’ is ill-chosen. It fails to convey, and even conceals, the essential thought of the account. The idea of disjunction does not contain the idea of an inner unity of the disjuncts. Thus an account called ‘disjunctive’ may appear to disregard the unity of the good case and the bad case. However, the disjunctive account is essentially an account of this unity: it denies that it resides in a common state and maintains that it lies instead in a common source: the power. Furthermore, the idea of disjunction does not contain the idea of an asymmetry among the disjuncts. Thus an account called ‘disjunctive’ may appear to hold that there are two kinds of experience on a par in their relation to the subject’s knowledge of them. However, the disjunctive account is essentially an account of the priority in knowledge that the good case enjoys over the bad case, a priority residing in the way in which each is related to the power that is their source. Therefore, a more apt name of the account in question is ‘power account’.9 In order to develop the power account of perception, let us first consider the concept of a power in the abstract. I shall follow Aristotle’s treatment of this concept in Metaphysics Theta. Power contrasts with act. To the extent that a power is distinct from an act of this power, it is not exhausted by it and thus not by any number of acts. A power is general with regard to its acts: it does not limit their number. In consequence, the representation of a power in language is a general statement. We say, for example, ‘The weaverbird builds a spherical nest’, describing a power of the weaverbird. The statement is general in that it may be exemplified on an indefinite 9. This is the name Andrea Kern gives it (in her Quellen des Wissens).

141

number of occasions, occasions for saying ‘Look, a weaverbird: it is building a spherical nest’ or ‘Look, a spherical nest: it is a weaverbird’s.’ Not only is a power general with regard to its acts; it explains its acts. We provide an account of an act of a power by referring it to this power. For example, when we see a weaverbird building a spherical nest, we recognize that this is no accident, knowing that weaverbirds build spherical nests. Of course, we may want to know why it is that weaverbirds build spherical nests. But as to this weaverbird here, we know why it is doing what it is doing: it is building its nest, and this is how weaverbirds do it. In contrast, if weaverbirds did not build spherical nests, we would be baffled by this weaverbird’s behaviour and would search for an explanation. General statements describing a power explain their instances. I shall, following Aristotle, I think, call the general statements describing a power the logos, the concept of the power. Aristotle says, in a passage we shall consider more closely in a bit, that the logos of a power explains its acts. So we may speak, if we think of explaining as the act of subsuming a given act of the power under its logos. However, as the logos explains by referring the act to the power, representing the power as the sufficient ground of the act, we can also say that the power explains the act. Thus I said in the preceding paragraph both that a power explains its acts and that general statements describing a power explain their instances. These are two ways of expressing the same idea. I shall continue to use them both.10 We can distinguish powers according to the logical form of the general statements that describe them, the form of their logos. These general statements may be hypothetical or categorical. They may say, simply, what a bearer of the power does, or what it does given a certain external condition. The condition is external, external to the power, if its obtaining is not explained by that same power, or the system of powers of which that power is an essential element. Consider, for example, the statement sugar dissolves in water. The power of sugar to dissolve is actualized under the condition that it be placed in water, and this condition is external in the sense that it 10. It may be that an act is possible only as an act of a power. That is, it may be that it is possible only as an instance of something general that underlies it. Thus it is with reading and adding. If it is an accident that the words spoken are the same as the words written, then she who is speaking them is not reading. And if it is an accident that the number pronounced is the sum of two given numbers, then she who is pronouncing it did not add. But that it is no accident that the words she is speaking are the words on the page and that the number she pronounces is the sum of the numbers she was given means that her speaking those words and her pronouncing this number manifest something general, something that may manifest itself on indefinitely many occasions: her power to read, or to add.

142

is not in the nature of sugar to find itself in water and thus an accident if it does. Another example: bobcats breed in spring. This describes an aspect of the reproductive power of bobcats. Although all manner of conditions must be satisfied if a given bobcat is to breed in a given spring, these conditions are not external to it in the sense that it is in the nature of bobcats, the nature expounded in judgments such as this one, bobcats breed in spring, to find themselves in these conditions. (Compare Thompson 2008, 70f.) This distinction will not further concern us. I mention it in order to distinguish it from a distinction orthogonal to it: the distinction of acts that are completely explained by a power from acts that are not. An act is completely explained by a power if, in order to explain it, it is sufficient to bring it under the general statement describing the power. If that statement is hypothetical, then this will involve seeing that the antecedent is satisfied. But once this is established, the act is explained by the power alone; no further inquiry into the circumstances of the act is necessary. By contrast, there are cases in which something interferes with the actualization of the power, and the power fails to be exercised, or fails to be exercised properly. It is not that, in these cases, the power is irrelevant to the explanation of the act. An explanation will still refer the act to the power. But the power does not suffice for its explanation. We must adduce conditions particular to this occasion of its exercise to explain what is happening. This structure is ubiquitous: When someone who can add writes down the number eight as the sum of three and five, then this is explained completely by his power to add. Were we to search for a cause of his writing the number eight among the present circumstances, we would show that we think he cannot add. By contrast, if he writes the number four, we may look for a cause among the circumstances. We may find the ‘5’ was badly written, so that he mistook it for a ‘2’. Or we may find something that distracted him, or something that put the number four into his head. The power to add is one whose subject can only be a rational being. But the structure remains as we descend the scala naturae. When a lion is sinking its teeth into the neck of an antelope, cutting through its spinal cord, this is explained completely by its power to kill its prey. However, if it bites its own paw, we must invoke particular circumstances to explain this use of its teeth. Aristotle describes the general structure as follows: The same logos explains a thing and its privation, only not in the same way. >…@; for, the logos is of the one kat’ hauto, of the other, in a sense, accidentally. For it is by negation and subtraction that it explains the contrary. (Metaphysics, Theta 2, 1046b8–14)

143

A power, or its logos, explains two kinds of cases, but not in the same way: acts of the one kind are explained by the power in itself, kat’ hauto; acts of the other kind are explained by the power accidentally, namely by negation and subtraction. This asymmetry constitutes the priority in being of the former kind of acts. It may be thought that the fact that it is necessary to invoke circumstances of the act to explain the failure of a power to be exercised, or to be exercised properly, circumstances that interfere with its exercise, shows that in no case does a power alone explain its act. For, one may think, when nothing interferes, too, the power is not the sufficient ground of its act. A complete explanation adds that certain conditions did not obtain, conditions that, had they obtained, would have interfered with the exercise of the power. Aristotle considers this thought and rejects it. For it is not necessary to add to the definition ‘nothing external interfering’. For a thing has a power as it is a power to act, and this it is not under all conditions, but in certain circumstances, in which externally interfering things will be excluded. For these will be barred by some of the things belonging to the definition. (Metaphysics, Theta 6, 1048a16–21) Aristotle here, I think, draws attention to the priority in concept of the act that the power explains completely. We can form no positive concept of conditions that interfere with the exercise of a power. The unity of these conditions is negative: what holds them together is this, that they are conditions that interfere with the exercise of the power. The concept of conditions that interfere with the exercise of a power thus contains and depends upon the concept of an act explained by the power alone. The act explained by the power alone is prior in concept. This can be put by saying, as Aristotle does, that the idea of a power and its causality excludes circumstances that interfere with its exercise; their absence is not a cause in addition to the power. This asymmetry of two kinds of acts, and the priority in being and concept of one of them, is found wherever there is a power distinct from its act. We may ask under what conditions this asymmetry is such as to give application to the concepts of good and bad, so that the act prior in being and concept is good. I shall not pursue this question here.11 I shall merely assert that the concepts of good and bad find application when the general statements describing the power are categorical as opposed to hypothetical. Such a power is a power of spontaneity. 11. I do pursue it in Rödl (2003).

144

3. Self-consciousness I said there is a priority in being and in concept, and, furthermore, in knowledge. I think this priority holds generally. But here I want to consider the case in which the priority in being is, as such, a priority in knowledge. So it is when the power is self-conscious. I can introduce the concept of a self-conscious power through the concept of a self-conscious causality. When I speak of causality, I mean what is represented by an explanation that represents one thing as the sufficient ground of the existence of another. Thus, when I speak of a kind of causality, I speak of a kind of explanation, a kind of ‘because’. A self-conscious causality is a causal nexus of acts of a subject that is constituted by the subject’s representation of this nexus, which representation on that account is knowledge and, specifically, self-knowledge. I shall give two examples: doing one thing for the sake of another, and judging one thing on the basis of another. One way of explaining why someone is doing something is to represent her as taking means to an end she pursues, for example, ‘She is going upstairs because she wants to fetch the camera’. This causality is constituted by the subject’s subsuming doing one thing as a means under her end of doing another thing, or, as Kant puts it, by her deriving doing it from this end.12 If she did not think that going upstairs was a means of fetching the camera, this would prove that she is not going upstairs because she wants to fetch the camera. The explanation is true only if the subject is able to give it; the causality is constituted by the subject’s representation of this very causality. Similarly, one way of explaining why someone thinks such and such is to represent her as resting her judgment on another judgment from which it follows: ‘She thinks A because she thinks B’. Again, the causality of the explanation is constituted by the subject’s recognition that A follows from B. If she did not think that A followed from B, and was unaware of a connection between these judgments, this would disprove the account of her judgment. The causality is constituted by her representation of this causality.13 In general, if an object is constituted by a representation of it, then this representation is knowledge. For, as the representation is constitutive of the object, there is a non-accidental unity of the representation and the object. Moreover, such knowledge is first personal. For, by nominal 12. Immanuel Kant, Grundlegung zur Metaphysik der Sitten, 412. 13. I develop the idea of this form of causality in Rödl (2007), chs. 2 and 3.

145

definition, a first person representation is one that is such that, not by accident, but in virtue of the form of the representation, she who represents is identical with her whom she represents. This condition is satisfied if the object represented is an act of the subject that is identical with the act of representing it. So it is in the examples we considered. Therefore the subject’s representation of the causality is first personal: ‘I am going upstairs because I want to fetch the camera’, she thinks, and, ‘I think A because I think B’. If we call self-conscious what is represented first personally, then this causality is self-conscious. The sufficient ground of the existence of an act may be another act, as in the cases we considered. It may also be a power.14 The causality of a power, too, may be self-conscious. If it is, then the causality of the power will be constituted by the subject’s knowledge of this causality. The power will be a source of its acts in such a way as to be a source of its subject’s knowledge of those acts as acts of the power. We called the general statements describing a power its logos. So we can say that a subject of a power whose causality is self-conscious, the subject of a self-conscious power, acts, not only according to the logos of its power, but according to a representation of this logos. In Aristotle’s words, she acts not only kata ton logon, but meta tou logou. Aristotle introduces the concept of a dunamis meta logou in the Metaphysics;15 he contrasts virtue as a hexis to act meta tou orthou logou with a hexis to act merely kata ton orthon logon in the Nicomachean Ethics.16 Consider the weaverbird and its power to build a spherical nest. The logos of this power is a general statement or system of statements describing how the weaverbird builds its nest. Observing weaverbirds, we may acquire this logos, that is, learn how the weaverbird builds its nest, and explain what a bird is doing on a given occasion by referring its act to this power, subsuming it under this logos. The weaverbird has the power, we have the logos. We have it from observing weaverbirds acting according to this logos (kata ton logon). Now consider a builder and its power to build a house. Again, there is a logos of the power, a system of general statements saying how the builder builds. And again, we may acquire this logos by observing builders. Then, on a given occasion, we may explain why a builder is doing what he is doing by subsuming his acts under 14. These kinds of causality do not stand opposed to each other. The first may be subsumed under the second, when a causal nexus of acts is explained by a power. 15. Aristotle, Metaphysics, Theta 2, 1046b1–2. 16. Aristotle, Nicomachean Ethics, Zeta 13, 1144b.

146

this logos. The builder has the power, we have the logos. We have it from observing builders acting according to it. Now, in order for the power to be a power with logos, meta logou, she who has the power must also have the logos. But this is not enough. For, it must not be per accidens that she who has the power has the logos. Rather, this must be a character of the manner in which she has the logos; equivalently, it must be a character of the kind of power of which it is the logos. That is, she who has the power must as such, by having the power, have the logos, and having the power and having its logos must be the same. Such is the power of the builder. The builder as such knows how he builds. His knowledge need not be articulate. It suffices that he can show, demonstrate, how this and that is done. In showing this, he presents, and thus comprehends, his act as an instance of something general, which is the power. The builder as builder knows how he builds because his knowledge is not derived from his acts of building, observing which he acquired the idea of a power that underlies them. Rather, conversely, his acts of building derive from his knowledge; he acts from this knowledge. In Aristotle’s words, he acts not only kata ton logon, but meta tou logou. She who exercises a self-conscious power knows the act of her power, and knows it as an act of her power. That is, she is in a position to explain it, and explain it completely, by the power. If the power is such that the asymmetry of the cases explained by the power alone and cases explained by the power by negation and subtraction gives application to the concepts of good and bad, then a subject of a self-conscious power, explaining her act by the power, represents her act as good. In other words, she justifies it. Her explanation, or justification, is not a distinct act from the act she justifies, but the self-knowledge internal to this act. I shall not further belabour it, for it seems sufficiently obvious that this is the fundamental form of justifying an act: explaining it by a self-conscious power. A self-conscious power is exercised in acts that include knowledge of these acts as acts of the power. So the subject of a self-conscious power knows its good acts, and knows them as good, in these very acts. This is not true of the bad cases. These are explained not by the power alone, but by the power together with circumstances impeding its exercise. Therefore, the subject knows these acts as bad only through knowledge of these circumstances. As this is knowledge of conditions external to the power, it is not provided by the power in this very act. The subject may gather this knowledge in a further act of this same power, as when she becomes aware of the falsifying light that made her misjudge the colour of the tomatoes in 147

the supermarket. So, while the good act of a self-conscious power includes its subject’s knowledge of this act as good, a bad act does not provide its subject with knowledge of it as bad. We can now return to the argument from illusion to see how the disjunctive account or, better, the self-conscious power account reveals it to be invalid. Its first premise is that, being under an illusion, I take myself not to be. This is contained in the idea of illusion. The second premise is that the basis on which I know something is something I know. If this seems questionable, then presumably because it is thought to entail a sceptical conclusion. However, there is no reason to deny the second premise of the argument in order to escape its conclusion because its third premise is faulty. It says that I cannot distinguish my situation as one of perceiving from one of illusion. Now, while it is true that, being under an illusion, I do not know that I am, it by no means follows that, not being under an illusion, but perceiving how things are, I do not know that I am. That I do not know that I am under an illusion marks my situation as one of illusion. There is no ground for thinking that the same lack of knowledge afflicts me in a situation in which I am, precisely, not under an illusion. Still one might be tempted to think this—for thinking which there is no ground—as long as one does not see how it is that, through perceiving, I know that I perceive. Our reflections remove this temptation. They show that perceiving is knowing that one perceives as perceiving is an act of a self-conscious power. ‘Disjunctive account’ is an ill-chosen name because it conveys only that we cannot conceive of the unity of perception and illusion as lying in an experience that as perception bears the additional feature of originating in the right way from the fact that it represents. This alone does nothing to ward off scepticism. We must understand how perception can be prior to illusion, and specifically, how it can be prior in knowledge. Perception can be a ground of knowledge only if knowledge that I perceive is not knowledge of something that holds in a case of perception and in a case of illusion alike, conjoined with further knowledge excluding illusion.17 If there is to be a disjunctive account, then knowledge that I perceive 17. Crispin Wright (2002) proceeds on the assumption that I can know the good disjunct “I perceive p” only by adding to my knowledge of the disjunction “Either I perceive p or I am under an illusion” knowledge that excludes the bad disjunct. He does not consider the possibility that knowledge of the good disjunct is prior to knowledge of the disjunction. That Wright could think that his reflections showed the disjunctive account to be defenceless against a reformulated argument from illusion is, I presume, to be explained by the unhappy name “disjunctive account”.

148

must be prior to knowledge of anything common to perception and illusion. We see how this may be as we understand perception as the act of a self-conscious power. It is for this reason that the disjunctive account is a self-conscious power account. 4. Conclusion The disjunctive account of experience insists that ‘I think p because I perceive it’ differs in form from ‘I think p because I think q’, holding that the case in which I perceive and thereby know p (the good case) and the case in which I think I perceive p, but do not do so (the bad case), are not joined by a psychic state I exhibit in both cases. Thus the disjunctive account must give a different explanation of the unity of the good case and the bad case, one that shows that I know I perceive something in a different way from that in which I know that which I perceive. An abstract specification of this way of knowing is: I know that I perceive something by perceiving it. Knowing that one perceives p is not a different act from knowing p by perceiving it.18 We understand the unity of the good case and the bad case as we relate both to a power, which completely explains the good case, and explains the bad case by negation and subtraction, that is, by circumstances that interfere with the proper exercise of the power. And we understand that the good case contains knowledge of itself, and of itself as good, as we see that the relevant power is self-conscious, or meta logou. There appears to be a vague idea that if I could distinguish the case of perceiving from the case of merely seeming to perceive (which, according to the account we presented, I can in the good case: I can, indeed I do, 18. Knowing something through perception is knowing that one so knows, as sensory knowledge is an act of a self-conscious power, a power whose acts include knowledge of these very acts. Saying this, we do not endorse a general principle to the effect that knowing something entails knowing that one knows it. In fact, a calculus of epistemic logic that licenses inferences exemplifying the schema “Kp Ÿ KKp” is incapable of representing our claim. It suggests that “Kp” is known in the same manner as “p”, as it employs the same letter to signify the nexus one bears to both these propositions. Thus, on the one hand, when p is known through perception, the calculus suggests that the fact that one knows something through perception is something one knows through perception, which is absurd. On the other hand, when p is known on account of being an act of a self-conscious power, the calculus suggests that “Kp Ÿ KKp” has content, while in truth it is of the form “p Ÿ p”. For, here, the object and the knowledge of it are the same reality, wherefore it is a tautology that who knows the former knows the latter.

149

distinguish the actual good case as one in which I am perceiving from a possible one in which I would merely seem to perceive; I can do that in virtue of the fact that I am perceiving), I would be infallible and not liable to error in my perceptual judgments. So it would be, if, per impossibile, my knowledge that I perceive were a distinct act from the act of which I thus know. However, as this is but one act, there is no difficulty understanding how my power of sensory knowledge may fail me, while yet providing me with knowledge of that and how I know when I do know. If I know p by perceiving it, I know that I do. But if I do not know p, as I do not perceive it, but only think I do, then neither do I know that I know, nor do I know that I do not know. I think I know, and think that I know that I do, but I am wrong in thinking this. My power to know that I know by means of the senses is as fallible as my power to know by means of the senses, for it is the same power.

REFERENCES Alston, William. 1980. “Level Confusions in Epistemology”. Midwest Studies in Philosophy 5, 135–150. Aristotle. Metaphysics. William D. Ross (ed.). Oxford: Clarendon Press, 1924. — Die Nikomachische Ethik. Rainer Nickel (ed.). Zürich: Artemis Winkler 2001. BonJour, Laurence. 1985. The Structure of Empirical Knowledge. Cambridge, (Mass.): Harvard University Press. Dretske, Fred. 1971. “Conclusive Reasons”. The Australasian Journal of Philosophy 49, 1–22. Heidegger, Martin. 1950. “Der Ursprung des Kunstwerks”. Reprinted in his Holzwege, Frankfurt/Main: Klostermann, 1–72. Kant, Immanuel. Grundlegung zur Metaphysik der Sitten. Gesammelte Schriften vol. IV. Berlin: Preussische Akademie der Wissenschaften 1900 ff. Kern, Andrea. 2006. Quellen des Wissens. Zum Begriff vernünftiger Erkenntnisfähigkeiten. Frankfurt: Suhrkamp. McDowell, John. 1982. “Criteria, Defeasibility, and Knowledge”. (Reprinted in his Meaning, Knowledge, and Reality. Cambridge, (Mass.): Harvard University Press 1998, 369–394.) Rödl, Sebastian. 2003. “Norm und Natur”. Deutsche Zeitschrift für Philosophie 51(1), 99–114. Rödl, Sebastian. 2007. Self-Consciousness. Cambridge, (Mass.): Harvard University Press.

150

S. Thomae de Aquino. Opera Omnia, Tomus XXII, Questiones disputatae de veritate. Rom 1970. Thompson, Michael. 2008. Life and Action. Cambridge, (Mass.): Harvard University Press. Williams, Michael. 2001. Problems of Knowledge. Oxford: Oxford University Press. Wright, Crispin. 2002. “(Anti-)Sceptics Simple and Subtle: G. E. Moore and John McDowell”. Philosophy and Phenomenological Research 65, 330–348.

151

Grazer Philosophische Studien 81 (2010), 153–172.

IN SEARCH OF CONCEPTS Katia SAPORITI University of Zürich Summary After some general remarks about the contemporary debate on concepts and about the justification of the so-called definition question, I try to participate in this debate by criticizing what I take to be an instructive approach to concepts and concept possession, viz. Hans-Johann Glock’s proposal to view concepts as rules or principles rather than abilities. I address his attempt to distinguish between sorting and classifying, his objections to the idea that concepts are abilities and his contention that thoughts and concepts are to be regarded as the content of thinking. I advocate a deflationary approach to concepts and try to raise doubts about the claim that thoughts have a language-independent structure.

1. Introduction In contemporary philosophy concepts are invoked to serve a great variety of purposes. One of these purposes is the explanation of a number of abilities or capacities which we ascribe to human beings. These will typically include the ability to entertain thoughts and propositional attitudes concerning kinds of things, the ability linguistically to refer to kinds of things and the ability to reason, i.e. to draw inferences. For it is widely held that only somebody who has mastered the concept of a tree can think about trees and entertain beliefs about trees; that only somebody who possesses the concept of a triangle can use the word ‘triangle’ in a meaningful way; and that only somebody who has acquired the concepts of a tree and of a plant can infer from something’s being a tree that it is a plant. One of several conceivable motives for enquiring into these abilities is the wish to know whether we can ascribe them to non-human beings (e.g. animals, machines, Martians). Answers to this question as well as one’s views on concepts will depend on how one takes the different abilities that are thought to involve the exercise of concepts or to presuppose concept possession to be related to each other and to further abilities (e.g. the ability

to sort things into classes). It is, therefore, of some importance to get clear about what project we engage in when thinking about concepts. In what follows I will try to substantiate this claim. If we wish to explain the capacity to think and reason by invoking concepts, assuming that, once their nature has been clarified, we will then be able to show how our linguistic abilities rest on our language-independent cognitive capacities, we may not, of course, conceive of concepts in terms of a wide range of abilities which include linguistic as well as non-linguistic abilities. This is acceptable for someone who, like Fodor, wants to explain the intentionality of the mental and our ability to reason by reference to an inner language of thought and assumes that semantic content of linguistic expressions ultimately rests on mental content but not vice versa (Fodor 1975, 1987 and 1990). But it also concerns anybody who hopes to ascribe concepts to non-linguistic beings, thereby proving them capable (at least in principle) of thought and reason. If you believe either that mental content rests on the use of a language or that the use of meaningful expressions ultimately depends on, and can be explained in terms of, mental content, you cannot run together two projects which traditionally have been closely linked, viz. to explain by invoking concepts both our ability to refer to things of a kind in speech and our ability to refer to them in thought. Locke designed his abstract ideas to serve this double purpose—and failed. 2. A question in need of justification Besides offering an opportunity to discover the constraints that different agendas will put on accounts of concepts, there is another reason to get clear about one’s means and ends in invoking concepts, namely, the risk of getting involved in a confused and partly empty discussion about the true or real nature of concepts. Some of the purposes concepts are supposed to serve in philosophy seem to have little to do with abilities. Concepts have been invoked to account for the unity of a proposition by philosophers who saw them as something unsaturated in the Fregean sense: as a function or the sense of a function that needs an argument to yield a proposition which has a truth value. Concepts have also been invoked to account for certain inferential relations that hold between propositions. If concepts are thought to apply to the things that fall under them, various relations of exclusion and inclusion will hold between them. It is in virtue of these relations that the propositions which concepts are held to figure 154

in stand in certain relations to each other. A large part of philosophy is claimed to consist in the analysis and clarification of concepts (if not in their definition). This sort of analysis is meant to spell out the relations certain concepts bear to other concepts. Furthermore, concepts have been invoked to account for linguistic phenomena. Synonymous words are held to express the same concept. A good translation of a word will express the same concept. For instance, the English word ‘rain’ is held to express the same concept as the German word ‘Regen’ or the French word ‘pluie’. Some philosophers believe that concepts are the meanings of words, they take them to be intensions as opposed to extensions. To continue to muddy the waters, one may want to remind people of the fact that the words ‘concept’ (in English) and ‘Begriff’ (in German) are often used synonymously with the expressions ‘word’ and ‘linguistic expression’, or ‘Wort’ and ‘sprachlicher Ausdruck’. When German speakers use the expression ‘Begrifflichkeit’ what they usually mean is the terminology that is under consideration—the kinds of words that are being used or the theoretical language specific to a kind of investigation. But if concepts were words or linguistic expressions, two people not sharing a language could not have the same concept. And in that case the vast majority of questions philosophers engaged in the contemporary debate on concepts seem to be interested in would make no sense. If concepts are not words, what exactly are they? In view of the remarks I have made so far it is to be expected that there is more than one answer to this question. Why should there be only one concept of a concept? The question as to the nature of a concept—the so-called definition question— seems somewhat strangely put. Compare the question ‘What exactly is an idea?’, which seems completely to hang in the void unless it is further specified by mentioning particular theories or authors or periods in the history of philosophy. An answer could be: eidos and idea in Plato is something different from idea and idée in Descartes as well as from Vorstellung and Begriff in Kant and Idee in Hegel. All one can do is to look at the works of these philosophers and try to find out differences and similarities between the various kinds of use that are there made of the relevant terms. Depending on what kind of theory in the philosophy of mind, the philosophy of language, linguistics, psychology or other disciplines you want to defend, you will have different views on concepts. As we are dealing with a theoretical term, the question what a concept really is amounts to asking which of several competing theories will prove the best one. What is not clear at all, though, is whether all the theories in question (that is, 155

all the theories making prominent use of the term ‘concept’) really stand in competition. Why is it, then, that, as George Rey once put it, ‘Given these diverse interests, there is considerable disagreement about what exactly a concept is’ (Rey 1998, Introd.)? In view of the fact that only people who share the relevant concepts can disagree about more than merely verbal questions, the definition question ‘What is a concept?’ itself seems to be in need of justification. Why should we be interested in an answer to this question—instead of only wishing to find out about the tenability and advantages of different theories in which the word ‘concept’ is used in various ways? This is not a trivial question, and the mere observation that ‘the participants in the dispute don’t generally view it as a terminological one’ (Margolis/Laurence 2006, sct. 1.4) clearly does not amount to a satisfactory answer. To be sure, if different theories in which the word ‘concept’ or a translation of it plays a significant role covered much of the same ground, this would be a good reason to raise the definition question. If they had the same subject matter, tried to account for at least part of the same phenomena and yet disagreed in various ways—if in other words they were competing theories—an investigation into the nature of entities named concepts would be worthwhile. But this is something which needs to be shown. As long as we do not know which philosophical problem we want to solve, which question or questions we want to answer by invoking concepts, the question about the nature of concepts is an empty one. Despite the fact that ‘the theory of concepts has become such a rich and lively topic in recent years’ (Margolis/Laurence 2006, Introd.) the unqualified question ‘What is a concept?’ does not seem to be a philosophical question in its own right. One way of qualifying the question would be to refer to a certain tradition in terms of which one wants to treat concepts. Our questions and the answers we come up with as well the concepts we invoke when we think about concepts are to some extent heirs of questions previously raised by others, they are the heirs of earlier conceptions and theories. We are likely to manoeuvre ourselves into even greater confusion if we disregard the history of philosophy and try to answer the question what a concept is independently of any theory as well as its predecessors and competitors. There is a long-standing tradition of bringing together what today would to many seem quite different projects. For example, some people wanted to give an account of the natures or essences of things as well as an account of our ability to apprehend particulars as being of a certain kind 156

and an account of our ability to convey our thoughts and to use meaningful words. The ambition to give all these accounts in one go is as old as ancient philosophy. Thus, we have reason to suspect that some of those who nowadays try to explain a number of related things by reference to concepts really share a concept (or have overlapping concepts) and in the context of their accounts try not to use the word ‘concept’ equivocally. But again, this needs to be spelled out. Intended continuity as well as attempts to break with certain aspects of a tradition would have to be indicated to clarify what is at stake when we think about concepts. Another possible justification of the definition question ‘What are concepts?’ would invoke the fact that there is an ordinary way of talking about concepts. In addition to figuring as a key concept in certain philosophical theories, the concept of a concept is employed in non-theoretical discourse. ‘Concept’, or ‘Begriff’, are not only technical terms but also words of our ordinary language. The German word ‘Begriff’ actually occurs in quite a lot of composite expressions (sich einen Begriff von etwas machen, keinen Begriff von etwas haben, alle Begriffe übersteigen, nach eigenen Begriffen, dem eigenen Begriff nach, schwer von Begriff, jemandem kein Begriff sein, ein dehnbarer Begriff). I will not try to translate these examples here. It is clear, however, that hardly any of these idiomatic expressions would normally be translated by using the word ‘concept’; ‘idea’, ‘notion’ and even ‘image’ or ‘picture’ would be better candidates, and some renderings would do without any noun corresponding to the German word ‘Begriff’. There are several things to be learned from this. For one thing, a closer look at adequate translations may help to identify philosophical traditions where concepts are supposed to play a certain role and reveal something about the relations the concept of a concept bears to other concepts. Secondly, the great number of translation problems of this kind adds to the plausibility of doubts about the existence of concepts as abstractions from uses of words. I will come back to this. For the time being I just want to point out that there is this idiomatic, ordinary kind of use of the word ‘Begriff’ in German, and I rather hope that native speakers will agree that in English too there are similar common or garden uses of the word ‘concept’, as for example in ‘the concept of a social class’, ‘he can’t grasp the basic concepts of mathematics’, ‘the concept that everyone should have equality of opportunity’.1 If there is an 1. Oxford Advanced Learner’s Dictionary, 6th edition; see Fowler’s Modern English Usage, revised 3rd edition: “He was the man who invented the concept of a weekly news magazine”, “We aim to sell a total furnishing concept based on the ‘one pair of eyes’ principle”.

157

ordinary use at all, different theories employing the concept of a concept can claim that they capture our ordinary use of the word ‘concept’ (or parts of it) and therefore meet one of the requirements for a good explication (in Carnap’s sense of an explicative definition). This is one sense in which theories can compete. Insofar as we are interested in explicating our concepts we may be interested in what a theory tells us about what exactly a concept is, independently of other things the theory may be designed to account for. In this case the qualified definition question could be, ‘What are the things we ordinarily call concepts?’. (Even if we are not interested in clarifying our non-technical concept of a concept, as competent speakers of our respective mother tongues we should be able to assess explicit as well as implicit claims about ordinary usage.) 3. To possess a concept is to possess certain abilities While the nature of concepts is a highly disputed issue, concept possession is a less controversial matter. And this is the point where I would like to join in the discussion about concepts. Many people agree that to possess or master a concept is to have a variety of abilities, faculties, or capacities. One reason for this may be that by acknowledging that possessing a concept is possessing a variety of abilities, faculties or capacities one does not concede anything as to which abilities exactly are relevant to concept possession, nor does one admit anything, for instance, about whether there are abilities whose possession can count as a necessary or a sufficient condition for concept possession. One even remains neutral on the question whether the relevant abilities are innate or acquired. A further reason may be that ‘ability’ is a very flexible term, and as long as one does not distinguish between different kinds of abilities as well as between different aspects of their possession and employment, one does not seem to commit oneself to much by saying that to possess a concept is to have certain abilities and that to acquire a concept is to gain such abilities. Even the representationalist can concede that to possess a concept is to have an ability, namely the ability mentally to represent something. If we grant that possession of a concept amounts to having certain abilities, we may wonder what abilities that could be. Many people hold the view that mere possession of recognitional abilities and discriminatory powers is not sufficient for the possession of a concept. A being that recognizes Fs and has the power to distinguish Fs or things that are F 158

from things that are not, i.e. from non-Fs, cannot merely in virtue of that be said to posses the concept F. On the other hand, as regards some of our concepts, possession of recognitional abilities does not even seem to be required for counting as having these concepts. Think of a theoretical concept like that of an electron for instance. Let us consider a concept whose possession implies the possession of certain recognitional capacities, the concept of something’s being red for instance. A creature to whom possession of the concept may be ascribed must be able to do more than just recognize red things and react to red things in ways that are different from her ways of reacting to things that are not red. If we look at our ordinary use of the word ‘concept’ and think of our fellow human beings, we are bound to come up with the following: She must, for instance, have grasped that red is a colour; that it applies to extended objects, but not to sounds or smells; that if something is red all over, it cannot simultaneously be green all over; and so on. Imagine a creature that does not possess any of these additional concepts (or lacks most of them) and therefore has no grasp of how they are interrelated and related to something’s being red. If we were dealing with a fellow human being we would certainly be reluctant to ascribe to her the concept of something’s being red. Here, an important aspect seems to be the logical geography of our concepts, as Ryle and Wittgenstein have called it (see Backer/Hacker 2005, 284–287). It does not seem possible for a being to count as someone who possesses just one concept. The possession of concepts entails the ability to draw certain inferences from and to propositions in which these concepts figure, and hence it seems plausible to hold that possessing a concept amounts to something very much like mastery of the rule-governed use of a word that expresses the concept. But maybe this was rash. What about animals that don’t use language? Many animals have better recognitional capacities than humans do, and some have recognitional capacities that humans lack altogether. If we agree that concept possession is a complex ability admitting of degrees, the question seems to pose itself whether it would not be a good idea to ascribe a rudimentary kind of concept mastery on the mere basis of the candidate’s having certain recognitional capacities. Surely no philosopher can stop us from proceeding this way. But philosophers can point out the price we are liable to pay for doing so. Peter Hacker is one of these philosophers. He says that if we ascribed concept possession on the basis of mere recognitional abilities, we

159

would detach the concept of concept possession from the cluster of abilities with which it is normally and usefully associated, and detach the concept of a concept from its connection with the concepts of application and misapplication, use, misuse and abuse, subsumption, extension, replacement and substitution. (Hacker 2007, 239 fn. 14)

Exactly how many concepts a creature needs to possess and exactly what kinds of ability it needs to have to count as possessing a concept may be a matter for debate and ultimately a matter for decision. But if we believe that the ability to draw certain inferences is relevant to concept possession, a possessor of concepts will have to be able to grasp propositions, which are true or false. And he will have to be able to entertain a number of true beliefs. 4. On Glock’s attempt to distinguish between sorting and classifying In an attempt to account for the fact that concept possession involves more than only the ability to discriminate, Hans-Johann Glock has suggested the following answer to the question what concept possession amounts to (which is of course related to the question of how to individuate concepts): To possess the concept F is not just the ability to sort things into Fs and nonFs, but the ability to classify them as Fs. And that is to distinguish Fs from non-Fs for the reason that the former possess certain features, namely those constitutive of being F. If A singles out triangles on account of the numbers of the angles, she exercises a different concept from B, who singles them out on account of the numbers of their sides. (Glock 2006, 59)

One could well imagine chimpanzees learning to pick out triangles by paying attention to the number of sides while others learn how to pick out triangles by paying attention to the number of angles. In Glock’s view, this is a case in which the two groups of chimpanzees have acquired different concepts that apply to the same things. So this suggestion would not only account for the possibility of different concepts’ applying to the same things; according to Glock it would have what he considers the additional advantage of allowing for the conceivability of ascribing concepts to nonlinguistic creatures. Here, I should like to comment on a few aspects of Glock’s suggestion. My first observation is a general remark that is addressed, not so much to Glock but rather to those philosophers, linguists, neuroscientists, biolo160

gists and other people who these days try to sell their theories by appealing to our feelings about ways of treating animals as well as to those who buy these theories for what seem to be the wrong reasons. As a matter of fact, philosophers do not usually link concept possession with language in order to exclude non-human beings from concept possession. The question whether non-human beings can have a language is not directly touched on by this idea of a link, nor is the question whether some non-human beings do in fact have a language. The reasons for thinking of concept possession in terms of linguistic abilities are of a different kind. Even Descartes did not justify his notorious position by simply stating that only languageusing creatures are capable of thought. What he claimed was that the use of a language is our best evidence for a creature’s having thoughts. And he tried to give additional reasons for his contention that animals do not think (Discours, V, AT VI, 56–59). Nowadays, to some philosophers only linguistic behaviour seems complex enough to account for perceptible differences between creatures manifesting different propositional attitudes and for differences made by changes in attitude or inferences that are drawn or thoughts that are had or judgements that are made and can turn out true or false. But if some other kind of behaviour could be shown to be complex enough to account in terms of manifest behaviour for the differences held relevant to concept possession and thought, it too would serve our purpose, even if one could come up with a reason for thinking that the behaviour in question should not count as a kind of language use. This, I think, is the spirit in which Glock’s suggestion ought to be read. Let us have another look at this suggestion. Glock wants to draw a distinction between sorting things into Fs and non-Fs, on the one hand, and classifying them as Fs, on the other. According to Glock, classifying amounts to distinguishing between Fs and non-Fs for a reason, namely for the reason that Fs possess certain features while non-Fs do not. Now, we would surely take a fellow human being who realizes that a particular thing has certain features and, starting from there, proceeds to take the thing to be of a certain kind to draw some kind of inference. We would take her to infer that the thing in question is of a certain kind from its displaying certain features. But at the same time it seems that she employs at least two concepts—the concept of the features she realizes the thing to have and the concept of the kind of thing she eventually takes the thing to be. On the one hand, this nicely fits Glock’s speaking of the creature’s doing something (i.e. distinguishing Fs from non-Fs) for a reason. On the other 161

hand, this cannot be the intended reading of his suggestion. For on this reading the assumption that to possess a concept amounts to an ability to classify things (to take things to be of a certain kind for a reason) seems to lead to an infinite regress. One would have to possess a concept in order to possess a concept. (This is a difficulty Locke’s account of abstract ideas runs into, too.) Let us assume therefore that classifying things as Fs (distinguishing Fs from non-Fs for a reason) does not involve the exercise of two concepts but only one, namely the concept F. In this case we would not take the creature classifying things to be drawing any kind of inference. But then our description of what went on was wrong or misleading from the beginning. For if the creature does not employ two concepts, which is what Glock seems to suggest when he says that the recognized features are those that are constitutive of being F, it is hard to see what classifying as or taking something to be F for the reason that it displays certain features amounts to beyond recognizing Fs and distinguishing them from non-Fs, i.e. in Glock’s terminology to sort things into Fs and non-Fs. So we would not be justified in speaking of the exercise of a concept at all. It seems clear that in the example given by Glock the chimpanzees either draw an inference, thereby employing two different concepts (namely having three sides and being a triangle or having three angles and being a triangle), or they do not draw an inference. In the latter case there is no reason to suppose that their behaviour amounts to more than discriminating or recognizing or reacting differently to certain kinds of things. (Here I want to add that the wisdom of describing these activities as ‘discrimination’ or ‘recognition’ seems to me dubious.) Any description of their behaviour as doing something for a reason would stand in need of further justification. For such a description invokes two things: what is done and a reason for which it is done. But are there really only these two alternatives: drawing an inference and thereby employing two concepts, on the one hand, and drawing no inference while exercising no concept but only reacting differently to different kinds of things, on the other? Couldn’t we for example say that paying attention to something does not amount to employing a concept? Maybe we could. But words to the effect that paying attention to something and then doing something else for the reason that the effort of paying attention had the outcome it in fact had do not easily seem to lend themselves to giving a description which presupposes the possession of only one concept. 162

Glock’s initial description of the chimpanzees is misleading in the following way. He claims that the chimpanzees that classify triangles by paying attention to their having three sides have a different concept from those that classify by paying attention, not to the number of sides things have, but to the number of angles. What are these different concepts? The concept of having three sides and the concept of having three angles? But then how can one say that both groups single out certain shapes or triangles without bringing in a third concept? Evidently they do not single out the same shapes. They single out different shapes. Some single out Fs (things that have three sides), the others single out Gs (things that have three angles). As it happens, all Fs are Gs and vice versa. Try to imagine what it would be like if things were different. Unlike chimpanzees, we can tell a priori that all three-sided figures (enclosing a surface) cannot help being figures with three angles. This shows just how much may be involved in the possession of concepts of geometrical figures. Can one be said to be paying attention to the number of angles a figure has if one does not have the concept of an angle? But to have the concept of an angle implies a good deal of geometrical knowledge, and this is knowledge that we find more or less difficult to acquire. Among the abilities the chimpanzees must possess if we want to ascribe the possession of the concept of having three sides to them would perhaps be the ability to distinguish between having three sides and having three angles. To be sure, if chimpanzees learned to discriminate in some way or other not only between things that are triangular and things that are not, but also between many other things that are relevant to what we call arithmetic and geometry (e.g., number and length of sides, number and size of angels), and if they began to draw inferences from the number of sides a thing has to the number of angles it has, for instance, or from its being a triangle to its not being square and its not being a circle, we would probably ascribe to them some knowledge of geometrical figures as well as all the relevant concepts. Drawing an inference may mean something like acquiring a belief on the basis of one or more other beliefs. It means coming to hold something to be the case (a proposition to be true) on the basis of holding something else to be the case (holding a different proposition or different propositions to be true). And it is yet another philosophical question what believing something, acquiring a belief and changing one’s beliefs amounts to. But if it is alleged that philosophers who introduce the ability to draw an inference and the ability to hold something to be the case into the debate 163

about concepts and concept possession are shifting ground to evade or prejudge the question whether animals can have concepts and are able to think, then this allegation is wrong. It just needs to be remembered that concept possession is closely connected with the abilities to have true and false beliefs and to draw inferences. The question to what extent certain animals are capable of entertaining beliefs and drawing inferences is a different matter. 5. An easy refutation of ability accounts? For the time being, let us put the possession question on one side and take a look at the definition question. If one believes that possession of a concept amounts to possessing certain abilities, one may be tempted to identify concepts with abilities. Glock objects to this idea for the following reasons: to explain and define a concept is not to explain and define an ability; concepts, unlike abilities, can be instantiated or satisfied by things; concepts, unlike abilities, have intensions and extensions; concepts, unlike abilities, can occur in propositions or statements (Glock 2006, 48f.). I do not wish to dispute these claims here. But I would like to point out that an assessment of Glock’s reasons for rejecting ability accounts of concepts will depend on what aim one pursues in invoking concepts. A linguist may take concepts to be intensions rather than as having intensions. A metaphysician may take concepts to be instantiated or satisfied by things, but he may be reluctant to say that they occur in propositions or statements. And without further argument it is not clear that anybody’s conception of concepts will meet all of the conditions Glock’s mentions. It is far from obvious that his own suggestion as to the nature of concepts does meet them. According to Glock, a concept is ‘the principle or rule according to which someone [who possesses the concept] can classify’ (Glock 2006, 49). Rules do not have extensions or intensions and they are not instantiated or satisfied by things. Nor do they in any obvious sense occur in statements or propositions. Defining and explaining a rule will in many cases amount if not to explaining an ability then to describing behaviour which accords with the rule. None of this tells against Glock’s proposal. Rather, it shows that we cannot ask questions about the nature of concepts independently of any use we want to make of the concept of a concept or of a theory in which this concept is embedded and which is designed to solve certain problems. 164

If we want to ascribe concepts to non-linguistic beings and can convince ourselves that a distinction between sorting and classifying can be drawn without referring to abilities that presuppose speech, Glock’s proposal may be just what we should adopt. The question that remains is why we would want to ascribe concepts to non-linguistic beings. If we hope thereby to pave the way to ascribing thought and reason to animals, Glock’s proposal is unlikely to do the trick. As the distinction between sorting and classifying cannot, as we have seen, rely on the subject’s drawing an inference, it remains to be seen whether the distinction, if it can be drawn at all, can be drawn in such a way that the ability to classify would count as evidence for the ability to think or to reason. 6. Thoughts and concepts as contents There is yet another problem with Glock’s proposal. Glock forcefully and convincingly argues against some versions of a representational theory of mind that neural firings cannot be symbols (as that theory has it), as there is no one who uses them to represent anything in a conventional way: ‘For something to be a symbol, it must have a rule-governed use, there must be a correct and an incorrect way of employing it’ (Glock 2006, 27; see Hacker 1987). If one accepts C. S. Peirce’s distinction between symbols, icons and indices, it is obvious that neural firings or patterns of neural firings are neither icons nor indices that could represent concepts. For icons would have to resemble what they represent, and indices would have to be taken as representing something by somebody who knows about their natural relation to or dependence on what he takes them to represent. In ordinary circumstances, no one can take his brain states or the patterns of neural firings in his brain to indicate anything, simply because we are usually not in the least aware of them. From his attack on the representational theory of mind Glock concludes that thoughts and concepts are not representations at all. They ‘are not what represents, but what is represented’ (Glock 2006, 45; see ibid., 46). They are shareable in the sense that different people may have the same concepts and entertain the same thoughts. Thoughts and concepts can be thought and made use of. They are contents rather than vehicles of content. But then, what are the vehicles of thoughts and concepts? What is a bearer of content? What item does the representing? We use words or linguistic phrases to express concepts. We use sentences to express thoughts. But 165

what does it mean to take thoughts and concepts to be the contents of our thinking? How can something be the content of an act or an episode? Representations, Glock holds, require a medium. They have representational properties in virtue of having certain non-representational properties. But if thoughts and concepts are what is represented rather than what represents, by what exactly are they represented when we think? This is the question Fodor’s hypothesis of a language of thought tries to answer. And it also is a question which philosophers since the middle ages have tried to answer by advancing representational theories of the mind—quite a number of them calling what they took to be mental representations ideae or notiones. According to them, in thinking we have ideas. Traditionally, ideas have a twofold nature as act and object of thought. Descartes famously spoke of ideas considered formaliter or objectivae (Meditationes, prefatio ad lectorem, AT VII, 8). They were taken to be modifications of the soul as well as intentional objects. One of the challenges would then be to explain the relation intentional objects bear to the modifications of the soul or thinking substance. What makes an idea of x an idea of x rather than of y? Traditional approaches refer to likeness, causation and teleological considerations. Nowadays hardly anybody seems to believe that mental representations have to be like the things they represent. For one thing, there seem to be different ways of representing something, e.g. by functioning as indices or symbols rather than icons. Secondly, it is hard to see how a mental entity like an idea or a mental symbol (or a brain state) could actually be like the thing it is supposed to represent in relevant respects. (The similarity between the concept F and Fs would have to be closer than the similarity between the concept F and anything that is not F.) Thirdly, likeness (resemblance, similarity) is a symmetric relation, while the relation between a mental representation and whatever it represents is not. Adherents of the hypothesis of a language of thought like Fodor tend to give a causal account of mental content, thereby rejecting the claim that for something to be a symbol it must have a rule-governed use. In their view, mental representation cannot possibly be a matter of convention. To allow for mis-representation Fodor advanced his idea of asymmetric dependency, according to which causal relations between a mental symbol and things that do not belong to the extension of the symbol depend on causal relations between the symbol and the things that do belong to its extension (Fodor 1990, ch. 3). If things belonging to its extension were not to cause tokens of a symbol to occur in a subject’s mind, things that 166

do not belong to its extension would not do so either, but not vice versa. It has been argued however, that this suggestion, ingenious as it may look at first sight, does not work. The reason given is that the set of things that would actually cause a token of a given mental symbol (which is so far individuated only syntactically) to occur is homogeneous with respect to its members causing an occurrence of the symbol. Unless it is presupposed that the symbol has a definite meaning, there is no reason to single some of them out as causing false or mistaken occurrences of the symbol. This is why mere causal relations cannot account for the semantics of the postulated language of thought.2 Descartes had yet a different reason for not adopting a purely causal account of mental representation. He pointed out that on such an account there would be no way of determining the beginnings of the relevant causal chains. This is why Descartes thought it necessary to invoke teleological considerations. Ultimately, he believed, only God would guarantee our ideas to be ideas of whatever they represent (Meditationes, VI, AT VII, 86–89). If we hold that neither likeness nor causality or teleology can explain intentionality, we may want to drop the idea of mental representation altogether. For one of the problems we then face is how to reconcile the fact that thinking seems to be something one does in private with the fact that representation seems to involve rule-governed uses of symbols. If it is impossible to follow a rule privately, this problem seems to have no conceivable solution. Of course, ideas have been taken to be icons and indices. But even icons and indices need somebody who uses them, and arguably even the use of icons and symbols ultimately cannot be private, nor can such use be made in one of the ways in which earlier philosophers thought we employed our ideas. But unless we abandon the idea that thoughts and concepts are something that is represented (by other than linguistic means or uses of signs), we are bound to yield to the temptations of one or the other version of the representational theory of mind. 7. A deflationary approach to concepts Let us assume for the time being that we agree on the following claim. The possession of concepts amounts to possessing a complex ability or a 2. See Saporiti 1997, chap. VI. for an elaboration of this argument.

167

number of interrelated abilities which we take any language-using creature to possess. Even if we agree on this, we have not yet answered the so-called definition question: What, then, are concepts? The way I like to think of concepts is in terms of the use of linguistic expressions. A concept, Wittgenstein says, ‘is a technique of using a word’ (LPP 50; cf. MS 163: 57r.). Those who think that this is too crude will say that concepts are abstractions from uses of words (Hacker 2007, 16). Note that one does not necessarily have to abstract from the use of a word in order to allow for speakers of different languages to have the same concept. For the use of a word in one language (e.g. ‘tree’) may be the same as the use of a different word in another language (e.g. ‘arbre’). The use of a linguistic expression is a rather complex affair, though. It will be interrelated in various ways with many other expressions and their use in a given language. It is therefore arguable that speakers of different languages do not share that many concepts after all. It will often be more accurate to say that they have, not the same, but overlapping concepts (where ‘overlap’ is a matter of degree). At any rate, the basic idea is that to possess a concept is to have mastered the use of a word or phrase. Surely a view of this kind will spare us the trouble of postulating obscure entities. All we do in talking about concepts and trying to explain or clarify them is to talk about uses of linguistic expressions—about people and the different ways in which they use words. In particular, this view does not postulate inner states, events or processes suited for representing concepts and thoughts which, in their turn, play the role of content. At the same time, this view seems to be strangely unsatisfactory. In a sense, the proposed view does away with concepts—it seems to deprive us of our topic. But what is at stake? What was it that we wanted concepts for? Instead of concentrating on the question how to define the word ‘concept’ we should ask ourselves why we want to talk of concepts in addition to, and independently of, the use of words. What purposes did we have in mind when we first thought of invoking concepts? Philosophers have hoped to explain our ability to classify particulars as being of a certain kind by reference to concepts that we grasp. According to this idea, possessing a concept is what enables one to recognize things as belonging to a certain kind. When it was suggested this idea made sense because it was combined with the claim that a concept is a representation of the defining features which would make a particular belong to a kind—its essential or characteristic features. To grasp a concept meant recognizing the nature of those things that fall under it. This approach, however, 168

becomes incoherent if it is combined with the empiricist thesis that we do not have innate ideas or concepts but acquire concepts by abstracting from experience and a nominalistic view according to which ideas and properties exist only insofar as they occur in a mind or are instantiated by something. 8. Do thoughts and concepts have a language-independent structure? Philosophers have hoped to be able to explain inferential relations between propositions as well as the systematicity and productivity of thought by invoking concepts. There is a long tradition of taking concepts (ideas or notions) as parts or constituents of thoughts. Regardless of whether concepts (and thoughts) are taken to be abstract entities or something that occurs in people’s minds or even brains, today many seem to agree that concepts are parts of propositions or thoughts (Rey 1998, Introd., Margolis/Laurence 2006, Introd.). Just as a sentence consists of words or phrases, so the proposition or thought expressed by the sentence is supposed to consist of concepts. Concepts are then said to be expressed by words or phrases and to occur as or to be parts of propositions or thoughts. If we hold on to the view that a concept is, or is closely connected with, the use of a word, none of this seems to make much sense any more. But one may doubt whether speaking of concepts as parts of propositions is a sensible thing to do—in particular, if it is meant to imply that a given proposition or thought determines which concepts the proposition consists of. For how are we to justify the claim that propositions are structured in any determinate way? It is supposed that the same proposition can be expressed by different sentences. These may differ with regard to their syntactical structure, as for instance the following two sentences do: ‘Hannah gave a book to Karl’ and ‘Karl was given a book by Hannah’. As the same proposition can be expressed in many different languages with considerably different kinds of syntax, it is obvious that the alleged structure of a proposition cannot be derived from the structure of a sentence which is used to express that proposition. One may expect to find an answer in Frege’s doctrine of thoughts as structured entities that always have to contain an unsaturated and a saturated part. Frege takes concepts to be expressed by linguistic expressions which occur in the sentences that we use to express thoughts. One would expect that the different parts of a thought (Gedankenteile) correspond to 169

different linguistic expressions—and this is in fact Frege’s view. But even according to Frege the same thought can be broken up in different ways into different parts. Which concept is part of a given thought depends on the way we divide it up; it depends on the way we choose to look at it or choose to describe or express it. Here there are two examples from Begriffsschrift § 9: ‘The circumstance that carbon dioxide is heavier than hydrogen’ and ‘the circumstance that carbon dioxide is heavier than oxygen’ are the same function with different arguments if ‘hydrogen’ and ‘oxygen’ are regarded as arguments; on the other hand, they are different functions of the same argument if ‘carbon dioxide’ is taken as the argument. (Frege 1879, 16; translation in Beaney 1997, 66)

Frege points out that whether the concept of being heavier than hydrogen or the concept of being lighter than carbon dioxide is a part of what is expressed by the first phrase is a question of how we choose to look at it. These different ways of looking at it do not according to him change the content of the whole phrase. They have to do with the way in which we grasp this content. Frege gives another example: The proposition that Cato killed Cato shows the same thing. Here, if we think of ‘Cato’ as replaceable at its first occurrence, then ‘killing Cato’ is the function; if we think of ‘Cato’ as replaceable at its second occurrence, then ‘being killed by Cato’ is the function; finally, if we think of ‘Cato’ as replaceable at both occurrences, then ‘killing oneself ’ is the function. (Frege 1879, 16; translation in Beaney 1997, 66)

Here I am not concerned with the problems this may pose for Frege, nor with the somewhat unfortunate way in which he expresses himself. The only point I want to make is that even in Frege’s view there is more than just one way of looking at a given thought or a proposition. Depending on how we decide to structure the thought, it will have different parts (Gedankenteile), and in accordance with that different concepts will figure in the proposition. Thus it appears that Frege holds that thoughts do not have one determinate structure. It is a disputed question whether we can account for conceptual relations between propositions without supposing them to be structured independently of the way we express them by linguistic means. To conclude: I think that there is a certain danger in asking questions about the nature of concepts independently of any theoretical background

170

and without taking into account at least some of the various purposes which concepts have been meant to serve in the history of philosophy. The danger is that of confusing different projects, some of which have a long tradition. There are, to be sure, many things contemporary philosophers want to do with concepts that I have not mentioned. And maybe there are good reasons for reviving certain metaphysical notions in order to complete some of these tasks. But if there is no specific task of that kind in view, that is, independently of any theoretical question that it might help to answer, I see no need to think of a concept as something over and above the use of a word. In particular, this deflationary approach to concepts recommends itself if one does not believe in representationalism. Someone who wishes to construe concepts, not as mental representations, but as constituents of abstract entities like propositions or Fregean thoughts faces a difficult task. For if he wishes to invoke parts of thought as existing independently of our linguistic ways of expressing thoughts, he has to justify the assumption that propositions (thoughts) have a language-independent structure.

REFERENCES Baker, Gordon and Peter Hacker. 2005. Wittgenstein. Understanding and Meaning, 2nd edition. Oxford: Blackwell. Beany, Michael (ed.). 1997. The Frege Reader. Oxford: Blackwell. Descartes, René. 1637. Discours de la méthode pour bien conduire sa raison, et chercher la vérité dans les sciences. Leiden. Descartes, René. 1641. Meditationes de prima philosophia. Paris. Fodor, Jerry. 1975. The Language of Thought. New York: Crowell. — 1987. Psychosemantics. Cambridge (Mass.): MIT Press. — 1990. A Theory of Content, Cambridge (Mass.): MIT Press. — 1998. Concepts: Where Cognitive Science Went Wrong. Oxford: Clarendon. — 2004. “Having Concepts”. Mind and Language 19, 29–47. Frege, Gottlob. 1879. Begriffsschrift, eine der arithmetischen nachgebildete Formelsprache des reinen Denkens. Halle: Nebert. Glock, Hans-Johann. 2006. “Concepts or Abilities?” In: Ezio Di Nucci and Conor McHugh (eds.). Content, Consciousness, and Perception. Essays in Contemporary Philosophy of Mind. Cambridge: Cambridge Scholars Press, 36–61. Hacker, Peter. 1987. “Languages, Mind and Brains”. In: Colin Blakemore and Susan Greenfield (eds.). Mindwaves. Oxford: Blackwell. — 2007. Human Nature: The Categorial Framework. Oxford: Blackwell.

171

Margolis, Eric and Stephen Laurence. 2006. “Concepts”. In: The Stanford Encyclopedia of Philosophy, February 2006. Rey, George. 1998. “Concepts”. In: Edward Craig (ed.). The Routledge Encyclopedia of Philosophy. London: Routledge. Saporiti, Katia. 1997. Die Sprache des Geistes. Berlin: de Gruyter. Wittgenstein, Ludwig. [LPP]. Wittgenstein’s Lectures on Philosophical Psychology 1946–47. Peter Geach (ed.). Hassocks: Harvester Press 1988. Wittgenstein, Ludwig. [MS]. Wittgenstein’s Nachlass. The Bergen Electronic Edition. Oxford: Oxford University Press 2000. (References according to manuscript number and page number.)

172

Grazer Philosophische Studien 81 (2010), 173–188.

CONCEPTUAL ANALYSIS FOR REPRESENTATIONALISTS Frank JACKSON La Trobe University / Princeton University The Australian National University Summary We use words to mark out patterns in nature. This is why a word like ‘nutritious’ is so useful. One way of thinking about conceptual analysis is as the business of capturing the structure in the patterns so picked out, for it is not credible that the patterns are one and all sui generis. This paper spells out this way of thinking about conceptual analysis. Along the way we discuss: the role of intuitions about possible cases with some reference to the experimental philosophy debate, why analyses are often hard to find, and whether conceptual analysis so conceived presupposes a controversial version of the description theory of reference.

1. Introduction People differ in where they draw the boundary between trees and shrubs. What’s a good way to find out where some particular person, Mary let’s suppose, draws the boundary? One way is to see how she classifies individual examples. This method has to be used with care (as does any method). For example, we need to be sure that she has the requisite information, that she’s not suffering from the kind of hangover Bob Newhart’s pilot suffered from, and that she’s bearing in mind possible consequences for other classifications that she makes—‘Hold on. In other contexts you take account of what’s normal. The plant you’ve just said is a shrub is of a type that is normally much bigger. Shouldn’t you, to be consistent, classify it as a (stunted) tree?’ But even with these caveats, there is something essentially right about the idea that way to find out the cases someone uses a word ‘W’ for, their concept of W, lies seeing how they classify putative examples. How else might we approach the question of which cases someone uses the term ‘dictator’ for, their concept of dictator, other than by seeing which cases they use the word for? It is their concept we are investigating after all.

What we have here is a bit of folk wisdom. We analytical philosophers might describe it as ‘conceptual analysis using the method of intuitions about possible cases’ but that should not disguise the fact that it is folk wisdom and not some arcane bit of technology special to our profession. The folk know perfectly well that very many terms—‘tree’ and ‘dictator’ being two examples among a host—serve to classify things, that what gets classified are putative examples, that people may differ in the classifications they make with some given term, and that a good way to find out about possible differences is to do some intelligent questioning focussed on putative examples. But folk wisdom can raise difficult questions, questions properly for philosophers. We’ll be concerned with three: the extraction problem; what to say about the case of knowledge (here we’ll touch on some of the questions raised under the heading of ‘experimental philosophy’); and, in a short final section, with what to say about the suggestion that the project of conceptual analysis presupposes a controversial—indeed, according to some, a discredited—theory of reference. I will set the discussion in the context of a representationalist picture of thought and language. It seems to me that conceptual analysis makes best sense against this background. I don’t see this as giving a hostage to fortune. That thought and language represent seems to me sufficiently obvious to be a proper starting point. 2. The representationalist background I am looking at a tree. My perceptual experience represents that there is a tall tree about five metres in front of me. It represents that things are a certain way, that the world I am in is a certain way—the way with a tall tree five meters in front of me. I might use a sentence to capture this representational content, most obviously the sentence ‘There is a tall tree about five meters in front of me’. True, my perceptual experience will be much richer than can be captured in a short sentence and its total content will outrun my linguistic capacities. But what will be true is that part of how my perceptual experience represents things to be can be given in a sentence. This means that there is a notion of content—call it representational content—that corresponds to a partition among possible worlds. Perceptual experience represents that the world—the huge complex structure that we inhabit for a very short time—is a certain way, and that ipso facto 174

means that our experience represents that we are located in a world of so and so a kind, as opposed to one of such and such a kind. (Or more accurately, experience represents our location and orientation within a location in logical space. Perception represents the kind of world one is in by representing how one is oriented within a world of that kind—facing towards a tall tree that is two meters away from one, as it might be. We will, however, suppress this complication. It is important but not for our purposes here.) There is another path I might have followed. I might have focussed on the question, what proposition does the sentence ‘I see that there is a tall tree about five meters in front of me’ express? and proceeded to discuss various theories about the propositional object that there is a tall tree about five meters in front of me—Fregean, Russellian etc. I might, that is, have cashed out how the experience represents things to be in terms of a proposition, and then discussed various accounts of propositions. On some accounts of propositions—those in terms of sets of possible worlds—this path ends up in the same place as the path I in fact took: representational content equals location in logical space. But on other, and probably more popular, accounts of propositions, we would end up in one or another different place. I took the path I did in order to emphasise what seems to me compelling: perceptual experience is world-directed. It represents what our world is like; indeed, that’s the most important thing it does for us. By representing what our world is like it delivers putative information about our world, and that’s what we need to get what we desire. Of course there are roles for propositions conceived of as other than sets of worlds—seeing that something is half-full can be distinguished from seeing that it is half-empty, despite the fact that being half full and being half empty are one and the same way for things to be—but that should not blind us to the evident fact that seeing is world-directed. And surely the same goes for sentences. The point follows from what we have just said for sentences that capture part of how experience represents things to be, but it is equally obvious for sentences like those produced by Einstein when he advanced the special theory of relativity in 1905, and the more recent sentences warning us about global warming. Einstein was telling us something important about the kind of world we are in, and the same goes for those telling us about global warming. If those who warn us about global warming turn out to be wrong, what will show this are investigations into what the world is like; likewise, if we want to know if ‘There is a tall tree about five meters 175

in front of me’ is true, we examine how things are in front of me. Or take the half-full half-empty example. Although we can distinguish (1) I see that the glass is half empty from (2) I see that the glass is half full for we can describe cases where (1) and (2) differ in truth value, what needs to be shown about how things are in order to check whether the experiences reported in (1) and (2) get the nature of the world right is exactly the same. (Incidentally, the appearance of words like ‘nature’ and ‘like’ in the foregoing should not be read as making a commitment to representational content being exhausted by the qualitative; that’s another issue, one we won’t be going into.) I labour the point because I often hear the criticism that I’m assuming a discredited account of content. I reply, what on Earth was Einstein doing in 1905 if he wasn’t telling us about the kind of world we’re in? The sentences that represent—it is near common ground that some sentences don’t represent, though it is contested which ones don’t—represent the world as containing things of various kinds: trees, dictators, inflation, free actions, knowledge and so forth. A sentence that represented the world as containing various things without representing them as being of any type whatever would hardly be worth uttering. In representing how the world is, we represent it as populated with items that fall into various categories. We represent the world as patterned, in a highly inclusive sense of patterned that counts any similarity that doesn’t collapse into a projection-less infinite disjunction to count, and we do this by using words that stand for these patterns. We are mainly interested in patterns that have some reasonable degree of unity but we make exceptions. Philosophers use ‘grue’ to mark out a pattern that lacks any obvious unity but do so in the interests of making a philosophical point, and researchers into colour vision use terms for certain unobvious patterns in reflectances in the interests of marking out the causal underpinning for colour experiences (and which are, on some views, the colours themselves). The two patterns I have just instanced are not sui generis. They have structure. Grueness, for example, is a disjunction of conjunctions of colour and time of observation. Or take the much more obvious and 176

unified pattern in nature for which we use the word ‘circle’. It is the pattern we get when we aggregate points so that they lie in a plane equidistant from some given point. Or take the pattern for which we use the word ‘inflation’. It is a complex of conjunctions of monetary patterns, a principal feature of which is that elements in the later monetary patterns take higher values than they do in the earlier ones. Our world is a huge four-dimensional structure in space-time (or if it isn’t, what I’m about to say would stand with appropriate revamping I trust) replete with patterns. Many of these patterns are constituted by aggregations of other patterns. Take the property of being the tallest person in the world. In order to make a world where some given person is the tallest person in the world, it is enough for God to give each and every person a specific height. She doesn’t need to add anything to make whoever it is be the tallest. The job is done when the individual heights are assigned. The situation is like the one materialists hold obtains for consciousness. Put together aright purely physical constituents in purely physical ways and you thereby have consciousness. There’s no more to do if materialism is true. For materialists, consciousness is a structured pattern in the purely physical. Often supervenience tips us off—one way or the other—about structure in patterns. Why are we confident that inflation is a structured pattern in the monetary? Because we find it immediately plausible that enough duplication in the monetary between two worlds ensures duplication with respect to inflation between those two worlds. Why do many resist regularity accounts of laws? Because they find it implausible that enough duplication in matters of particular fact alone between worlds ensures duplication with respect to laws. As they see things, duplication in matters of particular fact can always be, to one extent or another, a matter of chance and so consistent with variations in lawfulness. So, as we say above, some patterns are not sui generis. Maybe some are. Perhaps patterns in the intrinsic nature of the fundamental particles are examples of the sui generis, and some dualists hold the view that consciousness is a sui generis pattern in nature. What is important for us is the point that many patterns are not sui generis; indeed, I think we should regard this as obvious even if we were unsure of how to cash out what it is to fail to be sui generis. You could object to the aggregation style account I sketched above of what’s involved in being a structured pattern while granting that, in some good sense, many of the patterns in nature are structured—my mistake lay in the account of what it is to be a structured pattern, not in 177

my insistence that there are structured patterns. Once this is granted, we can give the representationalist account of conceptual analysis. 3. Conceptual analysis as a way of articulating structure The word ‘knowledge’ marks out a pattern in nature. If it didn’t, we couldn’t learn to apply it to new cases after sufficient exposure to examples, and we can. Moreover, it is plausible that there is something illuminating to be said about what’s in common to cases of knowledge over and above their being cases of knowledge. When we decide that something is a case of knowledge, we don’t deploy a ‘knowledge meter’; instead we aggregate information about belief, the facts, justification, causation, etc. To offer an analysis of knowledge is to say what’s to be said of an illuminating kind about the commonality. To analyse knowledge in terms of true justified belief is to say that the pattern picked out by ‘knows that p’ is what you get if you aggregate belief that p with the truth of p plus the belief ’s being justified; it is to say that if God makes a world where Fred believes that snow is white, where snow is white, and where Fred’s belief is justified, then there is no more for her to do to make it the case that Fred knows that snow is white. In general, to offer a conceptual analysis of being K is to offer a string of words that i) picks out the same pattern in nature as that picked out by ‘K’, and ii) does so in a way that reveals, to one extent or another, the structure in the pattern. Well not quite. Being philosophers we find it hard to resist the temptation to meddle. Often what philosophers offer is something seen as an improvement on the concept but I’ll largely set to one side the normative aspect of conceptual analysis. What does this tell us about the possibility of successful analyses? It tells us that our ability to make discoveries about the nature of abstract entities called ‘concepts’ is beside the point. Rather, the question turns on our linguistic abilities. In many cases we are confident that there is structure in some pattern or other picked out by some term. There is something illuminating to be said; the question is whether or not we are able to say it, to find the needed words and the right way to put them together. We can call this the extraction problem. In tackling it, we draw on two bits of knowledge concerning how our words relate to the world. First, we often know what our words stand for. If we didn’t, our words would not be much use in providing information about how things are, and they manifestly 178

are useful. Second, we often know what certain combinations of words stand for. If we couldn’t master what combinations of words stand for, combinations of words would not be much use in providing information about how things are, and they manifestly are. Words and combinations of words are often very useful sources of information about how things are (more on this in the final section). We have, that is, in principle the materials we need but, as we’ll see, that does not mean that the extraction problem is always easy to solve. 4. The extraction problem Sometimes the extraction problem is relatively easy to solve; it is relatively easy to find words for the pattern in a way that illuminates structure. Sometimes we acquire a term explicitly as a term for a structured pattern. We are introduced to the term via other terms in a way that gives the structure. Some people acquire the term ‘ellipse’ (and the corresponding concept) by hearing or reading words that run somewhat as follows: an ellipse is a locus of points in a plane such that the sum of the distances to two fixed points is a constant. If they agree to use the word ‘ellipse’ in accord with this definition, then they will be able to offer the following analysis of the term as they use it x is an ellipse if and only if x is a locus of points in a plane such that the sum of the distances to two fixed points is a constant. The same goes for many people for terms like ‘prime number’, ‘least upper bound’, and ‘monotonic increasing’ (and the corresponding concepts). Technical terms and concepts in logic and mathematics are a happy hunting ground for defenders of conceptual analysis. The same is occasionally true in the social sciences. ‘Sibling’ was introduced into anthropology via the following definition: x and y are siblings if and only if they have a parent in common. The hard cases are those where a term is not introduced via an explicit definition. We pick up terms like ‘knowledge’, ‘free society’, ‘pain’ and so on by being exposed to examples, or putative examples, and somehow latching onto the relevant commonality. We recognise the relevant patterns and thereby acquire mastery of the terms. (I presume that there is an evolutionary explanation for the speed and general reliability of this 179

process.) There need not be anything especially perceptual about the process of recognition. Descriptions in words of the workings of various societies may be enough to allow us to learn the partition ‘liberal society’ makes, the partition that has Sweden counting as a liberal society and Iran as an illiberal one. Typically, the learning process won’t be entirely recognitional. A school teacher who wants to explain what a democracy is will typically do more than describe some examples—Sweden, Britain and America, say—and some non-examples—Burma and Cuba, say—and hope the students pick up the relevant commonality unaided. The teacher may point out, for instance, that being able to vote for more than one party in elections and the way that getting more votes can affect the outcome of elections are crucial. What we have in these kinds of cases is the exercise of pattern-recognition abilities aided by prompts concerning the kinds of patterns to be looking for, combined with an ability to use terms in a language to make public the patterns once they have been mastered. In addition, we know that the patterns in question are structured and something about the structure in question. Someone who says that Sweden is a liberal society but has nothing to say about why, or who thinks that the key fact is that Sweden is cold in the Winter, hasn’t mastered the pattern we pick out with the term ‘liberal society’. They don’t know what the word is used for. What is the upshot for the possibility of conceptual analyses in these cases, cases where terms are learnt through grasping a pattern in a way that involves recognition? What can we say about the possibility of finding a string of words alike in representational content which reveals structure? There are a number of sub-cases to distinguish. The dimension of variation lies in the extent to which intuitions about individual cases allow us to find an analysis. It is sometimes asked, why should intuitions be relevant? But ‘intuition’ in this context is simply a reference to what happens when a recognitional ability is exercised. Part of mastering a language is acquiring the ability to recognise which sentences are grammatical and which aren’t. For those who have the ability with respect to English, ‘She are happy’ ‘feels’ wrong; intuition delivers the verdict, as we say it. I’ll start with a relatively easy case. Logic students can be trained to recognise well-formed formulae (wffs) by showing them examples. It is not necessary to give them a definition. In this kind of case they acquire the concept of a wff in the same kind of way that most of us acquire the 180

concept of being grammatical in our native language.1 Once they have this recognitional ability they can find the linguistic representation that captures the relevant commonality if they are smart enough. They try out various candidates until they have one that delivers all and only formulae that they recognise as wffs. There is a certain amount of hit and miss in the process. One reviews candidates waiting for that ‘Eureka’ moment and—somehow or other— one knows that there won’t be any counter-examples. But it need not be like that. Here’s a nice example I borrow from a conversation with Stephen Yablo many years ago (he should not be held responsible for my use of it or anything I say about it). People can be trained to reliably recognise ellipses by being shown enough examples and non-examples and ‘rewarded’ when they classify aright.2 After the training they reliably classify closed figures that satisfy the formula given some paragraphs back as ellipses, and those that may look much the same but don’t satisfy the formula, as not being ellipses. But they don’t know the formula. They’ll know that there is a formula but not what it is. Our logic students, if they are smart enough, can come up with the standard recursive definition by reflecting on the examples they recognise as wffs, and recognise as non-wffs. However, no amount of reflection on shapes they recognise as ellipses and as non-ellipses will deliver the mathematical formula for an ellipse. It seems, accordingly, as if we have a case that resists analysis. You might be tempted to respond by arguing that our subjects aren’t representing x to be elliptical when they say that x is elliptical, when they use ‘ellipse’ to describe it. They are instead representing that x has the shape, whatever it is, that they been trained to tag with the word ‘elliptical’. If this is right, we could argue that our subjects do have to hand an analysis of being elliptical in their sense of ‘elliptical’. For them, x is elliptical if and only if x has the shape that they are disposed to tag with the word ‘elliptical’, and surely this is something they know, for they know what was going on in the training process. However, it is hard to believe a meta-linguistic story about our subjects’ use of ‘elliptical’. The subjects see the objects they tag with the word ‘elliptical’ as having some specific shape intrinsic to the objects, and what specific shape can that be other 1. There’s an ambiguity in talk of wff-ness and grammaticality. It can mean the syntactic structure that unites the wffs and the grammatical sentences, respectively; or it can mean playing the well-formed role in a formal system or a natural language, respectively. We are talking about the first. 2. Or so we may suppose.

181

than being elliptical? Of course, the subjects know that the objects have a shape that they have learnt to recognise and tag with the word ‘elliptical’, but it is the shape as such that they ascribe to the objects. All the same, intuitions about possible cases can reveal to our subjects the analysis of being elliptical in the terms of the formula. Imagine that after they have acquired the ability to recognise being elliptical, our subjects are told about the formula. They aren’t though told that it is the correct one. They are asked to test it out. They can do this by checking which shapes satisfy the formula and which don’t, and then seeing whether or not the shapes they recognise as elliptical are the same as the shapes that satisfy the formula. If the answer is yes, the subjects have the answer. Their knowledge of the analysis will be fallible of course but that’s true of any analysis. Knowledge of correct analyses is fallible as is knowledge of correct additions or mathematical theorems. I emphasise that it is crucial to the subjects’ being able to arrive at the correct analysis by reflection on cases that they know i) that they are recognising a shape property, and ii) that it is a priori and necessary that the mathematical formula determines without remainder a shape property. Satisfying a formula can explain a recognitional capacity without its being what’s recognised. There is a formula covering the out of phase sound effects at our ears that explains our ability to recognise a sound’s location, but what we recognise is the location, not the out of phase effect. But in this case, what we know is that we are recognising a location property, and the out of phase property does not determine a priori the location property. 5. The analysis of knowledge saga I have just given some success stories for conceptual analysis. Once upon a time philosophers thought that knowledge was another success story. They thought that knowledge could be analysed in terms of true justified belief. Gettier (1963), and especially what happened following its publication, was a wake up call (see Shope 1983). I’ll start by discussing what to say if things are as many of us long thought—namely, that there is very substantial agreement among competent English speakers that the kinds of case of true justified belief described in Gettier’s paper are not cases of knowledge. (It is this presumption of very substantial agreement that has recently been challenged by some experimental philosophers.) The challenge for conceptual analysis posed 182

by Gettier’s paper under this supposition is that we’ve had so much trouble coming up with an alternative to the true, justified belief analysis of knowledge. Many smart people have come up with interesting suggestions for the blah blah in ‘S knows that p if and only if blah blah’, and in every case other smart people, or sometimes the very same people in a later article, have described cases where blah blah obtains but which are intuitively not cases of knowledge, or where blah blah doesn’t obtain but which are intuitively cases of knowledge, and in every case the intuitions are arguably firm ones; the cases aren’t ‘don’t care’ or ‘spoils to the victor’ ones. Here are the main options for what to say, or so it seems to me. First, we might say that there is an analysis to be had but we haven’t found it yet. Our situation is like that of the subjects who knew the meaning of ‘elliptical’ before they knew the formula. There is an analysis to be extracted but we have still to extract it. One day someone will give an account of blah blah that fits with all our firm intuitions. Second, we might say that there is an analysis to be had but we’ll never find it. The extraction job is beyond us. A situation of this kind plausibly holds for being a liberal society. Each of us uses ‘is a liberal society’ to capture a pattern in nature, and we know a good deal about which societies satisfy the pattern and which don’t, and why. However, we’ll never find anything of the form ‘x is a liberal society if and only if blah blah’ that’s of interest.3 It is enough that we can recognise a liberal society when we have enough data about it, and that we know the kinds of factors that drive our judgement. However what drives our judgement is too indeterminate and too plastic to be susceptible of extraction. Maybe a god-like being looking down on our classificatory practice could do the extraction job, modulo vagueness, but we can’t. This can create a problem for political theorists and they often get around it by stipulating for the purposes at hand—in this book, in this article—that they’ll mean by a liberal society one that, say, has a two party system with adult suffrage. Economists do the same when they stipulate that ‘recession’ means two quarters of negative growth. But this is stipulation, not arriving at a correct analysis. Third, we might say that we are under a recognitional illusion. We think our intuitions about cases are the recognition of a single complex and maybe disjunctive pattern. We are wrong. There are a number of patterns. What happens is that, as we contemplate one or another of the compelling counter-examples to one or another proffered analysis of knowledge, different potential epistemic virtues become salient. We 3. What we get substituting a synonym for ‘is a liberal society’ for ‘blah blah’ is not of interest.

183

shift from one concept to another. Under the influence of the to-ing and fro-ing between cases typical of debates about the analysis of knowledge, we shift between the various epistemically virtuous properties as the one we use ‘knowledge’ for. Each concept singles out true belief with one or another additional epistemic virtue but the additional epistemic virtue varies with context.4 Maybe (maybe) something like this happened to some of us when we first read Gettier. Our pre-Gettier concept was true, justified belief but reading Gettier made salient to us that there was an epistemic virtue that this concept failed to ensure—roughly, getting things right not by happenstance. We shifted as we read his paper to a different concept, mistakenly thinking we were discovering something about our original concept. Finally, we might go sui generis. We might insist that there is a pattern in nature that cannot be captured in words, except in terms of the word ‘knowledge’ or a synonym. This seems to me the least attractive option. It seems to me to be a metaphysical extravagance. (For more this point, see Jackson 2005). I’d rather go down the track of holding that there are many concepts that shift with context or even that our concept is confused and that we should replace talk of knowledge by talk of one or more of: true justified belief, true justified belief not sourced in a falsehood, true justified belief caused by what’s believed, true justified belief reliably obtained, etc. What should we say if the long-held presumption of very substantial agreement in intuitions is a mistake, as is suggested by some philosophers in the experimental tradition? Here I think we have to bear in mind that the classification we effect with a given term is a contingent, a posteriori matter—the point we emphasised at the very beginning with the treeshrub and dictator examples. We might, in particular, have used the word ‘knowledge’ for justified true belief. This means that our reason for doubting that we so use the word ‘knowledge’ has to be empirical evidence about which cases we use the word for, and Gettier provided evidence that bears directly on this question. He carried out a survey, courtesy of the editor of Analysis, and the evidence that came back strongly supported the view that the readers of Analysis, and those with whom they discussed Gettier’s paper, do not use the word ‘knowledge’ for true, justified belief (though, as we noted in the preceding paragraph, a possibility is that reading his paper, and the same would go for discussing it, made this the case rather 4. I presume here that true belief is common to all plausible attempts to analyse knowledge but some will want to weaken this to a true taking-to-be-the-case in some sense weaker than full-blown belief.

184

than discovering it). But what is sauce for the goose is sauce for the gander. If we take Gettier’s survey seriously, as we should, we must take seriously other surveys that suggest that different groups do or may use the word for true, justified belief.5 Having said this, I emphasise that I am not trying to do philosophers out of a job. Surveys, be they through the pages of a journal, mail outs to addresses selected at random, internet polls, or handouts in lectures, bear on what people use words for, the concepts they in fact have. They don’t bear on: what concepts are good for one or another theoretical purpose (is true, justified belief the gold coin of epistemology, or can we do better?); what concepts are coherent (is the libertarians’ concept of free action coherent?); how various concepts inter-connect (is giving causation a central role in justification consistent with a regularity account of causation?); and what concepts are instantiated (is objective, single case probability a feature of the world?). Experimental philosophers sometimes suggest that philosophers as a group are guilty of chauvinism in the way they privilege the intuitions of their colleagues in discussions of the concept of knowledge but we should allow, however we stand on the question of chauvinism, that philosophers are often, though not always, best placed to address questions like those just listed—that’s part of a philosopher’s job description. 6. Does conceptual analysis presuppose a controversial theory of reference? As noted at the beginning, it is sometimes suggested that conceptual analysis is wedded to the description theory of reference. This is seen as a serious problem. Hasn’t the description theory been refuted, or at the least surely we must allow that there are a range of theories of reference—including causal, description and teleological theories—that deserve serious attention? We should not, at this stage of the debate over reference, be assuming the description theory, and that’s what conceptual analysts are doing.6 5. Part of ‘taking seriously’ is subjecting any survey, including Gettier’s, to the kind of scrutiny familiar to social scientists. A possibility is that when we do this, the long-held belief of widespread agreement will turn out to be correct after all. The reported substantial differences over Gettier cases will turn out to be due to framing effects or misunderstandings or confusions or … But one would expect some degree of divergence to remain. Why should we all use the term ‘knowledge’ in the same way; why should we all, without exception, have the same concept? 6. I think this is one burden of Stephen Laurence and Eric Margolis (2003); see their discussion of conceptual analysis in the context of differing theories of ‘reference determination’.

185

I reply that there are two different things that can be meant by a theory of reference here and that on one, the one that is indeed presupposed by conceptual analysts, the description theory is not controversial—or perhaps I should say, should not be controversial. I will illustrate the two things that can be meant by a theory of reference with the example of the word ‘circle’. One question about the word ‘circle’ is what its content is. I mean this in the sense of what it takes for an object to be correctly characterised using the word. We all know the answer to this question. The object needs to be circular. It needs to have the property instantiated following the colon: O. In this sense the reference of the word ‘circle’ has an easy answer: it is that property, or the objects having that property, or the class of things having that property or … (Different theorists differ in how they like to say it.) The other question is how it comes to be the case that the word has this content or reference. The answer to this question is much more controversial (see Jackson 2004, 265f., Jackson 2001). However in the view of many, myself included, the answer involves certain causal-functional links inside heads and between heads and the world, especially those involving actual or possible circles, that make it the case that those heads have actual or possible thoughts concerning things being circular, plus the conventions of word usage whereby the word ‘circle’ comes to be a good word to capture these actual or possible thoughts. Others would add a teleological element to this kind of story. The details of the many accounts on the market are not important here. What is important is that the appeal of causal-cumfunctional accounts (and, for some, of teleological ones) in no way implies that we should not espouse a descriptive account of the content of the word ‘circle’. Also, it would be a mistake to move from the fact that words get to refer to what they do refer to in virtue of causal or selectional relations, say, to the conclusion that they represent the obtaining of the selectional or causal relations in question. To think that would be like thinking that if what makes it the case that the number of tree rings represents the age of a tree is the causal co-variance in normal circumstances between age and number of rings, then the number of rings represents that co-variance. The number of rings represents the age of the tree. The word ‘circle’ is one example among thousands. It is obvious, and should be in no way controversial, that many words have descriptive content. If that were not the case, we could not use words to describe the world, and we can and do. In that sense, it is non-controversial that a description theory of reference is true for very many terms in our language. 186

And that’s all that the kind of conceptual analysis I am discussing in this essay is presupposing. This essay is about capturing the patterns ascribed by certain descriptive terms in a language using other terms in a language in ways that illuminate the structure of the ascribed pattern. To say otherwise would mean that it would be sensible to respond to someone who offered the familiar recursive definition of a well formed formula in logic, or gave the standard analyses of being an ellipse or taking an average value, that what they were offering involves a risky commitment to a controversial theory of reference. That can’t be right. In discussing the questions raised in our often fraught attempts to do the capturing, to solve the extraction problem as I call it above, I presuppose, in common with most supporters of conceptual analysis, that we often know (fallibly) how we are representing things to be when we use one or another word or sentence. I suspect some will think that I am being unduly optimistic about our grasp of how things are being represented to be by sentences composed of words we understand, using modes of composition we understand.7 Indeed, some seem to take it that one message of recent discussions of reference is pessimism. But I am not being more optimistic than the players in the debates over reference. Their writings are full of detailed, sometimes complex descriptions of possible cases and claims about what refers to what and why in those cases. They take it for granted that they and their readers know the cases that are being presented. If they and their readers didn’t, discussions of the cases would not be worth much. They’d be boxing in the dark. Optimism over many examples is consistent with pessimism over some. Plausibly there is a sense of reference on which English speakers in 1750 did not know the reference of the word ‘water’, because they did not know that water is H2O. Is this a case where we should be pessimists about how much speakers using the word ‘water’ in 1750 knew about the content of the word in the sense of how they were representing things to be when they used the word? It might seem a small concession for a conceptual analyst to grant the case. What’s in a word? But the concession would open a floodgate. Very many words refer to kinds, and for large stretches of time, including the present, it is the case that we do not know the kind referred to by one or another kind term. In any case, I think we can and should resist the suggestion that people did not know what the word ‘water’ meant, in 7. We might define ‘understand’ in terms of grasping what’s represented; in which case my optimism is over the extent to which there is understanding.

187

the sense of its representational content, back in 1750. Someone in 1750 dying of thirst was rightly glad to hear the sentence ‘There is some water just over the hill’. How so if they did not know what was being claimed about how things are, if they did not know the information provided by the sentence? Again, reports of the key experiments that showed that water is H2O presumed that readers understood what sentences like ‘We placed some water in a beaker and proceeded to pass an electric current through it’ told them prior to knowing how the experiment turned out. Otherwise, we could accuse the experimenters of arguing in a circle or of arguing from an unsupported premise. Again, think of the implications of holding that the folk in 1750 didn’t know the meaning of ‘water’. How in that case could they know the meaning of words like ‘drought’ and ‘thirst’? The meaning of these words is tied to that of ‘water’. But surely people who said that the country was in the grip of a severe drought or that they were thirsty knew what they were saying about how things were?8

REFERENCES Gettier, Edmund L. 1963. “Is Justified True Belief Knowledge?” Analysis 23, 121–123. Jackson, Frank. 2004. “Why We Need A-intensions”. Philosophical Studies 118(12), 257–277. — 2001. “Locke-ing onto Content”. In: Denis Walsh (ed.). Naturalism, Evolution and Mind. Cambridge: Cambridge University Press, 127–143. — 2005. “Ramsey Sentences and Avoiding the Sui Generis”. In: Hallvard Lillehammer and D. H. Mellor (eds.). Ramsey’s Legacy. Oxford: Clarendon Press, 123–136. Shope, Robert. 1983. The Analysis of Knowing: A Decade of Research. Princeton (N.J.): Princeton University Press. Laurence, Stephen and Eric Margolis. 2003. “Concepts and Conceptual Analysis”. Philosophy and Phenomenological Research LXVII (2), 253–282.

8. Over the years I’ve discussed the issues surrounding conceptual analysis with more people that I can possibly remember but I must mention David Braddon-Mitchell, Simon Cullen, Michael Smith, and the discussion, especially from the discussant, Hans-Johann Glock, at the workshop in Zurich, 5–6 September 2008, where a version of this essay was presented.

188

Grazer Philosophische Studien 81 (2010), 189–214.

PHILOSOPHICAL THOUGHT EXPERIMENTS AS EXCERCISES IN CONCEPTUAL ANALYSIS Christian NIMTZ University of Erlangen-Nürnberg Summary In this paper, I defend the viability and importance of conceptual analysis to philosophical inquiry. My argument proceeds in two steps. In a first step, I argue that we rely on the notions guiding how we do and would apply our terms in order to evaluate the counterfactual conditionals we find at the heart of philosophical thought experiments. In a second step, I argue that our notions determine what the relevant terms mean in our mouth. In order to defend the resulting neo-descriptivist semantics, I put forth an epistemic argument for descriptivism—the argument from communication. I conclude that philosophical thought experiments are exercises in conceptual analysis.

1. Thought experiments and the fate of armchair philosophy Thought experiments1 are widely acknowledged to be essential to our armchair ways of doing philosophy. It is hardly surprising, therefore, that philosophical thought experiments take centre stage in the current debate about the viability of armchair philosophy. (See, e.g., Jackson, this volume; Williamson 2007; Nimtz 2007 and the papers in DePaul & Ramsey 1998, Gendler & Hawthorne 2002, and Knobe & Nichols 2008.) In this paper, I defend an epistemology of philosophical thought-experimenting. What entitles us to hold true the counterfactuals on which philosophical thought experiments pivot is tacit semantic knowledge we possess, or so I argue. On the account here defended, philosophical thought-experimenting is a priori in the sense in which determining what our words mean by reflection on their application is an a priori exercise. 1. This choice of label is not meant to imply that thought experiments literally are experiments, i.e. “controlled manipulations of events, designed to produce observations suited to confirm or disconfirm one or more rival theories or hypotheses“ (Blackburn 1994, 131).

My argument combines two lines of thought. The first is epistemic and focuses on the evaluation of counterfactuals we find at the heart of philosophical thought experiments. Having argued that philosophical thought experiments are best understood as pivoting on counterfactual conditionals (§§ 2–3), I argue that these counterfactuals—which I call philosophical counterfactuals—are epistemically distinctive. In contrast to mundane or scientific counterfactuals, our entitlement to hold philosophical counterfactuals true does not derive from empirical evidence, or from empirical theory (§§ 4–5). The second line of thought is semantic and focuses on how the conditions tacitly guiding how we do and would apply our terms—I sum up these conditions as notions—shape the semantic properties our terms have in our mouths. Having argued that we, in all plausibility, evaluate philosophical counterfactuals drawing on our notions (§ 6), I propose and defend a neo-descriptivist semantics (§§ 7–8). I explain that the key contention of neo-descriptivism consists in the idea that notions determine semantic properties, and I put forth the argument from communication in order to support neo-descriptivism and reject the rival Kripkean externalism. Combining both lines of thought, I conclude that philosophical thought experiments are exercises in conceptual analysis (§ 9). The account here presented runs counter to the ambitious rationalism espoused by Bealer (2000; 2002) or Katz (1998; 2002), who hold that we need a priori insight beyond mere semantic knowledge for philosophical thought-experimenting to work. It also puts me at variance with the armchair empiricism of Williamson (2007), who maintains that a priori knowledge plays no distinctive role in thought experiments. I won’t have much to say about ambitious rationalism. I will, however, take issue with Williamson’s armchair empiricism (see § 5). 2. Thought experiments and counterfactual thinking Thought experiments are ubiquitous in philosophy. Paradigm examples include Davidson’s Swampman (Davidson 1987), Gettier cases (Gettier 1963; Lehrer 1965), Putnam’s Voyages to Twin Earth (Putnam 1975), and Jackson’s Mary case (Jackson 2004b). These thought experiments share some distinctive features. They all comprise a narrative core—a story inviting us to consider a possible situation where something happens, typically to some subject. In telltale fashion, Lehrer’s Gettier-style case begins thus: ‘Imagine the following. I see two men enter my office whom I know to 190

be Mr. Nogot and Mr. Havit (…)’ (Lehrer 1965, 169). On this basis, we are expected to arrive at a specific upshot best read as being modal. In Lehrer’s case, we are expected to conclude that our subject is in a possible situation where he has a justified true belief that someone in his office owns a Ford, but does not know this. All the thought experiments listed are truly epistemic in that they aim to warrant (rather than just instil) such an upshot. They are not mere ‘intuition pumps’ in Dennett’s (1995) sense, i.e. quasi-mechanical devices designed to inculcate convictions. Thinkers devising philosophical thought experiments employ their upshots to draw general conclusions—say, that the JTB analysis of knowledge is false, or that physicalism is flawed. Contrary to Häggqvist (2009), these conclusions are best considered extraneous to the thought experiment devised to establish them. Whereas a thought experiment fails if it does not establish its intended upshot, it can be considered sound even if the philosophical conclusion does not follow. Philosophers typically agree that Jackson’s Mary case establishes that a physically omniscient person inhabiting a world like ours could still learn something. Still, many of them deny that physicalism fails (see the contributions to Ludlow, Stoljar & Nagasawa 2004). This is easily explained: Jackson’s thought experiment is only part of his more encompassing knowledge argument, and it is the part that apparently works. Our survey of paradigmatic philosophical thought experiments encourages the following general picture: In philosophical thought experiments, we are expected to come to the conclusion that a certain modal claim is true on the basis of considering a possible situation described in the experiment’s narrative core. Going by this finding and taking our cue from Williamson (2007, ch. 6; 2004), we may understand thought experiments to (paradigmatically) consist of three steps. In a first step, it is contended that a certain situation C, described by the experiment’s narrative core and typically comprising a subject or other, is possible. Let me represent this possibility premise thus: (1) ¡(s: s is in C )2 In Jackson’s case, it is claimed that there could be some physically omniscient person by the name of ‘Mary’ who has never perceived something 2. Williamson assumes that the modality in question is metaphysical. I do not want to commit to that. Please read ‘¡p’ and ‘p o q’ throughout as ‘It is possible, that p’ and ‘If p had been the case, then q would have been the case’, respectively, with an unspecific modality.

191

non black-or-white. On the basis of this possibility premise, we are expected to hold that our protagonist has some property F (that has not already been explicitly mentioned in characterizing C ). Contemplating Jackson’s Mary, we conclude that she would learn something on seeing something red for the first time. Purging the anaphoric element for ease of exposition,3 we can represent this modal upshot thus: (2) ¡(s: s is F ) How do we get from a thought experiment’s possibility premise to its modal upshot? The required link needs to be modal, since it is employed to establish a possibility on the basis of another possibility. A natural as well as popular4 idea for the modal link licensing us to hold (2) on the basis of (1) is the counterfactual statement that if some subject s had been in situation C, some subject would have property F: (3) s: s is in C o s: s is F Read along these lines, Jackson’s Mary-case and Lehrer’s Nogot/Havit scenario are understood to pivot on the following counterfactuals: (4) Some subject s is in a Mary type situation o Some subject s is physically omniscient, yet learns something on seeing something red for the first time. (5) Some subject s is in a Nogot/Havit situation o Some subject s has justified true belief that someone in her office owns a Ford, but no knowledge. In order to have a convenient label, I call such counterfactuals we find at the heart of philosophical thought experiments philosophical counterfactuals. The analysis offered takes philosophical thought experiments to be exercises in counterfactual thinking. In a philosophical thought experiment, we reason from a possibility premise to a modal upshot on the strength 3. If you want our representation to capture the intersentential anaphoric reference in “Imagine Mary to be in situation C. She would learn something”, you could employ Berger’s (2002) technique of indexed variables and represent it thus: ¡(s1: s1 is in C). ¡(s1 is F ). 4. See, e.g., Cooper 2005, 329; Norton 2004 and Quine 1995, 98.

192

of a modal link provided by a philosophical counterfactual.5 A survey of possible alternatives lends support to this account. On the one hand, one could point out that we apparently arrive at the modal upshot on the basis of the possibility premise without any discernible mediating step, and conclude that we should drop the modal link (3) from our analysis.6 However, to do so would mean to misunderstand the task an analysis is set to accomplish. It does not aim to retrace the sequence of psychological steps we find ourselves making in thought-experimenting. It rather aims to uncover the (apparent) justificatory structure in thought experiments. We clearly feel entitled to an experiment’s modal upshot because we judge it to follow from the possibility premise in a specific way. And a convincing way to capture this is to identify the modal link involved with a counterfactual. On the other hand, one could point out that the strict implication (6) (Some subject s is in C o some subject s is F) would also allow us to infer the modal upshot from the possibility premise. But (6) arguably is too strong (see Williamson 2007, 185). The narrative core won’t be specific enough to disallow all C-worlds where bizarre goingson prevent our subject to become F. Consider a world where, by a freakish law of nature, anyone seeing a colour evaporates milliseconds later. Mary would be hard pressed to learn something on seeing something red in such a world. Advocates of Jackson’s case are unmoved by such scenarios, and rightly so. Everyone agrees that in evaluating thought experiments, we may properly ignore such remote possibilities. Hence, the modal link cannot be a strict necessity such as (6). A similar argument threatens to undermine the idea that the modal link is counterfactual.7 I call it the challenge from unintended instances. Suppose that John actually satisfies the description of the subject as provided by the narrative core of the Nogot/Havit case, but rightly distrusts his beliefs about who owns what car. (Say, he knows that he is frightfully forgetful about these things.) Clearly, John does not have the justified true belief that someone in his office owns a Ford. Since the situation described—call it the ‘John-plot’—is by stipulation actual, the counterfactual (5) is false. 5. I am prepared to argue that the thought experiments in the sciences can be analysed along analogous lines. 6. I take Malmgren to take this position in her widely circulated, though still unpublished paper Staying in the Armchair. 7. This objection has been made popular by Malmgren. See previous footnote.

193

For although the John-plot satisfies the counterfactual’s antecedent, it fails to satisfy its consequent and, given centring, A & ™B entails that A o B is false. Hence, if the Nogot/Havit case pivots on a counterfactual, learning that the John-plot is actual should make us forego the thought experiment’s modal upshot. But it doesn’t. In fact, we clearly feel that John’s predicament, close to home though it is, is as irrelevant to the thought experiment as some fancy going-on in a remote world. Do we therefore have to conclude that the modal link we rely on isn’t a counterfactual after all? We do not. A thought experiment’s narrative core will inevitably underspecify the intended situation. So whenever we are confronted with a situation satisfying the letter of a narrative core, we still need to determine whether it is a relevant instance, or whether it may be properly ignored. The John-plot arguably falls into the latter category. As a Gettier case, Lehrer’s Nogot/Havit thought experiment is exclusively concerned with scenarios where a subject will arrive at a justified belief, given that this epistemic principle as espoused by Gettier holds true (and given that one may justifiably believe falsehoods): [F]or any proposition P, if S is justified in believing P, and P entails Q, and S deduces Q from P and accepts Q as a result of this deduction, then S is justified in believing Q. (Gettier 1963, 121)

The John-plot does not belong to these scenarios. John’s belief will therefore be unjustified even if Gettier’s principle holds true. The John-plot thus is an irrelevant instance of the Nogot/Havit case’s narrative core. Hence, (5) may well be true even if the John-plot happens to be actual. There is a more general way to accommodate the challenge from unintended instances. We have seen that we do not rely on strict implications to arrive at modal upshots from possibility premises. We rather draw on some restricted necessary (7) X(Some subject s is in C o some subject s is F) with ‘X’ marking the restriction to worlds that fit with how we intend the description of C provided by the narrative core to be read, a restriction varying with the thought experiment in question. This fits well with the idea that what provides the link is a counterfactual; for on the standard Lewisian analysis, counterfactuals just are variable strict conditionals (see Lewis 1973). But we can take (7) as a proposal in its own right. Since this proposal nicely avoids the challenge from unintended instances, we might 194

want to hold that our modal link is strictly speaking a restricted necessity, rather than a bona fide counterfactual. I do not mind. For none of this affects the arguments to come. Although I phrase them in the more familiar terms of ‘counterfactuals’, they equally apply to restricted necessities. 3. Counterfactuals and their epistemology The general recipe for evaluating a counterfactual seems simple enough. As Williamson puts it, ‘one supposes the antecedent and develops the supposition (…)’ (2007, 153; see also Edgington 2008) in order to see whether one arrives at the consequent. However, how we judge counterfactuals to be true is one thing. What entitles us to make such judgments is quite another. I venture that we can ascertain the latter by asking how we justify such judgments. More precisely, we can ascertain what entitles us to hold true a counterfactual A o C if we ask what licenses our move from its antecedent to its consequent, a move that, by standard counterfactual theory (see Lewis 1973; Stalnaker 1968; Bennett 2003), amounts to the claim that any relevant possible A-situation is indeed a C-situation. Here we find marked differences between mundane and scientific counterfactuals on the one hand, and philosophical counterfactuals on the other. Counterfactual thinking is ubiquitous in our everyday dealings with the world. We rely on mundane counterfactuals such as (8) If John had voiced his doubts at the board meeting, he would have been fired to assess choices, evaluate strategies, and learn from mistakes. What entitles us to move from the antecedent of a mundane counterfactual such as (8) to its consequent is well-confirmed empirical belief. We clearly take it for granted that these counterfactuals are subject to, as well as in need of, empirical confirmation.8 And we patently expect anyone issuing a verdict on (8) to back her claim by empirical evidence establishing that any relevant possible situation where John raises his doubts is one where he gets fired. Someone asserting (8) might for instance point out that John’s co-worker Bob did voice his doubts, and did get fired. 8. The same holds good for empirical disconfirmation. For simplicity’s sake, I exclusively mention confirmation throughout.

195

As a rule, the empirical evidence someone presents won’t by itself allow us to pass verdict on her counterfactual claim. We typically need to invoke theoretical elements. For mundane counterfactuals, we habitually need to rely on our background theory of folk truths and folk laws, the latter comprising causal generalizations, social rules, economic presumptions, psychological principles, and the like. In short, what licences our move from a mundane counterfactual’s antecedent to its consequent typically is empirical evidence together with empirical theory. Basically the same holds true for counterfactuals in the natural sciences. Consider this (rather uninspired) scientific counterfactual: (9) If someone formed a bare sphere of 7 kg Californium251, she would initiate a nuclear chain reaction. We manifestly take it for granted that such counterfactuals are subject to, and in need of, empirical confirmation. As before, we expect someone passing a judgment on (9) to provide empirical evidence in support of her verdict. In order to support our counterfactual, pointing out that Californium251 is a fissile material with a bare-sphere critical mass of 7 kg or less would do fine. For given these empirical facts, one needs but a rough grasp of basic nuclear physics to see that (9) is indeed true. Our evaluation of the scientific counterfactual (9) thus is informed by general empirical evidence together with an empirical background theory. This time, however, both fall within the purview of science. Our evaluation of counterfactuals such as (8) and (9) is critically informed by empirical evidence and by empirical theory, or so I have maintained. We take it for granted that these counterfactuals are subject to empirical (dis-)confirmation, we expect anyone issuing a verdict on them to provide empirical evidence, and we draw on empirical fact and/or empirical theory to move from their antecedents to their consequents. What in these cases licenses us to hold that any relevant possible A-situation is a C-situation is some body of beliefs b, where our entitlement to hold b is grounded in experience. This provides sufficient reason to hold that the counterfactuals themselves are to be classed as a posteriori, too.9 That does not hold true for all counterfactuals. Some counterfactuals are firmly to be classed as a priori, since the body of beliefs we draw on 9. I take it that S knows a priori that p iff S knows that p and S’s entitlement to believe that p is not grounded in experience; see Nimtz, Kompa and Suhm 2009. By this standard, a proposition p is a priori iff it can be known a priori.

196

to move from their antecedents to their consequents is manifestly not grounded in experience. Just consider: (10) If twelve guests had come to dinner, more than ten guests would have come to diner. We might of course ask someone asserting (10) to justify her judgment. But we do neither expect nor require empirical evidence. We will be perfectly happy with her pointing out that there being twelve Fs entails there being eleven, thus accepting a logico-mathematical truth as licensing her move from (10)’s antecedent to its consequent. Similarly, consider someone defending her judgment that (11) If Paul had been a bachelor (as he claimed to be), he would have been unmarried. This again requires no empirical evidence, since we straightaway see that the connections between ‘bachelor’ and ‘unmarried’, a connection we may for all that it’s worth dub ‘conceptual’, licenses her move from (11)’s antecedent to its consequent. In fact, if she were to defend (11) by stressing that all the bachelors she has heard of are unmarried, we would conclude that she had missed the point of her of own claim. The distinction between a priori and a posteriori counterfactuals does not coincide with Bennett’s (2003) distinction between dependent and independent counterfactuals who explains independent conditionals to be ‘the ones where the route from A to C owes nothing to any particular contingent fact or belief ’ (Bennett 2006, 174, see also 148f, 174f ). In other words, independent counterfactuals are those where we can move from A to C on the strength of the respective theory alone. Since this theory might be empirical, not all independent counterfactuals are a priori. But all a priori counterfactuals are independent. What renders (10) and (11) a priori is the combination of two facts: first, their truth (and hence their evaluation) is independent of particular contingent fact, since it exclusively relies on a background body of beliefs, and, secondly, our entitlement to hold that body is a priori. There is no doubt that philosophers habitually invoke a posteriori counterfactuals such as (8). But that is of no concern to us. What we need to ascertain is how philosophical counterfactuals such as (4) and (5) are to be classed. I hold that we have ample reasons to consider those a priori. 197

First of all, we do not treat these statements as in need of, or as subject to, empirical confirmation. In fact, it is hard to see what kind of empirical evidence could possibly settle the issue. Williamson (2007, 192f ) rightly points out that we can create actual Gettier-type situations. But those won’t settle the issue. Since we don’t have a device telling us whether the epistemic state our subject is in is indeed a state of knowing, we have to rely on the very same resources we rely on in passing verdict in hypothetical cases. Secondly, counterfactuals such as (4) and (5) are independent. Particular contingent facts do not bear on their truth; what licenses our move from their antecedents to their consequents are our background accounts. But by standard taxonomy, these accounts are non-empirical. Just as we defend the likes of (10) by stressing logico-mathematical entailment, we defend the likes of (5) by pointing out that this is what we take knowledge to be. Thirdly, please note that philosophical counterfactuals often are counterlegals in that they concern nomologically impossible situations (Goodman 1983, 7f ). (4) is a case in point. But we can hardly draw on empirical theory and empirical laws to license the move from the antecedents to the consequents of counterfactuals of that variety (Bennett 2003, 227f ). I conclude that the philosophical counterfactuals are often best classed as a priori. Let me be clear about my argument. I have diagnosed an epistemic difference between mundane and scientific counterfactuals on the one hand, and logico-mathematical, conceptual and philosophical counterfactuals on the other; a difference I have characterized drawing on the a priori/ a posteriori distinction. Those with little faith in this distinction will mind me putting the diagnosed difference in its terms. But they still have to acknowledge that there is a marked epistemic difference between these counterfactuals. In the end, this is all I need. 4. Against Williamson’s no-consequence assessment In arguing that philosophical counterfactuals are to be classed as a priori, I have presumed that such a classification would be of consequence. Williamson takes this to be a mistake (see Williamson 2004; 2007b; 2007, ch. 5 and 6). He maintains that ‘the question ‘A priori or A posteriori?’ is too crude to be of much epistemological use’ (2007, 169). To back this claim, Williamson argues that once we have settled that we can reliably evaluate 198

a posteriori counterfactuals, we can rest assured that the same holds true for counterfactuals best considered to be a priori: We have a general cognitive ability to handle counterfactual conditionals. (…) But we have no good reason to expect that the evaluation of ‘philosophical‘ counterfactuals such as [10] uses radically different cognitive capacities from the evaluation of ‘unphilosophical’ counterfactuals. (…) [W]e should not suppose [10] to involve fundamentally different cognitive capacities from evaluations of counterfactuals that we classify as a posteriori. Consequently, we should not suppose it to raise fundamentally new questions of reliability. (Williamson 2004, 13f., my italics)

Williamson argues in two steps. In a first step, he maintains that the very same cognitive capacities underlie our evaluation of all kinds of counterfactuals. He bolsters this claim by a general account of the cognitive procedures leading us to assert counterfactuals alluded to above. According to Williamson, we suppose the antecedent and then develop the supposition ‘adding further judgments within the supposition by reasoning, off-line predictive mechanisms and other off-line judgments’ (Williamson 2007, 152f.), and we assert the counterfactual if this development ‘eventually leads one to add the consequent’ (ibid., 153). This process will often involve what Williamson terms ‘imaginative simulation’ (ibid., 152).10 In a second step, Williamson draws an epistemic moral from this cognitive sketch: since one and the same cognitive procedure underlies evaluations of mundane and philosophical counterfactuals alike, the question of whether we can reliably evaluate counterfactuals of the latter kind is answered by our manifest ability to reliably evaluate counterfactuals of the former variety. I hold that Williamson’s inference from a cognitive premise to an epistemic conclusion fails. To begin with, there is a reading of Williamson’s cognitive premise on which it is false. On a narrow individuation of cognitive capacities, evaluating the mundane counterfactual (8) and the logico-mathematical counterfactual (10) involves different cognitive capacities. These statements both are of the form A o C. But they comprise different contents and I need to draw on different cognitive resources to reason from their respective antecedent and to their respective consequent. I need to employ factual knowledge in combination with folk-theoretical 10. Williamson initially assigns the imagination a key role in the process described, see his 2004; 2007b. He later downplays that role, see his 2007, 152f. That sits well with my assessment of how we evaluate counterfactuals which does not assign imagination any special role.

199

projection in the one case, whereas I need to rely on logico-mathematical entailment in the other. Hence, in order for Williamson’s cognitive premise to be true, we need to read it as concerning our cognitive capacities broadly individuated. But read thus, Williamson’s cognitive premise does not license his epistemic conclusion. From the fact that our evaluation of (8) and (10) relies on the same cognitive mechanisms broadly individuated, it does not follow that we are as reliable in evaluating (10) as we are in evaluating (8), or that we rely on the same sort of evidence. These statements comprise different contents. But we have every reason to believe that how reliable we are at evaluating a counterfactual conditional does critically depend on how good we are at counterfactually relating the contents they comprise; the same holds good for what evidence we rely on. Sameness of cognitive capacity broadly construed thus does not ensure sameness—or even likeness—of epistemic status. In fact, I have taken some pains in the last section to argue that we find marked differences in what it takes to evaluate counterfactuals. These epistemic differences are, of course, compatible with the idea that the same cognitive capacity broadly individuated underlies all such evaluations. I therefore conclude that Williamson’s argument fails. To support this conclusion, consider an analogy. On a broad individuation of cognitive capacities, we may agree that basically the same cognitive capacity underlies our evaluations of all statements of the form A  B. On this understanding of sameness of cognitive capacity, evaluating ‘Venus revolves around the sun & Pluto revolves around the sun’ does not involve cognitive capacities radically different from those we bring to bear in evaluating ‘7 is an odd number & 11 is an odd number’. Still, it does not follow that our reliability in evaluating conjunctions concerning astronomic facts guarantees that we are likewise reliable to evaluate conjunctions concerning mathematical facts. The analogous holds true for (8) and (10). 5. Notions and how we evaluate philosophical counterfactuals My case has hitherto been negative. I have contended that neither our empirical folk theory of the world, nor our empirically established scientific theory licences us to hold true philosophical counterfactuals such as (4) and (5). In the remainder of this paper, I present a positive account of what philosophical thought experiments amount to. Introducing the 200

notion of a ‘notion’ to sum up the conditions guiding how we do and would apply our terms, I maintain that our tacit knowledge of notions is an excellent candidate for what we draw on in evaluating philosophical counterfactuals (§ 6). I go on to defend a neo-descriptivist semantics that assigns notions centre stage in determining the semantic properties our terms have in our mouth (§§ 7–8). I conclude that our tacit knowledge of notions is semantic knowledge. As it turns out, then, philosophical thought experiments simply are exercises in conceptual analysis (§ 9). How we apply our terms is not arbitrary. Whether or not we, on reflection, class something as a planet or a sofa depends on the properties we take the object(s) in question to have. The same holds true of the way we would employ our terms. There should be no doubt that our considered application—I will exclusively be concerned with considered application and will henceforth drop this qualification—of ‘sofa’ or ‘planet’ across possible situations depends on the properties we take the relevant object to have. But if the application of terms by a speaker across possible situations is sensitive to the properties entities have, then she has to associate conditions with her terms determining how she applies them. Let me call the conditions a speaker S associates with an expression M that determine how S would apply M the notion S associates with M.11 For example, the notion I happen to associate with ‘grandmother’ is being a female parent of a parent, since it is the female parents of parents that I, on reflection, apply my term ‘grandmother’ to across all possible situations.12 I take the idea that we associate notions with our terms governing how we do and would apply them to be non-contentious. To begin with, everyone agrees that there must be some cognitive structure on the speaker’s side determining how she does and would apply her terms. Given that application is sensitive to the properties we take the objects to have, taking this cognitive structure as encoding a condition objects may meet seems straightforward to me. Secondly, note that my claim is deliberately neutral on the specifics of the cognitive structures making it the case that I associate, say, being a female parent of a parent with the term ‘grandmother’. 11. Jackson, this volume, talks of ‘patterns in nature’ rather than of ‘conditions’. 12. Let me emphasize that although notions guide how we apply our terms, we do not have to think of those as meta-linguistic. Put into words, my M-notion will consist of material-mode sentences that, taken together, specify the conditions that someone has to meet whom I count as a M. Thus “Grandmothers are female parents of parent” exhaustively captures my ‘grandmother’notion, whereas “Banana are fruit” partially characterize my ‘banana’-notion. This is in line with how we generally explain what our words mean, or apply to. Asked what ‘grandmother’ means, we simply say “Grandmothers are female parents of parents”.

201

In fact, it is intended to be compatible with all ideas on offer. There is a reason for this. As might be clear from my characterization, notions are contents (i.e. what does the representing). This is why I can leave issues concerning representational vehicles (i.e. what bears these contents) to one side. Finally, the claim that there are notions is not a semantic claim about aspects of what our terms mean. It is a cognitive claim about how we apply them. In order not to beg the question against the Kripkean externalists, who maintains that how we do or would apply, say, ‘heat’ or ‘gold’ does not affect what these terms mean in our mouths, we need (at least for the time being) to keep the semantic and cognitive dimension neatly apart. For many of our terms, we are hard pressed to specify the notions we associate with them without further ado. So it appears that we typically do not have explicit knowledge of which notions we associate with our terms. Yet the notion I associate with a term M must be manifest in how I do and would apply it, and this is something I can try to work out by considering possible situations and reflecting on how I apply M across those. By this method of possible cases, we aim to make our respective notion explicit. That is to say, we aim to determine the condition N such that, for any possible situations C (where M is applicable at all), we apply M to something if it is N in C.13 There is little doubt that we can reliably attain some clarity about our notions in this manner. We rely on the method of possible cases in our everyday diagnoses of differences in use, and hence in notion. Barring incoherent use, we do not question someone’s judgment that she would not count, say, a Basque-style beret as a hat. We agree that a speaker may be mistaken about what a term means, and hence applies to. But we by default grant her authoritative insight into how she would apply it.14 As I have explained it, the method of possible cases is an indirect procedure. I aim to spell out my M-notion by determining how I would apply M, rather than by retrieving it from some mental manual. Any proposal as to what my M-notion is will therefore typically be projected from a partial survey of possible cases, and a case not yet considered might force a correc13. There is no presumption that N breaks down into a neat conjunction of properties. To the contrary. Our notions will often be complex or even disjunctive, they may comprise recognitional conditions (“Ms look thus”), be indexed to specific items or events (“Ms are the animals I have seen this one day in Berlin Zoo”), or comprise rigidifying elements making the application of the terms they govern in possible situations dependent upon their actual application. 14. As a case in point, note that champions of experimental philosophy (see Knobe and Nichols 2008) do not doubt that their interviewees can reliably tell how they apply their terms.

202

tion. This familiar phenomenon explains why proposals are so much easier ruled out than confirmed (see Shope 1983). It might also explain why we are confident of having attained something approaching comprehensive clarity with respect to the notions of just a select few terms. As already said, I take it to be uncontentious that there are notions. What is controversial is whether our notions play a role in semantics. But let us leave this issue until the next section. Semantic or not, notions seem well-suited to guide our evaluation of philosophical counterfactuals such as (5), and we in all plausibility draw on our notions to move from the antecedent of a philosophical counterfactual to its consequent. First of all, suppose you find yourself, on reflection, withholding application of ‘knowledge’ to the epistemic state of our subject in Lehrer’s case. Suppose further that your most natural justification for this is quite simple: you point out that the subject does not count as a knower as you employ the term. This assessment will by default lead you to accept (5). Secondly, given the interplay of getting clear about our notions and the method of possible cases, evaluating the likes of (5) and getting clear about the notion you associate with ‘knowledge’ amounts to very much the same thing. More precisely, making up your mind on whether to count the protagonist as a knower just is part of determining which condition N you associate with ‘knowledge’ such that, for any possible situations C (where ‘knowledge’ is applicable at all), you apply ‘knowledge’ to something if it is N in C. Finally, your implicit knowledge of your notion can be classed as (weakly) a priori. You do not need to do empirical research to determine how you apply ‘knowledge’ across possible situation. Reflection works just fine. However, guiding our evaluations is one thing, warranting them is quite another. The latter requires that our application tracks our terms’ semantic properties such that our application of M reliably indicates what the term M in fact applies to. Suppose your considered use of ‘hat’ is such that you accept ‘Basque-style berets aren’t hats’. This judgment is unlikely to be true unless how you apply ‘hat’ tracks what ‘hat’ in fact applies to. Consider an analogy. If you employ stick M as a yardstick for the length of one meter, M is likely to guide your evaluation of ‘Saint Peter’s Square is 240 meters wide’. But your judgment won’t be warranted unless M happens to be (roughly) one meter in length (and you have grounds to think so). Your stick-based judgments need to track the length in question. Likewise, the considered application of your terms needs to track what they actually apply to. If your considered application is not a reliable guide to your 203

terms’ semantic properties, going by the former to pass judgment on the latter is not a sensible idea. 6. Neo-descriptivism and the argument from communication Embracing a semantics I label neo-descriptivism renders it easy to argue that considered application tracks semantic properties.15 Neo-descriptivism’s key contention is that notions determine semantic properties. Put a bit more circumspect, the idea at the heart of this semantics is this: ND The notion we associate with a term M, guiding the (considered) application of M across possible situations, determines the semantic properties M has in our mouths.16 The idea is that the term ‘hat’ in our mouths applies to the kind of headgear it does apply to because these items best satisfy the notion guiding our considered use of ‘hat’. This of course guarantees tracking. So if neodescriptivism happens to be the right semantics, our evaluation of philosophical counterfactuals drawing on our considered application warrants us to move from their antecedents to their consequents. In much the same vein, our employment of stick M will warrant our meter-judgments if (we have grounds to hold that) M happens to be the standard meter. Neo-descriptivism explains why our expressions mean what they do by embracing the descriptivist paradigm according to which the semantic properties of our expressions are in essence fixed by our cognitive states. In one respect, this is an uncontroversial stance to take. Barring the details, everyone agrees that many of our terms are notion-governed. Nobody thinks that what ‘shareholder’ applies to, or what ‘here’ refers to, is determined by factors beyond the conditions guiding their application. In another respect, neo-descriptivism is highly controversial. Kripke’s (1980) anti-descriptivist arguments have convinced many semanticists that the semantic properties of (at least) proper names and natural kind 15. I take it that David Chalmers and Frank Jackson are committed to neo-descriptivism. See Jackson, this volume, as well as Chalmers & Jackson 2001, § 3, Jackson 1998, 46–52; 2000; 2004; 2005; Chalmers 1996, 56–71; 2002; 2004. See also Nimtz 2004. 16. ND merely states the core idea. A fully worked-out neo-descriptivism needs to take the difference between speaker’s meaning and conventional meaning into account; see below the second objection in § 7.

204

terms are externally determined. This is key idea of Kripkean externalism. Kripkean externalists hold that semantic properties are fixed by factors such as causal-historical chains leading back to initial baptisms—factors that are beyond the associated notions, and hence beyond what speakers may uncover by mere reflection. I won’t here take issue with Kripke’s arguments.17 Focussing on natural kind terms, and drawing on ideas to be found in Jackson (2007b; 2005; 2004, 266–270; 1998, 40, Fn. 16), I will rather turn the table on the Kripkean externalist. I will argue that it founders on an epistemic problem: it cannot account for knowledge competent speakers need to possess, and manifestly do possess, in order to participate in our communicatory practice. Much of what we do aims at putting across information. We embellish our cheeks with small Swiss crosses to get across that we support the Swiss team, we put the name ‘Christian Nimtz’ on an office door to let people know that this is my office, and we put little red dots on boxes to convey that the article contained is red. But what is it to convey information—to get across an informational content? Here is a sensible answer: p1 To convey content c to an audience is to put the audience in a position to exclude possibilities in accordance with c. This fits well the general idea that to bear information is to exclude possibilities (see Stalnaker 1984, ch. 1; Jackson 2000, 331). It also fits with our examples. The Swiss emblem on the cheek allows you to rule out all other possible partialities, and the red dot allows you to rule out other possible colours for the contained article. There of course is no guarantee that things indeed are as the content conveyed presents them to be, and you have to be able to read the signs, as it were, for the content to get across. If you don’t realize that national emblems are worn to indicate whom you support, you might well end up sitting with the wrong crowd. Our most important means to convey information is language—we tell each other how things are (or rather: how we take them to be). Language is well-suited to that purpose. Asserting ‘Smoking kills’ allows me to convey that smoking kills. It does so by and large irrespective of context. That is to say, I can convey such a content by asserting such a sentence to just 17. See Spicer, this volume, for a statement and defence of Kripkean externalism as well as a summary of Kripke’s famed arguments.

205

about every competent speaker of my language, and doing so requires no background knowledge on the part of the audience beyond what realizing the pragmatic circumstances demands.18 That is easily explained, for what we do is we simply exploit the sentences’ truth-conditions and convey the semantic contents they bear.19 This is always an option: p2 We can employ our declarative sentences to convey their semantic contents to competent speakers mostly irrespective of context. You can convey to a perfect stranger that your wife just gave birth to twins asserting ‘My wife just gave birth to twins’, thereby putting her in a position to exclude possibilities in accordance with the sentence’s content. That is, you can do so given that she knows the sentence’s semantic content. Asserting ‘Vaimoni synnytti juuri kaksoset’ doesn’t reliably bring the very same information across around here since most people don’t speak Finnish and thus do not associate this sentence with that content. But if you happen to speak to a competent speaker of English, asserting an English sentence is a highly reliable means to convey its content. Sentences containing natural kind terms—or names, for that matter— are no exception to that rule. Asserting ‘Iran does not possess any weapons-grade plutonium’ puts competent speakers in a position to exclude a rather worrisome possibility, and asserting ‘There’s water on Mars!’ allows the audience to exclude a waterless Mars from the way they believe things to be. (As before, asserting the Finnish sentence ‘Marsissa on vettä!’ won’t work around here.) There should be no doubt that we can and do use such sentences to convey their semantic contents to competent speakers irrespective of context. In fact, part of the point of having those sentences in our repertoire is that they can be used without contextual support to get those very contents across. Conveying contents by way of assertions requires knowledge on the part of the audience. In fact, it holds true that:

18. You need to know enough about pragmatic circumstances to get clear about what Perry and Korta 2006 summarize as ‘near-side pragmatics’, i.e. ambiguity, context dependence and indexicality. 19. Those are equivalent. Suppose that sentence S is true iff smoking kills. Then the sentence’s content is that smoking kills. Accepting an assertion of S puts you in a position to exclude possibilities in accordance with that content—you may rule out that our world is one in which smoking doesn’t kill.

206

p3 We could not employ our sentences as p2 says we do unless competent speakers knew which semantic contents our sentences bear (and knew this in a non-trivial way). We all know who shot Kennedy: whoever shot Kennedy did. By the same token, we all know how someone asserting ‘Sataa’ presents things as being: she presents things as being such that ‘Sataa’ is true. Knowledge of this variety is trivial in that it does not put the knower in a position to actually exclude possibilities. Conveying contents by asserting sentences requires non-trivial knowledge of the sentences’ semantic contents on the part of the audience.20 Unless I know that ‘Sataa’ bears the semantic content that it’s raining, you cannot inform me about how things are (according to you) by asserting that sentence. Our very praxis of conveying information by assertive utterances thus requires competent speakers to have non-trivial knowledge of semantic contents. This holds true for sentences that do not contain natural kind terms or proper names, as well as for those that do. Consequently, no semantics can be an adequate semantics for our language unless it accounts for this knowledge. This provides a good reason to embrace neo-descriptivism: con The idea that the semantic properties of natural kind terms are notion-governed accounts for non-trivial knowledge of semantic contents. The idea that the semantic properties of natural kind terms are externally determined does not. If the semantic properties of ‘water’, ‘plutonium’, etc. are externally determined, even competent speakers might well be ignorant of the semantic contents of our sentences comprising these terms. In fact, since on the orthodox causal-historic account initial baptism determines subsequent meaning, this semantics allows that even all our experts might share the laymen’s ignorance. If, however, these properties are determined by the notions we associate with ‘water’, ‘plutonium’, etc., the required knowledge is guaranteed. In other words, the neo-descriptivist picture accounts for the non-trivial knowledge on the part of competent speakers requisite for our communicative praxis, whereas the externalist picture doesn’t. I conclude 20. I feel that in his critique of my argument, Spicer, this volume, does not sufficiently appreciate that the knowledge conveyed needs to allow the audience to exclude possibilities in this manner.

207

that we have every reason to hold ND. The notions we associate with our expressions do indeed determine the semantic properties our expressions have in our mouths across the board. 7. Objections and replies Kripkean externalism founders on an epistemic problem, or so I have argued. It cannot account for knowledge competent speakers need to possess, and manifestly do possess, in order to participate in our communicatory practice. Let me briefly deal with two objections to this argument and the neo-descriptivist stance it supports. Here is the first objection: x

Can’t we hold that the contents we unvaryingly convey with our sentences have nothing to do with their semantic contents?

The proposal is that ‘There is water on Mars’ has an externally determined semantic content (roughly: that there is H2O on Mars. Call this the ‘secondary content’) that is utterly independent from the non-semantic content unvaryingly conveyed with assertive utterances of that sentence (roughly: that there is watery stuff on the Red Planet. Call this the ‘primary content’). There is something right about this. We should indeed acknowledge that our sentences have both contents here distinguished. The resulting twodimensional semantic, assigning primary and secondary contents to our sentences and primary and secondary intensions to our subsentential terms, is of course congenial to neo-descriptivism (see Jackson 2004; Nimtz 2007, ch. 8). However, there is also something wrong about this. First, we have every reason to class the content conveyed with ‘There is water on Mars’ as semantic. After all, bearing that content is a representational feature of the sentence, it is a feature any competent speaker has to be aware of, and it is a feature that we can and do exploit in communication irrespective of the context. Secondly, the two contents in play are not independent of one another. That ‘water’ in our mouths picks out (i.e., secondarily designates) H2O depends on the fact that we use this very term to pick out (i.e., primarily designate) the watery stuff around here. Assuming otherwise doesn’t fit with what we find. What ‘elm’, ‘water’, ‘plutonium’ etc. pick out is constrained by what we use these terms to pick out. If our considered use of ‘elm’ picks out a variety of trees, the term’s semantic value cannot

208

turn out to be a kind of bird. How we use our terms and what they apply to cannot deviate that much. Worse still, assuming otherwise means risking methodological self-defeat. Kripkean externalists rely on our intuitive judgments as to what ‘water’ designates to determine what the term in fact designates. But our intuitive judgments are informed by how we apply that term, and what we convey with it. But if the term’s designation wasn’t constrained by our application of it, we had no reason at all to hold that our intuitions track the term’s semantic properties. The neo-descriptivist stance I have defended invites a more general worry. The objection here is this: x

Neo-descriptivists hold that what determines semantic properties are the notions individual speakers associate with their terms. But how can such an individualist account explain that we all speak English?

The driving force behind this objection is the lingering suspicion that neodescriptivism can neither account for the collective nature of speaking a language, nor for the semantic homogeneity we find amongst speakers of the same language. However, to think so would be to overlook, first, the emphasis neo-descriptivists put on deference. They emphatically embrace Putnam’s ‘division of linguistic labour’ (1975, 227f.) and acknowledge that many speakers defer to their co-speakers in their application of terms (Jackson 2004, 270–273; Chalmers & Jackson 2001, 327f.; Chalmers 2002, 170–173). The resulting socio-linguistic picture distinguishes nondeferring experts from deferring laymen and thereby avoids the threat of rendering semantics a private matter. To think so would, secondly, underrate the importance of conventions (see Jackson 2004). We mutually expect one another to associate notions with our expressions guaranteeing that communication proceeds smoothly. This requires that we convey essentially the same contents and associate essentially the same notions with our expressions. (I’ll come back to the ‘essentially’ in a moment.) This does not come about by itself. It rather requires social mechanisms and a collective effort. This is precisely what we find. We correct speakers whose notions differ from ours more than we think tolerable, we compile dictionaries and consult them to further consistency and settle controversies, we flag irreconcilable differences in notions if we think they matter, and we pass authoritative rulings on which conditions one should associate with a term, even if that means having only 209

eight planets in the solar system instead of the traditional nine. Furthermore, whenever we teach someone a term, we do not rest content unless she applies it in line with our notions. In short, we employ reliable means to make people associate the right conditions with their terms. To think so would, finally, be to succumb to what I dub the homogeneity myth. If we ignore context-dependence for the moment, it seems quite natural to hold that, since we all speak the natural language, viz. English, our words must mean the same. Embracing this is to presume that differences in meaning are incompatible with sameness of natural language. Natural though it may seem, this idea of semantic homogeneity is a myth. The natural languages we know aren’t semantically homogenous. We find all kinds of semantic differences within the natural language we share. Consider once more the term ‘hat’. According to the OED, there are two uses of this term. On the former, a hat is almost any covering for the head. On the latter, a hat is a specific kind of headgear that typically has ‘a more or less horizontal brim all round the hemispherical, conical, or cylindrical part which covers the head’. If you follow the former use and I follow the latter, you will and I won’t count a Basque-style beret as a hat. But there is no basis for holding that one of us must be making a mistake. And the fluency of our communication proves that both of us do, of course, speak English.21 Moreover, the natural languages we know need not be semantically homogenous. Reliable communication does not require that the content the audience takes up has to be precisely the one the speaker expresses. As long as the way I present things as being in uttering ‘Nice sofas are hard to find’ and the way you take things to be in believing me are sufficiently similar, everything is fine. Generally speaking, there is no need for all the notions in a community to be exactly the same; it suffices if they overlap substantially. Our praxis reflects that. On the one hand, we often find us willing to tolerate deviations we deem harmless. We usually don’t mind if how you apply ‘sofa’ or ‘hat’ slightly deviates from how I do. On the other hand, we habitually presuppose that our notions differ, and that we need to agree on our notions when something is at stake. It is no accident that our laws and contracts are full of meticulous classification.

21. I develop this in more detail in Nimtz 2009.

210

8. Thought experimenting as conceptual analysis In the paper, I have put forth two different lines of thought—one epistemic, the other semantic. As for the first line of thought, I have argued that there is a marked epistemic difference between mundane and scientific counterfactuals such as (8) and (9) on the one hand, and philosophical counterfactuals such as (4) or (5) on the other. Counterfactuals of the former variety are to be classed as a posteriori, since the bodies of belief licensing our move from their antecedents to their consequents comprise empirical evidence and/or empirical theory. By contrast, the counterfactuals we find at the heart of philosophical thought experiments are to be classed as a priori.22 I have also ventured that we draw on the notions guiding how we do and would apply our terms to assess such philosophical counterfactuals. For example, we consider (5) to be true because we find that the subject in the Nogot/Havit case does not satisfy the conditions guiding our application of ‘knowing’, and hence does not count as a knower. As for the second line of thought, I have proposed and defended a semantic theory I label neo-descriptivism. The semantics is labelled neodescriptivism since it maintains that what determines the semantic properties our terms have in our mouths are the notions we associate with these terms, guiding their (considered) application across possible situations. The semantics is labelled neo-descriptivism since by its two-dimensional structure, it incorporates key Kripkean insights such as that many of our terms designate rigidly, that there are necessary truths a posteriori, and that there are contingent truths a priori (see Kripke 1980, lecture I). For all that, the neo-descriptivist idea that notions determine semantic properties patently runs against Kripkean externalism. I have devised the argument from communication to defend neo-descriptivism against its externalist rival. I have argued that since neo-descriptivism can account for non-trivial knowledge competent speakers do possess, and need to possess in order to participate in our communicatory practice, whereas Kripkean externalism cannot, we have every reason to accept the neo-descriptivist account over the Kripkean alternative. Combining the two lines of thought yields a lucid account of philosophical thought experiments. What we go on in evaluating the counterfactuals at the heart of philosophical thought experiments are the notions 22. Mutatis mutandis the same holds true if we take modal links to be restricted necessities in line with (7), rather than bona fide counterfactuals.

211

guiding our considered application of our terms. But on neo-descriptivist premises, these notions are semantic in that they determine the semantic properties our terms have in our mouth. What we do in contemplating the likes of the Mary scenario or the Nogot/Havit case, then, is that we get clear about what our terms apply to by inquiring into how we would apply them, thereby rendering explicit the notions we associate with them. Philosophical thought experiments thus turn out to be instances of the method of possible cases, and philosophical thought experimenting is found to be a variety of what has traditionally been dubbed ‘conceptual analysis’ (see Grice 1958; Jackson 1998, ch. 2)—a variety, that is, of the endeavour to ascertain what our words mean in our mouths.23

REFERENCES Bealer, George. 1998. “Intuition and the Autonomy of Philosophy”. In: Michael DePaul and William Ramsey (eds.) 1998, 201–239. — 2000. “A Theory of the A priori”. Pacific Philosophical Quarterly 81, 1–30. Bealer, George. 2002. “Modal Epistemology and the Rationalist Renaissance”. In: Tamar Szabó Gendler and John Hawthorne (eds.) 2002, 71–125. Bennett, Jonathan. 2003. A Philosophical Guide to Conditionals. Oxford: Oxford University Press. Berger, Alan. 2002. “A Formal Semantics for Plural Quantification, Intersentential Binding and Anaphoric Pronouns as Rigid Designators”. Nous 36, 50–74. Chalmers, David. 1996. The Conscious Mind: In Search of a Fundamental Theory. Oxford: Oxford University Press. — 2002. “On Sense and Intension”. Philosophical Perspectives 16, 135–182. — 2004. “Epistemic Two-Dimensional Semantics”. Philosophical Studies 118, 153–226. Chalmers, David and Frank Jackson. 2001. “Conceptual Analysis and Reductive Explanation”. Philosophical Review 110, 315–361. Cooper, Rachel. 2005. “Thought Experiments”. Metaphilosophy 36, 328–347. Davidson, Donald. 1987. “Knowing One’s Own Mind”. In: Quassim Cassam (ed.). 1994. Self-Knowledge. Oxford: Oxford University Press, 43–64. 23. I would like to thank Jessica Leech and Wolfgang Schwarz for valuable comments on earlier versions of this paper. This material has also greatly profited from discussions in Zürich and Cologne, for which I am very grateful.

212

Dennett, Daniel. 1995. “Intuition Pumps”. In: John Brockman (ed.). 1995. The Third Culture: Beyond the Scientific Revolution. New York: Simon & Schuster, 167–181. DePaul, Michael and William Ramsey (eds.). 2000. Rethinking Intuition. The Psychology of Intuition and Its Role in Philosophical Inquiry. Lanham/Boulder: Rowman & Littlefield. Edginton, Dorothy. 2008. “Counterfactuals”. Proceedings of the Aristotelian Society 108, 1–21. Gendler, Tamar Szabó and John Hawthorne (eds.). 2002. Conceivability and Possibility. Oxford: Clarendon Press. Gettier, E. L. 1963. “Is Justified True Belief Knowledge?” Analysis 23, 151–153. Goodman, Nelson. 1983. Fact, Fiction, and Forecast. Cambridge (Mass.): Harvard University Press. Grice, Paul. 1958. “Postwar Oxford Philosophy”. In: Paul Grice. 1989. Studies in the Ways of Words. Cambridge (Mass.): Harvard University Press, 171–180. Häggqvist, Sören. 2009. “Modal Knowledge and Thought Experiments”. In: Nikola Kompa, Christian Nimtz & Christian Suhm (eds.) 2009, 53–68. Jackson, Frank. 1998. From Metaphysics to Ethics. A Defense of Conceptual Analysis. Oxford: Oxford University Press. — 2000. “Representation, Scepticism, and the A priori”. In: Paul Boghossian and Christopher Peacocke (eds.). 2000. New Essays on the A priori. Oxford: Oxford University Press, 320–332. — 2004. “Why We Need A-Intensions”. Philosophical Studies 118, 257–277. — 2004b. “Foreword: Looking Back on the Knowledge Argument”. In Peter Ludlow, Daniel Stoljar & Yujin Nagasawa (eds.) 2004, xv-xix. — 2005. “What Are Proper Names for?” In: Maria E. Reicher and Johann C. Marek (eds.). 2005. Experience and Analysis. Kirchberg am Wechsel, 102–114. — 2007. “Reference and Description from the Descriptivists’ Corner”. Philosophical Books 48, 17–26. — 2007b. “On Not Forgetting the Epistemology of Names”. Grazer Philosophische Studien 74, 239–250. — 2010. “Conceptual Analysis for Representationalists”. In this volume, 173–188. Katz, Jerrold. 1998. Realistic Rationalism. Cambridge (Mass.): MIT Press. — 2002. “Mathematics and Metaphilosophy”. Journal of Philosophy 99, 362–390. Kripke, Saul. 1980. Naming and Necessity, second edition. Oxford: Blackwell. Knobe, Joshua and Shaun Nichols (eds.). 2008. Experimental Philosophy. Oxford: Oxford University Press. Lehrer, Keith. 1965. “Knowledge, Truth, and Evidence”. Analysis 25, 168–175. Lewis, David. 1973. Counterfactuals. Oxford: Blackwell. Ludlow, Peter, Daniel Stoljar and Yujin Nagasawa (eds.). 2004. There’s Something

213

about Mary: Essays on Phenomenal Consciousness and Frank Jackson’s Knowledge Argument. Cambridge (Mass.): MIT Press. Miller, Richard. 2000. “Without Intuitions”. Metaphilosophy 31, 231–250. Kompa, Nikola, Christian Nimtz and Christian Suhm (eds.). 2009. The A priori and Its Role in Philosophy. Paderborn: mentis. Nimtz, Christian. 2004. “Reference Fixing, Conceptual Analysis, and the A priori”. In: Roland Bluhm and Christian Nimtz (eds.). 2004. Selected Papers Contributed to the Sections of GAP.5. Paderborn: mentis, 320–331. — 2007. A New Rationalism? Intuitions, Modal Knowledge, and the Analysis of Concepts. Unpublished manuscript. (Habilitationsschrift.) — 2009. “Conceptual Truth Defended”. In: Nikola Kompa, Christian Nimtz and Christian Suhm (eds.). 2009, 137–155. Norton, John 2004. “On Thought Experiments: Is There More to the Argument?” Philosophy of Science 71, 1139–1151. Putnam, Hilary. 1975. “The Meaning of ‘Meaning’ ”. In: Hilary Putnam. 1975. Mind, Language, and Reality. Philosophical Papers, Vol. 2. Cambridge: Cambridge University Press, 215–271. Quine, Willard V.O. 1995. From Stimulus to Science. Cambridge (Mass.): Harvard University Press. Shope, Robert. 1983. The Analysis of Knowing. Princeton (NJ): Princeton University Press. Stalnaker, Robert. 1968. “A Theory of Conditionals”. In: Nicholas Rescher (ed.). Studies in Logical Theory. Oxford: Oxford University Press, 98-112. — 1984. Inquiry. Cambridge (Mass.): MIT Press. Williamson, Timothy. 2004. “Armchair Philosophy and Counterfactual Thinking”. Proceedings of the Aristotelian Society 105, 1–23. — 2007. The Philosophy of Philosophy. Oxford: Blackwell. — 2007b. “Philosophical Knowledge and Knowledge of Counterfactuals”. Grazer Philosophische Studien 74, 89–123.

214

Grazer Philosophische Studien 81 (2010), 215–233.

KRIPKE AND THE NEO-DESCRIPTIVIST Finn SPICER University of Bristol Summary This paper looks at how neo-descriptivism grew out of Kripke’s anti-descriptivist arguments and examines two arguments for neo-descriptivism: one from Frank Jackson and one from Christian Nimtz. The former argument is that neo-descriptivism best explains how we are able to judge the referent of a term at a possible world when presented with a description of that world; the second argument is that only neo-descriptivism can account for our ability to gain new knowledge from testimony. The paper concludes that neither argument is successful.

1. A standard package of views Since Kripke’s Naming and Necessity (1980), it has been the orthodox view that the reference of a name is not fixed by any description that the speaker associates with the name; rather, the reference of a name is fixed by facts external to the speaker. The reference of a speaker’s use of the name ‘Gödel’ is fixed by the chain of use that goes back from current uses of the name ‘Gödel’ to the person Gödel himself. While Kripkeanism about names has become the orthodoxy, Kripkeanism about general terms has not caught on to the same extent. Many who accept Kripke’s line on names have chosen not to extend their Kripkeanism to general terms, or have done so only to a limited degree.1 Rather, they hold on to a kind of descriptivism about general terms, signing up to the view that speakers tacitly associate general terms with other general terms, forming descriptions or generalisations that help determine the meaning of these terms. So for example, speakers who understand ‘dog’ and ‘animal’ associate these terms in a way that embodies 1. These latter theorists extend their Kripkeanism only to a narrow class of general terms— natural kind terms fairly narrowly conceived. Chemical elements and biological species would be examples of such natural kinds; artefact terms and folk psychological terms might be excluded as not natural kind terms so conceived.

the generalization that dogs are animals, and this constrains the meaning of the term ‘dog’ such that necessarily the extension of ‘dog’ is a subset of the extension of ‘animal’. Likewise, speakers who understand the terms ‘knowledge’ and ‘belief ’ partly do so in virtue of their tacit knowledge of the fact that knowledge implies belief. Connections such as that between ‘dog’ and ‘animal’ and between belief ’ and ‘knowledge’ are sometimes called conceptual connections, and truths such as the truth that all dogs are animals and that knowledge implies belief are sometimes called conceptual truths.2 I won’t speculate on why this package of views (Kripkeanism about names, descriptivism about most general terms) has proved popular, beyond these few remarks. First, I suspect that although philosophers are not in general interested in analyzing terms like ‘dog’ and ‘animal’, they are interested in analyzing philosophically important terms like ‘knowledge’ and ‘belief ’; despite recent attacks,3 the old analytic practice of seeking to find the necessary features of philosophically important categories retains its currency. Belief in conceptual truths and descriptivism are the underpinnings of this practice. Perhaps for these reasons, there has been a descriptivist inertia that has halted the spread of Kripkeanism beyond the realm of names and natural kinds narrowly construed. Secondly, it is harder to work out the details of Kripke’s story about reference-fixing and rigid designation for the case of general terms than for names; Kripke gives a crisp summary of his position on general terms (1980, 135–8), but developing this summary into a fully worked out system has proved harder (see Salmon, 2005, §§ 4–6). Elaborating a notion of rigid designation applicable to general terms is problematic (Salmon 2005, § 6; Soames 2002, ch. 9). Also, whereas a particular (the referent of a name) can be the source of a chain of use that extends to current uses of a name, it is hard to see how a kind or category (the referent of a general term) could be the source of a chain of use in the same way; rather than the whole kind, it is instances of the kind that can stand at the source of a causal chain. Given this, theorists who have developed Kripkean accounts of kind terms, have had to appeal to a relation of sameness-of-kind that holds between the instance of the kind present at the baptism and the instances that the term truly applies to in use; this appeal has proved problematic for a wide swathe of terms. 2. These are used as terms of art throughout this paper. For a discussion of this notion of conceptual truth and its relationship to the notion attacked by Tim Williamson (2006), see my 2008. 3. With respect to analysing knowledge, the attacks have come from Kornblith (2002), Spicer (2008), Weinberg, Nichols and Stich (2001), and Williamson (2006).

216

Consider the term ‘gold’; some of the gold in the world has been discovered, mined, handled and named, and some of this gold has played a role in beginning and maintaining the causal chain of use that current uses of the term ‘gold’ are a part of—call this the ‘baptised gold’. The baptised gold is only a fraction of all the gold that forms the extension of the term ‘gold’; how does ‘gold’ come to refer to all the gold there is? Kripke’s answer is that all gold shares an essence with the baptised gold; this essence provides a sameness-of-kind relation that figures in the reference relation between ‘gold’ and gold. What, though, makes it the case that it is the relation of shared essence with the baptised sample that determines the reference of ‘gold’, rather than some other relation (such as identity with the baptised sample, or sameness of appearance with the sample)? Those philosophers who have partially embraced Kripkeanism are perhaps willing to concede that the identity relation (for particulars) and the shared-essence relation (for natural kinds) are sufficiently privileged relations as to make the question ‘why not some other relation?’ moot. For other terms, such as ‘shrub’, the question is not moot but pertinent, and is hard to answer. Shrubs do not share any chemical or biological essence, and so the question of in what respect a particular plant must be like the baptised sample shrubs cannot be given the immediate privileged answer ‘sameness of essence’. There must be some relation that a member of the category shrub must have to the baptised sample for it to be truly a ‘shrub’—what shrubs have in common, we might say. But there’s nothing particularly special about what shrubs have in common, we might have used the term ‘shrub’ for a wider category (or for an otherwise different category), had our needs been different; but we ended up reserving the term ‘shrub’ just for this category—for those things that have this (fairly motley) set of things in common. What makes this motley set hang together is us, our needs and our practices; in short, we decide what counts as a shrub, and we do so by tacitly associating a bunch of properties (and relations to paradigms) with the term ‘shrub’. Descriptivism seems right for ‘shrub’, the thought goes, and in fact for all terms beyond names for particulars and a fairly narrow class of kind terms referring to natural kinds such as ‘gold’. It seems that our terms lie on a line, with names at one end and natural kind terms near them, with ‘shrub’ further down the line; on this line there seems to be a point beyond which lie terms like ‘shrub’, terms where the category to which the term refers is up to us, settled by our intentions and tacit knowledge of the shape of the category. To the left of the ‘K-point’ are the Kripkean terms, to the right of the K-point are the descriptivist 217

terms. It’s not my plan in this paper to decide where to place the K-point; my argument is rather with those who seek to place the K-point to the left hand extreme: those who hold that there are no Kripkean terms; all terms are descriptivist—the neo-descriptivists. 2. Neo-descriptivism The ‘neo-descriptivists’ are represented most notably by Frank Jackson (see his 1998, 2005, 2007a and 2007b as well as his contribution to this volume). Neo-descriptivists do not restrict their descriptivism to a few general terms: they resist Kripke’s conclusions about the semantics of names and natural kind terms, offering replies to Kripke’s arguments and hence offering a descriptivist account of names and kind terms. In this paper, my focus is more on neo-descriptivism’s view of general terms, and on two arguments offered for it. Neo-descriptivism is a detailed theory that combines a view of what is psychologically involved in understanding general terms with a theory of what semantically determines these terms’ reference. These two ingredients—psychological and semantic—can be illustrated using the example of the term ‘knowledge’. As for the first ingredient, neo-descriptivism states that understanding the term ‘knowledge’ requires tacitly associating it with other notions, such as the notion of belief. Note that the association in the speaker’s mind is between the general term and other notions, not other terms (to use Nimtz’s helpful term ‘notion’, see this volume, 204); the speaker need not have a word for each or any of the notions he associates with the term ‘knowledge’.4 This association might include tacitly accepting the inference from knowledge to belief, tacitly accepting the inference from knowledge to truth, and tacitly accepting other inferences. The former two inferences are perhaps universally tacitly accepted among speakers in our community who possess the term ‘knowledge’; the further inferences might not be universal. Which inferences a speaker accepts will individuate the concept knowledge that the term ‘knowledge’ expresses on his lips. In the light of the history of disagreement there has been since Gettier on the analysis of knowledge, Jackson’s view about the 4. Jackson is explicit about this: “[When I associate the descriptive condition D with a term “A”] I may or may not have a word “D” in my language. The theory should probably have been called the property theory of reference.” (2007b, 240), as is Nimtz’. Thanks to Christian Nimtz for emphasizing this point to me.

218

term ‘knowledge’ is that there is a variety of concepts expressed by the term in our linguistic community (and that some speakers express different concepts with ‘knowledge’ on different occasions—see his 2005, and his paper in this volume, § 5). The second ingredient of neo-descriptivism is a semantic claim. Neodescriptivism states that tacitly associated inference from knowledge to belief plays a semantic role—descriptively constraining the reference of the term ‘knowledge’. ‘Knowledge’ is like ‘shrub’: there are many nearly overlapping categories that include all the paradigm instances of knowledge—many potential referent categories projectable from the baptism sample. It’s up to us to decide5 which of the possible referent categories knowledge in fact refers to and we decide this tacitly, by the tacit representation of what all instances of knowledge have in common. An example of how the tacitly accepted notions constrain the reference of ‘knowledge’ is the relation to the notion of belief: whatever ‘knowledge’ refers to, its referents must make true the principle that knowledge implies belief: the category knowledge must be a subset of category belief. ‘Knowledge’ refers to a set of mental states that all have the property of being beliefs. Because something we tacitly know is playing this semantic role, conceptual analysis can help us find out about knowledge. We tacitly know what instances of knowledge have in common, and conceptual analysis can make this explicit. In fact, in the case of ‘knowledge’, as mentioned above, Jackson’s view is that we don’t have a single tacit picture of what instances of knowledge have in common; for this reason our decision about which category is the referent of ‘knowledge’ is incomplete, and so part of the work conceptual analysts are doing in analysing knowledge is completing the decision, rather than discovering the category we already, universally and consistently call ‘knowledge’ (see Jackson’s paper in this volume, § 5). This, then, is neo-descriptivism: it is a thesis about understanding terms that combines with a semantic thesis to yield as a corollary a thesis about the potential of conceptual analysis. In this paper, I examine two arguments for neo-descriptivism, and I will conclude that they do not work. The first argument for neo-descriptivism is born out of Kripke’s own antidescriptivist arguments; the second argument concerns the issue of what is involved in understanding and communication. Taking these in order, I begin by returning to Kripke’s arguments against descriptivism. 5. This ‘up to us to decide’ locution recurs throughout Jackon’s writing (see for example his 2007a: 18, and this volume: § 6)

219

3. Kripke’s three arguments Kripke’s Naming and Necessity (1980) mounts a sustained attack on descriptivism—a view which at the time Kripke was writing was the orthodoxy and which had its roots in Frege and Russell’s work. Among the various anti-descriptivist points made in Kripke’s book, there are three major arguments. Because they are famous arguments, my sketches of them will be brief. Kripke’s first argument is his epistemic argument (1980, 87). Kripke argues that if understanding ‘Cicero’ involves associating it with the description ‘the denouncer of Catiline’, and if the name has its reference fixed by this description, then sentence (1) is knowable a priori to anyone who understands it. (1) Cicero denounced Catiline. But sentence (1) is not knowable a priori, so ‘Cicero’ does not get its reference fixed by the description ‘the denouncer of Catiline’. The second argument is his modal argument. Kripke argues (1980, 74ff.) that ‘Cicero’ cannot mean the same as ‘the denouncer of Catiline’, arguing as follows. If ‘Cicero’ and ‘the denouncer of Catiline’ mean the same, then (1) would mean the same as: (2) the denouncer of Catiline denounced Catiline. But if (1) and (2) mean the same, then because (2) is necessarily true, (1) must be necessarily true also. But (1) is not necessarily true, because Cicero might not have denounced Catiline. So ‘Cicero’ and ‘the denouncer of Catiline’ are not synonyms. Kripke’s third argument is the semantic argument (1980, 82–89). It involves counterfactual reasoning like the modal argument, but it’s important to see how it differs from the modal argument. Whereas in the modal argument a possible world is considered as counterfactual (Cicero might not have denounced Cataline), in Kripke’s semantic argument a possible world is considered as actual.6 Consider a thinker, Fred, who associates the description ‘the discoverer of the incompleteness of arithmetic’ with the name ‘Gödel’. Now suppose it turns out that despite the common view 6. See Stalnaker (1999) for more on this distinction.

220

that Gödel discovered that theorem, he did not. Not Gödel but someone else—Schmidt—did (imagine some story about Gödel being part of a large deception to hide Schmidt’s discovery and claim it as his own). Kripke’s semantic argument is: if reference is fixed by the description a speaker associates with the term, then given Schmidt proved the incompleteness theorem, ‘Gödel’ on Fred’s lips would refer to Schmidt. But this is not so: even if it happens to be the case that Schmidt discovered the incompleteness theorem, Fred still refers to Gödel with his name ‘Gödel’. Hence the reference of Fred’s name ‘Gödel’ can’t therefore be fixed by the description Fred associates with the name. Each of these arguments is presented using an arbitrarily chosen example of a name and description—‘Cicero’ and ‘denouncer of Cataline’, and ‘Gödel’ and ‘discoverer of the incompleteness of arithmetic’. Because they are arbitrary, Kripke takes himself as licensed to generalise the conclusion of each argument. Not only is (1) not a necessary truth, nor is any similar sentence, such as ‘Cicero was the most versatile Roman orator’; likewise not only is (1) not knowable a priori, nor is any similar sentence. So although the immediate conclusion of Kripke’s argument is that the description ‘denouncer of Cataline’ does not fix the reference of ‘Cicero’, Kripke goes on to draw the larger conclusion that nor does any other description fix the referent of this or any other name. 4. How neo-descriptivism replies to these three arguments Neo-descriptivists need to respond to these three arguments, and they do with vigour—building a case for their brand of descriptivism out of the very raw materials that Kripke provides. In response to the modal argument, the neo-descriptivist replies by pointing out that it is easy to rigidify a description, and so give it the correct modal profile to match that of the name. The modal argument claims that ‘Cicero’ cannot be synonymous with ‘the denouncer of Catiline’ since these terms have different modal profiles: the former refers to Cicero at all (Cicero-containing) possible worlds; the latter refers to other individuals at some worlds (those where someone else denounced Catiline). Given that Cicero did denounce Catiline (here in the actual world) ‘Cicero’ and ‘the actual denouncer of Catiline’ will corefer at every world. In response to the epistemic argument, the neo-descriptivist points out that the conclusion (that descriptivism implies widespread a priori knowl221

edge) only appears unwelcome in the light of Kripke’s chosen examples. Kripke’s examples are about famous people (Cicero, Aristotle, Gödel), and speakers associate the names of famous people with their deeds; but these associations are not the reference-fixing ones, neo-descriptivists claim. For everyday kinds terms such as ‘water’, the reference-fixing description will describe the syndrome of water’s manifest properties, such as the stuff that falls as rain and flows in rivers. For philosophical general terms like ‘knowledge’, principles such as knowledge implies belief will be part of the reference-fixing description. For names such as ‘Cicero’, Kripke’s own causal-historical theory of reference provides the reference-fixing description: the descriptions that determine the reference of a person’s name are not those describing their salient characteristics or deeds, rather they are descriptions that describe the person as being causally connected to the name. The neo-descriptivist will claim that, relative to these descriptions, it will be knowable a priori that water falls as rain; that knowledge implies belief; that Cicero is the origin of the causal chain of uses of ‘Cicero’. Jackson gives an example of a rather recherché piece of a priori knowledge about water: ‘It is a priori that the A-extension of ‘water’ is the watery stuff of our acquaintance …’ (1998, 46–52); a more simple piece of a priori knowledge (which follows from that given above by Jackson) is that water is the watery stuff of our acquaintance. The neo-descriptivist response to the last argument—the semantic argument—needs to be spelt out in more detail, because neo-descriptivists don’t regard the semantic argument merely as an argument to be resisted: they find in it the raw materials for building a positive argument for their position. The semantic argument describes a possible world (one like this except where Gödel does not prove the incompleteness theorem, Schmidt does) and invites us to judge whom the name ‘Gödel’ in that world refers to. And upon considering the imagined scenario, we judge that the name in that world refers to Gödel, not to Schmidt. Kripke’s conclusion is that the description ‘prover of the incompleteness theorem’ cannot be fixing the referent of ‘Gödel’; once again he uses the arbitrariness of the chosen example to draw a further corollary by universal generalization. There was nothing special about the description ‘prover of the incompleteness theorem’, Kripke assumes, and so the conclusion that that description does not fix the referent can be generalized to become the corollary that no description fixes the referent. Frank Jackson (1998, 212) claims that Kripke’s thought-experiment shows that Kripke’s corollary does not follow from his more specific con222

clusion. The thought experiment and speakers’ response to it, Jackson claims, show that speakers do tacitly know the reference-fixing description associated with the name. Here’s why: when Kripke invites us (competent speakers and users of the name ‘Gödel’) to take part in his thought experiment, we are presented with a description of a possible world (to be considered as actual); we are asked about the reference of the name ‘Gödel’ used at that world, and we get it right—‘Gödel’ refers to Gödel not Schmidt. That we speakers all get it right in our intuitions about this possible world shows two things, Jackson says. First that speakers tacitly know that the property of being the discoverer of the incompleteness theorem is not the reference-fixer of ‘Gödel’ (since we do not judge that in the described world, ‘Gödel’ refers to Schmidt). Second it shows that whatever the real reference-fixing feature is (call it FG), we as competent speakers must tacitly know that FG fixes the reference of ‘Gödel’. Jackson’s argument for the second claim is as follows; from a description of a possible world, speakers are able to identify the referent of ‘Gödel’, therefore speakers must have knowledge that allows them to deduce the referent from the description of a possible world; knowledge of FG is the knowledge that allows them to do this. Kripke’s conclusion, remember, was first that ‘discoverer of the incompleteness of arithmetic’ doesn’t fix the reference of ‘Gödel’, and then he added the corollary that no other description does either. Jackson agrees with the first conclusion; he agrees that the thought-experiment does establish that the feature being the discoverer of the incompleteness of arithmetic doesn’t fix the reference of ‘Gödel’. But Jackson rejects Kripke’s corollary; he takes Kripke’s own thought experiment to show that the speaker does tacitly know some other feature to be the reference-fixer for ‘Gödel’. This lesson born of Kripke’s semantic argument becomes a theme for Jackson’s program of defending and practicing conceptual analysis: ‘Indeed sometimes the very same data—about what refers to what—are taken by opponents of the description theory as making a powerful case against the description theory but by its supporters as serving to elucidate the descriptions that determine reference’ (2007a, 17).7 Jackson claims that from a description of a possible world, we are able to say which object is called ‘Gödel’ in that world—manifesting our ability to do so in the form of a ‘semantic intuition’. Likewise we have semantic intuitions about ‘water’: 7. Nimtz draws an analogous lesson from what he calls Kripke’s epistemic argument (different from the argument of Kripke’s I call the epistemic argument above); see Nimtz this volume.

223

from a description of a world, we are able to say which stuff is called ‘water’ in that world. This is just what we are doing in the Putnam thought-experiments: combining a description of a world (Twin Earth) with our tacit knowledge of the conditions a stuff must satisfy if it is to be the referent of ‘water’, to yield a conclusion about which stuff is the referent of ‘water’ on Twin Earth. Because the descriptive condition on being the referent of ‘water’ involves an indexical element (‘the stuff that flows in rivers, falls as rain and fills the lakes around here …’), our semantic intuitions deliver different answers about the stuff on Twin Earth depending on whether we consider it as a possible-world-considered-as-actual (on Twin Earth, we’d use ‘water’ to refer to XYZ) or consider it as a possible-world-consideredas-counterfactual (there’s no water there, only XYZ). Putting this the other way around: the fact that we have different semantic intuitions about Twin Earth considered as actual and counterfactual shows that the descriptive condition we associate with ‘water’ contains an indexical element—that is all that Putnam has shown, according to the neo-descriptivist. For Jackson, the above story provides a general model of what goes on when philosophers consult their intuitions about a hypothetical scenario— an activity that is a central part of the traditional philosophical practice of conceptual analysis. Consider the Gettier counterexamples: we are asked to consider a possible world (as counterfactual) in which a person forms a belief in a familiar Gettier-way. From that description of the possible world (and of that person and his belief in particular) and from our tacit knowledge of the necessary conditions a belief must satisfy if it is to be knowledge, Jackson claims, we are able to deduce that the man in the story does not know, but merely believes. The defence of neo-descriptivism that Jackson builds out of the resources he finds in Kripke, then, is intimately connected with the practice of conceptual analysis and its basis. 5. Responding to Jackson’s argument Jackson’s argument is best thought of as an instance of IBE—inference to the best explanation. Speakers exhibit the ability to produce a judgment in which they specify an object in a described possible world as the referent of a given term. So, what explains this ability? The answer (for Jackson) is that the best explanation of speakers’ ability to respond is that they have tacit knowledge of the descriptive condition that a thing must satisfy if it is to be the referent. They combine this tacit knowledge with 224

the description of the possible world, and deduce the conclusion about which object is the referent. Likewise, Jackson thinks that our ability to say that there is no water on Twin Earth (and that native’s uses of ‘water’ will refer to XYZ) is best explained by our tacit knowledge of the conditions a stuff must satisfy if it is to be what we call ‘water’ (and our knowledge of the conditions that a stuff must satisfy if it is to be what natives on Twin Earth call ‘water’). And finally, our ability to judge that a Gettier-belief is not knowledge is best explained by our tacit knowledge of the necessary conditions on a belief ’s being knowledge. If the right way to understand Jackson’s argument is as an instance of IBE, then the right way to criticize Jackson’s argument is to question whether the given explanation really is the best explanation of the data. Jackson’s explanation is not the best, I claim, because it is not the most parsimonious: Jackson’s explanation employs stronger assumptions than are required to deliver his explanation of the data. Recall from above that neo-descriptivism involves two claims: a psychological claim (P) and a semantic claim (S) (P) for many general terms, there is for each term a description that all speakers who understand the term tacitly associate with it, and which guides their reasoning about how the term refers (S) the reference of the general term is partly determined by this description: the category to which the term refers must satisfy the description Jackson argues for the conjunction of both (P) and (S) by giving both a role in the best explanation of speakers’ ability to respond with semantic intuitions to thought-experiments. But if one examines the details of his explanation, only (P) is really used; (S) is not used and so is eliminable. What Jackson is trying to explain is speakers’ ability to produce a judgment; he explains this by appeal to psychological claim (P) that speakers tacitly represent descriptions as being associated with the terms deployed in the judgment, and are guided by those descriptions in the process of reasoning that leads to the judgment. Illustrating this explanation for a specific example: in the case of Putnam intuitions, the datum to be explained is the fact that when presented with a description of Twin Earth and the speakers on Twin Earth, we judge that ‘water’ on Twin Earth would refer to XYZ; the explanation of this datum is that speakers tacitly represent 225

the description ‘the stuff around here that falls as rain, etc.’ and use this to reason about the reference of ‘water’, and they are guided by this representation to infer from the description of Twin Earth to the conclusion that ‘water’ on Twin Earth refers to XYZ. The semantic claim (S) that this description really does play a role in determining the reference of ‘water’ is not used in the explanation; all that’s used is the psychological claim (P) that speakers use the description in their reasoning about reference. If all that’s needed to explain speakers’ semantic intuitions is the psychological claim (P), then positing only (P) and remaining neutral on (S) is a simpler explanation of the data than positing the conjunction of (P) and (S). If the simpler explanation is the best explanation, then we have here an IBE argument for (P) alone—an argument that leaves (S) unsupported. Ceteris paribus, the simpler explanation is in general the better explanation, so unless Jackson can produce some reason for thinking that complicating the explanation by positing (S) makes it better in this case, he cannot claim to have any argument for (S). The only thing he might say in favour of complicating the explanation is that the simpler explanation is poorer because it uses (P) as an unexplained explainer. Without (S), Jackson might argue, (P) is mysterious: it’s mysterious why speakers would associate a description with a term (mentally giving the description a reference-fixing role, as it were) if in reality the description plays no reference-fixing role. But an externalist story about what fixes reference (one that says the description doesn’t really have a reference-fixing role) does not leave the existence of mentallyassociated descriptions a mystery: if what really fixes reference lies outside the head, it is not surprising that speakers have a representation of how to recognize the referent inside their heads—as a proxy for what’s outside. This representation will need to be accurate enough and computationally cheap enough to allow speakers to deal practically with their linguistic environment, but that’s all: we need not suppose that this representation which plays a reference-tracking role for the subject does so by representing the reference-determining relation that holds between the word and the object. Elsewhere I have argued that in the case of some philosophically interesting general terms, such as ‘knowledge’, the patterns of semantic intuitions that speakers produce are incoherent: when speakers are presented with a wide range of hypothetical scenarios and asked to make judgments about how the term ‘knowledge’ applies to these scenarios (i.e., when they are 226

given epistemological thought-experiments), the judgments they make fail to fall into any coherent pattern across modal space (see Spicer 2008). Such failure is just what one would expect on the externalist picture. If speakers represent just-enough-but-no-more about what knowledge is like in order to be able to get things pretty much right about knowledge in the actual world and in the nearby worlds where counterfactuals are evaluated, then we would expect them to start getting it wrong about what knowledge is like in the more distant worlds described in the more arcane epistemological thought-experiments. A neo-descriptivist picture that includes the claim (S) that speakers represent reference-determining information about knowledge will predict that speakers are right about how ‘knowledge’ applies in even the more remote possible worlds. It might be, then, that when we gather the data of speakers’ semantic intuitions more thoroughly, they exhibit a messiness that is better explained by the psychological claim that speakers associate a bunch of pretty-good, but not perfect rules of thumb for reference-tracking with their terms, than by Jackson’s psychological claim (P) that speakers associate a single, uniquely identifying description with each of their terms. 6. Nimtz’ new argument In his contribution to this special issue, Christian Nimtz offers a further argument for descriptivism (see Nimtz, 2007, 236–43, and this volume, sec. 6). Nimtz argues that speakers must know the reference of the terms they understand, if they are to be able to communicate with those terms. Nimtz then argues that this knowledge of reference must take the form of knowing a condition that the referent satifies. He proceeds as follows. Successful communication includes cases where hearers acquire new knowledge by testimony; for example, a hearer might be told ‘Gödel lived in Princeton’ and so gain the knowledge that Gödel lived in Princeton. In order to gain this knowledge, the hear must understand the sentence, which of course requires understanding each of the terms used in the sentence. Nimtz argues that since knowing that Gödel lived in Princeton is knowing something about Gödel and about Princeton, the understanding required to acquire this knowledge by testimony must include knowledge that the term ‘Gödel’ refers to Gödel and the knowledge that ‘Princeton’ refers to Princeton. Moreover, these pieces of knowledge-of-reference must be non-trivial: knowing that ‘Gödel’ refers to Gödel cannot merely 227

be knowing that ‘Gödel’ refers to the referent of ‘Gödel’. The non-trivial knowledge that Nimtz claims hearers must possess is knowledge of the reference-condition of ‘Gödel’ and of the reference-condition of ‘Princeton’: knowledge of the conditions that an object must satisfy if it is to be the referent of ‘Gödel’ or ‘Princeton’. Similarly, a hearer may learn that there is water on Mars by being told ‘There is water on Mars’; in order to acquire this piece of testimonial knowledge, the hearer must know what ‘water’ refers to, and know it in a non-trivial way. Nimtz argues that the way a hearer must know this is by knowing the conditions a stuff must satisfy if it is to be water. Only descriptivist knowledge of reference delivers a sufficiently rich kind of linguistic understanding to explain speakers’ and hearers’ abilities to transmit and receive testimonial knowledge through the practice of communication. 7. Replying to Nimtz’ argument Nimtz’ argument is epistemic; it isn’t like (Jackson’s argument) a mere inference to the best explanation. Had Nimtz merely been arguing that hearers are disposed to judge that Gödel lived in Princeton when they hear ‘Gödel lived in Princeton’, a reply along the same lines as the above reply to Jackson would suffice—distinguishing the claim that hearers associate various pieces of information with a term and the claim that this information plays a semantic, reference-determining role, and asking whether the second claim plays any role in the best explanation of hearers’ dispositions to judgment. Since Nimtz frames his argument in epistemic terms, though (focusing on communication’s capacity to transmit knowledge), I will answer the argument in epistemic terms. Nimtz claims that hearers acquire new testimonial knowledge by deploying their linguistic knowledge—knowledge of reference—to understand what they are told. I concede that knowledge of reference is required for communication, but still resist Nimtz’ neo-descriptivist conclusion. I claim that there is a form of non-trivial knowledge of reference that suffices for communication, but which does not take the descriptivist’s form. In short, the knowledge that a hearer has when they understand the name ‘Gödel’ is the following: (R*) If ‘Gödel’ refers, it refers to Gödel.

228

The conditional form of this knowledge (which allows speakers to understand a name while being neutral on whether the name is empty) doesn’t matter for the purposes of this paper, so for simplicity and without risk, we can think of the stronger (R) being involved in understanding (R) ‘Gödel’ refers to Gödel. If this suggestion about what knowledge is involved in understanding a name such as ‘Gödel’ is going to be a sufficient answer to Nimtz’ demand for non-trivial knowledge of reference, then a hearer can’t be thinking of Gödel on the right hand side of (R) via the description the referent of ‘Gödel’. I agree, a speaker does not think of Gödel as the referent of ‘Gödel’ when he knows (R). Nimtz assumes that if the hearer is not thinking of Gödel via this description, then she must have some richer notion associated with Gödel that provides the reference condition. To assume this, though, is to beg the question in favour of descriptivism, and hence to ignore an option which is available to the Kripkean externalist—an option which allows the externalist to say how speakers have a non-descriptive way of thinking about Gödel, which they can deploy in their knowledge of (R). I’ll spell out slowly how one can think about a particular or a kind non-descriptively, taking the whole of the next section to do so. 8. How to think about an object without describing it I’m denying that the only way one can think about an object or kind is via an associated description. For some objects, this claim is not controversial—objects we are acquainted with, in roughly Russell’s sense of acquaintance. Recalling Russell’s dichotomy between knowledge by acquaintance and knowledge by description will be illuminating, since I suspect that the descriptivist thinks of reference in the light of a similar dichotomy: one can either refer by acquaintance or by description. The Kripkean externalist thinks there is another way of referring, though—by being causal-historically hooked onto an object. Being acquainted is one way to be causal-historically hooked onto an object, but not the only way. When I’m acquainted with an object, I have both an epistemic and a semantic advantage with respect to that object; these advantages accrue from my having an open informational channel to it. Suppose I am giving 229

a lecture and a new person sits down in the front row; I am sizing up the newcomer (the paradigm case of acquaintance).8 While I’m confronted with this newcomer, I have him in my field of vision, and this provides an open informational channel to him. I can use the open informational channel to gather information about him; I can also control the information I gather by shifting my attention; and I can store the information I gather about him in a single locus—in the mental file that I’ve opened on the person. This mental file is the concept I have for this person—the concept him; using this concept I can think perceptually-based thoughts about this person (such as that he’s middle-aged), ponder possibilities about him (such as that perhaps he’s a visitor to the department), and reason about him to conclusions that go beyond the perceivable (such as that he is probably the oldest person in the audience). How am I able to think about the person—how does my concept him refer to the newcomer? The answer is that I am causally/informationally connected to him: acquaintance (having him in my sight) confers a semantic advantage. My concept him is like a demonstrative—fired back down the informational channel to its source. A descriptivist might think that the concept him refers descriptively, preferring to say that acquaintance delivers primarily only an epistemic advantage—one first gathers a lot of information about the new person, and then builds a description that uniquely describes him in order to refer to him in thought. This picture gets things the wrong way around, though: one needs already to have enough of an informational ‘hook’ on a person to direct attention and information-gathering, and one needs to already have a way of thinking about him in order to store the diverse information one gathers in a single locus. Suppose one is to use the description ‘the oldest person in the front row’; if one is to build this description, one must direct one’s attention from where the man is sitting to how old he looks and probably is; and one must store the two pieces in the same folder, so that one can combine them into the conjunction used in the description. These two things are achieved by having a referential hook on the person. Forming a uniquely identifying description follows from being able to refer, it does not precede it. In practice a descriptivist would probably not insist that we can only refer descriptively to people who are right in front of our eyes, but it is 8. I’m ignoring Russell’s idiosyncratic view that one can only be acquainted with oneself and one’s sense-data.

230

important to see that our referential ability does not here come from being able to do by acquaintance the epistemic task (of characterizing the person uniquely) that can normally only be done descriptively. In this case (and generally, I shall argue) the semantic advantage precedes and facilitates the epistemic, not vice-versa. Many of the points made above in relation to perception can be repeated for memory. Epistemologists have discussed the epistemic potential of preservative memory to maintain perceptually acquired knowledge beyond the end of a perceptual encounter. Likewise, memory can maintain perceptually acquired reference beyond the end of a perceptual encounter. Being in Princeton allows me to refer to it as here; having been there and remembering it allows me to refer to it as there. Again (as with perception), it is not right to think of memory as facilitating reference via description: it’s not that while I’m confronted with an object, I form a uniquely identifying description of it, which memory preserves so that I can refer in the future. Memory extends the informational channel that exists during perception between an object and a mental folder into the future, so that the mental folder continues to refer back to the object when it’s no longer present. Perhaps the descriptivist would concede that memory-based reference does not go via description; but Nimtz would rightly point out reference extends beyond what we perceive and remember. For the rest of what we refer to, we refer via description, he will insist. Not so. Languages are informational structures, in which words are interfaces between individual speakers’ cognition and the objects in the world to which they refer. Kripke describes how words in current use refer so long as they continue to maintain the chain of use that goes back to the object at the source of the chain. If a name refers to an object in this way, then a speaker can refer to the object too, by tapping into the informational channel that the chain of use of a name maintains—by becoming a user of the name. In picking up the name ‘Gödel’, a speaker opens a mental folder Gödel corresponding to the name, which becomes the locus for storing information about the referent. The folder comes to refer to Gödel not when it contains enough information about him to allow a uniquely specifying description to be built, but by extending the informational channel that exists from Gödel himself to current uses of the name into the cognition of the speaker. Being part of the practice of using the name allows a speaker to refer to Gödel with her concept (mental folder) Gödel, not by description, but by hooking the concept onto the chain that already connects the name to Gödel. 231

9. Conclusion Being part of the practice of using the name ‘Gödel’ allows a speaker to think about Gödel; she can also think about the very word that gave her this ability, and in particular she can think of this word that it refers to Gödel; that is she can think (R): (R) ‘Gödel’ refers to Gödel. (R) is not a trivial thing to think, though anyone who is sufficiently part of the ‘Gödel’ name-using practice to refer to Gödel can think it, and in doing so think it truly. There is nothing wrong with positing tacit knowledge of (R) to anyone who is part of this practice:9 doing so, as Nimtz has shown, allows us to give an explanation in epistemic terms of how hearers can gain testimonial knowledge about Gödel by being told such things as ‘Gödel lived in Princeton’. But as I have described, Kripkean externalists are not excluded from being able to give this explanation. Explaining the epistemic power of communication does not provide a new argument for neo-descriptivism. Neither Jackson nor Nimtz have given an argument that should force one to abandon the Kripkean picture of how terms refer and the externalist picture of how thinkers refer in thought. In response to Jackson, the externalist can point out that he too believes that speakers associate terms with descriptions, and are guided by these descriptions in their reasoning, while denying that these descriptions play a reference-fixing role. In response to Nimtz, the externalist can point out that he too is free to characterize understanding a term as knowing its referent, while denying that knowledge of reference requires employing a description to pick out the referent (even when the referent is an object one isn’t acquainted with). The better arguments remain with Kripke and his externalist followers, I contend.

REFERENCES Jackson, Frank. 1998. “Reference and Description Revisited”. Philosophical Perspectives 12, 201–218. 9. Or rather, tacit knowledge of (R*).

232

Jackson, Frank. 2005. “Ramsey Sentences and Avoiding the sui Generis”. In: Hallvard Lillehammer and D.H. Mellor (eds.). Ramsey’s Legacy. Oxford: Oxford University Press, 123–136. — 2007a. “Reference and Description from the Descriptivists’ Corner”. Philosophical Books 48, 17–26. — 2007b. “On Not Forgetting the Epistemology of Names”. Grazer Philosophische Studien 74, 239–250. — 2010. “Conceptual Analysis for Representationalists”. In this volume, 173–188. Kornblith, Hilary. 2002. Knowledge and Its Place in Nature. Oxford: Oxford University Press. Kripke, Saul. 1980. Naming and Necessity. Oxford: Blackwell. Nimtz, Christian. 2007. A New Rationalism? Intuitions, Modal Knowledge, and the Analysis of Concepts. Unpublished manuscript. (Habilitationsschrift.) — 2010. “Philosophical Thought Experiments as Conceptual Explorations”. In this volume, 189–214. Salmon, Nathan. 2005. Reference and Essence. expanded 2nd edition. Amherst: Prometheus Books. Soames, Scott. 2002. Beyond Rigidity. New York: Oxford University Press. Spicer, Finn. 2008. “Are there any conceptual truths about knowledge?” Proceedings of the Aristotelian Society CVIII(1), 43–60. Stalnaker, Robert. 1999. Context and Content. Oxford: Oxford University Press. Weinberg, Jonathan, Shaun Nichols and Stephen Stich. 2001. “Normativity and Epistemic Intuitions”. Philosophical Topics 29 (1&2), 429–460. Williamson, Timothy. 2006. “Conceptual Truth”. The Aristotelian Society Supplementary Volume 80, 1–41.

233

Grazer Philosophische Studien 81 (2010), 235–257.

FREGE ON CONCEPTUAL AND PROPOSITIONAL ANALYSIS Mark TEXTOR King’s College, London Summary In his Foundations of Arithmetic, Frege aims to extend our a priori arithmetical knowledge by answering the question what a natural number is. He rejects conceptual analysis as a method to acquire a priori knowledge (see section 1). Later he unsuccessfully tried to solve the problems that beset conceptual analysis (see section 2). If these problems remain unsolved, which rational method can he use to extend our a priori knowledge about numbers? I will argue that his fundamental arithmetical insight that numbers belong to concepts is based on the recognition that different sentences express the same thought. In Frege’s philosophy of arithmetic, propositional analysis does the main work. How it can do this work will be discussed in sections 3, 4 and 5. Sections 6 and 7 explore this approach further.

1. Frege on conceptual analysis and the paradox of analysis How can we, in a rational way, arrive at an answer to the question ‘What is a natural number?’ Certainly not by observation or experiment. In his Philosophy of Arithmetic, Husserl proposed to answer this question by an analysis of the sense of ‘natural number’. Frege criticised Husserl by saying: A definition is also incapable of analysing the sense, for the analysed sense is not the original one. In using the word to be explained, I either think clearly everything I think when using the defining expression: we then have the ‘obvious circle’; or the defining expression has a more richly articulated sense, in which case I do not think the same thing in using it as I do in using the word to be explained: the definition is then wrong. (Frege 1894, 199 [319])

In effect, Frege here discovers the paradox of analysis.1 Let us call any 1. Discovery of the paradox of analysis is usually credited to Langford (1942) and Moore (1942) or, some years earlier, to Wisdom (1934).

definition that aims to analyse the sense of a widely used and understood word like ‘number’ analytic. According to Frege, the paradox can be set up as follows: (A1) An analytic definition is correct if, and only if, the definiendum and definiens have the same sense. (A2) A and B are synonymous & (S grasps the sense of A & S grasps the sense of B) o S immediately knows that A and B have the same sense. Therefore: (K) An analytic definition is either trivial or false. The paradox arises for analytic definitions of general terms. It also arises for a kind of analysis that is sometimes called ‘propositional analysis’. Russell’s theory of definite descriptions is an example of propositional analysis. Russell held that definite descriptions are ‘incomplete signs’: unlike a general term, a definite description cannot be defined; one can only provide a paraphrase of a sentence containing a definite description that eliminates the definite description. If the paraphrase is supposed to preserve the sense of the original sentence and to clarify it, the paradox of analysis also arises for propositional analysis. I will come back to this point in the second half of the paper. Frege’s first premise (A1) is independently plausible: in an analytic definition, one doesn’t want to give a new sense to a word that is already in use. Loosely speaking, one aims to clarify the sense it already has. Frege’s second premise is essential to his theory of sense and reference. His criterion of thought identity assumes that one cannot grasp the sense of synonymous expressions without knowing that they are synonymous: Now two sentences A and B can stand in such a relation that anyone who recognizes the content of A as true must thereby recognize the content of B as true and, conversely, that anyone who accepts the content of B must straightaway accept that of A. (Equipollence). It is here being assumed that there is no difficulty in grasping the content of A and B. (Frege 1906, 197 [213])

If A and B have the same sense and one could grasp the sense of both without recognising their sameness, one could rationally have different attitudes towards the thoughts expressed. For example, I could assent to ‘John is bachelor’ and reject ‘John is an unmarried man’ without thereby becoming liable to criticism. If this were the case, the possibility of having different attitudes towards what A and B say would no longer track 236

differences in the thought expressed. (Equipollence) therefore presupposes that sense-identity is transparent. The paradox of analysis poses a problem not only for Husserl, but also for Frege himself. He tried to show that every truth of arithmetic is provable from basic laws of logic and suitable definitions of arithmetical signs in terms of logical ones. His definitions of arithmetical terms are informative. For instance, he defines ‘0’ as the extension of the concept equinumerous with the concept [z[’. (Frege 1884, § 72) This will come to most of us as interesting news. Hence, Frege’s definitions are either not analytic definitions or they are wrong and consequently, his Logicism is false. In his review of Husserl’s Philosophy of Arithmetic, Frege responds to the paradox by concluding that one cannot give an analytic definition, but that this is no reason for a mathematician to be disheartened: For the mathematician it is no more right and no more wrong to define a conic as the line of intersection of a plane with the surface of a circular cone than to define it as a plane curve with an equation of the second degree in parallel coordinates. His choice of one or other of these expressions (…) is made irrespective of the fact that the expressions have neither the same sense nor evoke the same ideas. (Frege 1894, 200 [321]. My emphasis.)

The paradox shows that one cannot give analytic definitions. But mathematicians are not interested in analytic definitions anyway. A mathematical definition is correct if, and only if, its definiens and definiendum are co-extensional. For example, ‘[ is a conic’ and ‘[ is a plane curve with an equation of the second degree in parallel coordinates’ have the same extension, but differ in sense or mode of presentation. The difference in sense explains why a mathematical definition can extend our knowledge. But since many non-synonymous predicates have the same extension, there are different correct mathematical definitions of the same definiendum. Frege does not mind this. A mathematical definition is not intended to decompose a concept into its constituents. Hence, it cannot be faulted for not achieving this aim. A mathematical definition provides us with a concept that enables us to give proofs. A truth that seemed unprovable may be proved only after one of its constituent concepts has been analysed.2 For example, we can only prove the axioms of arithmetic from logical laws if we have defined the numbers in terms of the concepts that figure in the laws of logic. 2. See Frege (1914), 209 [226].

237

But suppose that you have defined ‘x is the number of Fs =df. x is the extension of the concept equinumerous with F’. Even if definiendum and definiens are co-extensional, and the definiens makes new proofs possible, the definition will not help us to prove the axioms of arithmetic from the laws of logic. Arithmetic is a body of true thoughts. These thoughts are individuated by (Equipollence). According to (Equipollence), thoughts that contain the sense of ‘x is the extension of the concept equinumerous with F’ instead of ‘x is the number of Fs’ are different. Hence, Frege will have proved some truths from the laws of logic, but not the truths of arithmetic.3 Mathematical definitions don’t help his cause. A radical response to this problem is to say that arithmetical terms had no sense before Frege’s definitions provided one for them. We believe that there are general truths about numbers and that we know some of them. But we are wrong: Our everyday view that this sentence [‘0 is not equal to 1’] expresses a truth is not quite right. The content associated with it is not precise enough; the science is not yet sufficiently well worked out. (Weiner 2007, 705)

Weiner argues that signs like ‘number’, ‘0’, etc. have no public sense. Hence, these words can be given a sense by mathematical definitions in the first place. As an interpretation of Frege, this strikes me as wrong. For example, at the beginning of Conceptual Notation, as well as in Foundations of Arithmetic, he argues that we know many arithmetical truths, but are unclear about their ultimate truth-grounds. How could Frege appeal to this knowledge of truths like the truth that 1 + 2 = 3 if the words ‘1’ etc. did not even have a sense? If there are no truths about numbers and no knowledge about numbers, but only false meta-linguistic beliefs, how should we ever arrive at an analysis? Weiner maintains that we hold true certain sentences and that the development of science should make them come out true. But why should science make a false meta-linguistic belief come out true? How can taking a sentence to be true that is in fact senseless have any epistemic authority at all?

3. See Nelson (2008), 161.

238

2. Unclearly grasping a concept If Frege’s definitions were mathematical ones, Logicism would become philosophically uninteresting. Therefore Frege has a good reason to look for a solution of the paradox of analysis that allows Logicism to be an interesting thesis. In his manuscript ‘Logic in Mathematics’ (1914), analytic definitions are described as pairing a simple and a complex sign that have the same sense. When does the complex sign ‘bc’ have the same sense as the simple sign ‘a’ that has been in use before the analysis? Answer: I believe that we shall only be able to assert that it does when this is selfevident [unmittelbar einleuchtet]. And then what we have is an axiom. (Frege 1914, 210 [227])

Hence, every analytic definition must be trivial and therefore not worth knowing. However, Frege does not draw this conclusion. He argues that definiendum and definiens can be synonymous and yet one can still rationally doubt that they are if we do not have a clear grasp of the sense of the simple sign, but that its outlines are confused as if we saw it through a mist. The effect of the logical analysis will then exactly be this – to articulate the sense clearly. (Frege 1914, 211 [228])

Although definiens and definiendum have the same sense, someone who grasps both can nonetheless extend his knowledge if he has not clearly grasped the sense of the definiendum. (A2) is thus modified to: (A2*) If two expressions A and B have the same sense and one grasps the sense of A clearly and one grasps the sense of B clearly, one thereby knows that A and B have the same sense. If one does not grasp a sense clearly, one only attains such a grasp after the analysis is completed. Conceptual analysis improves our grasp of the sense of a concept-word: the analysis has psychological, but no logical importance. So far Frege’s attempt to work around the paradox is incomplete. ‘Unclear grasp’ of a concept is a suggestive metaphor. But in order for ‘unclear grasp’ to be more than a new name for the paradox of analysis, we need to provide an independently motivated answer to the question ‘What is it to grasp a concept unclearly?’ Frege answers:

239

We simply do not have the mental capacity to hold before our minds a very complex logical structure so that it is equally clear to us in every detail. For instance, what man, when he uses the word ‘integral’ in proof, ever has clearly before him everything that appertains to the sense of the word! And yet we can still draw correct inferences, even though in doing so there is always a part of the sense in penumbra. Weierstrass has a sound inkling (Ahnung) of what number is and working from this he constantly revises and adds to what should really follow from his official definitions. (Frege 1914, 222 [240]. In part my translation)

In essence: Thinker T grasps concept C unclearly if C is complex & T grasps C, but not every part of C. How is it possible that some parts of a sense are not grasped when one grasps it? Frege had previously argued that we often use words without grasping their sense when we use them. What we are conscious of when we reason is not the sense of, say, ‘integral’, but the word itself.4 We use the word, thereby relying on our ability to recall its sense when necessary. The view that reasoning, especially mathematical reasoning, is conducted by mentally manipulating signs while ‘ignoring’ their meaning is part of the motivation for introducing a calculus.5 Now it might indeed be an insight that a simple and a complex sign have the same sense when one has used a simple sign without being conscious of its complex sense. However, the knowledge-extending character of conceptual analysis cannot be explained, in general, by saying that one has not all the constituents of a complex concept in mind when it is expressed by a simple sign. Take the analysis of ‘x wagers’ as ‘x pledges something on a possible event’. The sense of ‘to wager’ is not so complicated that one could not use the expression in full awareness of what it means. Hence, the definition should be trivial. However, it is not. It takes some time to work it out. Can we improve Frege’s proposal and give an independently motivated explanation of unclearly grasping a concept that avoids the problem of his psychological explanation? Burge and more recently Nelson have tried to help Frege out and develop a better understanding of what unclearly grasping a sense is. Let’s see whether they are successful. 4. For further discussion see Horty (1993) and May (2006), 129ff. 5. See Whitehead (1898), 3f. on substitutive signs.

240

The basis of Burge’s account is his social model of partial understanding (see Burge 1979). In outline, he argues that one comes to grasp the sense of a general term by being inducted into a social practice of using it. One can count as a member of the practice of using a term and yet make systematic mistakes in its explanation and application. For example, you count as a partially competent member of the practice of using ‘arthritis’ in virtue of deferring to experts about the matter. You may explain ‘arthritis’ incorrectly, but as long as you defer to the right people we have reason to say that you explain the right concept in the wrong way.6 In ‘Frege on Sense and Linguistic Meaning’, Burge argues that there is a public sense of ‘number’ that Weierstrass and other mathematicians grasp, although they all explain ‘number’ differently and incorrectly. Prima facie, Weierstrass and Co. cannot defer to the experts about arithmetic; they are the experts. How can we then say that they have a partial understanding of the arithmetical vocabulary? Answer: we grasp the sense of ‘number’ etc. completely only if we know the ideal arithmetical theory: The senses of expressions could be fully grasped only by grasping equivalences given by ideal scientific explications, or by otherwise understanding the contribution of those expressions to a theory. Of course, since most of the envisioned ideal explications would come as discoveries, it is possible to doubt them (even if the doubt depends on less than full analytic mastery of the senses), while not doubting the corresponding self-identities. So by Frege’s test for the identity of senses, the senses of the explicans and explicandum would be different. (Burge 2005, 265, Fn. 16. My emphasis)

Burge’s account of (un)clearly grasping a concept cannot solve the paradox of analysis, though. For, as Burge himself makes clear, by Frege’s own lights, definiendum and ideal definiens differ in sense. Hence, the ideal definition is either not analytic or wrong. We have made no progress. Nelson (2008) has tried to help Frege in a different way. He proposes to add what he calls ‘conceptions’ to Frege’s theory: We grasp senses through what we can call conceptions. Conceptions, we can take it, are individuated by the agent’s current best explications of the sense grasped. More specially, they are individuated functionally. Conceptions are holistic, defined in terms of inter-relations to an agent’s other conceptions. For simplicity, we can simply say that conceptions (but not senses!) are indi6. A crucial part of Burge’s story is that we are right in taking people to say what their words mean in the public language, not what they mean in their idiolect. Whether this assumption is correct may be doubted with good reason. See Davidson (1987), 449.

241

viduated precisely in terms of acceptance patterns, just as ACCEPT would have senses individuated. (Nelson 2008, 165)

Nelson’s ACCEPT is Frege’s (Equipollence) under another name. In effect, Nelson uses Frege’s criterion to individuate conceptions. Conceptions and senses are supposed to be structured entities. If you grasp the sense of a word through a conception, and the parts of the conception don’t match the parts of the sense, your grasp of the sense is unclear. Nelson’s outline would help Frege only if we had a criterion to determine when the parts of a conception and the sense constituents match that does not rely on the notion of conceptual analysis. If we can only say that the structure of a sense and the conception through which we apprehend it do not match when the conception is not identical with the conception that contains all the parts that a successful analysis of the sense would specify, we have not solved the paradox of analysis, but only re-described it in a new terminology. What is needed is a plausible understanding of when a sense matches a conception that is independently motivated and explained. Nelson does not provide the required understanding of match. In order to do so, we would need to develop a criterion of sense identity different from (Equipollence). For, in order to say that the structure of a conception and the structure of a sense do not match, we need to be able to determine what the structure of a sense is independently of its correct analysis. Nelson’s account does not provide such a criterion, and therefore remains incomplete. 3. From conceptual analysis to the decomposition of thoughts Frege has no satisfactory account of conceptual analysis. I now want to suggest that he does not need one for his mathematico-philosophical project. Knowledge-extending definitions are not arrived at by conceptual analysis, but by decomposing thoughts. For example, Frege aims in his Foundations of Arithmetic to develop a scientific understanding of number by starting from judgements that answer the question ‘How many?’ (‘How many BMWs do you own? I have 2 BMWs’): In order to shed light on the matter [what is number?] it will be helpful to consider number in the context of a judgement in which its original application is prominent. (Frege 1884, 59. My translation)

242

This starting point of Frege’s investigation will concern us further in section 5. Let us notice here that his shift from concept-words to sentences that can be used to express a particular kind of judgement is in line with one of three ‘fundamental principles’ of the Foundations of Arithmetic, the context-principle: Never ask for the meaning of a word in isolation, but only in the context of a proposition. (Frege 1884, xxii)

But how can focussing on propositional analysis help with the paradox of analysis? After all, the paradox arises for concept-words and for sentences. How can finding out that two sentences express the same sense extend my knowledge? Frege has an answer to this question that does not assume that everyone who uses a word may grasp the sense expressed only incompletely. He takes his answer to apply the general role of recognition judgements in science to logic: There is no doubt that the first and most important discoveries in science are often a matter of recognizing something as the same again. (…) It is just as important not to distinguish what is the same as it is to be alive to differences when they don’t hit the eye. (Frege 1897, 141f. [153])

The recognition that the same sun rises every morning is of fundamental importance for astronomy. This is not different for logic, the science of the laws of truth. The acts of recognition that are fundamental in logic are judgements to the effect that P is the same thought as Q. According to Frege, such recognitions can be genuine discoveries: The first and most important task is to set out the object of investigation in pure form. Only by means of this is one able to make the acts of recognition that probably constitute the fundamental discoveries in logic, too. Therefore let us never forget that two different sentences can express the same thought, that we are only concerned with only that part of the sentence’s content that can be true or false. (Frege 1897, 143 [154]. My translation and emphasis.)

Frege here uses an analogy between chemistry and logic. He elaborates it in the following way: The distinction between what is part of the thought expressed in a sentence and what only gets attached to the thought is of the greatest importance for logic. The purity of the object of one’s investigation is not of importance only to the chemist. How would the chemist be able to recognize, beyond any

243

doubt, that he has arrived at the same result by different means, if the apparent difference of means could be traced back to impurities in the substances used? (Frege 1897, 141f. [154])

In chemistry, two samples may be samples of the same substance, but because they contain impurities, we can’t recognise the sameness in substance—it is masked by the addition of further substances. In logic, two sentences s and s* may express the same thought, but because they do something else, too, one can grasp the thought expressed by s and s* and doubt that they express the same thought. The analogy suggests that recognizing that different sentences express the same thought can extend our knowledge in that we come to know something we did not know before. It can be a genuine discovery because sense-identity is not transparent when thoughts are presented in an impure form. In order to develop this idea, we need to explain which impurities make sense-identity intransparent. The next section will answer this question. 4. Modes of apprehending a thought Sentences of Frege’s Conceptual Notation have a content, a content being a state of affairs. In his Conceptual Notation, Frege argues that such a content can be apprehended in different modes. For example, the distinction between function- and argument-expressions is a matter of apprehending one and the same content in different modes or ways.7 Let us consider Frege’s own example. We can take the sentence: Hydrogen is lighter than carbon dioxide as consisting of the argument-expression ‘hydrogen’, completing the function-expression ‘[ is lighter than carbon dioxide’, but we might equally say that the sentence consist of the argument-expressions ‘hydrogen’ and ‘carbon dioxide’ jointly completing ‘[ is lighter than ]’. The intuitive basis for holding that there are different decompositions of the same sentence that do not bear on the content expressed is that the decompositions are a matter of what linguists call ‘focus’ (see Rooth 1996). If Frege’s example 7. See Frege (1879), 15. Things get complicated if we take general sentences into account. I will ignore these complications for ease of exposition.

244

sentence is uttered as an answer to the question ‘Which gas is lighter than carbon dioxide?’, the singular term ‘hydrogen’ will be in focus, it is the prominent expression in the sentence. If the same sentence with the same meaning is uttered as an answer to the question ‘Which gas is lighter than which other gas?’, both ‘hydrogen’ and ‘carbon dioxide’ will be in focus. The expression(s) in focus is (are) the argument-expression(s), the sentenceremainder in the ‘background’ is the function-expression. Changing focus changes the decomposition of a sentence, but it does not change the content expressed by a sentence. Imagine that I said ‘I had [7]F pints tonight’ (where the subscript indicates the expression in focus) in order to contrast my beer intake today with my usual drinking of 5 pints per sitting. Suppose you understood every word I said when I said ‘I had [7]F pints tonight’, but you did not get the intended contrast. In order to get the message and its articulation across, I can use a transformation of my original utterance in which ‘7’ is moved (see Hofweber 2007, 147): The number of pints I had tonight was 7. I had [7]F pints tonight. Both sentences express the same thought. The only difference between them is that they put the same expression in focus in different ways. After the introduction of the sense/reference distinction, the important part of Frege’s idea remains in place. In ‘On Concept and Object’ Frege writes: [W]e must not fail to recognize that the same sense, the same thought, may be variously expressed; thus the difference does not here concern the sense, but only the apprehension (Auffassung), shading or colouring of the thought, and is irrelevant for logic. […] If all transformations of the expression were forbidden on the pleas that this would alter the content as well, logic would simply be crippled; for the task of logic can hardly be performed without trying to recognize the thought in its manifold guises. Moreover, all definitions would then have to be rejected as false. (Frege 1892, 185 [196])

Two sentences can express the same sense, but can differ in ‘apprehension (Auffassung), shading or colouring’. Let us set aside shading and colouring and concentrate on apprehension.8 In the same paper, Frege illus8. The view that definiens and definiendum differ in colouring can be found before Frege. The term ‘colouring’ is derived from the illumination of manuscripts or maps by colouring parts of them. In his Jaesche Logic Kant takes analysis to be the unclouding (‘Aufhellung’) of a concept

245

trates an important difference in apprehending a thought in the following way: But we must never forget that different sentences may express the same thought. For example, the thought we are considering [i.e., the thought expressed by ‘There is at least one square root of 4’] could also be taken as saying something about the number 4: ‘The number 4 has the property that there is something of which it is the square.’ Language has the means of presenting now one, now another, part of the thought as the subject; one of the most familiar is the distinction between active and passive forms. It is thus not impossible that one way of analysing a given thought should make it appear as a singular judgement, another, as a particular judgement; and a third, as a universal judgement. (Frege 1892, 188f. [199f.])

The thought expressed by ‘Everyone loves [Madonna]F’ is singular, the thought expressed by ‘[Everyone]F loves Madonna’ is general. Whether a thought is singular or general is a matter of our apprehension. These considerations about modes of apprehension help to answer our question ‘What are the impurities that make sense-identity intransparent?’ Thinkers like us grasp a thought as the sense of a sentence (see Frege 1924/5, 269 [288]). In putting a thought into the guise of a sentence, we decompose it into parts. Decomposing a thought has the same effect as adding an impurity to a chemical substance. It makes it a genuine task to establish that the thought in different decompositions is the same. For the decomposition of a thought given by a particular wording can be a misleading defeater of our belief that the same sense is expressed. For example, someone might perfectly grasp the thought expressed by ‘The number of pints I had tonight was 7’ and ‘I had 7 pints tonight’ and yet believe that they express different thoughts. Why? Because he thinks that ‘the number of ’ does not only create a focus effect but introduces reference to numbers, which is not present in ‘I had 7 pints tonight’. The misleading defeater needs to be explained away before one will be justified again in one’s belief that the same thought is expressed. The sentences s and s* seem to say different things, but this impression can be explained away and compares it to the illumination of a map. As illuminating a map does not add to the map, analysis does not add to the concept, see Kant (1800), 95. And one might continue: as illuminating a region on a map puts it into focus, a propositional analysis focuses on parts of a thought.

246

by appealing to syntactic or other features that create focus effects. If after the explanation the sentences strike us as satisfying (Equipollence), we can take the sentences to express the same thought. Frege remarks at the beginning of Foundations of Arithmetic: Often it is only through great intellectual labour, which can continue over centuries, that a concept is known in its purity, and stripped of foreign covering that hid it from the eye of the intellect. (Frege 1884, vii. My emphasis)

Thus Frege’s metaphor of ‘stripping-away’ becomes intelligible in the light of the considerations of this section. Sometimes, recognising the same thought in different linguistic guises involves two abilities—our ability to grasp thoughts expressed in a language and our ability to understand how intonation and syntactic structure affect focus. Is the second ability, or the knowledge that underlies it, a priori? If I know what ‘A’, ‘and’ and ‘B’ mean, I know without further empirical information what ‘A and B’ means. Is this empirical knowledge? No, it is knowledge we can acquire by exercising a special cognitive faculty without relying on perceptual knowledge in a justificatory role. If we classify this linguistic knowledge as a priori, we should also classify knowledge of focus as a priori. It seems plausible to take our knowledge of how focus works to be part of our general knowledge of language: A competent speaker of a language will have knowledge not only of which sentences are well formed, what the words of the language mean, and how the meaning of the simple parts determine the meaning of the complex parts, but also of how the intonation of a sentence affects focus, among other things, and how the syntactic structure of a sentence affects focus. (Hofweber 2007, 17)

Hence, recognising that the sentence s and s* apprehend the same thought is an a priori achievement. Let us compare and contrast the Fregean view of propositional analysis with the view of conceptual analysis attributed to Frege by Burge and Nelson. Burge accepts (Equipollence) and argues that according to Frege, the ideal arithmetic will contain the full explication of those arithmetical concepts we grasp incompletely. The combination of these theses is inconsistent. Nelson gives up on (Equipollence) as a criterion for the individuation of senses and takes it only to individuate the conceptions we have. Since he does not provide us with a criterion for the individuation of senses, his account of analysis is incomplete. 247

In contrast, Frege clearly appeals to the provision that (Equipollence) only applies to sentences whose sense can be grasped without a problem. Please read with me again the quote from section 1: Now two sentences A and B can stand in such a relation that anyone who recognizes the content of A as true must thereby recognize the content of B as true and, conversely, that anyone who accepts the content of B must straightway accept that of A. (Equipollence). It is here being assumed that there is no difficulty in grasping the content of A and B. (Frege 1906, 197 [213]. My emphasis)

(Equipollence) individuates the senses of sentences only if they can be grasped without a problem. But in many cases we have difficulties to grasp the content of A and B. (Equipollence) is only applicable to these sentences if the misleading indications of sense difference have been explained away. Frege does not need a new criterion of thought identity; he only needs to add a filter to (Equipollence) to do away with misleading appearances of thought-difference. 5. How ‘re-focussing’ yields philosophical insight We now have an answer to the question how finding out whether two sentences express the same thought can be a genuine discovery even for someone who grasps both thoughts. But does the explanation apply to Frege’s arithmetical investigations? It can only apply if focus plays an important role in these investigations. If a sentence is uttered as an answer to a question, the question determines which sentence-constituents are in focus. We have seen in section 3 that ‘How many?’ questions play a fundamental role in Foundations of Arithmetic. Frege introduces whole positive numbers as those things specified in suitably completed ‘How many?’ questions: The number answers the question how many [Die Zahl antwortet auf die Frage wieviel?] (Frege 1884, 44. Strictly speaking, it must be ‘wieviele’. See Frege 1884, 5, Fn. 1)

A number-statement (Anzahlaussage) is an answer to a suitably completed ‘How many?’ question. In such an answer, one will designate a whole number, often by using a numeral. Frege goes on to say:

248

In order to shed light on the matter [what is number?] it will be helpful to consider number in the context of a judgement in which its original application is prominent. (Frege 1884, 59. My translation)

Frege considers judgements that are responses to ‘How many?’ questions. Let us see which fruit this approach to the nature of number yields. Consider Frege’s own example, the sentence ‘The emperor’s carriage is drawn by four horses’. If it is uttered as an answer to the question: Who or what is drawn by four horses? ‘the emperor’s carriage’ must be in focus: [The emperor’s carriage]F is drawn by four horses. But if we use the sentence to answer a ‘Who or what’ question, the statement is no number-statement at all. If the same sentence is uttered as an answer to the question ‘How many horses draw the emperor’s carriage?’, it is a number-statement in which the number 4 has its original application: The emperor’s carriage is drawn by [four]F horses. The numeral is in focus and the number-statement contrasts with: The emperor’s carriage is drawn by five horses. The emperor’s carriage is drawn by three horses. … In a number-statement, the numeral is the argument-expression, while the rest of the sentence (‘The emperor’s carriage is drawn by ___ horses’) is the function-expression. A syntactic focus construction moves the numeral or number word to the end of the sentence: The number of horses that draw the emperor’s carriage is four. This construction has the same effect as intonational stress on the number designator. Frege uses this construction in Foundations of Arithmetic, but he interprets it wrongly as an identity statement (Frege 1884, 69. For critical discussion see Rumfitt 2001, 49). 249

We can express the result of Frege’s investigation of number-statements in a neutral terminology as follows: number-statements involve reference to numbers and put them in a relation to concepts, i.e. the referents of function-expressions. In order to see that we have made progress, assume that in theorising about number-statements you are guided by the subjectpredicate analysis of these statements. Then you will see the number as qualifying the object the grammatical subject stands for. For example, philosophers of arithmetic who take collections or manifolds to be numbered by the natural numbers misconstrue the focus in statements of the form ‘The so-and-so’s are n in number’. They go with grammar to determine what the statement is about, namely what is named by ‘the so-and-so’s’. But this is a mistake; so construed, we have no number-statement at all. Frege has exposed the problems of this view so convincingly that few take it now seriously. The cognitive effect of decomposing the thought expressed by ‘The emperor’s carriage is drawn by four horses’ differently is similar to a visual aspect change. When you see the duck/rabbit figure first as a rabbit and then, by switching your attention, you come to see the figure as a duck, you see something new, although the figure perceived does not change at all. What does change is which constituents of the picture are in the foreground, which in the background. A similar mechanism is in play when we come to realise that a statement of number is not ascribing a property to a plurality, but is a first-order concept represented as falling into a second-order one. The thought remains the same, the decomposition or mode of apprehension changes. If we decompose the thought differently, we can draw new inferences from it. Thereby we recognise what follows from what. In the new decomposition the thought can make contact with further knowledge that we possess. This is certainly an extension of our logical knowledge. Hence, recognising that different decompositions are decompositions of the same thought can extend our knowledge. So far, we have successfully applied the idea that one can make cognitive progress by recognising the same thought in different guises to Frege’s investigations. However, not all is well. Frege himself does not use the neutral terminology I have suggested when he states his result. For example, he says that his view about number-statements is ‘that the content of a statement of number contains a predication (Aussage) of a concept’ (Frege 1884, 59):

250

When I say ‘The emperor’s carriage is drawn by four horses’, I assign the number 4 to the concept ‘horse that draws the emperor’s carriage’. (Frege 1884, 59)

In Basic Laws, he will say that his whole theory of arithmetic rests on this view of number-statements (Frege 1893, IX). Although this is still some way from a definition of number and individual numbers, it is this view that motivates his definition of numbers in terms of logical objects that belong to concepts, their extensions. Now Frege’s view of number-statements as predications of concepts is not in harmony with the idea that these statements are answers to ‘How many?’ questions. In a number-statement, the number-designator, not the function- or concept-word is in focus. The number-designator designates the object of which something is predicated; the number is not something that is ascribed to a concept in such a statement. Compare: One answers the question ‘Of which type or kind of thing are there 4?’ correctly by saying ‘There are 4 [horses that draw the emperor’s carriage]F’. Here we have a predication of a concept, but not a number-statement, that is, an answer to the question ‘How many?’. I will come back to this point in due course. Frege is aware of this problem. He argues that the number 4 is not something that can be predicated of the concept The emperor’s carriage is drawn by [ horses. He claims that he has avoided taking numbers to be properties of concepts (Frege 1884, 68). But how can he avoid taking numbers to be properties of concepts if he says that in a number-statement something is predicated of a concept? How can he say that the concept is the true bearer of number (‘bearer’ in the sense of ‘property bearer’) if numbers are not properties of concepts (Frege 1884, 61)? Frege responds to this problem in his late work: Since a statement of number based on counting contains a predication of a concept, in a logically perfect language a sentence used to make such statement must contain two parts, first a sign for the concept of which the predication of number is made, and secondly a sign for a second level concept. These second level concepts form a series and there is a rule in accordance with which, if one of these concepts is given, we can specify the next. But still we do not have in them the numbers of arithmetic; we do not have objects, but concepts. (Frege 1919, 256f. [277]. My emphasis)

In English, the statement ‘The emperor’s carriage is drawn by [four]F horses’ is not a predication of a concept. There is no way around this. But in an ideal language, the numeral ‘4’ would be replaced by a second-level 251

concept word ‘There are 4 f’ which is predicated to the first-level concept horse drawing the emperor’s carriage. In the ideal language, a statement of number represents a first-level concept as falling into a second-level concept. In the reformed language one would answer ‘How many horses draw the emperor’s carriage?’ by asserting ‘[There are four]F horses that draw the emperor’s carriage.’ Now it is difficult to see why ‘There are four horses that draw the emperor’s carriage’ should be a sentence in an ideal language. In fact, Frege proposes a new starting point for arithmetical investigations. Instead of starting from number-statements, we should start with predications of concepts, in which numbers are no longer mentioned. This change of tack does indeed give him statements in which something is predicated of concepts. The price to pay is that these statements are no longer number-statements. For they don’t answer the question ‘How many?’ Frege’s attempt to defend the thesis that number-statements are predications of concepts has resulted in the refutation of this thesis. An outline of a different response goes as follows. If we want to develop the proposed decomposition of number-statements further, we should not assimilate them to predicative statements. Frege holds that extensions are objects that stand in a sui generis relation to concepts. A concept has an extension, but the extension is not a property of a concept. Why should he then not allow numbers to be objects that stand in a sui generis relation to concepts? After all, numbers will turn out to be extensions. A numberstatement is not a predication of a concept, but in the statement a number is put into a relation to a concept. 6. Analysis and the argument for sense/reference distinction The previous section has provided a good reason for Burge’s claim that especially in cases where the sense of an expression has logical structure, Frege thought that there is room for informativeness or scientific value in expressions of the same thoughts in different ways. (Burge 2005, 147)

Apprehending one and the same thought differently can lead to scientific discoveries. An essential premise of Frege’s argument for the sense/reference distinction in ‘On Sense and Reference’ is that difference in cognitive value implies difference in sense. Now we seem to have undermined this premise. For we have discovered an alternative source of differences in 252

cognitive value: difference in mode of apprehension. For example, why can’t we explain the difference in cognitive value between ‘Hesperus is the same planet as Hesperus’ and ‘Hesperus is the same planet as Phosphorus’ as a difference in mode of apprehension of the same content? If we can do so, Frege’s main argument for the introduction of the sense/reference distinction is blocked. How can one protect Frege’s crucial premise? Nelson answers that one can’t: What I doubt is that one can properly justify the claim that explaining these acceptance patterns requires a difference in sense without undermining the ‘unclouding’ view in the general process. To do so, one must provide a reason why the general sort of explanation for the differences in cognitive value in conceptual analysis sentences a proponent of the ‘unclouding view’ offers should not be extended to the differences in cognitive value in sentences with co-designating proper names. (Nelson 2008, 145)

On the view of propositional analysis proposed here, the question becomes the following: ‘Why should one not explain the difference between ‘Hesperus is the same planet as Hesperus’ and ‘Hesperus is the same planet as Phosphorus’ in terms of focus?’ For instance, if an utterance of ‘Hesperus is Hesperus’ is an answer to ‘Who is Hesperus?’, the second occurrence of ‘Hesperus’ must supply new information. However, if the same proper name is re-used, it can’t impart new information. By contrast, ‘Phosphorus’ can be in focus in the answer ‘Hesperus is Phosphorus’. However, such an explanation of the difference in cognitive value does not make the notion of sense superfluous; it only hides an appeal to it in the guise of ‘re-use of the same proper name’. Neither sameness of geometrical shape nor sameness of referent is sufficient for two physical objects to qualify as the same proper name. Frege argues that sameness of shape and mode of presentation are both required. A statement of the form ‘Hesperus is the same planet as Hesperus’ is informative if the two tokens of ‘Hesperus’ express different modes of presentation. The different modes of presentation may even determine the same object. There is further limitation to the application of the different apprehension/same thought strategy. I have argued that ‘The emperor’s carriage is drawn by four horses’ and ‘The number of horses that draw the emperor’s carriage is four’ are decompositions of the same thought. Frege wants to apply this description to further groups of sentences. The most important is

253

Basic Law V: (x) (Fx l Gx iff the extension of F = the extension of G) Frege tries to justify Basic Law V by arguing that the two sides of the bi-conditional are different decompositions of the same thought. However, the right hand side is not a focus construction. Hence, the theory of focus justifies some of Frege’s claims about thought sameness, but not others. Russell’s discovery of the set-theoretic paradox showed Basic Law V to be necessarily false. The proposed account of the informativity of propositional analysis does only justify those re-carvings of thoughts that are trouble-free. At the same time, we learn that one can make a mistake about what is a decomposition of a thought and what is not. 7. Is apprehending the same thought differently ontologically creative? Logicism consists of an ontological thesis (there are numbers and they are objects), an epistemological thesis (we can apprehend numbers independently of experience) and a thesis about truth-grounds (the laws of logic are the truth-grounds of the laws of arithmetic). Frege held and argued for all three theses. In decomposing a number-statement in a certain way, one discovers that it is about numbers and concepts and learns how to prove it from the laws of logic. One thereby has extended one’s a priori knowledge. To illustrate Frege’s idea, consider again Frege’s example. While (N–) The emperor’s carriage is drawn by four horses does not mention numbers, (N+) The number of horses that draw the emperor’s carriage is four apparently does. However, (N+) and (N–) shall express the same thought. How can this be? Neo-Fregeans try to turn this mystery into an answer to the question how we apprehend numbers. But if (N+) is a transformation of (N–) in which ‘four’ is moved to create a focus effect, how can it bring new ontological commitments with it? If I put an expression in focus in a sentence, I don’t create a new ontological commitment. How could focus create such commitments?

254

However, in (N+) the expression in focus completes a predicate and occupies a position that is accessible to a quantifier. Isn’t that saying that we have a new ontological commitment here? No. Take an independent example to make this clearer. You ask me: When did it rain today? I answer: It rained today [at four]F. The sentence is decomposed into ‘at four’ as argument-expression and the function word ‘It rained today ( )’. I can quantify into the argumentexpression position: It rained today sometime. But this does not commit me to days or times as new kind of objects. Why should then syntactic focus as in It was 4 when it rained today, do so? The same point is implicit in Frege’s own work. According to Frege, the focus construction ‘It is true that seawater is salty’ expresses the same thought as ‘Seawater is salty’. However, Frege insists that there is no property of truth involved in the utterance of the first sentence. If this is so, why should one treat ‘The number of horses that draw the Emperor’s carriage is four’ and ‘The Emperor’s carriage is drawn by 4 horses’ differently? In Foundations of Arithmetic, Frege argued that decomposing certain thoughts in a particular way gives one a reason to hold that numbers are objects and also points one to the way in which we come to apprehend numbers. Both these claims turn out to be controversial by Frege’s own lights. If thoughts are, as Frege stresses, multiply decomposable, a thought itself does not commit us to the existence of a particular object or kind thereof. Why should one decomposition of a thought be ontologically more revealing than another? If we take Frege’s view of multiple decomposability to its logical conclusion, thoughts are ontologically neutral.

255

In order to argue for the ontological and the epistemological theses that define Logicism, Frege and Fregeans need to appeal to further resources. But new decompositions of thoughts we already know allow us to see new logical connections between thoughts. Hence, Frege’s notion of analysis enables him to make progress in identifying the truth-grounds of number statements and the laws of arithmetic.9

REFERENCES Burge, Tyler. 1979. “Individualism and the Mental”. Midwest Studies in Philosophy 4, 73–121. — 2005. Truth, Thought, Reason: Essays on Frege. Oxford: Clarendon Press 2005. Davidson, Donald. 1987. “Knowing One’s Own Mind”. Proceedings and Addresses of the American Philosophical Association 60, 441–458. Frege, Gottlob. 1879. Conceptual Notation. In: Conceptual Notation and Related Articles. Oxford: Clarendon Press 1972. (Begriffsschrift und andere Aufsätze. Neudruck Darmstadt: Wissenschaftliche Buchgesellschaft 1977.) — 1884. The Foundations of Arithmetic. Oxford: Basil Blackwell 1974. (Die Grundlagen der Arithmetik. Christian Thiel (ed.). 1988. Hamburg: Meiner) — 1893. Grundgesetze der Arithmetik, I. Reprint Hildesheim: Georg Olms Verlagsbuchhandlung 1962. — 1894. “Review of E.G. Husserl, Philosophie der Arithmetik.” In: Collected Papers on Mathematics, Logic and Philosophy. Oxford: Blackwell 1984, 182–194. — 1897. “Logic”. In: Posthumous Writings. Oxford: Blackwell 1979, 126–151. (Nachgelassene Schriften, H. Hermes, F. Kambartel and F. Kaulbach (eds.). 1983. Hamburg: Meiner 137–163) — 1906. “A brief Survey of my Logical Doctrines”. In: Posthumous Writings. 197–202. (Nachgelassene Schriften, 213–218) — 1914. “Logic in Mathematics”. In: Posthumous Writings, 203–250. (Nachgelassene Schriften, 219–270) — 1919. “Notes for Ludwig Darmstaedter”. In: Posthumous Writings, 253–257. (Nachgelassene Schriften, 273–277) 9. I have presented the paper at the Workshop Concepts and their Analysis in Zürich (2008) and in the colloquium of the philosophy department of the university of Warwick (2009). I am grateful to both audiences for helpful feedback. Christoph Pfisterer’s comments on the paper in Zürich were especially valuable. I have also profited from discussions with Stephen Butterfill, Jessica Leech, Guy Longworth, Dawn Phillips and an email exchange with Stephan Krämer. I am grateful to Christian Nimtz for comments on the penultimate version.

256

Hofweber, Thomas. 2007. “Innocent Statement and their Metaphysically Loaded Counterparts”. Philosophers’ Imprint 7, 1–33. Horty, John. 1993. “Frege on the Psychological Significance of Definitions”. Philosophical Studies 72, 223–63. Kant, Immanuel. 1800. Imanuel Kants Logik. Ein Handbuch zu Vorlesungen. Königsberg: Friederich Nicolovius. Langford, Cooper H. 1942. “The Notion of Analysis in Moore’s Philosophy”. In: Paul A. Schilpp (ed.). The Philosophy of G. E. Moore. Open Court: La Salle, 321–42. May, Robert. 2006. “The Invariance of Sense”. The Journal of Philosophy 103, 111–144. Moore, George Edward. 1942. “A Reply to My Critics § 11”. In: Paul A. Schilpp (ed.). The Philosophy of G. E. Moore, Open Court, La Salle, 660–667. Nelson, Michael. 2008. “Frege and the Paradox of Analysis”. Philosophical Studies 137, 159–181. Rooth, Mats. 1996. “Focus”. In: Shalom Lappin, (ed). The Handbook of Contemporary Semantic Theory. London: Blackwell, 271–297. Rumfitt, Ian. 2001. “Concepts and Counting”. Proceedings of the Aristotelian Society, 41–68. Weiner, Joan. 2007. “What’s in a numeral? Frege’s answer”. Mind 116, 677–716. Whitehead, Alfred. 1898. A Treatise on Universal Algebra, with Applications. Cambridge: Cambridge University Press. Wisdom, John. 1934. “Is Analysis a useful method in Philosophy?” Proceedings of the Aristotelian Society 13, 65–89.

257

GPS Editors Johannes L. Brandl (Universität Salzburg) Marian David (University of Notre Dame) Maria E. Reicher (Universität Aachen) Leopold Stubenberg (University of Notre Dame) Managing Editor Martina Fürst (Universität Graz) Editorial Board Peter Baumann Monika Betzler Victor Caston Thomas Crisp Dagfinn Føllesdal Volker Gadenne Hanjo Glock Robert M. Harnish Reinhard Kamitz Thomas Kelly Andreas Kemmerling Jaegwon Kim Peter Koller Wolfgang Künne Karel Lambert Keith Lehrer Hannes Leitgeb Joseph Levine Alasdair MacIntyre

Georg Meggle Thomas Mormann Edgar Morscher Herlinde Pauer-Studer Christian Piller Edmund Runggaldier Heiner Rutte Werner Sauer Alfred Schramm Gerhard Schurz Geo Siegwart Peter Simons Barry Smith Thomas Spitzley Matthias Steup Mark Textor Thomas Uebel Ted Warfield Nicholas White

Subscription Rates There is no annual subscription rate. Each volume has its own price, varying from 60 to 90 US Dollars. Individual subscribers get a 50% discount. Informations for Contributors See the inside of the back cover.

Informa ion for Contributors GPS publishes articles on philosophical problems in every area, especially articles related to the analytic tradition. Each year at least two volumes are published, some of them as special issues with invited papers. Reviews are accepted only by invitation. Manuscripts in German or English should be submitted electronically as an e-mail attachment, either in MS Word or in rtf format, prepared for anonymous reviewing (i.e. without the author’s name and affiliation), together with an English abstract of 60–100 words. Footnotes should be kept to a minimum, and references should be incorporated into the text in (author, date, page) form. An alphabetical list of references should follow the text. A submitted paper will normally be sent to a referee. Authors are responsible for correcting proofs. Corrections that deviate from the text of the accepted manuscript can be tolerated only in exceptional cases. Authors will receive 25 free offprints of their article. Manuscripts should be sent to the following address: [email protected] Email Addresses Editors [email protected] [email protected] [email protected] [email protected] Managing Editor [email protected] GPS online Further information about GPS (back volumes, electronic publication, etc.) is available at the publisher’s web site: http://www.rodopi.nl, section “Series & Journals”.