Knowledge and Questions. (Grazer Philosophische Studien)

Knowledge and Questions

Grazer Philosophische Studien inteRnationale ZeitsCHRiFt FÜR analYtisCHe PHilosoPHie

gegRÜndet Von Rudolf Haller HeRausgegeBen Von Johannes l. Brandl Marian david Maria e. Reicher leopold stubenberg

VOL 77 - 2008

Amsterdam - New York, NY 2008

Knowledge and Questions

Edited by

Franck Lihoreau

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

In memoriam Georg Henrik von Wright

The paper on which this book is printed meets the requirements of “ISO 9706:1994, Information and documentation - Paper for documents Requirements for permanence”. Lay out: Thomas Binder, Graz ISBN: 978-90-420-2475-5 ISSN: 0165-9227 © Editions Rodopi B.V., Amsterdam - New York, NY 2008 Printed in The Netherlands

TABLE OF CONTENTS

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

vii

Christopher HOOKWAY: Questions, Epistemology, and Inquiries . . . .

1

Claudine TIERCELIN: The Fixation of Knowledge and the QuestionAnswer Process of Inquiry . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

23

Pascal ENGEL: In What Sense Is Knowledge the Norm of Assertion?

45

Ian RUMFITT: Knowledge by Deduction . . . . . . . . . . . . . . . . . . . . .

61

Paul EGRÉ: Q uestion-Embedding and Factivity . . . . . . . . . . . . . . .

85

Martin MONTMINY: Cheap Knowledge and Easy Questions . . . . . .

127

Berit BR OGAARD: Kno wledge-The and P ropositional A ttitude Ascriptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

147

Maria ALONI: Concealed Q uestions Under Cover . . . . . . . . . . . .

191

Daniele SGARAVATTI & E lia Z ARDINI: Knowing How to Establish Intellectualism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

217

Franck LIHOREAU: Knowledge-How and Ability . . . . . . . . . . . . . . .

263

Stephen HETHERINGT ON: Kno wing-That, Knowing-How, and Knowing Philosophically . . . . . . . . . . . . . . . . . . . . . . . . . . .

307

Duncan PRIT CHARD: Kno wing the Answ er, U nderstanding and Epistemic Value . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

325

Grazer Philosophische Studien 78 (2008), vii–xxiv.

INTRODUCTION Franck LIHOREAU Universidade Nova de Lisboa The topic of the r elationships betw een kno wledge and questions is of utmost impor tance to epistemology , philosophical logic, and the philosophy of language, as it raises a large number of issues in each of these fields and at their intersection. M ost of these issues ar e dealt with in the papers collected in this special v olume of Grazer Philosophische Studien. Lots of other highly r elevant and v ery inter esting topics ar e addr essed too. Surveying the papers in the v olume will thus amount, in a sense, to giving an overview of the current state of the debates o ver the issue(s) of knowledge and questions. Analytic epistemology, especially in the post-G ettier era, has put considerable emphasis on the “ standard” analysis of kno wledge as justifi ed true belief, the first formulation of which is, more often than not, credited to Plato’s Meno (98a). However, think a minute about the way in which Socrates, in the same Meno, makes his interlocutor “give birth” to a geometrical truth of which he seemed totally ignorant before, by asking him a series of questions that he answers correctly (82a–86a). This suggests an “alternative” approach to knowledge, whereby the ability to understand, to answer, and to formulate questions becomes of prime epistemological importance. Variants of such an approach have been endorsed by authors like Castañeda (1980), White (1982), Craig (1990), and others as well. The first two papers in the volume relate to this alternative line of thought. In the opening paper of the v olume, “Questions, Epistemology, and Inquiries,” Christopher HOOKWAY highlights four impor tant respects in which questions can be thought to be erlevant to epistemology. All four respects relate to the practices of epistemic ev aluation that occur when we are engaged in activities of inquir y. First, we use questions to set and identify our cognitive goals, what it is that is to be found out as the result of successful inquiry. Second, we can use questions to elicit information, not only when w e ask other people questions so as to gain kno wledge from their testimony, but also, presumably, when we ask ourselves ques-

tions in order to recover previously learned information. Thir d, questions play a cr ucial role in the ex ercise of r egulative control over our inquiries—one of the S ocratic roles that H intikka confers on them (H intikka 2007; see also Hintikka 1989’s “interrogative model of inquiry.”) On the one hand, asking people about the legitimacy of their opinions or about the grounds for the information they impar t is a way of assessing their value as testimonial sour ces; on the other hand, success in our inquiries and deliberations, whether individual or collectiv e, can be quickened by asking the right questions, or on the contrar y, be impeded b y asking the wrong questions. There is a fourth respect in which the formulation, the understanding, and the evaluation of questions are epistemologically relevant, namely, in virtue of the role they play in knowledge ascriptions. It is indeed v ery likely that most of our or dinary knowledge ascriptions are not of the some what canonical “‘kno w’ plus declarative that-clause” form, but of other forms, including the “‘kno w’ plus interr ogative whclause” form (i.e. the “ ‘know’ plus indirect question” form, explor ed in the papers by Aloni, Brogaard, and Egré). Given how relevant questions are to epistemology, Hookway proposes to investigate the complexity of our ability, qua epistemic agents, that is, as inquirers, to formulate, understand, and evaluate questions, a problem which, by the way, has interested epistemologists as well as logicians and philosophers of language at least since Hamblin’s seminal article, “Questions” (1958), inaugurated decades of debates between “semantic” and “pragmatic” approaches to questions and the meaning of interrogatives. The standard, semantic appr oach to the logic of questions, dir ectly in the line of Hamblin, relies on the assumption that understanding a question is a matter of kno wing the range of possible answ ers to it. ( Variants of this appr oach are described in B elnap and S teel 1976, H arrah 1984, Higginbotham 1996, Groenendijk and Stokhof 1997.) For Hookway, that approach misses some of the most relevant aspects of the understanding of questions. One such aspect, Hookway argues, is related to the importance of context in determining what counts as a er levant answer to a given question. Here, it is not just the context of utterance that counts, but also the context of inv estigation whose ev olution is mirr ored by the dynamics of the former kind of context. I nquiries and deliberations, Hookway insists, are “extended, structured goal dir ected activities,” so that “ the context of investigation changes rapidly, as possibilities ar e considered, investigated and eliminated, as challenges are raised and responded to.” This requires on our part that we be able to track the changing questions, and subsequently, viii

to track the “shifting range of answers that are relevant to them.” In other words, the notion of a relevant answer is to be approached from a dynamic perspective. But, Hookway notes, relevance of answers is not the final word, for we wouldn’t even bother to engage in an inquiry about a question if the question weren’t in some sense “real” for us. An impor tant part of Hookway’s paper aims at getting a better grip on what makes a question “ real,” a question that will prompt us to try to find an answer, as well as on what makes it reasonable to address such a question. The upshot of Hookway’s reflections on the matter is that understanding a question is, to a gr eat extent, a pragmatic matter, depending on such factors as “the motivations and interests of the person who asked the question; the shifting concerns that result from the development of a conversation, discussion, or inquiry; and the background knowledge of questioner and respondent.” A leitmotiv of Hookway’s epistemological work is that “the focus of our ‘epistemic lives’ is the activity of inquiry: we attempt to find things out, to extend our knowledge by carrying out investigations directed at answering questions, and to r efine our kno wledge by considering questions about things we currently hold to be true.” (Hookway 1990, 211) The point of epistemic evaluation, then, is the control of our inquiries and deliberations, instead of the mere evaluation of our cognitive states as being justified or as constituting knowledge. Issues about warrant and justification, central to orthodox epistemological theorizing, are thus conceived of as subordinated to issues r elated to what H ookway intends to occupy the centr e of the epistemological stage: inquiry. Sceptical challenges, then, can be er defined as “threatening our confidence that we can, indeed, inquire effectively and responsibly,” and not so much as thr eatening the possibility of kno wledge as such. So, it is not only a heterodox conception of knowledge that Hookway defends, but also a heterodox conception of the epistemological project. In “The Fixation of Knowledge and the Question-Answer Process of Inquiry,” Claudine TIERCELIN clearly acknowledges the merits of Hookway’s dynamic and pragmatist approach to knowledge. Yet, she has doubts about the “reality” of the need for depar ting, as Hookway invites us to do, fr om the or thodox epistemological perspectiv e. To make her point, combining lessons drawn fr om her discussion of other pragmatist and pragmatist-affiliated views, mainly Peirce and Wittgenstein, Tiercelin proceeds to outline a picture of knowledge and inquiry which, she thinks, reconciles the cor e pragmatist insights with the spirit of the traditional definition of knowledge as justified true belief. Tiercelin proposes to refine this defi nition b y making kno wledge, qua aim of inquir y, a matter of ix

“basically commonsensical and critical […], warrantedly asser tible […] intellectual and sentimental dispositions and asser tions […] for which the epistemic agent [possibly a community of inquir ers] should be held responsible.” Such a conception, she claims, per fectly compatible with some core pragmatist principles, is likely to shed much light on ho w we ought to think of the possibility of doubt, of the notion of inquir y, and of assertion. The connection between knowledge and the speech act of asser tion is precisely Pascal ENGEL’s main focus in his paper “In What Sense is Knowledge the N orm of Asser tion?” Engel considers the “kno wledge account of assertion,” developed by Slote (1979), Unger (1975), and Williamson (2000). O n Williamson’s v ersion of this account, the only constitutiv e norm of asser tion is that one must asser t that P only if one kno ws that P. To a number of objections to the eff ect that this kno wledge rule for assertion, and mor e generally, “epistemic” norms for asser tion (in terms of warrant, rational credibility, etc.), are either confused or inapplicable in practice, Engel replies that those objections rest on a misunderstanding of the nature of a norm. Those objections are no threat to epistemic accounts of assertion once “what it takes to be the norm of assertion” has been made clear. To object that an epistemic norm for assertion cannot be the norm, or that there can be no norm of assertion at all because otherwise it would too often be violated, or to object that an assertion presumably conforms to other, non-epistemic, norms, is, accor ding to E ngel, to amalgamate two notions of a rule. The knowledge rule, in particular, is meant to be a norm qua constitutive rule for assertion, not qua lexical rule for ‘assertion.’ Cases of assertions that fall short of corresponding knowledge do not, by any means, invalidate the knowledge rule, since the knowledge rule “does not say that it is a defining feature of assertion (a necessary and sufficient condition) that it is go verned by the norm of kno wledge.” On the other hand, Engel introduces a distinction betw een the “statement of a norm” and its “regulation.” The statement of a norm is what captures its objectivity, notably in an imperative form: “One must: perform practice P only if one satisfies feature C.” On the other hand, the regulation of a norm is a subjective matter: the regulation conditions of a norm “are the subjective conditions under which the [statement of a norm is] accessed b y a given individual and [is] implemented in his psychology.” So, while the statement of a norm for a given type of practice does not depend on what the subject believes to be the norm for that type of practice, theegulation r of the norm does depend on the subject’s beliefs about the norm. A pplied to the case x

of assertion, and assuming that kno wledge is the norm of asser tion, that means that a subject needn’t believe that it is indeed the norm of assertion in order to per form an asser tion according to that norm. S he may ev en believe that other norms are in place, or violate the norm, or let the norm be overridden by other candidate norms. But that does not, by any means, allow us to infer that the norm is not in place. E ngel concludes that the charge of confusedness, or inapplicability , against epistemic accounts of assertion, the knowledge account for one, does no harm. It is usually taken for granted that gaining new pieces of kno wledge is the point and pr oper outcome of successful inquir y, and that w e can do so, in a great number of cases, by means of our exercising a deductive capacity. In “Knowledge by Deduction,” Ian RUMFITT focuses on two questions r elated to the issue of kno wledge acquisition b y deduction: How can a subject gain knowledge by exercising her capacity for deductive reasoning? And: H ow can she gain new knowledge by exercising such a capacity? In answer to the fi rst question, Rumfitt endorses an externalist view about kno wledge, according to which one ’s belief that P is kno wledge if it is formed by using a reliable method, one that reliably yields the truth. Whenever a subject is “disposed to deduce a conclusion from some premises only when the conclusion does follow from them, and to recognize at least some of the mor e obvious cases of one statement’s following from others,” the externalist r equirement that the subject ’s belief in the conclusion be formed by a reliable method will be satisfied. To address the second question, Rumfitt opts for what he calls an “exclusionary” conception of a statement’s “logically relevant content,” which he develops from some r emarks b y Ramsey (1927). H e identifi es such content with the possibilities that the statement is understood to exclude, among those that are “live” possibilities in the context in which the declarativ e type-sentence constitutive of the statement is uttered. Rumfitt then defines a statement as tr ue when “ for ev ery possibility that the statement is understood to exclude, something obtains which pr ecludes the obtaining of that possibility.” This allows him to explain how a subject may come to know, by applying her deductive capacity to premises she already knows, a conclusion that she would not hav e come to kno w had she not ex ercised that capacity. When the subject knows the premises, for every “live” possibility that the conclusion is understood to ex clude, she knows something that precludes the obtaining of that possibility. But this is not enough for her to know the conclusion: she still “has to dosome conceptual work in order to see that each possibility ex cluded by the conclusion has alr eady been xi

excluded by the premises.” In this sense, kno wledge of the conclusion is not contained in knowledge of the premises. This view, Rumfitt continues, also explains how various sources of knowledge—perception, testimony, memory, etc.—can, jointly, establish a conclusion that none of them can, individually, establish. Allo wing us to combine deliv erances of diff erent sources of knowledge is precisely what, for R umfitt, makes for the v alue of logic. Based on this consideration, R umfitt concludes that systems of single-conclusion logic have that value, while multiple-conclusion logics don’t. A further epistemologically relevant aspect of inquiry has not so much to do with ho w we can come to kno w and with ho w we talk from what we know, as with how we talk about knowledge. As Jonathan Schaffer, taking inspiration from Hookway, writes, “one of the roles of the knowledge ascription is to keep scor e of the o verall progress of inquiry. We have an epistemic interest in tr uth. Inquiry is our method for seeking tr uth. So we have an epistemic inter est in keeping scor e of the o verall progress in inquiry, and we use our epistemic vocabulary to serve this interest—‘knows’ is an honorific for successful inquirers.” (Schaffer 2004, 84) Accordingly, knowledge ascriptions, and the role of questions in knowledge ascriptions, will be the unifying topic of the next four papers in the v olume. As alluded to above, analytic epistemology has usually been more concerned with kno wledge-that and kno wledge-that ascriptions than with knowledge-wh and knowledge-wh ascriptions. Knowledge-wh ascriptions are constr uctions in which ‘kno w’ takes a wh-clause as a complement. More precisely, a wh-clause is a clause that is intr oduced by a wh-word, like ‘who,’ ‘when,’ ‘where,’ ‘why,’ ‘how,’ and so on. N ow, embedded whclauses can basically be of two kinds: interrogative, and non-interrogative. Among non-interrogative embedded wh-clauses are “free relatives” (FR). (Jacobson 1995 and G rosu 2003 ar e among the main r eferences on the syntax and semantics of FRs.) A FR can always be paraphrased by means of an appropriate determiner phrase (DP), as in: (1) She ate [FR what was left in the cupboard]/[DP the food that was left in the cupboard] or an appropriate preposition phrase (PP), as in: (2) They left [FR when we arrived]/[PP at the time that we arrived]

xii

FRs show striking similarities with embedded wh-clauses of the interrogative kind, especially at the lexical lev el: except for ‘ whether’, all the wh-words that may be contained in a FR seem to be exactly those that may be contained in an embedded wh-interrogative. Examples of embedded interrogative wh-clauses are typically found in propositional attitude ascriptions, as in ‘He knows who came to the party,’ ‘She wonders why he was crying,’ ‘I remember where you have left your keys.’ Most of the papers in this second par t focus on embedded interr ogative wh-clauses, “interrogative complements” for short (as opposed to “declarative complements”, that is, embedded that-clauses), and the relationships those complements bear to propositional attitude verbs, ‘know’ being the pr opositional attitude verb of reference. A puzzling fact about propositional attitude verbs and their admissible complements is that a v erb like ‘know’ takes both declarativ e and interrogative complements, while a v erb like ‘believ e’ takes only declarativ e complements and a v erb like ‘ wonder’ takes only interr ogative complements. This is the core observation that Paul EGRÉ, in “Question-Embedding and Factivity,” sets out to inv estigate, by focusing on the behaviour of those v erbs with r espect to the embedding of whether-clauses. After discussing a number of candidate semantic as w ell as pragmatic explanations for that observation, Egré provides evidence for the hypothesis that among all v erbs licensing that-complements, those that ar e “veridical” also license whether-complements, wher e such a v erb is v eridical “if it entails the truth of its complement when used in the positiv e declarative form, namely if it satisfi es the schema Vp o p for all p, where p is a thatclause.” By appealing to this hypothesis in combination with the fur ther hypothesis, borrowed from Groenendijk and Stokhof (1982, 1990), that the extension of an interr ogative complement is a pr oposition (the tr ue answer to the corresponding question), Egré then explains the diff erence between ‘know’ and ‘believ e’ with r espect to question-embedding. Both ‘know’ and ‘believe’ basically take complements of the same semantic type, that-complements, which denote propositions. But because ‘know’, unlike ‘believe,’ is veridical, the denotation of the that-clause after ‘know’ coincides with the extension of the corresponding whether-complement—viz. the pr opositional answ er to the whether-question. This is why , unlike ‘believe,’ ‘kno w’ licenses whether-complements. E gré also explains the difference between ‘know’ and ‘believ e’ on the one hand, and ‘ wonder’ on the other hand. E gré assumes that ‘ wonder’ selects for the intension, as opposed to the extension, of a question, that is,grosso modo, for a set of xiii

propositions, instead of a proposition as in the case of ‘know’ and ‘believe’. For this reason, ‘wonder’ cannot take that-complements. Egré then goes on to assess the v alue of the proposed explanation in light of prima facie disconfirming data about other apparently veridical verbs as well as nonveridical ones. The explanation he proposes for the fact that ‘know’ licenses both thatand whether-complements, leads Egré to say that “to know that is to know whether, just as to know whether is to know that.” This conclusion can be seen as an instance of what Jonathan Schaffer, in “Knowing the Answer” (2007), labels the “ received” or “ reductive vie w” about kno wledge-wh, according to which knowledge-wh, just like knowledge-that, consists in a binary relation Ksp between a subject and a proposition: (3) s knows-wh if and only if Ksp, where p is the tr ue answer to the indirect question Q denoted by the wh-complement. (386) With this vie w can be associated the names of H intikka (1975), Le wis (1982), Boër and Lycan (1986), and more recently Stanley and Williamson (2001), for whom the meaning of (at least) cer tain constructions in terms of ‘know’ can be accounted for reductively in terms of the meaning of knowledge-that ascriptions. Schaffer puts forth an alternative view, the “question-including vie w” of kno wledge-wh, based on the idea that, in any construction in which it may occur , ‘know’ includes an irr educible reference to a question, and that to know is always to know something as an answer to a question. O n this view, knowledge-wh is understood as a ternary, rather than binar y, relation KspQ between a subject, a pr oposition, and a question: (4) s kno ws-wh if and only if KspQ, wher e Q is the indir ect question denoted by the wh-complement, and p the true answer to Q. (392) Because he takes questions to denote sets of contextually relevant alternatives, Schaffer equates the “question-relative” relation KspQ with the “contrast-relative” relation Kspq, “where q is the disjunction of non-p answers to Q.” (393) In particular, to know that p as the true answer to whether p or q is to know that p rather than q. As he intends his analysis to extend to all knowledge ascriptions, Schaffer’s question-including view is another name for what he calls the “contrastivist” view, or “contrastivism.” According to xiv

contrastivism, ‘know’ always denotes the ternar y relation Kspq, where p stands for a proposition and q for a contrast proposition. In the particular case of apparently binary knowledge ascriptions of the form ‘S knows that P ’, the v alue of the contrast v ariable is contextually pr ovided. This view, supposedly inspired by Dretske (1970) and found in other forms in Morton and Karjalainen (2003) and in S innott-Armstrong (2004), inter alia, is defended b y Schaff er in “From Contextualism to Contrastivism” (2004) and “Contrastive Knowledge” (2005). In his paper “Cheap Knowledge and Easy Questions,” Martin MONTMINY examines Schaff er’s contrastivist vie w and objects to it on the ground that it is unable to fully account for ordinary speakers’ judgments about kno wledge claims made in v arious contexts. O n the one hand, Montminy argues, some knowledge claims that are intuitively false given the context in which they ar e made turn out tr ue on Schaff er’s account. On the other hand, M ontminy continues, some kno wledge claims that are intuitively true given the context in which they are made are predicted to be false by Schaffer’s view. In particular, to make his point, Montminy considers knowledge of the denial of both moderate and radical sceptical hypotheses, and argues that in some “lo w standards” contexts, one can truthfully assert ‘I know that non-SH’—for SH a sceptical hypothesis—, whereas such a sentence would always be predicted to be false by Schaffer’s view. Montminy goes on to argue that when it comes to accounting for our intuitions about knowledge claims, another version of epistemological contextualism is likely to fare better than Schaffer’s contrastivist view. Epistemological contextualism is the view, championed by Cohen (1990), DeRose (1995), and Lewis (1996), according to which knowledge claims depend for their truth on the context in which they are made. “Epistemic pluralism,” the v ersion of contextualism fav oured by Montminy, allows the standards of justification to vary with the contexts in which knowledge claims are made: in a “lo w standards” context, the standards of justifi cation that are in place are “externalist,” which makes it truthful to assert ‘I know that non- SH’; on the contrar y, in a context wher e “high”, that is, “internalist” standards of justification are in place, the sentence cannot be truthfully asserted. According to Montminy, such a view is what is needed to do justice to ordinary speakers’ judgments about knowledge claims. Back to kno wledge-wh ascriptions, B erit BROGAARD, in “Knowledge-The and Propositional Attitude Ascriptions,” proposes to take a closer look at the semantics of the interrogative complements of ‘know.’ Earlier, a distinction between two kinds of wh-clauses was mentioned, that between xv

free relatives and wh-interrogatives. Brogaard’s suggestion is that all embedded wh-clauses, whether they are interrogative or consist of free relatives, be given the same uniform semantic tr eatment as pr edicates. Thus, for example, ‘when we arrived’ shall be tr eated as denoting the pr operty of being a time at which the people er ferred to by ‘we’ arrived; ‘who attended the lecture’ as denoting the property of being a person who attended the lecture in question. And this is supposed to be so whether the wh-clause is a wh-interrogative or a FR. (For another recent attempt, based on crosslinguistic evidence, at unifying the semantic treatment of wh-clauses, see Caponigro 2004; see also Caponigro and Heller 2007.) The possibility of accounting for wh-clauses as pr edicates allows Brogaard to pr opose the following truth-conditional analysis for knowledge-wh ascriptions: (5) ‘S knows wh-F ’ is true if and only if, for some x, S knows that x is wh-F. where, presumably, the range of x depends on thewh-word (persons in the case of ‘who’, ways in the case of ‘ho w’, locations in the case of ‘ where’, etc.). Brogaard extends this analysis to another type of knowledge ascription. In some such ascriptions, ‘kno w’ takes a defi nite or indefinite DP complement: ‘ S knows the F ’, ‘ S knows an F ’. In some cases, the DP in constr uctions of this form can be r eplaced b y a tr uth-conditionally equivalent wh-clause, as in: (6) Jessica knows [the capital ofVermont]/[what the capital ofVermont is]. When a paraphrase of this sort is available, the DP can be thought of as a “concealed question” (CQ), that expressed by the appropriate wh-clause. This being so, Brogaard proposes to apply an analysis analogous to (5) to ‘know-CQ’ constructions: (7) ‘S knows an/the F ’ is true iff for some x, S knows that x is an/the F. Maria ALONI offers a very different approach to the phenomenon of concealed questions, to which she dedicates her paper, “Concealed Questions Under Cover.” She focuses on the obser vation, initially made b y Heim (1979), that a knowledge-CQ ascription such as: xvi

(8)

John knows the capital that Fred knows.

is ambiguous between the following two readings: (9) John knows the same capital that Fred knows. (10) John knows what capital Fred knows. The pr oblem is to fi nd a suitable r epresentation of the ambiguity . I n order to solve it, Aloni assumes, like Egré in his paper, that ‘know’ always selects for a pr oposition. I n par ticular, when it takes an interr ogative complement, ‘know’ selects for the proposition that is the complete tr ue answer to the question. B ut a DP like that in (8), viz. ‘ the capital that Fred knows,’ is presumably an expression that denotes an individual entity, not a proposition. So, Aloni argues, there would be a mismatch between the semantic type of the complements licensed b y ‘know’ and the type of a DP like that in (8), if a type-shifting r ule didn’t enter the picture at that place, turning expr essions denoting individual entities into identity questions appropriate to those expr essions. So, for instance, if the DP is ‘the capital that Fred knows,’ the type-shifting rule applies and yields the question ‘What is the x such that x is identical with the capital that F red knows?’ This explains how (8) can be read as in (10), which can be paraphrased as saying that J ohn knows the answ er to the question ‘ What is the capital that Fred knows?’ Suppose Fred knows the capital of Italy. On this reading the sentence entails that J ohn knows that Fred knows what the capital of Italy is, but John himself need not know what the capital of Italy is. To explain this meaning Aloni assumes that an identity question is always interpr eted with r espect to a contextually selected “ conceptual cover.” A conceptual cover is a set of individual concepts, the idea being that each individual in the domain of discourse is identifi ed b y means of one and only one concept (in each world). Conceptual covers being allowed to v ary with the conv ersational circumstance, our ev aluation of the relevant fragment of discourse, in particular what counts as an answer to a wh-question like that r esulting fr om type-shifting a DP , will v ary accordingly. Putting conceptual covers to use, Aloni can then explain how (8) gets its two r eadings, as (10) can no w be paraphrased as saying that John knows (under one conceptual perspective) what capital Fred knows (under another conceptual perspective), and (9) can now be paraphrased as saying that ther e is a unique capital that F red knows and John knows it too under the same conceptual perspective as Fred. xvii

Of all constructions in terms of ‘know,’ those of the form S‘ knows how to F’ are of the most impor tant philosophically speaking, due to G ilbert Ryle’s (1949)denunciation of the prejudice in favour of knowledge-that and his subsequent insistence on the necessity of making kno wledge-how a kind of knowledge distinct from, and irreducible to a kind of knowledge-that. It has become standard practice to call “intellectualists” those who defend the reducibility of kno wledge-how to kno wledge-that, and to call “ antiintellectualists” those who deny that knowledge-how can be reduced to a kind of knowledge-that (disregarding the variety of intermediary positions that may be endorsed on the matter). (For a recent survey of the literature on knowledge-that and knowledge-how, see Fantl 2008.) A consequence of the aforementioned account of knowledge-wh by Brogaard is that she seems to be siding with the intellectualists. Indeed, arguing that ‘how’ is, semantically speaking, just a wh-word, Brogaard extends her analysis of knowledge-wh in (5) to the special case of knowledge-how: (11) ‘S knows how to F’ is tr ue if and only if , for some w, S knows that w is how to F. That is: knows that w is a way in which she mayF, or: knows that w is a way in which one may F, depending on whether ‘know how to F’ is interpreted as entailing the possession b y S of an ability to F—an “ability-entailing” reading—or not—an “ability-neutral” reading. In this respect, Brogaard’s account presents striking similarities with the str ong intellectualist position endorsed b y Jason Stanley and Timothy Williamson in their justly famous paper “Knowing How” (2001), which can certainly be praised for having recently revived the interest of analytic philosophers in the issue of knowledge-how and knowledge-that. The next three papers in the collection can be considered as challenges and/or alternatives to the reductivist or intellectualist view on knowledge-how. Focusing on the v ersion of intellectualism defended b y Jason Stanley and Timothy Williamson, Daniele SGARAVATTI and Elia ZARDINI present a number of criticisms intended to undermine some of tanley S and Williamson’s main arguments and assumptions. Sgarav atti and Zardini’s first objection is that S tanley and Williamson’s version of (11) will yield intuitively incorrect truth-conditions for ‘ S knows how to F ’ when the obvious answer to the question ‘How to F?’ is ‘There is no way to F.’ Their second objection is that ther e are cer tain logical r elations under which knowledge-how seems to be closed, but under which their knowledge-that xviii

counterparts appear not to be closed, especially when the relevant knowledge has “singular content,” that is, when it is dir ectly about a particular object. Their third objection rests on the conjecture that if knowledge-how were just a kind of knowledge-wh, and ultimately of knowledge-that, we should expect that pieces of kno wledge-how could be used as gr ounds for acquiring fur ther pieces of kno wledge-that by inference, just as w ell as obvious pieces of kno wledge-that and kno wledge-wh can be used to that eff ect. Unfortunately, Sgaravatti and Zardini argue, knowledge-how shows important dissimilarities with knowledge-wh and knowledge-that in that respect. Stanley and Williamson (and Brogaard in a similar fashion) assume as an intermediar y step in their argument for the r educibility of knowledge-how to knowledge-that, that knowledge-wh in general is reducible to knowledge-that. Sgaravatti and Zardini’s fourth and final objection is that Stanley and Williamson’s analysis of knowledge-wh ascriptions, in the vicinity of Brogaard’s analysis in (5), cannot fully deal with the availability of comparative knowledge-wh ascriptions (such as ‘Hannah knows how to ride a bicycle better than Mary does’) as opposed to the apparent unavailability of comparative knowledge-that ascriptions. Although they do not intend to pr ovide a positiv e account of kno wledge-how, a recurrent concern in Sgaravatti and Zardini’s paper has to do with the diffi culties faced b y the intellectualist when it comes to dealing with the apparent ability-like character of the knowledge ascribed by using ‘know how’ in some cases. I ndeed, a linguistic “ fact” about kno wledgehow ascriptions, already alluded to abo ve, is that in some cases, they will be giv en an ability-entailing r eading, while in other cases, they will be given an ability-neutral reading. In “Knowledge-How and Ability,” Franck LIHOREAU proposes to appr oach that fact, and the issue of kno wledge-how and ability mor e generally speaking, fr om an epistemological perspective, by focusing on the kinds of kno wledge and of ability that we may talk about when w e talk about kno wing how to do something. H e distinguishes two kinds of kno wledge of how to do something, a “ practical” and a “theoretical”, as well as two kinds of ability to do something, an “intrinsic” and an “ extrinsic”, and defi nes one’s “practical” knowledge of how to F as one’s acquired intrinsic, rather than extrinsic, ability to F, and one’s “theoretical” knowledge of how to F as one’s knowledge of, or about, how something that has the acquired intrinsic ability to F does F. Lihoreau intends to explain the aforementioned linguistic fact about knowledge-how ascriptions by putting to use that double epistemological distinction, and by saying that the schema ‘S knows how to F only if S is able to F’ is true xix

when ‘knowing how’ is intended to r efer to kno wledge of the practical kind and ‘being able’ to refer to an ability of the intrinsic kind. Lihor eau further argues that a number of theor etical benefits are to be gained from endorsing the resulting view about knowledge-how and ability, regarding issues related to the connection betw een knowledge-how and intentional action, to comparative attributions of knowledge-how and ability, the Gettierability of kno wledge-how, the opacity of kno wledge-how ascriptions, and possible counter examples to the vie w that to kno w how is to hav e an ability. Building from previous work of his (see Hetherington 2006), Stephen HETHERINGTON recommends an ev en more radical, albeit subtle, treatment of the intellectualist thr eat, as he goes so far as inviting us to consider knowing-that as a kind of knowing-how, conceived in turn as an ability. In his contribution to the volume, “Knowing-That, Knowing-How, and Knowing Philosophically,” Hetherington’s purpose is to elucidate parts of what makes for the complexity of that ability. He suggests that we turn from what he calls the “standard conception of knowing,” that is, the conception that to know that P is to be in a “state” of knowing that P (where the state in question is typically understood as one of believing thatP cum further appropriate properties that make it properly epistemic), to the more dynamic view on which knowing that P can be equally identified with any member of what Hetherington calls P’s “epistemic diaspora,” a grouping of cognitive phenomena related to P, such as believing accurately that P, asserting accurately that P, accurately providing evidence as to P, etc. A consequence is that one can know that P by being the subject of some of the P-related cognitive phenomena ev en if one is not the subject of any other P-related phenomena. None of the P-related phenomena deser ves the name ‘knowing that P’ more than the others members of the diaspora. What unifies P’s epistemic diaspora is the ability to generate the expr ession or manifestation of any of the members of the diaspora. As a special case, the ability to accurately question whether or about P will often, but not always, be part of P’s epistemic diaspora. But according to Hetherington, when it comes to philosophical knowledge, that ability becomes essential, as he suggests that philosophical kno wledge that P be considered a kind of know-how, that which consists for a large par t in the ability to pose philosophical questions about P and still to fi nd such questions ev en in philosophical answers to them. By developing these considerations and by reflecting upon the complex aspects of philosophical ability , Hetherington invites us to a tr ue reversal of the epistemological perspectiv e, and to a xx

rethinking of the natur e and structure of knowing, and of philosophical knowledge and inquiry. It is y et another, diff erent kind of r eversal of the epistemological perspective that Duncan PRITCHARD calls for in the paper that closes this volume, “Knowing the Answer, Understanding, and Epistemic Value.” The main problem that Pritchard addresses is the “problem of epistemic value.” (See Pritchard 2007 for a survey of the literature on the topic.) An underlying assumption of the whole epistemological enterprise, as traditionally conceived, is that knowledge is distinctively valuable, in that it is, as such, not only worthy of scrutiny, but more worthy of scrutiny than “that which falls short of knowledge not just as a matter of degr ee but as a matter of kind.” The assumption, Pritchard argues, is erroneous. The only plausible account available of why kno wledge should be of distinctiv e value in the foregoing sense is unsatisfactor y. The account in question, understood in standard virtue-theoretic terms, explains that kno wledge deserves of fi nal value because it is a “ cognitive achievement,” mainly a cognitiv e success achieved b y the ex ercise of one ’s cognitiv e ability. Pritchard objects to this account that kno wledge can be found without accompanying cognitive achievement in the pr oper sense, and on the other hand, cognitiv e achievement can be found without accompanying knowledge. To put things roughly, a cognitive achievement, in the pr oper sense, cannot be lost just because of the pr esence of what P ritchard calls “environmental epistemic luck,” which he distinguishes fr om Gettier-style epistemic luck. O n the contrary, knowledge can be lost for that v ery reason. Not so with understanding, which Pritchard regards as irreducible to mere knowing-why, and beyond this, to any form of knowing-that. Understanding is that which is distinctively valuable epistemologically speaking, and it is so because it is understanding, not knowledge, that is the relevant cognitive achievement. In light of this thesis, Pritchard goes on to reassess the relevance of inquiry to the problem of epistemic v alue. If inquiry might be thought to be r elevant to the problem of epistemic value, it is, Pritchard contends, because the product of successful inquiry, individual as well as collaborative, is not knowledge, but understanding. Acknowledgements Six out of the twelve contributions to the present volume originate from papers presented at the international wor kshop “Knowledge and Q uesxxi

tions” held at the Ar chives Henri Poincaré in N ancy, France, on M arch the 15 th and 16 th, 2007. Thanks to Gerhard Heinzmann, then the H ead of the Archives, for the unconditional trust he put in me when it came to organising the workshop. The workshop was financially supported by the Centre National de la Recherche Scientifique, Nancy-Université, the Conseil Général de M eurthe-et-Moselle, and the Communauté Urbaine du G rand Nancy. Financial assistance was also provided by the Department of Philosophy of the Université Nancy 2. The rest of the authors, contacted after the workshop had taken place, all accepted straight away to contribute a paper to the volume. I thank them a lot for that. I would like to thank Grazer P hilosophische Studien and its editors, in particular the managing editor , Maria E. R eicher, for publishing this special issue on kno wledge and questions. M any thanks too to M artina Fürst, Thomas Binder, as well as to a number of anonymous referees from the Grazer team, for their professionalism and efficiency, and for contributing to making the pr eparation of this v olume a much less worrisome experience for me. Finally, above all, and again, I would like all the authors to know how grateful I am that they so kindly, generously, and enthusiastically accepted to be part of the project.

REFERENCES Belnap, N. and T. Steel. 1976. The Logic of Questions and Answers, Yale University Press, New Haven. Boer, S. E. and W. Lycan. 1986. Knowing Who, Bradford, Cambridge. Caponigro, I. 2004. “The Semantic Contribution of Wh-words and Type Shifts: Evidence from Free Relatives Crosslinguistically,” in Proceedings from Semantics and Linguistic Theor y (SALT) XIV, (Ed.) R. B. Young, 38–55, CLC Publications, Ithaca. Caponigro, I. and D. Heller. 2007. “The Non Concealed Nature of Free Relatives: Implications for Connectivity C rosslinguistically”, in Direct Compositionality, (Eds.) C. Barker and P. Jacobson, 237–263, Oxford University Press, Oxford. Castañeda, H. N. 1980. “The Theory of Q uestions, Epistemic Powers, and the Indexical Theory of Knowledge,” in Midwest Studies in Philosophy, Volume V: Studies in E pistemology, (Eds.) P. French et al., 193–237, U niversity of M innesota Press, Minneapolis. xxii

Cohen, S. 1990. “Scepticism and Kno wledge Attributions,” in Doubting: Contemporary Perspectives on Scepticism, (Eds.) M. Roth and G. Ross, Dordrecht: Kluwer. Craig, E. 1990 Knowledge and the State of Nature : An Essay in Conceptual Synthesis, Clarendon Press, Oxford. DeRose, K. 1995. “Solving the Skeptical Problem”, The Philosophical Review 104, 1–52. Dretske, F. I. 1970. “Epistemic Operators,” The Journal of Philosophy 67, 1007– 1023. Fantl, J. 2008. “Knowing-How and Knowing-That, ” Philosophy Compass 3, 451– 470. Groenendijk, J. and M. Stokhof. 1982. “Semantic Analysis of Wh-Complements,” Linguistics and Philosophy 5, 117–233. — 1997. “Questions,” in Handbook of Logic and Language, (Eds.) van Benthem, J. and A. ter Meulen, 1055–1124, Elsevier, Amsterdam. Grosu, A. 2003. “A Unified Theory of ‘Standard’ and ‘Transparent’ Free Relatives,” Natural Language and Linguistic Theor y 21, 247–331. Hamblin, C. 1958. “Questions,” The Australasian Journal of Philosophy 36, 159– 168. Harrah, D. 1984. “The Logic of Q uestions,” in Handbook of Philosophical Logic II, (Eds.) Gabbay, D. and F. Guenther, 715–764, Reidel, Dordrecht. Heim, I. 1979. “Concealed Questions,” in Semantics from Different Points of View, (Eds.) R. Bauerle, U. Egli, and A. von Stechow, Springer, Berlin. Hetherington, S. 2006. “H ow to Kno w (That Knowledge-That is KnowledgeHow),” in Epistemology Futures, (E d.) S. H etherington, 71–94, Clar endon Press, Oxford. Higginbotham, J. 1996. “The Semantics of Questions,” in Handbook of Contemporary Semantic Theor y, (Ed.) S. Lappin, Blackwell, Oxford. Hintikka, J. 1975. “Different Constructions in Terms of the Basic Epistemological Verbs: A Survey of Some Problems and Proposals,” in The Intensions of Intentionality and Other New Models for Modalities, Reidel, Dordrecht. — 1989. “Knowledge Representation and the I nterrogative Model of I nquiry,” in Knowledge and Scepticism , (Eds.) M. Clay and K. Lehr er, Westview Press, Boulder. — 2007. Socratic Epistemology: Explorations of Knowledge-Seeking by Questioning, Cambridge University Press, Cambridge. Hookway, C. 1990. Scepticism, Routledge, London. Jacobson, P. 1995. “The Q uantificational Force of F ree R elative Clauses”, in Quantifiers in Natural Languages, Vol. II, (Eds.) E. Bach et al, 451-486, Kluwer, Dordrecht.

xxiii

Lewis, D. K. 1982. “Whether Report,” in Philosophical Essays Dedicated to Lennart Åqvist on his F iftieth Birthday, (Eds.) T. Pauli et al., 194–206, Filosofiska Studier, Uppsala. Lewis, D. K. 1996. “E lusive Knowledge,” The Australasian Journal of P hilosophy 74, 549–567 Morton, A. and A. Karjalainen 2003. “Contrastiv e Knowledge,” Philosophical Explorations 6, 74–89. Pritchard, D. 2007. “R ecent Work on Epistemic Value,” American Philosophical Quarterly 44, 85–110. Ramsey, F. P. 1927. “Facts and propositions,” Proceedings of the Aristotelian Society: Supplementary Volumes 7, 153–70. Ryle, G. 1949. The Concept of Mind, University of Chicago Press, Chicago. Schaffer, J. 2004. “F rom Contextualism to Contrastivism,” Philosophical Studies 119, 73–103. — 2005. “Contrastive Knowledge,” Oxford Studies in Epistemology, Vol. 1, (Eds.) T.S. Gendler and J. Hawthorne, 235-271, Oxford University Press, Oxford. — 2007. “Knowing the Answ er,” Philosophy and P henomenological Research 75, 383–403. Sinnott-Armstrong, W. 2004. Pyrrhonian Skepticism, Oxford University Press, Oxford. Slote M. A. 1979. “Assertion and Belief,” in Papers on Language and Logic , (Ed.) J. Dancy, Keele University Library, Keele. Stanley, J. and T. Williamson 2001. “Kno wing how,” Journal of P hilosophy 98, 411–444. Unger, P. 1975. Ignorance, Oxford University Press, Oxford. White, A. R. 1982. The Nature of Knowledge, Rowman and Littlefield, Totowa. Williamson, T. 2000. Knowledge and its L imits, Oxfor d U niversity P ress, Oxford.

xxiv

Grazer Philosophische Studien 77 (2008), 1–21.

QUESTIONS, EPISTEMOLOGY, AND INQUIRIES Christopher HOOKWAY University of Sheffield Summary Questions are relevant to epistemology because they formulate cognitive goals, they ar e used to elicit information, they ar e used in S ocratic r eflection and knowledge sentences often hav e indir ect question complements. The paper explores what capacities we must possess if we are to understand questions and identify and evaluate potential answers to them. The later sections explore different ways in which these matters depend upon pragmatic and other contextual considerations

1. Interrogatives and epistemology This section identifies four diff erent ways in which our ability to use and understand interrogatives has an important role in our practice of epistemic evaluation. In doing this, w e assume that, if w e are to understand our practices of epistemic evaluation, we must recognize that such evaluations normally occur when we participate in activities of inquiry, in attempts to find things out (Hookway 2006, Hintikka 2007). Sceptical challenges can often be understood as thr eatening our confi dence that w e can, indeed, inquire eff ectively and r esponsibly. In that case, w e need to identify the capacities we must possess if such inquiry is to be possible and see the challenges as questioning whether we genuinely possess these capacities. And we best understand concepts such asknowledge and warrant or justification by examining their role in the regulation of inquiries. First, when w e par ticipate in inquir y, w e tr y to fi nd a solution to a problem or an answ er to a question: questions ar e used to identify our cognitive goals. Even if our epistemic aim is to arrive at the truth or knowledge, this has to be understood in the perspectiv e just off ered. Not just any true proposition will satisfy our epistemic aims. I f truth is our aim, then we seek a tr ue proposition that answ ers our question or solv es our

problem. Since questions have this role, we need to understand how our understanding of questions provides us with a clarification of our goal that will enable us to monitor progress towards achieving it. Second, questions are epistemically important because they can be used to elicit information. First, a central (if not the only) way in which we gain knowledge through the testimony of others, is through asking them questions. The search for knowledge through testimony can go wrong because the informant lacks the required information or because she chooses to lie in her reply. Just as important is the possibility that, although the potential informant possesses the information I r equire, my question fails to elicit it. I hav e asked the wr ong question, or asked the right question in the wrong way. It can also go wrong because although my informant offers the information I require, she does so in a way that I cannot absorb or pr ocess. As we shall see below, it is not always transparent what sort of answer the utterance of an interr ogative sentence requires, nor is it always clear just how that information should be pr esented. If we are to understand the capacities we require in order to obtain (and provide) information through this sort of question-answer procedure, we need to take account of these kinds of sensitivity to how information can be elicited and how it can be transmitted. Second, it is not implausible that something similar to what occurs when I learn fr om testimony happens when I r ecover information that I have learned and r emembered. Most of the information stor ed in my memory is not being attended to most of the time. S uccess in inquir y depends upon my being able to attend to it when necessar y—memories should be av ailable for r eflective attention when this is r equired for the regulation of my inquiries. It is a useful idealisation of how this works that I elicit such information thr ough raising a question to which it pr ovides an answer. And, just as in the case of testimony, I can fail to elicit relevant information that I possess by, in effect, asking the wrong question—or by asking the right question in the wrong way. This may reflect the cognitive architecture of our memories: if w e think of memories as stor ed in fi les which are indexed in some way , then it is easy to see that w e could fail to recover that information because our self-dir ected question does not successfully connect with how the information is stored. The third epistemological r ole for questions concerns the ex ercise of regulative control over inquiries (see Hintikka 2007). This Socratic role for questioning also divides into diff erent cases. First, suppose I am ev aluating you as a source of testimony. I shall do this by asking questions: what 2

reason do you have for believing that pr oposition? How reliable are you as a source of testimony on this topic? H ave you taken account of some circumstances which may make you unreliable? And so on: I evaluate your value as a source of testimony by interrogating you about the legitimacy of your opinions. Second, such questioning has an important role in activities such as discussion in the course of cooperative attempts to solve problems. When engaged in discussions, w e question the pr oposals, opinions and arguments of other participants, recognizing this was the way to regulate joint inquiry and ensure that it conforms to defensible standards. Mutual questioning is a way of testing our diff erent contributions to the discussion, and this is necessary as a means of improving our chances of success in our inquiries. And thir d, this can occur without any one else being involved. I r egulate my o wn inquiries and deliberations b y asking questions about them: I wonder whether I should check how reliable evidence is, I consider whether I should be mor e careful in double checking my reasoning, I ask myself whether ther e are defeating considerations that I have ignored or whether I’ m right to tr eat some item of information as providing a potential defeater , and so on. R eflective inquiry—whether individual or cooperativ e—depends for its success upon our asking the right questions about its pr ogress. Asking the wr ong reflective questions can impede progress1. Once we take account of these phenomena, w e can recognize another capacity that is required if we are to carry out inquiries effectively. We need to be able to recognize when it is rational to raise particular reflective questions, when we have a reasons to ask them. Of course, there are interesting questions about what makes it rational to adopt a par ticular cognitiv e aim; we can investigate what makes a question or pr oblem interesting or important. But the solutions to such pr oblems may often be external to our practices of distinctively epistemic evaluation, which are concerned with how we can achieve the epistemic goals that w e happen to have. We can have all sorts of reasons for being interested in questions. In the cases I am concerned with here, identifying reasons for raising questions is internal to an on-going activity of inquiry. It is rational to ask these questions because it is epistemically appropriate to do so at this stage in the inquir y we are concerned with. Doing so will sometimes elicit information w e need, or it may lend salience to considerations that w e need to attend to if our 1. One suggestive move in this direction is found in the work of Charles S. Peirce who argues that the conclusion of an abductive inference is usually an interrogative: the inference establishes that the question arises, that it should be answered (see Hookway 2005).

3

inquiry is to prosper. We can be eff ective inquirers only if we know how to establish when it is rational to ask subordinate questions as the inquiry develops. We need to be sensitiv e to reasons for raising questions as w ell as to reasons for belief and doubt. The fourth reason for taking questions seriously in epistemology is not a matter of what we have to do in carrying out inquiries. It concerns the logical grammar of sentences about knowledge. It is commonly assumed that knowledge sentences ar e of a similar form to belief sentences: both express propositional attitudes, perhaps concerning a r elation between a subject and a pr oposition. One obvious reason for questioning the existence of this parallel between the grammar of belief and of knowledge rests on the observation that the word ‘knows’, unlike the word ‘believes’ can take an indirect question complement. As well as sentences such as: John knows that 71 is a prime number. There are also knowledge sentences such as: John knows when the next train will leave. John knows how the internal combustion engine works. Johns knows why water expands on freezing. It is an attractive idea that the indirect question form is the fundamental one. We can always fi nd an equiv alent way of expr essing what a pr opositional knowledge sentence expresses by using just the indirect question form. Our example above is equivalent to: 71 is prime and John knows whether 71 is prime. It is a far mor e controversial matter whether the indir ect question forms listed above can be paraphrased into statements of pr opositional knowledge. It is easy to see ho w to paraphrase ‘kno wledge whether’ locutions. If I kno w whether it is raining then I either kno w that it is raining or I know that it is not raining (see H ookway 1990). B ut once w e consider cases involving idioms like ‘knows why’, ‘knows where’, and ‘knows who’ (among others) it is v ery diffi cult to see ho w a r eduction to the ‘kno ws that’ paradigm can be obtained. I t is thus natural to conclude that if w e abandoned the ‘kno ws that’ locution, our expr essive powers would not be diminished, although we may have to use more cumbersome locutions 4

than we are accustomed to. But if we were to abandon the locutions involving ‘knows’ with an indirect question complement, our expressive powers would be seriously impair ed. Even if that w ere disputed, it is plausible that the indir ect question kno wledge sentences ar e the most impor tant and most widely used. I n that case, w e should begin our study of the importance of the concept of knowledge by focusing on the indirect question ‘knowledge’ sentences and, indeed, to tr eat knowledge as a r elation between a subject and a question. A subject possesses kno wledge when she possesses the answer to or ‘a correct answer to’ some question. Either way, as has been noted for some time, it is a r easonable conjecture that considering how and why we use these indirect question idioms should be a source of insight into how we should understand knowledge. (Williams 1973, Hookway 1990, 1995, Schaffer 2007) Whether knowledge sentences taking an indirect question complement are the fundamental ones or not, it is clear that the study of questions and their answers should contribute to our understanding of at least some of the sentences w e use to expr ess and describe states of knowledge. The first three considerations bring out ways in which the understanding and evaluation of questions is fundamental to the ways in which w e evaluate inquiries and deliberations. The role of questions in kno wledge sentences does not ob viously have this character. However the fact that questions have these two sorts of roles may provide a clue to how we use the concept of knowledge to make evaluations. The facts about knowledge sentences suggest that kno wledge is a r elation betw een an agent and a question which can be understood by exploring what counts as being in a satisfactory relation to a question when it is being deployed in one of the other three fashions. Perhaps I know some propositions when I have successfully completed an inquiry into it or would have no epistemic reason to carry out such an inquir y. Alternatively, exploiting the eliciting use of questions, perhaps w e have knowledge concerning some question when we can provide an accurate answer to it. Or thirdly, if we take note of the fact that questions hav e a S ocratic role in the r egulation of inquir y, we might suppose that knowledge with respect to a question requires that we can provide an answer to it that will not succumb in the face of fur ther questioning. Our explanation of knowledge may refer to the different uses that questions have in epistemic evaluation. So: what capacities must w e possess if w e are to use interr ogatives in epistemology in the ways described. Among these capacities must be an ability to recognize whether something is an answ er to a given question. 5

Another concerns when we understand a question well enough to be able to ask it and tr y to answ er it. S ection 2 identifi es some complexities in our understanding of questions by considering how we understand ‘who’ questions. Section 3 emphases the importance of the concept of a ‘relevant answer’ to a question and considers some of the capacities w e must possess in order to identify relevant answers. It will be a consequence of these discussions that the question/answer relation is more a matter of pragmatics than semantics and section 4, which considers the cir cumstances under which a question is ‘real’, develops some implications of this. 2. Two kinds of who-questions Much work on the logic of questions has made the assumption that understanding a question inv olves grasping the range of possible answ ers that it has. Indeed some theorists hav e wanted to identify questions with the set of their answers (Harrah 1984). And early work often concentrated on sentences that have a sort of quantificational structure (Belnap and Steel, Bromberger 1992). It is suggested that the interr ogative ‘When will you arrive home?’ contains the open sentence ‘You will arrive home at x’. The question is then expressed as: (Which x such that x is a time)(You will arrive home at x)? Then any sentence obtained b y replacing the variable in ‘ You will arrive home at x’ by an expression referring to a time will be a possible answ er; and, if the sentence is tr ue, it is the (or a) corr ect answer. Attractive as this approach is, it fails to engage with the most impor tant issues about the understanding of questions. This section will defend this claim, and identify some of the issues we do need to think about, through an examination of ‘who’ questions. Suppose I am the detectiv e who, at the end of my inv estigation, asks everyone who was in the house at the time of the shooting to gather together in the library, I then announce that I know which of the people waiting nervously in the library is the killer. And one of them asks me: So which of the people gathered here committed the crime? We can think of this as a ‘ Which-of-these-people’ question. A question of this form does fit the quantificational pattern:

6

(Which x such that x is a person in this room)(x killed the Colonel)? The question is sound so long as one of the people pr esent is the villain. And a correct answer to that question will identify which of those people it is.2 In giving my answer, I can pick this person out in any of a number of ways. I might identify them as the butler , as Jones, as the man sitting in the chair by the door, as ‘That man’—pointing at Jones—, as the only person who had served with the victim in the army, and so on. So long as all of these responses pick out the same person, it is best to see them all as different ways of giving the same answer. Although these are all different versions of the same answer, it does not follow that they are equally good. If no one but myself is aware that the villain served with the Colonel during the war, an answer that picks him out through this characteristic is correct but useless to those who ar e waiting for my r esponse. Indeed, my selection of one way of giving my answer in preference to others will normally implicate that the chosen one will be particularly salient for the audience. It is easy to imagine circumstances in which the audience is misinformed about the characteristics of their fellows and my correct answer sends most of them off in the wr ong direction. A good pr esentation of the corr ect answer should enable the audience to pick the mur derer out fr om their fellows themselves, a poor presentation of that answer may not give them this ability. If the answer is presented well, then the audience will kno w which of these people the murderer is. If it is presented poorly, then I am currently very unsure whether they would.We can easily see how someone else could learn who the murderer was from his or her testimony, so long as that person has appropriate background knowledge. What is important here is, first, that the structure of the question gives clear sense to what would count as a correct answer: it is an answ er that uniquely picks out the member of the universe of interrogative quantification who satisfi es the attributive presupposition of the question. S econd there can be a v ariety of ways of pr esenting the same answ er. This gives rise to pragmatic issues: some ways of presenting the answer will meet the questioner’s needs and others may not. And some may meet these needs better than others. Thus responding to attempts to elicit information b y trying to answ er questions of this sor t requires us to hav e the ability to 2. Such questions raise a host of interesting issues. Suppose my answer was: ‘The person who killed the Colonel killed the Colonel’. Is that an answer that is correct but useless? Or is it not even a correct answer because it is empty: unless my questioner already knows the answer to the question, he will not be able to make anything of the answ er. I shall not settle this issue here.

7

identify which answer is correct and to identify which way of pr esenting that answer will be appr opriate. We need to take account of what information the questioner already possesses, and also what they propose to do with the information that they obtain from me. Similar points hold if we think of the question as setting the aim of an inquiry rather than being used to elicit information. Suppose that, in my role as detective, I gaze at the assembled people and ask myself: which of these people killed the colonel? I reflect on the evidence I possess and try to work out which of them it is. O nce again, some ways of formulating the correct answer will not satisfy me: perhaps I need a formulation that will enable me to place my hand on the killer’s shoulder and read him his rights and some answers will not put me into this kind of demonstrativ e contact with him. In other circumstances I may just need a name, and I can then trust my colleagues to pick him out when he tries to boar d a plane. It is often assumed that who-questions work in the same way as whichof-these-people questions: correct answers must always succeed in picking out some member of a univ erse of discourse. I t is easy to see that this is not true. Suppose that we hear the doorbell ring and, after I return from answering it, you ask me who was at the door. In many cases, my answer will not succeed in picking out a unique individual y et will not be an evasion of the question. I may tell y ou that it is a J ehovah’s witness, that it is one of the children from next door, that it was a postman, that it was someone asking for directions and so on. The characterisation I offer is an indefinite one, but is exactly what is required. Other examples are: Who is Peter? Peter is a bricklayer. (Belnap 1982, 195) Who is Tully? A Roman statesman and orator. (Stampe 1974, 168–9) Denis Stampe calls such questions ‘ predicate-wanting’, and he follo ws Belnap in thinking that ther e is a syntactic ambiguity in who-questions, even if it is not evident from surface grammar. Some are predicate-wanting questions and others more closely resemble the which-of-these-people questions that we discussed above. Another possibility is that a ‘who’-question always requires a characterisation of the person at issue, and that pragmatic contextual factors determine what sort of characterisation is required: do we need one that uniquely picks out the person in question? O r do we need one that provides an attribute of the person?We do not need to resolve this issue of ambiguity here. The important point is that the predicate-wanting cases raise some issues that are interesting for epistemology. 8

As before, the same answer can be presented in different ways and the way in which we should present it depends upon which is more likely to enable the questioner to grasp the answer. I might describe some Roman politician as a friend of Tully’s, or I might describe him as a friend of Cicero’s. In either case, I ascribe a pr operty to him. M any would hold that, in each case, I ascribe the same property to him, in which case I give the same answer to the question. B ut it is easy to see that someone who was ignorant of the fact that Cicero was Tully may find one of these ways of presenting the answer much more useful than the other. However this example is less decisive than the earlier ones. This is because the criteria for same person are much clearer (at least in practice) than those forsame property. It is easy to see that some might dispute that, in the case described, the answers are the same and the properties they ascribe are the same. Indeed, identity of answers may often be an indeterminate matter. If I tell you that Peter works on a building site, is this a diff erent answer from the one in Belnap’s example or a less precise way of giving the same answer? Suppose we adapt Stampe’s example so the answer claims that Tully was a Roman politician and speechmaker, there may be room for dispute about whether that is the same answer (differently presented) or a different answer. In our which-of-these-people question, the range of possible correct answers was carefully circumscribed within the question: in answering the question, we can pick out anyone, so long as they possess the property of being in that room at that time. With a predicate-wanting question, the question itself gives very little guidance about what sorts of predicate can be used, about what sorts of pr operties and characteristics can be picked out in giving our answer. There may be no clear way of identifying whether something is an answ er to the question—or whether it is an answ er that is wor th taking seriously. Why should I tell you that Peter is a builder, rather than that he is a father of two, a M anchester United supporter, a keen cyclist, a Jehovah’s witness …? The list of candidates is potentially endless. I t seems to be possible to find circumstances in which each possibility would be a r elevant answ er and others wher e it appears not to be an answ er at all. These examples enable us to identify some different ways in which our understanding of a question is influenced by context. These differences are reflected in the capacities we must possess if we are to raise and respond to questions in inquiry. First, if I ask ‘Which of these people committed the crime?’, the range of possible answers depends upon who t‘hese people’ are, and this will vary from case to case. Familiar kinds of context-relativity in 9

the use of indexicals, demonstratives, vague words, etc., can all infl uence the set of possible answers associated with such a question. Second, once we have determined the range of possible answers in a case like this, there are further questions about how, in context, a particular answer should be expressed. Effective eliciting of information and inquiry will be hindered if w e lack the capacity to identify ho w answ ers should be formulated. But, third, who-questions (especially pr edicate-wanting ones) intr oduce a further level of context relativity. In these cases, we need the concept of a relevant answer: the question seeks information about someone, and the conversational context determines what sor t of information is r equired. Answers must provide salient information, and such salience is a function of the state of the conv ersation or inquiry. It is not just a matter of ho w the answer should be expressed; it is a matter of what counts as a serious answer to the question at all. Most interesting questions have this character: a why-question typically requires a salient or important cause or explanatory factor; a how question demands a relevantly important feature of how something happened. One source of clues concerning the r ole of context in guiding the use of interrogatives is provided by recent work in pragmatics on the nature of focus. A word or phrase is the focus of an utterance when it is given a sort of pitch accent, something that lends it prominence within the intonation contour of the utterance (G lanzberg 2005, 75). Consider the diff erence between the following two utterances: JOHN likes Jane John likes JANE It is natural to see the former as an answ er to the question ‘ Who likes Jane? ’ and as eliminating alternativ e statements to the eff ect that some other person likes Jane; and the latter may answers ‘Who does John like?’ and it rules out propositions to the eff ect that he likes people other than Jane. It is often suggested that when a wor d or phrase is focussed within an utterance, this introduces a range of alternative propositions with which it is being contrasted. The verb might also be focussed: ‘John LIKES Jane’ may rule out the possibility that John loves Jane, or regards her with contempt, and so on. This has a bearing on issues about when questions can legitimately be raised during discussion and inquir y: generally, questions intended to challenge a claim will hav e a diff erent target accor ding to which part of the claim is focussed. The observation that John was v ery 10

friendly towards Mary, for example, might challenge ‘J ohn likes JANE’ but not challenge either of the others. These patterns of pitch accent in an utterance r elate to our inter ests in the r ole of questions in sev eral ways. First, the focus of an utterances provides evidence of the questions (explicit or implicit) to which it is an answer. In the cases discussed abo ve, assertions are evaluated as answ ers to the questions ‘ Who likes Jane?’, ‘Who does John like?’, and ‘ What is John’s attitude to wards Jane? ’, respectively. Indeed theorists often insist that focus helps determine ‘question-answer congruence’. If we are asked ‘Who does John like?’, then ‘JOHN likes Jane’ is not a felicitous answer, while ‘John likes JANE’ is. (G lanzberg 2005, 77). S econdly, the pattern of focus of an utterance and hence the question (implicit or explicit) to which it responds, affects which questions can then be raised as felicitous challenges to the claim. I may challenge the claim that JOHN likes J ane by presenting evidence that J ane has other admirers; and I can challenge the claim that J ohn likes JANE b y suggesting that he likes other people too. We can see this as revealing how the identity of the question to which an assertion is a response can influence the kinds of challenges that can be raised to the asser tion: we challenge the claim as an answ er to a specifi c (implicit or explicit) question; and, in doing this, we often rely upon our sensitivity to the focus or intonation contour of the utterance. Although most discussions of these phenomena concern the ole r of focus in conversation, it is also found when w e try to solve problems through individual inquiries—focus has a place even in silent soliloquy. We can use focus, inter alia, to keep track of the questions curr ently under consideration in the course of an inquiry or deliberation. This helps us to identify some of the capacities we must possess if we are to respond to requests for information framed in questions of diff erent kinds. We must understand the interrogative sentence and thus identify the sort of thing that can serve as a candidate answer. We must draw on background knowledge and our understanding of the context to determine which answ ers are relevant to the question as posed. H aving ascertained which of these answ ers are correct, we must then work out which of them is the best to off er to the questioner, or to adopt at this stage in our deliberations. And, when an answer can be expressed in different ways, we must be able to determine which formulation is appropriate, given our cognitive projects. One issue, brought to the surface by our remarks on focus, concerns the different ways in which context, including the context provided by the current inquiry, influences our understanding of questions and their answers. 11

3. Relevant answers If asking and answering questions is fundamental to our practice of epistemic evaluation, then we need to ask how we can settle issues about what counts as a r espectable and relevant answer to a question. I f we have no confidence in our ability to do that, w e can hav e no confi dence in our ability to carry out inquiries and deliberations responsibly and effectively. We can now see that our ability to do this—whether in conv ersation or in the course of solitar y inquiry—depends upon our being able to solv e some distinct problems. First, then, ther e is a range of issues concerned with identifying sets of possible answers. The most straightforward way of doing this inv olves identifying a universe of discourse—or, more generally, with picking out a set of objects. When I ask ‘ Which of these people …?’, then I need to settle what counts as ‘these people’. I must determine who is a candidate for being the answ er to the question; whom must I be able to eliminate if I am to identify someone else as my answ er? Related problems about a ‘universe of discourse ’ arise when w e consider ‘where’ questions: if w e answer such a question by identifying a location, then there are questions about how large and appr oximate locations ar e. If you ask me wher e a particular book is, do I answer successfully when I tell you that it is in the library? Or is it required that the location be pinned down to a particular place on a particular shelf ? Questions like ‘predicate-wanting’ ‘who’ questions raise more difficult issues. We need to identify a range of alternatives here, a range of properties or predicates, rather than a range of objects. I n the case we have just considered, there is a sortal which determines what sorts of thing belong in the set of alternatives; with a ‘who’ question, the only constraint is that we are looking for a potential pr operty of a person, for example, which may, but need not, be uniquely individuating. What sort of property or predicate is appropriate is largely determined by contextual and pragmatic considerations.3 ‘Why’ questions ar e more complex still: pr esumably we look for a range of ‘possible explanations’ or something like that. When I 3. We can distinguish two strategies for dealing with these cases. Suppose I notice someone I don’t recognize in the department and ask you who it is. Suppose you answer: it is a woman in blue dress. This information is useless to me, and you are aware of this. Have you given me a correct answer, but an irrelevant, useless and uninteresting one? Or have you not given an answer to my question at all? I shall not take a stand on this issue her e—success at inquiry requires an ability to identify ‘relevant’ answers, and that is my concern.

12

ask a ‘who question’ or a ‘where question’, I am aware of a range of alternatives which can be eliminated, ev en if I am not awar e of many of the relevant predicates or of which places are possible locations for a book (for example). It may not occur to me that I may have left it on the bus.When I ask a ‘ why question’, I may not be awar e of any possible explanations, or of any that I cannot eliminate. That is why ‘why questions’ can be so hard. The other issues arise because the same answ er can be expressed in different ways, employing, for example, a different vocabulary. An answer can thus be infelicitously pr esented, even if it is corr ect. Thus effective inquirers need to be sure that they can formulate answers in ways that fit them for their intended use. How does context settle these things? How should we think about the context here? It is common to think of the context of an utterance as an evolving body of presuppositions, of things that are taken for granted by the participants, and, perhaps, are known to function as a body of shared background knowledge. When we consider solitary inquiries, we can recognize a limiting case of this. One possibility is that the context is composed of propositions that are accepted or matters of full belief , perhaps things we take to be certain. Alternatively, we can take the context as containing propositions that are treated as uncontroversial. What sort of information could be contained in these presuppositions which could be relevant to settling what sorts of answers to a given questions are to be taken seriously. There can be information about what sorts of answers are recognized as not real candidates, as not being r elevant at all. There can also be information about which candidate answ ers have already been r uled out. There might be information about what sor t of answer the questioner wants, and about what sort of answer the respondent expects the questioner to want. There may be assumptions about what the questioner wants to do with the information he hopes to obtain from his question or about what he ought to plan to do with that information. If we are concerned with the solitary case, the questioner and respondent will be the same, and this ur les out one source of error and misunderstanding. It is not possible that questioner and respondent have different views about what sort of answer will be appropriate. But, even in this case, it is possible that the agent will be mistaken about the sor t of answ er that is required. Suppose I ask myself who some person is, intending to use the information in order to predict what attitude they will take on some political issue. Suppose, too, that I believe that the best clue for making such a prediction is information about someone’s employment. So I assume that 13

the range of serious answers to my question concerns different sorts of job that the person could have. In fact I am wrong: information about where they live, or about their r eligious affi liations, will be far mor e valuable. This is the information I need when I ask my question; although it is not the information I actively seek. So long as the answer has an appropriate form, then whether an answ er is relevant or appropriate is a function of what it is to be used for, not of whether it is what the questioner explicitly has in mind. Talk of ‘ presuppositions’ here needs to be tr eated with caution. P hilosophers generally think of presuppositions as things that are believed to be true, and, when I described a context as providing a set of presuppositions, I went along with this. We need to allow that the presuppositions that guide the use of questions in inquiry can often be suppositions rather than presuppositions. Much counterfactual reasoning involves considering a hypothetical possibility, without being committed to its tr uth. Against the background of that hypothetical possibility , I can then inquir e into what the correct answers to various questions would be, if that hypothetical situation obtained. Within that part of the inquiry, we treat the supposition as fixed—at least for a short time—without being firmly committed to its truth. Indeed, if we are conducting a deliberation that involves a reductio ad absurdum, then we may treat something as true when we firmly believe it to be false. This does not aff ect the previous discussion. We have already noted that inquiries and deliberations ar e extended, structured goal dir ected activities. This means that the context of the investigation changes rapidly, as possibilities are considered, investigated and eliminated, as challenges are raised and responded to. This means that our understanding of what the context of an utterance is must be equally structured and complex. As well as taking note of the information erlevant to the interpretation of a par ticular assertion, judgment or question, w e need to keep track of its place in the structure of the inquiry as a whole. I can explain this by means of a simple example that illustrates how different who-questions can occur in the same inquir y while having diff erent sets of alternatives associated with them. Suppose I am a detective, trying to identify who murdered the colonel. I am told that just three people had the opportunity to commit the crime, so, I ask the question ‘ Who killed the colonel?’, against the backgr ound assumption that there are just these three candidates. Let us suppose they are Atkins, Burroughs, and Car ter, and I don ’t know much about them apart from that. I plan my inquir y, intending to fi nd which, if any , of 14

the three candidates had a motive to commit the crime. I need information about these people that will help me to solv e my problem. Perhaps I begin b y asking ‘ Who is Car ter?’, seeking an answ er that will suggest ways of establishing whether she has a motiv e to kill the colonel. But no such answer is forthcoming, so I turn to the second of my three questions, ‘Who is Atkins?’ and, we may suppose, arriv e at an answ er that helps to make it unlikely that Atkins is the criminal.When I then ask ‘Who is Burroughs?’, I am told that Carter knows more about him than anyone else, and I should seek information from her. In order to do so, I ask ‘ Who is Carter?’ again, this time as a means to obtaining information about Burroughs. In that case, ‘The woman o ver there’ may be a fully satisfactor y answer—it enables me to pursue my investigations about who Burroughs is. Having made some progress with that, I return to my question ‘Who is Carter?’, and once again, I have to admit that I don’t know. The information I possessed about her spatial whereabouts is irrelevant to my higherlevel inquiry into her likely guilt. We can picture the structure of the inquiry as comprising six levels: 1. Target: Who killed the colonel? 2. Subordinate inquiry: Which of A, B and C killed the colonel? 3. Subordinate inquiry: Which of A, B, and C had the motiv e to kill the colonel. 4. Three subordinate inquiries: Did A have a motive? Did B have a motive? Did C have a motive? 5. Three more subordinate inquiries (each subor dinate to the question above it) Who is A? Who is B? Who is C? 6. Yet another level (subordinate to the question immediately above it) Who is B? In understanding, and answering the questions, we need to keep track of this overall structure. The sort of information required of an answer to ‘Who is C?’ will v ary according to the lev el at which it is raised and the other questions to which it is subor dinate. I can answ er this question at level 6, but not as a question at lev el 5: accor ding to the lev el, it makes salient a diff erent set of alternativ e possibilities. Unless we can track the changing questions expressed by any given interrogative—unless we can track the shifting range of answers that are relevant to them—our ability to inquire effectively will be impaired. 15

4. Real questions and understanding We have discussed in general terms some issues about questions, answers and their roles in inquir y and deliberation. I n this section w e shall consider two fur ther questions. The first of these concerns what is inv olved in understanding questions. S ince we can use questions ev en if w e cannot identify their possible answ ers, it would be unr easonable to identify understanding a question with knowing what its possible answers are. Nor would it be realistic to claim that we understand a question only when we know how to answer it: much research in the sciences and philosophy is devoted to trying to answer intelligible questions that we don’t yet know how to answer. So, what is required for understanding a question? A erlated issue concerns what is required for a question to be real, for us to be able to ‘get a grip’ on it and take seriously the idea that we might try to answer it. Yet another issue concerns reasons for doubt: when is it reasonable for us to address a question, recognizing that it needs an answer and admitting that we do not currently have a secure answer to it? This section begins to address some of these questions. I shall star t by exploring some ideas and concepts used b y Nicholas Jardine in Scenes of Inquiry. We shall then use these as a tool for making sense of the cases w e have been considering. J ardine’s purposes ar e different from ours: his main concern is with the r ole of questions in the sciences and in the pr ospects for making the understanding of questions a central task in the historiography of science. He is interested in how we can understand the questions that were addressed by Galileo or Paracelsus or, his main example, eighteenth century natural history. Some questions ar e real, others ar en’t. If we ask who disco vered that the Thames was the longest river in Europe, the question is unreal because it rests on a false pr esupposition. When the logical positivists identifi ed questions such as ‘Why is there something rather than nothing?’ as nonsensical, we can read this as implying that they are unreal, that they have no true answers. Jardine begins with the mor e contextualised notion of a question being ‘locally real’ (1991, chapter three). A question Q is real for a community when they can ‘get a grip on it’. This rules out questions that are perceived as ‘utterly inscrutable’ or as carr ying false pr esuppositions, and is explained as follo ws. First, an answ er is straight when ‘it is both direct and adequate, in the sense that it conveys all the information that is called for’ (57). And an ‘evidential consideration E is relevant to a question Q? in community C just in case E is taken in C to fav our one 16

over (the disjunction of ) the other asser tions taken in C to be straight answers to Q?’. A question is locally real in C when there is an evidential consideration E that is held to be r elevant to Q? (57). S o a question is locally r eal (for a community) when that community r ecognizes some things as straight answers to it, and recognizes that some evidence would favour one of those answers over the rest. This gives some minimal sense to the communities knowing how to try to answer it. It is compatible with this that their inquiry may not get far, and also that the question actually has no correct straight answer: in the light of fur ther information it may emerge that it is not a real question at all. We can understand both questions that are real and those that are unreal—think of the example about the river Thames above. Jardine takes these in turn. His explanations make use of the idea of ‘cognitive frame’: this is the totality of the beliefs and commitments of a community . It includes general beliefs about what sort of evidential considerations are relevant to what sorts of questions, as w ell as the ‘beliefs and methodological commitments’ used by members of the community to justify those judgments of relevance. It thus includes what E rnest Sosa would call an ‘ epistemic perspective’ (1991, 97). J ardine suggests that w e understand locally r eal questions when we understand ‘the ways in which they ar e made real by our cognitiv e frame’. O ur understanding of unr eal questions is much more varied (Jardine 1991, 63ff), but we shall note just one of these. We understand the question ‘ Who discovered that the Thames was the longest river in Europe?’ by seeing how it would be real in a cognitive frame that differed from our own in including the belief that the Thames is the longest river in E urope. In general, w e can understand unr eal questions by considering variants of our cognitive situation in which they would be real; and often this will be very difficult. These questions of understanding and intelligibility of questions arise in two different contexts. First there are questions about the sort of understanding possessed by ordinary speakers involved in eliciting information from each other and ordinary inquirers (including ordinary scientists) trying to solve problems. Second, there is the sort of understanding required by historians and social scientists tr ying to understand the practices of inquiry employed in communities, perhaps in communities remote from their own. As philosophers we can describe both sorts of understanding. As Jardine records when discussing the difficulties of historical interpretation, most local methods are ‘routine and habitual, under normal circumstances rarely discussed or even articulated by their exponents’. And, he suggests, 17

it is these methods that hav e the largest role in determining which questions are real for a community. He mentions ‘tacit criteria by which [scientists] assign burdens of proof and judge the competence and wor th of other practioners and their findings’, tacit conventions of how to present materials to other practioners, and the role of ‘habitual competence in the use of instr uments’ in guiding experimental wor k (1991, 90). Although our concerns are not with scientifi c knowledge, however, it is easy to see that everyday inquiry and evaluation of argument is similarly guided b y what is routine, habitual, unarticulated, and, in different ways, tacit. The task of the historian is to make explicit what is tacit and habitual; but our everyday dealing with questions achieves understanding without this. We understand a question when our tacit and habitual understandings enable us to see ho w some evidential consideration or r eason is r elevant to its answer. It is clear, I think, that questions can be r eal for us that we don’t understand; we don’t see how or why the r easons favour some answer to our question. The examples of who-questions that we have just been looking at suggest a further dimension of understanding. I f we need to make judgments of relevance in identifying which things ar e actually candidate answ ers, we may also need to grasp what it is that makes something an appr opriate or good answer. Jardine’s account takes for granted the ideas of a ‘straight answer’, and I have been suggesting that for understanding many of our normal epistemic encounters with questions, this is not something w e can do. There are issues about how to identify what question is expressed by an interrogative, as well as issues about how, once we have identified a question, we can get to grips with it. The fact that a question is r eal for us does not entail that w e have any reason to tr y to answer it. We may not fi nd it interesting. There may be none of our other concerns to which its answer would be relevant, either by providing information that would be useful for practical purposes or yb offering information that will contribute to our ability to solve theoretical problems and better understand our surr oundings. Jardine distinguishes our scene of inquiry (the range of questions which is real for us) from our scene of response (the questions in our scene of inquir y that ‘engage with some aspect—cognitiv e, moral, aesthetic, etc.—of [our] attention and concern’) (1991, 68). These are the questions that ar e real in a str onger sense: they are experienced as needing an answer (rather than just as possessing one.) S uch questions ar e relevant to our concerns, and w e have standards (tacit as w ell as explicit) that go vern entr y into our scene of 18

response: the question ho w many grains of sand ther e are on B lackpool beach, for example, does not belong there. If we define the local reality of questions as Jardine has, then there are questions which are not locally real, but which engage with our attention and concern. It can be right for me to tr y to answer a question that does not belong within our ‘scene of inquiry’. The kind of case I am concerned about is one where the only answers I can think of (if any) can be eliminated, but I am confident that there is a correct answer to be found. And we might add, I have no real sense of where this unknown answer lies. Sylvain B romberger suggests that explanator y why-questions ar e responses to particular kinds of cognitive predicament. His example is his ignorance with respect to why kettles emit a humming noise just befor e the water begins to boil (1992, 27). The character of his pr edicament is that he believes that this question admits of a right answer (it is ‘sound’), but he can think of no possible answ er to which he cannot see decisiv e objections (1992, 29). Such a case may meet Jardine’s conditions for being locally real: Bromberger is aware of evidence that has been used to evaluate at least some answ ers; but he is not awar e of ho w that evidence can be used to evaluate answers that we may come up with in the future. His evidence eliminates all the options he can think of. It can still be rational to inquire into such questions, guided by the confident hope that answers will emerge, and by the hope that we will evaluate those answers in light of evidence. B ut in that case, our strategy of inquir y will be primarily focused, not on refuting or defending particular answers, but on looking for possible answers. So what do we (or should we) do in such cases? The motivational force rests on a substantial presupposition: there is a real question here, and we can discover what it is, and when we do, it will be a question that we can handle directly. And the first task of inquiry is to create a situation in which our interrogative expresses that real question. In order for this to happen, we may hav e to r evise answers that, w e had thought, had been r ejected. We may have to uncover the possibility of answers that, at present, we are not aware of. We can thus see the initial interr ogative, which expr esses no real question, as setting in motion a course of activities designed to enable us to find such a question. This might involve exploring the subject matter, hoping that useful ne w information may emerge. I t might also exploit our understanding of the unreal question. If we try to understand how the question could be r eal in counter factual circumstances, where we lack information that w e currently possess, then this can off er analo19

gies and models to exploit in making our diff erent attempts to make the question real. If a question is locally real, then, in appropriate circumstances, we can have reasons to try to answer it. It is not necessary, in such cases, that we know how to answer it, nor even that we think we know how to answer it. It is required that we can rationally hope that we can fi nd out how to answer it. And this hope can be supported by the fact that, at least in counterfactual situations, we can conceive of evidence that would be r elevant to answering it (or to answering analogous questions.) But our reaching a position where we are able to answ er it will depend upon fur ther cognitive advances or upon the cooperation of the world. O ur understanding of a question is thus heavily dependent upon our background knowledge and upon our confidence in the possibility of making appropriate progress towards a better understanding. In tackling the question we rely upon our background knowledge of how similar questions have been, or could have been, answered in other circumstances. Questions of this kind may be much mor e common, and much more important, than may be expected. Their features may be shared by questions that superfi cially look much mor e straightforward. The conclusion to draw here is that the semantics of interrogatives does little to guide us in recognizing whether something provides an answer to a given question. What the use of an interrogative is looking for in the way of an answer is largely determined by considerations that can be described as pragmatic: the motivations and interests of the person who asked the question; the shifting concerns that r esult fr om the dev elopment of a conv ersation, discussion, or inquiry; and the background knowledge of questioner and respondent. Answering questions, like asking questions, usually inv olves judgment, the ability to make assessments of particular cases while taking account of a broad context.4

4. I have given talks on this material on a number of occasions, most r ecently in Sheffield and Valencia, and at a confer ence on pragmatism in L ublin. I am v ery grateful for the help I have received from comments made on those occasions.

20

REFERENCES Belnap, N. 1982. ‘Declaratives are not enough’, Philosophical Studies, 59, 1–30. Belnap, N. and S teel, T. 1976. The Logic of Q uestions and Answers. New Haven: Yale University Press. Bromberger, S. 1992. On What We Don’t Know. Chicago: University of Chicago Press. Glanzberg, M. 2005. ‘Focus: a case study on the semantics-pragmatics boundary’, in Szabo, Z. G. (ed.) Semantics versus Pragmatics, Oxford: Oxford University Press: 72–110. Grimaltos, T. and Hookway, Ch. 1995. ‘When deduction leads to belief ’, Ratio, vol. VIII: 24–41. Harrah, D. 1984. ‘The logic of questions ’, in Handbook of P hilosophical Logic volume II, in Gabbay, D. and Guenther, F. (eds.) Dordrecht: D. Reidel. Hintikka, J. 2007. Socratic Epistemology: E xplorations of Kno wledge-Seeking b y Questioning. Cambridge: Cambridge University Press. Higginbotham, J. 1993. ‘Interrogatives’, in Hale and Keyser (eds.) The View from Building 20. Cambridge MA: MIT Press: 195–227. Hookway, Ch. 1990. Scepticism. London: Routledge. — 1995–96. ‘Questions of context ’, Proceedings of the A ristotelian Society , vol. XCVI, 1–16. — 2005. ‘Interrogatives and uncontrollable abductions’, in Semiotica, 158, 101– 116. — 2006. ‘Epistemology and inquir y: the primacy of practice ’, in Hetherington, S. (ed.) Epistemology Futures. Oxford: Oxford University Press: 95–110. Jardine, N. 1991. Scenes of Inquiry. Oxford: Oxford University Press. Willliams, B. 1972. ‘Knowledge and Reasons’, in G.H. von Wright (ed.) Problems in the Theory of Knowledge. The Hague: 1–11. Schaffer, J. 2007. ‘Kno wing the answ er’, in Philosophy and P henomenological Research LXXV, 383–403. Sosa, E. 1991. Knowledge in P erspective. Cambridge: Cambridge U niversity Press. Stampe, D. 1974. ‘Attributives and interrogatives’, in Munitz, M. and Unger, P. (eds.) Semantics and Philosophy, New York: New York University Press.

21

Grazer Philosophische Studien 77 (2008), 23–44.

THE FIXATION OF KNOWLEDGE AND THE QUESTION-ANSWER PROCESS OF INQUIRY Claudine TIERCELIN Institut Universitaire de France Université Paris XII, Institut Jean Nicod Summary The aim of the paper is to pr esent some impor tant insights of C. H ookway’s pragmatist analysis of kno wledge viewed less in the standar d (Gettier) way, as justified true belief, than as a dynamic natural and normative question-answer process of inquiry, a reliable and successful agent-based enterprise, consisting in virtuous dispositions explaining how we can be held responsible for our beliefs and investigations. Despite the merits of such an appr oach, the paper sho ws that it may be inefficient in accounting for some challenges posed by scepticism or by the nature of epistemic normativity. In which case it might be premature to propose it as a ne w conception of kno wledge against the standar d one and worth considering a different, though still pragmatist, strategy, in which inquiry would aim at the fi xation of knowledge, still viewed as justifi ed true beliefs, i.e critical commonsensical, warrantedly asser tible, intellectual and sentimental dispositions for which the epistemic agent, vie wed less as an individual person than as a scientifi c community of inquirers, should be taken as a kno wing and reliable agent, both answerable and responsible for her assertions.

1. Introduction C. Hookway’s conception of knowledge retains many pragmatist insights and develops itself as an original basically virtue-theoretic strategy, viewing knowledge not so much as a justifi ed true belief enterprise, but rather as a question-answer process of inquiry in which the regulation of our affective states is par t and par cel of our theor etical deliberation, stating the right limits to be imposed both on our reasons to doubt and on our justifications for believing, thus setting the r ules of our practice of epistemic evaluation. Despite the merits of H ookway’s approach, one may doubt

its capacity to justify such a change in our standar d view of kno wledge, and in particular, to overcome the threat of scepticism, as may be shown from other pragmatist strategies concerning the natur e of asser tion, of inquiry, or the position to adopt to wards doubt. This is why it might be worth suggesting a diff erent, though still pragmatist, strategy in which the fi xation of kno wledge would rather be taken as the a im of inquir y, and knowledge itself defi ned in terms of basically commonsensical and critical (justified), warrantedly assertible (true) intellectual and sentimental dispositions and assertions (beliefs) for which the epistemic agent, viewed less as an individual reliable person than as a collective agent, or scientific community of inquirers, should be held both answerable, in so far as her cognitive abilities ar e less r ule-governed than fi xed by the constraints of the real, and responsible, since, even if she cannot control the whole of her epistemic evaluations, she should still be able to contr ol and criticize the methods she is using, and be strongly committed to her assertions, in order to receive credit for them, and be vie wed not only as a vir tuous inquirer but as a knowing and reliable agent. 2. C. Hookway’s conception of knowledge as a question-answer process of inquiry From his 1990 book on Scepticism to his latest writings, C. Hookway has kept underlining that “the focus of our ‘ epistemic lives’ is the activity of inquiry: we attempt to fi nd things out, to extend our kno wledge by carrying out investigations directed at answering questions, and to refine our knowledge by considering questions about things we currently hold to be true” (1994, 211). Without being a “full-blooded virtue epistemologist”, Hookway stresses that the main concern of epistemology should not be with improving a G ettier-type defi nition of kno wledge as justifi ed true belief, but rather “with explaining the evaluations we must be able to make if we carry out inquiries in a r esponsible, self-controlled fashion” (1994, 212). For “we don’t understand a concept like justification until we grasp its role in evaluations employed in ‘ordering’ or regulating inquiries, and we have no need for a theory of justification until it is required to explain these evaluations” (ibid.). Indeed, a fundamental problem for epistemology concerns how it is possible for us to carry out the investigations required for effective and responsible inquiry, and this is threatened by the possibility of scepticism, as Hookway noted in his 1990 book on Scepticism, namely 24

the possibility of our being unable to take r esponsibility for whether w e inquire and deliberate well, or for a distinctive kind of autonomous inquiry. Now sceptical arguments could be r esisted only if w e could make sense of a way in which the role of reflection in the control of inquiry could be limited, since our investigations “depend upon a body of largely acquired habits, whose internal operations ar e opaque to us ”. Thus, at least in so far as Hookway might have been taken to defend a position close to some responsibilist-virtue epistemology, this was based on the claim, as noted in “How to be a virtue epistemologist?”, “not that we should define ‘justified belief ’ in terms of the vir tues, but, rather, that we shall only understand how we can regulate deliberations and inquiries, how we can control our opinions and inquir e well, by giving a central r ole to states such as virtues: indeed, the role of virtues in theoretical deliberation (in identifying reasons for belief , for example), parallels the r ole of vir tues in practical deliberation (in identifying reasons for action).” In search for “a better way of explaining how our deliberations could be so subject to infl uences we can think of as alien, perhaps as heter onomous”, Hookway thus came to “focus on their role in the regulation of inquiry and deliberation and took it that the primary focus of epistemic evaluation was control of activities of deliberation and inquir y, rather than the (largely thir d person) evaluation of states of belief as ‘justifi ed’ or as ‘kno wledge’” (ibid.). In other words, one should pass from the attempt to explain what knowledge and justified belief are, or to inv estigate how far we are able to possess states of knowledge and justified belief, to the following alternative attempt: to describe and explain our practice of epistemic ev aluation; to inv estigate how far our epistemic goals ar e appropriate and ho w far our ev aluative practice enables us to achieve our epistemic ends. More and more, and in the line of Dewey who took inquiry and questioning as being almost synonymous 1, Hookway comes to argue that if epistemic norms ar e to guide one ’s inquiries, then a cr ucial issue for all candidate analyses of epistemic ev aluation is what makes something count as a correct answer to a question (1999). I t is a sign of epistemic virtue that one asks the right questions when making a claim, and this also gives a way to av oid scepticism: for example, when making a claim about the time of the next train to London, it would be obsessiv e to feel 1. See Dewey’s account of inquiry in his Logic: inquiry responds to an “indeterminate situation”. Rational inquiry requires us first to institute a problem, in effect, to identify a question which focuses the indeterminacy which motiv ates us to inquir y (1938, 107–8); in H ookway 1999, 14.

25

obliged to check that there are no misprints in the printed timetable; but it would be rash to ignor e the possibility that y esterday’s rail strike has disrupted the schedules” (1994a, 1994b). “If I have no response to a challenge based upon this second possibility , I should r econsider my claim. Unless confronted by evidence of widespread inaccuracy in the timetable, I would be irrational to take the former into account. The possibility that I might be a brain in a vat grounds a challenge which I ought perhaps to take account of when engaged in a cer tain sort of philosophical inquir y, but I hope that my informants never regard it as relevant when asked for information about train times. “Context ” in that case, determines what questions (or challenges) I should hav e answ ers to befor e confi dently making a claim to knowledge. So understanding these context relativities requires us to understand which challenges to our kno wledge claims w e should take seriously” (1999, 4). “When we conduct an inquiry, or deliberate on some matter , we attempt to formulate questions and to answ er them correctly. We reflect on what we ought to do in order to arrive at a satisfactory answer. As rational, reflective agents, we monitor our inquiries asking further questions about how well we are conducting them, about the acceptability of interim results and so on. The epistemic status of the conclusions of our inquiry can depend upon how well it has been carried out. Have the right questions been asked and have they been investigated in a r esponsible manner? The obsessive over cautious agent may ask too many questions, considering challenges which a better cognitiv e agent knows can be ignor ed. A rash irr esponsible inquirer is too unr eflective, asking too few questions and unsuspicious of the possibilities of error. My claim to kno wledge expresses confi dence that no questions which need to be raised about my opinion hav e been left unaddr essed, and all those questions which have been considered have been resolved in a satisfactory manner.” (1999, 7–8). In other words, “philosophers have paid insufficient attention to the logic and semantics of questions ” (Hookway 1999, 8). Now, “most assertions are responses to questions” and “rather than seeing the question as containing (implicitly) a proposition, we might (with more justice) see the assertion as containing a question … In different contexts, the same interrogative sentence can be used to ask very different questions … Talking of my attitudes and r esponses to questions helps to str ucture the information I hold which is normally expr essed in terms of beliefs. ” (1999, 10–11). Besides, ignoring the role of questioning “encourages a very static conception of mind and disguises from us the extent to which thinking is an 26

active process of raising questions and trying different ways of answering them … We use the concept of knowledge to identify informants. When I discover that you know something, I can accept y our testimony on the matter…Assertions, when directed at others or to oneself in solitary reflection, typically purport to express knowledge. We should try to understand the concept of knowledge by investigating its role in such practices” (1999, 11). Hence, “the general pattern of kno wledge”, according to Hookway, should rather take the following form: 1. X knows Q 2. X’s answer to Q is p 3. So, p. How does the schema r eflect the classical one in terms of “justifi ed true belief ”? “The three components of the traditional analysis of kno wledge relate to diff erent stages of this schema. The requirement that X believ e that p is a r eflection of the r equirement that p be X ’s answer to Q. The requirement that the belief be true derives from the demand that 3 should follow from 1 and 2; and the justification requirement expresses the need for some pr operty which warrants the asser tion of 1—a pr operty that ensures that whatever answer is offered to our question will be the correct one (cf. Scepticism, chap. 10). These remarks suggest that we should treat knowledge as a relation between an individual and a question rather than between an individual and a pr oposition. The question determines what sorts of challenges should be taken into account and what sor ts of inquiries should be conducted” (1999, 11). From his analysis, Hookway emphasizes the need to focus on adynamic rather than static view of knowledge as inquiry, deliberative investigation, but also on an ev aluation of our epistemic practices, and fi nally on an agent-based model, in which skills in deliberation and inquiry are viewed as vir tues or character-traits (like attentiv eness, fair mindedness, openmindedness, intellectual tenacity and courage) engaging the ersponsibility of the agent taken in a question-answer situation involving the necessary account of the asser tive context: indeed, our abilities ar e relative to our environment and to a lot of background knowledge. Ther efore, we should try to r egulate our inquiries in terms of guidance and pr escription, and determine the right type of conduct we should have: this implies knowing what questions to raise or to avoid, being observant, but also, and importantly, being aware of the meaninglessness of certain claims to knowledge, 27

or, contra the radical sceptic, of some attempts to fi nd endless justifi cations. Furthermore, it involves paying attention to the process of theoretical deliberation, viewing it in parallel with practical deliberation. All this supposes control but also a lot of uncontrolled, unreflective processes. Indeed, much of the knowledge and information that guides us in performing our activities and practices does so without being represented in deliberation. Therefore, we should take account of a twofold consciousand unconscious process, namely a conscious r eflection—crucial to understanding ho w it shapes our habits—though occurring against a background of what seems to be habits of thought and reflection. Indeed, unless we possess the appropriate body of habits, w e seem to be incapable of the kinds of r eflection required for effective deliberation (cf. Dewey): in other words, there seems to be a kind of normal sensitivity to the normativ e demands of r eason. This explains the importance of such phenomena as salience, appropriateness and relevance, but also of epistemic akrasia (Hookway, 2001): indeed, unless we have mastered such normative standards in the form of habits or skills, we will be unable to ex ercise the deliberative capacities that are required for effective actions. As another classical pragmatist (Peirce) had already pointed out, this shows how decisive are such virtues as observance, open-mindedness, etc. In short, on the one hand, “possession of epistemic virtue depends upon the possession of skills and habits whose possession is largely independent of the recognition that some state is, in fact, such a virtue. And possession of these capacities seems to be what is required for confidence in one’s deliberative skills, for example in one’s sense of salience, not to be an impediment to one’s freedom of mind”. On the other hand, such form of habits of thought, skills in the use of concepts and argument forms, ways of ex ercising judgement in w eighing considerations ar e not available to intr ospection or consciousness. As a consequence, complex cognitive achievements can have a sort of phenomenological immediacy which “mimicks foundationalism” (1993, 2003) and ought to be analyzed as such. This is why, contrar y to a conception of disengaged “inquir y”, Hookway is more and more inclined to focus on the analysis of such affective (or metacognitive) states as emotional involvement, anxiety, “feeling of knowing”, “sentiment of rationality” (W. James), which play such an essential role in raising genuine ( contra paper) doubts, in characterizing our general cognitive practices, and even in explaining how our epistemic evaluations are possible at all (2002, 2003, 2008a and 2008b).

28

3. The merits of Hookway’s virtue-theoretic and pragmatist approach Such a pragmatist view of knowledge, modelled on the classical pragmatist account as the general pr ocess which leads fr om an unsettling state of belief-disposition which gave rise to a genuine doubt, to a satisfactor y state of belief (Peirce, Dewey) has many merits: rather than directly focusing on such problematic concepts as “justifi cation”(always threatened by vicious regress, circularity, dogmatic stopping-point, and often leading to some form or other of neo-pyrrhonism), “truth”(with all its paraphernalia of metaphysical obscurities linked with such concepts as correspondence, coherence, but also strict utility or v erification), and “kno wledge” (with all the Plato-Gettier aporia it leads to, whatev er the—coherentist, foundationalist, counterfactual, reliabilist, internalist or externalist—proposed definitions may be), w e might w ell be in a better situation b y vie wing knowledge, not as static belief-pr opositions, but rather as some mental agents’questioning states, related to inquirers directly engaged in practices of epistemic evaluation. After all, as the classical pragmatists have claimed, it might be more economical and less problematic to analyze “truth” in a metaphysically neutral way, either along “redundantist” lines (F. Ramsey) or in terms of “the ideal limit of inquiry” (Peirce) or as stating some minimal “warranted assertability” conditions (Dewey), and insist more on the advantages of a genuine account of the semantics of assertion and of our various methods of inquir y. B esides, if kno wledge is vie wed in r elation with our assertive and questioning practices, not only does it throw light on what counts as epistemic agency, but it aff ords new interesting links between epistemology and the philosophy of mind and even with cognitive psychology and metacognition, since it pr oposes better accounts of the etiology of our epistemic states, thr ough the study of ho w a person formed her beliefs, how sentiments and affective states are linked with our best rational judgements and not opposed to them (as often underlined by both W. James and C. S. Peirce), thus providing plausible explanations of the ways in which some epistemic norms may be related with (or even may emerge from?) our nature (Tiercelin 1997); again, in sho wing how, for example, our investigations still aim at knowledge and truth, or how deliberation is both theoretical and practical, it provides, as all pragmatists recommend, a more convincing account of the ways in which knowledge is tied to action. Finally, in looking at belief-dispositions or habits not only through Aristotelian lenses, but thr ough Peircian lenses, w e get an interesting explanation of the reasons why, although our belief-habits are 29

stable and rigid, they ar e also radically indeterminate, thus allo wing for changes, evolution and learning through new experiences and unexpected encounters which often oblige us to “ overthrow the whole car tload” of our previous beliefs (fallibilism, as the pragmatists claim, being par t and parcel of any knowing experience). 4. Some problems with the question-answer or knowledge-as-inquiry conception However illuminating such an approach may be, it encounters some obvious difficulties: after all, one may object, even though knowledge may be equated with inquiry or with a question-answer process, it does not mean that this is all there is to knowledge or that justification should no longer be viewed as an impor tant issue to settle. A quick look at another pragmatist approach such as Peirce’s, as far as his vie ws on assertion, inquiry or doubt are concerned, might be helpful here. 1. Let us take assertion first. For Peirce, “any symbol involves an assertion, at least r udimentary” (CP. 2.34). E nabling one to distinguish betw een the volitional content fr om the r epresentative or pr opositional content, the symbol ar ticulates the iconic elements ( i.e. the elements which hav e some formal r esemblance with the object) and the indexical elements (i.e .those which have some physical resemblance with the object) of the proposition. Indeed, even if it is upon the symbol that the whole w eight of assertion bears (Brock 1981; Tiercelin, 1993a, 281–305), an assertion has no meaning except through some designation that shows whether one refers to the real universe or what universe of fiction it is about (CP.8.368). Hence the importance of the indexical element of the proposition. More precisely, an asser tion is an act in which the speaker addr esses a listener, formulates a propositional symbol and assumes some r esponsibility concerning the truth of that symbol. Any assertion implies, on the part of the speaker, that he believes or knows what he asserts (knowledge is the norm of assertion) and that he intends to conv ey the same belief and the same knowledge to his listener . Thus, it is fi rst of all the speaker who has the main responsibility: it is his task to eliminate any imprecision that might be an obstacle to communication. This involves, on the utterer’s part, “a voluntary self-subjection to penalties ” in the ev ent that the pr oposition turns out to be false (Ms 517). And P eirce goes so far as to say that such 30

penalties ar e comparable to the legal penalties associated with making a false statement under oath (Ms 517; NEM IV , 249). Why is that so? Because assertion takes place within the context of what Peirce, following the terminology of the M odists, calls a “Speculative Grammar”, i.e. that part of Logic which deals with the formal conditions of symbols that have a meaning (CP 1.559; CP 4.116). Not only has a sign or symbol meaning within the propositional and assertive context in which it is inserted (CP 4.583; cf. CP 4.56, 551): “Thought must have some possible interpretation for some possible interpreter”, wherein lies the very being of its being (CP 4.6), in other words, its dialogical character. But, more generally, Speculative Grammar is identified with an Erkenntnistheorie or epistemology which considers “in what sense and ho w there can be any tr ue proposition and false proposition, and what are the general conditions to which thought or signs of any kind must conform in order to assert anything” (CP 2.206). Its formal or “ quasi-necessary” task being to establish what must be tr ue of the representamina that ar e being used b y a Scientifi c Intelligence, in order for them to embody any meaning whatsoev er (CP 2.229) , i.e. an Intelligence (linked with thecommunity of investigators rather than limited to an individual) which is incapable of intuition, in accor dance with the conclusions established by the three 1868 articles of The Journal of Speculative Philosophy (W2, 193–272), and which cannot possibly learn except by following the rules of inductive, abductive and deductive inference, as applied to experience; an Intelligence too, which has accepted certain aims and methods, among others the principle on which a discourse is meaningful only if its deliberate aim is to put the process of rational inquiry at the service of knowledge and tr uth, which can only be r eached through self-control processes2. All of this presupposes on both parts that the speakers involved in assertion, have a cer tain competence and that they par take to a community of ideals and aims of speech. Both want to communicate, to learn and to know, that is, to try and suppress all kinds of ambiguity that might creep into the rational pr ocess and br eak communication (due to the fact, in particular, that no sign, for Peirce, is either totally determinate, or totally indeterminate). Indeed, any symbol, or sign is capable of determining a further symbol which interpr ets it or translates it, so that it is at least 2. This is why, for Peirce, such a Grammar involves at least, a theory of communication, a theory of the norms that go vern communication, a theor y of pr opositional symbols, a theor y of truth, of meaning, of belief , and of kno wledge and a theor y of vagueness as applying to all signs (Brock 1975, 129; 1979; Tiercelin 1993a, 258–334).

31

potentially indeterminate.3 At the same time, “Honest people, when not joking, intend to make the meaning of their wor ds determinate, so that there shall be no latitude of interpr etation at all” (5.447). Thus, it is the context or the situation of asser tion that gives the rules of the right functioning of vagueness and generality. But the situation is far fr om edenic: the question is not so much to describe a situation of communication or of dialogue between speakers that care about one another than to provide the rules of a game. As R. H ilpinen (1982) has pointed out rightly, such analyses have much in common with the strategy adopted b y Hintikka in his G ame-Theoretical semantics (1979). B ut why is that so? B ecause, what is at stake, is not any kind of communication or some Habermasian quiet ethics of discussion: it is the communication of truth. Now, for Peirce, truth goes hand in hand with the adoption of beliefs: ther efore, the speaker must, one way or another, have his belief adopted by his listener. For the aim of communication is nothing, but an “ endeavour to make the person addressed (i.e. the interpreter) think in a cer tain way”, that is believe something.4 No wonder then if the asser tion fi nds its expr ession in descriptions which have more in common with confl ict than dialogue. The speaker 3. “A sign (under which designation I place ev ery kind of thought, and not alone external signs), that is in any respect objectively indeterminate (i.e. whose object is undetermined by the sign itself) is objectively general in so far as it extends to the interpreter the privilege of carrying its determination fur ther. Example: ‘Man is mor tal’. To the question, ‘ What man?’ the r eply is that the proposition explicitly leaves it to you its assertion to what man or men y ou will (cf. 2.357). A sign that is objectively indeterminate in any respect is objectively vague in so far as it reserves further determination to be made in some other conceiv able sign, or at least does not appoint the interpr eter as its deputy in this offi ce. Example: ‘A man whom I could mention seems a little conceited. ’ The suggestion here is that the man in vie w is the person addr essed; but the utterer does not authoriz e such an interpr etation or any other application of what she says. She can still say, if she likes, that she does not mean the person addressed. Every utterance naturally leaves the right of fur ther exposition in the utter er; and therefore, in so far as a sign is indeterminate, it is vague, unless it is expressly or by a well-understood convention rendered general” (5.447). A sign is objectiv ely general, in so far as, leaving its eff ective interpretation indeterminate, it surr enders to the interpr eter the right of completing the determination for himself. ‘Man is mortal’. ‘What man?’, ‘Any man you like’. A sign is objectively vague in so far as, leaving its interpretation more or less indeterminate, it reserves for some other possible sign or experience the function of completing the determination. ‘This month’, says the almanachoracle, ‘a great event is to happen’. ‘What event?’ ‘Oh, we shall see. The almanach does not say that.’” (5.505). 4. Ms 284: “The assertion consists in furnishing of evidence b y the speaker to the listener that the speaker believ es something, that is, fi nds a cer tain idea to be defi nitively compulsory on a certain occasion” (Ms 787; see CP 2.335).

32

who asserts a proposition accepts to be held responsible for it, and subjects himself to possible penalties, in case the proposition turns out to be false.5 The speaker is a defender of his o wn position; as for the listener, it is his interest to try and detect a possible falsehood committed b y the speaker, since “the affirmation of a proposition may determine a judgement to the same effect in the mind of the interpreter to his cost” (Ms 517; NEM IV, 249). Hence the utter er and the interpr eter hav e opposite interests and attitudes with regard to the truth of any proposition asserted by the former (Hilpinen 1982, 185), so diffi cult is it to give up one’s beliefs.6 This is why the interpr eter of a pr oposition is at times called its “ opponent” (e.g. in Ms 515) 7. Indeed, although all communication implies the mutual r espect of a number of tacit assumptions, some agreement on the aim of communication on both parts, no “latitude of interpretation” (CP 5.447), the necessarily asymmetrical situation which prevails between both speakers does not make the elimination of indetermination in all cases desirable. The speaker may have some interest in remaining in a certain fuzziness (CP 5.505n1; cf. CP 3.94). However, if the speaker wants to convince or to communicate an information, it is up to him to qualify (though not eliminate) it, either by using indexical signs indefinite enough so that “the sign is not sufficiently expressing itself to allo w of an indubitable determinate interpr etation” (CP 5.448n1) or b y accepting to extend “to the interpreter the privilege of carrying its determination fur ther” (CP 5.447), thus turning a pr eviously vague assertion into a general one, as when the utterer leaves it to the interpreter to complete the determination of the implications of such an assertion as: “This being filthy, in every sense of that term”, thus counting on the “ collateral” (CP 8.178–9) or backgr ound information linked to the system of conv entions (such as determinate index es (proper names) or indefinite ones (such as common nouns) to facilitate interpretation and to determine who, the utter er or the interpr eter will be in a position of defense or attack (Ms 283; cf . Hintikka’s interpretation of quantifi ers in 5. “To assert a proposition means to accept responsibility for it, so that if it turns out ill, or as Mr. Schiller says (by implication) unsatisfactory, in a certain way which we need not defi ne, but which is called proving to be false, he who asserted it regrets having done so” (Ms 280). 6. “The utterer is essentially a defender of his o wn position and wishes to interpr et it so that it will be defensible. The interpreter, not being so interested, and being unable to interpret it fully without considering to what extreme it may reach, is relatively in a hostile attitude, and looks for the interpretation least defensible” (Ms 9, 3-4). 7. Thus, the language-game occurring between the speaker and the interpreter with respect to an indeterminate proposition is very close to what Hintikka calls a zero-sum game (1979, 51).

33

game-theoretical semantics 1979, 51)). B ut the context of the asser tion (in other words its pragmatic dimension) is the thir d means invoked by Peirce—since “no general description can identify an object”—to explain how an object of experience can be identified, besides its being singularized: “The common sense of the interpreter of the sign will assure him that the object must be one of a limited collection of objects ”.8 Thus the puzzle is solved in a way, closer to “rank pragmatism” than to a straightfor ward “commonsensism”, much in keeping with the famous P eircian maxim: “Consider what eff ects that might conceiv ably hav e practical bearings, we conceive the object of our conception to hav e. Then our conception of these eff ects is the whole of our conception of the object ”. But again, if assertion is tied to such pragmatist constraints, it is because it operates within an epistemic context in which what is at stake, aimed at, in the question-answer process, is not any kind of communication, but the communication of knowledge and truth. 2. What about inquir y? How should it be depicted? I ndeed, as rightly insisted on b y C. H ookway, as a r esponsibilist inv estigation, in which some epistemic traits of vir tuous character such as honesty , economy, confidence, and such aff ective states as emotions, sentiments, feelings of knowing have to be instanced in order, not only to determine whether we are engaged in a genuine process or a mere “paper” doubt or fake enterprise, but to explain ho w we can be vie wed as r eliable and r esponsible agents: 8. “Suppose, for example, two E nglishmen to meet in a continental railway carriage. The total number of subjects of which there is any appreciable probability that one will speak to the other perhaps does not ex ceed a million, and each will hav e perhaps half that million not far below the surface of consciousness, so that each unit is r eady to suggest itself. If one mentions Charles the S econd, the other need not consider what Charles the S econd is meant. I t is no doubt the English Charles the Second. Charles the Second was a quite different man on different days; and it might be said that without fur ther specifi cation the subject is not identifi ed. But the two E nglishmen have no purpose of splitting hairs in their talk; and the latitude of interpretation which constitutes the indeterminacy of a sign must be understood as a latitude which might affect the achievement of a purpose. For two signs whose meanings are for all possible purposes equivalent are absolutely equivalent. This is, to be sur e, rank pragmaticism; for a purpose is an aff ection of action” (CP 5.448n1). “S uppose the chat of our E nglishmen had fallen upon the colour of Charles II’s hair. Now that colours are seen quite differently by different retinas is known. That the chromatic sense is much more varied than it is positively known to be is quite likely. It is very unlikely that either of the trav ellers is trained to obser ve colours or is a master of their nomenclatur e. But if one says that Charles II had dar k auburn hair, the other will understand him quite pr ecisely enough for all their possible purposes, and it will be a determinate predication” (ibid.).

34

however, these are not the only requirements to be met by an inquirer: as far as doubt is concerned, P eirce claims that the only way to justify our reasons for doubting does not so much per tain to feelings, emotions or affectives states (which belong to what he calls in his categorial jargon, “Firstness”) as to reactions, existential shocks (“Secondness”) encountered when we meet some “recalcitrant” experience, which forces us to revise our previous beliefs. Again, the only way to make sure that our beliefs are not mere prejudices or are not fixed in the wrong way—because, for example, we would be using such spurious methods as the a priori (or psy chologistic, subjectivist) method, or the method of “ tenacity” or the method of “authority” (see “The Fixation of Belief ”, or “How to make our ideas clear” 1878–79)—is to rely on the “scientific method”, the justification of which is provided by the fact that there are “real things” which are stable, external enough (though not totally independent of our thinking about them) to constrain and “fi x” our beliefs in the appr opriate way. In other words, inquiry is part and parcel of a strong realistic commitment and of an allegiance not to mere conversational “maxims” or to some set of “questions” and “answers”, but to the strict rules of the scientifi c method itself with all its inferential requirements in terms of deduction, induction and abduction, and its methodological procedures (as prescribed, in particular, by the “economy of research”). So inquiry implies more than deliberation and discussion: it implies a strong commitment to norms and it is liable to penalties. Indeed, for Peirce, some “ethics of inquiry” is needed. But this should not dispense us either with some e“thics of belief”, as can be shown from Peirce’s severe condemnation of James’s position in the Clifford-James debate (Haack 1997; Tiercelin 2005a, 196ff .) and his siding rather with an attenuated Cliffordian (or basically evidentialist) position. 3. What about doubt? One of the challenges any account of kno wledge has to face is the threat of scepticism. How does Hookway’s account bear upon scepticism? Or does it even face the issue? 9 As he notes himself, “it would be nice to defuse scepticism b y arguing that when w e leav e the 9. This is an objection addressed to Hookway by Marie McGinn in her reply to Hookway 2003 (2003, 99–100): “But [a consequence of Hookway’s view] means that I can take my current emotional evaluations as a proper ground for rejecting the sceptic’s questions only by assuming the very thing that the sceptical voice in me doubts—that is, by arguing in a circle”. “Moreover, given that Hookway’s account of our ordinary practice acknowledges that the relation between the emotional evaluations on which it r ests and objective truth is contingent, it is har d to see how the work of resisting skepticism in a philosophical context is to be achiev ed” (99).

35

backgammon table and raise global questions about whether w e know anything at all, our interrogative sentences cease to express genuine questions. I am doubtful that such an argument can be made to work—and if it can it is likely that we should conclude only that our sceptical anxieties should not be expr essed using the concept of kno wledge. Deeper worries about our ability to carr y out inquiries responsibly or to for mulate questions effectively might remain (italics mine). M ore promising is the suggestion that since an interr ogative sentence will be used to ar ticulate diff erent questions (or formulate different problems) in different contexts, there is no direct route (italics mine) from the impropriety of claiming knowledge of some fact in a distinctively philosophical context to the impropriety of doing so in an ev eryday context” (1999, 16). However is this enough to constitute an effi cient “parry” to scepticism under its most serious forms (in particular the Cartesian sceptical scenario), at least if we are not totally convinced by any contextualist suggested solution to it? Here it might be worth comparing two types of strategies that have been provided by Wittgenstein and Peirce, along what sounds undoubtedly like a “ pragmatist” line of thought, close, in many ways, to the one suggested b y Hookway himself. Briefly, all have a lot in common in terms of the diagnosis and the reaction they deem necessary to the Cartesian sceptical scenario (or radical doubt): a doubt must have practical effects otherwise it is vain and vacuous; radical doubt is not so much unpractical or unr easonable as intrinsically incomprehensible: it is contrary to the logic of judgement, which needs to be put in context; a mere logical possibility of doubt does not constitute a real one. Doubt presupposes a system of beliefs and previous certainties: “If I want the door to turn, the hinges must stay put ” (OC, 343). The Cartesian scenario, Wittgenstein claims, is based on a confusion between what is empirical and grammatical. Therefore, a Moorean type of answer, relying on Commonsense is not the right one. There are many problems with the use of the wor d “kno w”. I ndeed, Moore’s “propositions” ar e neither true nor false: they ar e just accepted. This is also why it is useless to ask to justify them. Justification must end some where. So not only is the Cartesian sceptical challenge meaningless but so is Agrippa’s sceptical challenge (the infinite regress sceptical scenario). However, as has been often noted, a large part of Wittgenstein’s answer relies on ambiguities related to the epistemic status of the “hinge” propositions: are they logical, analytical truths, rules, principles, practical norms, norms “in context”, “standing” or “unearned” certainties (Wright 2004)? It is not easy to tell. Quite different is Peirce’s off ensive strategy: fi rst, he has his o wn (and naturalistic rather 36

than “grammatical”) r easons against radical doubt: as w e noted pr eviously, for him, doubt must have an external cause in order to be real. This leads him to insist more indeed on the links with the doubt-belief (habit) structure (or struggle) of inquiry (rather than of knowledge proper). What is important is tofi x belief, the “settlement of opinion”: “A true doubt is a doubt which really interferes with the smooth working of the belief-habit” (CP 5.510). Secondly, whereas Wittgenstein’s reply consists in underlining the groundlessness of beliefs—and stressing their “mobile” epistemic status (hinge-propositions)—Peirce has a two-levels answer: 1. He adopts a form of C ritical Common S ensism (inspired by both R eid and Kant, and mixes “indubitable” beliefs with a some what extreme fallibilism. 2. He follows a strategy consisting, as noted earlier , in sor ting out, among several methods for fi xing belief, the method which is the only one able to fi x belief (i.e to calm the uneasiness cr eated by doubt): the Scientifi c Method based on the hypothesis of Reality (both fixed by experience and by the community of investigators).This explains why, in the end, we have two rather different “parries” to the sceptical challenge. Both pragmatists, generally speaking, r eject the Car tesian sceptical scenario: a doubt (as Hookway also emphasizes) must be contextualized. All seem to accept the view that “at the foundation of w ell-founded belief lies belief that is not founded” (OC, 253). In other words, some beliefs do not per tain to the vocabulary of foundation, justifi cation, knowledge. Hinge propositions are not strictly speaking known, since they ar e outside epistemic ev aluation. From which it seems easy to conclude that one can admit the infi nite regress argument without scepticism follo wing therefrom. Indeed, for a doubt to be possible, does not mean for it to be necessary. Even when we could doubt, we simply do not (cf. OC, 509): “The difficulty is to realize the groundlessness of our believing ” (OC, 166). This is why their r eply to the sceptic is at times paradoxical, since it induces a form of scepticism towards justifi cation, very much in the spirit of neo-pyrrhonism which refuses to accept the philosophical perspectiv e adopted b y both dogmatic sceptics and their opponents (Fogelin 1994, 99). In the same way, Peirce’s extreme fallibilism at times only separates him fr om scepticism by a hair’s breadth (as often noted by Hookway himself). However, while specific difficulties are posed by the epistemic status ofWittgenstein’s hinge propositions (especially if meant as r ules defi ning a practice), in terms of any empirical inquiry which is traditionally conceived as aiming at truth, Peirce’s “parry” seems mor e “efficient”, in so far as, for him, any belief may be criticized, at least if there are positive reasons for that. A Common 37

Sensical answer? Hardly, since Commonsensism remains irreducibly critical and normative. Finally, one may wonder whether Hookway’s approach, focusing, as it does, on the description of our evaluative practices, is suffi cient, not only to defuse scepticism, but to get the right type of epistemic normativity we are looking for.10 5. Conclusions and suggestions from a standard and pragmatist approach Hookway is surely right to stress the importance of viewing knowledge as a question-answer process of inquiry. However, he may not have provided strong enough arguments in fav or of a total r evision of our concept of knowledge or, by the same token, of the primary task of epistemology itself. It may well be that the “standard” account has to be modifi ed, in view of the various aporia it faces. But one might also suggest a different “pragmatist” strategy, which would start from the traditional (though modified and improved)11 definition of kno wledge, rather than fr om a vir tue theoretic approach, question-answer, or knowledge as inquiry approach and would add two impor tant “pragmatist” principles: 1) The refusal of the principle 10. See M. McGinn’s criticism in 2003. 11. In Tiercelin 2005a, 267 ff., I have given a detailed analysis of what might be such “constraints” on knowledge (rather than necessary and sufficient conditions) and suggested a definition along the follo wing lines. Briefly: (K) S knows that P iff : a) S believes that P (belief being less an internal mental state than a disposition to action)(as noted by A. Bain, Peirce or Ramsey); b) P is true (truth being itself conceived either on the mode of warranted assertibility (Dewey), at the ideal limit of inquiry (Peirce), or in a merely redundantist fashion (Ramsey): truth is success; it is in itself metaphysically neutral; what r eally matters is what goes together with it: our assertions, our investigations); c) S is (most often: because of the ever present risk of fallibilism) justified in believing that P, namely, essentially, that: 1. S would believ e that P iff P were true (obeying E. Sosa’s principle of safety; along Reidian, Peircian, neo-Moorean principles of commonsense). 2. If P were not true in relevant circumstances, S would not believe P (principle of “sensitivity”); in case our until now best established beliefs encountered the shock of recalcitrant experience, then we should be ready to “overthrow our whole cartload of beliefs“ (Peirce’s critical commensism). 3. If, in other relevant circumstances, P were still true, S would still believe that P (“counterfactual” conditions formulated by Nozick—translatable into the lessons to be drawn from the pragmatist maxim, along a subjunctive conditional reading: in order to get the meaning of the hardness of the diamond, one can always translate it into a set of conditionals: “If the diamond was pressed upon, it would not break” (reality of the “would-be” or disposition)). 4. S is justifi ed in believing that P b y a proper reliable causal process (Goldman): hence, irrelevant counterfactual situations are excluded, the reliability of the process being itself a matter of trust on our part in our cognitive (intellective and active) faculties (Reid, Peirce, Sosa, Greco).

38

of radical generality, which goes hand in hand with the sceptical argument: as a consequence, an analysis is required of the conditions of possibility of doubt itself, and not only of knowledge: “Doubt itself needs to be justified” (Reid, Peirce and Wittgenstein); we should always find a way to distinguish between paper (or chamber) doubt and genuine r easons for doubt; r ely either on certain rules or norms (hinge propositions in Wittgenstein’s style) or on principles (closer to Common S ense), but constantly submit such principles—which, incidentally, are not fi xed (Reid) but ev olutive: “ We outgrow the applicability of instinct ” (Peirce)—to our criticism and control; and show that the sceptical objections ar e most often strong because they refuse any kind ofcontextualization. However, invoking contextualism does not amount to holding that the epistemic status of a proposition may vary according to purely conversational, cultural, social (or other) factors: it is to hold that, independently of such infl uences, a proposition has no epistemic status whatsoever. Thus formulated, contextualism implies a form of externalism, for even if proper contextual constraints have to be satisfied in order for such and such a proposition to be able to count as knowledge, such constraints need not be actually claimed, kno wn, nor ev en believed, even if some minimal sensitivity to such constraints (because of their causal impact) is unav oidable (as str essed by Williamson 2000). I n that respect, a pragmatist r efuses the irr elevant choice—due to an (accor ding to him) err oneous split betw een the internal and the external—between internalism and externalim. I n particular, not accepting the principle of priviledged access to our mental states (access-internalism), through some form of conscious awareness, intuition, introspection, does not imply that we cannot (in fact we must) exercise some self-control and criticism on our beliefs. The question r emains open as to the natur e of such a contr ol: is it something irreducibly normative and reflexive, or is such a normativity already present in terms of some metacognitiv e capacities at the lev el of nature itself? In which case, we would not have to distinguish a “reflexive” level from an “animal” level (Sosa) in order to have a “perspective” on our beliefs, but we should merely view epistemic normativity as being in continuity with (emerging from) nature (Peirce’s solution). But then we would have to view our belief-dispositions not only asstable habits but as involving the necessar y “habit-changes”—Peirce; cf. Aristotle’s distinction betw een “hexis” (stable habit) and “diathesis” (moving disposition)—entailed by the education of our “feelings of knowing” or “sentiment of rationality” (James, Peirce) in order for our dispositions to become virtues, hence to play their expected role in our practices of epistemic evaluations and ethical valuings 39

(Dewey’s distinction, which we have to make if we want to keep the insight that there is more value in knowledge than in justified true beliefs). In the same manner, not to accept the principle of an independent reality, totally external to our beliefs (see the criticism of metaphysicalealism r present both in Peirce and in P utnam), does not imply that it is not possible (indeed, it is necessary, if we want to explain the cooperation of the (tr uth-maker) world with our beliefs), either to maintain the causal strength of r eality upon our beliefs (Peirce’s Secondness and semiotic abductive realism applied to the analysis of per ception and opposed to a R eidian direct realistic or a Putnamian natural r ealistic approach); or to dev elop a form of r ealism enabling one to explain how, on the one hand, the real does constrain our beliefs, and on the other hand, is to be vie wed as the final opinion agreed upon by the scientific community, hence justifies our scientific method of fixing our beliefs in a warranted way (at the roots of Peirce’s scholastic and even Scotistic realism). 2) The principle of fallibilism is the second pragmatists principle that should be added to such constraints: it is closely allied with the presumption of the possibility of knowledge but also with a necessarily un-dogmatic definition of knowledge and an epistemic (and ev en ontological) radical indeterminism. But it is a fallibilism which may itself happen to be questioned and is constantly submitted to the r ules of scientifi c method and to the constraints of inquiry. But note that it is precisely such a fallibilism (which is consubstantial to pragmatism) which constitutes the ev er present risk of the sceptical drift (several times, indeed, the pragmatists come very close to scepticism, in a neo-P yrrhonian or dogmatic way). Contrar y to Hookway’s suggestion (2008b) that “fallibilism” is not really “disturbing” for Peirce, since it has the status of a mer e “abstract” possibility, such an alliance of “ fallibilism and anti-scepticism ” might w ell constitute “ the” insight of American pragmatism (P utnam), but also a genuine and constant threat to knowledge. This is why it seems advisable to bet on some form of weak foundationalism which might be looked for, along such lines as those pr oposed by Tyler Burge’s concept of “ perceptual entitlement”: our perceptions may not pr ovide us warrants nor justifi cations, however they entitle us or giv e us prima facie justifi cations to believe what we do believe, as Peirce tries to make clear in his analysis of perception through abduction12. 12. See Tiercelin 2005b, for a detailed account of Peirce’s views on abduction and perception.

40

Although remaining within a basically standard model, such a pragmatist strategy about kno wledge would impose the r evision of some of our former views. Involving a naturalistic though not “naturalized” approach, it would tend to bring epistemology 1. closer to ethics: stress the need to interpret normative facts (without totally indulging into virtue epistemology); account for normativity, in terms of an emergence of norms (viewed as both stable and ev olutionary dispositions (habits and habit-changes) from nature, akin to Kant ’s model of a system of pr eformation of pur e reason); 2. back to the philosophy of mind and cognitiv e psychology: account for such mental states as both belief-dispositions and aff ective states as being constitutiv e of mental and epistemic agency; 3. closer to the philosophy of science: insist on experience and experimentation within the context of an inquir y ruled by the scientifi c method; 4. closer to the philosophy of language: the inquir y part of kno wledge stresses how our assertions aim at kno wledge (which works as their norm) within a pragmatist though limited contextualism; 5. closer to metaphysics, thr ough the need for a better understanding of the nature of dispositions (are they mental, physical, basic, super venient on categorical pr operties, mer ely functional), and of the r eal, if per ceptual entitlement may be vie wed as a prima facie justifi cation, constraining our beliefs, “ mimicking foundationalism”, how does perception hook into the world? I n what way does it inform us about the ontological furnitur e of the states of aff airs? In so doing, the new strategy would provide a decidedly anti-pyrrhonian model of agency and a guide to action, in accordance with the strong rejection of ethical neutrality of abstention condemned by James. This morning, when we woke up, we stopped dreaming (at least, we are entitled to believe so). The sun, like y esterday, had arisen, the ear th was under our feet. This is not infallibly certain, but it remains highly probable, until proven wrong. At any rate, it would bemeaningless to ask us to justify it or to pretend that we are in no way entitled to think it. Because we aim at truth, because we value knowledge more than mere justified and true belief, because we view ourselves as answerable and responsible (epistemic and mental) agents who are constrained by the real and commit themselves to their assertions, we presume (not only as a regulative but as a living hope) that knowledge is possible, and this is indeed suffi cient to dispose us to act.

41

REFERENCES Brock, J. 1975. “Peirce’s Conception of Semiotic”, Semiotica, 14: 2, 124–141. — 1979. “Principal Themes in P eirce’s Logic of Vagueness”, in Peirce Studies, Lubbock, Texas, 41–50. — 1980. “Peirce’s anticipation of Game-theoretical Logic and Semantics”, in M. Herzfeld & M. Lehnart (eds.), Semiotics, Plenum Press, New York and London, 55–64. — 1981. “An Introduction to Peirce’s Theory of speech-acts”,Transactions of the C. S. Peirce Society, vol. 17, 319–326. Dewey, J. 1938. Logic: The Theor y of Inquiry. Henry Holt, New York. Fogelin, R. (1994) Pyrrhonian Reflections on Knowledge and Justification. Oxford, Oxford University Press. Haack, S. 1997. “ ‘The Ethics of Belief ’ Reconsidered”, in L. E. Hahn (ed.), The Philosophy of R. Chisholm, La Salle, Ill., Open Court, 120–44. Hilpinen, R. 1982. “On Peirce’s Theory of the Proposition: Peirce as a Precursor of Game-Theor etical Semantics”, The Monist 65, 182–8. — 1995. “Peirce on Language and Reference”, in Peirce and Contemporary Thought , K. L. Ketner(ed.), Fordham University Press, New York, 272–303. Hintikka, J. 1979. “Quantifiers in Logic and Q uantifiers in Natural Language”, in J. S aarinen (ed.), Game Theor etical Semantics, Reidel, Dordrecht, 27–47. “Quantifiers versus Quantification-theory”, ibid., 49–79. Hookway, C. 1985. Peirce. London, Routledge and Kegan Paul. — 1990. Scepticism. London, Routledge. — 1993. “Mimicking Foundationalism: on Sentiment and Self-Control”, European Journal of Philosophy 1, 156–74. — 1994a. “Cognitive Virtues and E pistemic Evaluations”, International Journal of Philosophical Studies, 2, 211–27. — 1994b. “Naturalized Epistemology and Epistemic Evaluation”, Inquiry, Vol. 37, n°4, 465–85. — 1998. “Doubt, Aff ective States and the R egulation of I nquiry”, in C. M isak (ed.), Pragmatisms, Canadian Journal of Philosophy, Supplementary volume 24, 203–26, reprinted as chapter ten of Hookway 2000, 246–264. — 1999a. “How to be a vir tue epistemologist”, reprinted in M. D ePaul & L. Zagzebski (eds.) Intellectual Virtue. Perspectives from Ethics and E pistemology, Oxford, Clarendon Press, 2003, 111–134. — 1999b. “Questions of Context”, Proceedings of the Aristotelian Society, 1–16. — 1999c. “Epistemic Norms and Theor etical Deliberation”, in Ratio vol. XII, 380–97. Reprinted in Normativity, J. Dancy (ed.) Oxford, Blackwell, 2000.

42

— 2000. Truth, Rationality and Pragmatism, Themes from Peirce. Oxford, Clarendon Press. — 2001. “Epistemic Akrasia and E pistemic Virtue”, in A. F airweather and L. Zagzebski (eds.) Virtue Epistemology, Essays on Epistemic Virtue and Responsibility, Oxford University Press, 178–199. — 2002. “Emotions and Epistemic Evaluations”, in P. Carruthers, S. Stich and M. Siegel, (eds.) The Cognitive Basis of Science, Cambridge, Cambridge University Press, 251–262. — 2003. “Affective States and E pistemic Immediacy”, Metaphilosophy, v ol.34, n°1–2, 78–96. — 2008a, to appear. “Epistemic Immediacy, Doubt and Anxiety: On the Role of Affective States in Epistemic Evaluation” in G. B run, U. Doguouglu and D. Kuenzle (eds.), Epistemology and the Emotions, Ashgate. — 2008b, to appear. “Peirce and Scepticism”, in J. Greco (ed.) The Contemporary Anthology of Scepticism, Blackwell, Oxford. James, W. 1979. The Will to B elieve, in The Works of William James (vol. 6), F. H. Burckhardt, E. Bo wers and I. K. Skrupskelis (eds.), Cambridge, H arvard University Press (17 vols.), 1975–88. McGinn, M. 2003. Reply to C. Hookway’s “Affective States and Epistemic Immediacy”, Metaphilosophy, vol.34, n°1–2, 96–101. Peirce, C. S. 1931–58. The Collected Papers of C. S. Peirce (CP). Ch. Hartshorne, P. Weiss & A. Burks (eds.), Cambridge, Mass., Harvard University Press, cited by volume and paragraph numbers. — 1967. The Annotated Catalogue of the Papers of C.S. Peirce (Ms). R. Robin (ed.), Amherst, Mass., University of Massachusetts Press. — 1976. The New Elements of Mathematics (NEM). C. Eisele (ed.), Mouton, The Hague (4 vols.). — 1982. Writings of C.S. Peirce, a chronological edition (W). M. Fisch, Ch. Kloesel, E. C.Moore, D. D. Roberts (eds.), I ndiana University Press, Bloomington (6 vols.). Tiercelin, C. 1993a. La pensée-signe: études sur Peirce. Éditions Jacqueline Chambon, Nîmes. — 1993b. Peirce et le pragmatisme. Presses Universitaires de France, Paris. — 1997. “Peirce, on norms, ev olution and knowledge”, Transactions of the C. S. Peirce Society, vol.33, n°1, 35–58. — 2005a. Le doute en question: parades pragmatistes au défi sceptique. Paris, éditions de l’Eclat. — 2005b. “Abduction and the Semiotics of Perception”, Semiotica, F. Merrell and J. Queiroz (eds.), 389–412. Williamson, T. (2000) Knowledge and Its Limits. Oxford, Oxford University Press.

43

Wittgenstein, L. 1969. On Certainty (OC). G. E. M. Anscombe and G. H. von Wright (eds.), Oxford, Blackwell. Wright, C. 2004. “Wittgensteinian Certainties”, in D. McManus (ed.), Wittgenstein and Skepticism, London, Routledge, 22–55.

44

Grazer Philosophische Studien 77 (2008), 45–59.

IN WHAT SENSE IS KNOWLEDGE THE NORM OF ASSERTION? Pascal ENGEL Université de Genève Summary The knowledge account of assertion (KAA) is the view that assertion is governed by the norm that the speaker should know what s/he asserts. It is not the purpose of this ar ticle to examine all the criticisms nor to tr y to giv e a full defence of KAA, but only to defend it against the charge of being normatively incorrect. It has been objected that assertion is governed by other norms than knowledge, or by no norm at all. It seems to me, however, that a number of these criticisms are based on a number of misunderstandings of the notion of a norm and of the way it can regulated a given practice. Once we spell out in what sense knowledge can play a normative role in this context, the KAA appears much more plausible.

1. Preliminary formulations of the knowledge account The knowledge account of assertion (KAA), although it has probably been present within philosophy at least since the P yrrhonian sceptics 1, was introduced into contemporary analytic philosophy by G.E. Moore (1962), revived by Michael Slote (1979) and P eter Unger (1975) and Timothy Williamson (2000) 2. It has since r eceived as much appr oval as criticism. It consists in at least the following three claims: 1. Although it has, to my kno wledge, not been often r emarked, the P yrrhonian sceptic’s refusal to assent fully to pr opositions is based on the implicit r ecognition that kno wldedge is the norm of assertion. See Frede’s (1987) classical article. 2. The knowledge account sur faces also in some writings b y Ruth Marcus, such as her (1983), where she holds that to believ e that P implies that P is possible, and one ’s realisation that P is impossible leads to a withdrawall of one ’s claim to believe, and in appropriate conditions of one ’s having asser ted that P. In other wor ds, realisation that one does not kno w that P, since P cannot but be false when P is impossible, amounts to the r ealisation that one is not entitled to assert that P.

(i) There is a norm for the speech act of asser tion (ii) This norm is unique and constitutive of assertion (iii) This norm is that one must assert that P only if one knows that P Each of these claims can be contested. O ne can reject the idea that there is a norm for the practice of assertion: maybe there are no rules or norms for this practice, and just things that people say in various circumstances, which can all equally count as assertions. One can also reject the idea that if there is a norm it is unique and constitutiv e. There might be several kinds of norms, including ones which do not involve any epistemic position on the part of the asserter. And one can reject the knowledge norm itself, and propose a weaker one, while accepting that the norm of assertion is some epistemic status. The other ob vious candidates ar e the w eaker norms of truth and belief: (KT) One must assert that P only if P is true (KB) One must assert that P only if one believes that P or perhaps some higher condition than mere belief, alternatively: (KRB) One must assert that P only if one rationally believes that P (KWB) One must assert that P only if P is warranted (KPB) One must assert that P only if P is highly probable In this ar ticle, I will not r eally discuss the issue of whether one should defend (KT ) or one of the norms (KB), which all pr esuppose that the norm of asser tion inv olves some epistemic status, lo w or high, on the part of the speaker. I shall be mainly concerned with the sense in which either of these can be said to be a norm of assertion. For that, I shall also suppose that the norm for asser tion is at least as str ong as (KN) belo w. Even if we assume that (i)–(iii) are correct, there are, nevertheless, several formulations of (iii), which ar e not equiv alent. The most basic one says that an asser tion that P is corr ect if and only the speaker kno ws that P, hence obeys the prescription: (KN) One must: assert that P only if one knows that P This is Williamson’s (2000, 243) formulation. M oore’s o wn v ersion is 46

distinct: “By asserting that P, you imply, though y ou do not asser t, that you know that P” (Moore 1962, 277). (MKN) To assert that P is to imply that one knows that P Peter Unger’s formulation diff ers too: “ Asserting that something is so entails not just representing the thing as being so, but er presenting oneself as knowing that it is.” (Unger 1975, 256) (SPKN) To assert that P is to represent oneself as knowing that P Let us call this, with P agin (2006), the “ self-presentation account”. The self- presentation account is clearly the counterpart of similar claims which have been proposed for the belief account of assertion (KB). Thus Dummett has suggested: A man makes an assertion if he says something in such a manner as deliberately to convey the impression of saying it with the o verriding intention of saying something true (Dummett 1973, 300)

and Davidson similarly: What is understood [b y the hear er of an asser tion by a speaker] is that the speaker, if he has asserted something, has represented himself as believing it —as uttering a sentence he believes true, then. But it is not a convention, it is merely part of the analysis of what asser tion is. To assert is, among other things, to represent oneself as believing what one asserts”. (Davidson 1979)

The self-presentation analysis transposes this claim from belief to knowledge. As one sees, these formulations are not equivalent and their differences can generate a lot of ambiguities. In the first place it is not clear whether they are supposed to give a defining feature of assertion, or at least to give a necessary condition of it, as (KN) does, or whether they ar e giving the condition of a successful assertion. On the first reading, a statement which would not conform to the (KN) norm would fail to be an asser tion. On the second r eading, it might be an asser tion but a defectiv e one. I n the second place (KN) and the other two formulations (MKN) and (SPKN) differ in that the latter two do not require that the speaker actually knows that P, but only that he conv eys to his audience the belief that he does, whereas the former does r equire that the speaker actually kno ws that P, 47

and is compatible with the following: (KNB) One must: asser t that P only if one believ es that one kno ws that P There is at least one version of the self-presentation account (SPKN) which is clearly defectiv e, for r easons which hav e been clearly br ought up b y Peter Pagin (2006). Suppose that a speaker represents herself as knowing that P by simply saying (1) I hereby represent myself as knowing that P But an utterance of (1) is obviously not an assertion that P, for conveying to an audience, successfully or not, the belief that one kno ws that P cannot amount to asserting that P. When one asserts that P, what one says implies that P, and is incompatible with its falsity. (1), however, is compatible with the falsity of P. This is not simply because one may lie: one may commit oneself, through some sor t of act—for instance a statement befor e a tribunal, which legally ties one to the tr uth of P, without it being the case that P is actually true. Asserting that P is more than committing oneself to some audience, through some sort of social commitment. In other words, it is something different from assuring someone that P. It is saying that P with the aim of saying something true, and not simply saying something which one represents to others as tr ue, and that one believ es as true. An actor on the stage might, in this sense, represent himself as knowing that P without asser ting that P. For the same r eason, the self-pr esentation account does not wor k when the speaker is supposed to commit herself to the belief that P: (2) I hereby represent myself as believing that P is just as inadequate as (1) for the same r easons. The same goes for: (3) I hereby represent myself as committed to the truth that P which is no more an assertion that P than (1). 3 3. Pagin 2006. P agin dev elops in his 2004 a general strategy for assessing claims about speech acts.

48

This may hav e to do with the r eason why M oore says that to asser t that P is not to assert that one knows that P, but only to imply it. But it remains to be seen in what sense there is such an implication in assertion. The inadequacy of the formulations (1), (2) and (3) does not rule out the self-presentation account in itself, since there may be more adequate formulations of it (I shall come back to that below). I shall however, assume first that the correct account is (KN). 2. Familiar objections to KAA Several kinds of arguments hav e been adv anced in fav our of (KN). The most familiar ones appeal to linguistic evidence. Asser ting that P invites the question “How do you know?” when the assertion is improper, or “You don’t know that”, when it is false. S imilarly asking the question whether P seems to imply that one ignores whether P. The point is that when a speaker asserts that P, and does not kno w that P, he can be criticised, or has to withdraw his asser tion. Similarly if y ou ask whether P while y ou already know that P, you have in some sense failed to ask a question, or have violated G rice’s maxim of quality . This linguistic evidence is not very compelling, though. In a number of contexts, utterances that P can elicit questions like: “What are your reasons for believing P?” or “Do you believe that?” which suggest that reasonable grounds for belief, or simply truth, are the pr oper presuppositions of asser tion. And some questions are just ir onical. For instance the follo wing dialogues make per fectly good sense: (4) “Lake Leman is polluted. ”—What ar e y our r easons to believ e this? (5) “Hillary Clinton will be the next president”—Do you believe this? (6) “Have you not received the Nobel prize yet?”—You are kidding! An argument based on Moore’s paradox has also been advanced in favour of KN. It consists in remarking that sentences like: (7) It is raining, but I do not know it are just as absur d as the familiar M oorean sentences such as “I t is raining but I don’t believe it” or “ I believ e it is raining but it ’s not”. Indeed 49

sentences like (7) are much more clearly absurd than these. But one may remark that: (7c) It is raining, but I do not have very good reasons for believing it. does not make sense either. So why should the presence of “know” in (7) make the difference? The lottery argument is perhaps the most discussed (Williamson 2000, 246–47). The point is that an assertion made only on the basis of a v ery high degree of belief can be straightforwardly rejected. In a fair lottery of 10 000 tickets where A has bought one ticket, if B says: “ Your ticket did not win”, A is entitled to asser t: “But you do not kno w that!” H ere too there is a defence in terms of rational belief . Someone could as well answer: “But you do not have enough grounds to believe this!” Another related argument appeals to the role of knowledge in practical reasoning. Hawthorne (2004) pr esents the follo wing piece of r easoning coming from an individual participating in a fair lottery: (1) (2) (3) (4)

My ticket will not win If I keep my ticket, I shall win nothing If I sell my ticket, I’ll get 10 pence Hence I must sell my ticket.

The reasoning is wrong, because the individual is not entitled to (1) in the first place, for his premise is not known. The point is that knowledge plays an indispensable r ole in practical r easoning. But here too it is not clear that rational belief could not be the appr opriate norm: we seem entitled to affi rm things of which w e are not absolutely cer tain in the sense of knowledge (Douven 2006). Other objections to (KN) concentrate on the many cir cumstances where assertion seems to be proper even when one does not know that P (Weiner 2005). They include cases of lying (a liar does not ev en believe what he says, hence does not know it) , ironical assertions, gossips where one merely “says” that P in the coffee house without asserting it seriously, cases of assertions about future events (“The next American president will be Barack Obama”) which we feel perfectly entitled to assert even though we do not kno w that they ar e true, cases of agents who ar e in a G ettier situation, where they ar e justifi ed to believ e something tr ue merely b y accident (“Nogot owns a Ford”, “This is a barn façade ”), or cases wher e 50

the cir cumstances allow us to asser t something for r easons which hav e nothing to do with any epistemic status. For instance, I may assert “That ’s your train!” in a cir cumstance where I am not sur e that it is y our train, although it is v ery important for you to take a train (if it w ere not your train, you would not be entitled to criticiz e me for not kno wing) (Williamson 2000, 256). Or I may say to you, when you are worried by a very serious problem: “That’s not a big deal ”, when I actually kno w that it’s a big deal. Another similar kind of objection inv olves situations where assertions are pr oper in the pr esence of other norms than pur ely epistemic ones. Jennifer Lackey (2007) has presented cases of what she calls “selfless assertions”, where a subject, for purely non epistemic reasons does not believe that P, but is awar e that P is w ell supported by epistemic reasons, hence is led to asser t that P, although he still does not believ e that P, hence does not kno w that P. She gives the case of a racist jur or who, because of his prejudices, is led to believ e that a black man is guilty although he recognises the racist origin of his pr ejudice and asser ts that the man is innocent, or the case of a cr eationist teacher who nev ertheless is able to recognise the scientifi c evidence against her religious beliefs and is led to assert evolutionist claims. According to Lackey, in those cases, the agent, qua asserter is not subject to criticism—she is actually subject to praise for having surmounted her previous beliefs—and offers an assertion which is both true and epistemically flawless, although she can be criticised for her beliefs, and because of these, she does not hav e knowledge. The upshot of these examples is that it is a mistake to require proper assertion to pass through the doxastic states of the asserter. Notice also that they are meant to be counterexamples both to KN and to KB (or one of its version such as KRB), hence to any account of assertion according to which the asserter has to occupy a certain epistemic status. It is not clear, however, that such cases are genuine counterexamples to KAA and to KN in particular. In the first place it is not clear that although the subjects have religious or racist beliefs which run counter their assertions, they do not make these asser tions on the basis of their evidence and of their kno wledge (of ev olutionary theory, of the innocence of the black man). I n the second place, it is not clear that these cases ar e not cases where someone accepts that P although he or she does not believ e that P. O ne can accept, for pragmatic or for non epistemic r easons of some kind (pr ofession, role), and hence asser t that P, without believing that P. At this point both the defender of KAA and its critic can hav e an 51

answer4. On the one hand, the defender of KAA will want to say that to assert that P on the basis of an act of acceptance is mer ely to simulate assertion, or giving a fake asser tion (not necessarily lying). O n the other hand, the critic of the KAA will insist that these cases count as genuine assertions, although the subject is not in the appropriate epistemic status required by KAA or KB (or one of its v ersions). We meet again here the ambiguity already noticed abo ve, betw een the claim that KAA states a norm for the correctness of assertions and the claim that it defines what an assertion is. The critic of KAA can claim that the case of selfless assertions, although they do not conform to the norm of corr ectness of asser tion, count as assertions nevertheless. At this point is becomes important to turn to the issue which is implicit from the beginning: in what sense is assertion subject to a norm? 3. Which norm do we follow in assertion? The objections above show that there is a strong disagreement on this point. Two problems have to be distinguished. The first one is whether knowledge (or for that matter some sort of high epistemic status) is the only norm of assertion. In other terms, are our reasons for asserting a given sentence only epistemic and are the reasons for our being criticized in our assertions? One of the main r easons adduced in fav our of the idea that kno wledge is the norm of asser tion by its defenders is that asser tions which ar e not made in presence of pr oper knowledge can be “ criticised”. But are our r easons to criticize an assertion only epistemic reasons? The second problem is the problem of the relationship between the norm and its application. Timothy Williamson addresses the first problem by replying to the kind of counterexamples mentioned in the previous section involving cases of assertion made in the absence of kno wledge or belief in the asserted sentence, such as lying, speaking non seriously or making “selfless”assertions: Such cases do not show that the knowledge rule is not the rule of assertion. They merely show that it can be o verridden by other norms not specifi c to assertion. The other norms do not giv e me warrant to asser t p, for to hav e such warrant is to satisfy the rule of assertion. (Williamson 2000, 256) 4. For the distinction between believing and accepting, see Cohen 1992, Engel 2000. Jennifer Lackey considers the case of pr ofessional assertions, of lies and other cases, but considers that they are clear violations of the norm knowledge of assertion.

52

The problem is whether an asser tion can r emain an asser tion when it is uttered in the absence, or in the r elative unimportance, of the epistemic status of the sentence utter ed. Indeed the person who says “That ’s your train” intends to say this because there is an obvious cost in not asserting it, and because it is better in such cases to err on the side of pr udence. There is no doubt that we make assertions for reasons which can be practical. In many of these cases we make, as I noticed above, acts of pragmatic acceptance without belief or without knowledge. It is clear that there can be assertions made for such pragmatic reasons, just as there can be (pragmatic) reasons for wanting to believe distinct from (epistemic) reasons for believing. There is a clear sense in which a belief which is held for reasons which fall short of being epistemic—for instance a self deceptive belief or one which we aim to have to secure a form of comfor t—still counts as a belief, so why could not assertions which are made for reasons which fall short of being epistemic, or which happen to be epistemically w eak, fail to count as assertions? Commenting upon the example of “That’s your train” and of assertions out of Gettierised beliefs, Jennifer Lackey says: There is a clear sense in which speakers who assert reasonably believed falsehoods and Gettierized beliefs are not subject to criticism. F or the faultiness of the asser tions in such cases r esult fr om infelicitous cir cumstances, not from any sort of blameworthy behaviour on the part of the asserters.( Lackey 2007, 597)

But the issue is precisely this: what kind of blameworthiness is at stake here? In the case of selfless assertions, the speakers are not blameworthy because they make assertions which are, although epistemically defective, perfectly proper. In the same manner, the liar, the joker, the ironist or the man who wants to r eassure y ou b y saying “That’s no big deal ” can be praised or blamed for their assertions for reasons which are distinct from their being epistemically valuable or not valuable. There is no doubt that they make assertions. This is not the point at issue, and we can here remove the first ambiguity noticed in §1 above: the KAA does not say that it is a defining feature of asser tion (a necessar y and suffi cient condition) that it is go verned by the norm of knowledge. It is perfectly compatible with KAA that some assertions violate the norm, and that other norms or dimensions of evaluation of assertions come into play. But from the fact that such norms can come into play in a number of asser tions it does not follo w that the other norm, the KN (or for that matter the KB or KRB norms) ceases to operate. Compare again with the contrast between reasons to believe and 53

reasons for wanting to believe: the two can coexist or can confl ict within a subject (for instance in self-deception as it is usually characterised), but the fact that it can be in some cases rational in a pragmatic sense to hold a belief for which one has only v ery weak epistemic reasons or evidence does not show that the main standard of evaluation of a belief is not truth or evidence. Similarly the fact that one can praise or blame an asser tion in some cases for pragmatic reasons does not mean that the dimension of evaluation remains its epistemic status. One might object that this answer relies too much on a parallel between belief and assertion, and that it is not a norm of asser tion that it aims at truth or at knowledge contrary to the parallel claim about belief.5 Indeed it seems that in the case of asser tion, we are entitled to adopt whichev er norm w e intend to adopt, depending on our specifi c aims, hence that the norms her e are hypothetical imperativ es. In the case of belief , they may be categorical. But is it the case? Certainly the liar, the ironist or the friendly asserter do make asser tions, and they make them in the absence of knowledge (and of belief ) of what they asser t. But does it means that the point of their assertions was not truth or knowledge? If the objective of asserting that P were not to aim at one’s audience recognising the truth of P (or believing or knowing that P—the kind of epistemic norm is not her e in question—) assertions would be pointless, ev en for the liar. Certainly the ties betw een belief and tr uth or kno wledge are stronger than those between assertion and truth or knowledge, but it does not imply that they stop to operate in the case of assertion. The KN norm on assertion seems sometimes to be the view that our assertions should be so serious that only a Victorian clergyman—or an overly scrupulous scientist—could actually make assertions. But even the joker and the ir onist have to r ecognise, if only implicitly, the norm of truth and knowledge on assertion. 4. What it takes to obey the norm of assertion But it is pr ecisely at this point—and it is our second pr oblem with the very idea of a norm of asser tion—that we may ask: is not the norm KN too strong? Who will ever be able to obey it? Isn’t it actually so strong that 5. See e.g. Boghossian 2003 p. 39 : “The linguistic version of the normativity thesis, in contrast with its mentalistic version, has no plausibility whatsoever; and the reason for this is that it is not a norm of asser tion that we should aim at the tr uth, in the way in which it is a norm of belief to do so.” I have defended the normativity claim for belief in Engel 2005.

54

no one will ever be able to apply it? This was in part what was involved in the lottery case: there are actually many propositions (perhaps most of our beliefs) upon which we can have only a very high degree of belief, and no certainty or knowledge: if it were required that we know them, we would be entitled to assert nothing, or most of our assertions would fail6. Hence the KN norm would be trivial. After all, ought is supposed to imply can. Bernard Williams says: It has been claimed that the norm attached to asser tion is kno wledge, in the sense that in asser ting that p one r epresents oneself as kno wing that p. The force of the claim might be thought to lie in the idea that one can be criticized for asserting what one does not kno w (“you should not hav e said this if you did not know”). This represents the criticism as a criticism of an assertor, but the speaker , after all, may not be in position himself to apply the norm effectively, because, at the point of asserting that p, he may reasonably think that he kno ws that p when he does not. I n such a case, it is not reasonable to criticize him, in the sense, at least ofblaming him for not having remained silent rather than speaking, or for not having said that p only in some qualified or doubtful way. It may be said, as it can in the case of many other rules, that he has broken the norm, but that in these circumstances he is not to blame for doing so . But then w e still need to kno w the supposed norm applies, and what the consequences of its obtaining ar e supposed to be. (Williams 2002, 76–77)

Here some defenders of the KAA account often use a distinction put forward by Keith DeRose between the primary and the secondary propriety or impropriety of a norm: As it happens with other rules, a kind of secondary propriety / impropriety will arise with respect to KN. While those who assert appropriately (with respect to this rule) in a primary sense will be those who actually obey it, a speaker who broke this rule in a blameless fashion (one who asserted something she didn’t know, but reasonably thought she did know, would in some secondary sense be asserting properly. (De Rose 2002, 180)

In other words, one could violate a norm of asser tion in the primar y 6. It is inter esting her e to note that this kind of objection has also been raised against normative accounts of belief in terms of tr uth: it is said that the norm for belief cannot be truth, since many truths cannot be believed and many truths are too trivial to be believable (see Boghossian op. cit., Engel 2005). But since the norm of belief for asser tion is arguably weaker than the knowledge norm KN, the charge of being too strong would apply even for the weaker norm (KB)

55

sense, while not violating it in the secondar y sense, hence not being open to criticism in the primar y sense; alternativ ely one can violate the norm in the secondary sense and not be criticisable in the primary sense. Some writers, like Matt Weiner (2005), use this distinction to defend the norm of truth (KT) for assertion against the KN norm: the speaker may not be able to r espect the str onger norm (KN), but he has to r espect in any case the weaker one to make a pr oper assertion: the only case wher e one is liable to criticism, on the vie w in question, is the case wher e the asserter has no gr ounds for believing the utter ed sentence to be tr ue (as in the paradoxical sounding (7c) above). Now I agree with Lackey (2007, 603–607) that this distinction is spurious, because ther e is only one way of violating a norm: simply b y not doing or believing what it pr escribes. The fact that I am unawar e that I cr ossed the yellow line does not make me less liable to receive the fine from the policeman who notices my bad driving behaviour. That may excuse me, but that does not change the fact that I have violated the rule. The fact that one is unaware of violating the norm does not change one’s status with respect to the norm. Williams’ objection and DeRose’s distinction, however, although they do no signal diff erent ways of r especting or violating a norm, point to another distinction which it is important to make, between the statement of a norm and its regulation. On the one hand a norm can be formulated in an objective sense, as a correctness condition of the form: Practice P is correct only if it has feature C or as an imperative: One must: perform practice P only if one satisfi es feature C. KT, KB and its variants, as well as KN as formulated above are formulated in terms of such correctness and imperative conditions. But on the other hand there are the conditions under which a norm is regulated, or its regulation conditions. These are the subjective conditions under which the correctness conditions are accessed by a given individual and are implemented in his psy chology.7 The former are independent of what the subject believes to be the norm, whereas the latter are so depen7. Although there are many ways of drawing this distinction, I was inspir ed by the way in which Shah (2003) formulates a similar distinction with respect to the norm of truth for belief. See Engel (2008).

56

dent. Of course the two do not hav e to be completely independent, for otherwise one could not see how the norm is regulated, but they need not necessarily coincide. In the case of the kno wledge norm for asser tion, the relevant distinction is between (i) the question whether a subject’s assertions are conform to the norm, and (ii) whether the subject actually believ es that his or her assertions are conform to the norm. M any criticisms of the KN r ule actually assume that conformity to the norm implies (ii). This is why, as we saw abo ve in §1, a number of formulations consider that the norm KN has to include the correctness conditions that the subject “represents himself ” as knowing the content of the asser tion, or some other kind of “self-presentation”. As we saw, the self pr esentation account in the form (1) above, is clearly inadequate. B ut the KNB form, which Williamson (2000, 260) formulates as: (The BK rule): One must assert P only if one believes that one knows P is not obviously so inadequate. AsWilliamson notes, saying that the asserter has only to believe that he knows, or take himself to know, can explain many conservational phenomena which the KN account explains, without asking for a correctness condition as strong as knowing the content of the assertion. But Williamson rightly objects to it that ev en if we add to BK the condition that the belief has to be rational (for one might irrationally believe that one kno ws), one can rationally believ e oneself to kno w that P while not kno wing that P, hence violating the norm KN ( Williamson 2000, 262). I agree with him that the right move, for the defender of KN is not to r elativise it to a belief or to some mental attitude wher eby the subject represents to himself the norm. It is certainly diffi cult to say that an agent does not need in any sense to have access to a norm in or der to follo w it, and familiar attempts to reduce rules and norms to dispositions or rules to tacit knowledge of rules fail in this r espect. But it is not clear that one needs to actually believe anything about the norm in or der to apply it and to per form assertion. Someone who asserts a sentence in the futur e tense—“there will be a sea fight”—obviously does not know, and does not believe that he knows, the existence of the future state of affairs described by the sentence. Nevertheless he makes an asser tion, and he conforms himself to the KN norm. A liar obviously does not believ e what he asser ts, hence he does not kno w it. He nevertheless makes an asser tion. It does not matter her e whether 57

we say that the KN norm is a conv entional feature of assertion in a language in general, or whether it is a commitment of individual speakers. The point is that the norm is pr esupposed even when other norms, such as truthfulness, social norms, or other determinants of what we may want to say, are in place. The fact that the norm can be violated in many cases, overridden by other norms, or be applied in a very loose and relaxed way in many conv ersational cir cumstances does not sho w that the norm is not in place.8 Similarly the fact that it is most often (in most of the cases) very hard to implement does not show that it is not in place.We certainly sometimes do not hav e the slightest idea of what it would take for us to know the things that we assert. But this does not imply that we do not have any idea of what it would take to know anything. In so far as the standard of knowledge is with us in quite many cases (I would argue for many so called “Moorean” propositions, such as that ther e are trees in the garden across my window), it is enough for us to have it for most assertions. It is not because we cannot attain certain kinds of knowledge that the norm is not in place. In this sense, ought does not necessarily imply can. 5. Conclusion I have not, in this paper, tried to examine all the objections that hav e or can be raised against the KAA. Neither have I dealt with the issue whether another epistemic norm than knowledge (warrant or rational credibility) should not be preferred to KN. In fact most of the arguments that I have presented are compatible with the view that some high epistemic status, be it knowledge or not, is a norm of assertion, although there are some good reasons, which I hav e not giv en here, to pr efer the kno wledge account. My main point has consisted in objecting that the kno wledge norm is confused or inapplicable does not thr eaten the claims of KAA, once w e understand what it takes to be a norm of asser tion.9

8. I here concur completely with Williamson 2000, 258–59. 9. Versions of this paper have been read in 2007 in York, Nottingham, Geneva and Nancy. For their comments and encouragements I thank Tom Baldwin, Steven Barker, Berit Brogaard, Greg Currie, Igor Douven Julien Dutant, Frank Lihoreau, Kevin Mulligan, Paul Noordhof, Otto Bruun, Tom Stoneham. I am also indebted to the par ticipants of a wor kshop that I organised at the University of Paris IV in January 2006, Manuel Garcia Carpintero, David Owens, Maria Van der Schaar, and Peter Pagin, whose 2006 article inspired me much.

58

REFERENCES Boghossian, P. 2003. “The Normativity of Content ”, Philosophical Issues, 13, 2003, 32–45. Cohen, J. 1992. An Essay on B elief and A cceptance. Oxford, Oxford University Press. Davidson, D. 1979. “Communication and Convention”, in Inquiries into Meaning and Interpretation, Oxford, Oxford University Press. De Rose, K. 2002. “ Assertion, Knowledge and Context ”, Philosophical Review, 111, 2, 167–203. Douven, I. 2006. “Assertion, Knowledge and Rational Credibility”, Philosophical Review, 115, 449–485. Dummett, M. 1973. Frege, Philosophy of Language. London, Duckworth. Engel, P. 2000. (ed.) Believing and accepting. Kluwer, Dordrecht. — 2005. “Truth and the Aim of B elief ”, in D. G illies (ed.) Laws and Models in Science, London, King’s college Publications. — 2008. “Belief and Normativity”, Disputatio, 23, 153–177. Frede, M. 1987. “The Skeptic’s Two Kinds of Assent and the Question of the Possibility of Knowledge”, in Essays in Ancient Philosophy, Minneapolis, University of Minnesota Press. Hawthorne, J. 2004. Knowledge and Lotteries. Oxford, Oxford University Press. Lackey, J. 2007. “Norms of Assertion”, Noûs 41, 594–626. Marcus, R. B. (1983) “Rationality and Believing the Impossible”, Journal of Philosophy, LXXX, 6, 321–338, repr. in her Modalities, Oxford, Oxford University Press, 1993, 143–161. Moore, G. E. 1962. Common Place Book. C. Lewy (ed.), Cambridge, Cambridge University Press. Pagin, P. 2004. “Is Assertion Social?”, Journal of Pragmatics 36, 833–859. — 2006. “ Knowledge Account of Assertion”, talk at the workshop “Assertion and Knowledge”, University Paris IV, January 2006, ne w version 2007 at http:// people.su.se/%7Eppagin/papers/astnorm.pdf. Shah, N. 2003. “How Truth Governs Belief”, Philosophical Review , 112, 447–82. Slote, M. A. 1979. “Assertion and Belief ”, in J. Dancy (ed.), Papers on Language and Logic, 177–90, Keele, Keele University Library. Unger, P. 1975. Ignorance. Oxford, Oxford University Press. Weiner, M. 2005. “M ust We Know What We Say? ” Philosophical Review,114, 227–251. Williams, B. 2002. Truth and Truthfulness. Princeton, Princeton University Press. Williamson T. 2000. Knowledge and its Limits. Oxford, Oxford University Press.

59

Grazer Philosophische Studien 77 (2008), 61–84.

KNOWLEDGE BY DEDUCTION Ian RUMFITT Birkbeck College, University of London Summary It seems beyond doubt that a thinker can come to kno w a conclusion by deducing it from premisses that he knows already, but philosophers have found it puzzling how a thinker could acquire knowledge in this way. Assuming a broadly externalist conception of kno wledge, I explain why judgements competently deduced from known premisses are themselves knowledgeable. Assuming an exclusionary conception of judgeable content, I further explain how such judgements can be informative. (According to the exclusionary conception, which I dev elop from some remarks in Ramsey , a judgement ’s content is giv en by the hither to live possibilities that it excludes or rules out.) I propose that the value of logic lies in its allowing us to combine diff erent sources of knowledge, so that we can learn things that w e could not learn fr om those sour ces individually. I conclude b y arguing that while single-conclusion logics possess that value, multiple-conclusion logics do not.

1. It seems beyond doubt that an individual can gain knowledge by exercising his capacity for deductive reasoning. Indeed, it seems clear that he can thereby gain knowledge that he could not attain without ex ercising that capacity. Suppose I am strapped to the chair in my study. From that chair, I cannot see the street below. I do, however, see that it is raining, and thus know that it is raining. Moreover I know, ultimately on inductive grounds, that if it is raining the street is wet. Accordingly, I reason as follows:

This paper benefited greatly from the discussion at the Nancy colloquium, and I am duly grateful to the organizers and participants. I also wish to thank Crispin Wright for stimulating correspondence about § 2. *

1.

It is raining 2. If it is raining, the street is wet So 3. The street is wet. The case appears to show how, by exercising a deductive capacity, a thinker can gain knowledge that he would not hav e been able to gain other wise. In the case described, I know that it is raining by virtue of seeing that it is raining. And I know through induction that if it is raining, the str eet is wet. By making the deduction, I thereby come to know that the street is wet. Ex hypothesi, though, I cannot see that the street is wet, so I cannot come to know the conclusion simply b y exercising my per ceptual capacities, which is how I came to kno w the fi rst premiss. Similarly, I cannot come to know the conclusion on inductive grounds alone, which is how I came to know the second premiss. Even in England, so pessimistic a view of the weather (or of the wastefulness of the water companies) would not yield knowledge. But by exercising my deductiv e capacity on the kno wledge that has been delivered by perception and induction, I can come to know something that I could not know on either of those bases severally. Obvious as this may seem, some philosophers hav e been puzzled b y how a thinker can gain knowledge by deduction. There are two problems here: first, how exercising a deductive capacity can yield knowledge; second, how it can yield knowledge that the thinker does not possess already. I take these in turn. The second problem is the harder one, but it can be solved, I argue, if we adopt a certain conception of a statement’s logically relevant content. I conclude with some remarks about the bearing of the resulting conception of logic on the choice between single-conclusion and multiple-conclusion logical systems. 2. It is not, I think, so har d to explain ho w applying a deductiv e capacity to premisses that one kno ws can yield kno wledge of the conclusion. A t least, this is not har d to explain on a br oadly externalist conception of knowledge, whereby a belief has the status of knowledge if it is produced by a method which r eliably yields the tr uth. Let us say that a thinker is logically capable if and only if he is disposed to deduce a conclusion from some premisses only when the conclusion r eally does follow from them, and to recognize at least some of the more obvious cases of one statement’s 62

following from others. Now suppose that a logically capable person knows the premisses of the argument in § 1, and deduces its conclusion fr om those premisses. B ecause he kno ws the two pr emisses, those pr emisses are true. And because the pr emisses are true, and the conclusion follo ws from them, the conclusion is also tr ue. Ex hypothesi, our thinker is logically capable, so he will deduce a conclusion fr om some pr emisses only if it really does follow from them. Accordingly, when a logically capable thinker deduces a conclusion from premisses he knows, the belief thereby formed will be tr ue, and will hav e been pr oduced in a way that r eliably yields true beliefs. That belief, then, has the status of knowledge, or something close to it. This explanation, it may be noted, does not presume that a logically capable thinker kno ws that ‘The street is wet’ follows from ‘It is raining’ together with ‘If it is raining then the str eet is wet’. Indeed, it does not presume that the thinker has the concept of consequence at all. Being able reliably to trace what follo ws from premisses comes fi rst, and is all we need to know things by deduction. Conceiving of statements or propositions as standing in relations of consequence comes later, when we start to do logic, if it comes at all. It may be objected that the explanation just pr oposed fails to explain how w e can know a conclusion deduced fr om kno wn pr emisses. The only operative ingredient in the explanation seems to be that the thinker makes a competent deduction from true premisses. But it will not be true in general that a thinker kno ws a conclusion that he has competently deduced from true premisses: he may not kno w the pr emisses. The last observation is of course corr ect, but it misses the point of the appeal to methods which r eliably yield the tr uth. The ‘method’ of forming beliefs that consists in only believing what logically follo ws from true beliefs is reliable, indeed infallible: for a false belief cannot follow from true beliefs. All the same, actually applying this ‘ method’ pr esumes some antecedent criterion for judging whether the pr emisses are true. So, in the fi rst instance, what master y of a deductiv e capacity pr ovides is not a ne w, reliable method for forming beliefs per se, but rather a means of combining reliable methods of belief-formation that one alr eady possesses so as to yield a ne w method which has a wider range than its components. O f course, the second-or der method may itself be applied to establish the truth of some statements—namely, the logical truths. But the value of our deductive capacities does not lie in their potential for establishing logical truths, which are merely a by-product of our interest in deductive consequence. Rather, it lies in their power to splice together reliable methods of 63

belief-formation to yield further such methods which have a wider range of application. First say it, then qualify it. The first qualification to the explanation just given is needed because risks can accumulate in the course of a deduction in such a way as to undermine kno wledge of the conclusion. A v ersion of the Paradox of the Preface provides a simple example of this. S uppose that an author has composed a book comprising only tr ue statements. Suppose too that the author knows each statement in the book to be true. Now a plausible necessary condition for knowing a statement to be true is that there should be very little risk, given the subject’s evidence, that the statement is false. Ex hypothesi, then, the author will meet this condition in respect of each individual statement in his book. However, he need not meet the condition in respect of the conjunctive statement that is reached by successively applying the logical r ule of ‘ and’-introduction to all the statements in the book. F or even when the risk of each conjunct ’s being false is low, the risk of the conjunction’s being false will be higher, and if the book contains sufficiently many statements the latter risk can be high enough to disqualify the author from knowing the truth of the conjunction, even though it is true and he believes it. This qualification, however, does not undermine the proposed explanation of how a logically capable thinker can come to know a conclusion by deducing it from premisses he already knows. For in many deductions—including short, simple deductions from few premisses—the accumulation of risks will not suffi ce to undermine knowledge. A second qualifi cation is needed on account of what epistemologists call ‘D retske cases’ (the one that follo ws is taken fr om D retske 1970, 1016ff. ). Thus, at the z oo one day, you correctly identify an animal that you see as a z ebra. So you know, it seems, that the animal befor e you is a zebra. That premiss logically entails that the animal befor e you is not a non-zebra carefully disguised to look like a z ebra. So a logically capable thinker may legitimately move from the premiss that the animal is a zebra to the conclusion that it is not a non-zebra carefully disguised to look like a zebra. Many philosophers, however, share Dretske’s intuition that you do not know the conclusion: in or der to know it, you would need evidence that excluded the possibility of the animal’s being a non-zebra disguised as a zebra, and your non-expert glance into the animal’s pen fails to provide such evidence. The precise analysis of these cases is a hotly contested topic in epistemology, but I think we can exclude them without compromising the value of the explanation proposed three paragraphs back. They present 64

a problem because the space of epistemically relevant possibilities expands between premiss and conclusion: in or der to kno w the conclusion, the subject has to exclude a possibility (that the animal is a non-z ebra disguised as a zebra) that is not relevant when we are assessing whether he knows the premiss. So we still have an explanation of how a logically capable thinker can come to know a conclusion by deducing it from known premisses in cases where there is no such expansion in the range of epistemicallyelevant r possibilities. Since many instances of deductiv e argument patently meet this condition, Dretske cases restrict, rather than undermine, the proposed explanation of the epistemic value of deduction. That explanation, though, depends crucially on the factivity of ‘knows’: this is what under writes the infer ence fr om the thinker ’s knowing the premisses to their being true. So it does not extend to explain why a logical capacity would enable a thinker to extend warrants or justifi cations for pr emisses that fall shor t of kno wledge to comparable warrants or justifications for conclusions that are correctly deduced from those premisses. Senses of ‘warrant’ or ‘justification’ that fall short of knowledge will precisely not be factive, so the corresponding arguments for them would break down. But this, I think, is as it should be. We certainly exercise our logical capacities to pr oduce inconclusive justifi cations for beliefs, and I am not suggesting that none of these exercises is legitimate. All the same, I see no reason to assume that the explanation of their legitimacy should be essentially the same as in the case of kno wledge. One straw in the wind here is the way that non-factive justifications, so far from combining to yield a new justification for a statement deducible from justifi ed premisses, frequently undermine each other . Thus, I may have warrant or justifi cation for the claim that Charles is no w in P aris: he always stays ther e at this time of y ear, I saw him heading off to the Eurostar, &c. I may also hav e warrant or justifi cation for the claim that Charles is not in Paris: I have just received a postcard, apparently sent by him from Berlin. If I notice the confl ict of evidence, I may make further investigations to resolve it. But what I will not do—what I cannot do—is to apply ‘and’-introduction to splice my various warrants or justifications together, so that collectiv ely they suppor t the claim that Charles both is and is not in Paris. For nothing at all warrants or justifies that. So far from combining to warrant the conjunctive conclusion, my justifications for the two premisses undermine each other . Cases of this kind ar e ubiquitous, and any explanation of why deduction is v aluable when applied to warrants that fall short of knowledge needs to accommodate them. Theories 65

that model strength of evidential support on conditional probabilities can, I think, accommodate them, while explaining why deductiv e reasoning is still valuable in the non-factive case. Perhaps other theories of non-factive warrant can too . Since my pr esent topic is knowledge by deduction, I cannot go into this matter here. But I hope to have said enough to cast doubt on the assumption that the v alue of deduction must be explained in essentially the same way for factive and non-factive warrants. 3. We have, then, an explanation of how deduction can yield knowledge. It is far harder to explain how it can yield new knowledge. The sense of puzzlement that pr ompts the demand for explanation has an ancient lineage, but fi nds its most powerful articulation in John Stuart Mill’s A System of Logic. ‘We have now to inquire’, he wrote there, whether the syllogistic process, that of reasoning from generals to particulars, is, or is not a process of inference, [that is,] a process from the known to the unknown: a means of coming to a kno wledge of something which w e did not know before (Mill 1891, Book II, chap. iii, § 1).

The difficulty he discerns in an affi rmative answer does not arise only for properly syllogistic reasoning, but for watertight or conclusive deductions more generally: Logicians have been remarkably unanimous in their mode of answering this question. It is universally allowed that a syllogism is vicious if there is anything more in the conclusion than was assumed in the premises. But this is, in fact, to say that nothing ev er was, or can be, pr oved by syllogism which was not known, or assumed to be known, before (ibid.).

This leads Mill to conclude that ‘in every syllogism, considered as an argument to pr ove the conclusion, ther e is a petitio principii’ (II iii 2). The problem is then to reconcile this apparently ‘irrefragable’ doctrine with the indisputable fact that we do sometimes gain knowledge by deduction. Mill’s own solution was to deny that the premisses of a deductive inference are ever, strictly speaking, the gr ounds on which kno wledge of its conclusion rests. His leading example is the argument ‘All men are mortal. The Duke of Wellington is a man. Therefore the Duke of Wellington is mortal’—propounded at a time when the (fi rst) Duke of Wellington 66

was aliv e. ‘The pr oposition that the D uke of Wellington is mor tal’, he comments, ‘is evidently an infer ence; it is got at as a conclusion fr om something else; but do w e, in r eality, conclude it fr om the pr oposition, All men are mortal? I answer, No’ (II iii 3). A thinker may infer that the Duke of Wellington is mortal. But he will reach this conclusion ‘from the death of J ohn and Thomas, and ev ery other person w e ever heard of in whose case the experiment has been fairly tried ’ (ibid.). It is on kno wledge of such individual cases that knowledge of the Duke’s mortality will rest—not on knowledge that all men are mortal. In the present case, the inductive methods that take us fr om knowledge that J ohn and Thomas are mortal to kno wledge of the D uke’s mortality also enable us to infer more generally that all men are mortal. But ‘it is not in the…[deductive] descent from all men to the Duke of Wellington that the inference resides’ (ibid.). Rather: All inference is from particulars to particulars; General propositions are merely registers of such infer ences already made, and shor t formulae for making more; The major pr emise of a syllogism, consequently , is a formula of this description; and the conclusion is not an inference drawn from the formula, but an inference drawn according to the formula; the real logical antecedent or premise being the particular facts from which the general proposition was collected by induction (II iii 4).

Few today will be convinced by Mill’s own solution to his problem. As he says, it rests on the claim that every inference—every instance in which new knowledge is gained on the basis of old—mo ves from particulars to particulars. This claim ignores the fact that the acceptability of an inductiv e inference can depend on deductions from scientific theories. Mill’s theory also breaks down over cases in which the thinker who reaches the eventual conclusion of an apparent deductive inference does not himself possess the evidence that inductiv ely supports the r elevant generalization—as when I come to know by testimony that all men ar e mortal, and infer that the Duke of Wellington (whom I observe to be a man) is mortal. Mill recognizes that such cases pr esent prima facie diffi culties for his account, but insists that they can be accommodated, if we understand testimony to the truth of a general premiss to be a ‘memorandum’ which reminds us, that fr om evidence, more or less car efully weighed, it formerly appeared that a cer tain attribute might be inferr ed wherever we perceive a certain mark. The proposition, All men are mortal, for instance, shows that we have had experience from which we thought it followed that the attributes 67

connoted by the term ‘man’ are a mark of mortality. But when we conclude that the Duke of Wellington is mortal, we do not infer this from the memorandum, but from the former experience (II iii 4).

The difficulty here is obscured by Mill’s casual use of the pronoun ‘we’. Suppose you have observed John, Thomas, and the rest dying. On that evidence, you inductively infer that all men are mortal; and you tell me that all men are mortal. If I take y our word for this, and then learn that the D uke of Wellington is a man, I may infer that he is mortal. But my ground for that conclusion cannot be y our experiences which, in the case described, I do not share. Nor is my ground the claim that ther e is suffi cient evidence to infer the Duke’s mortality from his humanity. Unless there is such evidence, my belief that he is mor tal will fail to qualify as kno wledge. All the same, the claim that ther e is such evidence is not the gr ound for my infer ence. My ground is what you told me—namely, that all men are mortal. Inadequate as Mill’s solution is, he was right to be puzzled: the possibility of gaining knowledge by deduction calls for explanation. He was also right to reject some purported explanations as mere restatements of what needs to be explained. For example, it will not do to follow Bishop Whately and simply invoke a distinction between what is actually asserted and what is asserted ‘merely by implication’: the natur e of this distinction is what calls for explanation (op. cit., II iii 2). A full explanation of ho w we gain knowledge by deduction would be complicated. It would certainly include the point that Frege set out so clearly in ‘Booles rechnende Logik und die Begriffsschrift’ (Frege 1880–81): in making deductions using the rules of the calculus of relations, we are often led to form complex predicates that correspond to new concepts (new modes of classifying things), concepts that are only latent in our pr emisses. But while this obser vation helps to explain the conceptual fertility which is one source of our deductive capacities’ cognitive utility, it does not explain that utility in simpler cases—such as those discussed here—where no new concepts are introduced, and where the rules being applied are only those of the propositional calculus or the monadic fragment of the predicate calculus. 4. I wish to argue that a certain conception of a statement’s logically relevant content (its conceptual content , as I shall follo w Frege in calling it) can provide the missing explanation in these simpler cases. I n this section 68

and the next, I outline the theor y of content and the attendant account of consequence. In § 6, I apply the result to Mill’s problem. I take the relata of consequence, and the bearers of conceptual content, to be statements. But what do I mean by a statement? Let us consider those ordered pairs whose first element is a meaningful, indeed disambiguated, declarative type-sentence and whose second element is a possible context of utterance; by a possible context of utterance, I mean a determination of all the contextual features which can bear upon the truth or falsity of a declarative utterance. Some of these ordered pairs will be such that, were the declarative type-sentence that is the fi rst element uttered in the context that is the second element, a determinate pr oposition, or complete thought, would then be expressed. As I shall use the term, astatement is an ordered pair that meets this condition. Not every ordered pair of declarative type-sentence and possible context of utterance will qualify as a statement in this sense. For example, a pair whose first member is the sentence ‘You are ill’ will not count as a statement unless the context that is its second member supplies an addressee. On this definition, each statement belongs to a language, namely, the language of the sentence that is its first element. Furthermore, each statement possesses a sense or pr opositional content: this will be the thought that would be expressed by uttering the statement’s fi rst element (the declarativ e sentence) in the context that comprises its second element. It then makes sense to classify a statement as true or false simpliciter, according as that thought is true or false. Both assertions, and utterances made within the scope of suppositions, may be instances of statements: utterances of either kind will instantiate a giv en statement when the utterance is of the declarativ e sentence that is the statement’s first member, and is made in the context that is the statement ’s second member. It is, I admit, some what infelicitous to hav e unasserted statements, for ‘states’ often means ‘asserts’; however, alternative terms are more likely to mislead. The term ‘proposition’, for instance, would hav e served better (one may propound something without asserting it) were it not for the fact that so many philosophers already apply the term to what a declarative utterance expresses. In what does a statement’s conceptual content consist? In the relevant context of utterance, a number of possibilities will be taken to be open or ‘liv e’. I hold that a statement ’s content is giv en b y the hither to liv e possibilities that the statement is understood to ex clude. A statement is understood to exclude a possibility if anyone who understands the statement will know that, were the possibility to obtain, then an assertive use 69

of the statement would be incorr ect. A speaker who made such an assertion would have misled his hearers as to the facts. This gloss on exclusion relates the notion to assertive uses of statements, and the obtaining of an excluded possibility would not show a speaker who had made a suppositional use of the statement to hav e misled people. A speaker who makes a supposition, though, will be inviting his audience temporarily to set aside certain hitherto live possibilities, perhaps for the sake of argument, and the possibilities he is inviting his audience to set aside ar e precisely those that ar e excluded in the specifi ed sense. They are, in other wor ds, precisely the possibilities whose obtaining would show the corresponding assertion to be misleading. Thus, when I expr essly suppose ‘A city will never be built her e’, I temporarily set aside the same possibilities whose obtaining would render an assertive use of the same statement incorr ect. (Compare Dummett 1959.) It is because the v ery same possibilities ar e, respectively, outright excluded and temporarily set aside, that the assertion and the supposition shar e a conceptual content. I call this pr oposal the exclusionary theory of conceptual content. Why should we accept this theory? The best reason I know is suggested in the paper where the theory was first clearly propounded. In ‘Facts and propositions’, his pragmatist manifesto of 1927, F . P. Ramsey noted that any attitude of full belief or full disbelief can be defined in terms of the truth-possibilities of atomic propositions with which it agrees and disagrees. Thus, if we have n atomic propositions, with regard to their tr uth and falsity there are 2 n mutually exclusive possibilities, and a possible attitude is giv en by taking any set of these and saying that it is one of this set which is, in fact, realized, not one of the remainder. Thus to believe p or q is to express agreement with the possibilities p true and q true, p false and q true, p true and q false, and disagr eement with the r emaining possibility p false and q false. To say that feeling belief to wards a sentence expresses such an attitude is to say that it has certain causal properties which vary with the attitude, i.e. with which possibilities ar e knocked out and which, so to speak, are still left in. Very roughly the thinker will act in disr egard of the possibilities r ejected, but ho w to explain this accurately I do not kno w (Ramsey 1927, 45–46).

Ramsey’s last sentence suggests the follo wing general argument for an exclusionary theory of conceptual content. When trying to identify what assertive and suppositional instances of statements hav e in common, it makes little sense to focus on gr ounds, for while w e expect speakers to 70

have grounds for their assertions, we do not expect them to have grounds for their suppositions. Matters are diff erent, however, when we focus on statements’ consequences. Most assertions belong not in soliloquy but in colloquy, and are utterances in which the speaker tells a hearer something. And in most cases of telling, the speaker intends his audience to believ e what he says. F or this r eason, it makes sense to identify an asser tion’s content with that of the belief that would be inculcated in a hear er who understood and accepted the asser tion. Ramsey’s pragmatist insight is then that the content of this belief consists in the way that it bears on possible courses of future action. And it bears on futur e action by virtue of ‘knocking out’ possibilities of which a plan of action would other wise have to take account. S o, if an asser tion’s content is that of the corr esponding belief, and if belief in understood in Ramsey ’s pragmatist way, then the assertion’s content will be given by the hitherto live possibilities that it knocks out or ex cludes. We cannot give exactly the same analysis for suppositional instances of statements. For in expressly propounding a supposition, a speaker does not usually intend his hear er to believe what is thereby supposed. I n performing such a speech act, ho wever, he does intend his hearer temporarily to set aside certain possibilities, perhaps for the sake of seeing what follows when they are set aside. And an express supposition will share its content with an assertion if the possibilities thereby set aside are the same as those that someone who accepted the asser tion would thereby disregard when making plans. The exclusionary conception, then, giv es us a way of explaining ho w a supposition and an asser tion can share their logically r elevant content. So far as I can see, it is the only av ailable explanation that pr oceeds by identifying the particular features of the two speech acts in virtue of which their instances can be said to share contents. Although I cannot argue for this here, it seems to me that an explanation of this kind underlies others. Some philosophers will want to say that a supposition and an asser tion can share their logically relevant content by virtue of sharing their truthconditions. Perhaps so, but the Ramseyan explanation underlies this one, for the ex clusionary conception opens the way to a general account of what a statement’s truth-conditions are. Three features of my version of the exclusionary theory of content are worth noting at this stage. (1) According to the theory, a statement’s conceptual content is given by the possibilities that it isunderstood to exclude—not by the possibilities which are in fact inconsistent with its truth, or by the possibilities which 71

could be worked out to be inconsistent with its truth by pure deduction. I intend a statement’s conceptual content to be an aspect of its meaning; that is, I intend it to be something that someone who understands the statement will apprehend. Since the notions of content and understanding are coeval, the theory does not pretend to reduce conceptual content to putatively more primitive terms. It also respects the indeterminacy of content: even when a possibility is inconsistent with a statement, it will often be indeterminate whether the statement is understood to ex clude it. A theory of content must deal with such indeterminacy; it should not try to legislate it away. (2) The theory speaks of possibilities. By a possibility, I mean a way in which things are, or might be, or might hav e been, where the ‘might’ is that of broadly logical possibility. Robert Stalnaker uses a similar formula to explain his conception of a possible world, but he means a completely determinate way in which things—all things—might be. I do not; I mean only a way in which some things might be or might hav e been. S o my possibilities, which might also hav e been called possible states of aff airs, correspond to aspects of Stalnaker’s possible worlds. Stalnaker is a fellow exclusionist. ‘ To make an asser tion’, he writes, ‘is to r educe the context set [sc., the set of possible worlds r ecognized to be live options in a conversation] in a particular way, provided that there are no objections from the other par ticipants in the conv ersation. The particular way in which the context set is r educed is that all of the possible situations incompatible with what is said ar e eliminated’ (Stalnaker 1978, 86). B ut because he takes possibilities to be possible worlds, and neglects the distinction between what is incompatible with what is said and what it understood to be incompatible, S talnaker has gr eat tr ouble explaining why w e do not believe all the logical consequences of our beliefs. (F or his attempt to explain this, see S talnaker 1991, 1999a.) The present version of the exclusionary theory does not face this diffi culty. (3) At the root of the exclusionary theory is the idea that ‘our notions of right and wrong, for assertions as for actions, are asymmetrical, and it is the apparently negative notion which is primary’ (Dummett 1972, 22). For many simple statements, however, the possibility which we understand the statement to exclude is precisely the truth of its negation, and the possibility which we understand the negation to exclude is precisely the truth of the original statement. Thus, if we set aside v agueness, the possibility which we understand ‘It is raining’ to exclude is precisely that of its not raining, and the possibility which w e understand ‘I t is not raining ’ to 72

exclude is precisely that of its raining. For these statements, understanding a statement and understanding its negation are exercises of the same intellectual capacity: they are two sides of one coin. So these are statements to which we could attach the mediaev al tag Eadem est scientia oppositor um, and for them, the asymmetry Dummett describes becomes nugatory, or at least invisible. Not all statements are like this, however. Dummett’s own example of ‘A city will nev er be built her e’ (uttered in some r ural location) nicely brings out the intuitiv e difference. We understand the statement as excluding each of the follo wing possibilities: a city ’s being built here after one day, a city’s being built her e after two days, … I n each of these possibilities ‘A city will one day be built here’, which is the original statement’s negation, is tr ue. For were it the case that a city is built her e after a thousand days, ‘ A city will one day be built her e’ would be tr ue. All the same, it is unclear quite what possibilities the latter statement is understood to exclude. It is surely wrong to answer: the possibility that, at the end of time, a city will not hav e been built at the r elevant place. For if time continues indefi nitely, there will be no such thing as the end of time. I cannot ar ticulate here the principles that implicitly guide our intuitive judgements about this case. B ut the statement is one for which Dummett’s asymmetr y is r eal, so that the ex clusionary theor y off ers a distinctive account of its content. E xcessive concentration on statements that conform to eadem est scientia oppositor um has, I fear , led people to overlook the distinctiv e character of the ex clusionary theory of content, despite the powerful general considerations in its favour. 5. What account of consequence does the exclusionary theory yield? To answer this question we must fi rst say what it is for a statement to be true, and the natural exclusionary account is as follows: A statement is tr ue if and only if , for ev ery possibility that the statement is understood to exclude, something obtains which precludes the obtaining of that possibility. This explanation (and especially the component notion of pr eclusion) itself needs glossing. E ven in adv ance of that gloss, though, the explanation accounts for the validity of our initial inference. As we saw, ‘It is raining’ 73

is a simple statement which conforms to eadem est scientia oppositor um: the possibility the statement is understood to exclude is simply the possibility that it is not raining. So if this statement is true, something obtains which precludes the possibility that it is not raining. As for the conditional statement ‘If it is raining, the street is wet’, this is understood to exclude any possibility in which it is raining while the str eet is not wet. So if the conditional is true, something obtains which pr ecludes the possibility of its raining while the str eet is not w et. Suppose now that both of these statements are true, and consider the possibility that the street is not wet. If this possibility were to obtain, then either it would be the case that it is raining and the street is not wet, or it would be the case that it is not raining and the street is not wet. Now because the argument’s second premiss (the conditional) is tr ue, something obtains which pr ecludes the obtaining of the first of these sub-possibilities. And because the argument’s first premiss is tr ue, something obtains which pr ecludes the obtaining of the second of these sub-possibilites (which is a possibility in which it is not raining). So, for any possibility in which the street is not wet, something obtains which precludes the obtaining of that possibility. However, those possibilities are precisely what the conclusive statement ‘The street is wet’ is understood to exclude. So the condition for the truth of the conclusion is met, and we have an account of the inference’s validity: the truth of the premisses guarantees the truth of the conclusion. Some may find this account of the validity of modus ponens perplexing. In supposing that any possibility in which the street is not wet must either be a possibility in which the str eet is not wet and it is raining, or be one in which the street is not wet and it is not raining, the explanation r elies on an instance of ex cluded middle. Accordingly, it will extend to yield a general account of the v alidity of modus ponens only if ex cluded middle is taken to be a logical law. But, it may be observed, intuitionist logicians adhere to modus ponens unrestrictedly while accepting only a r estricted version of ex cluded middle. The obser vation is corr ect, but all it sho ws is that an intuitionist logician cannot accept the ex clusionary theor y of content. The intuitionist will have to give a very different account of why modus ponens is v alid, as indeed will a classical logician who r ejects the exclusionary theory of conceptual content. That, however, does not detract from the interest the above account of modus ponens has for a classical logician who accepts the theor y. For, as I now argue, the account pr ovides a solution to Mill’s problem.

74

6. Mill’s problem concerns knowledge, not truth. How does the ex clusionary theory help to explain the possibility of a thinker’s gaining knowledge through a simple deduction? Let us revert to our original case and suppose that a thinker knows (by seeing) that it is raining. Then he will hav e per ceptual knowledge that rules out any possibility in which it is not raining. S uppose too that he knows that the street is wet if it is raining. Then he will have knowledge (inductive knowledge in this case) that rules out any possibility in which it is raining and the str eet is not w et. Because any possibility in which the street is not wet is either a possibility in which it is raining or one in which it is not, any possibility in which the str eet is not wet is ruled out by something he kno ws. So we can agr ee with M ill that a thinker who knows the premisses of our argument is in a position to know the conclusion: for any possibility that the conclusion is understood to ex clude, he knows something that pr ecludes the obtaining of that possibility . The knowledge our thinker possesses by virtue of knowing the premisses already excludes all the possibilities that need to be excluded in order to establish the conclusion. Moreover, Mill is right that this will always be tr ue of a deductively watertight inference. However, w e are now in a position to explain why the thinker may not actually kno w the conclusion. I n order to kno w that, he has to do some conceptual wor k in or der to see that each possibility ex cluded b y the conclusion has alr eady been ex cluded by the pr emisses. For he will only see that the pr emisses jointly establish the conclusion if he hits on a way of partitioning the possibilities that the conclusion excludes which makes it manifest that all these possibilities ar e already excluded by the premisses. In the present case, this is achieved by the simple partitioning of cases in which the str eet is not w et into cases in which it is not w et and it is not raining, and cases in which it is not w et and it is raining. A thinker’s deductive capacity rests on his ability to fi nd suitable partitions of this kind, and this ability goes bey ond anything that is strictly implicated in his kno wledge of the pr emisses. So although ther e is a sense in which a sound argument ’s conclusion is contained in its pr emisses, and although knowledge of the pr emisses puts us in a position to kno w the conclusion, actual kno wledge of the conclusion need not be contained in knowledge of the premisses. We need to find the appropriate partition of the excluded cases. Deductive rules are, in eff ect, rules for doing this. 75

That is enough for a deductive capacity to be epistemically useful even in these simple cases. The analysis also makes it clear ho w some sour ces of kno wledge can collectively establish a conclusion that none of them establishes b y itself. The possibility that the street is not wet is not ruled out by knowledge that it is raining: for there may be circumstances in which it rains without the street’s getting wet. Neither is the possibility that the street is not wet ruled out by the knowledge that the str eet is wet when it rains: it may not be raining. But the partition of possible cases shows that each case in which the street is not wet is excluded by one source of knowledge or the other. This shows how we should individuate the premisses of arguments: namely, by reference to the sour ces of the pr oponent’s knowledge of them. P hilosophers and logicians sometimes write as though it makes no difference whether an argument ’s premisses are treated separately or amalgamated into a conjunction. Thus, in the famous passage in which he appropriated the term ‘entails’ from the lawyers, we fi nd G. E. Moore writing casually about ‘the sense in which the conclusion of a syllogism in Barbara follows from the two pr emisses, taken as one conjunctiv e pr oposition’ (Moore 1922, 291). B ut it does make a diff erence, ev en over arguments with finitely many premisses. Since a deductive capacity’s epistemic power lies in its ability to combine knowledge from various sources to yield knowledge that would not be obtainable without it, an accurate r epresentation of one of its ex ercises must present as distinct premisses the deliverances of different sources of knowledge—as they may be, perception, memory, testimony, induction, abduction, or the results of previous deductions. The two premisses of a syllogism in Barbara may easily be grounded in different sources of knowledge (induction and testimony, perhaps). A thinker may apply ‘and’-introduction to the premisses and thereby reach a conjunctive conclusion. But in doing so, he will be ex ercising (albeit very gently) his capacity for deduction. The present theory explains, indeed, how a thinker may gain knowledge via a syllogism in Barbara. The possibilities that the conclusion ‘All As are C ’ is understood to exclude comprise all possible states of affairs in which some A is not C. Let us then consider an arbitrary such possibility, that in which a particular A object—call it a—is not C. Now any such possibility will either be a possibility in which a is B but not C, or it will be one in which a is neither B nor C. Consider first any possibility of the first kind. Such a possibility will be one in which someB is not C. The second premiss of Barbara (‘All Bs are C ’) is understood to ex clude any such possibility, 76

and thus excludes all possibilities of the fi rst kind. As for any possibility of the second kind, it will be a possibility in which some A is not B. The first premiss of the argument (‘All As are B’) is understood to exclude any such possibility, and thus excludes all possibilities of the second kind. As before, then, we can say the following. (1) Any instance of Barbara is valid: any possibility that the conclusion is understood to exclude is excluded by the truth of the premisses. (2) A thinker who kno ws the premisses of an instance of Barbara is in a position to know its conclusion: for any possibility that the conclusion is understood to exclude, he knows something that precludes its obtaining. (3) All the same, such a thinker need not actually know the conclusion. The conceptual work needed to fi nd the right way of partitioning the class of possibilities ex cluded by the conclusion goes beyond what is required to know the premisses. Minimal as that work is in the present case, the necessity for it is enough to explain how a thinker can gain knowledge by making the deduction. As before, the explanation assumes excluded middle in supposing that each possibility in which some A is not C is either one in which some A which is B is not C or one in which some A which is not B is not C. 7. Our analysis bears, I think, on the v exed question of whether systems of multiple-conclusion logic have the same status as the more familiar singleconclusion calculi. We may agree that in some areas of our intellectual economy, multipleconclusion consequence relations can play rôles recognizably akin to those played by single-conclusion consequence relations. An important example is the way that consequence relations constrain combinations of acceptance and rejection of statements. The mark of an attitude of acceptance, as I shall use the term, is that a thinker who bears such an attitude to a statement will thereby be mistaken unless the statement to which he bears it is true. Truth is thus the norm for acceptance. Belief, plainly, is one such attitude, but there are others—notably the kind of ‘ acceptance’ which some philosophers of science say is the proper attitude to adopt towards currently well-confirmed scientific theories, given the strong ‘pessimistic’ inductive evidence that those theories will eventually be refuted. It will be clear from this that the relevant notion of being mistaken is being mistaken as to the facts. A thinker may be mistaken in this sense even though his acceptance 77

of a statement is in no way irrational, nor even epistemically irresponsible. Similarly, the mark of an attitude of rejection is that a thinker who bears such an attitude to a statement will thereby be mistaken unless the statement to which he bears it is untrue. Just as belief is a paradigm attitude of acceptance, so disbelief is a paradigm attitude of rejection. And just as we express acceptance of a statement b y asserting it, so w e express rejection of a statement by denying it. This latter speech act may be performed, for example, by answering ‘no’ to the corresponding yes-or-no question. Part of the interest of single-conclusion consequence relations certainly lies in the way they bear on combinations of instances of acceptance and rejection. That bearing is partly captured in the following principle: (N) If a statement B follows from a statement A, then a thinker who accepts A and rejects B will be making at least one mistake as to the facts. If B follows from A, then a certain combination of acceptance and rejection—viz. accepting A and rejecting B—will involve at least one mistake as to the facts. (It is a further question how one tries to correct this mistake.) Indeed, giv en the glosses placed on ‘ accept’ and ‘r eject’, principle ( N ) admits of proof in a classical metalogic. For suppose that B follows from A. Then we certainly have that if A is actually tr ue then so is B. Given a classical metalogic, it follo ws from this that either A is not tr ue or B is true. Now if A is not tr ue, then any acceptance of A is mistaken. And if B is true then any rejection of B is mistaken. Either way, then, a thinker who both accepts A and rejects B will be making a mistake. But that is just what (N ) says. This proof of (N), it may be noted, requires only that either B is actually tr ue or A is actually untr ue, given that B follows from A. It does not exploit any necessity there may be in the way B’s truth depends on A’s. So the proof of (N) goes through even for the weakest consequence relation—the relation of Philonian consequence, in which A stands to B except when A is actually true and B is actually untrue.1 1. What of the conv erse of (N)? If, in the specifi ed senses, it is a mistake to accept A and reject B, will B follow from A? The answer is ‘no’; a version of Moore’s paradox provides a counterexample. A thinker who accepts that it is raining but denies that he accepts that it is raining will be making a mistake as to the facts. B ut the statement ‘He accepts that it is raining’ is not a consequence of the statement ‘It is raining’. It may not even be a Philonian consequence: we may suppose that it is in fact raining, but that no one accepts that it is. This sort of case is a problem for those who seek to explicate consequence in normativ e terms, but no such pr oject is contemplated here.

78

Principle (N ), and its proof, extend straightforwardly to the case of multiple premisses. If B follows from A1,…, An then a thinker who accepts all the Ai and rejects B will be making at least one mistake as to the facts. What is striking, however, is that the principle, and the proof, also extend without strain to yield an account of the force of instances of multiple-conclusion consequence. Multiple-conclusion consequence obtains between two sets of statements X and Y when it is impossible for all the members ofX to be true without at least one member ofY ’s being true. An argument parallel to the proof of (N ) shows that if a setY is a multiple-conclusion consequence of a set X then a thinker who accepts all the members of X while rejecting all the members of Y must be making at least one mistake. This result has signifi cance for the enterprise of multiple-conclusion logic. The rarity, to the point of extinction, of naturally occurring multiple-conclusion arguments has always been the r eason why mainstr eam logicians have dismissed multiple-conclusion logic as little mor e than a curiosity. (See e.g. Tennant 1997, 320.) And attempts b y enthusiasts to alleviate the embarrassment her e have often ended up compounding it. In the introduction to their textbook on the subject, Shoesmith and Smiley concede that multiple-conclusion pr oofs can scarcely be said to form part of the ev eryday repertoire of mathematics. ‘Perhaps the nearest one comes to them’, they go on, ‘is in proof by cases, where one argues “suppose A1 … then B; …; suppose Am … then B; but A1 › … › Am, so B”. A diagrammatic r epresentation of this argument exhibits the do wnward branching which we shall see is typical of formalised multiple-conclusion proofs … But the ordinary proof by cases is at best a degenerate form of multiple-conclusion argument, for the diff erent conclusions ar e all the same (in our example they are all instances of the same formulaB)’ (Shoesmith and Smiley 1978, 4–5). ‘At best degenerate’, though, hardly says it. I do not know how the word ‘multiple’ is used in Cambridge, but in the rest of the E nglish-speaking world it is understood to mean ‘ more than one’. So an example of an argument in which all the conclusions (sic) are identical provides little justifi cation for taking multiple-conclusion logic seriously. But since this is all that S hoesmith and Smiley provide by way of a positive case for deeming their system to be a branch of logic, readers of their book may be forgiv en for closing it with a sigh on r eaching p. 5 of the introduction. Our discussion suggests a better justifi cation for their subject. Even if an instance of multiple-conclusion consequence yields no direct appraisal of any naturally occurring argument, it bears on the evaluation of certain 79

combinations of acceptances and r ejections in a way that r ecognizably generalizes (N ). Seen from this angle, then, ther e seems to be no good reason to privilege multiple acceptances o ver multiple r ejections. Let us grant that there are no naturally occurring multiple-conclusion arguments or proofs. If logic were primarily the theory of sound argument or proof, that would matter. But, it might be claimed, its import lies in the way in which it constrains combinations of acceptances and r ejections. On this view, there is no warrant for a lopsided r estriction of the conclusion sets to singletons. I think that this is the best case one can make for multiple-conclusion logic.2 Here, though, the best is not good enough. The basic problem with the defence is that ( N ) does not capture anything like the full force of single-conclusion consequence. It says that when B follows from A, a thinker who accepts A and rejects B will be making a mistake. So he will. But—just to take one example among many—the principle says nothing about the case of someone who accepts A, who knows that B follows from A, but who r efuses to accept B. A thinker who is in this position need not be making any mistake as to the facts. But the force of single-conclusion consequence is still eluding him. This is why we get irritated when we encounter such thinkers in our seminars. (‘What do you mean, you refuse to accept B? You continue to adhere to A, and I’ve shown you that B follows from A.’) Yet this aspect of the for ce of consequence does not transfer to the multiple-conclusion case. A thinker who accepts all the statements in a set X, who knows that a set Y is a multiple-conclusion consequence of set X, but who refuses to accept any statement in Y need not be making any mistake. Of course, he will be making a mistake if he refuses to accept the claim that some member of Y is true. But that point is grist to the mill of sceptics about multiple-conclusion logic. Yet again, they will say, we can only understand an instance of multiple-conclusion consequence as an instance of single-conclusion consequence in which the conclusion is a disjunctive or existentially quantifi ed claim.3 2. Something like this case for multiple conclusions is pr esented in R estall 2005. B ut he overplays his hand in suggesting that ‘ Y is a multiple-conclusion consequence of X ’ can be explained as meaning ‘The mental state of accepting all of X and rejecting all of Y would be selfdefeating’. The mental state that consists of accepting that there will never be sufficient grounds for accepting or rejecting ‘There is a god’, while rejecting that very statement, is self-defeating. But ‘There is a god’ is in no sense a consequence of ‘There will never be suffi cient grounds for accepting or rejecting “There is a god”’. 3. A cognate problem confronts Allan Gibbard’s attempt to extend the notion of validity to practical arguments—arguments whose conclusions are decisions rather than the bare acceptance

80

The analysis of the previous section brings out, indeed, how impoverished a conception of the force of consequence (N ) and its generalizations afford. Those generalizations r ecord a priori constraints on combinations of acceptances and r ejections, and it is of some inter est to hav e rules which yield such constraints. B ut constraints of that kind ar e only a small par t of what makes single-conclusion logic impor tant. As w e have seen, its r ules ar e also of inter est because, b y follo wing them, a thinker can combine diff erent sour ces of kno wledge to come to kno w things that those sour ces do not establish sev erally. B ut this splicing together of diff erent sour ces of kno wledge to yield ne w pr opositional knowledge presumes many premisses but only one conclusion. The epistemic importance of that splicing justifies logic’s traditional focus on laws governing consequence r elations that r elate many pr emisses to a single conclusion. It is important to settle this matter, for the differences between multipleconclusion and single-conclusion logical systems r un deep. In particular, the pressure exerted by certain logical paradoxes is directed very differently in single-conclusion and multiple-conclusion frame works. Ian Hacking (1979, 292–3) draws our attention to the seemingly ‘ magical fact’ that the same operational r ules yield intuitionist logic in their single-conclusion form and classical logic in their multiple-conclusion form. But this is just one illustration of the way the str uctural rules of a multiple-conclusion calculus bear far more of the deductive weight than do the analogous rules in a single-conclusion system, and hence come mor e into contention in disputes between rival logical schools. An even more striking illustration of this concerns the multiple-conclusion cut rule, which Shoesmith and Smiley formulate as follows:

of statements. His leading example is Sherlock Holmes’s reasoning: ‘Either packing is now the thing to do, or by now it’s too late to catch the train anyway. It’s not even now too late to catch the train. Therefore packing is now the thing to do.’ ‘We can see’, Gibbard writes, ‘what makes Holmes’s practical argument valid. It took the form: F or P, not F, therefore P. An argument of this form is valid, even if to accept P is to come to a decision. To accept the premises and reject the conclusion would be to rule out every way that Holmes could become opinionated factually and fully decided in his hyperplan’—i.e., could come to have a opinion on every question of fact and to have made a decision about ev ery question of what to do (G ibbard 2003, 59). Holmes would indeed be inconsistent if he accepted the practical argument ’s premisses and rejected its conclusion. But there remains a gap betw een rejecting the r ejection of a practical conclusion and accepting it, so Gibbard does not capture the sense in which a successful deduction for ces one to accept its conclusion.

81

(Multiple-conclusion cut) If there exists a set of statements Z such that X, Z1 : Z2, Y for each par tition of Z, then X : Y (op. cit., 29). (A partition of Z is an or dered pair such that Z1 ‰ Z2 = Z and Z1 ˆ Z2 = ‡.) Now in the envisaged justifi cation of multiple-conclusion logic, all that matter ed was that the r elevant ‘logical’ possibilities should include the actual circumstances. For that was enough to ensure that (N) and its generalizations ar e tr ue. Accordingly, one of the multiple-conclusion consequence r elations to which the laws of multiple-conclusion logic apply must be the r elation in which the only r elevant possibilities are actual. This will be the material, or P hilonian, multiple-conclusion consequence relation B, which obtains betw een a set of pr emisses X and a set of conclusions Y when either at least one member of X is not true or at least one member of Y is true. Thus the negation H of this relation will obtain between X and Y when each member of X is true but no member of Y is true. Instantiating the schematic ‘:’ in (Multiple-conclusion cut) with this Philonian relationship and contraposing, we reach (Philonian cut) If X H Y then for every set Z of statements, there exists a partition of Z such that X, Z1 H Z2, Y. The observation I wish to make is that (Philonian cut) contains a strong semantical assumption which goes far bey ond the transitivity of consequence. This comes out most clearly if w e apply the r ule in a language with vague predicates: anyone who wishes to deny that v ague predicates possess sharp boundaries or cut-offpoints is forced to deny (Philonian cut) and hence to deny the str uctural rule (Multiple-conclusion cut) of which it is a contraposed instance. To see this, consider a sequence a1, …, a100 of colour samples which range gradually fr om a1, which is clearly red, to a100, which is clearly orange and hence clearly not r ed. And consider the corresponding sequence of statements P1, …, P100 in which Pi predicates redness of ai . Let us then instantiate (Philonian cut) by taking the set X to be the singleton {P1} and Y to be the singleton {P100}. The case has been constructed so that P1 is true and P100 is not true, so we certainly have P1 H P100. By (Philonian cut), then, we infer that For every set Z of statements, there exists a partition of Z such that P1, Z1 H Z2, P100. 82

In par ticular, then, ther e must be a par tition < Z 1, Z2> of the set {P2, …, P99} such that P1, Z1 H Z2, P100. That is to say, there must be a partition of the set {P2, …, P99} such that all the members of Z1 are actually true while none of the members ofZ2 is. Now clearly the only partition that could possibly have this property is one where Z1 = {P2, …, Pk } and Z2 = { Pk +1, …, P99} for some integer k between 2 and 98. S o we are committed to the existence of a k for which P1, …, Pk H Pk+1, …, P100. But that is to say: we are committed to the existence of a k such that each statement up to and including Pk in the sequ ence is true, while no statement after Pk is true. We are committed, in other words, to a sharp cut-off point in the tr uth-values of the Pi s, and hence a sharp cut-off point in the redness of the ai s. So anyone who wishes to deny that there is such a cut-off point must deny (Multiple-conclusion cut). The moral I draw fr om this is that the content of ( Multiple-conclusion cut) includes much mor e than the transitivity of consequence. The concentration of logical po wer in the str uctural rules is characteristic of multiple-conclusion systems, which may be likened to those high-po wered sports cars whose engines need to be taken apar t and rebuilt almost from scratch if one is to make ev en a minor adjustment to them: fun to drive, but a nightmar e to maintain. B ut we have found r eason to leav e multiple-conclusion logics to the boy racers, and focus on the single-conclusion rules, by following which we can splice together the deliv erances of various sources of kno wledge to come to kno w things that w e could not know otherwise.

REFERENCES Dretske, F. I. 1970. ‘E pistemic operators’. The Journal of P hilosophy 67, 1007– 23. Dummett, M. A. E. 1959. ‘Truth’. Proceedings of the Aristotelian Society 59, 141–62. Page references to the reprint in Dummett 1978, 1–19. — 1972. ‘Postscript (1972) to “Truth”’. In J. M. E. Moravcsik, ed., Logic and Philosophy for Linguists: a Book of R eadings (The Hague: Mouton, 1974), 200–5. Page references to the reprint in Dummett 1978, 19–22. — 1978. Truth and Other Enigmas. London: Duckworth. Frege, F. L. G. 1880–81. ‘Booles r echnende Logik und die B egriffsschrift’. In Frege, ed. H. Hermes et al., Nachgelassene Schriften (Hamburg: Felix Meiner, 1969), 9–52. 83

Gibbard, A. F. 2003. Thinking How to Live. Cambridge, Mass.: Harvard University Press. Hacking, I. M. 1979. ‘What is logic?’ The Journal of Philosophy 76, 285–319. Mill, J. S. 1891. A System of Logic, Ratiocinative and Inductive, 8th edition. London: Longman. Moore, G. E. 1922. Philosophical Studies. London: Routledge and Kegan Paul. Ramsey, F. P. 1927. ‘Facts and propositions’. Proceedings of the Aristotelian Society Supplementary Volumes 7, 153–70. P age references to the r eprint in Ramsey 1990, 34–51. — 1990. Philosophical Papers, ed. D.H. Mellor. Cambridge: Cambridge University Press. Restall, G. 2005. ‘Multiple conclusions’. In P. Hajek et al., eds., Logic, Methodology, and Philosophy of Science: Proceedings of the Twelfth International Congress (London: King’s College Publications), 189–205. Shoesmith, D. J. and T. J. Smiley. 1978. Multiple Conclusion Logic . Cambridge: Cambridge University Press. Stalnaker, R. C. 1978. ‘Assertion’. Syntax and Semantics 9, 315–332. P age references to the reprint in Stalnaker 1999b, 78–95. — 1991. ‘The problem of logical omniscience, I’.Synthèse 89, 425–440. Reprinted in Stalnaker 1999b, 241–54. — 1999a. ‘The problem of logical omniscience, II’. In Stalnaker 1999b, 255–73. — 1999b. Context and Content: E ssays on I ntentionality in S peech and Thought . Oxford: Oxford University Press. Tennant, N.W. 1997. The Taming of the True. Oxford: Clarendon Press.

84

Grazer Philosophische Studien 77 (2008), 85–125.

QUESTION-EMBEDDING AND FACTIVITY Paul EGRÉ Institut Jean-Nicod, Paris Summary Attitude verbs fall in diff erent categories depending on the kind of sentential complements which they can embed. I n English, a v erb like know takes both declarative and interrogative complements. By contrast, believe takes only declarative complements and wonder takes only interrogative complements. The present paper examines the hypothesis, originally put forward by Hintikka (1975), that the only verbs that can take both that-complements and whether-complements are the factiv e verbs. I argue that at least one half of the hypothesis is empirically correct, namely that all v eridical attitude v erbs taking that-complements take whether-complements. I distinguish veridical verbs from factive verbs, and present one way of deriving the generalization. Counterexamples to both directions of the factivity hypothesis ar e discussed, in particular the case of emotive factive verbs like regret, and the case of non-veridical verbs that licence whethercomplements, in par ticular tell, guess, decide and agree. Alternativ e accounts are discussed along the way, in particular Zuber (1982), Ginzburg (1995) and Saebø (2007).

1. Introduction: the factivity hypothesis Attitude verbs in English fall in different categories depending on the kind of sentential complements which they can embed. In English, a verb like know takes both declarative and interrogative complements. By contrast, a verb like believe takes only declarative complements. A verb like wonder, on the other hand, takes only interrogative complements: (1)

a. Pierre knows that it is raining b. Pierre believes that it is raining c. *Pierre wonders that it is raining

(2)

a. Pierre knows whether it is raining b. *Pierre believes whether it is raining c. Pierre wonders whether it is raining

These contrasts in embedding behavior do not only hold for E nglish but ar e widely suppor ted by cross-linguistic data (Kar ttunen 1977, Ginzburg 1995, Lahiri 2002). The contrast betw een wonder and know, in particular, plays an impor tant role in the theor y of questions put forward by Groenendijk and Stokhof (1982), and serves as the basis of their distinction betw een the intension and the extension of a question (to which we shall return). To some extent, the opposition between know and believe has been less systematically studied, despite a number of obser vations on the contrast, both b y philosophers and b y linguists ( Vendler 1972, Hintikka 1975, Lewis 1982, Zuber 1982, Groenendijk and Stokhof 1982, B erman 1991, G inzburg 1995, Lahiri 2002, and mor e recently, Saebø 2007). In his paper, however, Hintikka observed that the diff erence in question-embedding behavior between know and believe seems to be correlated with the factiv e vs. non-factiv e feature of each of the two v erbs. Thus, Hintikka conjectured that “only verbs which can have a success force may enter into a wh-construction” (Hintikka 1975, 21). For instance, non-factive verbs used to report belief, like pretend or claim, are non-factive and do not appear to embed questions. On the other hand, attitude verbs like see, remember, forget, which also bear a semantic connection to know, are both factive and question-embedding. Despite this, H intikka did not pr ovide a detailed explanation of the link between factivity and question-embedding behavior. The aim of the present paper is to take a closer look at this connection, namely to examine how robust it is and how it may be derived. Let us call Hintikka’s hypothesis the factivity hypothesis, that is the hypothesis according to which the only verbs that take both declarativ e and interr ogative complements ar e the factive verbs. The factivity hypothesis immediately raises two pr oblems: the first concerns the defi nition of the notion of factivity that is r elevant in the formulation of the hypothesis. Among philosophers and logicians, a predicate is generally called factiv e if it is simply veridical, namely if it entails the tr uth of its complement (see e.g. Williamson 2000). Among linguists, a predicate is usually called factiv e if it presupposes the tr uth of its complement (Kiparsky & Kiparsky 1970, Karttunen 1971). Which of these two defi nitions is most relevant is of central importance for testing 86

the empirical adequacy of the generalization, which is the second main problem raised by the hypothesis. Thus, v erbs like tell or decide ar e not v eridical, and y et they admit whether-complements. This suggests that v eridicality is not a necessar y condition in order for a verb that takes that-complements to also embed questions. Conversely, as pointed out by Lahiri (2002), a verb like regret is generally considered factive and veridical, but it does not admit whethercomplements. This suggests that factivity is not even a sufficient condition for verbs that take that-complements to embed questions. In what follows, however, I want to suggest that at least one half of the factivity hypothesis is correct. More precisely, my suggestion is that all veridical attitude verbs are question-embedding. The converse direction, as w e will see, is mor e problematic, and it not obvious that all verbs that admit boththat-complements and whether-complements should be factive. The aim of this paper, however, is to try and defend the hypothesis in its strong form, if only to see where it is most likely to break down. Before we proceed, I want to make the cav eat that this paper is fundamentally on the link between that-complements and whether-complements, rather than on wh-complements in general.To be more precise, what I want to examine is to what extent factivity can be seen as necessar y and sufficient in order for a v erb that takes that-complements to take whether-complements. The restriction to whether-complements has to do with the fact that some verbs that do not licence whether-complements do seem to licence wh-complements. This happens with emotive predicates like surprise, and even with verbs like believe and regret in appropriate contexts:1 (3) Peter will never believe who came to the party (4) Peter was surprised by who came to the party (5) Peter really regrets who came to the party Not all native speakers of English find (4) and (5) completely acceptable, but for some of them these sentences ar e fi ne. On the other hand, for both kinds of speakers the same contexts clearly r ule out the embedding of whether-complements: (6) *Peter will never believe whether Mary came to the party 1. See Lahiri (2002) for a discussion of whether wh-complements should be seen as questions (rather than free relatives, or exclamatives) after such verbs.

87

(7) *Peter was surprised by whether Mary came to the party (8) *Peter really regrets whether Mary came to the party In what follo ws, we will r efer to whether-complements as the basic test for question-embedding, setting aside the explanation of why v erbs like surprise are nevertheless able to licence wh-complements.2 The paper is structured as follows. In Section 2, I first discuss three different accounts of the impossibility of embeddingwhether-complements after believe, namely the accounts of Zuber (1982), Ginzburg (1995) and Saebø (2007). As we will see, all accounts make direct or indirect use of the notion of factivity to explain why believe, in par ticular, fails to embed questions. None of those, however, explains why factive verbs would embed questions in the fi rst place. I n Section 3, I make a diff erent proposal to deriv e the generalization that all v eridical attitude verbs embed questions. I contrast veridical and factive verbs and explain on what assumptions a veridical verb taking that-complements may be expected to embedwhether-complements. The behavior of pr edicates like regret and it is tr ue that, which challenge the assumption that v eridicality is suffi cient for question-embedding, is discussed in detail. Section 4 examines the conv erse of the generalization, namely the case of non-v eridical verbs that ar e question-embedding (like tell, decide, agree on). In the case of non-veridical verbs like agree and decide, I argue that the presence of an overt or covert preposition (like on, or about) is needed to account for the embedding of whether-complements. The case of tell is diff erent, but it is at least consistent with the pr esent account to assume that it is systematically ambiguous betw een a v eridical and nonveridical reading. To get to the heart of my proposal, the reader may wish to go dir ectly to the positiv e account laid out in S ection 3, and to come back only later to the discussion of alternative accounts given in Section 2. 2. Facts and factivity In this section I compare three different accounts of the distribution ofthatand whether-complements among attitude verbs. I start with the examination 2. See Saebø 2007 for a r ecent discussion of whether-complements after surprise. One of Saebø’s examples is par ticularly intriguing, namely “D on’t read this installment befor e seeing the episode if you want to be surprised at whether or not H ercules makes it” (example (25) in Saebø’s paper). Saebø uses the example to argue that the exclusion of whether-complements after surprise is pragmatic, and not syntactic or semantic.

88

of a little kno wn note b y Zuber 1982, and go on to examine G inzburg’s 1995 theor y, and the pragmatic account r ecently put for ward by Saebø 2007. In all of these theories, w e will see that the notion of factivity plays an explanatory role, but only indir ectly. In particular, none of these theories explains why factiv e verbs should embed questions in the fi rst place. 2.1 Neg-raising and factivity (Zuber 1982) In a brief note, Zuber observed that neg-raising verbs cannot take whethercomplements, and pr oposed to explain this failur e on the basis of their neg-raising behavior. Believe, in particular, is neg-raising, which means that a sentence of the form “ x does not believ e that p”, where negation takes wide scope o ver the v erb, can be used to mean the str onger “x believes that not p”, in which the verb raises above negation: (9)

Mary does not believ e that it will rain º Mary believes that it will not rain

A verb like believe clearly differs from know in this respect, since for know that inference is not available: (10) Mary does not know that it will rain i Mary knows that it will not rain The inference associated to neg-raising is not usually consider ed to be a semantic entailment, but rather a semantic or pragmatic infer ence of a different sort (namely a pr esupposition, or an implicatur e).3 Zuber does not say what he takes the status of the infer ence to be, but he basically handles it as a semantic entailment for the purpose of his argument. Zuber’s explanation makes fur ther assumptions about the semantic behavior of whether-complements. In particular, Zuber assumes that for any verb V that takes both that- and whether-complements: (a) “ V that p” implies “V whether p” (Vp º
V ?p) and likewise: 3. See Gajewski 2005, chapter 1, for a review of different views about neg-raising. Gajewski himself handles neg-raising as a semantic infer ence, based on a cancellable pr esupposition of the corresponding verb.

89

(b)

“not V that p” implies “not V whether p” (™Vp º ™V ?p)

In par ticular, both infer ences appear to be satisfi ed with a v erb like know: (11) Pierre knows that it is raining º Pierre knows whether it is raining (12) Pierre does not kno w that it is raining º Pierre does not kno w whether it is raining Zuber’s argument pr oceeds b y reductio to sho w that if believe whether was a grammatical construction, this would confl ict with the neg-raising behavior of believe, based on the assumptions made about the behavior of whether-complements. Thus one would have: (1) (2) (3) (4) (5)

™Bp º
B™p B™p º B?p ™Bp º B?p ™Bp º ™B?p ™Bp º
B?p š™B?p

(neg-raising) (from (a)) (from (1) and (2)) (from (b)) (from (3) and (4))

The conclusion is that if “believe whether p” was grammatical “Pierre does not believe whether p” would be impossible to utter without contradiction. Granting Zuber’s premises, his account would explain why other negraising verbs, like want for instance, do not takewhether-complements. On the other hand, we may note that his theory leaves unexplained why a verb like regret, which is not neg-raising, does not take whether-complements. This is not b y itself an objection, of course, since for such a v erb other mechanisms could v ery well come into play to explain the ex clusion of whether-complements. More problematically, as pointed out by B. Spector (p.c.), the account does not explain why a evrb close in meaning tobelieve, like be convinced, cannot take whether-complements. The problem in that case is that a predicate like be convinced is not neg-raising: (13) John is not convinced that it is raining i ( John is convinced that it is not raining) (14) *John is convinced whether it is raining

90

Furthermore, the account does not make explicit on what basis one should expect whether-complements to satisfy properties (a) and (b) when embedded under any v erb V. In particular, Zuber does not specify in his note what the meaning of whether-complements should be in general, nor the kind of implications that (a) and (b) correspond to. Clearly, however, the inference patterns (a) and (b) cannot both be simple semantic entailments, for this would predict that “V that p” and “V whether p” are semantically equivalent in general, which is not the case. One possible way of deriving Zuber’s constraints, which anticipates the proposal we shall make more explicit in Section 3.2 below, might be the following. Let us assume that “whether p” denotes the true answer to the question whether p (an assumption made b y Karttunen 1977 as w ell as Groenendijk and Stokhof 1982), so that for any V, Vx?p is true in i if x is in the relation denoted by V to the true answer to ?p in i. Assume that V is factive in the presuppositional sense, namely such that Vx p is defined in i only if p is true in i. Then one can verify that V will satisfy both properties (a) and (b). As for (a), if Vx p is true in i, then p(i)=1, by factivity, and so x is indeed in the relation to the true answer to the question whether p in i, that is Vx?p holds in i. As regards (b), assume ™Vx p is true in i. Since V is factive, this presupposes that p(i)=1, so x fails to be in the relation denoted by V to the true answer to whether p, and therefore ™Vx?p holds too. Thus, if V is assumed to be factive, the constraints (a) and (b) are met. As observed by Kiparsky and Kiparsky (1970), fur thermore, factivity (in the sense of pr esupposing the tr uth of the complement) appears to be incompatible with neg-raising: if ™Vp presupposes that p is tr ue, then ™Vp cannot be str engthened into V™p, for this would imply that ™p is true (see Gajewski 2005). What Zuber’s theory ends up sho wing, under this interpretation of the constraints (a) and (b), is that if a neg-raising verb like believe were to pattern like a factiv e verb, then this would giv e rise to a contradiction. At this point, however, one should ask whether the patterns (a) and (b) are always met with verbs that take both that-complements and whethercomplements, and whether they only appear with factiv e v erbs. Prima facie, it seems the patterns can be observed with a non-veridical questionembedding verb like say. Consider for instance: (15) John said that Mary had left (16) John did not say that Mary had left

91

A sentence like (15) cannot easily be followed by: “but John did not really say whether M ary had left ”. It can be follo wed, however, by: “but J ohn did not say the tr uth about whether M ary had left ”. If so, this suggests that the (a) constraint can be met even for non-veridical verbs, a point to which we will return in Section 4. A sentence like (16), b y contrast, is compatible with the continuation: “John said that M ary did not leav e”. This is enough to sho w that “John did not say that M ary had left” does not entail “J ohn did not say whether M ary had left ”. However, it may be the case that b y default a sentence like (16) implicates: “J ohn did not say that M ary had not left”. O ne way to deriv e the implicatur e is to obser ve that “J ohn said that Mary had not left ” is mor e informative than the actual utterance: “John did not say that Mary had left” (assuming John does not contradict himself ). Assuming a G ricean reasoning where the two sentences would stand as alternativ es, an informed speaker who utters the latter implicates the negation of the former sentence, namely “John did not say that Mary had not left”. The conjunction of the two sentences, finally, namely “John did not say that M ary had left and J ohn did not say that M ary had not left”, seems equivalent to: “John did not say whether M ary had left”. Another way of deriving Zuber’s constraints, consequently, would be to assume that V ?p means Vp ›V™p. On that interpretation of the meaning of whether-complements, the constraint (a) is automatically met, since Vp trivially implies Vp ›V™p.4 But then, the entire weight of the explanation rests on condition (b), which, as w e see her e, turns out to be an implicature for a non-veridical verb like say. Under either of the proposed derivations of Zuber’s constraints, we conclude that the inferential status of the constraint (b) above is not clear. As such, therefore, the account is not empirically adequate, nor suffi ciently grounded. 4. An alternative account of the infelicity of “believ e whether p” sentences, also based on neg-raising, would be to assume, as in Gajewski 2005, that neg-raising originates from the idea that believe presupposes opinionatedness on the par t of the attitude-holder . Thus, “John does not believe that p” should presuppose that either John believes p or that he believes not p; consequently, this is taken to imply that he believ es ™p. Assuming a w eak semantics of “ whether p” whereby “V ?p” would mean: “ Vp ›V™p”, then “believe whether p” would end up making a vacuous assertion, since it would only asser t what the use of believe presupposes. As pointed out to me b y B. S pector, however, again an account like this one does not generaliz e in any obvious way to be convinced whether, since be convinced does not satisfy the presupposition that the subject is convinced that p, or convinced that ™p.

92

2.2 Facts and propositions (Ginzburg 1995) The second account of question-embedding we shall examine in this section, Ginzburg’s 1995 account, r ests on the assumption that the cr ucial feature of factive verbs is that they express a relation to a fact, and that the notion of fact ought to be taken as a primitiv e. According to a tradition that goes back to R ussell, the semantic distinction between a v erb like know and a v erb like believe has to do with the fact that knowledge expresses a relation between an agent and a fact, while believe only expr esses a r elation between an agent and a pr oposition. Thus, Russell (1918, 93) writes: “I am inclined to think that perception, as opposed to belief , does go straight to the fact and not thr ough the proposition. When you perceive the fact you do not, of course, hav e error coming in, because the moment it is a fact that is y our object error is excluded … Therefore, the logical form of per ception will be diff erent from the logical form of believing, just because of that circumstance that it is a fact that comes in ”. While Russell talks only about per ception in this quote, his account can readily be extended to knowledge. The central aspect of the opposition between facts and true propositions, finally, is that facts are conceived as irr educible to tr ue propositions (see Kratz er 2002 and Holton 2006 for more about facts). Following R ussell, Vendler (1972, 105) makes the claim that thatclauses are ambiguous, depending on the kind of v erb under which they appear.5 Vendler’s thesis that belief and kno wledge are attitudes to ward distinct types of entities is based on several linguistic contrasts including, in particular, the evidence thatbelieve, unlike know, does not select questions as its complements. Vendler points out two other sets of contrastive data, which are worth rehearsing here. The first of these, originally put forward by 5. The idea that that-clauses are ambiguous is also defended by Kiparsky and Kiparsky 1970, who contrast “regret that” and “believe that” sentences. They notice that “I regret that John is ill” licences “I regret the fact that John is ill”, while “I believe that John is ill” cannot be expressed as “*I believe the fact that John is ill”. According to them, “Simple that-clauses are ambiguous, and constitute the point of overlap (neutralization) of the factive and non-factive paradigm” (1970, 356). Surprisingly, however, Kiparsky and Kiparsky do not compare “know” and “believe”. For although know is factive, “I know the fact that it is raining ” is not quite felicitous to mean “I know that it is raining ”. Another diff erence between that-complements after regret and after believe or know concerns the fact that the complementizer “that” cannot be omitted after regret. Thus: “*John regrets it is raining ” is incorr ect, as opposed to “J ohn knows/believes it is raining” (I am indebted to Lea N ash for this obser vation). For all these reasons, the contrasts that Kiparsky and Kiparsky point out betw een regret and believe are not necessarily r elevant when one compares know and believe.

93

Austin (1954), concerns the opposition between why- and how-questions. Questions concerning the justification of belief are typically why-questions, while questions concerning the justifi cation of kno wledge are typically how-questions: (17) a. b. (18) a. b.

Why do you believe that it is raining? *How do you believe that it is raining? *Why do you know that it is raining? How do you know that it is raining?

The second contrast concerns the behavior ofknow and believe with respect to what-clauses (to the extent that they are questions at all in the relevant contexts). Thus, it is possible to say: (19) Mary knows what Peter said (20) Mary believes what Peter said If Peter said “it is raining”, then from (20) one can only infer that Mary believes that it is raining. From (19), however, it does not necessarily follow that Mary knows that it is raining (she may think P eter said something false). More precisely, what (19) typically means is: “Mary knows that Peter said that it is raining”. No such interpretation is available with believe in (20), however.6 None of these contrasts, w e should note, provides clear linguistic evidence for the idea that that-clauses are ambiguous. G inzburg, however, adds to Vendler’s data some tests inv olving the nominalization of thatclauses. Thus, from “Peter believes that it is raining”, one can infer “Peter believes a cer tain hypothesis”. By contrast, fr om “Peter knows that it is raining”, one cannot infer in the same way: “P eter knows a certain hypothesis”. This suggests that the complement of know and the complement of believe fall in distinct categories. Vendler’s claim that that-clauses are ambiguous plays a central r ole in Ginzburg’s 1995 theory of question-embedding. Thus, Ginzburg proposes a classifi cation of attitude v erbs into thr ee main classes, depending on the kind of objects denoted b y their complements. In Ginzburg’s formal 6. In (20), the what-clause is obviously a free relative, and not an interrogative complement. In (19), the what-clause can be both a fr ee relative, or a question, namely “ What did P eter say?”. If questions are prohibited after believe, then the contrast can easily be accounted for . I am indebted to P. Schlenker for this observation.

94

ontology, questions, pr opositions and facts corr espond to thr ee distinct kinds of entities. Accordingly, attitude verbs are divided into three main categories, namely resolutive v erbs (such as know, and also tell), which can take facts in their denotation and licence both that- and wh-complements; interrogative verbs (such as wonder), which take questions in their denotation, and only allow wh-complements; and truth-falsity verbs, namely verbs of asser tion and opinion such as believe, which only allo w that-complements. On Ginzburg’s theory, an attitude v erb taking that-complements can take interrogative complements if and only if it expr esses a relation to a fact. In order to explain why believe cannot take questions, Ginzburg postulates that a question can be coerced to a fact, but can never be coerced to a proposition. The technical notion of coercion, introduced by J. Pustejovsky, can be grasped informally in the following way: to say that a question can be coerced into a fact means that the question-denoting complement can be expressed in terms of a fact-denoting complement. 7 For instance, to know whether Mary left is to kno w the fact that M ary left if she did, or to know the fact that Mary did not leave if she did not. The class of resolutive verbs, finally, is divided into two classes, namely resolutive factive verbs, like know, and resolutive non-factive verbs, like tell. An important point about tell is that tell can embed questions, despite the fact that it is not factive with respect to that-complements. On Ginzburg’s account, however, tell does expr ess a r elation to a fact whenev er it also expresses a relation to a question. I n that, Ginzburg follows the remarks of Baker (1968), who pointed out that “John told Mary who left” is typically judged false in a situation in which J ohn made a mistake or gav e some false information about who left. Thus resolutive non-factive verbs are ambiguous, since unlike believe, they can express relations to propositions or to facts. In comparison to the account discussed in the previous section, Ginzburg’s account makes a clear connection betw een factivity and questionembedding. It is unclear , however, to what extent the account is r eally explanatory, since Ginzburg only stipulates that questions can be coerced into facts, and cannot be coer ced into pr opositions. Lahiri points out two further limitations of the account (Lahiri 2002, 290–91). The first concerns the case of regret. Using one of Ginzburg’s nominalization tests, 7. See Pustejovsky 1993, 83, who defi nes type coercion as “a semantic operation that converts an argument to the type which is expected by a function, where it would otherwise result in a type error”.

95

it seems one can infer “Peter regrets the fact that it is raining” from “Peter regrets that it is raining ”. Prima facie, this suggests that regret denotes a relation to facts, y et regret does not embed questions. Another objection concerns the behavior of pr edicates like agree (on), which are not factive with respect to that-complements, and which are not veridical either with respect to interrogative complements. For instance: (21) a. Jack and Mary agree that Peter left b. Jack and Mary agree on whether someone left For (21)-a to be tr ue, Jack and M ary need only agr ee on the tr uth of a proposition (the pr oposition that P eter left), which may per fectly turn out to be false. Clearly, however, (21)-a entails (21)-b, and on Ginzburg’s account the complement of the preposition “on”, namely “whether someone left” denotes a question (Lahiri 2002, 290). As such, therefore, a verb like agree threatens the generalization whereby questions cannot be coerced into propositions. 2.3 A pragmatic account (Saebø 2007) In a recent paper Saebø presents a pragmatic account of the ex clusion of whether-complements after believe. The presupposition of factivity associated to know also plays a central r ole in the explanation, but the main aspect of Saebø’s account is the idea that know and believe are competing alternatives, and that a sentence like “Peter believes whether it is raining”, if it could be uttered, should be systematically dispreferred to a sentence like “Peter knows whether it is raining”. Saebø’s account is presented in the framework of bidirectional optimality theory, but makes use of general principles of the theor y of implicatures and presuppositions. More precisely, Saebø’s theory is based on an account of the competition betw een believe and know in fr ont of that-complements, which he generaliz es to the case of whether-complements. To use an example from Sauerland (2007), from a sentence like: (22) John believes that 313 is a prime. one typically infers that the speaker does not believ e that 313 is a prime. One way to account for this inference is to notice that if “313 is a prime” was believed by the speaker, it would be mor e appropriate for her to say 96

“John knows that 313 is a prime”. Saebø presents the inference as based on the Gricean Maxim of Quantity (“make your contribution as informative as possible”). As Sauerland (2007, 7) argues, for this and related phenomena, it is even more adequate to talk about a principle of maximization of presuppositions, as fi rst formulated by Heim 1991, namely “M ake your contribution presuppose as much as possible!”. Indeed, in a case in which the speaker takes it to be part of the common ground that 313 is a prime, uttering a sentence like (22) would be misleading. In Saebø’s framework, the sentence “J ohn believes that I” is like wise predicted to be less optimal than “John knows that I” in all contexts in which I follows from the common ground, namely in all contexts in which the pr esupposition of know is satisfi ed. Conversely “John believes that I” will be more optimal than “John knows that I” in all contexts in which I does not follow from the common gr ound, since the pr esupposition is not satisfi ed for know. Thus, Saebø predicts that any sentence of the form “J ohn believes that p” will implicate that p does not follo w from the common gr ound. For future reference I call this implicature the “BT-implicature” (for “Believe That ”): BT-implica tur e: a sentence of the form “John believes that I” implicates that the that-clause I does not follow from the common ground. In the case of whether-complements, Saebø postulates that a similar implicature should hold in principle. His account goes on to show that if a sentence of the form “J ohn believes whether I” could be uttered, then it should implicate in the same way that the presupposition associated to “John knows whether I” does not hold. On his theory, however, the presupposition of “John knows whether I” turns out to be a tautology, and holds vacuously. From this, it follo ws that a sentence like “J ohn believes whether 313 is a prime” would implicate a contradiction, which explains why the sentence is blocked. The way S aebø deriv es this r esult is b y assuming that any sentence of the form “J ohn knows \” (where \ is either of the form “ that I” or “whether I”) is defined only if the proposition Oi\(i) (which he calls the rectified complement proposition or RCP) follows from the common ground. When \ is a that-clause, namely of the form \= Ow(I(w)= 1), then the RCP Oi\(i) reduces to Oi(I(i)= 1), which corresponds to the denotation of the that-clause. When \ is a whether-clause, taken to be of the form Ow(I(w)= I(i)) (Groenendijk & Stokhof 1982), the RCP Oi\(i) reduces 97

to Oi(I(i)= I(i)), namely to a tautology . The BT-implicature associated to believe-that sentences is then generaliz ed by Saebø into a generaliz ed implicature for believe-that and believe-whether sentences alike (w e shall call it the BRCP-implicature): BRCP-implica tur e: a sentence of the form “John believes \” implicates that the RCP Oi\(i) does not follow from the common ground. From this assumption it follows that when \ is a that-clause, what is implicated is that \ does not follo w from the common gr ound, which yields the BT-implicature. When \ is a whether-clause, on the other hand, what is implicated is a contradiction, since any tautology necessarily follo ws from the common ground, and this should explain why “believe whether” is not grammatical. As we can see, Saebø’s account of the infelicity ofwhether-complements after believe does not depend dir ectly on its non-factiv e behavior, but more centrally, on the competition with know. The factivity of know, on the other hand, is needed to explain the BT-implicature of believe, which Saebø generalizes to the case of whether-complements. One possible objection concerns the or der of explanation in S aebø’s account. In the case of the B T-implicature, one deriv es the existence of this implicature, based on the fact that a sentence of the form “X believes that p” was actually utter ed, and on the principle of maximization of presuppositions. In other words, an implicature is derived from the competition between two sentences that can both actually be uttered. For the generalization to whether-complements, however, there is no evidence that any implicature should take place in principle, other than by analogy with the case of that-complements. A second issue about Saebø’s account concerns the way in which it can be generalized. For the case of a verb like surprise, in particular, Saebø suggests that the exclusion of whether-complements is based on a competition between that-clauses and whether-clauses (rather than between alternative matrix verbs), in relation to the factivity of the verb. If we consider a verb like regret (which is closely r elated to emotiv es like “be surprised ”, “be sorry” or “be sad”), Saebø’s account should predict that “John regrets that I” is strongly optimal over “John regrets whether I” whenever I follows from the common ground, due to the factivity of regret. In the same way, his account will predict that “John knows that I” is strongly optimal over “John knows whether I” whenever I follows from the common ground. 98

This prediction is correct, since for instance “John knows that Mary left” is clearly more informative that “John knows whether Mary left” in a situation in which it is part of the common ground that Mary left. But then, how can it be explained that “know whether” is nevertheless a grammatical form, while “regret whether” is systematically excluded? To account for the diff erence, Saebø observes that the use of a pr edicate like surprise generally bears a str onger factivity pr esupposition than the use of know in front of wh-complements. One can say “J ohn knows who left” without kno wing who left (“J ohn knows who left, let us ask him”). By contrast, according to Saebø, “John was surprised by who left” tends to convey that the speaker kno ws who left. Saebø calls this feature super-factivity. Independently, Guerzoni (2007) also makes use of a elated r notion of speaker-factivity, on the basis of similar examples. The difference between factivity and super-factivity is in fact cr ucial to explain why “be surprised whether” is less optimal that “be surprised that”, since according to Saebø the only relevant contexts that would occur for the evaluation of “Mary is surprised whether it is raining”, for instance, should be contexts in which either the speaker knows that it is raining, or contexts in which the speaker knows that it is not raining (by speaker-factivity). By application of the principle of maximization of pr esuppositions, the speaker should therefore prefer to use “M ary is surprised that it is raining ” or “M ary is surprised that it is not raining ” each time. The situation is supposed to be different with “John knows whether it is raining”, since that sentence becomes optimal in a situation in which the speaker herself does not kno w which of “it is raining” or “it is not raining” is true (in that case, the use of the that-complement is simply impossible). Despite this and notwithstanding the convergence between Saebø’s and Guerzoni’s intuitions, the evidence of super-factivity is not entirely clear. For instance, it seems one can say: (23) I met John this morning. He was very surprised by who had failed the exam in his class. I did not dare ask him which students had failed, but he seemed to be really disappointed. Clearly, the sentence “I did not dare ask which students had failed” conveys that the speaker does not know who failed the exam. This casts doubt on the robustness of the notion of super-factivity . As a r esult, the explanation of the exclusion of whether-complements after emotive verbs like be surprised or regret remains inconclusive. If regret is simply factive, and not 99

super-factive, then some other mechanism must be used to explain the difference with know with regard to whether-complements. 2.4 Summary Saebø’s theory of the exclusion of whether-complements is explicitly presented as pragmatic, rather than syntactic or semantic. As w e saw, his account appeals to two kinds of competition (betw een complementizers on the one hand, and between matrix verbs on the other hand, depending on the embedding verbs). This, however, may also be an issue, if one can give evidence that the exclusion of whether-complements after believe and regret might be based on one uniform mechanism, rather than on diff erent mechanisms. While his inspiration is more directly semantic, we saw that Zuber does not account for the ex clusion of whether-complements after regret, which is not neg-raising. Like wise, Ginzburg’s account does not make clear why the complements of regret, if they ar e facts, do not put regret on a par with know with respect to questions. All the accounts we reviewed, moreover, propose to explain why whether-complements are excluded in specifi c contexts, but they do not pr ovide a positive account of the cases in which whether-complements are licenced by verbs that take that-complements. In the next section, I pr opose to see ho w Hintikka’s factivity hypothesis may be justified to account both for the admission and for the exclusion of whether-complements. The way we will formulate the hypothesis will end up putting believe and regret in the same category. 3. Veridicality and question-embedding In this section I present one way of deriving Hintikka’s factivity hypothesis.8 More exactly, I examine one dir ection of the hypothesis, accor ding 8. I should note her e that the account I pr opose is most likely not exactly what H intikka had in mind, nor does it purport to be. Hintikka (1975, 21) explicitly comments only onknowing who constructions, as in “D r Welby knows who has a pneumonia ”, which he analyz es as: x(Px o Dr Welby kno ws that Px). He writes that “ on the univ ersal-quantifier r eading of subordinate wh-questions the main v erb … receives a kind of success for ce, independently of whether it has one in the that-construction.” His theory is that only verbs that already have this success force with that-complements can therefore enter in a wh-construction. Berman (1991) gives a more elaborate view of this account in terms of presupposition accommodation. I leave a discussion of B erman’s proposal aside (see Lahiri 2002). To be completely fair to H intikka’s account, finally, I should note that his claim is that only factive verbs can embed questions, and

100

to which all veridical attitude v erbs are able to embed whether-complements. First, we need to say mor e about the diff erence between factivity and veridicality. 3.1 Veridicality and factivity We shall call a v erb that takes that-clauses veridical if it entails the tr uth of its complement when used in the positiv e declarative form, namely if it satisfies the schema Vp op for all p, where p is a that-clause. Following Kiparsky and Kiparsky (1970), w e shall call a v erb V factive if asser ting Vp presupposes the tr uth of the complement p. A v erb like know is both veridical and factive, since “John knows that it is raining” implies that it is raining, and “John does not kno w that it is raining ” also commits the speaker to the belief that it is raining.9 Some predicates like “it is clear”, or “prove”, appear to be veridical and yet not factive. Although the positive form of the v erb entails the tr uth of the complement, this entailment is not preserved under negation, in particular: (24) John has proved that Mary was not there that night [# but in fact she was there] (25) John has not proved that Mary was not there that night [and in fact she may have been there] (26) It is clear that J ohn is the one who did it [# but J ohn may not have done it] (27) It is not clear that John is the one who did it [and in fact he may not have done it] An important issue concerns the r elation between the two notions of veridicality and factivity . O ne possibility is to see factivity as implying veridicality. Another is to see the two notions as independent in principle, albeit coinstantiated in some v erbs. The choice betw een the two options matters, since it also depends on how one thinks factivity presuppositions are generated. Usually, a factive predicate is considered ipso facto veridical, since factivity is defi ned as pr eservation of the tr uth of the complement does not explicitly say that all factive verbs should. Hence only verbs like tell, but not regret, are a putative counterexample to his account. 9. Of course, it is always possible to say: “J ohn can’t know that it is raining, since it is not raining”, but this is exactly a phenomenon of cancellation of the pr esupposition (also kno wn as local accommodation).

101

in specifi c contexts (under negation, modals, and so on), but taking the positive context as the baseline. In their grammar of the E nglish Language, on the other hand, H uddleston and Pullum (2002) go for the second option. They call entailing the predicates we here call veridical, and consider that some v erbs, like regret, admit, and resent, are factive but non-entailing. Usually, a verb like regret is presented as a paradigm case of factive verb, since from “John regrets that Mary left him”, one infers that the speaker believ es that M ary left J ohn, and similarly for “J ohn does not r egret that M ary left him” (Kiparsky & Kiparsky 1970, Karttunen 1971). Several examples, first put forward by E. Klein (1975), challenge the idea that regret (and other emotive predicates like “be glad”) is really veridical: (28) Falsely believing that he had infl icted a fatal wound, O edipus regretted killing the stranger on the road to Thebes (Klein 1975, quoted in Gazdar 1979, 122) (29) Ed believ ed that he had off ended his par ents and v ery much regretted that he had done so, but it turned out that he had been mistaken: they had not in the least been off ended. (Huddleston and Pullum 2002, 1007) (30) John wrongly believes that Mary got married, and he regrets that she is no longer unmarried. (Egré 2004, 2005, based on Schlenker 2005, fn. 14) Interestingly, these examples bear some similarity to examples put forward by Iatridou (2000) inv olving the v erb wish. Iatridou observes that “X wishes she were p” generally implies that “X is p” is not true. Iatridou argues that this element of counterfactuality is only an implicature, however, based on the following kind of example: (31) In the movie True Lies, Jamie Lee Curtis wishes she were married to an exciting person and she is. The behavior of wish followed by the subjunctive mood may be compared to that of regret in the same sentence, considering that regretting p implies wishing not p: (32) In the mo vie True Lies, Jamie Lee C urtis regrets that she is not married to an exciting person, but in fact she is. 102

In the same way in which Iatridou’s examples suggest that counterfactuality with wish and the subjunctiv e is only an implicatur e, conversely the example may suggest that the factivity of regret is also a cancellable implicature, rather than a semantic presupposition of the verb. One issue raised by such examples, however, is whether the use ofregret could possibly be a scare-quote use, as in a sentence like “Before Copernicus, people knew that the sun was revolving around the earth”, in which the verb “knew” is usually giv en a special intonation indicating pr etense on the part of the speaker.10 According to Gazdar (1979), examples similar to (28) are found with cognitive factive verbs, for instance: (33) Falsely believing that he had infl icted a fatal wound, O edipus became aware that he was a murderer. A sentence like (33) appears to be a case of what Holton (1997, 626) calls protagonist projection: in this par ticular case O edipus becomes awar e of a property that he is self-ascribing, using the wor ds “I am a mur derer”. Gazdar concludes fr om the parallel with (28), in par ticular, that Klein ’s examples do not pr ovide “sufficient grounds for arguing that some or all factive verbs do not entail their complements” (1979, 123). Likewise, one may argue that sentences (28)–(30) also inv olve the same mechanism of protagonist projection, by expressing in the that-clause the content of what the attitude holder would express directly by saying, for instance: “I regret that Mary got married”. One diff erence which r emains, ho wever, is that the substitution of “know” for “regret” in a sentence like (30) produces a contradiction. Thus, it seems just inconsistent to utter: (34) (??) John wrongly believes that Mary got married and he kno ws that she is no longer unmarried. This suggests, at any rate, thatknow and regret do not behave in exactly parallel ways. In (30), it seems one can ascribe to John a true mental state of regret. In the case of (34), one cannot ascribe tr ue knowledge in the same way . Further elements suggest that regret and other emotive predicates, like be sad or even be surprised, may be closer to believe than to know (Schlenker 2005, E gré 2004, 2005). O ne indication concerns the mood of the 10. This suggestion was made to me by T. Williamson in particular.

103

embedded complement in French. In French, emotive predicates like être triste (be sad) and être surpris (be surprised ) require the v erb of the thatclause to be in the subjunctive, like regretter (regret), in contradistinction to savoir (know), which selects for the indicative:11 (35) a. P b. P (36) a. P b. P

Pierre est surpris que Marie soit partie sans dire au revoir ierre is surprised that Marie isSUBJ left without saying good bye. *Pierre est surpris que Marie est partie sans dire au revoir. ierre is surprised that Marie isIND left without saying good bye. Pierre sait que Marie est partie sans dire au revoir ierre knows that Marie isIND left without saying good bye. *Pierre sait que Marie soit partie sans dire au revoir. ierre knows that Marie isSUBJ left without saying good bye.

Contrary to (36)-a, a sentence like (35)-a does not necessarily imply that Pierre knows that Marie left without saying good bye, not even that the speaker assumes that M arie left without saying good b ye. What the sentence presupposes is that it seems (at least to some par ticipants in the conversation) that M arie left without saying good b ye. The best way to approach the corr ect meaning would be to say: “P ierre is surprised that Mary might have left without saying good bye”. Thus (35)-a can very well be pursed with: (37) Moi aussi cela me surprend. A mon avis, elle a dû lui dir e au revoir, mais il ne s’en souvient pas. It surprises me too . In my opinion, she must hav e said good b ye to him, but he does not remember.

11. Mood plays a role more generally in the triggering of factivity. For instance, in French, the verb “se souvenir” (remember) is veridical in positive declarative contexts, and r equires the indicative, as in “J ean se souvient que M arie est par tie” (Jean remembers that M arie is IND left), which entails that Marie left. Quite strikingly, however, the use of the subjunctive mood under negation suspends factivity, while it is pr eserved with the indicativ e (see E gré 2004 and E gré 2005). Thus “Jean ne se souvient pas que M arie est partie” (Jean does not remember that Marie isIND left) implies that M arie left. B y contrast, “J ean ne se souvient pas que M arie soit par tie” (Jean does not remember that Marie isSUBJ left) does not entail that Mary left. Interestingly, “savoir” (know) does not allow for the subjunctive under negation, except in the first singular person, in which case the v erb savoir too is put in the subjunctiv e, as in: “Je ne sache pas que M arie soit partie” (I don’t knowSUBJ that Marie isSUBJ left). In that case, the sentence means something close to: “I don’t have any evidence that Marie would have left”.

104

In what follows we shall therefore consider that regret, like other emotive predicates, and despite appearances to the contrar y, is not v eridical, unlike know.12 On the present account, x regrets p presupposes neither p, nor x knows p, but only x believes p (Huddleston & Pullum 2002, Schlenker 2005, Egré 2004). Following Huddleston and Pullum’s grammar, one may still wish to call regret a factive verb, based on the consideration that the default interpretation of a sentence of the form x“ regrets that p”, in most contexts, including in questions and under negation, is that p is true. To say this, however, is to commit oneself to a pragmatic understanding of the notion of factivity , seen as an implicatur e or as a pragmatic pr esupposition, and to giv e up the semantic account of presupposition whereby factivity is conceived as veridical entailment preserved in different contexts.13 In my opinion, it may be pr eferable to maintain that factivity does imply veridicality, and to conclude that because regret is not v eridical, it is not factiv e tout cour t. Whichever vie w one takes of the link betw een veridicality and factivity, however, what we owe is an explanation of why, if regret is indeed not veridical, it appears to pattern like a factiv e verb in most contexts. That explanation will most likely involve some pragmatic mechanism. In particular, if “x regrets p” only presupposes “x believes p”, what we need to account for is how we get from the semantic presupposition “x believes p” to the conclusion “the speaker believes p”. Huddleston and Pullum make a suggestion about this problem, by saying that “just as one cannot regret some proposition p unless one believes that p is true, so one would not normally ask whether someone else egrets r that p unless one believes that p is true” (2002, 1008). While this intuition seems plausible, it seems clear that some principle about pr esupposition accommodation is needed more generally to account for this shift of perspective. 12. Another observation on the contrast between know and regret can be seen by applying Vendler’s tests. The why- vs. how-questions suggests that regret behaves more clearly like believe than know in “Why do you regret that Mary left?”, which is corr ect, as opposed to “*H ow do you regret that Mary left?” Using Vendler’s own standards, this should mean that the objects of regret are more likely to be propositions than facts. 13. See Gazdar (1979, 120), who describes Karttunen’s position along the same lines: “some writers, such as Karttunen, who have embraced a pragmatic definition of presupposition, seem implicitly to hav e abandoned the entailment r elation between affi rmative sentences and their presuppositions”. Gazdar adds in a footnote (fn. 12) that “In fact, Karttunen (p.c. 1977) thinks that cognitive factives like know do entail their complements, but he is agnostic with respect to emotive factives like regret”. On the r elation between entailment and pr esupposition, see also Stalnaker (1974).

105

Interestingly, this pr oblem has a connection to what is kno wn as the proviso pr oblem for pr esuppositions, originally put for ward b y G eurts (1996). The proviso problem is that on standard accounts of presupposition projection, a conditional sentence of the form “if Theo likes sonnets, so does his wife” is predicted to semantically presuppose that if Theo likes sonnets, Theo has a wife . However, the pr esupposition accommodated in “out of the blue” contexts is the unconditional pr esupposition that Theo has a wife. Heim (1992, 206) discusses a similar accommodation problem concerning sentences like: “John believes that it stopped raining”, where we standardly accommodate “it had rained ”, rather than the pr edicted presupposition: “John believes that it had rained ”. The problem in this case is to fi gure out ho w we are able to infer: “it had rained ”, from the semantic presupposition “John believes it had rained”. As such the problem is even more similar to what happens with “John regrets that it is raining”, assuming that “John believes it is raining” is the semantic presupposition, and that “it is raining” is what we pragmatically accommodate. 14 I will set aside a further discussion of this issue here. What matters for the present account is the idea that veridicality may indeed correlate better with question-embedding than factivity. In particular, veridical non-factive predicates like “prove” and “it is clear” can embed whether-questions: (38) As yet, there is probably no evidence that would definitively prove whether or not some dinosaurs were warm-blooded. (39) It is not y et clear whether such phenomena can be obser ved in patients with non-tumor diseases. We may note that for such v erbs, not all contexts allo w for felicitous whether-embedding. For instance, in a context in which the speaker kno ws what John has proved about dinosaurs, it is obviously odd to say: (40) John has proved whether dinosaurs were warm blooded. Rather, one would expect the speaker to use “prove that”, followed by the appropriate answer to the question. Like wise, in a context in which the speaker knows whether Mary left or not, it sounds incorrect to say: 14. I am indebted to P. Schlenker, B. Spector and K. von Fintel for independently pointing out to me the connection with Heim’s problem. I am further indebted to R. Singh and D. Fox for pointing out the analogy with the proviso problem more generally. See von Fintel (2006) for an outline of the accommodation mechanism in the case of the pr oviso problem.

106

(41) It is clear whether Mary left. Here again, one would expect “it is clear that ” to be pr eferred by the speaker, unless the speaker wants to hide the information from the hearer, or signal to the hearer that he has an answer (as in: “it is completely clear to me whether M ary left, but I won ’t tell y ou, you need to fi nd out for yourself ”). For such cases, ther efore, the infelicity of using “ whether” is readily explained by the competition withthat-complements, exactly along the lines of Saebø’s pragmatic account. But it remains that such verbs can embed whether-complements in principle, as opposed to verbs like believe or regret, which seem to exclude them completely. 3.2 From veridicality to whether-embedding Assuming veridicality might be a necessar y and suffi cient condition for a verb that takes that-complements to take whether-complements, let us now try to derive this connection mor e formally. To do this, w e need to specify what the meaning ofwhether-complements is, and what we should expect the meaning of a sentence of the form “ x V whether p” to be in general. In principle the meaning of embedded questions could evry well depend on the embedding v erb. For instance, it is not ob vious that “whether p” makes the same contribution in “wonder whether p”, “decide whether p”, and “know whether p”, a point to which w e will return below. Here we shall assume that the contribution of “ whether p” is uniform, and that it denotes the tr ue answer to the question whether p. More precisely, we assume Groenendijk and Stokhof ’s denotation for questions (Groenendijk and Stokhof 1982, 1990). I n their frame work, “whether p” denotes the proposition that p if p is true, and the proposition that not p if p is false. More formally, awhether pb (i) =Ow(p(w) = 1) if p(i) = 1, and awhether pb (i) =Ow(p(w) = 0)if p(i) = 0,that is awhether pb (i) =Ow(p(w) =p(i)).15 The expression Ow(p(w) =p(i)) corresponds to what GS call theextension of the question whether p. On GS’s account, “whether p” can also denote the intension of the question, for instance after wonder, namely the function that, in each context, gives the true propositional answer to whether p in that context (OiOw(p(i) =p(w))). By contrast, “whether p” denotes the 15. W e write Ow(p(w) = 1) instead of Ow(a p b(w) = 1), and dr op double brackets to ease notation when it does not lead to confusion.

107

extension of the question after know, namely Ow(p(w) =p(i)), assuming i is the evaluation context. Thus, the diff erence between “whether p” after wonder and after know is fundamentally a type distinction, which leav es the meaning of the embedded question essentially inv ariant. O ur fi rst, assumption, therefore, is that, after a verb that takes both declarative and interrogative complements, whether-complements denote the true answer to the corresponding question: (42) Meaning of WHETHER: for a verb V that takes that-clauses, “x V whether p” is true in i iff the referent of x is in the relation denoted by V to the true answer to the question “?p” in i. Our next assumption about the meaning of “ whether”-clauses is that they systematically weaken the meaning of that-clauses when embedded after verbs that admit both kinds of complements. This corresponds to Zuber’s assumption (a), namely to the idea that all v erbs that take both that- and whether-clauses satisfy the entailment from “that p” to “whether p”. For instance, one can check that the inference goes through for know and decide: (43) a. John knows that Mary left b . John knows whether Mary left (44) a. John has decided that he will go to London b. John has decided whether he will go to London We call this second assumption the “that-to-whether” constraint: (45)

THAT-TO-WHETHER

c onstraint : a v erb V that takes declarativ e complements takes whether-complements provided it satisfies the entailment from “x V that p” to “x V whether p”.

Now, it is easy to see that if a erb v is veridical, then it systematically satisfies the that-to-whether constraint, given our assumption on the denotation of whether-complements. Indeed, by defi nition, if a v erb V is v eridical, then one has: (46) a. i[V(i)(a, Ow(p(w)=1))=1 o p(i) = 1] (veridicality), hence b. i[V(i)(a, Ow(p(w)=1))=1 o V(i)(a, Ow(p(w)=p(i)))=1]

108

In other words, whenever a veridical verb selects a that-clause, because the that-clause has to be tr ue, it thereby corresponds to the extension of the corresponding whether-question. For verbs that are not veridical, like believe or regret, on the other hand, the that-to-whether constraint cannot be satisfi ed in the same way . For instance, to believe that it is raining is to believe a proposition that could very well turn out to be false. Likewise, on the present account, to regret that it is raining does not by itself entail that it is raining. If x regrets that it is raining, x believes that it is raining and wishes it were not raining. But as we argued, x’s regret can very well rest on a false belief. 3.3 it is true that, be right that The constraint (45) is a stipulation of the theory, just like the assumption made above about the meaning of “ whether”. As stated, mor eover, (45) may appear as only a suffi ciency condition for question-embedding. F or the condition to be fully explanator y, we need to suppose that if a v erb does not automatically satisfy the entailment fr om “that” to “ whether”, then it does not licence whether-complements. B efore turning to this problem, what we need to examine is whether, as a sufficiency condition, the condition does not already overgenerate. Some problem cases for the account here are non-factive predicates like “it is true that” and “be right that”, which by defi nition entail the tr uth of their complement, and y et do not take whether-complements: (47) (48) (49) (50)

Mary is right that John left [# but John did not leave] *Mary is right whether John left It is true that John left [# but John did not leave] *It is true whether John left

One common aspect of these two pr edicates is that they are not attitude predicates per se. As Abusch (2002) observes about be right that, this predicate is in a sense symmetric toknow, since “Mary is right thatp” asserts that p, and presupposes that Mary believes that p. Hence, stricto sensu “Mary is right that p” does not assert that Mary believes that p, and to that extent it does not ascribe an attitude to Mary. By contrast, “Mary knows that p” asserts that Mary believes p, and presupposes that p. Similarly, “it is true that it is raining” asserts that it is raining, but does not presuppose it. The same holds of the predicate “it is right that”. 109

An important difference with predicates like “prove” or “be clear” is that whether-complements after prove or be clear, in the relevant contexts, ascribe a non-vacuous property to the true answer to the question. For instance: (51) We will soon be able to prove whether some dinosaurs were warmblooded (52) By the end of the semester, it will be clear whether the budget was sufficient asserts that the true answer to the question will soon be proved or become clear. By contrast, sentences like “it is right whether it is raining ”, or “it is true whether is is raining” would assert that the true answer to the question is true/right, both of which being vacuous statements.16 Another way to see it is to look at the negation of whether-sentences. “John does not kno w whether it is raining” says that the true answer to whether it is raining is not known to John, which is consistent without being trivial. B ut “it is not true/right whether it is raining” should mean that the true answer to the question is not true/right, which would then be contradictory. This suggests that the corr ect way of putting the that-to-whether constraint is to say that a pr edicate V that takes that-complements takes whether-complements, provided the inference is non-vacuous. Alternatively, one may state this as a separate constraint, namely the constraint whereby an expression can appear as argument of another pr ovided the r esulting expression is semantically non-vacuous.17 Finally, a predicate close to be right that is the predicate correctly believe that, which, b y vir tue of the adv erb, is also v eridical. In principle, our account should pr edict that such a pr edicate can take whether-complements, which is not the case: (53) Only 17 percent of Canadian adults correctly believe that/*whether approximately 45,000 Canadians die each year as a result of tobacco use. 16. Assume that aTrueb(i)=OPstT(i)(P), where T(i)(P)=1 iff P(i)=1. Then aTrue(whether p)b(i) =T(i )(Ow(p(w)=p(i ))=1, since Ow(p(w)=p(i))(i)=(p(i )=p(i))=1. 17. I here rely on unpublished r emarks by J. G ajewski, and also D. F ox. The view that a sentence will be ungrammatical if it pr oduces systematic contradictions and tautologies can be traced to Barwise & Cooper (1981), who use that idea to explain the infelicity of sentences like “there is every student in the garden”.

110

The embedding of “ whether” after “ correctly believ e” is just as bad as under “believ e”. In this case one cannot say that the embedding of “whether” would be vacuous. Rather, the badness of the sentence obviously originates from the impossibility of embeddingwhether-complements after believe. This means that the that-to-whether constraint stated above cannot be taken to apply to arbitrary predicates, but only to the head of the verb phrase containing the that-clause (“correctly” is only an adjunct). 3.4 Summary In this section we have given evidence in favor of the hypothesis that all veridical attitude verbs licence whether-complements. We have proposed to derive the generalization by assuming that whether-complements denote the corresponding tr ue answer, and b y obser ving that v eridical attitude verbs, because of their veridical character, in principle make it possible to express that-clauses as whether-clauses in the positive form. Moreover, we have argued that emotive predicates like regret or be surprised should not count as veridical, despite the fact that in many contexts the truth of their complements is pragmatically presupposed. So far, however, we have said nothing about the converse direction of the factivity hypothesis. That is, while the present account predicts that all veridical attitude verbs licence whether-complements, it leav es open the possibility for non-v eridical verbs also to take whether-complements. The next section is dev oted to this problem. 4. The case of non-veridical verbs We have formulated the that-to-whether constraint by saying that a v erb which takes that-complements takes whether-complements provided it satisfies the entailment from “V that p” to “V whether p”. For the constraint to be fully explanatory, however, the constraint should be read as “if and only if ”. If so, this would explain not only why v eridical verbs can take both that- and whether-complements, but also why non-v eridical verbs like believe and regret can not take whether-complements, since for such verbs the entailment is not satisfi ed. One way of defending the biconditional formulation of the constraint is to observe that in all contexts in which “V whether p” is true, if “V that p” is tr ue, p must be tr ue too (giv en our semantics for whether). This, however, falls shor t of establishing that in all contexts in which “ V that 111

p” is true, p should be true. In other words, it does not follow analytically that if a v erb taking whether-complements is extensional in G roenendijk and Stokhof ’s sense, then it is v eridical with respect to that-complements in all contexts. And indeed, counter examples are given by verbs like tell, say, which are not veridical with respect to that-complements, but which admit whether-complements. Other problem cases for the present account include verbs like guess, but also doubt, agree (on), conjecture (about) and predicates like be certain (about) or be sure (on), which can all be followed by whether-clauses. The first observation to make about such verbs, however, is that they fall in at least three different categories. As argued by Karttunen (1977b), whetherclauses after doubt are special and do not behave like questions, which sets doubt apart. Regarding the other verbs, we can see, following Lahiri (2002), that they fall into two distinct classes: most of the time, v erbs like tell, say and guess appear to be veridical with respect to questions. By contrast, verbs like agree (on), be certain (about) are non-veridical with respect to questions. I examine these diff erent verbs in turn. F or the latter v erbs, I argue that prepositions like about and on play a semantic role. Regarding the former, I argue that they are ambiguously veridical already with that-clauses. 4.1 D oubt As discussed b y Kar ttunen (1977b), whether-clauses after doubt behave idiosyncratically. In English the two sentences (54)-a and (54)-b can be used interchangeably: (54) a. I doubt that they serve breakfast. b. I doubt whether they serve breakfast. Karttunen also observes that doubt cannot take wh-questions (“*I doubt who is coming for dinner ”), and does not licence alternativ e readings of disjunctive whether-questions (“I doubt whether they serve tea or coffee” can only mean “I doubt whether they serve either tea or coffee”). Moreover, “whether” cannot be replaced by “whether or not” after doubt: (55) *I doubt whether or not they serve breakfast. Karttunen therefore concludes that whether-clauses after doubt do not denote real questions. I concur with Karttunen’s analysis here. One obser112

vation we may add to this is that in F rench, “douter si” is marginally acceptable and found in some dialects. I n that case, ho wever, the construction has a meaning close to “ se demander si” (wonder), and it does allo w for alternative readings and for “ whether or not ” constr uctions; moreover, “douter si” is followed by the indicative mood, unlike “douter que” (doubt that), which requires the subjunctive mood: (56) Les gens doutent si c’est la peste ou pas. (Google) People doubt whether it is the plague or not. (57) Nous voyons des gens qui doutent si l ’univers n’est point l ’effet du hasard ou d’une nécessité aveugle. (Google) We see people who doubt whether the univ erse is not the eff ect of chance or of blind necessity. (58) Les gens doutent que ce soit la peste / *que c ’est la peste People doubt that it be the plague / that it is the plague. These data strengthen Karttunen’s observation about English, since in French a sentence like “P ierre doute qu’il pleuve” (Pierre doubts that it is raining) cannot entail nor be expr essed by “Pierre doute s ’il pleut”. We therefore agree with Karttunen that doubt does not embed whether-questions, exactly like believe. 4.2 Tell, Guess and non-veridical readings While tell and agree both ar e non-veridical with r espect to that-complements, one can observe that after tell or guess, embedded questions usually have a veridical reading, which means that the interrogative complement seems to denote the tr ue answer to the question. F or that reason, Lahiri (2002) puts them in the class of what he callsveridical responsive predicates. For instance: (59) John told Mary who was at the party. seems to mean that John told Mary the true complete answer to the question “who was at the party” (Baker 1968, Karttunen 1977).18 In a situation 18. See Tsohatzidis (1993) for further references, in particular Vendler, and for a criticism

113

in which only S ue and B ill were at the par ty, and J ohn only told M ary that Sue was at the par ty, or told her mistakenly that Alfr ed was at the party, the sentence seems to be false. The same seems to hold of whetherquestions. The sentence: (60) John will tell us whether it is raining in London. conveys that J ohn is expected to tell the tr ue answ er to the question. Likewise, guess in most contexts appears to for ce the veridical reading of the embedded question: (61) John had guessed whether it would rain. In such examples, tell seems to behave like reveal, which is factive with respect to that-complements, and like wise guess gets the meaning of the factive divine (cf. the French “deviner”, which is factive with that-complements, and is one common way of translating “guess” in many contexts). Interestingly, Lewis (1982) obser ved that “this veridical sense [of telling whether] may or may not be the only sense of ‘tell whether’; it seems at least the most natural sense ”. At the same time, ho wever, the r emark indicates that Lewis had in mind the possibility of non-veridical readings of whether-complements after tell. E gré and S pector (2007) inv estigate the possibility that non-v eridical readings of embedded questions might indeed exist after tell and other communication verbs (such as predict, or announce), on the basis of examples like: (62) Everyday, the meteor ologists tell the population whether it will rain the next day, but they are often wrong. In such a case, it seems the meteorologists needn’t tell the actual answer for “tell whether” to count as tr ue.19 The hypothesis examined in E gré and Spector (2007) is that “tell whether p” could then mean: to tell some potential complete answer to the question whether p, namely to tell p or to tell ™p, rather than to tell the actual answer to the question. of the view that questions denote their true answers after tell, and Holton (1997) for a defense of the view that tell wh- constructions are veridical. 19. See earlier examples by Tsohatzidis (1993) on the idea that to tell wh- does not necessarily mean to tell the truth. On Tsohatzidis’ account, the veridicality of tell wh-constructions is treated as an implicature.

114

Further support for this idea also comes from the behavior of verbs of conjecture such as conjecture, guess, and predict:20 (63) Out of all of the plays run on aSturday, I incorrectly guessed whether it was going to be a run or pass on just a handful. (Google) (64) For each draw , I had to conjectur e whether an odd number or an even number would come out. All my guesses pr oved to be wrong. (65) In one study, 56% of r ecipients guessed incorr ectly about an email’s intended tone, while 78% of the senders incorr ectly predicted whether or not the recipients would get it right. (Google) In (63) the verb guess seems to have exactly the meaning of predict, and ends up meaning: “I made an incorr ect guess as to whether it was going to be a run or pass”. Sentence (64) suggests that to conjecture whether p would simply mean “ to conjecture one of p or ™p”, rather than “fi gure out the true answer”. Due to the presence of “incorrectly”, more generally, (63) and (65) seem to argue against B aker’s idea that tell and guess allow only for veridical readings of embedded questions. Such examples r emain compatible with sev eral interpretations, however. Le wis (1982), for instance, suggests that ther e may be a sense of “telling falsely whether ” that “ does not count as telling whether at all, but only as purpor ting to tell whether ”. Likewise, “conjecture whether” or “guess whether” might r eally mean “ try to fi nd out whether ”. This hypothesis seems adequate for a sentence like (62), but does not squar e well with an example like (63), ho wever, wher e the v erb “guessed” is modified by “incorrectly”. If guess meant “ try to make a corr ect guess”, then one might expect the adverb “incorrectly”, in “guessed incorrectly”, to modify “ try” semantically, which does not seem to get the corr ect meaning. Another hypothesis, which w e are about to see in the next section, is to view the non-veridical readings of questions after tell, guess and related verbs as parasitic on the veridical reading of the question, but as involving some coercion mechanism due to the presence of a preposition (possibly unexpressed). Such an account does not easily apply to tell or predict, however, for which no preposition seems ever to overtly show up.21 20. Sentences taken from Google were checked by native speakers. 21. For instance, “John will tell us about whether Mary left” sounds odd, or otherwise has a meaning much weaker than “John will tell us whether Mary left” (see below).

115

Setting aside a further discussion of these examples, I conclude that for the present account to be fully explanator y, we need to endorse the idea that tell is ambiguous betw een a factive and a non-factiv e interpretation already with that-clauses (see Egré and Spector 2007, and already Berman 1991, Ginzburg 1995). Furthermore, we are in principle committed to the view that questions after tell primarily have the veridical reading. 4.3 Agree, be certain and the preposition about By contrast to verbs like tell or guess, verbs like agree or be certain are not only non-veridical with respect to that-complements, but also with respect to embedded questions. For instance, one can say: (66) a. b. (67) a. b.

Mary and John agree on who left. Mary and John agree on whether Sue left. John is certain about who left. John is certain about whether Sue left.

In the first two examples, the object of Mary and John’s agreement need not be the true answer to the question. It is enough for John and Mary, for instance, to agree that Sue left in order to agree on whether Sue left, even if in fact, Sue did not leave. The case is the same for (67)-b: it is enough for John to be cer tain that Sue left in or der for him to be cer tain about whether Sue left. Consequently, not only do such verbs licence both thatand whether-complements, but they appear to satisfy the that-to-whether constraint stated in (45) above. These verbs are a problem for the account spelt out in Section 3, since they are not-veridical. From a syntactic point of view, however, these verbs clearly differ from the verbs we have discussed so far. For one thing, agree requires a preposition in front of embedded questions. In the case of be certain, the preposition can sometimes be dr opped, but nativ e speakers r eport that the sentences are not as good as when the preposition is expressed:22 (68) *John and Mary agree whether Sue left. (69) *John is certain whether Sue left.

22. The presence of a negation seems to help, as in “John is not certain whether Mary left”, “John is unsure whether Mary left”.

116

Another difference, observed by B. Spector, is that such verbs allow for the paraphrase “the question of whether ”. This is not so with know, for instance: (70) *John knows the question of whether Mary left. (71) John and Mary agree on the question of whether Mary left. (72) John is certain about the question of whether Sue left. Interestingly, tell patterns like know relative to this test. B y contrast, verbs like guess and conjecture allow for the pr esence of a pr eposition in contexts in which they giv e rise to an explicitly non-v eridical reading of the embedded question: (73) *John told Mary the question of whether Sue left. (74) For each draw, I had to conjecture about whether an odd number or an even number would come out. (75) You’ll never again have to guess about whether or not a plant needs water. (Google) In (74) and (75), the presence of the preposition “about” clearly forces the verbs to mean “ make a conjectur e” and “make a guess ” respectively. The interesting fact, in the light of our pr evious discussion, is that the sentences seem to r etain the same meaning when the the pr eposition is not expressed. This is not so with know or tell, for which “tell about” and “know about”, when they can be used, alter the meaning of the original sentence. For instance, the second sentence belo w conveys that John has some information relevant to answer the question, not necessarily that he knows the answer: (76) a. John knows whether Mary left. b. John knows about whether Mary left. Another v erb, which appears not to be v eridical, at least not in any obvious sense, is decide, which also embeds whether-complements and appears to satisfy the that-to-whether constraint: (77) a. John has decided that he will go to London. b. John has decided whether he will go to London.

117

While decide can hav e an epistemic r eading in some contexts, it does not appear to be veridical. For instance, one can say: (78) John had decided that he would go to London before Christmas. But he never went there. As a consequence, to decide whetherp seems to mean simply “to decide p or to decide ™p”, and not necessarily “to decide the true answer to the question whether p” (meaning the true answer relative to the actual world). For decide too, however, it is possible to add the preposition “about” without altering the meaning of the sentence: (79) John has decided about whether he will go to London. (80) John has decided about the question of whether he will go to London. One way of accounting for these contrasts is to see the pr eposition “about” as allo wing both for such non-v eridical v erbs to embed questions, and as explaining the non-v eridical interpr etation of the questions in these contexts.23 Let us consider, for instance, the following valid inference: (81) a. John believes that Mary left. b. John has a belief about whether Mary left. To account for the inference, we shall suppose that the sentence (81)-b is analyzed according to the following parse tree, where “John has a belief about whether Mary left” is understood as “J ohn believes some proposition about whether Mary left”:

23. The same will hold of the prepositions “on” or “concerning” whenever they can be used with the same meaning as “about”, for instance in agree on.

118

1

some

John believes

about

P1

whether Mary left

We assume the following semantics (following Heim and Kratzer’s 1998 treatment of quantifi er raising, wher e the DP mo ves and leav es a trace, meaning that there is some proposition about whether Mary left such that John believes it): aaboutb
= OQs(st)OPstwc(P=Q(wc)) awhether Mary leftb = OiOw(L(wc)(m)=L(w)(m)) asome P1b = OX(st)t OY(st)t Pst(X(P) =1š Y(P) = 1) abelievesb
= OPst Oxe Bel (w)(x,P)) = 1 Applying the r ules of functional application, one can check that the sentence is tr ue in w if and only if: P wc(P = Ow(L(wc)(m) =L(w)(m)) š Bel(w)(j,P) = 1), namely if there is a proposition that is a potential answer to the question whether M ary left, and J ohn believes that pr oposition. This analysis allows us to capture the inference from (81)-a to (81)-b, for (81)-a is true in w if and only if Bel(w)(j,Ow(L(w)(m) = 1)) = 1.Crucially, however, this inference no longer rests on the veridicality of the verb, but only on the semantics assumed for “about”. We can generalize this proposal to other v erbs followed by a preposition like “about”, such as agree, or decide. Adapting the previous tree, the proposed analysis for “John and Sue agree on whether Mary left” also yields that there is some potential answ er to the question of whether M ary left on which John and Sue agree:

119

1

some

John and Susan agree about

P1

whether Mary left

In these examples w e assumed that for a pr oposition to be about a question, the pr oposition must corr espond to some potential complete answer in G roenendijk and Kar ttunen’s sense (as in E gré and S pector 2007). Arguably, this meaning for “about” may be too strong for embedded wh-questions after some pr edicates. For instance, in or der to have a belief about who came, it is pr obably suffi cient to believ e a pr oposition that only partially answers the question (rather than a potential complete answer in GS’s sense).24 For a verb like agree, on the other hand, the meaning assumed seems adequate, since in or der to agree about who left, it is not enough to agr ee on some par tial answer: John and M ary may agr ee that Sue left, but disagr ee about J anet, for instance, in which case they disagree about who left. 4.4 Summary In this section we have seen that question-embedding verbs that are nonveridical with r espect to that-complements fall in two main categories, following Lahiri’s 2002 typology (and setting aside the case of doubt, which turns out not to embed questions). F or the verbs that Lahiri calls non-veridical responsive, like agree, our account relies in an essential way on the overt or covert presence of a preposition like “about”. For that rea24. A complete answer, in Groenendijk and Karttunen’s sense, corresponds to a cell in the partition determined by the question. I f John left and M ary didn’t, and they ar e the only two individuals in the domain, the complete answ er to “who left?” is “John left and M ary didn’t”. By contrast, “J ohn left”, “Mary didn’t leave” count as par tial answers. For whether-questions, partial answers happen to be complete answers however.

120

son, such v erbs cannot be taken to satisfy the that-to-whether constraint directly. For verbs like tell, which Lahiri classes as veridical with respect to questions, we have seen that we need to postulate a systematic ambiguity, and to assume that there is a veridical use of tell which allows for whetherembedding. As w e have seen, however, the data in the case of v erbs like tell, predict do not obviously support the present account, if non-veridical readings of embedded questions occur after these v erbs, which ar e not constructed with a pr eposition like about. This, as a matter of fact, may be the most serious problem for the present account, and the point where one may have to revise the semantics of embedded whether-complements proposed in (see Egré and Spector 2007). 5. Conclusion Let us review briefly the main elements of our account and conclude with a discussion of some problems. Why is it that wonder selects only for interrogative complements, believe only for declarative complements, while know allows for both? F ollowing Groenendijk and S tokhof, we have assumed that the difference between know and wonder is that wonder fundamentally selects for the intension of a question; it does not take that-clauses because this would produce a type mismatch. By contrast, know is a propositional verb, like believe, but because know is v eridical, that-complements after know coincide with the extension of the corr esponding whether-question: to know that is to know whether, just as to know whether is to know that.25 Here we have argued that with the ex ception of assertion predicates (like be true/right that), the same inference pattern is satisfied in principle by all veridical attitude verbs. As for the converse, we have argued that verbs like believe and regret do not embed whether-complements because that-clauses after such v erbs cannot be expr essed in the same way as whether-clauses. When non-veridical predicates like agree seem to satisfy the inference from that to whether, some coercion mechanism is in fact triggered by the presence of a preposition in front of interrogative complements, in the same way in which believe that licences the inference to have a belief about whether. How does the pr esent account relate to the ones w e examined above? To some extent, our account agr ees with G inzburg’s intuition that that25. S ee Schaffer 2007, 399 for similar emphasis on the equiv alence between knowing that and knowing whether, but in the context of a discussion of contextualism in epistemology.

121

clauses behave differently after believe and after know. But on the present view, that-clauses continue to denote pr opositions after know and after believe in both cases: the only diff erence is in the lexical semantics of the verbs, not in the objects they select for . O ur account also agr ees with Saebø’s account on the idea that the use ofwhether-clauses is pragmatically restricted in a number of contexts, due to a competition withthat-clauses. However, on our theory these pragmatic restrictions are not suffi cient to explain those cases wher e the v erbs cannot be used. O ur account of the admission and exclusion of whether-complements remains fundamentally semantic in this respect. Here we have assumed that interr ogative complements always denote the tr ue answ er to the question. An alternativ e possibility, explored in Egré and S pector (2007), would be to assume that questions systematically denote some potential complete answer to the question, rather than the true answer to the question. U nder that assumption, “ V whether p” should mean “Vp or V ™p”, rather than “V whichever of p and ™p is the true answer” (namely (poVp)š(™poV™p)), a distinction that is collapsed, precisely, when the verbs are veridical (see Egré and Spector 2007). If so, however, the that-to-whether constraint would be trivially satisfi ed, since trivially Vp entails Vp›V™p. To explain why some propositional verbs like believe or regret do not take whether-complements, we would then need a different story, whereby some semantic featur e of those v erbs blocks the semantic interpretation in front of questions (as in the accounts reviewed in section 2 and as in the strategy used in section 3.3 to ule r out sentences like “it is tr ue whether p”). Such a line of explanation may pr ove more explanatory and economical in the end. On the present account, the reason why verbs like regret or believe do not embed whether-complements is because they lack a semantic featur e (namely v eridicality). It remains to be seen whether this explanation can be stated mor e directly in terms of whether-complements ending up making vacuous or contradictory assertions after such verbs.26 A related weakness of the present theory is that we do not explain the unacceptability of believe whether and it is true whether in the same way, where one may feel that the exclusion is of some common origin in both cases. Like wise, as announced in the intr oduction, our account leav es unexplained why some verbs appear to take wh-questions, despite the fact that they do not take whether-complements (like surprise). Assuming wh26: See Aloni (2007) for quite a diff erent account along such lines.

122

complements really express questions in such contexts, this may suggest that veridicality is not the right criterion, and that further lexical properties of the embedding verbs come into play (as in Guerzoni 2007).27

REFERENCES Abusch D. 2002. “Lexical Alternatives as a Source of Pragmatic Presuppositions”, in SALT XII, Brendan Jackson, ed. CLC Publications, Ithaca NY. 2002. Aloni M. 2007. “F ree choice and exhaustifi cation: an account of subtrigging effects”, in E. P uig-Waldmueller (ed.), Proceedings of Sinn und Bedeutung 11. Barcelona, Universitat Pompeu Fabra, May 2007. Austin J. L. 1954. “Unfair to Facts”, in Warnock et al. (eds.), Austin: Philosophical Papers, Oxford, Oxford Universiy Press. Baker C. L. 1968. Indirect Questions in English, PhD. Dissertation, University of Illinois at Urbana-Champaign. Barwise J. and Cooper J. 1981. “Generalized Quantifiers and Natural Language”, Linguistics and Philosophy 4, 159–219. Berman S. 1991. The Semantics of Open Sentences, PhD. Dissertation, University of Massachusetts. Egré P. 2004. Attitudes propositionnelles et paradoxes épistémiques, PhD. Dissertation, IHPST and Université Paris I, Paris. — 2005. “Savoir, Croire et questions enchâssées”, in P. Schlenker & D. Sportiche (eds.), Proceedings of Division of Linguistic Labor: The La Bretesche Workshop, 2003; also available in Electronic Publications of Philosophia Scientiae. 27. This paper is based in part on chapter 5 of my PhD thesis and derives from earlier work on this topic (Egré 2005). Versions of this paper were presented at diff erent places and conferences since 2005, including at the JSM2005 in Paris, at talks given in linguistics seminars at the University of Paris 7 and Paris 8, at the conference in Linguistics and Epistemology in Aberdeen in 2007, and at the Workshop on Knowledge and Questions organized by F. Lihoreau in Nancy that same year. I am indebted to many people for ex changes and discussions on this topic, including M. Aloni, C. B eyssade, S. B romberger, A. Carballo P erez, D. Bonnay, B. G eurts, K. von Fintel, D. Fox, I. Heim, N. Klinedinst, R. Holton, R. Katzir, F. Lihoreau, L. Nash, D. Pesetsky, N. Richards, F. Roelofsen, I. Rumfitt, K. J. Saebø, J. Schaffer, P. Schlenker, Y. Sharvit, R. Singh, B. Spector, D. Sportiche, R. Stalnaker, T. Williamson, H. Zeevat and R. Zuber. None of them, of course, is r esponsible for the content or form of the ideas defended in this paper . Special thanks, fi nally, are due to my colleagues P . Schlenker and B. Spector for their comments and immensely helpful feedback since the v ery beginning of this r esearch. I also ackno wledge the MIT-France Seed Fund for Collaborative Research, which made it possible for me to write this paper during a stay at the MIT Linguistics & P hilosophy Department (Exchange on Presuppositions and Implicatures).

123

Egré P. & S pector B. 2007. “E mbedded Questions Revisited: An Answ er, not necessarily The Answer”, manuscript, Harvard & IJN, presented at the Journées de Sémantique et Modélisation 2007, Paris. von Fintel K. 2006. “ What is Presupposition Accommodation, Again?”, manuscript, MIT (version of 06/29/2006). Gajewksi J. 2005. Neg-Raising: Polarity and Presupposition, PhD Dissertation, MIT. Gazdar G. 1979. Pragmatics: Implicature, Presupposition, and Logical F orm, Academic Press, New York. Geurts B. 1996. “Local Satisfaction Guaranteed: A Presupposition Theor y and its Problems”, Linguistics and Philosophy 19 (3), 259–294. Ginzburg J. 1995. “Resolving Questions : I & II”, Linguistics and Philosophy, 18, 459–527 & 567–609. Groenendijk J. & S tokhof M. 1982. “S emantic Analysis of Wh-Complements”, Linguistics and Philosophy, 5, 117–233. — 1990. “Partitioning Logical Space”, ESSLLI 1990 Course Notes. Guerzoni E. (2007), “Weak Exhaustivity: A Pragmatic Account”, SALT 17. Heim I. 1991. “Artikel Und Definitheit”, in A. von Stechow and D. Wunderlich (eds.), Semantik: Ein internationales Handbuch der z eitgenössischen Forschung, 487–535, de Gruyter, Berlin. — 1992. “Presupposition Projection and the Semantics of Attitude Verbs”, Journal of Semantics, 9, 183–221. Heim I. & Kratzer A. 1998. Semantics in Generative Grammar, Blackwell. Hintikka J. 1975. “Different Constructions in Terms of the Basic Epistemological Verbs”, in The Intentions of Intensionality, Dordrecht, Kluwer, 1–25. Holton R. 1997. “Some telling examples: a reply to Tsohatzidis”, Journal of Pragmatics 28, 625–628. — 2006. “Facts, Factives and Anti-Factives”, manuscript, MIT. Iatridou S. 2000. “The Grammatical Ingredients of Counterfactuality”, Linguistic Inquiry 31 (2), 231–270. Huddleston R. and P ullum G. 2002. The C ambridge Grammar of the E nglish Language, Cambridge University Press. Karttunen L. 1971. “Some Observations on Factivity”, Papers in Linguistics 4:1, 55–69. — 1977. “Syntax and S emantics of Q uestions”, Linguistics and P hilosophy 1, 3–4 4. Karttunen L. 1977b. “To Doubt Whether”, The CLS Book of S quibs, Chicago Linguistics Society (scanned version available from Karttunen’s webpage). Kiparsky P. and Kiparsky C. 1970. “F act”, in M. B ierwisch and K. H eidolph (eds.), Progress in Linguistics, The Hague, Mouton. Klein E. 1975. “ Two S orts of F active Predicate”, Pragmatic M icrofiche, I, 1: B5–C14.

124

Kratzer A. 2002. “F acts : P articulars or I nformation Units ? ”, Linguistics and Philosophy, 25, 655–670. Lahiri U. 2002. Questions and Answers in Embedded Contexts, Oxford Studies in Theor etical Linguistics, Oxford. Pustejovsky J. 1993. “Type Coercion and Lexical Selection”, in J. Putejovsky (ed.), Semantics and the Lexicon, Kluwer. Russell B. 1918. The Philosophy of Logical Atomism, Open Court Classics, LaSalle, Illinois, 1985. Saebø K. J. 2007. “ A Whether Forecast”, in B. ten Cate and H. Z eevat (eds.), TbiLLC 2005, Berlin, Springer 2007, 189–199. Sauerland U. 2007. “Implicated Presuppositions”, in A. Steube (ed), Sentence and Context, Mouton de Gruyter, Berlin. Schaffer J. 2007. “Knowing the Answer”, Philosophy and Phenomenological Research 75 (2), 383–402. Schlenker P. 2005. “The Lazy Frenchman’s Approach to the Subjunctive (Speculations on Reference to Worlds and Semantic Defaults in the Analysis of Mood)”, Proceedings of Going Romance XVII. Stalnaker R. 1974. “Pragmatic Presuppositions”, repr. as chapter 2 in R.Stalnaker, Context and Content, Oxford Cognitive Science, Oxford, 1999. Tsohatzidis, S. L. 1993. “Speaking of Truth-Telling: the View from Wh-complements”, Journal of Pragmatics 19, 271–279. Vendler Z. 1972. Res Cogitans, Ithaca, NY, Cornell University Press. Williamson T. 2000. Knowledge and its Limits, Oxford UP. Zuber R. 1982. “Semantic Restrictions on Certains Complementizers”, Proceedings of the 13th International Congress of Linguists, S. Hattori & K. Inoue (eds), Tokyo 1982.

125

Grazer Philosophische Studien 77 (2008), 127–146.

CHEAP KNOWLEDGE AND EASY QUESTIONS Martin MONTMINY University of Oklahoma Summary Contrastivism is the idea that kno wledge is question-relative: to know is to be able to answer a contextually salient question. Constrastivism’s main selling point is that it pr omises to r espect ordinary speakers’ judgments about kno wledge claims made in v arious contexts. I sho w that contrastivism fails to fulfi ll this promise, and argue that the vie w I call epistemic pluralism does much better in this respect.

0. Introduction In a highly stimulating series of articles, Jonathan Schaffer has developed a sophisticated v ersion of contextualism, which he calls contrastivism.1 According to this vie w, knowledge is the capacity to answ er questions: a subject S counts as kno wing that p in a giv en context just in case S can answer a contextually salient, multiple-choice question, to which ‘ p’ is the correct answer. Contrastivism is thus a form of r elevant alternatives theory, for it holds that knowledge consists in the elimination of the r elevant alternatives that ar e included in the contextually salient question. Although contrastivism has much to recommend it, this view fails on two fronts. First, the contrastivist’s condition for knowledge is, in some cases, too easily satisfied. This problem occurs in contexts in which the question is too easy to answer. Second, contrastivism is too restrictive. It seems correct, at least in some contexts, to credit subjects with the knowledge that certain skeptical hypotheses do not obtain. Contrastivism implausibly makes such knowledge attributions false, when the subject’s evidence does not strictly entail that the skeptical hypotheses ar e false. I will consider 1. S ee Schaffer (2004, 2005a, 2005b, 2007, forthcoming). Different versions of contrastivism are discussed in Johnsen (2001), Morton and Karjalainen (2003) and Sinnott-Armstrong (2004). See also Dretske (1970). My focus here will be solely on Schaff er’s version of this view.

various responses contrastivists could make regarding these problems, and show that they are unsatisfactory. I will then argue that contextualists, if they wish to r espect the wide div ersity of our intuitiv e judgments about knowledge claims, ought to be epistemic pluralists, that is, they ought to admit a plurality of epistemic standar ds relative to which kno wledge claims can be made. 1. Contextualism and contrastivism Contextualism is the vie w that kno wledge attributions of the form ‘S knows that p’ have context-sensitive tr uth conditions. What vary from context to context ar e the epistemic standar ds that S must meet for the knowledge claim to be tr ue. In “ordinary”, or “low standards”, contexts, the tr uth of ‘S kno ws that p’ requires, in addition to S’ s having a tr ue belief that p, that S satisfy r elatively low epistemic standards, whereas in “skeptical”, or “high standar ds”, contexts, the kno wledge claim will be false unless S satisfi es higher epistemic standards. There ar e diff erent linguistic accounts of the context sensitivity of knowledge claims. O ne possible vie w, which w e may call the indexical account, treats the verb ‘to know’ as an indexical in the br oad sense: this verb belongs to the family of context-sensitiv e expressions that include “pure” indexicals such as ‘I’, today’ ‘ and ‘tomorrow’, as well as third-person pronouns, demonstratives and other expr essions whose denotations v ary depending on the context. A ccording to the indexical account, ‘kno w’, relative to a giv en context, designates a r elation between a person and a proposition corresponding to certain salient epistemic standards. But contextualists need not be saddled with the indexical account. Instead of holding that ‘kno w’ is a context-sensitiv e expression, contextualists could appeal to a “ covert variable” analysis, accor ding to which the context sensitivity of kno wledge claims is traced to the pr esence of a structural position in logical form that is occupied b y a hidden v ariable. On the hidden variable account, ‘know’ is a 3-place predicate with a covert argument place that takes epistemic standards as values. Thus, the logical form of ‘S kno ws that p’ is something like ‘S kno ws that p according to standards E’, where ‘E’ is a free variable whose value is fixed in context. Another option for contextualists is what we may call the unarticulated constituent account. According to this account, kno wledge sentences ar e syntactically complete, but fail to express complete propositions indepen128

dently of a context of utterance. The unarticulated constituent account holds that an utterance of ‘S kno ws that p’ conveys a proposition of the form S knows that p according to epistemic standards E. The standards E are a constituent of the content of a kno wledge claim, but this constituent is articulated neither at the surface level nor at a deeper level of syntactic analysis. I will not try to adjudicate among these three accounts here. I will thus not address the objections that sev eral critics, including Schaff er (2004; 2005a, 261), have raised against the indexical account. It should be clear, though, that if the indexical account turns out to be unviable, there are at least two other linguistic accounts contextualists can invoke. Contrastivism is the idea that knowledge is question-relative: to know is to have the ability to answer a contextually salient question.2 Not every conversational context explicitly poses a question, but, Schaff er claims, a multiple-choice question is always r ecoverable from context. A ccording to contrastivism, kno wledge is a ternar y relation among a subject S, a selected proposition p (the correct answer to the question), and a contrast proposition q, which can be regarded as a disjunction of relevant alternatives. Hence, a simple kno wledge claim of the form ‘S kno ws that p’ is properly understood as conveying the content that S knows that p rather than q. Alternatively, the claim ‘S kno ws that p’ can be read as: S knows that p as the tr ue answ er to the contextually salient question ‘ p or q?’. This means that kno ws-that can be r egarded as contextually equiv alent to knows-whether.3 In a context in which the question is whether M oore has hands or stumps, the claim ‘M oore knows that he has hands ’ is true just in case Moore knows whether he has hands or stumps (assuming that Moore does have hands). Contrastivism is a form of contextualism that identifies epistemic standards with questions to be answered. Furthermore, contrastivism invokes the hidden v ariable account: knowledge sentences of the form ‘S kno ws that p’ contain a co vert variable for a contrast pr oposition. This means that the logical form of ‘S knows that p’ is something like ‘S knows that p rather than q’. On the question-relative version of contrastivism, the logical form of the knowledge claim is something like ‘S knows that p relative to question Q’. For our purposes, it will not matter which logical form the contrastivist opts for. 2. N ote that guessing rightly does not count as having the required ability. 3. S ee Schaffer (forthcoming).

129

Contrastivism can account for a gr eat deal of the context sensitivity of our kno wledge attributions. I n many contexts, w e willingly attribute knowledge that p to a subject on the basis of her capacity to discriminate between p and a set of salient alternatives. However, I will argue, there are many correct uses of knowledge claims that contrastivism fails to account for. In my view, contextualists had better be epistemic pluralists. Epistemic pluralism holds that the space of epistemic standards is multidimensional: there is a wide range of epistemic standards according to which knowledge claims can be correctly made.4 For example, one can deny knowledge that p because the subject was right by accident, or because the subject cannot eliminate some error possibilities, or because the subject would still believ e that p if p were false, or because the subject cannot provide reasons for her belief that p, etc. I n order to accommodate the full div ersity of ways in which knowledge claims can be suppor ted or contested, a single type of epistemic standard will not suffice. I will discuss this objection to contrastivism in the second half of the paper.My other objection to contrastivism, which I will present in the next two sections, concerns easy questions. 2. Easy questions Assimilating knowledge to the ability to answ er a multiple-choice question is problematic. As John Hawthorne points out, “the very asking of a question may provide one with new evidence regarding the subject matter at hand” (2004, 78). In typical conversations, in asking whether p or q, a questioner conveys the information that either p or q to interlocutors who may not already possess this information. Hilary cannot tell an elm from a beech, but has no trouble distinguishing either from a willow. Hence, if someone shows Hilary a picture of an elm and asks him ‘Is this an elm or a willow?’, he can come to know that the photo depicts an elm. But since Hilary acquired this kno wledge in par t thanks to the v ery asking of the question, the fact that he can correctly answer the question should not be regarded as manifesting knowledge he already possessed. We would not be inclined to hold that Hilary knew that the picture depicted an elm before he was asked the question. 5 4. See Unger (1986) for a more detailed exposition of this view. Neta (forthcoming) holds a similar view. 5. Did Hilary know that the pictur e depicted an elm rather than a willo w? I will addr ess this question in the next section.

130

Schaffer is awar e of this pr oblem, and makes two points in r esponse. First, he writes, w e need not suppose that the subject is a par ticipant in the conv ersation: the subject “ may not be privy to the conv ersation at all. O thers might be discussing what she kno ws” (2005a, 250, n. 20). But this does not addr ess the pr oblem. Suppose Mary has no idea who the President of Tajikistan is: she vaguely remembers that Tajikistan used to be a S oviet Republic, and is awar e of the fact that Tajikistan is no w a country. But this is the extent of what w e would say she kno ws about Tajikistan. It thus seems correct to say, regardless of the context, that she does not know that Emomali Rahmon is the President of Tajikistan. She has not ev en heard the name ‘E momali Rahmon’ befor e. Yet, if asked ‘Is the P resident of Tajikistan E momali Rahmon or K oko the gorilla? ’, she would giv e the corr ect answer. Hence, when talking about M ary in absentia, we would judge that although (i) she kno ws the answ er to the question whether Emomali Rahmon or Koko the gorilla is the President of Tajikistan, (ii) she does not know that Emomali Rahmon is the President of Tajikistan. Schaffer’s second point is that the subject need not trust the questioner to select the likely options. The subject, Shaffer writes, “might just play along” (2005a, 250, n. 20). But this does not help either. If the subject is just playing along and does not believ e that p, after being asked whether p or q, then she does not satisfy one of the conditions for kno wing that p. This means that cases in which the subject does not tr ust that likely options have been selected by the questioner should be excluded from the equation between knowing that p and the ability to answ er the question whether p or q. When the subject is “just playing along ”, her ability to give the right answer to the question ‘p or q?’ should not be construed as constituting her knowing that p, for she may not believe that her answer is true. Contrastivism is thus too liberal: it attributes knowledge to subjects who would more plausibly be regarded as ignorant. 3. Conditions for knowledge As we saw in section 1, contrastivism holds that the kno wledge claim ‘S knows that p’ is properly understood as conveying the content that S knows that p rather than q. According to Schaff er (2005a, 255), S knows that p rather than q just in case: (i) p; (ii) S has pr oof that p rather than q; (iii) S is certain that p rather than q on the basis of (ii). 131

Schaffer acknowledges that counter examples to his analysis ar e inevitable, but adds that it is at least a promising and illuminating gloss of knowledge. It seems to me, ho wever, that the pr oblems with Schaff er’s analysis are deep and serious. F irst, the analysis faces the pr oblem discussed in the previous section: in some cases, a subject cannot be plausibly credited with knowing that p just in vir tue of her ability to eliminate q, for q is too easily eliminable. Al does not kno w what an aar dvark looks like. However, he is certain that the animal depicted on the pictur e is an aardvark rather than a dog. He also has proof of that, for Al knows what dogs look like. H ence, Schaff er’s analysis entails that ther e are contexts in which the contrast pr oposition is easily eliminable and Al counts as knowing that the animal depicted on the picture is an aardvark. Schaffer’s conditions for knowledge are thus too easily satisfi ed. Does Al know that the animal depicted on the pictur e is an aar dvark rather than a dog? I have no clear intuition, but both y‘ es’ and ‘no’ answers are pr oblematic for the contrastivist. I f the contrastivist insists that Al does know that the animal depicted on the pictur e is an aar dvark rather than a dog, then her claim that utterances of the form ‘S kno ws that p’ should be understood as meaning that S knows that p rather than q cannot be corr ect. This is because ther e is no context in which it would be correct to asser t, ‘Al knows that the animal depicted on the pictur e is an aardvark’, for b y assumption, Al has no idea what an aar dvark looks like. And of course, if the contrastivist concedes that Al does not kno w that the animal depicted on the pictur e is an aardvark rather than a dog, then she must ackno wledge that this case pr ovides a counter example to her view. Schaffer (2005a, 257) suggests that he is willing to bite the bullet and accept that, in some contexts, it may be corr ect to assert, ‘Al knows that the animal depicted on the pictur e is an aar dvark’. However, he points out, this bullet is less unsavory than one may think. Al can eliminate the possibility that the picture depicts a dog, and this is an epistemic achievement that is marked by the knowledge claim. Claiming that Al knows that the picture depicts an aardvark distinguishes Al’s epistemic standing from that of, say, Ed, who is incredibly ignorant and would not even be able to eliminate the dog possibility. However, there is no need to attribute to Al this knowledge in order to mark his epistemic achievement. This achievement can simply and plausibly be characteriz ed by asserting that Al, but not Ed, knows that the picture does not depict a dog. Schaffer (2005a, 258; 2007, 251 n. 20) considers another possible 132

response: in or der to count as kno wing that p (rather than q), a subject must hav e a nominal justifi cation, or some positiv e evidence, for p. Unfortunately, this r esponse undermines the spirit of contrastivism, for it entails that knowing that p is not equivalent to knowing the answer to ‘p or q?’. Knowing the answer to the question is no longer suffi cient for knowledge. But more importantly, there is another serious difficulty with Schaffer’s analysis that his response fails to address. Schaffer (2005a, 256–257) claims that his analysis resolves Gettier cases. Suppose that the clock stopped on 3 p.m. 24 hours ago . Sally looks at the clock and, on this basis, forms the belief that it is 3 p .m. Our inclination is to say that S ally does not know that it is 3 p.m., and this inclination does not appear to be subject to context-sensitive variations. The contrastivist may tr y to explain this inclination by invoking the fact that Sally cannot eliminate the possibility that it is, say, 4 p.m. and the clock is off . But the contrastivist solution is unsatisfactory, for there are many possibilities that Sally can eliminate: it is 4 p.m. and the clock is on time, it is 4 p.m. and the clock is stopped on 4 p.m., etc. Furthermore, Sally clearly has positive evidence for the fact that it is 3 p.m. Hence, on Schaffer’s view, in a context in which the question is whether it’s 3 p.m., or 4 p.m. and the clock is on time, it is corr ect to assert that Sally knows that it’s 3 p.m. This example shows that for some contrast propositions q, the equation between knowing that p and being able to answ er the question ‘ p or q?’ correctly cannot be maintained, ev en if the subject has positiv e evidence for p. However, the equation may hold for a cer tain class of contrast propositions. Hence, the contrastivist could perhaps hold that although utterances of ‘S knows that p’ made in different contexts may involve different contrast propositions, there are context-independent constraints on the admissibility of a contrast proposition. For example, the contrastivist could hold that any alternativ e that is similar to actuality should fi gure among the disjuncts of a contrast proposition. In other words, the contrastivist could invoke something like D avid Lewis’s (1996) Rule of Resemblance, according to which an alternative is relevant if it saliently resembles actuality. Invoking such a r ule would allo w the contrastivist to handle G ettier cases more successfully. The claim ‘Sally knows that it is 3 p.m.’ cannot be true, regardless of the context in which it is made, because there are many possibilities resembling actuality that Sally cannot eliminate, such as the possibility that it is 4 p.m. and the clock stopped on 3 p.m. Similarly, there 133

would be no context in which Al counts as kno wing that the picture depicts an aardvark, for there are possibilities similar to actuality that Al cannot eliminate, for example, the possibility that the picture depicts a pangolin. The Rule of R esemblance also makes Schaff er’s additional r equirement that the subject have positive evidence for p unnecessary, for a subject who can eliminate alternatives that resemble actuality automatically counts as having positive evidence for her belief. 4. Eliminating alternatives The revised version of contrastivism introduced at the end of the previous section is a step in the right direction. Unfortunately, this view also faces serious problems. One’s evidence that an alternative q is false may be more or less conclusive. But what is required to eliminate (or rule out) q? Most supporters of r elevant alternativ es account hold that in or der to eliminate q, one’s evidence must entail that q is false. 6 But nothing, it seems, should prevent the r elevantist fr om subscribing to a w eaker condition, namely, that a subject S has eliminated q if S has good inductive evidence against q. Now, to be sur e, the wor d ‘eliminate’ is a technical term, and one is free to stipulate how one wishes to understand it. However, as I will show, given other aspects of the contrastivist account, some stipulations should be r ejected. Recall that accor ding to contrastivism, a subject S counts as knowing that p in context C just in case S can eliminate the contrast proposition q in C. But S’s counting as knowing that p in context C is also equated with S’s knowing the answer to the question ‘ p or q?’, and to S’s knowing that p rather than q. If they want to maintain these equations, I will show, contrastivists ought to be contextualists about ‘S has eliminated q’; in other words, there should not be a particular, context-independent condition for the elimination of an alternative. In some contexts, a subject who mer ely has good inductiv e evidence against q may be considered as knowing that p rather than q (or as knowing the answer to the question ‘p or q?’). Do I know that I am looking at my cat, rather than a r obot look-alike that was cr eated by scientists and substituted for my cat when I was asleep? I t seems v ery plausible to say that I do, at least in some contexts. I hav e good reasons to believe that I 6. See Dretske (1981, 364), Lewis (1996, 553) and Schaff er (2005a, 255).

134

am not looking at a r obot look-alike. The construction of such an entity seems incompatible with the curr ent state of technology . And ev en if it turned out to be possible, it would r equire a large eff ort for no plausible purpose: why would people decide to replace my cat with their robot? So, in some contexts at least, it seems correct to assert that one knows that p rather than q, even though one’s evidence is, strictly speaking, compatible with q. But this does not mean that good inductiv e evidence against q is always considered sufficient. In some contexts, the conversational participants’ practical interests may be such that it is of the utmost impor tance not to get things wrong. They may thus adopt a stricter notion of evidence, and hold that a subject does not know that p rather than q unless she has infallible evidence that not-q. There ar e, no doubt, many other cases in which w e ar e inclined to attribute knowledge to a subject S, ev en though S’s evidence is compatible with some salient err or possibility. Consider the follo wing example imagined by Jonathan Vogel: Hole-In-One Case. Sixty golfers are entered in the Wealth and Privilege Invitational Tournament. The course has a short but difficult hole, known as the “Heartbreaker”. Before the round begins, you think to yourself that surely, not all sixty players will get a hole-in-one on the “Heartbreaker”. (1999, 165)

Our intuition, Vogel writes, is that you know that not all sixty players will get a hole-in-one. Clearly , the possibility that all sixty play ers will get a hole-in-one is salient in this context. This means that you count as having eliminated that possibility, despite the fact that y our evidence does not entail that it is false. 7 It is wor th noting that the pr obability that all sixty play ers will get a hole-in-one could be identical to the pr obability that Smith will win the next lottery. However, we are inclined to assert that we do not know that Smith will not win the lottery. In the lottery case, unlike in the Hole-InOne case, one’s evidence must entail that an alternative is false in order to count as having eliminated that alternative. A view that allows for contextsensitive standards for the elimination of alternatives is thus desirable. 8 7. In another paper, Vogel invites the r eader to “consider whether y ou know that a plane about to crash into a mountain will not emerge unscathed on the other side, or whetherouy know that the blood of every person in San Francisco will not freeze in the next minute. According to quantum mechanics, there is a tiny probability of the former, and according to thermodynamics, there is a tiny probability of the latter” (2007, 82, n. 19). 8. My point here is not that we count as knowing that not all sixty players will get a hole-inone in all contexts. Perhaps in a given context, a speaker could argue that the Hole-In-One Case

135

One of Schaffer’s main arguments in support of his view invokes the role that knowledge claims play with respect to the project of inquiry.9 A subject counts as kno wing if she can answ er the question at hand. The problem is that often a subject who is consider ed to have correctly answered ‘p’ to the question ‘ p or q? ’ has at best good inductiv e evidence against q. In other words, progress in inquiry does not always require infallible evidence against the rejected propositions. Hence, if Schaff er wishes his vie w to fi t our practice of attributing knowledge to a subject on the basis of her ability to answer questions, then he ought to admit a context-sensitive parameter that concerns the subject’s evidence against the contrast proposition. Another of Schaffer’s arguments in favor of his view supports this idea. Consider the following example: Suppose that Student, Assistant and Professor are visiting the zebras at the zoo. Student is remarkably ignorant and can’t even discern a z ebra from a mule; Assistant can discern a z ebra from a mule, but cannot discern a z ebra from a cleverly painted mule; P rofessor can discern a z ebra even from a clev erly painted mule b y anatomical featur es that no mer e paint job can disguise. (2005a, 260)

Schaffer writes that a desirable feature of a theory is the ability to capture the distinctions among these thr ee cases. Contrastivism does that, he adds. Student does not know that the beast is a zebra rather than a mule. Assistant knows that the beast is a z ebra rather than a mule, but does not kno w that the beast is a z ebra rather than a painted mule. P rofessor knows that the beast is a zebra rather than a mule, and that the beast is a z ebra rather than a painted mule. (2005a, 264)

I agree with Schaffer that a good theory should accommodate the distinctions he signals. The problem, though, is that his o wn theory misses other distinctions among subjects ’ epistemic achiev ements. H ere is a modified version of his story: Student and Assistant can both discern a zebra from a mule, but only Assistant is aware of the fact that zoo keepers do not typically go around painting mules, and that zoos usually exhibit genuine specimens of the indicated types is just a lottery case, and that by parity of reasoning, we should deny that we know that not all sixty players will get a hole-in-one. I am not disputing the legitimacy of such a denial: my point is merely that Vogel’s positive knowledge attribution is correct in at least some contexts. 9. S ee Schaffer (2005a, 237, 241–242; 2005b, 117; forthcoming).

136

of animals. Hence, Assistant has good inductive evidence that the beast is not a painted mule, wher eas Student lacks any evidence of the sor t. Professor’s evidence that the beast is not a painted mule is better than Assistant ’s, for Professor can discern a zebra from a cleverly painted mule.

Does Assistant know that the beast is a zebra rather than a painted mule? A view that holds that there is a fixed, context-independent, threshold for eliminating an alternative is forced to collapse the distinction between Student and Assistant, or the distinction between Assistant and Professor. The foregoing considerations suggest that, at least in the contrastivist framework, the conditions for eliminating q are best regarded as identical to the conditions for knowing that not-q. Unfortunately, this makes contrastivism explanatory idle. According to this view, ‘S knows that p’ is true in context C just in case ‘S has eliminated q’ is true, where ‘q’ is a salient alternative (or a disjunction of salient alternatives) in C. However, we just saw that ‘S has eliminated q’ is true in C just in case ‘S knows that not-q’ is true in C. And ‘S knows that not-q’ is true in C iff S has eliminated the alternative to not-q that is salient in C. By assumption, this alternative is q. Hence, ‘S knows that p’, ‘S has eliminated q’ and ‘S knows that not-q’ all have the same tr uth conditions in C, but this does not tell us what these truth conditions are. This also means that the contrastivist’s idea that the context sensitivity of simple knowledge claims can be accounted for in terms of ‘kno wingrather’ or ‘knowing the answer’ has little explanatory value. As the various examples suggest, claims of the form ‘S knows that p rather than q’ and ‘S knows the answer to the question “p or q?”’ are just as context-sensitive as simple knowledge claims. Contextualists generally consider the presentation of error possibilities as an important contributing factor involved in contextual shifts in epistemic standards: a subject S counts as kno wing that p in a given context only if S can eliminate salient err or possibilities. B ut as the for egoing discussion shows, a closer examination of this factor quickly r eveals that at least two context-sensitiv e parameters aff ect the tr uth conditions of knowledge claims: what err or possibilities a subject must eliminate, and what evidence that subject needs in or der to count as having eliminated a given error possibility. One lesson I draw fr om the examples discussed in this section is that a contextualist account that admits only one context-sensitive parameter cannot r espect our intuitiv e judgments about knowledge claims made in various contexts. I take this to be a crucial step towards epistemic pluralism. 137

Contrastivists could of course intr oduce another context-sensitiv e parameter in their account. Such a parameter would concern how strong a subject’s evidence needs to be in or der to count as having eliminated a relevant alternative. In other words, contrastivists could hold that ‘S knows that p’ is true in C just in case S satisfies salient epistemic standards E with respect to the elimination of the salient alternativ e q. But this account is unnecessarily complicated. A simpler theor y, appealing to only one context-sensitive parameter the epistemic standards S must satisfy in order to count as knowing that p, seems preferable. If one’s account of knowledge attributions is to inv oke contextually salient epistemic standar ds, it is more sensible to make them knowledge standards, as opposed to elimination standards. 5. Cheap knowledge In the previous section, I discussed a few cases in which amoderate skeptical hypothesis SH is contextually salient, for example, the possibility that my cat has been replaced by a look-alike robot. We saw that in these cases, it seems correct, at least in some contexts, to assert that one knows that notSH, even if one merely has good inductive evidence against SH. According to some authors, cases involving radical skeptical hypotheses are different, for, they argue, w e do not hav e any evidence against such hypotheses. Contrastivists concur, and on their vie w, one cannot know that not-SH, when SH is a radical skeptical hypothesis. Consider the sentence: (1) Moore knows that he’s not a brain-in-a-vat (BIV, for short). According to Schaffer (2004, 91; 2005a, 263), (1) cannot be truthfully asserted. This is because, pr operly understood, an asser tion of (1) is to the effect that Moore knows that he’s a non-BIV rather than a BIV.10 This entails that any assertion of (1) is false, since Moore cannot eliminate the alternative that he is a BIV . Moore, Schaff er writes, has no evidence at all against this alternativ e. Contrastivism thus blocks what Schaff er calls cheap knowledge, that is, kno wledge that is without any evidential basis (2007, 230). 10. An assertion of (1) cannot be understood as meaning, for example, that M oore knows that he’s a non-BIV rather than a man with stumps. Having stumps is not an alternative to not being a BIV, for the two propositions are compatible.

138

Contrastivism can accommodate the intuition shared by many that it is correct to deny (1). The problem is that not everyone shares this intuition: many speakers would fi nd asser tions of (1) intuitiv ely plausible. 11 As a matter of fact, many epistemologists, including so-called ‘neo-Mooreans’, consider it a desideratum for a theor y that it entails the tr uth of (1). Consider, for example, the safety condition for knowledge: S’s belief that p is safe just in case in the nearby possible worlds in which S believes that p, p is tr ue.12 Moore’s belief that he is not a BIV is safe, since in nearb y possible worlds in which he believ es that his belief is tr ue. Advocates of safety would consider it an advantage that on their view, assertions of (1) come out true. Contrastivism fails to respect the judgment, accepted by many, that (1) is true. It thus involves a kind of error theory. This is unfortunate, for there is no obvious explanation as to why this judgment is mistaken. It appears that, at best, w e have a clash of intuitions betw een neo-Mooreans, who hold that (1) is true, and contrastivists, who deny that. A theory that can accommodate both intuitions seems pr eferable. In other wor ds, it seems preferable to hold that there are different legitimate interpretations of an assertion of (1), some accor ding to which it is tr ue, and some accor ding to which it is false. I should add that the contrastivist’s position regarding (1) not only goes against an intuition shar ed by many speakers, but also inv alidates some plausible inferences. For example, fr om the pr emise ‘I kno w that I hav e hands’, Moore cannot infer ‘I know that I am not a BIV ’. Contrastivism thus flouts a pre-theoretically appealing closure principle for knowledge, according to which normally , if S kno ws that p, then S kno ws logically weaker propositions that obviously follow from p. Moore’s epistemic position with respect to the logically weaker proposition that he is not a BIV seems to be at least as good as his epistemic position with r espect to the logically stronger proposition that he has hands. It seems strange that he can be said to know the latter, but not the former. Contrastivists can provide an explanation as to why the inference from ‘I know that I have hands’ to ‘I know that I am not a BIV’ is invalid, but this explanation inv okes non-standard closure principles. The inference is invalid, Schaff er writes, because “one cannot use one’s knowledge that one has hands rather than stumps, to come to kno w that one is not a 11. See Neta (forthcoming), for a similar point. 12. See, for instance, Sosa (2002) and Pritchard (2005).

139

brain-in-a-vat rather than a creature with stumps” (2007, 247). Schaffer’s explanation is based on a set of general epistemic closur e principles that are compatible with the contrastivist framework. Interestingly, according to these principles, the infer ence from ‘I do not kno w that I am not a BIV’ to ‘I do not know that I have hands’ is valid. In a context in which the fi rst sentence is utter ed, the contrast pr oposition is that I am a BIV (with vat-images of hands). Hence, what I mean in uttering this sentence is that I do not kno w that I am a non-BIV rather than a BIV , and from this, I can infer that I do not kno w that I hav e hands rather than v atimages of hands. 13 The fact that Schaffer’s closure principles validate the inference from ‘I do not know that I am not a BIV ’ to ‘I do not kno w that I have hands’, but invalidate the inference from ‘I know that I have hands’ to ‘I know that I am not a BIV’, is problematic, for the second inference is arguably just as plausible as the fi rst. This consequence of Schaff er’s closure principles would be mor e palatable if it could be sho wn to be unav oidable, given some pre-theoretically plausible adequacy conditions for closure. But this is not the case. Schaff er’s defense of his closur e principles is based on a number of adequacy conditions, one of which he calls epistemic modesty: “Surely we possess modest knowledge of the external world. For instance, Moore knows that he has hands. H e need only look. And sur ely we also suffer modest ignorance of the external world. F or instance, Moore does not know that he is not a brain-in-a-vat” (2007, 236). Epistemic modesty clearly begs the question in fav or of contrastivism, and thus lacks the prima facie neutrality expected fr om an adequacy condition. Furthermore, Schaffer’s statement of epistemic modesty appears to commit him to what DeRose (1995) calls theabominable conjunction, since it combines the claim that Moore knows that he has hands with the claim that he does not know that he is not a BIV.Schaffer would no doubt protest that this is not a full characterization of epistemic modesty: fully spelled out, epistemic modesty asserts that Moore knows that he has hands rather than stumps, and he does not know that he is a non-BIV rather than a BIV . But this further supports the idea that epistemic modesty begs the question in favor of contrastivism, for the palatability of this constraint erquires that we interpret it the way a contrastivist would. Epistemic modesty thus lacks any pre-theoretical appeal. Schaff er’s unduly restrictive closure principles, which are steeped in epistemic modesty, should thus be rejected. 13. See also Schaffer (2005a, 260).

140

There is a simple way to addr ess this issue. O ne need not consider Moore’s knowledge that he is not a BIV as immodest. The contextualist can hold that fully characteriz ed, an asser tion of (1) is to the eff ect that Moore knows that he is not a BIV according to lo w, ev eryday, epistemic standards. This does not commit the speaker to attributing to M oore the immodest knowledge that he is not a BIV according to the skeptic ’s high epistemic standards. This means that in contexts in which lo w epistemic standards ar e in place, one can tr uthfully asser t (1). F urthermore, the inference from ‘I know that I have hands’ to ‘I know that I am not a BIV’ is valid, provided that the two kno wledge claims ar e made accor ding to low epistemic standards. And the inference from ‘I do not know that I am not a BIV ’ to ‘I do not kno w that I hav e hands’ is also v alid, if the two claims invoke high epistemic standar ds. Unlike contrastivism, this vie w can respect both the intuition that one can tr uthfully assert (1), and the pre-theoretically appealing closure principle according to which normally, if S knows that p, then S knows logically weaker propositions that obviously follow from p. 6. Epistemic pluralism In the previous section, I have argued that there are contexts in which one can truthfully assert ‘I know that I am not a BIV ’; in such contexts, this assertion means that one kno ws that one is not a BIV accor ding to lo w epistemic standards. But some may reject the idea that I know that I am not a BIV according to low epistemic standards. Since I lack any evidence for the proposition that I am not a BIV, it may be argued, I cannot be said to satisfy ev en low epistemic standards with r espect to that pr oposition. For this reason, one may add, the idea that knowledge claims are relative to epistemic standards does not help solve the problem of cheap knowledge: it seems incorrect to attribute knowledge that one is not a BIV, whether it is according to high or low standards, for this alleged knowledge is unsupported by any evidence. This line of r easoning is r epresentative of a cer tain form of radical skepticism. Some radical skeptics not only deny that w e have knowledge about the external world, but also hold that our beliefs about the external world lack any justification. Following Richard Feldman (1999, 111), let us call these justification skeptics. According to such skeptics, I hav e no justification, or no evidence, for the belief that I have hands, since I have 141

no non-circular grounds for favoring this belief over, say, the belief that I am a BIV undergoing qualitativ ely identical perceptual experiences. Justification skeptics reject as question-begging any r eliance on the kind of evidence an epistemic externalist would favor, for instance the propensity of my perceptual mechanisms to form accurate r epresentations of the environment. Such evidence is illegitimate, justification skeptics contend, since it presupposes the tr uth of cer tain propositions about extra-mental r eality. Only propositions about my impressions and experiences, constr ued internally, are acceptable; but clearly, these are insufficient to justify even my belief that I have hands, since I would have the same impressions and experiences were I a BIV. Contextualists hav e two options r egarding justifi cation skepticism: rejection and accommodation. A number of contextualists hav e opted for rejection, by appealing to an externalist construal of epistemic standards. On their view, our beliefs about the external world do meet ordinary epistemic standards, for such standards are externally construed. For example, accor ding to D eRose’s (1995) account, the asser tion ‘S knows she has hands’, made in a low-standards context, is true just in case S’s belief that she has hands issensitive; that is, if S had not had hands, then she would not have believed that she has hands. Hence, on DeRose’s view, S counts as knowing that she has hands according to low epistemic standards. DeRose (1995, 50) acknowledges that his account assumes certain things the skeptic claims are not known: that S has hands, and is thus not a BIV; that in the closest possible worlds in which S does not have hands, S is aware of that; etc. But, as DeRose puts it, he is not playing “King of the Mountain”; that is, he is not attempting to establish anti-skepticism, by appealing to premisses that the skeptic would accept. DeRose’s account thus discards justification skepticism right off the bat. This, of course, does not mean that all forms of radical skepticism ar e rejected. A skeptic can still truthfully deny that we have knowledge about the external world, provided that high epistemic standards are in place. For example, the skeptic can correctly assert that we do not know that we have hands according to high epistemic standards, by invoking the fact that in the distant possible world in which S is a BIV, S would believe that she has hands. This is compatible with holding that in or dinary, low-standards, contexts, S counts as knowing that the remote error possibility that she is a BIV does not obtain. And this “low-standards” knowledge is supported by evidence, constr ued externally. Many of S’s beliefs about the external world are supported by such evidence. S thus kno ws many things about 142

her world according to low standards, and, by epistemic closure, S knows that she is not a BIV according to low standards. It is, it seems to me, per fectly legitimate for contextualists to inv oke externalist epistemic standar ds. However, the idea that the only acceptable constr ual of epistemic standar ds is externalist strikes me as o verly r estrictive, and as going against the spirit of contextualism, for this view purpor ts to r espect uses of kno wledge sentences that ar e plausible. I f one holds that the skeptic who denies kno wledge because the subject fails to meet high standar ds is making legitimate use of ‘kno w’, then one should say the same thing about the justifi cation skeptic who denies knowledge because the subject’s belief lacks non-circular evidence: the latter form of skepticism does not appear mor e objectionable than the former. However, as I wr ote above, contextualists need not r eject justifi cation skepticism; they have the option of accommodating this point of view. By this, I mean that contextualists can hold that what counts as legitimate evidence is a context-sensitive matter. In other words, there may be different ways in which the requirements associated with knowledge attributions can be made mor e stringent. O ne can hold that a subject fails to kno w, because her level of justification is not sufficient to eliminate remote error possibilities. Alternativ ely, one can follo w the justifi cation skeptic and deny knowledge because the subject lacks any justifi cation for her belief. By admitting a multidimensional space of epistemic standards, epistemic pluralism makes room for a parameter that concerns what counts as legitimate evidence. I should note that nothing prevents contextualists from adjusting their terminology and describing both forms of skepticism as inv oking high epistemic standards. First, a skeptic can raise epistemic standards by drawing attention to the fact that the subject’s evidence does not guarantee that some remote error possibilities do not obtain. The justification skeptic can also be regarded as elevating a standard for knowledge, namely the standard that concerns what counts as legitimate evidence for a belief . Hence, on the proposed terminology, justification skepticism is just another form of high-standards skepticism. It is also wor th noting that contextual shifts r egarding what counts as legitimate evidence ar e not confi ned to issues r elated to skepticism. Externalists hold that a subject counts as kno wing simply in vir tue of instantiating a certain relationship (reliability, truth tracking, elimination of relevant alternatives, safety, etc.) with her envir onment. On this view, 143

a subject may be tr uthfully said to know, even though she is ignorant of that relationship. Internalists reject this point of vie w, and require more for knowledge: a subject knows that p only if she can provide justification for her belief that p. In other words, the capacity to articulate reasons for one’s beliefs is deemed necessary for knowledge.14 This internalist perspective on knowledge has been criticized for “overintellectualizing” kno wledge. It seems that unsophisticated epistemic subjects, who ar e incapable of understanding justifi catory arguments or engaging in reasoning, can be credited with knowledge. The internalist criterion for knowledge may also turn out to be too demanding with respect to many of the things w e typically take ourselves to know. The existence of such counterexamples has led many to discard internalism. But this is too quick: the counterexamples are damaging to an invariantist version of internalism, but they leav e a contextualist appr oach that accommodates internalism unscathed. A ccording to epistemic pluralism, w e need not choose between externalism and internalism: both externalists and internalists invoke legitimate requirements for knowledge. Their mistake is to believe that a single perspective on knowledge is suitable for all contexts.15 To paraphrase David Lewis, (1994, 424), w e can follow the internalists’ (or externalists’) lead for the duration of the conversation, but we should oppose their attempts to settle the question once and for all. 7. Conclusion I hav e objected to contrastivism on the gr ounds that it fails to r espect many of our judgments about kno wledge attributions. I n some r espects this view is too liberal: when the question is easy , contrastivism licenses knowledge claims that appear to be false. B ut contrastivism is also too restrictive, for it implausibly prevents us from correctly claiming to know that some skeptical hypotheses, moderate and radical, do not obtain. uch S knowledge, I have argued, is not cheap, for it can be said to be supported by adequate evidence. 14. This characterization of the diff erences between externalism and internalism is r ough, and does not do justice to the many v ariants of these two vie ws. However, it will ser ve our purposes. 15. I should insist that I am not recommending an “anything goes” approach to knowledge claims. We sometimes use knowledge sentences incorrectly or non-literally, and epistemic pluralism is not designed to treat such uses as strictly speaking correct.

144

I have defended epistemic pluralism, a version of contextualism according to which there is a diverse range of epistemic standards that can be in place in different contexts. The main motivation for this view is entirely in the spirit of contextualism. I have shown that a form of contextualism, such as contrastivism, that admits only one type of epistemic standar d cannot respect our intuitive judgments about knowledge claims made in various contexts. The diversity of such judgments is too wide to be captur ed by a single context-sensitiv e parameter. The contextualist should thus hold that the space of epistemic standards is multidimensional. I should insist that epistemic pluralism need not ex clude all appeal to relevant alternatives. In some contexts, the prevailing epistemic standards may be best described in terms of the putativ e knower’s ability to eliminate relevant alternatives. For example, in some contexts, our inclination to deny knowledge to a subject S may be best explained b y the fact that S’s evidence is compatible with some salient err or possibilities. My point in this paper has been that this type of explanation cannot wor k for all cases.16

REFERENCES DeRose, K. 1995. ‘S olving the S keptical Problem’, Philosophical R eview 104, 1–52. Dretske, F. 1970. ‘Epistemic Operators’, Journal of Philosophy 67, 1007–1023. — 1981. ‘The Pragmatic Dimension of Kno wledge’, Philosophical Studies 40, 363–78. Feldman, R. 1999. ‘Contextualism and Skepticism’, Philosophical Perspectives 13, 91–114. Hawthorne, J. 2004. Knowledge and Lotteries, Oxford University Press, Oxford. Johnsen, B. 2001. ‘Contextualist S words, Skeptical Plowshares’, Philosophy and Phenomenological Research 62, 385–406. Lewis, D. 1994. ‘David Lewis: Reduction of Mind’, A Companion to the Philosophy of Mind, (eds.) S. Guttenplan, 412–431, Basil Blackwell, Oxford. — 1996. ‘Elusive Knowledge’, Australasian Journal of Philosophy 74, 549-567. Morton, A. & A. Karjalainen 2003. ‘Contrastive Knowledge’, Philosophical Explorations 6, 74–89. 16. I am grateful to Ray Elugardo and Jonathan Schaffer for very useful comments and criticisms on an earlier version of this article. I also want to thank Sherri Irvin for her feedback.

145

Neta, R. (for thcoming). ‘Undermining the Case for Contrastivism ’, Social Epistemology. Pritchard, D. 2005. Epistemic Luck, Oxford University Press, Oxford. Schaffer, J. 2004. ‘F rom Contextualism to Contrastivism ’, Philosophical Studies 119, 73–103. — 2005a. ‘Contrastive Knowledge’, Oxford Studies in Epistemology, vol. 1, (eds.) T.S. Gendler and J. Hawthorne, 235–271, Oxford University Press, Oxford. — 2005b. ‘ What Shifts? Thresholds, Standards, or Alternativ es’, Contextualism in Philosophy: Knowledge, Meaning, and Truth, (eds.) G. P reyer & G. P eter, 115–130, Clarendon Press, Oxford. — 2007. ‘Closure, Contrast, and Answer’, Philosophical Studies 133, 233–255. — (forthcoming). ‘Kno wing the Answ er’, Philosophy and P henomenological Research. Sinnott-Armstrong, W. 2004. Pyrrhonian Skepticism, Oxford University Press, Oxford. Sosa, E. 2002. ‘Tracking, Competence, and Knowledge’, The Oxford Handbook of Epistemology, (ed.) P. Moser, 264–286, Oxford University Press, Oxford. Unger, P. 1986. ‘The Cone Model of Knowledge’, Philosophical Topics 14, 125– 178. Vogel, J. 1999. ‘Relevant Alternatives Theor y’, Philosophical Perspectives 13, 155– 180.

146

Grazer Philosophische Studien 77 (2008), 147–190.

KNOWLEDGE-THE AND PROPOSITIONAL ATTITUDE ASCRIPTIONS Berit BROGAARD Australian National University Summary Determiner phrases embedded under a pr opositional attitude v erb have traditionally been taken to denote answ ers to implicit questions. F or example, ‘the capital of Vermont’ as it occurs in ‘John knows the capital of Vermont’ has been thought to denote the proposition which answers the implicit question ‘what is the capital of Vermont?’ Thus, where ‘know’ is treated as a propositional attitude verb rather than an acquaintance verb, ‘John knows the capital ofVermont’ is true iff John knows that Montpelier is the capital of Vermont. The traditional view lost its popularity long ago, because it was thought to r est on the controversial assumption that determiner phases embedded under a pr opositional attitude verb function semantically in the same way as the corr esponding wh-clauses. Here we defend the traditional assumption against objections. We then argue that wh-clauses are not to be giv en a uniform tr eatment as indir ect questions. When occurring under a propositional attitude verb, wh-clauses are better treated as having a pr edicate-type semantic v alue. We conclude b y considering some possible objections to the predicate view.

The most frequently cited examples of knowledge sentences in the philosophical literature are sentences of the form ‘s knows that p’ (e.g., ‘Jessica knows that Montpelier is the capital of Vermont’). However, few of the sentences we actually use to make knowledge attributions take this standard form. As examples of sentences that do not, consider: (1) (a) (b) (c) (d)

Jessica knows the capital of Vermont. John knows who knows the capital of Vermont. Tim knows a doctor who can treat your illness. Ralph knows where to fi nd a clinic that specializ es in or thopedic disorders.

(e) Mary knows every capital in Europe. (f ) Peter knows whether Mary knows every capital in Europe. The use of sentences like those in (1) diff ers from the use of that-clause sentences. Unlike the corresponding that-clause sentences, the sentences in (1) can be used to attribute to a subject knowledge which the attributor does not possess. For example, a correct assertion of (1)a does not require that the speaker kno ws that M ontpelier is the capital of Vermont. The use of such sentences to make kno wledge attributions thus puts fe wer demands on the speaker than the use of the corr esponding that-clause sentences. In this paper we off er an account of kno wledge sentences of the nonstandard form exemplified in (1) which reflects this difference. We argue that, unlike the more frequently discussed examples, sentences of the nonstandard form attribute a special kind of de er knowledge. The structure of the paper is as follows. In the first section we show that determiner-phrase complement clauses embedded under pr opositional attitude v erbs such as ‘know’ function semantically in the same way as the corr esponding wh-clauses. We then off er reasons against pr evious analyses of wh- and determiner-phrase complement clauses embedded under pr opositional attitude verbs. In the subsequent sections we defend the view that uses of sentences containing either kind of complement clause embedded under ‘know’ attribute a form of de re knowledge. Finally, we reply to some possible objections to the proposed analysis. 1. Knowledge-CQ vs. objectual knowledge Few would deny that there are knowledge attributions that do not attribute knowledge-that, for example, ‘Stephanie knows Kripke’.1 An utterance of the latter sentence attributes objectual knowledge to Stephanie: it is true iff Kripke is one of Stephanie’s personal acquaintances. Objectual knowledge attributions uncontroversially attribute non-pr opositional knowledge to the denotation of the subject term. O ne mark of the objectual ‘kno w’ is that it does not translate into the same word as the non-objectual ‘know’ in languages such as French, German and Italian. In German, for example, 1. Exceptions include Quine (1955), Fodor (1979, 319–28), Larson, Ludlow and den Dikken (1996) and Parsons (1997).

148

the objectual ‘kno w’ translates as ‘kennen ’, wher eas the non-objectual ‘know’ translates as ‘wissen’. If it could be argued that utterances of the sentences in (1) attributed objectual knowledge to the denotation of the subject term, ther e would be little controversy as to whether or not the sort of knowledge attributed is a form of kno wledge-that. But, the sentences in (1) admit of r eadings that do not attribute objectual kno wledge. (1)a, for example, does not state that Jessica is acquainted with the state of Vermont. This can be seen from the fact that the occurrence of ‘know’ in (1)a translates into a form of ‘wissen’ in German. Some knowledge sentences are ambiguous between a “kennen” (or objectual) r eading and a “ wissen” (or non-objectual) r eading. Consider, for instance: (2) (a) Stephanie knows the author of Naming and Necessity. (b) Tim knows a doctor who can treat your illness (= 1c). (c) Claire knows Jill’s favorite actress. (2)a can be read as saying that the author of Naming and Necessity is one of Stephanie’s personal acquaintances but it also has a eading r that requires for its tr uth that S tephanie know that Kripke is the author of Naming and Necessity. Likewise, (2)b can be r ead as saying that one of Tim’s personal acquaintances is a doctor who can tr eat the addressee’s illness, but (2)b also has a r eading that requires for its tr uth that Tim can identify a doctor who can tr eat the addr essee’s illness. F inally, (2)c can be r ead as saying that Jill’s favorite actress is one of Clair e’s personal acquaintances but it also has a r eading that requires for its truth that Claire have a way of picking out J ill’s favorite actress (e.g. b y naming her). When knowledge sentences exhibit an ambiguity of this sor t, the ‘know’ translates as ‘kennen’ (objectual), on the fi rst reading, and as ‘ wissen’, on the second reading. Here we shall be concerned only with knowledge sentences that translate using ‘ wissen’, that is, w e shall set aside objectual kno wledge attributions. If we set aside objectual kno wledge attributions w e are left with two sorts of knowledge sentences which are not explicitly a form of knowledgethat, viz. kno wledge sentences which contain a wh-complement clause (e.g., ‘what the capital of Vermont is’) and kno wledge sentences which contain a determiner-phrase complement (e.g. ‘ the capital of Vermont’). 149

Following the literatur e, let us call determiner-phrase complements, on their non-objectual readings, ‘concealed questions’ (CQ). There are then two kinds of kno wledge attributions which do not exhibit a that-clause structure: knowledge-wh and kno wledge-CQ. There is a thir d kind of knowledge attribution which is not explicitly a form of kno wledge-that, viz. knowledge-how, and we shall have something to say about it belo w. At this point, it will suffice to note that ‘how’ does not differ syntactically from words philosophers do not hesitate to call ‘wh-words’. It has been argued that concealed questions function semantically in the same way as the corr esponding wh-clauses (Baker 1968). The main evidence for this thesis comes fr om the fact that for any kno wledge-CQ sentence, it is possible to fi nd a kno wledge-wh sentence with the same meaning. Here are some examples: (3) (a) Jessica knows the capital of Vermont (= 1a). Jessica knows what the capital of Vermont is. (b) Tim knows the author of Naming and Necessity. Tim knows who the author of Naming and Necessity is. (c) Jacob knows the price of milk. Jacob knows what the price of milk is. If concealed questions function semantically in the same way as the corresponding wh-clauses (and ‘ho w’ is a wh-word), then ther e is just one kind of non-objectual knowledge sentence which is not explicitly a form of knowledge-that. There are several objections to this simple proposal. However, as we will see, none of them turns out to be fully successful. One common objection is directed against the proposal that concealed questions are constituents of wh-clauses at the lev el of logical form. If the syntactical v ersion of the disguised knowledge-wh approach is correct, then the surface forms given in (3)a–c ar e the r esult of deleting the underlined material in the paraphrases. But this proposal allegedly runs into difficulties. For, what holds for knowledge-wh sentences should hold for other sentences containing attitude or speech act verbs and a wh-complement clause. But the result of deleting the wh-word and the main verb in sentences containing verbs that take wh-clauses as complements is sometimes infelicitous. Consider , for instance:

150

(4) (a) John wondered who Smith’s murderer was. (b) Mary inquired what the capital of Vermont was. The sentences in (4) are perfectly acceptable but the result of deleting the wh-word and the main v erb phrase of the complement clause is infelicitous: (5) (a) #John wondered Smith’s murderer. (b) #Mary inquired the capital of Vermont. This suggests that concealed questions are not wh-clauses that have undergone deletion. I n response to this sor t of worr y Jane Grimshaw (1979) suggests that concealed questions are not constituents of wh-clauses at the level of logical form but determiner phrases that occur in ‘legitimate NPpositions’ (1979, 303). Grimshaw’s suggestion receives support from the fact that concealed questions occur as the subject terms in the follo wing sort of construction: (6) The number of moons of Jupiter is four. It is arguable that the definite description in (6) occurs in a legitimate NPposition and has a concealed question interpretation just like the complement clause in ‘John knows the number of moons of Jupiter’.2 If Grimshaw’s proposal is right, then we have a straightforward explanation of why ‘wonder’ and ‘inquire’ do not take concealed questions as complements. I f concealed questions occur in legitimate noun-phrase positions, we should not expect them to be selected b y verbs that do not combine with that-clauses (e.g. ‘wonder’ and ‘inquire’). Hence, we should expect (4)c and (4)d to be infelicitous. On Grimshaw’s proposal, concealed questions ar e not constituents of wh-clauses at the lev el of logical form. N onetheless, concealed questions embedded under v erbs that select for interr ogative complements (e.g. ‘know’) are semantically interpr eted as wh-clauses. The hypothesis that concealed questions ar e interpreted as wh-clauses explains why ‘believ e’ and ‘deny’, which do not takewh-clauses as complements, do not ordinar2. See Schlenker (2003) for a defense of this proposal and Brogaard (2007b) for a reply.

151

ily combine with concealed questions (witness ‘#John believes the capital of Vermont’). However, Grimshaw’s proposal has come under recent attack. As Lance Nathan (2006) has argued, if concealed questions function semantically in the same way as the corr esponding wh-clauses, then we should expect a wide range of interpr etations to be av ailable for concealed questions embedded under a complement-taking verb-phrase. But concealed questions embedded under such v erb phrases admit of a rather narr ow range of interpretations. For example, as (7) illustrates, ‘I told him the capital of Vermont’ cannot be interpr eted as ‘I told him wher e the capital of Vermont is’ but can only be interpreted as ‘I told him what the capital of Vermont is’: (7) (a) Jacob was wondering what the capital of Vermont is. I told him the capital of Vermont. (b) Jacob was wondering wher e the capital of Vermont is. I told him where the capital of Vermont is. (c) Jacob was wondering where the capital of Vermont is. #I told him the capital of Vermont. In (7)c the preceding linguistic context makes it salient that the information John was asking for is information about the location of the capital. So, if concealed questions were interpretable as wh-clauses, then a whereclause interpretation should be available for (7)c. But the discourse fragment in (7)c is infelicitous. There is, however, a straightforward reply to this line of argument. Suppose concealed questions and wh-complement clauses function semantically as predicates when embedded under propositional attitude verbs like ‘know’. Predicates denote sets in truth-functional contexts and properties in attitude contexts. S o ‘the capital of Vermont’ and ‘ what the capital of Vermont is’ both denote the pr operty of being (what) the capital of Vermont (is). S ince they co-denote, they ar e intersubstitutable. B ut ‘the capital of Vermont’ and ‘where the capital ofVermont is’ do not co-denote. The former denotes the property of being what the capital of Vermont is; the latter denotes the pr operty of being where the capital of Vermont is. Since the two complement clauses do not co-denote, they ar e not intersubstitutable. Hence, definite concealed questions are not interpretable as where-clauses. We will develop the hypothesis that concealed questions and 152

wh-complement clauses embedded under propositional attitude verbs like ‘know’ function as pr edicates in fur ther detail belo w. Before developing this proposal, however, let us offer a more detailed explanation of why not all verbs that select wh-complements take concealed questions. It will be noted that v erbs that take only interr ogative clauses like ‘wonder’ and ‘inquire’ and verbs that take both declarative and interrogative clauses combine diff erently with interrogative clauses. ‘Wonder’ and ‘inquire’ combine with direct reports of questions, as in ‘John wondered, “Who is the murderer of Smith?” ’ and ‘Mary inquired, “What is the capital of Vermont?” ’. ‘Know’, on the other hand, does not combine with direct reports of questions, as can be seen fr om the infelicity of ‘M ary kne w, “What is the capital of Vermont?”’. Moreover, interrogative clauses that occur as the complements of ‘kno w’ are interpreted differently from interrogative clauses that occur as complements of ‘wonder’ and ‘inquire’. ‘Mary knew what the capital of Vermont is’ is true only if Mary knows that Montpelier is the capital of Vermont. ‘Mary wonders what the capital of Vermont is’, on the other hand, is false if Mary knows that Montpelier is the capital of Vermont. This difference in the interpretation of wh-clauses calls for an explanation. Here is a tentative hypothesis which we will test below. We hypothesize that wh-clauses function either as r estrictors of quantifi ers or as indir ect questions. Verbs like ‘wonder’ and ‘inquire’ select complements that function as indirect questions, whereas verbs like ‘know’ select complements that function as restrictors of quantifiers (a similar hypothesis is set forth by Berman 1991). If the hypothesis just offered is correct, then it may well be that concealed questions function semantically in the same way as wh-clauses, which is to say, they function as r estrictors of quantifi ers at the lev el of logical form. As a r esult, concealed questions combine only with v erbs that select whcomplements which function as r estrictors of quantifi ers. Since ‘wonder’ and ‘inquire’ do not select wh-complements that function as restrictors of quantifiers, they do not select concealed questions as their complements. But a problem remains. As Nathan (2006, chap. 2) points out, the verb ‘ask’ seems to cause trouble for theories which take concealed question to be interpretable as wh-clauses (Nathan 2006). I n the ‘unsure that’ sense rather than in the ‘r equest’ sense ‘ask’ behaves like ‘wonder’ with respect to that-clauses and wh-clauses, that is, it selects wh-clauses but does not select that-clauses. Yet, as N athan notes, unlike ‘ wonder’, ‘ask’ selects concealed questions: 153

(8) (a) Councilmember Brown asked what the time frame for implementing these goals is. (b) Councilmember Brown asked the time frame for implementing these goals (c) #Brown asked that the time frame for implementing these goals is 2 years3 However, there is no cause for concern. Concealed questions ar e not widely allowed in the scope of ask’, ‘ 4 and when they are, they are somewhat idiomatic. We agree with Nathan (2006, 5.3.1) that concealed questions embedded under ‘ ask’ do not hav e the same meaning as concealed questions embedded under v erbs that select that-clause complements. Concealed questions embedded under ‘ask’ are indeed still interpretable as what- and who-clauses but unlike other concealed questions their meaning is a question (where a question is a set of possible propositional answers). Further, as a simple Google search will show [search terms: “* wondered the *”], even ‘wonder’ occasionally combine with concealed questions with a question meaning, as in: 5 (9) (a) (b) (c) (d)

Ever wondered the answer to any of these questions? Ever wondered the “right” way to describe swordfights? Stratford wondered the time line. If you’ve ever wondered the value of partner programs from software/hardware companies, let me give you my perspective from the front lines. (e) Geisenberger wondered the price range. (f ) Hopkinson wonder ed the height of the tallest building in this neighborhood. 3. With the subjunctive mode ‘be’ this is acceptable, but with the subjunctiv e mode, ‘ask’ means ‘request’, which is not the meaning we are interested in here. 4. Compare ‘John knows the student who kno ws the capital of Vermont’ vs. ‘J ohn asked the student who knows the capital of Vermont’. The noun-phrase complement may function as a concealed question only in the fi st case. 5. ‘I’ve often wondered the same thing’ seems to be a fur ther example. However, as Lance Nathan has pointed out to me, ‘ the same thing ’ is pr obably not a concealed question but an anaphoric determiner phrase (http://lemmingsblog.blogspot.com/2007/07/i-hav e-often-wondered-same-thing.html). Nathan also points out that the examples in (9) might not be genuinely acceptable.

154

As both ‘wonder’ and ‘ask’ take concealed questions with a question meaning in ordinary language (even if the result is somewhat idiomatic), I would hesitate to discard the position that concealed questions function semantically in the same way as the corresponding wh-clauses on the grounds that ‘ask’ differs from ‘wonder’.6 There is, ho wever, a fur ther objection to the pr oposal that concealed questions function in the same way as the corresponding wh-clauses. The objection is due to I rene Heim (1979), who attributes it to G reenberg, and it r uns as follo ws. ‘Found out’ is a v erb phrase that takes concealed questions as complement clauses. S o if concealed questions function in the same way as the corresponding wh-clauses, we should expect concealed question embedded under ‘ found out’ to be interpr etable as wh-clauses. But consider now the following two sentences: (10) (a) John found out the murderer of Smith. (b) John found out who the murderer of Smith is. If (10)a and (10)b were interpretable in the same way, then we shouldn’t expect there to be a difference in meaning. But there is an obvious difference in meaning. Heim explains: [(10)b] cannot only be used to expr ess that J ohn solved the question who murdered Smith, but has a further reading which is perfectly compatible with John’s being entirely ignorant of Smith’s murder, and which only amounts to the claim that John found out some essential fact or other (e.g. that he was his brother) about the person referred to as ‘the murderer of Smith’, but this is not an av ailable reading for [(10)a], which can only be used in the fi rstmentioned way. (1979, 53)

Heim’s point is this. If John found out James is his brother, and the speaker and hearer know that James is the murderer of Smith, then (10)b is acceptable but (10)a is not. J ohn didn’t fi nd out the murderer of Smith but he found out who the murderer of Smith is. Since (10)a and (10)b diff er in 6. ‘Care’ seems to be another counterexample to our hypothesis, as it selects pr opositional and wh-complements but not CQ-complements. However, as Nathan (2006, chap. 5) observes, arguments such as ‘M ary cares who left. I t is the case that J ohn left. H ence, Mary cares that John left’ are invalid. This seems to suggest that ther e are indeed two distinct pr edicates, one that selects wh-clauses and one that selects that-clauses. If he is right about this, then ‘ care’ is not a counterexample to our hypothesis.

155

meaning, the concealed question in (10)a cannot be interpr eted as ‘who the murderer of Smith is’. The argument is forceful. But it doesn’t prove what it claims to prove. Heim is certainly right that (10)a and (10)b do not have the same meaning. But there is no r eason to think that that undermines the position that concealed questions function in the same way as the corr esponding wh-clauses. For, it is arguable that the implicit copula and the defi nite description in (10)a are inverted: (10)c John found out who is the murderer of Smith. Like (10)a, but unlike (10)b, (10)c is true only if there is some x such that John found out x is the murderer of Smith. Thus, (10)c is false if oJ hn found out that James (who happens to be the murderer of Smith) is his brother. Indefinite concealed questions may also seem to cause tr ouble for the view that concealed questions function in the same way as the corr esponding wh-clauses. If indefinite concealed question were interpretable as whclauses, it might seem that we should expect the following two sentences to have the same meaning (see Frana 2006). (11) (a) John knows a doctor who can treat your illness. (b) John knows who is a doctor who can treat your illness. But (11)a and (11)b clearly do not have the same meaning. (11)b requires for its tr uth that J ohn know of ev eryone in the domain whether or not he or she is a doctor who can tr eat your illness. B ut (11)a r equires only that you know of a certain person that he or she is a doctor who can treat your illness. However, the argument does not undermine the vie w that concealed questions function semantically in the same way as the corresponding whclauses. All it sho ws is that indefi nite concealed questions do not always function in the same way as who- or what-clauses. But we already knew that. Consider the following discourse fragment: (12) A: I haven’t been able to find any Belgian chocolate. Do you know where I can get it? B: No, I don’t. But you should ask John. He knows a store that sells it. 156

B’s last remark can be read as ‘John knows where there is a store that sells Belgian chocolate’. That is, concealed questions of the form ‘an F’ sometimes have the same meaning as ‘where an F is’ or the equivalent but less idiosyncratic ‘where there is an F ’. 7 Thus, it is reasonable to suppose that (11)a has the same meaning as ‘J ohn knows where there is a doctor who can treat your illness’, and so diff ers semantically from (11)b.8 It seems, then, that there are no serious objections to the proposal that concealed questions function in the same way as the corr esponding whclauses. 2. Previous approaches Before off ering an account of wh-clauses embedded under pr opositional attitude verbs like ‘kno w’, it will be fr uitful to look at pr evious accounts of concealed questions. As we will see, none of the previous approaches is entirely successful. 2.1 The complements-as-questions approach On the standar d account, wh-clauses that occur as the complements of verbs that take boththat-clauses and wh-clauses are indirect questions, and concealed question are literally concealed questions.9 Following Higginbotham (1996, 381), knowledge-wh sentences may be assigned the following meta-linguistic truth-conditions:10 there is a proposition p such that s knows that p, and p answers the indirect question of the wh-clause. 7. I ndefinite concealed questions do not always hav e the same meaning as where-clauses. For example, it seems that ‘John knows a hotel that permits guests to keep pets’ may be true if John can name a relevant hotel (e.g., ‘Holiday Inn #767’) but doesn’t know where it is. 8. It is arguable that (11)a could be tr ue even if J ohn could only name a r elevant doctor (see the pr evious note). B ut the diff erence in meaning betw een (11)a and (11)b can then be explained on the assumption that (11)a has the same meaning as the slightly more idiosyncratic sentence ‘John knows who a doctor who can treat your illness is’. 9. For a defense of the disguised-questions appr oach to wh-complement clauses, see e.g. Hintikka (1975), Boër and L ycan (1986), Higginbotham (1996), Bach (2005), Braun (2006, manuscript), Kallestrup (forthcoming and manuscript). For a variation on this view, see Schaffer (2007). 10. The meta-linguistic truth-conditions are also sometimes called the ‘truth-maker truthconditions’. They specify what the world (or logical space) must be like for the sentence in question to be tr ue. But they do not specify which pr oposition is expr essed by the sentence. For example, ‘it is possible that there are blue swans’ can be given the following meta-linguistic

157

Knowledge-wh: s knows-wh iff there is a proposition p such that s knows p, and p answers the indirect question of the wh-clause. Meta-linguistic truth-conditions may be assigned to knowledge-CQ in a parallel fashion: Knowledge-CQ: s knows-CQ iff there is a pr oposition p such that s knows p, and p answ ers the concealed question of the noun phrase complement. ‘John knows (who) the author ofNaming and Necessity (is)’, for example, is true iff there is a proposition p such that John knows that p, and p answers the question ‘ who is the author of Naming and N ecessity? ’ S o, if J ohn knows that Kripke is the author of Naming and Necessity, then he knows the author of Naming and Necessity, in the relevant sense of ‘knows’. On this approach, wh-clauses and concealed questions function in the same way semantically: they denote true answers to the questions to which they correspond. Unfortunately, the disguised-questions appr oach is not empirically adequate. First, it is unable to distinguish semantically betw een inverted and non-inverted wh-clauses. Consider: (13) (a) John found out who (exactly) is the murderer of Smith. (b) John found out who (exactly) the murderer of Smith is (= 10b). The disguised-question appr oach predicts that (13)a and (13)b ar e true under the same circumstances. (13)a and (13)b are true iff there is some p such that John found out that p, and p answers the question ‘who (exactly) is the murderer of Smith?’ But, as we have seen, (13)a and (13)b do not have the same meaning. Unlike (13)a, (13)b may be true if John found out that James (who happens to be the murderer of Smith) is his brother. Second, the disguised-questions approach will deliver the wrong verdict in cases of the following sort. Suppose John mistakenly believes that Mary is leaving because she dislikes her depar tment, and suppose J ohn knows truth-conditions: ‘it is possible that ther e are blue swans ’ is tr ue iff there is a world in which there are blue swans. B ut if ‘it is possible that ’ functions as a sentential operator rather than as an object-language quantifi er over worlds, ‘it is possible that ther e are blue swans’ does not express the proposition that there is a world in which there are blue swans.

158

that Mary has accepted a job in California. If ‘Mary has accepted a job in California’ answers the question ‘what is Mary’s reason for leaving?’, then John knows Mary’s reason for leaving, in spite of the fact that he mistakenly believes she is leaving because she dislikes her depar tment. Or to take another case: suppose Alice kno ws that her student M ary left at 1 AM. Alice doesn’t know whether her students Bob and Carl also left at 1 AM. So, Alice doesn’t know which ones of her students left at 1 AM. Yet if ‘Mary left at 1 AM’ answers the question ‘which ones of Alice’s students left at 1 AM?’, then the standar d approach predicts that Alice knows which ones of her students left at 1 AM, despite the fact that she doesn’t know whether Bob and Carl also left at 1 AM. To avoid this problem, defenders of the disguised-questions appr oach are required to place restrictions on answers that may serve as the denotation of concealed questions. F or example, they ar e required to say that the only admissible answ ers to ‘ what is M ary’s reason for leaving? ’ ar e answers of the form ‘ r is Mary’s reason for leaving’. Since John does not know that her acceptance of a job in California is her r eason for leaving, John doesn’t know Mary’s reason for leaving. Likewise, they are required to say that the only admissible answers to ‘which ones of Alice’s students left at 1 AM?’ are answers of the form ‘X are the only ones among Alice’s students who left at 1 AM’. We might reformulate the disguised-questions approach as follows: S knows wh/CQ iff there is a p such that s knows that p, and p is an admissible true answer to the indirect/concealed question of the complement clause. However, without independent motiv ation, this account is ad hoc. To back up the account, it may perhaps be said that only full and exhaustive answers are answers in a technical sense. But this hypothesis is highly questionable. Except in the case of multiple choice questions, it is not clear that there is such a thing as a full and exhaustive answer (Groenendijk and Stokhof 1997). John knows that he can buy a can of cranberr y sauce in Family Delight. So, he knows where he can buy a can of cranberry sauce. But ‘John can buy a can of cranberr y sauce in F amily Delight, Country Farm, Stop One, and …’ (listing all the contextually r elevant places) is clearly more exhaustive than ‘J ohn can buy a can of cranberr y sauce in Family delight’. Precisely the same point can be made with respect to socalled open questions. John knows what natural-language semantics is but 159

there is no exhaustiv e answer to the question ‘ what is natural-language semantics?’ It remains a substantial question whether any motivation could possibly be given for imposing admissibility constraints on questions ex cept that it yields the correct verdict in the cases just considered. A third problem for the disguised-questions approach concerns iterated knowledge ascriptions.11 Consider, for instance: (14) John knows that Mary knows the capital of Vermont. (14) can be true even if John doesn’t know that Montpelier is the capital of Vermont. Yet this is not the result delivered by the disguised-questions approach. On the disguised-questions approach, ‘the capital of Vermont’ denotes the pr oposition Montpelier is the capital of Vermont. So, (14) is true iff (15) is true: (15) John knows that M ary knows that M ontpelier is the capital of Vermont. If John is a fairly competent speaker of E nglish, it is plausible that he knows whether certain basic inferences hold in English, for instance, the inference from ‘Mary knows that Montpelier is the capital of Vermont’ to ‘Montpelier is the capital of Vermont’. The following is then true: (16) John knows that ‘M ary knows that M ontpelier is the capital of Vermont’ entails ‘Montpelier is the capital of Vermont’. By Closure (if s knows p, and s knows that p entails q, then s knows q), (15) and (16) entail: 12 (17) John knows that Montpelier is the capital of Vermont. But, ex hypothesi, John doesn’t know this. 11. Brogaard (2007d) and Kallestr up (manuscript) raise r elated objections for what K ent Bach (2005) calls the ‘stupid view’. 12. Some prefer the following version of Closure: if s knows p, and s competently deduces q from p and thereby comes to believe q while retaining knowledge of p, then s knows q. If this is the preferred version, let us add that John competently deduces q from p and thereby comes to believe q while retaining (implicit) knowledge of p.

160

It may be objected that as Closure has been questioned on several occasions, defenders of the disguised-questions approach can simply reject it. This move has a considerable degr ee of initial plausibility . Fred Dretske and Robert Nozick familiarly disallow corollaries of Closure where p is an ordinary light-weight proposition, and q is the negation of a heavy-weight skeptical hypothesis (Dretske 1970 and Nozick 1981).13 For example, they disallow the infer ence from ‘I kno w I hav e hands’ to ‘I kno w I am not a disembodied brain in a v at’. Someone sympathetic to D retske/Nozick considerations might thus suggest that w e reject the instance of closur e utilized in the above argument. However, Dretske/Nozick considerations do not lend support to a rejection of the instance of Closure utilized in the above argument. For that instance does not license inferences from lightweight pr opositions to the negation of heavy-w eight skeptical hypotheses. It licenses inferences from ‘s knows that r knows that p’ to ‘s knows that p’. It may also be objected that the substitution of ‘that Montpelier is the capital of Vermont’ for ‘the capital of Vermont’ is illegitimate. Following the standard account, ‘the capital of Vermont’ denotes the proposition that Montpelier is the capital of Vermont but, it may be said, ‘that Montpelier is the capital of Vermont’ does not denote a proposition; following Frege, it denotes a truth-value. However, this reply is amiss. As ‘that Montpelier is the capital of Vermont’ occurs in an attitude context, it does not denote a tr uth-value. If it denoted a tr uth-value, its tr uth-value would matter to the tr uth-value of the whole. But the truth-values of sentences embedded under an attitude v erb do not matter to the tr uth-value of the whole (ex cept when the attitude v erb is factiv e).14 ‘Fermat’s last theor em is tr ue’ cannot be substituted for ‘2 + 2 = 4’ in ‘ almost everyone believes that 2 + 2 = 4’. When embedded under an attitude v erb, ‘2 + 2 = 4’ denotes the pr oposition that 2 + 2 = 4. Like wise, when embedded under an attitude v erb ‘that Montpelier is the capital of Vermont’ denotes the pr oposition that Montpelier is the capital of Vermont. So, when embedded under ‘know’, the that-clause ‘that Montpelier is the capital of Vermont’ and the concealed question ‘ the capital of Vermont’ denote the same pr oposition. 13. The terms ‘light-w eight proposition’ and ‘heavy-w eight skeptical hypothesis ’ are borrowed from Hawthorne (2005). 14. Of course, the tr uth-value of the that-clause matters when the attitude v erb is factiv e but the point still stands that the truth-value of the whole is not determined by the truth-value of the that-clause. Factivity is a property of the attitude verb.

161

As they denote the same pr oposition, they ar e intersubstitutable salv a veritate. It may also be urged that the substitution of ‘ that Mary knows that Montpelier is the capital of Vermont’ for ‘that Mary knows the capital of Vermont’ is illegitimate. However, it is difficult to see what could possibly be the cause of this sort of substitution failure. As we have just seen, when embedded under ‘know’, the that-clause ‘that Montpelier is the capital of Vermont’ and the concealed question ‘the capital of Vermont’ denote the same pr oposition. Moreover, blocking substitution is unlikely to help . For, if the standar d account does not generaliz e to iterated kno wledge ascriptions, then it does not off er a fully general account of concealed questions. Finally, the defender of the standard account may grant that the standard account is mistaken but insist that ther e is a modifi ed and mor e plausible form of it which tr eats the meta-linguistic tr uth-conditions for knowledge-the provided by the standar d account as object-language truth-conditions. Given this suggestion, ‘ s knows the F’ is to be tr eated as expressing the proposition that there is a p such that s knows p, and p (correctly) answers the concealed question of the complement clause. This move avoids the problems presented by iterated knowledge ascriptions. ‘Mary knows the capital of Vermont’ expresses the proposition that there is a p such that M ary knows that p, and p answers the question ‘what is the capital of Vermont?’ Similarly, ‘John knows that Mary knows the capital of Vermont’ expresses the pr oposition that J ohn knows that there is a p such that Mary knows p, and p answers the question ‘what is the capital of Vermont?’ From this we cannot infer that J ohn knows the capital of Vermont. Though I have no knockdown objection to this proposal I believe that there is some reason to resist it. The reason is that it violates compositionality. If we take ‘logical form’ to mean ‘underlying syntactical structure’ or ‘LF’ (in linguistic jargon), ‘s knows the F’ simply does not have the logical form ‘there is a p such that s knows p, and p answers the question “what is the F?” ’ But if not, then the (propositional) meaning of occurrences of ‘s knows the F ’ is not the result of assigning semantic values to the syntactic constituents of the sentence. O f course, someone sympathetic to unarticulated constituent theories of pr opositional attitude ascriptions might find appeals to strict compositionality principles rather insipid. B ut it is one thing to posit unarticulated constituents in the sentence structure of attitude ascriptions and quite another to say that ‘s knows the F’ reduces 162

to ‘there is a p such that s knows p, and p answers the question “what is the F ?”’. Unarticulated constituent theories posit a limited number of additional constituents. The proposal under consideration, on the other hand, takes knowledge-CQ sentences to be reducible to sentences whose grammatical structure does not even remotely resemble that of the original. We conclude that the disguised-questions appr oach does not off er a satisfactory account of concealed questions. 2.2 I ndividuals Definite concealed questions hav e the sur face form of r egular defi nite descriptions. It is ther efore tempting to think that concealed questions denote individuals. This is the pr oposal (cautiously) defended b y Heim (1979). O n her pr oposal, ‘the capital of Vermont’ denotes M ontpelier, ‘the author of Naming and Necessity’ denotes Saul Kripke, ‘the president of America’ denotes Bush, and so on. ‘John knows the capital of Vermont’ is not to be analyzed as ‘John knows Montpelier’, because concealed questions are then indistinguishable from objectual noun-phrase complements. Rather, the verb that embeds the concealed question is associated with a variable whose value is determined by linguistic or extra-linguistic context. ‘Know’ is to be understood as expr essing the three-place relation: knowx-as-y. ‘John knows the capital of Vermont’, for example, may expr ess, relative to context, the proposition expressed by ‘John knows Montpelier as the capital of Vermont’. While Heim’s proposal seems to fare better than the disguised-questions approach, it’s not entirely happy. As Maribel Romero (2005) and Lance Nathan (2006, chap. 3) have argued, if knowledge is a relation between a contextually determined guise and two individuals, then it becomes difficult to explain why ‘I know Mark Twain’ cannot be read as saying that I know Samuel Clemens as Mark Twain but can only be read objectually. Furthermore, as N athan points out (2006, chap . 3), if determiner phrases have the same denotation regardless of whether they are in a concealed-questions context or in a normal context, then w e should expect the following sentences to be equally acceptable: (18) (a) Jim told Kim, and Kim told her mother, the murderer of Smith. (b) Jim has seen pictures of, and Kim has actually visited, the murderer of Smith. 163

(c) ?Jim found out, and Kim visited, the murderer of Smith. (d) ?Kim visited, and Jim found out, the murderer of Smith. The occurrence of ‘the capital of Vermont’ in (18)a serves as the complement of both ‘J im told Kim ’ and ‘Kim told her mother ’. Like wise, the occurrence of ‘the capital of Vermont’ in (18)b serves as the object of both ‘Jim has seen pictur es of ’ and ‘Kim has actually visited ’. (18)a and 18)b are felicitous. H ence, if determiner phrases hav e the same denotation regardless of the context in which they occur , we should expect (18)c–d to be acceptable as well. But (18)c–d are rather stilted. Finally, the individuals approach yields the wrong result in contexts in which one and the same v erb embeds two complements. Consider , for instance: (19) Jacob knows a lot: he knows the capital of Vermont, what a concealed question is and ho w the world can suppor t the gr owing global population. (19) is felicitous. Yet if the occurr ence of ‘kno w’ is H eim’s concealedquestions ‘know’, then it follows that Jacob knows-as-Montpelier what a concealed question is and how the world can support the growing global population. But (19) clearly cannot be interpreted in this way. Given these challenges it seems fair to conclude that concealed questions do not denote individuals. 2.3 I ndividual Concepts The thesis that definite concealed questions denote individuals has much to r ecommend it. U nfortunately, it has cer tain fl aws which ar e better avoided. There is, ho wever, a pr oposal in the vicinity which, initially at least, seems to far e better than H eim’s individuals approach, namely the individual-concepts vie w (R omero 2005). O n the individual-concepts view, ‘s knows the F’ is true iff s is able to identify the referent of ‘the F’ at the world she occupies. For example, ‘John knows the capital of Vermont’ is true iff John is able to identify the r eferent of ‘the capital of Vermont’ at the world he occupies. This proposal has several advantages compared to the disguised-questions appr oach and the individuals appr oach. First, it appears to solv e the pr oblem of explaining the diff erence betw een ‘John found out the 164

murderer of Smith’ and ‘John found out who the mur derer of Smith is’ (see Heim 1979). The former r equires that J ohn be able to identify the referent of ‘the murderer of Smith’, whereas the latter r equires that John know p, where p answers the question ‘who is the murderer of Smith’? The proposal has the further advantage that it is able to explain the difference between objectual and non-objectual noun-phrase complements. O bjectual noun-phrase complements denote individuals, whereas non-objectual noun-phrase complements denote individual concepts. Third, it seems required anyway to explain the inv alidity of the follo wing w ell-known argument (due to Partee):15 The temperature is ninety. The temperature is rising. Therefore, ninety is rising. If the two occurr ences of ‘ the temperatur e’ hav e the same denotation, then the argument should be valid. But it is clearly invalid. It is arguable that the invalidity of the temperature puzzle owes to the fact that ‘rise’ is a predicate of individual concepts rather than a pr edicate of individuals. If the first occurrence of ‘the temperature’ denotes an individual, but the second occurrence denotes an individual concept, i.e. a function fr om <world, time> pairs to individuals, then the premises clearly do not entail the conclusion. These are all good prima facie reasons for endorsing the individual-concepts approach. However, upon fur ther scrutiny, the motiv ation for the individual-concepts approach is rather weak. First, as we have seen, ‘John found out the mur derer of S mith’ may be interpr eted as meaning that John found out who is the murderer of Smith. Second, as Nathan (2006) points out, if ‘the temperature’ in the second premise of the temperature puzzle must be interpreted as an individual concept, then numerous other complex expressions must be similarly interpreted, for instance, m ‘ y home’ as it occurs in (from Partee) (20) My home was once in Maryland, but now it’s in Los Angeles. If the ‘it’ goes proxy for ‘my home’, then (20) would seem to assert that my home moved from Maryland to Los Angeles but it didn’t literally move. So, 15. “R eflections of a Formal Semanticist” in Partee (2004).

165

one might argue, m ‘ y home’ must be interpreted as denoting an individual concept, a function from times to referents, which yields a different referent at different times. In fact, any noun phrase that combines felicitously with ‘rises’ or ‘changes’ (in the substitution sense) will have an individual concepts denotation, including ‘the capital of Vermont’, ‘the color of her hair’, ‘the live-in boyfriend’, ‘the mayor of Boston’, ‘the coach’, ‘the picture on Alice’s wall’ (see Nathan 2006, chap. 3 for further examples). This proliferation of individual concepts, however, seems implausible. And, as Peter Lasersohn (2005) has sho wn, there is no need to tr eat the meaning of most noun phrases as inher ently time- or world-dependent. Consider the following argument which David Dowty, Robert Wall and Stanley Peters (1981) attribute to Anil G upta (see also N athan 2006, chap. 3): Necessarily, the temperature is the price. The temperature is rising. Therefore, the price is rising. Unlike Partee’s temperature puzzle, this argument seems valid. But if the first occurrence of ‘the temperature’ denotes an individual, and the second occurrence denotes an individual concept, then it is invalid. The reason it is invalid is that ‘the temperature’ need not rigidly denote an individual concept. Suppose there are just two indices: t1 and t2. At t1 the temperature concept assigns 99 at t1 and 100 at t2, and the price concept assigns 99 at t1, and 98 at t2 but at t2 the temperature concept assigns 89 at t1 and 90 at t2, whereas the price concept assigns 91 at t1 and 90 at t2. At every index, the temperature is then identical to the price. But at t1 the temperature is rising, whereas the price is not. So, the premises are true but the conclusion false. Dowty, Wall and Peters (1981) argue that the problem goes away if we require that the second occurr ences of ‘ the temperature’ and ‘ the price’ rigidly denote the individual concepts they denote. H owever, there are several reasons to resist this move (Lasersohn 2005, Nathan 2006, chap. 3). One forceful objection against it turns on the fact that some determiner phrases are not rigid designators. Consider the following argument: Necessarily, John’s favorite color is Amy’s favorite color. John’s favorite color keeps changing. Therefore, Amy’s favorite color keeps changing. 166

If the second occurrences of ‘John’s favorite color’ and ‘Amy’s favorite color’ denote individual concepts (i.e. functions from times to colors), then the argument is invalid, contrary to what we should expect. For, the following scenario is plausible. A t t1 John’s favorite color concept assigns r ed at t1 and blue at t2, and Amy’s favorite color concept assigns red at t1 and t2. At t2, John’s favorite color concept assigns blue at t1 and red at t2, but Amy’s favorite color concept assigns red at t1 and t2. At each index, Amy’s favorite color is then identical to John’s favorite color. But at t1 John’s favorite color changes (red at one time, blue at the next), wher eas Amy’s favorite color is constant. S o, the premises are true and the conclusion false. Yet the rigidifi cation strategy suggested b y Dowty, Wall and P eters will not work in this case, as ‘J ohn’s favorite color’ and ‘Amy’s favorite color’ are non-rigid (see e.g. Soames 2002, 261). Lasersohn uses this sort of argument to motivate a view of definite descriptions as referring terms (in Strawson’s sense). For example, ‘the temperature’ takes the intension of ‘temperature’ and returns, relative to a time, an individual as its extension. Moreover, Lasersohn argues that predicates like ‘rise’ can take intensions as their arguments (‘rise ’ intensionalizes the standar d denotation). The appearance of an individual-concepts meaning of ‘ the temperature’ is thus built into the meaning of ‘rise ’. Since the intension of ‘temperature’ is constant across worlds, no rigidification maneuver is required to guarantee the validity of the Gupta-style arguments.16 All that is required is that predicates such as ‘rise’ take intensions as their arguments. Besides lacking in motiv ation the individual-concepts appr oach runs into diffi culties that giv e us r eason to r esist it. O ne diffi culty is that of accounting for all of the natural eradings of the following kind of sentence, due to Heim (1979): (20) John knows the price that Fred knows. (20) admits of two readings. On one reading, John knows the same price as Fred. So, if Fred knows that the price of milk is $3.74/gallon, then ohn J too knows that the price of milk is $3.74/gallon. On the second reading, John knows what the price that Fred knows is. So, if Fred knows the price of milk, then John knows that Fred knows the price of milk but he himself need not know what the price of milk is (in, say, US Dollars). 16. Ultimately, Lasersohn’s explanation will not do (N athan 2006, chap . 4), as or dinary quantifiers also intensionalize, as witnessed by ‘each month, every picture on Alice’s wall changes’, but Lasersohn’s point that some kind of intensionalization is required still stands.

167

The individual-concepts account can accommodate only the first reading. For, on the individual-concepts approach, John knows the price that Fred knows only if he is able to identify an individual (e.g., $3.84) as the referent of ‘the price that Fred knows’ at the world he occupies. B ut the second reading does not require any such ability. Here is another example to illustrate. Brown and Smith have been killed. Tom, Dick and Harry are working on the two cases. After a couple of w eeks Tom fi nds out who murdered Brown and tells Dick. Dick tells Harry that Tom solved one of two cases, and that he, Dick, knows which one it is. In these circumstances Harry may truly utter the following sentence: (21) Dick knows the murderer Tom knows. Dick is not in the same epistemic position now as he will be when he later finds out that it was E vil Eye who murdered Brown. Yet the individualconcepts approach predicts that (21) is tr ue only if D ick knows how to identify an individual (e.g. Evil Eye) as the referent of ‘the murderer Tom knows’. Some defenders of the individual-concepts approach have argued that one can avoid this difficulty by taking the meta-level individual concepts readings required to account for (20) and (21) to be deriv able from the meaning of a meta-level form of ‘know’ (see e.g., Romero 2005). That is not an implausible proposal but there are other reasons for rejecting the individual-concepts appr oach. If individual-concepts interpr etations can be accounted for along the lines suggested b y Lasersohn, then the individual-concepts r eading of kno wledge-CQ is deriv able from the meaning of ‘know’, that is, ‘know’ takes the intension of the noun-phrase complement as its argument. But as Nathan points out (2006, chap. 3), it then becomes mysterious why some determiner phrases that do admit of an “individual-concepts” interpretation cannot ser ve as the complement clauses of ‘know’. For example, when ‘the poster on Alice’s wall’ occurs in ‘the poster on Alice’s wall keeps changing’, it gets an individual concepts reading, as its intension is the argument of ‘keeps changing ’. But ‘John knows the poster on Alice’s wall’ can, in most circumstances, receive only an acquaintance r eading, that is, it can only be interpr eted as meaning that John is acquainted with the poster on Alice ’s wall (though it might receive a CQ reading in circumstances in which the speaker intends to say, for instance, that John knows that the poster on Alice’s wall is a Toulouse Lautrec). Another example: as ‘keeps changing ’ takes the intension of 168

‘Tim’s armchair’ as its argument in ‘ Tim’s armchair keeps changing ’, we know ‘Tim’s armchair’ is capable of pr oviding its intension as the argument of a predicate. So we should expect a CQ reading to be available for ‘Harry knows Tim’s armchair’. Yet ‘Harry knows Tim’s armchair’ can only be read as saying that Harry is familiar with Tim’s armchair. We conclude that the individual-concepts appr oach, while initially compelling, is not in the end viable. 2.4 The propositions view Having considered and rejected the above alternatives Nathan (2006, chap. 4) suggests that concealed questions denote propositions rather than questions or individual concepts. O n Nathan’s vie w, concealed questions of the form ‘the F ’ denote the unique (maximal) proposition p such that, for some x, p is the proposition that x is the F. For example, if Mary’s reason for leaving is that she dislikes her depar tment, then ‘M ary’s reason for leaving’ denotes the unique proposition ‘ “Mary dislikes her department” is her r eason for leaving ’. So, ‘John knows Mary’s reason for leaving ’ is true iff John knows that ‘Mary dislikes her department’ is her reason for leaving. Nathan’s proposal does away with many of the problems facing previous accounts. First, as Nathan does not take concealed questions to denote true answers, his proposal does not yield the result that John may know Mary’s reason for leaving even if he mistakenly believes she is leaving because she dislikes her depar tment. Second, because he takes concealed questions to denote propositions, he is able to account for the semantic diff erence between objectual and non-objectual knowledge. Third, unlike the simple version of the individual-concepts approach he is able to account for the ambiguity in ‘John knows the price Fred knows’. If Fred knows the price of milk, John knows the price Fred knows if John knows the proposition that Fred knows the price of milk. But the propositional account has one drawback. It runs into the same sort of diffi culties with r espect to iterated kno wledge ascriptions as the disguised-questions approach. Consider the following scenario: Scenario: John doesn’t know the capital of Vermont. But one of his class mates, Alice, does, and John knows that she does.

169

When interpreted against this background, (22) is true: (22) John knows that Alice knows the capital of Vermont. Enter Nathan’s proposal. On Nathan’s proposal, ‘the capital of Vermont’ denotes the unique (maximal) proposition p such that, for some x, p is the proposition that x is the capital of Vermont. That is, ‘the capital of Vermont’ denotes the proposition that Montpelier is the capital of Vermont. Thus, (22) is true iff (23) is. (23) John knows that Alice kno ws that M ontpelier is the capital of Vermont. Assuming that John knows that ‘Alice knows that Montpelier is the capital of Vermont’ entails ‘Montpelier is the capital of Vermont’, the following is true: (24) John knows that ‘Alice knows that M ontpelier is the capital of Vermont’ entails ‘Montpelier is the capital of Vermont’. From (23) and (24), w e can then infer (b y Closure: if s knows p, and s knows that p entails q, then s knows q): (25) John knows that Montpelier is the capital of Vermont. But, ex hypothesis, J ohn doesn’t know that M ontpelier is the capital of Vermont. How might Nathan respond to this objection? H e might turn to the possible rejoinders we considered and r ejected a couple of sections ago . Alternatively, he may r eject the general pr oposal that the denotation of concealed questions of the form ‘ the F’ is always the unique (maximal) proposition p such that for some x, p is the proposition that x is the F. For example, Nathan might insist that (25) ascribes to J ohn the kno wledge that for some x, Alice knows that x is the capital of Vermont. The main pr oblem with this suggestion is that it is ad hoc. N athan’s proposal makes the follo wing predictions. If Bill is the student who got an A on the exam, then ‘J ohn knows the student who got an A on the exam’ is true iff John knows that Bill is the student who got an A on the exam. Likewise, since M ontpelier is the capital of Vermont, then ‘Alice 170

knows the capital of Vermont’ is tr ue iff Alice knows that Montpelier is the capital of Vermont. But then, by standard rules of compositionality, Nathan’s proposal should predict that (22) is true iff John knows that Alice knows that Montpelier is the capital of Vermont. We conclude that while the propositional approach seems to be on the right track, it hasn’t gotten things exactly right. 3. wh -clauses as predicates We turn now to our positive proposal. For the remainder of this paper we will defend the hypothesis that concealed questions and wh-clauses that occur under pr opositional attitude v erbs like ‘kno w’ function semantically in the same way as fr ee (or “nominal”) relative clauses, that is, r elative clauses without an explicit antecedent noun phrase (for example, ‘I haven’t bought what I am going to w ear’ as opposed to ‘I haven’t bought the clothes I am going to wear’). There is good reason to think that wh-clauses that occur as free relatives function as restrictors of unpronounced antecedent quantifiers. Consider, for instance, the following sentences. (26) (a) I haven’t bought what I am going to wear at the wedding. (b) I cooked what I found in the almost empty fridge. (c) I love when you pay me an unannounced visit. Where ‘x’ is a gr oup variable, (26)a can be paraphrased ‘ for all x, if x is what I am going to w ear at the wedding, then I haven’t bought x’, (26)b can be paraphrased ‘ for all x, if x is what I found in the almost empty fridge, then I cookedx’, and (26)c can be paraphrased for ‘ all x, if x is when you arrive unannounced, then I love x’. As the wh-clauses in (26) occupy the restrictor position of the unpr onounced antecedent quantifi ers, they function as predicates which denote sets of things that satisfy the condition imposed by the wh-clause. For example, ‘what I am going to wear at the wedding’ denotes the maximal set of gr oups of things I am going to wear at the w edding, ‘what I found in the almost empty fridge ’ denotes the maximal set of gr oups of things I found in the almost empty fridge and ‘when you arrive unannounced’ denotes the maximal set of groups of situations that are unannounced arrivings by you. 171

As several thinkers have pointed out, free relatives appear to have universal rather than existential force (Jacobson 1995, Grosu 2003). But they sometimes receive existential force. Consider, for instance (from Sternefeld 2006): (27) Everyone who takes what does not belong to him is a thief. (27) cannot be r ead as saying that ev eryone who takes ev ery item that does not belong to him is a thief , but must be interpr eted as saying that everyone who takes some item that does not belong to him is a thief . So (27) is of the form ‘for all x and some y, if y is what does not belong to x, then x is a thief ’. Here is another example: (28) The pilgrims settled where water could be found. (28) cannot be read as saying that the pilgrims settled at every place where water could be found but must be read as saying that the pilgrims settled at a place where water could be found. So, (28) is of the form ‘for some x such that x is where water could be found, the pilgrims settled at x’. One explanation of the variable force of the antecedent quantifiers is that whclauses function as predicates which are bound by a quantifier that receives existential force. The universal force of the quantifier antecedents in (26) may be thought to derive from the nature of the wh-clause. For example, it is possible that the wh-clauses in (26) denote singleton sets containing a group that satisfi es the condition specifi ed by the wh-clause. 17 In some cases the universal force may be due in part to a restriction on the denotation of the wh-clause to a singleton set containing a contextually relevant group of items. 18 Now, it is tempting to think that wh-clauses function in the same way regardless of whether they occur as constituents of fr ee relatives or as the complement clauses of ‘know’. Wh-clauses then function as the restriction of a wide-scope existential quantifier. If the existential quantifier is a kind of existential closure, we should expect it to take wide scope with respect to ‘know’. We thus arrive at the following logical forms for knowledge-the, knowledge-a, and knowledge-wh (see Brogaard 2007d, forthcoming). 17. That is, it is arguable that ‘what-F ’ is of the form ‘[Ox (x = (Ly)what-F y)]’. 18. Alternatively, we might treat the phenomenon as a case of semantic quantifi er variability. On the latter, see Heim (1982).

172

Logical form (know-the F ): x(s knows(x is the F )) Logical form (know-an F ): x(s knows(x is an F )) Logical form (know-wh-F ): x(s knows(x is wh-F )). The following are instances of these schemas: Knowledge-the ‘Rosa knows Ted’s reasons for leaving’: For some r, Rosa knows that the rs are Ted’s reasons for leaving. ‘Dorothy knows the way to Kansas’: For some w, Dorothy knows that w is the way to Kansas. Knowledge-a ‘Rebecca knows a place that sells Italian newspapers’: For some l, Rebecca knows that l is a place that sells Italian newspapers. Knowledge-wh ‘John knows what Mary did at 3 p.m.’: for some e, John knows that e is what Mary did at 3 p.m. ‘Rebecca knows where one can buy an I talian newspaper’: for some l, Rebecca knows that l is where one can buy an Italian newspaper. ‘Simon knows who attended the lecture’: for some s, Simon knows that s is who attended the lecture. ‘Alice knows why John is upset’: for some r, Alice knows that r is why John is upset. ‘Ruth knows which ones of her students left’: for some xs, Ruth knows that the xs are which ones of her students (that) left. ‘Where one can buy an Italian newspaper’ denotes the property of being a place at which one can buy an Italian newspaper, ‘who attended the lecture’ denotes the property of being a person who attended the lecture, ‘why Mary is upset’ denotes the pr operty of being a r eason that M ary is upset, and so on. The domain of quantification is, of course, likely to be contextuallyestrictr ed. For example, if Mary fed the dog at 3 p .m. and smoked a cigarette at 3 p.m., but the cigarette-smoking is not salient in the context, then ohn J knows what Mary did at 3 p .m. iff John knows that M ary fed the dog at 3 p .m. The account just off ered does not straightfor wardly apply to kno wledge-whether, for where ‘e is what Mary did at 3 p.m.’ is well-formed, ‘p is 173

whether q or r’ is not. At first glance, it may seem that knowledge-whether is a different species. We can, however, offer the following paraphrase which is reasonably close to the ones alr eady off ered for the other instances of knowledge-wh (Brogaard forthcoming) s knows whether q or r: for some p, s knows that p is a true proposition identical to q or r. We have a unified proposal if we assume that whether-clauses of the form ‘whether q or r’ are predicates with the meaning ‘is a tr ue pr oposition identical to q or r’. One advantage of the present proposal is that it straightforwardly extends to pseudo-clefts (Brogaard forthcoming). Consider, for instance: 19 (29) (a) What Mary did at 3 p.m. was feed the dog. (c) Where he lives is on the other side of the ocean. (d) Who we thought he would end up marrying was Alisa Brown. (f ) That Mary left is why John is upset. (g) This is how you open a can with one hand. (h) Whether Mary gets a raise is none of your business. A plausible account of pseudo-clefts is the semantic appr oach defended by Pauline Jacobson (1994). Following Partee (1986),20 Jacobson suggests that the wh-clauses are best tr eated as pr edicates which denotes sets of actions (‘what Mary did’), sets of properties (‘what John is’), sets of times (‘when the talk is’), sets of locations (‘where the talk is’), etc., whereas the post-copular constituents of pseudo-clefts ar e treated as either pr operty designators or predicates. Consider, for instance: 21 (30) (a) What John is is admirable. (b) What John is is proud of himself. (c) What John is is boring. 19. 29(b)–(c) are from Partee (1986, 200). 20. See also Bach and Partee (1980) and Szabolsci (1987). 21. Clauses with a pr operty-denoting subject and an entity-denoting post-copular element are also kno wn as ‘ specificational clauses’. For discussion, see M ikkelsen (2004) and B rogaard (2007b).

174

The wh-clause ‘what John is’ functions as a pr edicate which denotes the set of (salient) properties John has. In (30)a ‘is admirable’ functions as a predicate. Since predicates cannot occur in argument position at the level of logical form, sentences with pr edicates in argument position at the level of surface form require for grammaticality existential closure (Heim 1982, Fara 2001, 2006, Brogaard 2007a). 22 (30)a thus says that a salient property of John has the property of being admirable. In (30)b ‘is pr oud of himself ’ does not function as a pr edicate. For, (30)b does not say that a salient property of John has the property proudof-self. Rather, it says that a salient pr operty of J ohn is identical to the property proud-of-self. So, ‘proud of himself ’ undergoes a type-lowering from predicate type (type <e, t>) to individuals type (type e). The same sort of type-shifting is known to take place in the case of ‘J ohn’s favorite color is red’, which cannot be interpreted as meaning that John’s favorite color has the property red, but only as meaning that John’s favorite color is identical to the pr operty red. So, ‘red’ here shifts fr om predicate type (type <e, t>) to the type of individuals (type e). S omething similar goes on in the case of ‘M ary’s dress is the color of the sky ’, which cannot be interpreted as meaning that Mary’s dress is identical to the color of the sky but only as meaning that M ary’s dress has the pr operty denoted b y ‘the color of the sky’. So, ‘the color of the sky’ type-shifts from quantifier-type (type <<e, t>, t>) to predicate type (type <e, t>). (30)c is ambiguous betw een the two r eadings ex emplified b y (30)a and (30)b. On one r eading, it says that a salient pr operty of J ohn (e.g. the property of being prudent) has the property of being boring. On the second reading, it says that a salient pr operty of John is identical to the property boring. So, on the second r eading, ‘boring’ type-shifts from the predicate type (type <e, t>) to the individuals type (type e), and ‘is’ typeshifts from the ‘is’ of predication to the ‘is’ of identity. On Jacobson’s proposal, wh-clauses with embedded quantifiers receive a functional interpretation.23 Consider, for instance: 22. Jacobson (1994) argues that pseudo-clefts ar e inverted. For example, ‘ what John is is proud of himself’ has the same logical form as t‘he proud-of-self property is what John is’. Ther e is then no need for existential closur e. But this approach is unmotivated, as existential closure is required for ‘what John is is what Mary is’. 23. This idea has been dev eloped in further detail in e.g. J acobson (1994), Sharvit (1999) and Brogaard (2007b). It rests on a variable-free approach to binding proposed in Quine (1966). Jacobson’s goal is to develop a general variable-free approach to binding. That is not my intention. I am simply using Jacobson’s idea to account for the appearance of binding in sentences with a clear functional interpretation, such as ‘the person every student admires is his mother’.

175

(31) What every freshman admires most is her father. Following Jacobson, ‘what every freshman admires most’ can be treated as a predicate which denotes the set of functions f such that f takes individuals in the domain as input and yields individuals every freshman admires most as output. ‘H er father’ is a function-designating expr ession which denotes the father-of function (on female individuals if ‘her’ is gendered). So, (31) says that a function that takes individuals in the domain as input and yields individuals every freshman admires most as output is the fatherof function (on female individuals). If w e adopt J acobson’s treatment of wh-clauses in pseudo-clefts and a variation on the standar d treatment of fr ee relatives, we have a unifi ed treatment of wh-clauses in pseudo-clefts, as complement clauses embedded under verbs like ‘know’ and as free relatives. In the remainder of this paper we will develop this proposal in further detail and respond to objections. 4. De re knowledge On the predicate view, many things that puzzled us in this paper fall into place. First, as we have seen, knowledge-CQ and knowledge-wh are special kinds of de r e knowledge. s knows wh-F iff s knows that x (for some x) is wh-F. For example, John knows what Mary’s reason for leaving is iff John knows that r (for some r) is what Mary’s reason for leaving is. As it stands, the disguised-questions approach does not prove to be a particular promising way of making good on that vie w. The predicate view, on the other hand, automatically delivers this result. On the predicate view, ‘what Mary’s reason for leaving is’ denotes the property of being a (salient) reason for Mary’s leaving. To know Mary’s reason for leaving is to kno w that r (for some reason r) has that property. Second, unlike the view that concealed questions denote individuals, the present proposal is able to explain why only determiner phrases occur as concealed questions. On the present proposal, concealed questions function as predicates. Names, on the other hand, do not function as pr edicates. Third, unlike the disguised-questions and the pr opositional view, the predicate view is able to off er an account of iterated kno wledge ascriptions. On the predicate view, ‘John knows that Mary knows the capital of Vermont’ has the underlying logical form: ‘J ohn knows that for some y, Mary knows that y is the capital of Vermont’. 176

A similar approach can be extended to account for Heim’s price-of-milk case. ‘John knows the price F red knows’ may be interpr eted as meaning that John knows a particular price known to Fred (e.g., the price of milk). On this interpretation, ‘John knows the price Fred knows’ does not entail that there is x such that John knows that $ x is the price of milk. G iven the predicate view, it is a simple matter to account for this reading. ‘John knows the (salient) price Fred knows’ is true iff John knows that there is a y such that F red knows that y is the (salient) price. This is tr ue if, for example, John has a knowledge mental state with the content of ‘there is a y such that Fred knows that y is the price of a gallon of milk’. 24 Despite its clear advantages to previous approaches, the predicate view runs into tr ouble. On the pr esent analysis, kno wledge-CQ and kno wledge-wh are special cases of de r e knowledge. It is sometimes thought that even in the scope sense, de r e knowledge requires a special kind of acquaintance relation between the subject and the entity the subject is said to have knowledge of. For example, it is sometimes thought that on its de re reading, ‘John knows that Smith’s murderer is insane’ entails that John knows of a particular person—whom John could identify demonstratively —that he is insane. That is not quite right (see e.g. Kripke 1977, Kaplan 1978, L udlow and Neale 1991). 25 The de r e/de dicto distinction (in its scope sense) is syntactic, not psychological. To say that Smith’s murderer is such that John knows he is insane does not imply that John stands in any kind of direct acquaintance r elation to S mith’s murderer. The de r e r eading of ‘J ohn knows Smith’s murderer is insane’ is preferred just when for some person x thought to be Smith’s murderer by the conversationalists, John knows x is insane. Thus, if the conversationalists think Bill is Smith’s murderer and think John knows Bill is insane, then it is appropriate for one of them to assert that John knows Smith’s murderer is insane. The same considerations carry over to kno wledge-CQ and kno wledge-wh. Kno wledge-CQ and knowledge-wh are de re syntactically, not psychologically. But all is not well. The predicate view seems to yield the wr ong result in the following sort of case (Boër and Lycan 1986, Sterelny 1988): (32) Alice, you don’t know who you are: you’re the rightful heir to the Swedish throne. 24. On the notion of a knowledge mental state, see Williamson (2000, intro). 25. The cited articles are mostly concerned with the attributiv e/referential distinction but precisely the same points can be made with respect to the de re/de dicto distinction.

177

If the pr edicate view is corr ect, then the sentence ‘ you don’t know who you are’ has the following logical form: (32)a There is an x such that Alice doesn’t know that x is who Alice is. But unlike (32), (32)a would seem to equire r for its truth that Alice doesn’t know that she is Alice. In such cases a version of the disguised-questions approach would seem to do better. On the disguised-questions approach, (32) is true iff there is a p such that Alice doesn’t know p, and p answers the question ‘who is the rightful heir to the S wedish throne?’ So on the pr opositional view, (32) is true iff Alice doesn’t know that she is the rightful heir to the S wedish throne. One way out of this pr oblem would be to argue that (32)a is the right interpretation of (32) but that the quantifi er in (32) ranges o ver individual concepts rather than individuals. I f ‘who Alice is’ denotes the property of being an individual concept of who Alice is, v erifiers of (32) will then be individual concepts. (32), then, is tr ue iff there is an x (e.g. the concept of being the heir to the w S edish throne) such that Alice doesn’t know that it has the pr operty of being who Alice is. B ut the individualconcepts interpretation of wh-clause complements is not independently motivated. As it turns out, ho wever, the problem posed by (32) is an instance of a more general pr oblem of de r e attitude ascriptions (B rogaard 2007d). Consider the following variation on Kaplan’s S.O.B. case (1973, 555, note 71).26 Suppose detective Brown just disco vered that the infamous N ew Jersey carjacker is identical to the less well known New Jersey pickpocket. One afternoon Brown overhears John say to his pal ‘the person who stole my wallet just sent me a letter from Newton’. Upon his return to the office Brown says to his colleague: (33) Remember the New Jersey carjacker? Well, I just met a guy who thinks that the S.O.B. sent him a letter from Newton. Suppose the New Jersey carjacker (alias the S.O.B.) is the person who stole John’s wallet. In that case, Brown’s utterance could be true. But this is not what the Russellian approach gives us. As John does not have a belief 26. Kaplan’s original example was ‘John thinks the S.O.B. who took my car is honest’.

178

with the content ‘the S.O.B. sent me a letter fr om Newton’, he does not have a general belief with the content of the that-clause of (33). S o, we cannot assign narr ow scope to the description ‘ the S.O.B.’. On a widescope reading, the second sentence in (33) is of the form ‘ the S.O.B. is an x such that I just met a guy who thinks that x sent him a letter fr om Newton’. This is true only if John has a belief with the content of ‘ x sent me a letter from Newton’ for the assignment of an individual (who happens to be the New Jersey carjacker) to ‘x’. That is, it is true only if John has a singular belief dir ectly about the N ew Jersey carjacker (alias the S.O.B.) to the eff ect that he or she just sent him a letter fr om Newton. But John does not hav e a belief dir ectly about the N ew Jersey carjacker in the envisaged cir cumstances. He has a general belief with the content of ‘the person who stole my wallet (whoev er that may be) sent me a letter from Newton’, that is, he has a belief that is only indirectly about the New Jersey carjacker. Here is another case, a variation on Kripke’s (1977) Hoover case.27 After a press conference at which Henry Kissinger, Nixon’s impending security advisor, is mentioned by Hoover, Hoover tells his assistant: (34) The Berrigans believed that their accomplices would kidnap the official I mentioned at the press conference. Intuitively, (34) could be true if the Berrigans had guesstimated that their accomplices would kidnap the impending pr esident’s security advisor (whoever that would turn out to be), and so did not have a belief directly about Kissinger. Yet on the standar d appr oach to attitude ascriptions, (34) can be tr ue only if the B errigans had a belief with the content of the that-clause or a belief dir ectly about Kissinger to the eff ect that their accomplices would kidnap him. So, on the standard approach, (34) is false in the envisaged circumstances. The problem arises only on the assumption that attitude ascriptions, if correct, specify the Russellian content of a mental state. This assumption is normally r egarded as uncontr oversial (see e.g. Richar d 1990: chaps. 3 and 4). N eo-Russellians, for example, take the belief r elation to be a two-place relation between a subject and a R ussellian proposition,28 and 27. Kripke’s original example was ‘H oover charged that the B errigans plotted to kidnap a high American official’. 28. See e.g. M cKay (1981, 1991), S almon (1986), S oames (1988, 1995), B raun (1998, 2000), and Nelson (2002, 2005).

179

hidden indexicalists take it to be a three-place relation among a subject, a Russellian proposition and a guise.29 But, as the above examples illustrate, attitude ascriptions need not specify the R ussellian content of a mental state ev en if corr ect. While an utterance of the sentence ‘ the B errigans believed their accomplices would kidnap the offi cial I mentioned at the press conference’ does not r equire for its tr uth that the B errigans had a belief with the content of thethat-clause or a singular belief directly about the mentioned official, it does require that they had a belief which isabout the individual the r eport is about in some sense y et to be specifi ed. On the Russellian account, a belief is about an individual only if the content of the belief is singular . For example, a belief is about H enry Kissinger only if its content is a R ussellian proposition containing Kissinger . But this notion of aboutness is too strong. To arrive at a w eaker notion, let us intr oduce what w e shall call the ‘structured extension of a sentence’. The structured extension of a sentence, as we envisage it, is a composite of the extensions (denotations, referents) of its syntactic constituents. The structured extension of ‘John is male’, for example, consists of John and the set of men, and the structured extension of ‘the author of Naming and Necessity is male’ consists of Saul Kripke and the set of men. 30 Given the notion of str uctured extension, w e can off er the follo wing account of attitude reports (‘t’ is a name or quantifi er phrase) (Brogaard 2007d). ‘s believes that t is F ’ is true iff s has a belief with an t‘ is F ’-appropriate Fregean component and with the structured extension of a sentence which has ‘t is F ’ as an obviously relevant and necessary consequence. This account integrates the plausible idea that belief is closed under obviously relevant and necessary consequence (see Brogaard 2007d). There are several ways to arriv e at these tr uth-conditions. One way is to take ‘believe’ to express a three-place relation among a subject, a guise and a structured extension. Another way is to go two-dimensional. Epistemic two-dimensionalists think every expression has two different kinds of meaning: a non-F regean and a F regean meaning (see e.g. Chalmers 29. See e.g. Schiff er (1977, 1992, 1996), C rimmins and Perry (1989), Crimmins (1992), Recanati (1993), Jaszczolt (1999, 2000). 30. This is a simplifi cation. Strictly speaking, the str uctured extension of ‘J ohn is male ’ consists of the set containing the r eferent(s) of ‘John’, the set of males, and a r elation between them, and the str uctured extension of ‘the author of Naming and Necessity is male’ consists of two sets (viz. the set of authors of Naming and N ecessity and the set of males) and a r elation between them. I n both cases the r elevant relation (R) can be defi ned as follo ws: Rxy iff x has exactly one member, and x is a subset of y.

180

1996, 2002, 2006, forthcoming, for discussion see also Brogaard 2007c). On standard two-dimensional accounts, the two kinds of meaning ar e a Russellian intension (a function from worlds considered counterfactually to extensions) and a Fregean intension (a function from worlds considered as actual to extensions). The extension of an expression at a world can be derived from either kind of meaning by keeping the world constant. And the str uctured extension of a complex expr ession can be deriv ed fr om the extensions of its syntactic constituents. S o, if ev ery expression has a Russellian intension and a F regean intension, then arguably it also has a structured extension. Given a two-dimensional framework, we can say that ‘believe’ expresses two kinds of relations: a relation between a subject and a structured extension, and a relation between a subject and a Fregean sense. The main difference betw een the hidden-indexical v ersion and the two-dimensional version of the extensional view is this. For the hidden-indexicalist, ‘believe’ ascribes a three-place relation among a subject, a structured extension and a Fregean sense. For the two-dimensionalist, ‘believe’ expresses two distinct two-place relations: a relation between a subject and a structured extension, and a relation between a subject and a Fregean sense. Given the extensional account of attitude ascriptions w e can handle the problematic cases as follows. As ‘the person who found John’s wallet’, and ‘the New Jersey carjacker’ denote the same person in the S.O.B. case, they have the same structured extension. So, if John has a belief with the exact content of ‘the person who found my wallet sent me a letter fr om Newton’, it may be corr ect to say that J ohn believes that the New Jersey carjacker sent him a letter from Newton. Similar r emarks carr y o ver to the H oover case. As ‘ the impending president’s security advisor’, and ‘the official Hoover mentioned at the press conference’ both denote Henry Kissinger in our envisaged circumstances, they have the same structured extension. So, if the Berrigans had a belief with the exact content of ‘ our accomplices will kidnap the impending president’s security advisor’, it may be correct for Hoover to say ‘the Berrigans believed that their accomplices would kidnap the official I mentioned at the press conference’. Finally, let us consider our pr oblematic knowledge-who case. I n the envisaged circumstances, ‘Alice: you don’t know who you are’ cashes out to ‘there is an x such that Alice doesn’t know that x is who Alice is’. ‘The rightful heir to the Swedish throne is who Alice is’ is an obviously relevant and necessar y consequence of ‘ Alice is the rightful heir to the S wedish 181

throne’. Moreover, ‘Alice’ and ‘the rightful heir to the Swedish throne’ have the same structured extensions in the envisaged circumstances. So if Alice doesn’t know she is the rightful heir to the Swedish throne, then she fails to know something with the str uctured extension of a sentence that has ‘Alice is who Alice is’ as an obviously relevant and necessary consequence. So, ‘there is an x such that Alice fails to know that x is who she is’ is true in virtue of Alice’s lack of the knowledge that she is not the rightful heir to the Swedish throne. 5. Finding out who We still need to say something about the follo wing problematic pair of sentences: (35) (a) John found out who the murderer of Smith is. (b) John found out who is the murderer of Smith. Recall that the disguised-questions appr oach runs into trouble with this pair of sentences because it treats the embedded wh-clauses in exactly the same way. (35)a and (35)b ar e true iff there is a p such that J ohn found out that p, and p answers the question ‘ who is the mur derer of S mith?’ Yet, intuitively, (35)a and (35)b have different truth-conditions. (35)a may be true if John found out that James (who happens to be the murderer of Smith) is his brother but (35)b is not true under these circumstances. The difference between (35)a and (35)b, we will argue, is that the subject term of the wh-clause is ‘the murderer of Smith’ in (35)a but ‘who’ in (35)b. In fact, (35)b is acceptable only when who’ ‘ is focused (for instance, by adding stress, or by inserting an adverb like ‘exactly’, as in ‘who, exactly, is the murderer of Smith’). In (35)b ‘is the murderer of Smith’ functions as a genuine predicate and not as a device to single out an individual. This means that (35)b is acceptable only when it is r ead as equivalent to: (36) John found out who murdered Smith Unlike (35)a, (36) isn’t true if John found out that J ames is his br other. The reason (35)a admits of this reading is that quantification into attitude context is possible. (35)a may be read as saying: 182

(35)c The murderer of Smith is an x such that John found out who x is. (35)b, on the other hand, may not be read as saying: (35)d The murderer of Smith is an x such that John found out who is x. The reason the wide-scope r eading is not av ailable for (35)b is that the occurrence of ‘the murderer of Smith’ occurs in predicate position. When a description occurs in pr edicate position, it does not hav e its or dinary scope-taking properties (Fara 2001, B rogaard 2007a). For example, it is unable to take wide scope with respect to other operators. The lesson to be learned from (35)a and (35)b is a familiar one. A ttitude ascriptions which quantify into attitude contexts need not specify ho w the subject believes what she does. Whereas the occurrences of ‘the murderer of Smith’ in (35)a describes in more or less precise terms one of the ways in which J ohn thinks of the mur derer of S mith, the occurr ence of ‘ the murderer of Smith’ in (35)b does not pick out J ohn’s way of thinking of the murderer of Smith. We thus have further evidence for the hypothesis that de er belief reports in the syntactic sense are to be distinguished from de re belief reports in the psychological sense. As de re belief reports in the syntactic sense purpor t to be partial descriptions of what someone believes, they need not involve any singular or direct thoughts about an object. 6. Know-How Before concluding we should say something about how to extend the proposed analysis of knowledge-wh to knowledge-how. As mentioned above, there is good reason to think that knowledge-how is a kind of knowledgewh. First, semantically speaking ‘how’ just is a wh-word (linguists will give you the blank stare if you say it is not). Second, it is a fl uke that ‘how’ is not explicitly a wh-word in English. In many other languages (e.g. G erman) ‘how’ translates into the equivalent of a wh-word. If we extend the abo ve analysis for kno wledge-wh to kno wledge-how we end up with the following logical form:

183

s knows how to F: for some w, s knows that w is how to F. For example, Amy kno ws how diffi cult this task is iff for some x, Amy knows that x is how difficult this task is. This proposal is remarkably similar to that offered by Stanley and Williamson (2001). According to them, ‘s knows how to F ’ is tr ue iff for some contextually r elevant way w which is a way for s to F, s knows that w is a way for her to F. If we require that the way w be contextually salient and allo w some instances of kno wledge-how to be about degr ees rather than ways, the S tanley/Williamson proposal reduces to the proposal we have just off ered. And we now have an explanation of why this proposal is correct. It is correct, because ‘how to F ’ functions semantically as a predicate. For example, ‘how to curse out someone in French’ functions as a pr edicate which denotes the pr operty of being a way of cursing out someone in French, and ‘how to open a can with one hand ’ functions as a pr edicate which denotes the pr operty of being a way to open a can with one hand. In philosophy, however, there has been a lot of reluctance toward taking knowledge-how to be a kind of knowledge-that. It has been noted by several authors that kno wing-how need not entail possessing an ability (see e.g. Stanley and Williamson 2001, Bengson and Moffett 2007). For example, the Olympic figure skater Irina Slutskaya knows how to perform a quintuple salchow, but she cannot per form one herself (the example is from Bengson and Moffett). But some knowledge-how attributions seem to entail the possession of an ability. That seems to threaten the account of knowledge-wh offered in this paper. Suppose I have never practiced playing the piano but that I hav e taken numer ous theory lessons. There is then an x such that I know x is how to play the piano. Still, it would seem that someone could correctly claim that I don’t know how to play the piano . Likewise, if Mary—a mono-lingual speaker of English—sees Danny curse out his cousin in I talian, she might corr ectly say [while pointing] ‘ that is how to curse out someone in I talian’. Yet someone could corr ectly say ‘Mary doesn’t know how to curse out someone in Italian’. After all, Mary doesn’t even speak Italian. There is, however, a straightforward reply to these sorts of objections (for details see S tanley and Williamson 2001—their appr oach varies slightly from the one taken here).31 Unlike knowledge-how sentences that do not embed infinitive clauses, knowledge-how ascriptions that do embed infini31. For example, they invoke the notion of a practical guise.

184

tives are ambiguous between a reading that requires that the subject possess an ability (first-person) and a reading that does not require that the subject possess an ability (thir d-person). For example, ‘John knows how to play the piano’ may be r ead as saying that J ohn knows how JOHN may play the piano, or as saying that J ohn knows how ONE may play the piano . So, on the analysis off ered here, ‘John knows how to play the piano’ can be read as saying that ther e is a w such that J ohn knows that w is ho w John may play the piano or as saying that ther e is a w such that J ohn knows that w is how one may play the piano. If John has never practiced playing the piano, it is false that ther e is a w such that J ohn knows that w is how John may play the piano’ but it may well be true that there is a w such that John knows that w is how one may play the piano. So, ‘John knows how to play the piano’, is false when given the fi rst reading but it may well be true when given the second reading. Likewise, if Mary doesn’t speak Italian, then it will be false that there is a w such that Mary knows that w is how Mary may curse out someone in Italian but it may be true that there is a w such that Mary knows that w is how one may curse out someone in Italian. Unlike knowledge-how sentences with infinitive clauses, knowledge-how sentences without infinitive clauses are not ambiguous between a first-person and a third-person reading. For example, ‘John knows how Mary got home’, ‘John knows how the sandwich ended up in the refrigerator’, and ‘John knows how diffi cult this task is ’ do not hav e readings that require that John possess an ability to do any particular thing. 7. Conclusion Knowledge attribution sentences such as ‘John knows the capital of Vermont’ and ‘M ary knows the price of milk ’ diff er from standard knowledge-that attribution sentences such as ‘John knows that Montpelier is the capital of Vermont’ by having a determiner phrase as complement. S uch knowledge attribution sentences are fairly puzzling. It might be thought that we can simply interpret ‘John knows the capital of Vermont’ as shorthand for ‘John knows that Montpelier is the capital of Vermont’ but this proposal runs into trouble with examples such as ‘Elisa knows that Mary knows the capital ofVermont’. The latter is not shorthand for ‘Elisa knows that Mary knows that Montpelier is the capital of Vermont’. For the latter entails that Elisa knows that Montpelier is the capital of Vermont; yet it 185

may be that Elisa knows that Mary knows the capital of Vermont, even if she doesn’t know that Montpelier is the capital of Vermont. We argued that concealed questions function in the same way semantically as the corresponding wh-clauses, and that wh-clauses are to be interpreted as predicates. ‘John knows the capital of Vermont’ has the logical form: there is an x such that John knows that x is what the capital of Vermont is. We further argued that these wide-scope attitude ascriptions are best dealt with on a two-dimensional semantics that treats mental states as relations to structured extensions and Fregean senses. A two-dimensional semantics is independently motiv ated, as it is r equired to solv e cer tain puzzles about de re attitude ascriptions.32

REFERENCES Bach, K. 2005. “Q uestions and Answ ers”, comments on J onathan Schaff er’s “Knowing the Answer”, Bellingham Summer Philosophy Conference, August, 2005. Baker, C. L. 1968. Indirect Questions in English, Doctoral Dissertation, University of Illinois. Bengson, J. and M. Moffett. 2007. “Know-How and Concept-Possession”, Philosophical Studies 136, 31–57. Berman, S. 1991. The Semantics of O pen Sentences, Doctoral Dissertation, University of Massachusetts, Amherst. Braun, D. 1998. “Understanding Belief Reports”, The Philosophical Review 107, 555–595. — 2000. “Russellianism and Psychological Generalizations”, Noûs 34, 203–236. — 2006. “Now You Know Who Hong Oak Yun Is”, Philosophical Issues 16, E. Sosa and E. Villanueva, ed., 24–42. — Manuscript. “Implicating Questions”. 32. This paper has benefited from correspondence with Barbara Abbott, Kent Bach, David Braun, David Chalmers, Jim Higginbotham, Peter Lasersohn, Peter Ludlow, Ernie Lepore, Matt McGrath, Lance Nathan, Ram Neta, Duncan Pritchard, Joe Salerno, Jonathan Schaffer, Barry Schein, Philippe Schlenker, Ernie Sosa, Jason Stanley, Adam Taylor, and Kai v on Fintel, and comments provided by Maria Aloni, Kent Bach, Mike Barber, Scott Berman, Jim Bohman, Julien Dutant, Paul Egré, Pascal Engel, John Greco, Franck Lihoreau, Tim Pawl, Scott Ragland, B ill Rehg, Ian Rumfitt, Kent Staley, Eleonore Stump, Claudine Tiercelin, and audiences at Aberdeen, St. Louis, ANU, and the Kno wledge and Questions Conference, March 16, 2007 on par ts of this paper. Special thanks to Franck Lihoreau for organizing a great conference.

186

Brogaard, B. 2007a. “Descriptions: Predicates or Quantifiers?”, Australasian Journal of Philosophy 85, 117–136. — 2007b. “Number Words and Ontological Commitment”, Philosophical Quarterly 57, 1–20. — 2007c. “That may be J upiter: a Heuristic for Thinking Two-Dimensionally”, American Philosophical Quarterly 44, 315–328. — 2007d. “Attitude Reports: Do you Mind the G ap?”, in B. Weatherson, ed., Philosophy Compass. — Forthcoming. “ What Mary Did Yesterday: Reflections on Kno wledge-wh”, Philosophy and Phenomenological Research. Boër, S. and W. Lycan. 1986. Knowing Who, Cambridge: Cambridge University Press. Chalmers, D. 1996. The Conscious M ind: In Search of a F undamental Theor y, Oxford: Oxford University Press. — 2002. “On Sense and Intension”, Philosophical Perspectives 16, 135–82. — 2006. “The Foundations of Two-Dimensional Semantics”, in M. Garcia-Carpintero and J. Macia, ed., Two-Dimensional Semantics: Foundations and Applications, Oxford University Press, 55–139. — Forthcoming. “Propositions and A ttitude Ascriptions: A F regean Account”, Nous. Crimmins, M. 1992. Talk about Beliefs, Cambridge, Mass.: MIT Press. Crimmins, M. and J. Perry. 1989. “The Prince and the Phone Booth: Reporting Puzzling Beliefs”, Journal of Philosophy 86, 685–711. Dowty, D., R. Wall, and S. Peters. Introduction to Montague Semantics, Dordrecht: Reidel. Dretske, F. 1970. “Epistemic Operators”, Journal of Philosophy 67, 1007–23. Fara, D. G. 2001. “D escriptions as Predicates”, Philosophical Studies 102, 1–42. Originally published under the name ‘Delia Graff’. Fara, D. G. 2003 “Desires, Scope andTense”, Philosophical Perspectives 17: Philosophy of Language and P hilosophical Linguistics, 141–163. Originally published under the name ‘Delia Graff’. Fodor, J. D. 1979. The L inguistic D escription of O paque Contexts , New York: Garland. Frana, I. 2006. “Wondering about Concealed Questions”, SALT 16. Frege. G. 1892. “O n Sense and R eference”, in P.T. Geach and M. B lack, ed., Translations fr om the P hilosophical Writings of G ottlob Frege, Oxfor d: B asil Blackwell, 1952, 56–78. Grosu, A. 2003. “A Unified Theory of ‘Standard’ and ‘transparent’ Free Relatives”, Natural Language and Linguistic Theor y 21, 247–331. Grimshaw, J. 1979. “Complement Selection and the Lexicon”, Linguistic Inquiry 10, 279–326.

187

Groenendijk, J. and M. Stokhof. 1997. “Questions”, in J. van Bentham, and A. ter Meulen, Handbook of Logic and Language, Amsterdam/Cambridge, Mass.: Elsevier/MIT Press, 1055–1124. Hawthorne, J. 2005. “The Case for Closure”, in M. Steup and E. Sosa, ed., Contemporary Debates in Epistemology, 26–42. Heim, I. 1979. “Concealed Questions”, in R. Bäuerle, U. Egli and A. von Stechow, Semantics from different points of views, Berlin: Springer, 51–60. — 1982. The Semantics of Definite and Indefinite Noun Phrases, doctoral dissertation, Amherst, MA: University of Massachusetts. Higginbotham, J. 1996. “The Semantics of Q uestions”, in S. Lappin, ed., The Handbook of Contemporary Semantic Theor y, Oxford, 361–383. Hintikka, J. 1975. “Different Constructions in Terms of the Basic Epistemological Verbs: A Survey of Some Problems and Proposals”, The Intensions of Intentionality and Other New Models for Modalities, Dordrecht, 1–25. Jacobson, P. 1994. “Binding Connectivity in Copular S entences”, in Proceedings of the Fourth Conference on Semantics and Linguistic Theor y, Cornell University, Cornell Working Papers in Linguistics. — 1995. “The Quantificational Force of Free Relative Clauses”, in E. Bach et al, eds., Quantifiers in Natural Languages, Vol. II, Dordrecht: Kluwer, 451–486. Jaszczolt, K. M. 1999. Discourse, B eliefs, and I ntentions: Semantic D efaults and Propositional Attitude Ascriptions, Oxford: Elsevier Science. Jaszczolt, K. M. 2000. “The Default-Based Context-Dependence of Belief Reports”, in K. M. Jaszczolt, ed. The Pragmatics of Propositional Attitude Reports, Oxford: Elsevier Science, 169–185. Kallestrup, J. F orthcoming. “Knowledge-Wh and the P roblem of Conv ergent Knowledge”, Philosophy and Phenomenological Research. — Manuscript. “Knowing-Wh and the P roblem about Kno wledge of Kno wledge”. Kaplan, D. 1973. “D emonstratives”, Almog, P erry and Wettstein, eds., Themes from Kaplan, New York: Oxford University Press, 1989. Kripke, S. 1977. “Speaker Reference and Semantic Reference”, in Peter A. French, Theodore E. Uehling, Jr . and Howard K. Wettstein, eds., Midwest Studies in Philosophy, vol. II: Studies in the Philosophy of Language, Morris, MN: University of Minnesota, 255–276. Kripke, S. 1979. “ A Puzzle about B elief ”, Meaning and U se, ed. A. M argalit, Dordrecht, D. Reidel, 239–83. Larson, R., P. Ludlow and M. den Dikken. 1996. “Intensional Transitive Verbs”, Rivista di Linguistica 8, 29–46. Lasersohn, P. 2005. “The Temperature Paradox as Evidence for a Presuppositional Analysis of Definite Descriptions”, Linguistic Inquiry 36, 127–34.

188

Lewis, D. 1982. “Whether Report”, in T. Pauli et al, eds., 320311: Philosophical Essays Dedicated to Lennart Åqvist on his Fiftieth Birthday, Uppsala, Filosofiska Studier, 194–206; cited as reprinted in Papers in Philosophical Logic, Cambridge: Cambridge University Press, 1998, 45–56. Ludlow, P. and Stephen, N. 1991. “Indefinite Descriptions: In Defense of Russell”, Linguistics and Philosophy 14, 761–825. McKay, T. 1981. “On Proper Names in Belief Ascriptions”, Philosophical Studies 39, 287–303. McKay, T. 1991. “R epresenting D e Re B eliefs”, Linguistics and P hilosophy 14, 711–739. Nathan, L. E. 2006. On the Interpretation of Concealed Q uestions, Doctoral Dissertation, Department of Linguistics and Philosophy, Massachusetts Institute of Technology. Nelson, M. 2002. “Puzzling Pairs”, Philosophical Studies 108, 109–119. — 2005. “The Problems of Puzzling Pairs”, Linguistics and P hilosophy 28, 319– 350. Nozick, R. 1981. Philosophical Explanations, Cambridge, MA: Harvard University Press. Parson, T. 1997. “M eaning Sensitivity and G rammatical Structure”, in M. L. Dalla Chiara, K. Doets, D. Mundici, and J. van Bentham, eds., Structures and Norms in Science. Volume Two of the Tenth International Congress of Logic, Methodology and Philosophy of Science, Florence, August 1995. Dordrecht: Kluwer, 369–383. Partee, B. 1986. “Ambiguous Pseudoclefts with Unambiguous Be”, in Proceedings of the NELS XVI , S. B erman, J. Choe, and J. M cDonough, Amherst, MA: University of Masachusetts at Amherst, 354–66, as r eprinted in Partee 2004: 190–202. Partee, B. 2004. Compositionality in Formal Semantics, Oxford: Blackwell. Perry, J. 1979. “The Problem of the Essential Indexical”, Nous 13, 3–21. Quine, W. V. 1956. “Quantifiers and Propositional Attitudes”, Journal of Philosophy 53, 177–187 Recanati, F. 1993. Direct Reference: From Language to Thought , Oxford: Blackwell. Richard, M. 1990. P ropositional Attitudes, Cambridge: Cambridge U niversity press. Romero, M. 2005. Concealed Questions and Specificational Subjects”, Linguistics and Philosophy 28, 687–737. Salmon, N. 1986. Frege’s Puzzle, Cambridge, Mass.: MIT Press. — 1988. “Reflexivity”, in N. Salmon and S. Soames, eds., Propositions and Attitudes, Oxford: Oxford University Press, 240–274. — 1992. “Reflections on Reflexivity”, Linguistics and Philosophy 15, 53–63.

189

Schaffer, J. 2007. “Knowing the Answer”, Philosophy and Phenomenological Research 75, 383–403. Schiffer, S. 1977. “N aming and Kno wing”, Midwest Studies in P hilosophy 2, 28–41. — 1987. “The ‘Fido’-Fido Theory of B elief ”, Philosophical Perspectives 1, 455– 480. — 1992. “Belief Ascription”, Journal of Philosophy 89, 499–521. — 1996. “The Hidden-Indexical Theory’s Logical-Form Problem: a Rejoinder”, Analysis 56, 92–97. Schlenker, P. 2003. “Clausal Equations (A Note on the Connectivity Problem)”, Natural Language and Linguistic Theor y 21, 157–214. Soames, S. 1988. “Direct Reference, Propositional Attitudes, and Semantic Content”, in N. S almon and S. S oames, eds., Propositions and A ttitudes, Oxford: Oxford University Press, 240–274. — 1995. “B eyond Singular Propositions”, Canadian Journal of P hilosophy 25, 515–549. — 2002. Beyond Rigidity: the U nfinished Semantic Agenda of Naming and Necessity. Stanley, J. and T. Williamson. 2001. “Knowing How”, Journal of Philosophy 98, 411–44. Sterelny, K. 1988. “ Knowing Who by Stephen E. Boer, William G. Lycan”, Philosophy of Science 55, 654–656. Sternefeld, W. 2006. “Do Free Relative Clauses have Quantificational Force?”, in S. Beck and H.-M. G ärtner, eds., 30–40 Puzzles for Manfred Krifka. Online Festschrift, 1–3. Williamson, T. 2000. Knowledge and its L imits, Oxfor d: Oxfor d University Press.

190

Grazer Philosophische Studien 77 (2008), 191–216.

CONCEALED QUESTIONS UNDER COVER* Maria ALONI University of Amsterdam Summary Our evaluation of questions and kno wledge attributions may v ary relative to the way in which the relevant objects are identified. In the first part, the article proposes a theory that represents different methods of trans-world identification and is able to account for their impact on interpr etation. In the second par t, the same theory is used to account for the meaning of concealed questions. On the proposed account, the interpr etation of a concealed question r esults from the application of a type-shifting operation mapping an individual denoting expression into an identity question interpreted relative to a contextually selected identification method.

1. Introduction Consider the following situation. In front of you lie two face-down cards. One is the Ace of Hearts, the other is the Ace of Spades, but you don’t know which is which. You are playing the following game. You have to choose one card: if you choose the Ace of Spades, you win 10 euros. If you choose the Ace of Hearts, you lose 10 euros. Consider now the following sentence: (1) You know which card is the winning card. Is this sentence tr ue or false in this situation? O n the one hand, the sentence is true. You know that the Ace of Spades is the winning card, so you know which card is the winning card. On the other hand, however, as * The account presented here has been inspired by Harris (2007) and Schwager (2007), who are gratefully ackno wledged. I would also like to thank Kata B alogh, Floris Roelofsen, Tikitu de Jager, Paul Dekker, Paul Egré and Jeroen Groenendijk for insightful comments. The research reported here has been fi nancially suppor ted b y the N etherlands O rganization for Scientifi c Research (NWO).

far as you know, this card on the left might be the winning car d, or that one on the right. S o you do not kno w which car d is the winning car d. Intuitively, there are two different ways in which the cards can be identified in this situation: b y their position (the car d on the left, the car d on the right) or by their suit (the A ce of Hearts, the Ace of Spades). Which of these identifi cation methods is adopted seems to aff ect our evaluation of sentence (1). I f identifi cation b y suit is adopted, (1) is tr ue. But, if demonstrative identification is used, the sentence is false. This example illustrates the central idea I wish to defend in this article. Different methods of identification are operative in different conversational circumstances and our ev aluation of fragments of discourse, in par ticular knowing-wh constr uctions, may v ary relative to these methods. I n a number of previous works, I proposed a formalization of this old insight in a possible-world semantics (Aloni 2000, 2001, 2002, 2005a–b). The main idea consisted in r epresenting methods of identifi cation by what I called conceptual co vers (sets of concepts satisfying a number of natural constraints) and relativizing interpretation to a contextual parameter selecting diff erent conceptual covers on diff erent occasions. The main goal of the present article is to apply these ideas to another puzzle concerning the interpretation of so-called concealed questions (Heim 1979). A concealed question is a noun phrase naturally r ead as an identity question. As an illustration, consider the italicized nouns in the following examples: (2) a. b. (3) a. b. (4) a. b.

John knows the price of milk. { John knows what the price of milk is. Mary found out the murderer of Smith. { Mary found out who the murderer of Smith is. Ann told me the time of the meeting. { Ann told me what the time of the meeting is.

In her classic article, Heim (1979) observed that there is an ambiguity in sentences like the following: (5) John knows the capital that Fred knows. a. [Reading A] John knows the same capital that Fred knows b. [Reading B] John knows what capital Fred knows Suppose Fred knows the capital of Italy. Then on reading A the sentence entails that also J ohn knows the capital of I taly. On reading B, instead, 192

(5) lacks this entailment. It only follows that John knows that Fred knows what the capital of Italy is. John himself may lack this knowledge. Of the pr omising existing appr oaches to concealed questions (e.g. Romero 2005, N athan 2005, F rana 2006), R omero (2005) is the only one who provides a detailed account of H eim’s ambiguity. Her solution, however, has a cost: it r equires an undesirable cr oss-categorial account of the embedding verb know which can take complements of types ( s,e), (s,(s,e)), (s,(s,(s,e))), … In this article I will show that by adopting conceptual covers we can perspicuously represent the two readings of (5) without type inflation. The idea of using conceptual co vers in the r epresentation of concealed questions has been r ecently defended in H arris (2007) and Schwager (2007). The account I wish to present here is very close in spirit to these two previous approaches, although technically it is quite different, in particular in its solution to Heim’s puzzle. As a starting point I will assume Groenendijk & Stokhof ’s (henceforth G&S)(1984) account of questions and kno wledge-wh. The next section briefly introduces their analysis. Section 2 introduces the notion of a conceptual cover and employs it in the interpretation of (embedded) questions. Sections 3 and 4 extend this analysis to concealed questions. S ection 5 concludes the article and describes future lines of research. 2. Groenendijk and Stokhof on questions and knowledge In this section I will briefly introduce Groenendijk and Stokhof ’s (1984) analysis of questions and kno wledge-wh ascriptions. In G&S theory, the meaning of a question is identified with the set of meanings of all its possible exhaustive answers. More formally, interrogative sentences are represented by formulae of predicate logic preceded by a question operator, ?, and a sequence x1,…,xk G (henceforth x ) of k variables. Sentences are evaluated with respect to models M = (D, W, I ) consisting of a set D of individuals, a set W of possible worlds and a world dependent interpretation function I for the non-logical fragment of the language. A classical interpretation is assumed for indicative sentences. The denotation of an indicative in a model M, relative to a world w and an assignment function g is a truth value: aI bM,w,g  {0,1}. Interrogatives are analyzed in terms of their possible answers. The denotation of an interrogative in a given world is the proposition expressing the 193

exhaustive (or complete) true answer to the question in that world. Definition 1 [Q uestions] G G G G = aI b G G a?x I bM,w,g = {v  W |d  Dk : aI bM,v,g[x/d ] M,w,g[x/d ]} G An interrogative ? x I collects the worlds v in which the set of sequences G of individuals satisfying I is the same as in the evaluation world w. If x is G empty, ?x I denotes in w the set of the worlds v in which I has the same truth value as in w. For example, a polar question ? p denotes in w the proposition that p, if p is true in w, and the proposition that not p otherwise. As for who-questions, suppose a and b are the only two individuals in the extension of P in w, then the proposition that a and b are the only P is the denotation of ? x Px in w, that is the set of worlds v such that Iv(P ) = Iw(P ) = {a, b}. While indicatives express propositions, interrogatives determine partitions of the logical space. I will write aI bM to denote the meaning of a closed sentence I with respect to M, identified with the set of all possible denotations of I in M . While the meaning of an indicativ e corresponds to a set of worlds, i.e. a pr oposition, the meaning of an interr ogative is identified with the set of meanings of all its possible complete answ ers. Since the latter is a set of mutually ex clusive propositions the union of which exhausts the set of worlds, we say that questions partition the logical space. Partitions can be perspicuously visualized in diagrams. p ™p

Ow [nobody is P in w] Ow [d1 is the only P in w] Ow [d2 is the only P in w] Ow[d1 & d2 are the only P in w] … Ow[all d  D are P in w]

In the fi rst diagram, the polar question ? p divides the set of worlds in two alternatives, the alternative in which p is tr ue and the alternativ e in which p is false. I n the second diagram, the single-constituent question ?xPx divides the set of worlds in as many alternatives as there are possible denotations of the pr edicate P within M. Intuitively, two worlds belong 194

to the same block in the par tition determined by a question if their differences are irrelevant to the issue raised by the question. To express knowledge-wh, we extend the language with a kno wledge operator Ka selecting questions as complements. A sentence like ‘a knows G wh-I’ will translate as Ka(?x I ). A model for the extended language is a quadruple (D, W, Bel, I), where D, W, and I are as above and Bel is a function mapping individual-world pairs (a, w) into a subset of W. Intuitively Bel(a,w) represents the belief state of a in w. The semantics of the knowledge operator Ka is specified in the following clause: Definition 2 [Kno wledge-wh] G G aKa(?x I )bM,w,g = 1 iff Bel (a,w) Ž a?x I bM,w,g KaQ is true in w if a’s belief state is contained in the denotation ofQ in w. Since Q’s denotation in a world is equivalent to Q’s true exhaustive answer in that world, KaQ is true iff a believes the true exhaustive answer to Q. Consider now our initial sentence (1), here rewritten as (6)-a: (6) a. You know which card is the winning card. b. Ka(?xI ) Given the standard method of individuating objects adopted in the G&S analysis, the embedded question will determine the following partition: (7) {that d1 is the winning card, that d2 is the winning card, …} Sentence (6) is true in w if you believe the only proposition in this partition that is true in w, i.e. the unique true exhaustive answer to the embedded question in w. As w e hav e obser ved in the intr oduction, however, our evaluation of (6) (and of what counts as a good answ er to its embedded question) depends on the adopted method of identifi cation. Clearly, Groenendijk and S tokhof ’s standard treatment fails to account for this dependence. An ev en mor e serious pr oblem arises with the follo wing sentences: (8) a. b. (9) a. b.

Which card is which? ?xy. x = y You don’t know which card is which. ™Ka(?xy. x = y) 195

It’s easy to see that, since in each world each individual d is identical to itself, (8) is v acuous on G roenendijk and S tokhof ’s account and (9) is predicted to entail that your belief state is inconsistent. Both predictions clash with our intuitions about these examples. The analysis presented in the following section will solve these problems by introducing the notion of a conceptual cover. 3. Questions under cover In this section, I present a refinement of the G&S semantics in which different ways of identifying objects are represented and made available within one single model. I dentification methods ar e formaliz ed b y conceptual covers. Conceptual covers are sets of individual concepts which represent different ways of per ceiving one and the same domain. Q uestions ar e relativized to conceptual covers. What counts as an answer to a wh-question depends on which conceptualizations of the universe of discourse are assumed in the specifi c circumstances of the utterance. 3.1 Conceptual covers Consider again the card situation discussed at the beginning of the article. In front of y ou lie two face-do wn cards. One is the A ce of S pades, the other is the A ce of H earts. You don’t kno w which is which. Ther e are two diff erent ways of identifying the two car ds in this scenario: b y their position on the table (the car d on the left, the car d on the right) and b y their suit (the Ace of Spades, the Ace of Hearts). These two identification methods are typical examples of what the notion of a conceptual cover is meant to formalize. A conceptual cover is a set of individual concepts that satisfi es the following condition: in each world, each individual constitutes the instantiation of one and only one concept. More formally: Definition 3 [Conceptual co vers] Given a set of possible worlds W and a univ erse of individuals D, a conceptual cover CC based on (W, D ) is a set of functions W oD such that: w  W: d  D: !c  CC: c(w)=d.

196

Conceptual covers are sets of concepts that exhaustiv ely and ex clusively cover the domain of individuals. In a conceptual cover, each individual is identified by means of at least one concept in each world ( existence), but in no world is an individual counted twice ( uniqueness). In a conceptual cover, each individual in the universe of discourse is identified in a determinate way, and diff erent conceptual covers constitute diff erent ways of conceiving of one and the same domain. Illustration Consider again the card situation discussed above. To formalize this situation, we just need to distinguish two possibilities. The following diagram visualizes such a simple model: w1 6   w2 6
  The domain D consists of two individuals  and . The set of worlds W consists of w1 and w2. As illustrated in the diagram, either  is the car d on the left (w1) or  is the card on the right (in w2). There are only two possible conceptual co vers defi nable over such a model, namely the set A that identifi es the cards by their position on the table and the set B that identifi es the cards by their suit: A = {Ow [left]w , Ow [right]w } B = { Ow [Spades]w , Ow [Hearts]w } Set A contains the concepts the card on the left and the card on the right ; set B the concepts the Ace of Spades and the Ace of Hearts. They stand for two different ways of conceiving one and the same domain. C below is an example of a set of concepts that is not a conceptual co ver: C = { Ow [left]w , Ow [Hearts]w } Formally, C is not a co ver because it violates both the existential condition (no concept identifi es  in w1) and the uniqueness condition (  is counted twice in w1). Intuitively, C is ruled out because it does not provide a proper perspective over the universe of individuals. That C is inadequate is not due to properties of its individual elements, but to their combina197

tion. Although the two concepts the card on the left and the Ace of Hearts can both be salient, they cannot be regarded as standing for the two cards in D. If taken together, the two concepts do not constitute an adequate way of looking at our domain. 3.2 Quantification under cover I propose to r elativize interpretation to contextually selected conceptual covers. I add a special index n  N to the variables in the language. These indices range over conceptual covers. A model for this richer language is a fi ve-tuple (D, W, I, B el, C ) where D, W, I and Bel are as abo ve and C is a set of conceptual co vers based on (W, D). A conceptual perspective „ in M is a function fr om N to C. Sentences are interpreted with r espect to assignments under a perspectiv e. An assignment under a perspectiv e g„ is a function mapping v ariables xn to concepts in „(n), rather than individuals in D. Q uantification under conceptual co ver is defi ned as follows: Definition 4 [Q uantification under conceptual cover] axnI bM,w,g„ = 1 iff c  „(n) : aI bM,w,g„ [xn/c] = 1 On this account, variables range over elements of a conceptual cover, rather than over individuals simpliciter. The denotation of a variable in a world, however, will always be an individual, and never a concept. Definition  [The denotation of variables] axn bM,w,g„ = ( g„(xn ))(w) The denotation axn bM,w,g„ of a variable xn with respect to a modelM, a world w and an assignment under a perspective g„ is the individual (g„(xn))(w), i.e. the value of concept g„(xn) in world w. On this account, then, variables do not refer to concepts, but to individuals. They do refer, however, in a non-rigid way: different individuals can be their value in different worlds. To avoid the w ell-known problems that arise when w e treat variables as non-rigid designators, 1 we put r estrictions on their possible trans-world values via the notion of a conceptual perspective. 1. As an illustration of these pr oblems consider the v alidity of the expor tation from ‘Ralph believes that there is a spy’ to ‘There is someone Ralph believ es to be a spy’ (Kaplan 1969, 220) and the solution proposed in Aloni (2005a–b).

198

Quantification under conceptual co ver is neither bar e quantifi cation over individuals nor quantification over ways of specifying these individuals, rather it is quantification over individuals under a perspective. Relativization to a perspectiv e „ will only aff ect the interpretation of variables occurring free in an intensional context, and our evaluation of constituent questions. In this system, questions inv olve quantifi cation over elements of „-selected conceptualizations. In case of multi-constituent questions, G different variables can be assigned diff erent conceptualizations. (B y xn I mean the sequence x1,…,xk indexed by the indices n1,…,nk respectively. By G G „(n ) I mean the product –ik(„(ni )). And by c (w) I mean the sequence c1(w), . . . , c k(w).) Definition 6 [Q uestions under Cover] G G G a?xnI bM,w,g„ = {v | c  „(n ) : aI bM,w,g„[xGn /cG]=aI bM,v,g„[xGn /cG]} The idea formalized by this definition is that by interpreting an interrogative sentence one quantifi es over tuples of elements of possibly distinct conceptual covers rather than dir ectly over tuples of individuals in D. If analyzed in this way, a question like ?xn Pxn groups together the worlds in which the denotation of P is identifi ed by means of the same set of elements of the conceptual cover selected for n. A multi-constituent question like ?xn ym Rxn ym groups together those worlds in which the pairs (d1,d2) in the denotation of R are identified by means of the same pairs of concepts (c1,c2), where c1 and c2 can be elements of two different conceptualizations. The following diagram visualizes the partition determined by ?xn Pxn under a perspective „ such that „(n) = {c1,c2,…}. Ow [no ci(w) is P in w] Ow [c1(w) is the only P in w] Ow [c2(w) is the only P in w] Ow [c1(w) & c2(w) are the only P in w] … Ow[all ci (w) are P in w]

Due to the definition of conceptual covers, in the first block of this partition no individual in D is P; in the fourth block exactly two individuals in D are P; and in the last block all individuals in D are P. 199

Illustration Consider again the card situation described above. In front of you lie two face-down cards. One is the Ace of Hearts, the other is the Ace of Spades. You don’t know which is which. Furthermore, assume that one of the cards is the winning car d, but y ou don’t know which one. We can model this situation as follows (the dot indicates the winning card): w1 w2 w3 w4

6  6

 6 • 6

•

• •  

Our model now contains four worlds, r epresenting the possibilities that are compatible with the described situation. N ow consider two possible conceptual perspectives: „1 and „2. The former assigns to the index of the variable xn the cover that identifies the cards by means of their position on the table, „2(n) identifies the cards by their suits: (10) a. „1(n)={Ow[left]Z, Ow[right]Z}; b. „2(n) ={Ow[Spades]Z, Ow[Hearts]Z}. Consider the following interrogative sentence: (11) a. Which is the winning card? b. ?xn . xn=L yn Pyn Example (11) structures the set of worlds in two different ways depending on which perspective is assumed: under „1:

w1 w2 w3 w4

under

„2:

w1 w4 w2 w3

Under „1, (11) disconnects those worlds in which the winning car d occupies a diff erent position. Under „2, it groups together those possibilities in which the winning card is of the same suit. In other words, in the first case, the relevant distinction is whether the left car d or the right car d is the winning car d; in the second case the question expr essed is whether Spades is winning, or H earts. Since diff erent par titions are determined 200

under diff erent perspectiv es, w e can account for the fact that diff erent answers are required in different contexts. For instance, (12) counts as an answer to (11) only under „2: (12) The Ace of Spades is the winning card. Consider now the situation described at the beginning of this article. You know that the A ce of S pades is the winning car d, but y ou don’t know whether it is the card on the left or that on the right. nI this situation your belief state corresponds to the set {w1, w4}. Sentence (13) is then correctly predicted to be true under „2, but false under „1. (13) a. You know which card is the winning card. b. KD(?xn . xn = L yn Pyn ) Finally, consider again the following sentences: (14) a. b. (15) a. b.

Which card is which? ?xn ym . xn = ym You don’t know which card is which. ¬KD(?xn ym . xn = ym )

As we saw, in standar d theories, (14) and (15) ar e wrongly predicted to be vacuous and to entail that your belief state is inconsistent, respectively. On our account, instead, since different wh-phrases in a multi-constituent question can range over different sets of concepts, (14) can be signifi cant and (15) can fail to entail inconsistence. To see this, assume „ assigns different covers to n and m, for example: (16) a. „(n) = {Ow[left]w , Ow[right]w }; b. „(m) = {Ow[Spades]w , Ow[Hearts]w }. If interpreted under such perspectiv e, (14) groups together those worlds that supply the same mapping fr om one co ver to the other , and is not vacuous in our model. The determined par tition is depicted in the following diagram:

201

under „:

w1 w3 w2 w4

The question divides the set of worlds in two blocks: w{ 1, w3} and {w2, w4}. The first alternative corresponds to the possible answ er (17), the second to the possible answer (18): (17) The Ace of Hearts is the card on the left and the Ace of Spades is the card on the right. (18) The Ace of Hearts is the card on the right and the Ace of Spades is the card on the left. If your belief ’s state is specified as above, i.e. as {w1, w4}, then (15) would be tr ue in w1 without entailing inconsistency . In the follo wing section we will use this analysis of kno wing-wh to explain our interpr etation of concealed questions. 4. Concealed questions under cover In this section I propose an analysis of Concealed Questions (henceforth CQs) defi ning a type-shifting operator mapping nominals into identity questions. The latter are then interpreted relative to a conceptual perspective as explained in the previous section. As w e saw in the intr oduction, CQs ar e nominals naturally r ead as identity questions. As an example, consider the italicized part in (19): (19) a. John knows the capital of Italy. b. John knows what the capital of Italy is. Example (19)-a is ambiguous between an epistemic CQ-reading exemplified in (19)-b and an acquaintance erading in which the italicized nominal is not a CQ. 2 Only on this second r eading, substitution of identicals is allowed: 2. Languages like German or Italian use different lexical items for the two r eadings: wissen and kennen in German (Heim 1979); sapere and conoscere in Italian (Frana 2007). When wissen and sapere take a nominal argument, only the CQ reading is available.

202

(20) Mary knows the capital of Italy. a. Acquaintance: º She knowsAC Rome b . CQ-reading: i She knowsCQ Rome In this article we will only be concerned with CQ-r eadings of these sentences. A detailed analysis is proposed in the following section. 4.1 The proposal On this account, CQs are syntactically nominals, but semantically questions. Their interpretation crucially involves the application of a type-shifting operator nn which transforms an entity denoting expression D into the identity question ?xn . xn = D (whon /whatn is D?).3 Definition  [The type-shift rule] nnD =def ?xn . xn = D The type-shift rule nn applies to avoid the type mismatch otherwise arising from the application of a question-embedding verb like know to an entity denoting expression. The value of n in nn is pragmatically supplied. The resulting identity question is interpreted relative to a conceptual perspective as explained in the previous section. Analyses of CQs can be grouped in three classes (see Romero 2006 for a detailed ev aluation): pragmatic theories (e.g. F rana 2006), individual concept theories (e.g. R omero 2005), and pr opositional theories (e.g. Nathan 2005). The present account, technically, can be seen as a combination of all three approaches. Since question denotations are propositions we share the positiv e sides of the pr oposition theor y in that no special notion of kno wledge-CQ must be posited, standar d knowledge-wh will suffice. On the other hand, identity questions ar e here interpreted with respect to contextually selected sets of concepts allo wing us to account for (a) the contextual dependence of CQs as in the pragmatic appr oach and (b) the intuition formaliz ed in the individual concept appr oach that their interpr etation requires comparing v alues assigned in diff erent worlds.

3. Cf. Harris (2007, chapter 4) which pr esents psycholinguistic evidence that is br oadly compatible with a view of concealed questions that involves a shift of interpretation.

203

Illustrations A sentence like (20), on its CQ-reading, receives on this account the following representation: (21) a. John knows the capital of Italy. b. Kj (nmL xn Pxn ) (=

Kj (?ym. ym = L xn Pxn ))

When a question-embedding verb like know applies to an entity denoting expression like the capital of Italy, the type-shift rule nm must apply to avoid type mismatch. The resulting sentence is then interpreted according to the analysis of knowing-wh under cover given in the previous section. (22) aKj (?ym . ym = L xn Pxn )bw,g„ = 1 iff Bel ( j, w) Ž a?ym . ym = L xn Pxn bw,g„ The intended meaning is obtained if m is mapped to the following cover representing identification by name:4 (23) „(m) ={Ow[Berlin]w , Ow[Rome]w , Ow[Paris]w ,…} Assuming the question semantics intr oduced in the pr evious section, the embedded question ?ym . ym = L xn Pxn denotes in w the proposition that Rome is the capital of I taly, if in w Rome is indeed the capital of I taly. Sentence (21) then is true in w if John believes in w this true proposition. Since no shift of cover is necessary for this example, the same interpretation would have obtained if we had assumed the classical G&S theor y. In the following example, instead, conceptual covers play a more crucial role. Example (24) illustrates the case of a quantifi ed CQ. Quantified CQs are typically problematic for an individual concept account. On the present theory, they can be perspicuously analyzed as follows: (24) a. John knows all European capitals. b. xn(Pxn o Kj (nm xn )) (24)-b can be roughly paraphrased as for each European capital John knows it. The most natural resolution for n and m here is the following:

4. The value „(n) assigned to n is irrelevant in this case, because xn does not occur free in an intensional context. By an economy principle, we can then assume „(n) = „(m).

204

(25) a. „(n) ={the capital of Germany, the capital of Italy, …} b. „(m) ={Berlin, Rome, …} The sentence is then predicted to be true if John knows the true exhaustive answer to the question ?ym . xn = ym (Whatm is xn?) for each xn  „(n), that is: (26) a. What is the capital of Germany? b. What is the capital of Italy? c. ... This prediction is intuitiv ely correct. Note that contrar y to the pr evious example, the quantifi ed case cr ucially requires a shift in perspectiv e, n and m cannot be assigned the same v alue here, otherwise the quantifi ed questions would be trivialized. In the following section, we discuss Heim’s ambiguity where conceptual covers again play an essential role. 4.2 Heim’s ambiguity As we saw in the intr oduction, Heim (1979) describes two r eadings for sentences like (27), normally labeled as reading A and reading B. (27) John knows the capital that Fred knows. a. [Reading A] John knows the same capital that Fred knows b. [Reading B] John knows what capital Fred knows The representation of this ambiguity is a challenge for most appr oaches to CQs. R omero (2005) was the fi rst to pr ovide a detailed account. I n Romero’s analysis, CQs denote individual concepts. Reading A is obtained by letting kno w apply to the extension of the description ‘ the capital that Fred knows’, an object of type ( s,e). Reading B is obtained b y letting know apply to the intension of the description ‘the capital that Fred knows’, an object of type (s,(s,e)). Suppose Fred only knows the capital of Germany, then the extension of ‘ the capital that Fred knows’ will be the concept the capital of G ermany, its intension will be the ‘ meta-concept’ the capital that Fred knows. Only reading A then is correctly predicted to entail that J ohn knows the capital of G ermany. Reading B only entails that John knows what capital F red knows. Although R omero’s analysis captures the two r eadings of our sentence, it does it on a high cost: a 205

cross-categorial account of know is needed which can take complements of types ( s,e), ( s,(s,e)), ( s,(s,(s,e))), … I n what follo ws I will sho w that this type inflation can be avoided if we use quantifi cation under conceptual covers. In our frame work, Heim’s ambiguity can be r epresented as follows: (28) John knows the capital Fred knows. a. Reading A: yn(yn = L xn (Pxn š Kf (nm xn )) š Kj (nm yn )) b . Reading B: Kj (nn L xn (Pxn š Kf (nm xn ))) Reading A can be paraphrased as saying that there is a unique capital that Fred knows and John knows it too under the same conceptual perspective. Reading B simply asserts that John knows the answer to the question ‘What is the capital that F red knows?’. Heim’s intended meanings ar e captured by assuming the following resolution for the indices n and m: (29) a. „(n)={the capital of Germany, the capital of Italy, …} b. „(m)={Berlin, Rome, …} On this resolution, on reading A, both Fred and John can identify one and the same capital in „(n) by name. On reading B, instead, only Fred has this knowledge. John can only giv e a descriptiv e answer to the question ‘What is the capital that Fred knows?’. Illustration As an illustration consider the following situation: (30) a. In w1, Berlin is the capital of Germany and Paris is the capital of France. Fred knows that B erlin is the capital of G ermany, but doesn’t know the capital of France. b. In w2, Paris is the capital of G ermany and F red knows that Paris is the capital of G ermany, but doesn’t know the capital of France. c. In w3, Berlin is the capital of Germany and Paris is the capital of France. Fred knows that Paris is the capital of F rance, but doesn’t know the capital of Germany. Suppose Bel (j, w1) = {w1, w2}, that is, John believes in w1 that Fred knows the capital of G ermany, but he himself wonders whether it is B erlin or 206

Paris. Intuitively this is a situation in which reading A is false and reading B is true. Suppose now instead Bel ( j, w 1) = {w1, w 3}. In w1, John knows all the capitals, but he doesn’t know which capital Fred knows: Fred might know Berlin, or he might know Paris. In this case, intuitively, reading A is true and reading B is false. Let us call these two cases M1 and M2: (31) a. M1: Bel1( j, w1) = {w1, w2} 6 reading A is false inw1 and reading B is true. b. M2: Bel 2( j, w1) = {w1, w3} 6 reading A is true in w1 and reading B is false. In what follows I will show that these are indeed our predictions given the representations in (28). First of all, let us consider the concepts relevant in these situations. Let b, p and r stand for the individuals B erlin, Paris, and Rome respectively. Let g, f, i be the concepts the capital of Germany, the capital of France and the capital of I taly respectively. The following table r epresents the v alues of these concepts in the r elevant worlds. To fully r epresent Fred’s belief, we consider for each world w also a world w and assume Bel(f, w ) = {w , w }.

(32)

w1 w1 w2 w2 w3 w3

g

f

i

b b p p b r

p r b r p p

r p r b r b

Let us consider now the description ‘the capital Fred knows’. (33) a. The capital Fred knows. b. L xn(Pxn š Kf (nm xn )) As above, w e assume that „(n) is the descriptiv e cover { the capital of Germany, the capital of F rance, the capital of I taly}. Whereas „(m) is the cover {Berlin, Paris, Rome} representing identification by name, which in 207

this model corresponds to the rigid cover RC = {Ow d | d  D}. It’s easy to see that under this conceptual perspective, (33) corresponds to the concept k getting the values listed in (34).

(34)

w1 w1 w2 w2 w3 w3

k b b p p p p

As an illustration of the wor king of this semantics, we show in details the case of w1. Let us assume the following treatment of iota terms: (35) aL xnI bM,w,g„ = c (w), if !c  „(n) : aI bM,w,g„[x/c] = 1, undefi ned otherwise. Recall that „(n) is the descriptive cover. We want to show that there is only one concept c in this cover that satisfi es the clause in (36), and that its value in w1 is b, as illustrated in table (34): (36) aP xn š Kf (nm xn )bM,w1,g„[x n/c] = 1 In this logic, (36) holds iff the following holds: (37) c (w1)  Iw1(P) & Bel(f, w1) Ž a?ym . xn = ym bM,w1,g„[x n/c] Bel (f, w1) is {w1, w1} by definition of the model. The following table gives us the denotation of ?ym . xn = ym (what is xn?) in w1 for each possible values for xn. (38) a. xn l the capital of Germany b. xn l the capital of France c. xn l the capital of Italy

6 {w1, w1 , w3} 6 {w1, w3 , w3} 6 {w1, w2, w3}

The set in (38)-a corresponds to the denotation inw1 of the question What is the capital of G ermany? and contains all worlds in which the capital of 208

Germany is identifi ed by the same name as in w1, i.e. as B erlin. The set in (38)-b corresponds to the denotation in w1 of the question What is the capital of F rance? and contains all worlds in which the capital of F rance is identified by the same name as in w1, i.e. as Paris. And so on. Of these three sets, only the one in (38)-a contains Bel ( f, w1). Therefore, there is a unique concept satisfying the clause in (37), namelythe capital of Germany, and its value in w1 is indeed b, Berlin. Consider now reading A, represented by (39). (39) yn ( yn = L xn(P xn š Kf (nm xn )) š Kj (nm yn )) Example (39) is tr ue in w1 if there is a concept c in the descriptive cover „(n) such that (i)c(w1) is equivalent to the value of the concept the capital Fred knows in w1, i.e. c(w1) = b, and (ii) J ohn knows how to map c into the naming co ver „(m), i.e. ther e is a concept cc in „(m) such that in each world v in John’s belief state c and cc have the same value. Consider now the two models M1 and M2 in (31). In both models, the only concept in „(n) satisfying the fi rst condition is the capital of G ermany. It is easy to see, however, that the second condition is only satisfi ed in M2, where Bel2( j, w1) = {w1, w3} is a state where c, the capital of Germany, is identified as Berlin. In M1, instead, which formalizes a situation in which John wonders whether the capital of G ermany is Berlin or Paris, the sentence is false. Bel1( j, w1) = {w1, w2} is a state where the capital of Germany cannot be identified. Consider now Reading B, represented by (40). (40) Kj (nn L xn(P xn š Kf (nm xn ))) Sentence (40) states that J ohn knows the tr ue exhaustive answer to the following question: (41) a. What is the capital Fred knows? b. ?yn . yn = L xn(P xn š Kf (nm xn )) Consider the par tition determined b y this question in both our models M1 and M2:

209

(43)

w1 w2 w3

w1 w2 w3

In the first block of this partition, Fred knows the capital of Germany. In the second block, he kno ws the capital of F rance. Reading B states that John knows the true exhaustive answer to this question, i.e. that his belief state is contained in one and only one block in this par tition. Consider now again the characterization of John’s belief state in w1 in our two models in (31): Bel1 ( j, w1) = {w1, w2} in M1, and Bel2 ( j, w1) = {w1, w3} in M2. Sentence (42) is then true in w1 given M1, but false in w1 given M2. Aside: an alternative pragmatic account In a logic assuming quantifi cation under conceptual co ver, a de dicto sentence always has an equivalent de re representation. This holds also for our de dicto rendering of reading B in (40). This means then that in the present system we have an alternative way of representing Heim’s ambiguity solely in terms of different index resolutions. Our ambiguous sentence could receive one and only one de re representation as in (43). (43) a. John knows the capital that Fred knows. b. y0(y0 = L x1(Px1 š Kf (n2 x1)) š Kj (n3 y0 )) The readings A and B would then be obtained by assuming the resolutions in (44) and (45) respectively: (44) R eading A: a. 0, 1 l the capital of Germany, the capital of Italy, … b . 2, 3 l Berlin, Rome, … (45) R eading B: a. 0 must include the capital that Fred knows; b . 1, 3 l the capital of Germany, the capital of Italy, … c. 2 l Berlin, Rome, … Note however, that a de re representation of reading B is more costly than our de dicto alternative in (40) in that it involves a third conceptual cover as value of „(0), a cover containing the concept the capital Fred knows. A purely pragmatic account of Heim’s ambiguity along these lines has been proposed by Schwager (2007), although her formalization is quite

210

different from the one I present here. In order to fully evaluate a structural or a pragmatic account, w e probably will need to hav e a closer look at various disambiguated variants of (43): (46)

a. John knows the capital Fred does. b. John knows the same capital Fred knows. c. John knows the capital Fred knows too.

Example (46)-a is from Harris (2007). Examples (46)-b and c are attributed to Irene Heim in Schwager (2007). In all three cases in (46) only reading A survives. A structural account to the ambiguity might be better equipped to account for these facts than a purely pragmatic account. A full analysis however must be left to another occasion. 5. Two observations and their exceptions Section 2 of the present article accounted for the variability of interpretation of questions and their answers by relativizing their evaluation to contextually selected methods of identification. The previous section applied the same theor y to the case of concealed questions. G reenberg (1977), however, observed that CQ interpretations are typically not ambiguous in the way their question paraphrases ar e. Consider the following CQ with its question paraphrase: (47) a. Officer Hopkins found out the murderer of Smith. b. Officer Hopkins found out who the murderer of Smith was. Both (47)-a and (47)-b can be interpr eted as (48)-a, but, as G reenberg observed, only the full embedded question v ersion (47)-b enjo ys also reading (48)-b, where an identifying property or fact is enough to resolve the embedded question. (48) a. Officer Hopkins resolved the question of who murdered Smith by identifying the individual. b. Officer Hopkins resolved the question of who murdered Smith by finding out some essential facts (e.g. that he was his br other) about the individual denoted by the murderer of Smith.

211

To capture this fact about CQs, in most existing analyses, sentence (47)-a is interpreted as r equiring that one and the same individual is the murderer of Smith in each world in Offi cer Hopkins’ belief state. This is the standard way of modeling identification. For a term t to be identified by a subject a, t has to denote one and the same individual in all ofa’s doxastic alternatives (Hintikka 1969). In this framework, we could follow the same strategy and account for Greenberg’s observation by fixing as value of n, in our type-shift operation nn , the rigid co ver RC = { Ow d | d  D}. Note that if w e adopted nRC , still conceptual covers could play a cr ucial role to account for quantifi ed CQs like those in (24), which, as we saw, involved quantification under a perspective. We will not follow this strategy though and in the following paragraphs I will briefly explain why. First of all, if w e assumed nRC , we would lose our perspicuous r epresentation of H eim’s ambiguity. Reading B of a sentence like John knows the capital F red knows typically inv olves identifi cation by the non-rigid cover {the capital of Germany, the capital of Italy, …}. Indeed all previous theories that defi ne identification in terms of rigidity fail to account for reading B of Heim’s sentence. Secondly, as various double vision or mistaken identity puzzles sho w, identification is always relative to a perspective (e.g. Quine’s (1953, 1956) Tully/Cicero or Ortcutt situations, but also our initial car d scenario). An analysis of identifi cation in terms of rigidity would fail to account for these situations. As an illustration involving a CQ, consider the following example from Schwager (2007, 3): (49) scenario J ohn gives you a name and address of Dr. Maria Bloom (the individual DMB) who is indeed a doctor who can help you. That same night, John and DMB happen to be at the same party and she is introduced to him as ‘Mary’. They start chatting and, since she is a spare-time semanticist, she starts explaining to him some classical puzzles of mistaken identity. John is very fascinated and ends up thinking she must be some sor t of philosopher (or maybe, philologist?), but certainly not a doctor. Intuitively both (50)-a and (50)-b can be understood as true in the given scenario.

212

(50) a. John knows a doctor who can help you. b. John thinks his interlocutor is not a doctor. As Schwager obser ves, on a rigid account of CQs (e.g. F rana 2006 and Nathan 2005), (50)-a entails that ther e is one and the same individual, DMB, who is a doctor who can help you in all of John’s doxastic alternatives. But then since DMB is also John’s interlocutor, (50)-b is incorrectly predicted to be false in this situation by these theories. Schwager’s theor y, instead, which uses conceptual co vers, avoids this problem, as well as the present analysis. On these two accounts, identification by name and demonstrative identification can be represented by two different conceptual covers. For example, the rigid cover RC can be used for demonstrative identification, and the non-rigid cover NC = {Ow[Maria Bloom]w , …} for naming (in an epistemic setting it is most natural to treat names as non-rigid designators, cf. Hintikka 1975). Sentence (50)-a then, naturally interpreted under naming, can be compatible with (50)b, interpreted under demonstrative identification. But then adopting nRC would not be adequate to account for this situation. To captur e Schwager’s mistaken identity example, but, at the same time, account for Greenberg’s observation, we could then assume nRC/NC , rather than nRC . CQs would then require identification by ostension or by name. Identifications by description would be r uled out in this account, in accordance to Greenberg’s observation. Reading B of H eim’s example would have then to be explained on a diff erent level (e.g. as in R omero 2005). Harris (2007), however, discusses v arious examples of CQs inv olving identification by description, rather than acquaintance or naming, showing that given the right context many exceptions to Greenberg’s observation can be found. Here is one of his convincing examples (Harris 2007, 18): (51) scenario J ohn is a statistician r esearching management tr ends in professional or academic organizations. H e discovers that the person elected president of the LSA is always the person who has published the most articles in Language the year before. Perhaps this fact is merely an odd anomaly. Perhaps it is stated as such in the by-laws of the organization. In any event, interpreting (a) as (b) is perfectly felicitous in this situation: a. John predicted/guessed/recalled the president of LSA. 213

b. John predicted/guessed/recalled it is the linguist with the most articles published in Language. Examples like (51) constituted the main motivation for our choice to leave the resolution of n to pragmatics in our defi nition of nn . Romero (2005, 2006), however, discusses a potential problem arising for such a pragmatic approach. On a pragmatic account, when placed in the right context, we should be able to interpr et Ro me in (52) as a concealed question. B ut, as Romero observed, ‘this is contrar y to fact: no matter ho w much one plays with the context, (52) does not have a CQ reading’ [Romero, 2006, example (12)]. (52) #J ohn knowsCQ Rome. To further support her argument Romero also mentions that in languages like Spanish in which, contrary to English, epistemic know and acquaintance know are lexically distinct the epistemic, or CQ, v ersion of (52) is simply ungrammatical: (53) a. #Juan sabe Roma. b . ‘John knowsCQ Rome’ The following example however shows that again given the right context exceptions to Romero’s observation can be found.5 Suppose you are given the list of world capitals in (54) and you are asked to say which state each city is the capital of: (54) Rome, Naypyidaw, Mbabane, Ouagadougou. In this situation I could say in Italian: (55) a. So solo Roma. b . ‘I knowCQ only Rome’. Italian sapere, like Spanish saber only has the CQ reading (cf. footnote 2). Example (55) then shows that given the right context even proper names could have a CQ interpr etation, an ex ception to R omero’s obser vation. 5. Other examples of this type have been brought to my attention by Paul Dekker (p.c.).

214

This last example, again, can only be tackled if we let the value of n in nn be pragmatically supplied, as has been pr oposed in the pr evious section. Of course, still, CQs are ordinarily interpreted as requiring identification by name, and cases like (55) ar e very marginal. A theor y of ho w covers are contextually selected is urgently needed to account for these facts, but unfortunately such a theor y must be left to another occasion (see Aloni 2005b, and Schwager 2007 for a fi rst attempt). 6. Conclusion A domain of individuals can be observed from many different angles. The first part of this article presented a theory in which these different ways of identifying objects are represented and their impact on our interpretation of (embedded) questions is accounted for. In the second par t, the same theor y has been used to account for the meaning of concealed questions. In this proposal, the interpretation of a concealed question results from the application of a type-shifting operation mapping an individual denoting expression into an identity question interpreted relative to a contextually selected identifi cation method. The most urgent question that needs to be addr essed now is how different identification methods are selected on different occasions. Another issue that deser ves fur ther inv estigation concerns the disappearance of reading B in elliptical or other variants of Heim’s sentences (see examples in (46)). Other empirical properties of CQs that have not been discussed yet include their link with relational nouns, and the fact that not all question-embedding verbs can embed concealed questions contrary to what the present analysis predicts (Mary knows / #wonders the capital of tIaly). But again, an explanation of these facts must be left to another occasion.

REFERENCES Aloni, M. 2000. “Conceptual co vers in dynamic semantics ”, in L. Cav edon, P. Blackburn, N. Braisby, and A. Shimojima (eds.), Logic, Language and Computation, Vol. III, CSLI, Stanford, CA. — 2001. Q uantification under Conceptual Co vers, P h.D. thesis, U niversity of Amsterdam. 215

Aloni, M. 2002. “Questions under cover”, in D. Barker-Plummer, D. Beaver, J. van Benthem, and P. S. de L uzio (eds.), Words, Proofs, and Diagr ams, CSLI, Stanford, CA. — 2005a. “Individual concepts in modal predicate logic”, Journal of Philosophical Logic 34(1), 1–64. — 2005b. “A formal tr eatment of the pragmatics of questions and attitudes ”. Linguistics and Philosophy, 28(5), 505–539. Frana, I. 2006. “The de r e analysis of concealed questions ”, in Semantics and Linguistic Theor y XVI. Greenberg, B. 1977. A semantic account of relative clauses with embedded question interpretations. Ms. UCLA. Groenendijk, J. and Stokhof, M. 1984. Studies on the Semantics of Questions and the Pragmatics of Answers, Ph.D. thesis, University of Amsterdam. Harris, J. A. 2007. Revealing Concealment: A (Neuro-)Logical Investigation of Concealed Questions, MA thesis, University of Amsterdam. Heim, I. 1979. “Concealed questions”, in R. Bauerle, U. Egli, and A. von Stechow (eds.), Semantics from Different Points of View, Berlin, Springer. Hintikka, J. 1969. “S emantics for pr opositional attitudes”, in D avis, Hockney, and Wilson (eds.), Philosophical Logic, Reidel, Dordrecht. — 1975. The Intensions of I ntensionality and O ther New Models for M odalities, Reidel, Dordrecht. Kaplan, D. 1969. “Quantifying in”, in D. Davidson and J. Hintikka (eds.), Words and Objections: Essays on the Work of W. V. Quine, 221–243, R eidel, Dordrecht. Nathan, L. 2005. On the interpretation of concealed questions, Ph.D. thesis, MIT. Quine, W. V. O. 1953. From a Logical P oint of View, Chapt. VII Reference and modality, Harvard University Press. — 1956. “Q uantifiers and pr opositional attitudes”, Journal of P hilosophy 53, 101–111, reprinted in: W. V. O. Quine, 1966, The Ways of Paradox and Other Essays, Random House, New York. Romero, M. 2005. “Concealed questions and specificational subjects”, Linguistics and Philosophy 28, 687–737. — 2006. “On concealed questions”, in Semantics and Linguistic Theor y XVI. Schwager, M. 2007. Keeping prices lo w: an answ er to a concealed question . Talk presented at Sinn und Bedeutung 2007.

216

Grazer Philosophische Studien 77 (2008), 217–261.

KNOWING HOW TO ESTABLISH INTELLECTUALISM*

Daniele SGARAVATTI

Arché, University of St Andrews & University of Eastern Piedmont ‘Amedeo Avogadro’

Elia ZARDINI

Arché, University of St Andrews Summary In this paper, we present a number of problems for intellectualism about knowledge-how, and in par ticular for the v ersion of the vie w developed by Stanley & Williamson 2001. Their argument draws on the alleged uniformity of ‘kno w how’and ‘know wh’-ascriptions. We offer a series of considerations to the effect that this assimilation is pr oblematic. Firstly, in contrast to ‘kno w wh’-ascriptions, ‘know how’-ascriptions with known negative answers are false. Secondly, knowledge-how obeys closure principles whose counterparts fail for knowledge-wh and knowledgethat. Thirdly, as opposed to knowledge-wh and knowledge-that, knowledge-how is inferentially isolated from further knowledge-that. We close by providing some evidence against the further reduction of knowledge-wh to knowledge-that, which is presupposed by the intellectualist theory under discussion.

1. Introduction and overview Is there a unique, basic form of human cognition, to which ev ery other form of human cognition, no matter how apparently dissimilar it may seem * Earlier versions of the material in this paper were given in 2007 at the UCLA Epistemology Workshop and at the Ar ché Knowledge and Language S eminar (University of St Andrews); in 2008, at the Arché Reading Party in Carbost (University of St Andrews) and at a research seminar at the University of Göttingen. We would like to thank all these audiences for v ery stimulating comments and discussions. Special thanks go to Herman Cappelen, Stewart Cohen, Dylan Dodd, Mikkel Gerken, Felix Mühlhölzer, Nikolaj Pedersen, Adolf Rami, Jonathan Schaffer, Jason Stanley and Crispin Wright. In writing this paper, the first author (the order being merely alphabetical) has benefitted from funding from the University of Eastern Piedmont ‘Amedeo Avogadro’ and the University of St Andrews, the second author from an AHRC Postdoctoral Research Fellowship.

in thought and talk, can ultimately be reduced? Despite the etymological suggestion, human cognition—the wide range of diff erent phenomena in which reality discloses itself to the human mind—need not be (and arguably is not) exhausted by knowledge: for example, irreducible warrant, perception and understanding—to name but a few epistemic properties that are recalcitrant to any straightfor ward assimilation to kno wledge—may all well occur essentially in a full description of human cognition. Restricting henceforth our attention to (human) kno wledge, our question becomes whether there is a unique, basic form of knowledge, to which every other form of knowledge, no matter how apparently dissimilar it may seem in thought and talk, can ultimately be r educed. If there were such a basic form of knowledge, a good case could be made that this would have to be knowing that something is the case (also known as ‘knowledge-that’),1 or so many theorists have traditionally thought. Many tough challenges await attempts at r educing all kno wledge to knowledge-that. For example, knowing things (also known as ‘knowledgeof’) has long been thought to be irreducible to knowledge-that (see Russell 1912, 72–92). Elia knows Daniele, but he would still kno w him even if he had very diff erent beliefs (and hence kno wledge-that states) from the ones he actually has. Maybe it is necessary for him to know Daniele that he know that D aniele exists, but this is cer tainly not suffi cient, since he also knows that Carla B runi exists, but, much to his o wn chagrin, does not know Carla. These and similar considerations present a challenge to a reduction of knowledge-of to knowledge-that. More related to the topic of the special issue of this journal, knowing how things ar e (also kno wn as ‘knowledge-wh’)2 has surprisingly pr oven to be not so easily amenable to reductions in terms of kno wledge-that (see Schaff er 2007 for a r ecent challenge to such r eductionist attempts). I n barn-façade countr y (see Goldman 1976), H enry would seem to kno w which barn he ’s looking at, but, as the majority of epistemologists hav e held, H enry would not seem to kno w that the object is looking at is a barn. These and similar 1. Knowledge-that is v ery frequently glossed as inv olving a r elation between the r elevant subject of knowledge and a proposition. Given the extreme plasticity of the notion of proposition in contemporary philosophy, we are very unclear as to what this gloss is supposed to amount to, and hence very unclear as to whether it is even extensionally adequate (see e.g. Lewis 1979 for a view of propositions (and attitudes) according to which not every case of knowledge-that would be a relation between a subject and a proposition). We will thus shun it in the following. 2. As usual, w e use ‘ wh’ as a place-holder for the interr ogative pro-forms ‘who’, ‘whom’, ‘whose’, ‘what’, ‘which’, ‘why’, ‘when’, ‘where’ etc. For the reasons mentioned in fn 3, we don’t wish to include ‘how’ in this list (when it is used to r efer to knowledge-how).

218

considerations pr esent a challenge to a r eduction of kno wledge-wh to knowledge-that (without committing ourselv es to its fi nal unfeasibility, we will touch later upon some more, hitherto unexplored, difficulties that this reduction must face). Our leading question in this paper will concern the pr ospects for reduction to kno wledge-that of y et another form of kno wledge which is apparently distinct fr om it: knowing how to do something (also kno wn as ‘knowledge-how’).3 Indeed, knowledge-how has been understood by many philosophers as consisting in a practical competence that outr uns every possible knowledge-that, kicking in exactly when kno wledge-that giv es out. For example, it has been assumed b y many philosophers that one can know every relevant fact concerning how one could and should ride a bicycle, and yet still not be in a position to ride a bicy cle: very intuitively, one falls to the gr ound when put on the bicy cle’s saddle not because one does not know that something is the case, but because one fails to put one ’s wealth of knowledge-that into action—one still does not know how to ride a bicycle. This kind of consideration is prominent especially in the hermeneutic tradition, emerging fr om its sustained r eflection on the notorious problem of the gap betw een principles and their applications to particular cases. One may know a certain principle but still fail to apply it correctly in a particular case. Not even knowledge of a particular way to the effect that that way is a corr ect way to apply the principle would seem to guarantee that one will apply the principle correctly—one may still interpret that piece of knowledge as requiring a very different application from the one which it in fact requires. Only knowledge of how to apply the principle would seem to be able to bridge the gap between it and its applications. The reduction of all knowledge to knowledge-that thus arguably requires the truth of the following reductionist thesis concerning knowledge-how: (R) Knowing how to do something consists in (is a species of) knowing that something is the case. Going back to a usage introduced by Ryle 1946, in our debate (R) is also known as ‘intellectualism’ (see Stanley & Williamson 2001), while its anti3. As we explain in section 2, ‘know how’-ascriptions are at least syntactically on a par with ‘know wh’-ascriptions. Still, since we don’t want the terminology to obfuscate important issues that will be relevant to our discussion, w e’re going to use ‘kno wledge-wh’ and its like in a way that does not include knowledge-how, even though, of course, this terminological choice leaves fully open that knowledge-how might be identical to knowledge-wh in all relevant respects.

219

reductionist denial is corr espondingly known as ‘ anti-intellectualism’.4,5 In the follo wing, we intend critically to examine the pr ospects for (R), focussing—by the way of a case study—on a par ticularly promising version thereof which has recently been offered and developed by Stanley & Williamson (2001). In the bulk of the paper, we aim to provide a battery of arguments which show that Stanley & Williamson’s view is untenable, and which do so without actually r elying very much on the peculiarities of their vie w: hence, ev en though, for lack of space, w e won’t attempt to recast our arguments in a mor e general fashion, w e believe that they 4. We should like to str ess that anti-intellectualism is r estricted to ‘know how’-ascriptions taking an infinitive rather than fi nite clause (see section 2). Clearly, Jacques’s knowing how the exam went is no more problematic for (R) than standard knowledge-wh is. We should also like to stress that, even restricted to ‘know how’-ascriptions taking an infinitive clause, anti-intellectualism need not hold an anti-r eductive thesis for every such ascription. For example, it would seem desirable to allo w that J acques’s knowing how to pr ove the completeness of fi rst-order logic does consist after all in a cer tain (possibly very complex) knowledge-that state: if Jacques knows how a proof of the completeness of fi rst-order logic goes (where such a state apparently need not involve anything but a (possibly v ery complex) knowledge-that state, whose content is something of the form: ‘A proof of the completeness of fi rst-order logic goes by first directly inferring Pi from P0, P1, P2… Pi–1, and then dir ectly inferring Pi+j from Pi, Pi+1, Pi+2… Pi+j–1, and then directly inferring Pi+j+k from Pi+j, Pi+j+1, Pi+j+2… Pn+j+k–1…, and then directly inferring the completeness of first-order logic from Pi+j+k…+m, Pi+j+k…+m+1, Pi+j+k…+m+2… Pi+j+k…+m+n’), this would seem enough to establish that he kno ws how to pr ove the completeness of fi rst-order logic. The fact that Jacques might not be able (or kno w how) to produce a token of the pr oof (either by vocalising it, or b y inscribing it, or b y some other means of expr ession), or that he might not even be able (or know how) to “run through it” in his own occurrent thoughts, would seem to be neither her e nor ther e. While recognising that the issue invites fur ther reflection, we emphasise that what matters for anti-intellectualism is the thesis that at least some cases of knowledge-how are not reducible to cases of knowledge-that—for example, paradigmatic cases involving relatively simple material actions such as knowing how to ride a bicycle, knowing how to swim, knowing how to play football etc. In the following, ‘knowledge-how’ and its like will be implicitly understood as carrying this further restriction. 5. We should like to stress that, on our view, it is an open question whether the same kind of practical competence that the anti-intellectualist reads paradigmatic ‘know how’-ascriptions as targeting is also targeted b y some ‘kno w wh’-ascriptions in infi nitive form (on at least one possible reading). For example, one can kno w every relevant fact concerning when and wher e one could and should stamp one’s feet in order to dance flamenco, and yet still not be in a position to dance fl amenco: very intuitively, one ends up dancing something that looks rather like a waltz not because one does not know that something is the case, but because one fails to put one’s wealth of theoretical knowledge-wh into action—one still does not know when and where to stamp one’s feet in or der to dance fl amenco (nor, of course, does one kno w how to do it). Indeed, one could employ some of the arguments that in the follo wing we’ll use to prise apar t knowledge-how from both knowledge-that and theoretical knowledge-wh to buttress this intuition. We won’t pursue this issue further here, and will henceforth rule out such cases from our understanding of ‘knowledge-wh’ and its like.

220

present an interesting challenge not only to Stanley & Williamson’s specific view, but to a wide range of theories subscribing to (R). The plan for the rest of the paper is as follows. Section 2 introduces the essential features of S tanley & Williamson’s reductionist view. Section 3 attacks their assimilation of kno wledge-how to kno wledge of an answ er, exploiting the case of negativ e answers. Section 4 unco vers a disanalogy between knowledge-how and kno wledge-that concerning their diff erent behaviour in relation to singular contents. Section 5 argues that, as a matter of principle and contrary to knowledge-that states, knowledge-how states are not available to a subject as a basis for acquiring inferential knowledge. Section 6 offers evidence against Stanley & Williamson’s presupposition that knowledge-wh is reducible to knowledge-that, rejecting a possible attempt at conciliating the data with the presupposition. Section 7 briefly recapitulates our case and draws some conclusions about the prospects for Stanley & Williamson’s reduction and, more generally, for intellectualism. 2. A reductionist strategy In their seminal (2001), Stanley & Williamson have attacked anti-intellectualism claiming that it is in tension with contemporar y linguistic theory. They note that, from a syntactical standpoint, ‘know how’-ascriptions simply belong to the wide class of reports (so-called ‘infinitive indirect-question reports’) of the form ‘s S WH to A’, where ‘s’ is schematic for an appropriate noun phrase, ‘S ’ for an appropriate attitude verb, ‘WH’ for an interrogative pro-form and ‘A’ for an appropriate infinitive verb phrase. Examples are: (1a) Rudolf remembers whom to thank (1b) The mafiosi knew where to find the money (1c) Some students learnt why to vote for the Communist Party. As they recall, the accepted syntactic structure of infinitive indirect questions is:6 (2a) Rudolf remembers [whom PRO to thank t] (2b) The mafiosi knew [where PRO to find the money t] 6. We follow Stanley & Williamson in ignoring here many syntactic niceties. Square brackets indicate clausal boundaries.

221

(2c) Some students learnt [why PRO to vote for the Communist Party t], where ‘PRO’ is an unpronounced pronoun and ‘t’ the trace of the movement of the relevant interrogative pro-form from its original position. The syntactic structure of infinitive indirect-question reports is thus no different in the relevant respects from that of so-called ‘finite indirect-question reports’. The latter are of the form ‘s S WH P’, where ‘P’ is schematic for a sentence whose main verb phrase is fi nite. Examples are: (3a) Rudolf remembers whom Gustav had to thank (3b) The mafiosi knew where the policemen would fi nd the money (3c) Some students learnt why Karl and Friedrich voted for the Communist Party. Hence, the syntactic structure is ultimately not relevantly different from that of propositional-attitude reports (that is, reports taking a ‘that’-clause): in all three cases, the attitude verb is analysed as taking a sentential complement. Stanley & Williamson proceed then to apply to ‘know how’-ascriptions the main features of contemporary semantic accounts of indirect questions.7 According to such accounts, an indir ect-question report of the form ‘ s S WH P ’ (with ‘P ’ in either infi nitive or finite form) is tr ue iff ,8 for ever y (some) correct answer Q to question9 as to WH P, s S that Q.10 For example, 7. For simplicity and concreteness, we follow Stanley & Williamson in assuming Karttunen (1977)’s w ell-known theor y of indir ect questions. As they note, nothing impor tant for our discussion hinges on the specifics of this choice. We also ignore some features of their proposal that are not relevant to our discussion. 8. From now on, some ‘iff ’-claims will crucially occur in our discussion of intellectualism. This is however purported to be as strong as a constitutive thesis—claiming that knowledge-how consists in (is a species of) knowledge-that—and not simply anextensional or modal thesis—claiming that there is some sort of correlation between the instantiation of knowledge-how states and the instantiation of knowledge-that states. The relevant ‘iff ’-claims will thus have to be understood accordingly as carrying the force of a constitutive claim (even though our arguments against some of these claims, if successful, will in eff ect undermine even the weaker correlation readings). 9. Throughout, w e’ll be rather cav alier in our use of ‘ question’ and ‘ answer’, sometimes referring to the non-linguistic entities entertained when someone asks something and someone answers something, sometimes referring to the abstract linguistic entities that are usually tokened when someone asks something and someone answ ers something, sometimes r eferring to the speech acts of asking and answering. 10. Note how such a semantics would quite generally reduce knowledge-wh to knowledgethat. See our cursory comments on this in section 1 and our discussion of recalcitrant data that such a reduction has to face in section 6.

222

‘George remembers who was at the party’ is true iff, for every (some) correct answer Q to question as to who was at the par ty, George remembers that Q; ‘Vinny knows where to buy good Italian food’ is true iff, for every (some) correct answer Q to question as to where to buy good Italian food, Vinny knows that Q (some r eports, like the former , seem most natural under the ‘every’-reading of their truth conditions; others, like the latter, under the ‘some’-reading, see Hintikka 1976). 11 Stanley & Williamson’s application has it then that a ‘know how’-ascription like: (4) Diego knows how to play football is true iff , for some corr ect answer Q to the question as to ho w to play football, Diego knows that Q (they observe that for ‘know how’-ascriptions the ‘some’-reading is very plausible). But what does Diego know when he knows any such Q? Stanley & Williamson observe that, for ‘know how’-ascriptions, ‘PRO’ is most naturally interpreted as referring to the subject of the attitude, and the modality expressed by the infinitive as being possibility-like. They also assume that, just as one knows a correct answer to the question as to who was at the par ty iff , for some person p that was at the par ty, one kno ws that p was at the par ty, so one knows a correct answer to the question as to how P iff, for some way w in which P, one knows that [P in w]. Under these assumptions, S tanley & Williamson’s application has it that (4) is true, iff , for some way w in which he could play football, D iego knows that he could play football in w.12 Moreover, there is a possible situation 11. As these examples might suggest, a lot must be packed into ‘correct’ in order to achieve an at least prima facie plausible account, and of course a par ticular semantic theory of indirect questions which follo ws this v ery broad style of analysis can only be ev aluated for adequacy once this and other important details have been settled. In particular, ‘correct’ had better mean something str onger than ‘ true’, for other wise G eorge and Vinny could far too easily kno w, respectively, who was at the party and where to buy good Italian food, since they know, respectively, that whichever person who could tr uly be believed to have been at the par ty was at the party and that one can buy good I talian food wherever one can buy good Italian and Mexican food. How best to specify this str onger sense of ‘correct’ is a notoriously v exed issue (see Boër & Lycan 1986). We’ll discuss in Section 3 another respect in which the delimitation of the set of “correct” answers is very important for our topic. 12. ‘Could’? As Stanley & Williamson are well aware, there are compelling examples which show that kno wledge-how does not r equire the kind of ability usually denoted b y ‘can’: for instance, a football player whose legs have been chopped off may still know how to play football, even though, in the most natural sense, he can no longer do so (this kind of example is due to Ginet 1975). S imilarly, in the most natural sense, it is no longer the case that he could do

223

where Diego is a complete disaster with a football, but, having attentively watched Pelé play football on the television, pr esumably knows, of the way Pelé played, that he could play football in that way. In order to avoid being committed to saying that, in such a situation, (4) is true, Stanley & Williamson add the epicy cle that a ‘kno wledge how’-ascription like (4) either semantically or pragmatically implies that, for some wayw in which he could play football, D iego knows that he could play football in w by entertaining this proposition13 under a so-called p‘ ractical mode of presentation’—that is, a mode of pr esentation that bears a privileged connection to action just like the first-personal mode of presentation associated with ‘I’ has long been argued to do (P erry 1979 is one of the loci classici for the latter claim). G iven this fi nal analysis, knowledge-how is claimed to have been revealed to be simply a species of knowledge-that,14 and (R) is sharpened as: so. Now, suppose that Diego is such an unfor tunate football player. Then, in the most natural sense, it is not the case that, for some way w in which he could play football, Diego knows that he could play football in w, for the simple r eason that, for no w, Diego could play football in w (because unlucky D iego could no longer play football at all) and kno wledge-that is factive. But (4) is arguably still true in such a situation. The lesson is that ‘could’ had better be read in a slightly technical and idealised sense if Stanley & Williamson’s theory is to stand a chance of being correct. But Stanley & Williamson themselves have objected to Ryle (1946); (1949)’s antiintellectualist theory by (rather uncharitably) reducing Ryle’s rich and nuanced talk of a“bilities”, “skills”, “capacities” etc. to the kind of non-idealised ability that a person in D iego’s situation would lack. O nce the R ylean anti-intellectualist is allo wed to use the slightly technical and idealised sense of ‘ability’-talk that Stanley & Williamson themselves need in order to save their theory from easy counterexamples, their objection to Rylean anti-intellectualism evaporates. 13. Throughout, we’ll be assuming for defi niteness a broadly Russellian theory of propositions. All the discussion could easily be r eframed in the other main theoretical frameworks for thinking about propositions. 14. As Crispin Wright has emphasised to us in conversation, there is a strong worry that the epicycle about “practical modes of presentation” risks to give the game away. For it implies that, in the fi nal analysis endorsed by Stanley & Williamson (at least taking the option of semantic implication, which seems in other respects much more preferable, given the problem raised by disastrous Diego), knowledge how is not simply any old kno wledge-that. It is knowledge-that plus a cr ucial additional featur e in the subject ’s cognitive system—knowledge that something is the case plus a privileged connection betw een this state and action . Is this something the antiintellecetualist needs to disagree about? Certainly, anti-intellectualism can allow that some form of knowledge-that is a necessar y condition for (and possibly, in some sense, “part of ”) knowledge-how! There is also a worry about Stanley & Williamson’s knowledge-that requirement once “practical modes of presentation” are introduced (see Zardini 2007). In the following, however, we propose to ignor e these pr essing worries, pursuing other av enues of tr ouble for Stanley & Williamson‘s theory, which relate more to semantic and epistemic (rather than mental) aspects of our topic.

224

(RSW)

s’s knowing how to F consists in, for some way w in which s could F, s’s knowing (under a practical mode of presentation) that she could F in w.

3. Negativity In S ection 2 w e hav e seen that, accor ding to contemporar y semantic accounts of indirect questions, a ‘know wh’-ascription like: (5)

George knows who was at the party

is true iff , for ev ery (some) corr ect answer Q to the question as to who was at the party, George knows that Q. While, as applied to ‘know how’ascriptions and barring the worries mentioned in fn 14, this v ery broad style of analysis, if corr ect, would presumably already suffice to establish (R), it is legitimate to ask about the admissible values of ‘ Q’ in or der to assess the analysis’ fi nal plausibility. As has emerged in S ection 2, it’ll be useful to examine whether the knowledge-of-answer statements: (5KAE) George knows every correct answer to the question as to who was at the party (5KAS) George knows some correct answer to the question as to who was at the party are equivalent with the corresponding knowledge-of-objects statements: (5KOE) For every person p that was at the party, George knows that p was at the party (5KOS) For some person p that was at the party, George knows that p was at the party (we have already had occasion in Section 2 to note the ambiguity between the ‘every’- and the ‘some’-reading). We have serious doubts about the correctness of almost all these putative equivalences. By way of building up to our main argument in this section, we star t with the putativ e entailments fr om (5KOE) to (5KA E) and from (5KOS) to (5KAS). We first note that such putative entailments (in particular, the one fr om (5KOE) to (5KA E)) stand a chance of being 225

correct only given an appropriately restrictive understanding of what the range of corr ect answers to the question as to who was at the par ty is. Such an appropriately restrictive understanding is required in order to rule out counterexamples of the follo wing form: suppose that, inter estingly enough, all and only the former US Presidents (those who are still alive!) were at the party, but that George, even though knowing, of each person (former US President) p that was at the party, that p was at the party, does not know that p is a former US P resident. Then, under some extr emely natural assumptions, George does not know that all and only the former US Presidents were at the party, even though that all and only the former US Presidents were at the party may contextually be an ex cellent answer to the question as to who was at the par ty. However, sometimes what would hav e to be an appr opriately restrictive understanding of what the range of corr ect answers to the question as to who was at the par ty is in or der for the entailments fr om (5KOE) to (5KA E) and fr om (5KOS) to (5KA S) to be guaranteed is actually of dubious adequacy. Alr eady with r egard to the pr evious example, ther e are possible contexts where the interest in whether former US Presidents were at the par ty is so salient that an utterance of (5) in such contexts would be infelicitous and indeed false. To strengthen this point, consider a situation where the party is a masquerade with, unbeknownst to George but under his carefully observing eyes, Bill Clinton attending masked as Abraham Lincoln. Suppose further that the day after George is watching a documentary on former US P residents and G eorge is told that one of them was attending the party the night before. Suppose finally that, for no other person p that was at the party, George knows that p was at the party. It is very plausible that George’s careful observation of the person masked as Lincoln gives him knowledge of Clinton that he was at the party, even though, in the context of the day after the par ty, George knows no correct answer to the question as to who was at the par ty (and even though an utterance of (5) in the context of the day after the party is infelicitous and indeed false). Even more straightforward and related to our upcoming argument from negativity, the putative entailment from (5KOE) to (5KAE) is plainly incorrect in a case where no one was at the party and George does not have the faintest idea about the party’s attendance (because, say, being completely disconnected from the relevant people and groups involved in the par ty, he is not ev en aware of the fact that that par ty had been planned). I n such a case, (5KOE) is vacuously true (since no person was at the par ty), 226

but (5KA E) (and (5)) is clearly false (since G eorge has no clue about the party’s attendance). Notice that this is so independently of one’s favoured representation of the quantificational structure of (5KOE), either as involving a unary universal quantifier along the lines of: (5KOEU) For every person p (p was at the party Š George knows that p was at the party) or as (mor e plausibly) inv olving a binary universal quantifi er along the lines of: (5KOEB) For every person p (p was at the par ty, George knows that p was at the party) (see the next paragraph for discussion of logically str onger, existentially committing quantifi ers). Although we have no space to under take a full survey of the options av ailable, w e should mention that v acuous tr uth will still r esult on most r epresentations (extensional or intensional) of the conditionality possibly implicit in (5KOE), and that many intensional representations under which (5KOE) would come out false are dubiously adequate as general r epresentations of the kind of statements to which (5KOE) belongs (in the sense that they have the opposite flaw of also counting as false many statements of the kind to which (5K OE) belongs that should rather come out true). Opting e.g. for a representation in terms of strict implication (denoting it with ‘<’) would certainly make: (5KOE<) For every person p (p was at the party < George knows that p was at the party) false in our last example (since at least one personp is such that it is metaphysically possible that [ p was at the par ty and G eorge does not kno w that p was at the party]), but the existence of the very same metaphysical possibility would also make (5KOE<) false even if George knew full well who was at the party. It is important to observe how the last counterexample to the putative entailment from (5KOE) to (5KA E) cannot be avoided by simply beefi ng up (5KOE) in such a way that it carries an existential commitment to someone’s having been at the party, as in:

227

(5KOES) Some person was at the party and, for every person p that was at the party, George knows that p was at the party. This is so because, ev en if—setting aside the earlier worr y illustrated b y the Clinton/Lincoln example—it may no w be granted that (5K OES) is strong enough as to entail (5KAE), (5KOES) is in effect so strong as not be any longer plausibly entailed b y (5KA E). For suppose that no one actually was at the party but that this time George knows full well about this circumstance. Then George does have an excellent answer to the question as to who was at the par ty—namely, ‘No one’! We may fur ther suppose that George knows of all the people who possibly could have gone to the party (Janet, Mary, Moira etc.) why they did not go to the par ty, and so that he is also in a position to give an even more informative answer along the lines of ‘Neither Janet nor Mary nor Moira etc. did, because such-andsuch was the case ’. Indeed, whatever the range of corr ect answers to the question as to who was at the par ty may exactly amount to in this case, the scenario can clearly be enriched (if it need be) in such a way as to hav e George in possession of all such answ ers. In such a situation, (5KA E) is true but (5K OES) is false, since its fi rst existentially committed clause is not satisfied.15 15. As suggested by e.g. Groenendijk & Stokhof (1982), in some contexts the truth of (5) may well demand a stronger condition than (5KOE), requiring knowledge of both positive and negative cases: (5KOE±) For every person p that was at the par ty, George knows that p was at the par ty, and, for every person p that was not at the party, George knows that p was not at the party. For instance, if there is an early-morning lectur e the day after the par ty and it is impor tant to determine how many handouts to print out, but G eorge falsely believes that Janet, Mary and Moira were at the party (while he also knows of every person that was at the party that she was at the par ty), an utterance of (5) could naturally be interpr eted as false, which pr ovides good evidence that its truth condition is something along the lines of (5K OE±) (of course, with suitable restrictions on the domain of its universal quantifiers). (5KOE±) is not existentially committing to someone’s having been at the par ty, and so does not fall afoul of our criticism to (5K OES). Nor does it fall afoul of our v acuous case where no one was at the par ty and George does not have the faintest idea about the par ty’s attendance. Still, we submit that (5KOE±) represents an implausibly strong truth condition for many uses of a sentence like (5): for instance, if all that matters is to thank people for coming to the par ty the day after the par ty and G eorge knows of every person that was at the party that she was at the party (while he is also agnostic about a couple of people who actually did not show up), an utterance of (5) is most naturally interpreted as true. Moreover, (5KOE±)’s analogues for indirect questions fronted by other interrogative proforms are even less plausible truth conditions in most contexts of use. nI any event, we can afford to remain neutral here about the extent to which (5KOE±) and its analogues provide correct truth

228

Moreover, not only is (5KAE) true in the situation just described—in it, (5) itself is arguably true. As in many cases of untypical truth, this semantic judgement may be obfuscated by some pragmatic noise.Yet, it can be made to emerge in its strength in various ways. Firstly, and most simply, all the natural negations of (5)—such as ‘George does not know who was at the party’, ‘It is not the case that G eorge knows who was at the par ty’, ‘The following is not tr ue: George knows who was at the par ty’—just sound plainly false. Secondly, the implication that someone was at the party can be explicitly cancelled in such a way that (5) becomes ev en more clearly acceptable, as in: ‘G eorge knows well who was at the par ty: he has seen with his own eyes that no one was there!’, ‘I don’t know whether anyone was at the party, but ask George—he knows who was there’ and ‘No matter whether one, or ten, or billion or no people at all are at the party: George always knows who is ther e’. Thirdly, (5) is straightfor wardly entailed b y some perfectly acceptable implicit comparison. Suppose for instance that we are comparing George, whose epistemic state is as just described, with James, who does not hav e the faintest idea about the par ty’s attendance. Given such comparison, ‘G eorge is the one who kno ws who was at the party’ is compelling. Fourthly, (5)’s truth is entailed b y the most natural way of connecting indirect questions with dir ect questions and answ ers, which requires that (5) be true (under the e‘ very’-reading) iff George knows every answer to the question as to who was at the par ty (with suitable restrictions on the domain of the universal quantifier). The appeal of such a direct connection is ev en stronger once one considers that statements like ‘George knows the answer to the question as to who was at the party, but ‘George knows who was at the party’ is not true’, ‘George knows that no one was at the par ty, but ‘George knows who was at the par ty’ is not true’ and ‘I know that George knows the answer to the question as to who was at the party, but I don’t know whether he knows who was at the party’ sound plainly contradictory. We encounter the same patterns with what’-, ‘ ‘which’-, ‘when’- and ‘where’-indirect questions.16 conditions for ‘know wh’-ascriptions, since our argument from negativity will go through even under the exacting conditions required by them. There are other truth conditions stronger than (5KOE) which share the relevant features of (5KOE±) discussed in this footnote—the pr oblem of which of these tr uth conditions are suited to interpr et which ‘knowledge wh’-ascriptions is known in the literature as ‘the exhaustiveness problem’ (see Groenendijk & Stokhof 1997). 16. ‘Why’-indirect questions do not seem to align in this respect (with respect to the fourth consideration, arguably not because knowledge of the answer to a ‘why’-question does not imply the tr uth of the r elevant ‘know why’-ascription, but because ther e seems to be no answ er to the question as to why P if it is not the case that P). An utterance of ‘G eorge knows why the

229

Substantially the same pattern is also displayed for ‘how’-indirect questions. Suppose that G eorge’s par tner Elizabeth did not dr ess at all the night of the party but instead went naked straight to bed and that all the relevant details of this circumstance are known to George. Consider then the ascription: (6)

George knows how Elizabeth dressed the night of the party

and the related knowledge-of-answer statements: (6KAE) George knows every correct answer to the question as to how Elizabeth dressed the night of the party (6KAS) George knows some correct answer to the question as to ho w Elizabeth dressed the night of the party. Again, George does hav e an ex cellent answer to the question as to ho w Elizabeth dressed the night of the par ty—namely, ‘In no way ’! Furthermore, he is also in a position to giv e an ev en more informative answer along the lines of ‘In no way—she didn’t dress at all, but had a shower and then went straight to bed’. Indeed, whatever the range of correct answers to the question as to ho w Elizabeth dressed the night of the par ty may exactly amount to in this case, the scenario can clearly be enriched (if it need be) in such a way as to have George in possession of all such answers. In such a situation, both (6KA E) and (6KAS) are true. Moreover, not only ar e (6KA E) and (6KA S) true in the situation just described—in it, (6) itself is arguably tr ue even if it is also in many ways just as infelicitous as (5) is in the situation wher e George knows that no one was at the party. Again, the semantic fact can be made to emerge inarious v ways. Firstly, and most simply, all the natural negations of (6)—such as ‘George does not know how Elizabeth dressed the night of the par ty’, ‘It is not the case that George knows how Elizabeth dressed the night of the party was a success’ seems untrue if George knows that the par ty was not a success and hence that there is no reason in virtue of which the par ty was a success. No analogues of the considerations in favour of the truth of (5) with George knowing a negative answer seem to apply to ‘why’-indirect questions (see also fn 18). We wish to r emain neutral here as to ho w to explain this peculiarity (maybe some version of the semantic-presupposition view to be mentioned later in the text could pr ofitably be applied to this case)—as w e proceed to obser ve, in this r espect ‘how’ seems to follow ‘who’ rather than ‘why’. (Hence, some of the following claims in the text about ‘know wh’-ascriptions are to be understood as r estricted to non-‘why’ cases.) Thanks to Jason Stanley for discussion on this issue.

230

party’, ‘The following is not tr ue: George knows how Elizabeth dressed the night of the party’—just sound plainly false. Secondly, the implication that Elizabeth dressed in some way the night of the party can be explicitly cancelled in such a way that (6) becomes even more clearly acceptable, as in: ‘George knows well how Elizabeth dressed the night of the party: he has seen with his own eyes that she didn’t dress at all!’, ‘I don’t know whether Elizabeth dressed in some way the night of the party, but ask George—he knows how she did ’ and ‘N o matter whether E lizabeth dresses red, or blue, or black, or doesn’t dress at all the night of the party: George always knows how she dresses’. Thirdly, (6) is straightforwardly entailed by some perfectly acceptable implicit comparison. Suppose for instance that we are comparing George, whose epistemic state is as just described, with James, who does not hav e the faintest idea about whether and what E lizabeth dressed the night of the party. Given such comparison, ‘George is the one who knows how Elizabeth dressed the night of the par ty’ is compelling. Fourthly, statements like ‘G eorge knows the answ er to the question as to how Elizabeth dressed the night of the par ty, but ‘George knows how Elizabeth dressed the night of the par ty’ is not true’, ‘George knows that Elizabeth dressed in no way the night of the par ty, but ‘G eorge knows how Elizabeth dressed the night of the party’ is not true’ and ‘I know that George knows the answ er to the question as to ho w Elizabeth dressed the night of the party, but I don’t know whether he knows how Elizabeth dressed the night of the party’ sound plainly contradictory. An alternative view on the perceived infelicity of some ‘know wh’-ascriptions in situations wher e the subject possesses a corr ect negative answer would attribute to these ascriptions, and to the corr esponding interrogations, a false semantic presupposition (such vie w has been explicitly taken for direct questions b y Katz (1972) and K eenan & H ull (1973) and is actually briefl y endorsed b y Stanley & Williamson (2001), 421, fn 19 at least for the case of ‘kno w who’-ascriptions). Such an account could be supported by appealing to cases which pr esent a more acute infelicity than those we have just discussed, like e.g. ‘George knows how Elizabeth passed the exam’ or ‘George knows which numbers are a counterexample to Fermat’s Last Theorem’ (‘how’ is not unique in exhibiting such cases). On this vie w, utterances of (5) and (6) in a situation wher e the subject possesses a correct negative answer would be relevantly akin to a pr esent utterance of ‘George met the present king of France’ (if you don’t like this example, plug in your favourite case of semantic presupposition). However, it seems to us that these cases are clearly different from the point of view of 231

intuitive acceptability. Even ‘George knows how Elizabeth passed the exam’ sounds much better than ‘G eorge met the pr esent king of F rance’ (with ‘George knows when the present king of France was born’, on the relevant narrow-scope reading of the definite description, lying in between).17 More importantly, an attribution of semantic pr esupposition would be v ery hard to motivate given that the same indirect questions clearly don’t carry any such pr esupposition when embedded under other v erbs like ‘hav e a view about’, ‘bet on’, ‘depend on’ etc. 18 In conclusion, given the various theoretical vir tues of the kind of semantic theor y of indir ect questions we have been wor king with and giv en the o verwhelming wealth of data we’ve just presented in favour of the truth of ‘know wh’-ascriptions with known negative answers, we think it clear that the best explanation for the infelicities in question cannot be but pragmatic rather than semantic (the same conclusion is reached, among others, by Groenendijk & Stokhof (1982); (1997)). It is an inter esting question how exactly to characterise the pragmatic mechanisms involved—here, however, we shall not speculate further about this (see e.g. Kar ttunen & Peters (1976) for a specifi c proposal). That being said, our upcoming argument from negativity is compatible with a version, which we take to be the least implausible, of the semanticpresupposition view. On the v ersion of this vie w that w e have in mind, direct and indirect questions could take on diff erent semantic presuppositions depending on the conv ersational context. This would happen via the (direct or indirect) question’s somehow contextually selecting a range of available answers. If one logically possible answ er is not selected, its negation is semantically presupposed by the question. Thus, while in some contexts ‘No one’ would be a corr ect answer to the question ‘ Who was at the party?’, in other contexts it would rather be a way of “rejecting the question” (analogously for indirect questions). If you should be attracted by such a vie w, just assume that the cases w e are discussing ar e cases in which the question does not carry any relevant semantic presupposition. With so much b y way of stage setting, our argument fr om negativity should begin to emerge. F or, while it might be granted that the pr evious considerations in fav our of the tr uth of ‘kno w wh’-ascriptions with 17. Thanks to Herman Cappelen for suggesting to us to consider embeddings into indirect questions of expressions carrying semantic presuppositions. 18. Again, ‘why’-indirect questions behav e diff erently here: for example, an utterance of ‘Whether Elizabeth went to the party depends on why the party was a success’ seems untrue if the party was not a success.

232

known negative answers do not strictly speaking impugn the entailment from (5KAE) to (5KOE) (since, in the situation where George knows that no one was at the party, (5KOE) is vacuously true), they do show that the converse entailment fails, and, mor e importantly, that both (5KA E) and (5KOE) can be true in a situation where no one was at the party (with, on the one hand, the tr uth of (5KA E) requiring a substantial cognitiv e feat by George and, on the other hand, the tr uth of (5KOE) being vacuously guaranteed by the facts of the matter). E ven more importantly, they also show that the entailment from (5KAS) to (5KOS) fails: while George has some correct answer to the question as to who was at the party (‘No one’), there is no person p such that George knows that p was at the party. And, in such cases, we have argued that both theory and data couple the truth of (5) with the truth of (5KA S) rather than (5KOS). Now, as we have already remarked in Section 2, some ‘know wh’-ascriptions lend themselves better to the ‘every’-reading while some others lend themselves better to the s‘ome’-reading. Moreover, as Stanley & Williamson themselves seem to acknowledge, an analysis of ‘know how’-ascriptions in terms of ‘know wh’-ascriptions had better use the ‘some’-reading on pains of leaving precious little knowledge-how around. If in order for Diego to know how to play football he has to kno w, of each and ev ery way w in which he could play football, that he could play football in w, then, even taking into account possible reasonable restrictions on the domain of the universal quantifi er, a standard for knowledge-how would hav e been set that even el pibe de oro could hardly ever satisfy. It is thus quite essential to the success of S tanley & Williamson’s analysis that the ‘some’-reading is adopted at least as default for ‘kno w how’-ascriptions. This, however, makes the analysis par ticularly prone to the risk of the negativity glitch that has just been unearthed. Indeed, the analysis is in effect fatally affected by the negativity glitch. For concreteness, suppose that G eorge knows full w ell about all the r elevant details concerning the impossibility of squaring the circle. Consider then the ascription: (7)

George knows how to square the circle,

the related knowledge-of-answer statements: (7KAE) George knows every correct answer to the question as to how to square the circle 233

(7KAS) George knows some correct answer to the question as to ho w to square the circle and the corresponding knowledge-of-objects statements: (7KOE) For every way w in which he could squar e the circle, George knows that he could square the circle in w (7KOS) For some way w in which he could squar e the circle, George knows that he could square the circle in w. If we were to treat (7) like a ‘knowledge wh’-ascription, then, as we have argued (and assuming as default the ‘ some’-reading), (7) would hav e (7KAS) as its truth condition. Unfortunately, given the facts of the matter, George does have an excellent answer to the question as to how to square the circle—namely, ‘In no way’! Furthermore, he is also in a position to give an even more informative answer along the lines of ‘In no way, since, as von Lindemann has proved, π is transcendental’. Indeed, whatever the range of corr ect answers to the question as to ho w to squar e the cir cle may exactly amount to in this case, the scenario can clearly be enriched (if it need be) in such a way as to hav e George in possession of all such answers. In such a situation, (7KA S) (and (7KA E)) is true (with (7KOE)’s being vacuously true and (7KOS)’s being false), and so, if semantically on a par with other ‘know wh’-ascriptions, (7) would have to be true as well. Unfortunately, in no way can (7) possibly be true—George does not know how to square the circle!19 It should be clear that the disaster is not av oided by switching, somehow arbitrarily, from the ‘some’-reading to the ‘every’-reading: given that, whatever the range of correct answers to the question as to how to square the circle may exactly amount to in this case, the scenario can clearly be enriched (if it need be) in such a way as to hav e George in possession of all 19. J onathan Schaffer has suggested to us that the question ‘H ow could you square the circle?’ could in some contexts carry the false semantic presupposition that there is a way to square the circle: hence, in such contexts, George could not know the answer to the question (since, in such contexts, every admissible answer would be positive). As we have already discussed in the text, we are rather sceptical about the pr ospects of ever counting such pr esuppositions as semantic. Be that as it may, even on this view there will also be contexts in which the question does not carry the false semantic presupposition (for example, if someone were offered a multiple choice of answers including ‘In no way. It is impossible to square the circle’). We think it clear that in these contexts, while it is true that the subject knows the answer to the question, attributing to the subject knowledge of how to square the circle would still be absurd.

234

such answers, (7KAE) will be true in some extension of the scenario—thus forcing Stanley & Williamson’s analysis to admit the truth of (7)—while, in such an extension, George will still not know how to square the circle. True, (7KOS) is false in all these pr oblematic situations, since it r equires the existence of a way in which George could square the circle but no such way exists. Could Stanley & Williamson then simply by-pass the intuitive analytical step constituted by (7KAS) and rather opt for a straightforward analysis of (7) in terms of (7K OS)? Indeed they could. B ut, no matter what the independent merits of such analysis might be, this move would come at a v ery high dialectical price in our context, for it would completely undercut Stanley & Williamson’s only argument in favour of their analysis—namely, its conformity with contemporary semantic theory (and with general data concerning ‘kno w wh’-ascriptions). Knowledge-how is not knowledge of a correct answer. Finally, in contrast to knowledge-that and knowledge-wh, it is interesting to note that an analogous requirement that some way exist is present for a wide variety of abilities, like e.g. those variously denoted by ‘can’: if George can square the circle, then there is some way of squaring the circle (just like if a car can u r n 200 km/h, then there is some way of running 200 km/h). The existence requirement is thus an impor tant aspect in which, against the intellectualism encapsulated in (RSW), knowledge-how falls in the realm of practice rather than in the realm of theory. 4. Singularity A prominent feature of propositional attitudes like knowledge-that, beliefthat, desire-that etc. is their capacity for what have come to be known as singular contents.20 As is w ell-known, an exhaustive specifi cation of what these exactly amount to is a matter of great controversy in the philosophy of mind and language, but a minimal, fairly neutral and perforce sketchy characterization will suffi ce for our purposes. Let us say that a pr opositional attitude has a singular content iff it is directly about some particular objects. Of course, the crucial work in this characterization is being done by the phrase ‘directly about’, which, as it stands, is not much more than a place-holder for a fuller articulation of the relevant intentional relation.21 20. We’ll be using ‘content’ and its like interchangeably with ‘proposition’ and its like. 21. We stress that our understanding of direct aboutness is deliberately minimal (and hence hopefully sufficiently neutral). In particular, we don’t wish to presuppose any strict correlation

235

Still, we take it that the philosophically trained r eader has an intuitiv e understanding of and good judgement about such relation, and that these will suffi ce in what follows. To calibrate intuitions, on such understanding, a piece of knowledge most accurately expressed by an utterance of the sentence ‘Carla is gorgeous’ (where ‘Carla’ is hereafter used as an ordinary proper name referring to the wife of the French President in 2008)—i.e. the piece of kno wledge that Carla is gorgeous, which is typically arriv ed at by perceiving Carla and its relevant physical features—is singular. It is directly about a par ticular individual, Carla B runi. On the other hand, a piece of kno wledge most accurately expr essed b y an utterance of the sentence ‘The wife of the F rench President in 2008 is gorgeous ’ (where the definite description ‘The wife of the French President in 2008’ is used “attributively”, in Donnellan (1966)’s sense)—i.e. the piece of knowledge that the wife of the French President in 2008 is gorgeous, which is typically arrived at by hearing or r eading that the same person who is the wife of the French President in 2008 is also a gorgeous person—is not singular . Even though ther e is a par ticular individual, Carla B runi, that uniquely satisfies the description, that piece of knowledge is not directly about her, consisting rather in the general information that whoever is the wife of the French President in 2008 is also gorgeous: such information is so much about Carla Bruni as it is about Juliette Binoche. Arguably, propositional attitudes with singular contents pr esent spectacular failures of closur e principles under known logical consequence. Talking with you about your new acquaintance Carla, who—unbeknownst to him—is in fact Carla Bruni, Elia might know from your testimony that, if Carla is gorgeous, her husband Nicholas will be jealous of her. Elia might also know by looking at pictures in newspapers and magazines that Carla is gorgeous, but these two pieces of kno wledge (plus his kno wledge that the proposition that Carla is gorgeous and the proposition that, if Carla is between direct aboutness at the level of thought and what has been called ‘direct reference’ at the level of language (that is, very roughly, the property that an expression might have of picking out its referent without the mediation of a condition that the referent satisfies, see Kaplan (1989)). Direct aboutness is thus compatible with the object’s being thought about with the mediation of a (possibly sui generis) condition that the object satisfies (like for example a “singular sense”, see McDowell (1977)). Nor do we wish to equate without further ado propositional attitudes towards singular contents with so-called “ de re” attitudes: for example, a typical I talian peasant’s belief that Bangkok is an Asian city, expressed by using the proper name ‘Bangkok’, is directly about Bangkok, but it’s not clear whether the peasant is thereby believing of Bangkok that it is an Asian city (or, possibly equivalently, whether the peasant is ther eby believing Bangkok to be an Asian city). De re thought might have stronger access requirements than merely singular thought.

236

gorgeous, her husband Nicholas will be jealous of her entail the proposition that Nicholas will be jealous of Carla) will not put him in a position to know22 that Nicholas is jealous of Carla by drawing the relevant modus ponens inference. This is so, we may suppose, even though Elia is an acute logician. Thus, arguably, the following principle of closure of knowledge-that under known logical consequence fails: (CKTK) If s knows that P0, that P1, that P2… and s knows that ¢P0², ¢P1², ¢P2²… entail ¢Q²,23 then s is in a position to know that Q once we allow singular contents as values of ‘P0’, ‘P1’, ‘P2’…24 Equally inter esting failur es of closur e principles under obvious logical consequence are present once one pays due attention to a pr ominent feature of singular contents, namely that one cannot entertain any propositional attitude towards them simply on the basis of , roughly speaking, one’s general linguistic and conceptual competence: in addition, one needs to be in some non-trivial kind of cognitive relation to the objects that they are directly about. For example, in or der for one to enter tain the singular content that Carla is gorgeous, it is not suffi cient that one be able in some sense to single out Carla B runi, for instance b y means of the defi nite description ‘the wife of the French President in 2008’—one needs to be in some non-trivial kind of cognitiv e relation to Carla Bruni, such as the one aff orded by the competent use of the name ‘Carla ’ as ultimately grounded in Carla Bruni herself, or the one afforded by seeing Carla Bruni on the str eet, or the one aff orded by Nicholas’ talk about his wife etc. 25 22. Throughout, b y ‘being in a position to kno w that P ’ w e’ll mean ‘being in such an epistemic position that only a straightforward reflection is still needed for one to come to know that P—that is, if anything is still needed at all’ (see Williamson 2000, 95). Admittedly, such a notion remains very vague, but will do for our purposes. 23. We use ‘ ¢M ²’ to denote the pr oposition denoted b y the phrase ‘ the pr oposition that M ’. 24. The example might be r esisted if one adopts an extr emely fi ne-grained individuation of singular contents, such that the content expr essed by an utterance of the sentence ‘Carla is gorgeous’ in the context of our conversation about your new acquaintance is different from the content expressed by an utterance of the v ery same sentence in the context of my r eading and thinking about celebrities (even though the name ‘Carla’ refers to the very same person in both contexts). Some Fregean approaches to the famous Paderewski case of Kripke (1979) do adopt such an individuation. An assessment of the ultimate tenability of this extr eme measure lies beyond the scope of this paper and is quite unnecessar y for the argument of this section. 25. There are at least two factors on which one’s ability to entertain a singular content about Carla Bruni (or anything else!) depends. O ne factor concerns the fine-grainedness of singular

237

Consider a Scotsman and an Inuit, Calum and Kirima respectively, such that Calum does not stand in the r elevant cognitive relation to Kirima, and hence cannot enter tain singular contents about Kirima. I nformally speaking, Calum has nev er heard about Kirima. S uppose fur ther that Calum, an impeccable logician, kno ws that ev erything is self-identical. The proposition that everything is self-identical entails, w e may assume, the proposition that Kirima is self-identical, indeed ob viously so b y the fundamental rule of inference of universal instantiation.26 Yet, Calum is not in a position to know that Kirima is self-identical by drawing the relevant universal-instantiation inference, since he lacks the ability to entertain the content that Kirima is self-identical. Thus, arguably, the following principle of closure of knowledge-that under obvious logical consequence fails: (CKTO) If s knows that P0, that P1, that P2… and it is ob vious that ¢P0², ¢P1², ¢P2²… entail ¢Q², then s is in a position to kno w that Q once we allow singular contents as values of ‘Q’.27 contents: the more fi ne-grained one individuates the content that Carla is gorgeous (contrasting it for example with the content that that woman (pointing at Carla B runi on the street) is gorgeous), the more stringent the requirements will be for enter taining the content that Carla is gorgeous. Another factor concerns the strength of the cognitiv e relation to the r elevant objects: the stronger the cognitive relation to Carla Bruni has to be (fr om merely having a description or name for Carla to being closely acquainted with her), the mor e stringent the r equirements will be for entertaining the content that Carla is gorgeous. The exact nature and possible context dependence of the r elevant non-trivial kind of cognitiv e relation need not concern us her e. It will suffi ce for the argument in this section to go thr ough that the cognitiv e relation be of a non-trivial kind or other, so that a subject with a general linguistic and conceptual competence may still fail to be able to entertain a certain range of singular contents (such as those involving Carla Bruni). 26. A lot of distracting noise is caused here by the fact that the (suitably regimented version of the) argument form ‘Everything is F. Ther efore, a is F’, even though valid in classical first-order logic, is arguably strictly speaking inv alid (consider the instance ‘E verything exists. Ther efore, Vulcan exists’—setting aside Meinongian views, the premise is true while the conclusion is false). A free logic of some sort or other should be used in order to detect invalidities generated in natural language by the use of what arguably are non-referring proper names (and other singular terms). However, as these problems are clearly orthogonal to our concerns in this paper, we simply wish to ignore existence issues, assuming that every singular term has a referent. For those who remain suspicious, we should like to point out that a more clumsy example would anyway go through with no assumption about existence: the argument ‘S now is white. Therefore, either sno w is white or Kirima is fishing’ is valid in all standard (bivalent) free logics, and obviously so, and its premise is known by Calum, yet he is not in a position to know its conclusion. 27. Note that it is singular contents as premises that create trouble for (CKT K), while it is

238

The stage is no w ready for lodging our argument fr om singularity. In short, the trouble with (RSW) is that it entails that knowledge-how should suffer from closure failures analogous to the ones kno wledge-that suffers from (in particular, analogous to failures of (CKTO)), whereas there seems to be all evidence to the contrary. We show this by focussing on a peculiarly strong and yet quite widespread closure feature of knowledge-how, involving indefinite descriptions used with universal (or generic) force. Go back to Calum and Kirima. Enrich the scenario by adding that, alas, Calum is also quite a pr oficient and experienced serial killer , very well acquainted with all sorts of murdering methods, ranging from strangling to shooting. In such a situation, the sentence: (8) Calum knows how to kill a man is undoubtedly true. Moreover, the sentence: (9) Calum knows how to kill Kirima is also arguably true, and just for the same reason: the general competence acquired by Calum with the relevant murdering methods, if sufficient for giving him kno wledge of ho w to kill a man in general, is cer tainly also sufficient for giving him knowledge of how to kill Kirima in particular. Of course, on the by far most natural reading,28 (8) would still be true even in a situation wher e an individual exists who, while belonging to the species Homo Sapiens, is for some reason or other invulnerable by all, as it w ere, common or gar den murdering methods master ed by Calum: even though having generalising for ce, the indefi nite description ‘a man’ as occurring in (8) has generic rather than universal force, just as ‘A penguin is black and white ’ is not false in a situation wher e, in addition to the standard wealth of black and white penguins, an extraordinary genetic mutation has caused one single penguin to be red. Thus, in order for the inference from (8) to (9) to be tr uth preserving, we should assume that Kirima is a typical specimen ofHomo Sapiens. That assumption is unproblematic and will be tacitly understood as holding in what follo ws. Under singular contents as conclusions that create trouble for (CKTO). 28. We are inclined to believ e that the most natural r eading is in this case also the only admissible reading (so that on no reading would (8) be false in the situation to be described in the text), but wish to remain officially neutral on this issue as nothing in our argument hinges on it.

239

these conditions, the truth-preserving nature of the inference from (8) to (9) can be even better appreciated by further enriching the Calum/Kirima scenario with the additional assumption that a species exists all of whose specimens are indeed invulnerable by all murdering methods mastered by Calum. Let us pick one par ticular such specimen and call it ‘ Athanatos’. In such a situation, the sentence: (10)

Calum knows how to kill Kirima, but does not know how to kill Athanatos

is clearly true, but its truth evidently requires the truth of (9). 29 Extrapolating fr om this par ticular case, w e can conclude that, quite generally, knowledge-how obeys the following principle of closure of knowledge-how under indefinite instantiation: (CKHII) If s knows how to Aan F and d is a typical 30 F, then s knows how to Ad/an F ,31 29. Since, under the present assumptions, there is nothing peculiar about Kirima (as far as killing goes) and the standar d grounds for asser ting (9)—the general competence acquir ed by Calum with the relevant murdering methods—are just as good grounds for asserting (8) (which, as we’re arguing, is strictly stronger than (9)), an assertion of (9) would usually violate the plausible conversational maxim of asser ting the stronger. Hence, an assertion of (9) would usually implicate that there is something peculiar about Kirima (as far as killing goes). We submit that whatever oddity an assertion of (9) might be felt to have is due to this fact (as can be tested by cancelling the implication). Thanks to Felix Mühlhölzer for discussion of these issues. 30. To stress, the qualification ‘typical’ is supposed to take care of cases in which the object has some special featur e that prevents application of the kno wledge-how to it, thus constituting a counterexample to an unqualified version of the principle (for example, (8) would still be true even if a man existed who is invulnerable to all mur dering methods mastered by Calum). As Dylan Dodd pointed out to us, the specifi cation could perhaps still be deemed insuffi cient, as an object might be a typical F in most respects, while being still peculiar in very few others, where, again, such a “local ” peculiarity happens to pr event application of the kno wledge-how (for example, the man invulnerable to all murdering methods mastered by Calum may well be typical in most other respects relevant to humanity). Such far-fetched cases could be ruled out e.g. by relativising typicality to respects and consequently weakening (CKH II) with the qualifi cation that d is a typical F in the relevant respects. The details of how this can best be done are however clearly inessential to our main line of argument (which ultimately only r equires the truth of (9) against the falsity of (9 SW)). 31. ‘ A’, ‘an F’ and ‘d’ are schematic, respectively, for a bare infinitive verb phrase, an indefinite description and a singular term. ‘Aan F’ is schematic for an infinitive verb phrase with designated occurrences of the indefinite description instantiating ‘an F’. ‘Ad/an F’ is schematic for the result of substituting occurrences of an instance of ‘ d ’ for the designated occurrences of the instance of ‘an F’ in the instance of ‘Aan F’.

240

where the designated occurrences (see fn 31) of the instance of ‘ an F ’ are “safe”. Interesting and challenging as a fully precise characterisation of the operative notion of safety of a designated occurrence may be, for our purposes we can leav e such notion at an intuitiv e level; suffi ce it to say that the restriction to safe designated occurrences is needed in order to rule out blatant falsehoods arising from complex embeddings of the instance of ‘an F ’ in the instance of ‘ A’, such as ‘I f s knows how to make Calum believ e that a man is approaching, and Kirima is a typical man, then s knows how to make Calum believe that Kirima is approaching’. On the other hand, (R SW) offers the following truth condition for (8) in terms of knowledge-that: (8SW) For some way w in which he could kill a man, Calum kno ws that he could kill a man in w, and the following truth condition for (9) in terms of knowledge-that: (9SW) For some way w in which he could kill Kirima, Calum kno ws that he could kill Kirima in w. For the very same reasons as those that determine the failure of (CKTO), (8SW) cannot however possibly entail (9 SW). Letting ‘Murdery’ denote a suitable witness for (8 SW), Calum’s knowledge of the general content that he could kill a man in Murdery by no means guarantees his knowledge of the singular content that he could kill Kirima in M urdery, given that, in our situation, Calum fails to be in the erlevant non-trivial kind of cognitive relation to Kirima, and hence is not even in a position to entertain—let alone know—the singular content about Kirima that Calum could kill her in M urdery. Indeed, these considerations sho w not only that (8 SW) fails to entail (9 SW)—they also show that, in our situation, the latter is false. The former circumstance refutes (R SW), since, accor ding to that principle, (8 SW) should entail (9 SW) (given that, as we have argued, (8) entails (9)). The latter circumstance also refutes (RSW), since, according to that principle, (9SW) should be true (given that, as we have argued, (9) is true). Notice that, for the very same reasons, (8SW) fails to entail (9SW) even if it is assumed that ‘Calum could kill a man in Murdery’ in some sense actually “ entails” ‘Calum could kill Kirima in Murdery’:32 the general failur e of (CKT O) shows that ev en an entailment 32. Such an assumption would hold in eff ect that instantiation of a safe occurr ence of an

241

from ‘Calum could kill a man in Murdery’ to ‘Calum could kill Kirima in Murdery’ (indeed, even an obvious entailment from the former to the latter) will not suffice to generate an entailment from (8SW) to (9SW). We have developed our argument fr om singularity exploiting kno wledge-how’s closure under indefi nite instantiation. H owever, with some more ingenuity, the argument could hav e been centr ed upon a closur e principle under ob vious logical consequence for kno wledge-how. We briefly sketch how such a modified argument would go. In order to state the knowledge-how analogue of (CKTO), we would ideally need a defi nition of what it is for the semantic values (modal possibility-like properties) of a set of ‘to’-infinitive verb phrases to entail the semantic value of another ‘to’-infinitive verb phrase (see S ection 2 for the modal interpr etation of ‘to’-infinitive verb phrases in the constr uctions of interest here). We take it that such a notion of entailment is naturally modelled on the notion of entailment among the semantic values of sentences (propositions) and that, for our present purposes, we have enough of an intuitive grasp of it as to obviate the lack of a precise definition: ‘to eat’ entails ‘to eat or drink’, ‘not to shout’ entails ‘not to shout and play football’, ‘to shut the door’ entail ‘to shut something’ etc. (but ‘to sit’ and ‘to stand’ do not entail ‘to sit and stand’, just like the proposition that it is possible that Calum sits and the proposition that it is possible that Calum stands do not entail the proposition that it is possible that Calum sits and stands). Availing ourselves of this new notion of entailment, w e can then state the follo wing principle of closure of knowledge-how under obvious logical consequence: (CKHO) If s knows how to A0, how to A1, how to A2… and it is obvious that ¢to A0², ¢to A1², ¢to A2²… entail ¢to B², then s is in a position to know how to33 B.34 indefinite description with a term denoting a typical specimen is indeed an entailment. I n so doing, the assumption would go against contemporary orthodoxy in at least two ways: in maintaining that entailment sometimes depends on the semantic values of occurrences of non-logical expressions (such as ‘Kirima’), and in maintaining that entailment sometimes does not r equire necessary truth preservation (since Kirima might well not have been a typical man). 33. We write ‘in a position to kno w how to’ merely to r espect uniformity with (CKT O). Admittedly, contrary to knowledge-that and knowledge-wh, we don’t have a clear idea of what the distinction between knowing how and being in a position to know how could possibly amount to (in the relevant sense of ‘being in a position to know’, see fn 22). We take this to be yet another symptom of the diff erence between knowledge-how on the one hand and kno wledge-that and knowledge-wh on the other hand. 34. We use ‘¢to ) ²’ to denote the property denoted by the phrase ‘the property of possibly ) ing’.

242

(CKHO) is arguably correct, for reasons similar to those for which (CKHII) is correct. Rather than expanding on a defence of (CKH O), we note that it justifies the inference from (8) to: (11)

Calum knows how to kill a man or please Kirima,

while the truth condition in terms of knowledge-that offered by (RSW) for (8), namely (8SW), fails to entail the truth condition in terms of knowledgethat offered by (RSW) for (11), namely: (11SW) For some way w in which he could kill a man or please Kirima, Calum knows that he could kill a man or please Kirima in w. Moreover, (11) is tr ue in the Calum/Kirima scenario while (11 SW) is false. It is important to note that failure of knowledge-that to be closed under indefinite instantiation is mirrored by an analogous failure of knowledgewh to be so closed. We may fur ther enrich the Calum/Kirima scenario by adding that, having diligently follo wed the geography class while at school, Calum knows where an Inuit lives. However, and again for the very same reasons as those that determine the failure of (CKTO), this piece of knowledge-wh on Calum’s part by no means guarantees the other piece of knowledge-wh on his part which would consist in knowing where Kirima lives: clearly, in our situation, failing to be in the relevant non-trivial kind of cognitive relation to Kirima, Calum does not know where Kirima lives (indeed, failing to be in the relevant non-trivial kind of cognitive relation to Kirima, Calum is not even in a position to entertain any hypothesis as to where Kirima lives).35 Similar examples also show that knowledge-wh is not closed under obvious logical consequence (presupposing a suitable notion of entailment for the semantic v alues of ‘ wh’-clauses): informed by a knowledgeable and highly reliable source that Hamish is the one and only one person that likes everything, Calum knows who likes everything, but does not know who likes Kirima and everything else.36 35. A slight v ariation on the example suffi ces to driv e the point home also for infi nitive knowledge-wh (as opposed to finite knowledge-wh): Calum knows where to find an Inuit, but does not know where to find Kirima. 36. Again, a slight variation on the example suffices to drive the point home also for infinitive knowledge-wh (as opposed to finite knowledge-wh): informed by the same knowledgeable and highly reliable source that Hamish is the one and only one person to whom ev erything should

243

Finally, in contrast to their obser ved failures for kno wledge-that and knowledge-wh, it is again interesting to note that analogous closure principles under indefinite instantiation and under obvious logical consequence do hold for a wide variety of abilities, like e.g. those variously denoted by ‘can’: if Calum can kill a man, then he can kill Kirima and, fur thermore, he can kill a man or please Kirima (just like if a poisonous mushr oom can kill a man, then it can kill Kirima and, furthermore, it can kill a man or please Kirima). The validity of (CKH II) and (CKH O) is thus another important aspect in which, against the intellectualism encapsulated in (RSW), knowledge-how falls in the r ealm of practice rather than in the realm of theory. 5. Inference If knowledge-how were a species of kno wledge-that, it would be such as to put a subject in a position also to kno w that cer tain other things ar e the case, just as more usual knowledge-that does—it would serve as basis for acquiring, via deductive and non-deductive inference, further knowledge-that. The undeniable role of knowledge-that as a basis for acquiring further knowledge-that via inference is one of the main factors that makes closure principles for kno wledge-that so appealing. E ven if, as shown in Section 4, the most simple-minded formulations of closur e principles for knowledge-that are subject to straightfor ward counterexamples, we’ll assume that a mor e sophisticated, satisfactor y formulation of a closur e principle of knowledge-that under logical consequence can be achiev ed. We won’t attempt to give such a formulation here, but we’ll simply assume that it exists and call it ‘ Closure’. We stress that, while Closur e has just been introduced in a such a way as not to be subject to the cheap counterexamples of Section 4, it too may well turn out to be in the end untenable in its full generality (recall e.g. the very deep and structural problems raised by preface-paradox situations). S till, something like Closur e does have to hold in a suffi ciently wide range of circumstances if deduction is to preserve its traditional role in the acquisition of new knowledge-that, and this indisputable restricted validity of Closure is all we’ll need to assume in the following.37 be mentioned, Calum kno ws whom to mention ev erything, but he does not kno w whom to mention Kirima and everything else. 37. We’ll continue to talk of “closure” even though the issues involved with the acquisition

244

(RSW) thus pr edicts that kno wledge-how should equally w ell ser ve as a basis for acquiring fur ther knowledge-that. However, this prediction is problematic in a number of ways. Consider fi rst the following conjunction: (ACKT)

John does not know that he is not a brain in a vat (BIV), but he knows that he has hands.

DeRose (1995) dubbed this ‘ the abominable conjunction’, pointing out that it constitutes an unacceptable result of certain denials of Closure (see e.g. Nozick 1981, 167–288). Whatever your ultimate vie w on Closure’s validity in such cases is, we think it clear that (ACKT) at least sounds very bad. This judgement is shar ed b y many epistemologists. F or instance, Stanley (2005, 67–69) objects to epistemological contextualism on the grounds that conjunctions like (ACKT) should sometimes be acceptable if ‘know’ were a genuine context dependent expression, but they never seem to be such. S ubstituting a ‘know wh’-ascription for the second conjunct would not at all make the conjunction more acceptable, as witnessed by: (ACKW) John does not know that he is not a BIV, but he knows where his friend lives. By contrast, consider: (ACKH) John does not know that he is not a BIV, but he knows how to swim. The striking fact is that (A CKH), far fr om being abominable, does not sound bad at all. To be clear, it does not “sound bad” not in the sense that it is not the case that, in most usual situations where the second conjunct is true, the fi rst one is not, for , presumably, most usual situations wher e John knows how to swim are also situations where John knows that he has hands. Rather, (ACKH) does not “sound bad” in the sense that it does not strike us as presenting an internal contradiction, or even a tension, between its two conjuncts—what, quite to the contrary, both (ACKT) and (ACKW) of knowledge-that via deduction cr y out for drawing the usual distinction betw een closure and transmission (see e.g. Wright 2000) and for focussing on the latter rather than the former . Fortunately, the scope of our argument is so restricted as to allow us to disregard the otherwise important differences between these two epistemic properties.

245

do. One appealing explanation of this disanalogy is, of course, that while knowledge-wh at least inv olves relevant knowledge-that, and ther efore similarly licences kno wledge-that extending infer ences, knowledge-how does not involve any relevant knowledge-that, and therefore there is nothing relevant one could infer from it. Such an explanation is however clearly unavailable to a defender of (R SW). Still, of course, the defender of (R SW) can look for alternativ e explanations. As we see things, there are in particular two other explanatory strategies that enjoy some initial plausibility. According to (RSW), John knows how to swim iff , for some way w in which John could swim, John knows that he could swim in w. Let’s call ‘Swimmy’ the particular way that witnesses the (alleged) truth of the analysans. The first explanatory strategy consists then in maintaining that the belief that J ohn could swim in S wimmy is typically implicit, and ther efore not typically av ailable as a star ting point for inferences. This would however be a very weak reply. Firstly, the intuition about (ACKH) would not change at all if it w ere specifi ed that John is a swimming teacher (w e are assuming that swimming teacher J ohn has an explicit belief that he could swim in Swimmy if anyone has). Secondly, the pattern of intuitions relating to (ACKT), (ACKW) and (ACKH) would not change if we were to substitute uniformly ‘is in a position to know’ and ‘is not in a position to know’ for ‘knows’ and ‘does not know’ respectively. SW A further thought in this connection is that the defender of (R ) could here appeal again to practical modes of pr esentation. Since, according to (RSW), the pr oposition that J ohn could swim in S wimmy is kno wn b y John under a practical mode of presentation, it may be argued that John is somehow blind to inferences involving it essentially together with other propositions that he kno ws under a different mode of pr esentation (see Section 4 for an illustration of failure of a closure principle in such cases). If this is supposed to be a fully general principle, however, it would seem ad hoc: sometimes, w e are not blind in this way to infer ences involving propositions that w e know under diff erent modes of pr esentations. For example, one of the authors can eff ortlessly infer fr om the pr oposition that Daniele is not in London (presented under a third-personal mode of presentation) to the (same) pr oposition that he himself is not in London (presented under a fi rst-personal mode of presentation) and vice versa. It would thus be utterly mysterious why , in contrast, pr opositions known under a practical mode of presentation are never similarly available to the speaker’s cognitive system in any (ACKH)-like case, and it would cast doubt on them being known at all in anything like the standard sense of ‘know’. 246

Of course, as we’ve already seen in Section 4, in some cases a difference in modes of presentation does prevent the inference from being available. So, for example, the author just mentioned might be an amnesiac and fail to know (under the relevant mode of presentation) that he himself is Daniele, in which case he will neither be inclined to infer (under the elevant r mode of presentation) the pr oposition that he himself is not in London fr om the proposition that Daniele is not in London nor vice versa. Typically, we submit, the unav ailability of a cer tain inference will be easily explained by failure on the part of the subject to know, or even believe, an identity claim (such as the claim expressed by ‘I am Daniele’). However, no reason seems to be for thcoming as to why a similar failur e should occur in all (ACKH)-like cases. We will come back to issues revolving around different modes of presentation later in this section. Meanwhile, the more promising explanatory strategy consists in maintaining that the proposition that John knows when he knows how to swim, namely the proposition expressed by: (12)

John could swim in Swimmy,

does not after all entail—not even together with trivial propositions such as the proposition that BIV don’t have hands—the proposition that John is not a BIV. This might be because (12) is modal. One interpretation of its truth condition that would block the entailment is given by the metaphysical-possibility claim: (12MP) It is metaphysically possible that John swims in Swimmy. Of course, (12 MP) would be tr ue even if John were a BIV, because ther e surely is a metaphysically possible world—for instance, any world that is similar to (what we take to be) the actual one—in which Swimmy, which we are imagining to be one of the standard ways in which human beings could swim in normal conditions, is a way for J ohn to swim. H owever, to suppose that, accor ding to the defender of (R SW), (12 MP) exhausts the content of John’s knowledge-how leads to quite unacceptable results. For suppose that ther e is a way (let ’s call it ‘S wimmy*’) that is not a way in which John could swim, but is nevertheless such that it is metaphysically possible that J ohn (given appropriate changes in human biology , in the environment, in the laws of nature etc.) swims in it. Suppose also that John practised swimming in a virtual-reality machine specifically devised so as 247

to make Swimmy* a good way of virtually swimming. Suppose finally that John is well aware of all this and does not know of any other possible way of swimming. Then, presumably, John knows (under a practical mode of presentation) that it is metaphysically possible that he swims in wimmy*, S even though he does not know how to swim, which would be very much against the reductionist spirit of (R SW).38 The defender of (RSW) could improve by offering a truth condition for (12) intermediate in str ength between a simple metaphysical-possibility claim and a simple claim about the actual world, for example b y going counterfactual, as in: (12C) If John were to try to swim in Swimmy, he would succeed. This kind of interpr etation cer tainly seems to wor k for some cases: for example, in or der for J ohn to kno w how to driv e his car, it will not be required that he be in a position to know that the car has not been stolen, or destroyed, since the time he par ked it this morning. I ndeed, the car might have been stolen, or destr oyed, and it would still be tr ue that he knows how to drive it. On the present proposal, this is so because, for some way w in which he could drive his car, he knows that, if he were to try to drive his car in w, he would succeed. In effect, to help ourselves to a very influential semantic analysis of counterfactual talk, in the closest possible worlds in which he tries, he does succeed. Still, the counterfactual reading is not generally adequate. Consider again the case of poor D iego whose legs have been chopped off (see fn 12): it is arguably false that, for some way w in which he could play football, if he w ere to tr y to play football in w, he would succeed; y et, as S tanley & Williamson concur, it is still plausibly true that he knows how to play football. It is therefore not clear to us how exactly (12) should be analysed. However, any reading that, as it seems unavoidable in order to avoid (12MP)-like troubles, is strong enough as to be incompatible with the envir onment’s having certain unfavourable properties, will not help with our problem. To see that, think of a possible world, call it ‘NSW ’ (for ‘No-Swimmy World’), in which trying to swim in Swimmy would result in certain drowning: this is so because, while human beings ar e relevantly similar in NSW, water, or whatever “waterish” substance exists in NSW, behaves very differently 38. If not the letter, since, as w e’ve stated it, in or der for it to be satisfi ed the analysans of (RSW) would also require that Swimmy* be a way in which John could swim, which it is not.

248

from the way water behaves in the actual world. Let us then consider the possibility that J ohn is in NSW . Given the lesson of (12 MP) about the required strength of (12), w e take it that such a possibility is incompatible with (12) (the r eader who thinks other wise is welcome to substitute in the following whatever possibility she does think is incompatible with (12)). Let us also suppose that, being awar e of the incompatibility, John is considering this possibility. Consider the conjunction: (ACKH+) John is not in a position to kno w he is not in NSW, but he knows how to swim. We think that (ACKH+) does not sound bad (we’ll discuss shortly one possible source of distraction about this). I f this is so, the initial disanalogy between (ACKT) and (A CKW) on the one hand and (A CKH) on the other remains ultimately unexplained fr om the point of vie w of a defender of (RSW). There is, admittedly, an impor tant complication in the case just discussed, pursuing which will both strengthen our intuitive judgement about (ACKH+) and lead us to another related argument. In a case of the kind just described, John will typically know that he kno ws how to swim. H e also knows that, if he is in NSW, he could not swim in Swimmy. Typically, he will also realise that if he could not swim in S wimmy, he does not know how to swim. Therefore, by Closure, if he kno ws that he kno ws how to swim, he should be in a position to know that he is not in NSW. Now, any residual doubts against the coher ence of (ACKH+), we submit, are due to this further typical and yet distracting feature of the case (that is, that ohn J knows that he knows how to swim). To illustrate the point, let us consider a different case, in which the second-order knowledge is absent. 39 Mary, the physicist, believes that she does not know how to swim. As a matter of fact, she was actually taught how to swim when she was 3-year old, but she has somehow lost her memory of this learning experience. She has not however lost her competence: if she were to be thrown in water, she would instinctively perform the right movements. Indeed, she sometimes dreams of being in water and swimming by performing those movements. In this situation, it seems true that she knows how to swim. According to (RSW), Mary knows how to swim iff, for some way w in which Mary could 39. We should note that, assuming (RSW), the following case would also therefore constitute a plausible counter example to the “KK-principle ” (roughly, if one kno ws that P, one is in a position to know that one knows that P).

249

swim, Mary knows that she could swim inw. Let’s call ‘Floaty’ the particular way that witnesses the (alleged) truth of the analysans, presumably the way involving the movements that Mary instinctively makes. Let it also be understood that, while both denoting F loaty, ‘FloatyP’ is associated with the relevant practical mode of presentation, whereas ‘FloatyT’ is associated with a theoretical physical description fully grasped by Mary. A few times, we may suppose, Mary has deliberately tried out Floaty’s movements out of sheer curiosity—while walking around, or perhaps even in the air (while she was suspended, on purpose, on the roof of her laboratory). Despite being in many other r espects a v ery good and w ell-informed physicist, let us also suppose, Mary does not know much about water: she knows of course that water is H2O, but she does not know enough about the properties of H 2O to know whether it permits swimming in F loatyT (we may suppose that Mary has spent her recent years locked in her laboratory, that most of her knowledge dates from this period and that she hasn’t conducted any experiment concerning water: for all Mary presently knows, water is a sticky treacherous liquid in which any terrestrial animal would drown). Given her general good knowledge of physics, Mary does however know the proposition expressed by the following sentence to be true: (WN) If Mary could swim in Floaty, then water is Non-Sticky, where ‘Non-Sticky’ stands for the negation of a disjunction of properties that Mary knows are incompatible with any form of natation and floating. It seems clear that, while kno wing the pr oposition expressed by (WN), Mary—who doesn’t know very much about the pattern of instantiation of properties in the world outside the lab , and in particular about which properties are co-instantiated with water—is in no position to infer the proposition expressed by its consequent, which represents a very substantial piece of kno wledge about water . That water is N on-Sticky just isn’t the kind of thing that Mary can come to know simply by deduction without conducting fur ther experiments. G iven Closur e (and M ary’s ex cellent mastery of modus ponens), that in turn seems to entail that she does not know the proposition expressed by (WN)’s antecedent to be true. But she knows how to swim, and hence (R SW) is false. Here, the intellectualist is likely to appeal again to differences in modes of presentation: indeed, it might be said, Mary does not even believe that she could swim in FloatyT (as opposed to believing that she could swim in FloatyP), and so it’s not at all surprising that she is not in a position to infer 250

that water is N on-Sticky. She believes that, if she could swim in FloatyT, water is Non-Sticky, but she believes that she could swim inFloatyP. She fails to recognise that the latter proposition is the antecedent of the former. Now, granting for the sake of argument that the notion of belief under a practical mode of presentation can be made good sense of, we think that this reply faces the follo wing problem. We obser ved above that, in general, failur e of availability of inferences involving propositions known under different modes of pr esentation is explained b y failure of r ecognising, under these different modes of presentation, some components of the pr opositions as being the same. But there is no very plausible candidate for such a failure of recognition in our example. The most plausible one would be Floaty itself. We specified however that Mary has tried out Floaty’s movements (which presumably gives her access to Floaty as FloatyP) and that she can also give an accurate description of the movements in a physical vocabulary (which presumably gives her access to Floaty as FloatyT); we can now add that she does recognise the identity between FloatyP and FloatyT. Indeed, she might accept the proposition expressed by (WN) while thinking about Floaty as FloatyP: she might be tr ying out Floaty’s movements and then realise that that way (FloatyP) is just the such-and-such physical way (FloatyT), and so is such that, if she could swim in it, water is Non-Sticky. Therefore, the reply is ad hoc: there is no plausible r eason why M ary should fail to r ecognise that the proposition that she could swim in FloatyP is the antecedent of the proposition that, if she could swim in FloatyT, water is Non-Sticky. It might be thought that w e’ve been o verlooking a fur ther feature of Mary’s belief: its being only implicit. H owever, all Closur e presumably requires is that Mary be in a position to know that water is Non-Sticky. If she does have knowledge that she could swim in Floaty, it seems that this requires that, if she w ere competently to infer that water is N on-Sticky from her (implicit) belief that she could swim in F loaty (and her belief that, if she could swim in F loaty, water is N on-Sticky), she would come to know this. But that counterfactual is clearly false. That water is NonSticky just isn’t the kind of thing that Mary can come to know simply by deduction without conducting fur ther experiments. This is another way to see that, if knowledge-how were a species of knowledge-that, it would be mysteriously barren—incapable of serving as basis for acquiring further knowledge-that via inference. Finally, in contrast to the kno wledge-that and kno wledge-wh, it is once again interesting to note that an analogous inferential isolation does exist for a wide v ariety of abilities, like e.g. those v ariously denoted b y 251

‘can’: even if John can jump over the hedge in your garden, this does not put him in a position to kno w that there is a hedge in y our garden (just like even if a dustbin can contain y our garbage, this does not put it in a position to know that you have garbage). Inferential isolation is thus y et another important aspect in which, against the intellectualism encapsulated in (R SW), knowledge-how falls in the r ealm of practice rather than in the realm of theory. 6. Gradability As has been exposed in Section 2, it is crucial for Stanley & Williamson’s argumentative strategy for sho wing that knowledge-how is a species of knowledge-that that the intermediar y step that knowledge-wh is a species of knowledge-that be secured. All of our objections so far hav e been directed at the move from this assumption to the conclusion that knowledge-how is also a species of kno wledge-that. Indeed, w e hav e been at pains to emphasise common aspects with erspect to which knowledge-that and knowledge-wh behave diff erently from knowledge-how. In closing, we wish however to discuss briefly some evidence that casts doubt on the assumption that knowledge-wh is a species of knowledge-that. Although this evidence falls shor t of pr oviding a conclusiv e argument against the reducibility of knowledge-wh to knowledge-that, we think that it does at least give some reason to think that the r eduction is not as ob vious as it is usually thought to be, and that fur ther arguments are needed in order to establish it. Stanley (2005) considers briefl y a possible objection to the thesis that knowledge-that is not gradable, based on the fact that knowledge-wh and knowledge-how seem to be clearly gradable. The same data, holding fixed the assumption that kno wledge-that is not gradable, could be used to question the assumption that kno wledge-wh is r educible to kno wledgethat. We think that S tanley (2005) pr ovides a good case for the former assumption, and we therefore take it for granted. Here are a few examples of felicitous grading that Stanley (2005, 42) considers: (13a) John knows how to swim well (13b) John knows how to ride a bicycle better than Mary does (13c) Hannah knows where Texas is better than John does.

252

Stanley’s own explanation of the felicity of (13a)–(13c) is that what is being compared (in the case of (13a), ev aluated) is the quality of the answ ers that the subjects have to the indirect questions. Let us fi rst consider grading of kno wledge-wh. Surely, we do use sentences like (13c) in the way S tanley suggests, e.g. in a situation wher e Hannah can say that Texas borders on Mexico plus the other US states of Lousiana, Arkansas, Oklahoma and New Mexico, while John can only say that Texas is some where in the US. H owever, it seems that an utterance of (13c) would also be felicitous in a situation wher e Hannah and John have the same answer to the question as to where Texas is (say, the answer that Texas is in the US), but H annah knows that answ er on v ery good grounds (say, she has been told so by a world expert of US geography and has double-checked this with another ten authorities), while J ohn only knows the answer on worse but still kno wledge-conferring grounds (say, he has r ead so on Wikipedia). Note that this cr eates a clear disanalogy between knowledge-that and knowledge-wh. In the latter situation, while an utterance of (13c) would be felicitous, an utterance of the follo wing sentence would not be: (13cKT) Hannah knows that Texas is in the US better than John does. Here is one more example that clearly confi rms the pattern: (13d)

Hannah knows who will come to the par ty better than J ohn does.

An utterance of (13d) can be felicitous in a situation wher e Hannah has a more accurate answer to the question as to who will come to the par ty (say, the answ er that at least G eorge, Bill and R onnie will come to the party, which is more accurate than John’s answer that at least George will come to the party). However, an utterance of (13d) can also be felicitous in a situation wher e H annah and J ohn hav e the same list of names of people coming to the party (say, they both had it from Mary, a trustworthy source), but, while J ohn only r ead the list, H annah actually called each person on the day of the party to double-check. The evidence indicates that S tanley’s explanation of the gradability of ‘know wh’-ascriptions is (partially) inadequate. What is being compar ed can be both the quality of the answers and the quality of the epistemic relations that the subjects have to those answers. Similar, if slightly more unusual, 253

examples can be giv en where what is being compar ed are not diff erent answers that diff erent subjects hav e to the same question, but diff erent answers that the same subject has to different questions, or even different answers that different subjects have to different questions. One of the most frequent arguments against knowledge-that and knowledge-wh expressing different relations is that they seem to fail some familiar tests for ambiguity, such as disallowing zeugmas where a single occurrence is used in the same sentence in the two senses and being expr essed by different words in some other languages.Yet, one who wants to defend the nonreducibility of knowledge-wh to knowledge-that needn’t posit any strong form of ambiguity here: clearly, the senses would have to be systematically related. It would therefore be a case of polysemy rather than homonymy;40 the tests mentioned before, however, are notoriously not very effective in detecting polysemy—they systematically yield false negatives. Take the case of ‘book’: an utterance of ‘That thought-provoking book has been written by Gilbert Ryle in 1949 and has dust on its co ver’ is typically felicitous, and there are no languages that w e know of which r eserve two diff erent words for book types and book tokens. To take a case more related to our main topic, consider the sentence ‘Mary is justified both in believing that 1 + 1 = 2 and in eating healthy food ’. The prevalent view, we take it, is that the kinds of correctness that the two conjuncts are concerned with are fundamentally diff erent: epistemic justifi cation in one case and practical justification in the other one. Maybe the prevalent view is not correct; but it is certainly not refuted by the felicity of the sentence just quoted, nor by the fact that, as far as we know, there are no languages that reserve two different words for epistemic and practical justifi cation. 7. Conclusion Our aim in this paper has not been to giv e a positive account of, or positive arguments for, a view of knowledge-how on which it does not turn out to be a species of kno wledge-that. Rather, we have developed a number of arguments against the opposite vie w, intellectualism, and in par ticular 40. Some theorists wish to count both homonymy and polysemy as (diff erent) forms of ambiguity, others prefer to reserve the word ‘ambiguity’ for the former phenomenon (for a survey, see Ravin & Leacocke (2000)). ‘ Ambiguous’ is ambiguous. The choice is ho wever clearly irrelevant for our purposes: whether or not it is a form of ambiguity , polysemy is not usually detected by the proposed tests.

254

against the specifi c version of the vie w which w e judge most inter esting (and which is most prominent in the contemporary literature)—namely, the one ingeniously developed and defended by Stanley & Williamson (2001). Stanley & Williamson’s argument for intellectualism is semantic: it draws on the alleged uniformity of ‘kno w how’- and ‘know wh’-ascriptions, both of which are best treated, they claim, along the lines of standard semantics for indirect questions (with a little help from practical modes of presentation). We believe that we have provided good reasons to think that this assimilation is pr oblematic. To sum up , the thr ee main r easons to this eff ect are the follo wing. Firstly, in contrast to ‘kno w wh’-ascriptions, ‘kno w how’-ascriptions with kno wn negativ e answ ers ar e false: for instance, while ‘No one’ can be a corr ect answer to the question as to who was at the party and ‘I n no way ’ is a corr ect answer to the question as to ho w one could square the circle, possession of the former answer provides sufficient grounds for attributing to someone kno wledge of who was at the party, but possession of the latter answ er provides no gr ounds at all for attributing to someone knowledge of how to square the circle. Secondly, knowledge-how obeys closur e principles whose counterpar ts clearly fail for knowledge-that and kno wledge-wh when the contents inv olved are singular: for instance, Calum ’s knowledge that he could kill a man in a certain way w does not give him knowledge that he could kill Kirima inw, but Calum’s knowledge of how to kill a man does give him knowledge of how to kill Kirima. Thirdly, in complete disanalogy with knowledge-that and knowledge-wh, knowledge-how is inferentially isolated from further knowledge-that: for instance, however the content of the knowledge-that which allegedly constitutes kno wledge of how to swim is specifi ed, such knowledge does not put one in a position to kno w propositions that are known to follow from that content. There is precious little, arguably, that you can infer from your knowledge-how (although you can clearly infer a lot from your knowledge that you have knowledge-how). Finally, we’ve indicated some evidence which casts doubt ev en on the reducibility of kno wledge-wh to kno wledge-that. It’s a w ell-known fact that knowledge-wh enjoys some forms of gradability, while knowledge-that plausibly does not. S tanley tries to explain away this semantic evidence arguing that the gradability of kno wledge-wh only concerns the quality of the answer that the subject has to the indirect question rather than the quality of her epistemic r elation to the answ er. However, it seems that felicitous grading persists in cases in which two subjects can only giv e exactly the same answ er to a cer tain question, which pr ovides good evi255

dence that the gradability of kno wledge-wh can also concern the quality of the epistemic relation that the subject bears to the answer. Even though we haven’t had space to develop the point at length here, we take our arguments—pointing at crucial semantic and epistemic disanalogies between knowledge-how and knowledge-that—not only to cast into serious doubt the adequacy of Stanley & Williamson’s particular analysis, but also to present a quite general challenge to intellectualist views. Unifying all forms of human cognition is a grandiose task; our suspicion however is that there is a humble and too simple answ er to the question of how to accomplish it—pretty much the same one as the answer to the question of how to square the circle.

REFERENCES Boer, S. E. & Lycan, W. (1986), Knowing Who, MIT Press, Cambridge MA. DeRose, K. 1992. ‘Contextualism and Kno wledge Attributions’, Philosophy and Phenomenological Research 52, 913–929. — 1995. ‘Solving the Sceptical Puzzle’, The Philosophical Review 104, 1–52. Donnellan, K. 1966. ‘R eference and D efinite Descriptions’, The Philosophical Review 75, 281–304. Ginet, C. 1975. Knowledge, Perception, and Memory, Reidel, Dordrecht. Goldman, A. 1976. ‘D iscrimination and Perceptual Knowledge’, The Journal of Philosophy 73, 771–791. Groenendijk, J. & Stokhof, M. 1982. ‘Semantic Analysis of WH-Complements’, Linguistics and Philosophy 5, 175–233. — 1997. ‘Questions’, in v an Benthem, J. & ter M eulen, A. (eds), Handbook of Logic and Language, MIT Press, Cambridge MA, 1055–1124. Hintikka, J. 1976. The Semantics of Q uestions and the Q uestions of S emantics, North-Holland, Amsterdam. Kaplan 1989. ‘Demonstratives’, in Almog, J., Perry, J., Wettstein, H. (eds), Themes from Kaplan, Oxford University Press, Oxford, 481–563. Karttunen, L. 1977. ‘S yntax and S emantics of Q uestions’, Linguistics and P hilosophy 1, 3–44. Karttunen, L. & Peters, S. 1976. ‘What Indirect Questions Conventionally Implicate’, in M ufwene, S., Walker, C., S teever, S. (eds), Papers from the Twelfth Regional Meeting, Chicago Linguistic Society, Chicago, 351–368. Katz, J. 1972. Semantic Theor y, Harper & Row, New York. Keenan, E. & H ull, R. 1973. ‘The Logical P resuppositions of Q uestions and 256

Answers’, in Petöfi, J. & Franck, D. (eds), Präsuppositionen in Philosophie und Linguistik, Athäneum, Frankfurt a.M., 441–466. Kripke, S. 1979. ‘A Puzzle about Belief ’, in Margalit, A. (ed), Meaning and Use, Reidel, Dordrecht, 239–283. Lewis, D. 1979. ‘ Attitudes De Dicto and De S e’, The P hilosophical Review 88, 513–543. McDowell, J. 1977. ‘O n the Sense and Reference of a P roper Name’, Mind 76, 159–185. Nozick, R. 1981. Philosophical Explanations, Harvard University Press, Cambridge MA. Perry, J. 1979. ‘The Problem of the Essential Indexical’, Nous 13, 3–21. Ravin, Y. & Leacock, C. 2000. ‘Polysemy: An Overview’, in Ravin, Y. & Leacocke, C. (eds), Polysemy. Theoretical and Computational Approaches, Oxford University Press, Oxford, 1–29. Russell, B. 1912. The Problems of Philosophy, Williams and Norgate, London. Ryle, G. 1946. ‘Knowing How and Knowing That ’, Proceedings of the Aristotielian Society 46, 1–16, reprinted in Ryle, G. (1971), Collected Papers II, Hutchinson, London, 212–225. Schaffer, J. 2007. ‘Knowing the Answer’, Philosophy and Phenomenological Research 75, 383–403. Stanley, J. 2005. Knowledge and P ractical I nterests, Oxfor d University Press, Oxford. Stanley, J. & Williamson, T. 2001. ‘Knowing-How’, The Journal of Philosophy 98, 411–444. Williamson, T. 2000. Knowledge and its L imits, Oxfor d U niversity P ress, Oxford. Wright, C. 2000. ‘Cogency and Question-Begging: Some Reflections on McKinsey’s Paradox and Putnam’s Proof ’, Philosophical Issues 10, 140–163. Zardini, E. 2007. ‘Knowledge-How, True Indexical Belief, and Action’, ms.

APPENDIX: KNOWLEDGE–HOW AND CONTEXT DEPENDENCE ON NON-TRADITIONAL FACTORS Many contemporar y epistemological discussions ar e shot thr ough with examples that seem to demonstrate the context dependence41 of ‘know that’41. To be clear , by ‘context dependence of a sentence M ’ we simply mean that diff erent utterances of M in different contexts can have different truth values.

257

ascriptions on non-traditional factors such as salience of counterpossibilities and practical stakes. In this appendix, w e aim to sho w that, while ‘kno w how’-ascriptions too are arguably sensitive to some non-traditional factors or others, they do not seem to be sensitiv e to the non-traditional factors on which ‘know that’-ascriptions do seem to be sensitiv e, at least not to the same extent. To wit, the intuitions that arise in cer tain kinds of cases for ‘know that’-ascriptions do not arise in analogous kinds of cases for ‘know how’-ascriptions. Independently of the hard question as to whether such intuitions should be tr usted in the case of ‘kno w that’-ascriptions and hence lead to a revision of the standard invariantist picture of such ascriptions, this asymmetr y in intuitions will suffi ce to constitute y et another piece of evidence against (R SW). We start by observing that ‘know how’-ascriptions seem to be dependent on context: an utterance of ‘John knows how to play chess’, if John knows the rules and nothing mor e, would seem tr ue when uttered in a context where we are only looking for someone who can offer a first introduction to the game to a six-year old child, while it would seem false when uttered in a context where we are looking for someone whom Kasparov can train with. Indeed, the example shows that ‘know how’-ascriptions are sensitive to factors (such as the purposes of the situation at hand) that hav e not been traditionally thought to be r elevant to epistemic assessment. O f course, this kind of prima facie context dependence can be explained in diff erent ways, just as the abundant literature of the last decade has shown for the prima facie context dependence of ‘know that’-ascriptions. The point we are interested in her e is whether the prima facie context dependence of ‘know how’-ascriptions is sensitive to the same factors and to the same extent as the prima facie context dependence of ‘know that’-ascriptions is. Such factors can for our purposes be usefully summarised as consisting in either the salience of counterpossibilities or in practical stakes. 42 Both kinds of factors can then be tested for context dependence b y changing their v alue for either the ascriber of ‘kno w’ or for the subject to which ‘know’ is ascribed. In the space of this appendix, w e limit ourselves to off ering a couple of what w e believe are simple and y et telling examples to test the alleged context dependence of ‘kno w how’-ascriptions on practical stakes, in particular cases where the ascriber coincides with the 42. To stress, what makes the alleged examples of context dependence so inter esting is at least in part the fact that such factors—as against many other factors, such as e.g. time—hav e not been recognised by traditional epistemological theorising as influencing the truth values of ‘know that’-ascriptions.

258

subject. We believe that the insensitivity uncovered can be shown to hold quite generally both for ascriber and subject, and that it can also be sho wn to extend to the salience of counterpossibilities (again, for both ascriber and subject), but will not attempt to generalise these r esults here—the relative insensitivity of knowledge-how (against the seeming sensitivity of knowledge-that on the same range of cases) will be suffi cient to provide the final piece of evidence against (R SW). Consider fi rst two v ariant cases of the much discussed bank example (taken with only very few alterations from Stanley (2005, 4–5)—the bank example goes back to DeRose (1992)): High Stakes. Hannah and her wife Sarah are driving home on a Friday afternoon. They plan to stop at the bank on the way home to deposit their paychecks. Since they hav e an impending bill coming due, and very little in their account, it is v ery important that they deposit their paychecks by Saturday. But as they drive past the bank, they notice that the lines inside ar e very long, as they often ar e on Friday afternoons. Hannah notes that she was at the bank two weeks before on a Saturday morning, and it was open. B ut, as Sarah points out, banks do change their hours. Hannah says, ‘I guess you are right, I don’t know that the bank will be open tomorrow’. [...] Ignorant High Stakes. Hannah and her wife Sarah are driving home on a Friday afternoon. They plan to stop at the bank on the way home to deposit their paychecks. Since they have an impending bill coming due, and very little in their account, it is v ery important that they deposit their paychecks by Saturday. But neither Hannah nor Sarah is aware of the impending bill, nor of the paucity of av ailable funds. Looking at the lines, Hannah says to Sarah, ‘I know the bank will be open tomorrow, since I was there just two weeks ago on Saturday morning. So we can deposit out paychecks tomorrow morning’. The intuition we have according to Stanley is that Hannah’s utterance of ‘I don’t know that the bank will be open tomorrow’ in High Stakes is true, while her utterance of ‘I kno w that the bank will be open tomorr ow’ in Ignorant High Stakes is false. Taking these intuitions to be solid, at least for the sake of the argument, we wish to see whether and how they change when the object of the attribution is changed fr om knowledge-that to knowledge-wh or knowledge-how. 259

The cases ar e very easy to adapt for kno wledge-wh: just imagine that what Hannah says is rather ‘I don’t know [I know] when the bank will be open tomorrow’. It seems clear to us that in this case ther e is no diff erence in intuitions: the context-dependence intuition persists that Hannah would speak truly in (suitably modifi ed) High Stakes and falsely in (suitably modified) Ignorant High Stakes. Knowledge-wh perfectly aligns with knowledge-that. However, we find that the pattern drastically changes once we move to knowledge-how. Let us consider the two following modified cases: High Stakes Kno wing-How. H annah and her wife S arah ar e driving home on a Friday afternoon. They plan to stop at the bank on the way home to deposit their pay checks. Since they hav e an impending bill coming due, and v ery little in their account, it is v ery important that they deposit their pay checks by Saturday. But as they driv e past the bank, they notice that the lines inside ar e very long, as they often ar e on Friday afternoons. Hannah and Sarah know full well that the bank is going to be open on the Saturday. However, since they moved quite recently to the area, there might be a worry about getting the way from their new place to the bank wrong on Saturday morning. Of course, on Saturday morning they would get to the bank sooner or later , but, if they were to get lost while driving there, they would probably miss the bank’s opening hours. Hence, being there, they might take the opportunity to deposit the paychecks now. As it happens, however, Hannah actually did manage very well to drive from their new place to the bank two weeks earlier. Sarah nevertheless points out that people sometimes misremember routes and end up getting the wrong way. Hannah replies, ‘I guess you are right, I don’t know how to get to the bank’. Ignorant High Stakes Knowing-How. H annah and her wife S arah are driving home on a Friday afternoon. They plan to stop at the bank on the way home to deposit their paychecks. Since they have an impending bill coming due, and v ery little in their account, it is v ery important that they deposit their paychecks by Saturday. But neither Hannah nor Sarah is awar e of the impending bill, nor of the paucity of av ailable funds. Hannah and Sarah know full well that the bank is going to be open on the Saturday. Looking at the lines, Hannah says to Sarah, ‘I’ll come to deposit the pay checks tomorrow. I kno w how to get to the bank from our new place, I’ve done that two weeks ago’. 260

Keeping in mind that, as in the original case it was supposed to be true that the bank was open on Saturday morning, in this case Hannah would drive to the bank without making err ors on Saturday morning, it seems clear to us that the intuitions that are now elicited are exactly the opposite to those elicited for kno wledge-that and knowledge-wh: Hannah would speak falsely in High Stakes Knowing-How and truly in Ignorant High Stakes Knowing-How. What intuitions track for knowledge-how, in general, seems to be simply the presence of some sort of competence. What counts as a high enough level of proficiency to possess the competence, in turn, can depend on the context (as for the competence in playing chess), but not with respect to the same factors and to the same extent as the correctness of ‘know that’-ascriptions seemingly does. The same conclusion could be r eached b y looking at the follo wing further data. Consider the r etrospective assessment w e would obtain in High Stakes if Hannah and Sarah decided to take the risk, and found, as we know, that the bank is open on that S aturday morning: it does not seem that H annah could then tr uly say, ‘ What I said y esterday turned out to be true: I knew that the bank was going to be open!’ (analogously for knowledge-wh). Even if H annah’s belief turned out to be tr ue, this does not seem to sho w that she kne w all along that the bank would be open on Saturday. By contrast, in High Stakes Knowing-How, having successfully driven to the bank on the S aturday following the Friday of the conversation, it would seem that H annah could then tr uly say, ‘What I said yesterday turned out to be true: I knew how to get to the bank!’.

261

Grazer Philosophische Studien 77 (2008), 263–305.

KNOWLEDGE-HOW AND ABILITY Franck LIHOREAU Universidade Nova de Lisboa Summary A knowledge-how attributing sentence of the form S‘ knows how to F’ may yield an ‘ability-entailing’ reading as well as an ‘ability-neutral’ reading. The present paper offers an epistemological account of the availability of both readings, based on two conceptual distinctions: fi rst, a distinction betw een a ‘practical’ and a ‘theoretical’ kind of kno wledge of how to do something; second, a distinction between an ‘intrinsic’ and an ‘extrinsic’ kind of ability to do something. The first part of the paper pr esents the double distinction that constitutes the pr oposed account; the second part presents a number of theoretical, mainly epistemological motivations for accepting the account.

What is the connection between knowledge-how and ability? As a jumping-off point, here is a ‘fact’ about knowledge-how ascriptions. A speaker can make at least two different uses of the locution ‘know how’. On the one hand, if I say that I kno w how to wiggle my ears, or that I know how to blow smoke rings, or that the I talian chef knows how to cook an excellent risotto, or that the eye surgeon knows how to operate a cataract, or that the Belgian monk knows how to brew a perfect beer, and so on, I will most often want to be understood as ascribing the ability to per form the relevant task to the relevant subject.1 On the other hand, although I am perfectly aware that being extremely allergic to it, you are perfectly unable to drink the smallest amount of alcohol, a fortiori unable to get dr unk, I might still make a correct use of ‘know how’ when I say that you know how to get drunk (your friends do it on a regular basis, say.) Likewise, although I am perfectly aware that having never tried to ride a bicycle in her life, that world-leading expert in the biomechanics of cycling is perfectly unable to 1. Throughout the paper, I shall side with Hawley (2003) in finding it more convenient to talk about the ‘objects’ of knowing-how in terms of ‘tasks’ and of ‘task performing.’ The reasons for the convenience shall shortly become clear.

ride a bicycle, I might still make a correct use of ‘know how’ when I say that she knows how to ride a bicycle (she could give professional cyclists relevant advice for improving their performances).2 In short, there is a possible use of ‘know how’ in which ‘kno wing how’ seems to entail ‘being able ’—an ‘ability-entailing’ use—and another in which ‘knowing how’ does not seem to entail ‘being able ’—an ‘ability-neutral’ use. This is an inter esting ‘fact’ about knowledge-how ascriptions, a linguistic fact then. One might thus naturally expect the explanation for that fact to be of a linguistic nature itself. For instance, a suggestion might be that the availability of both uses of ‘know how’ is traceable to the existence of two distinct senses of the locution, an ability-entailing sense, and a distinct ability-neutral sense. Or on the contrary, it might be suggested that there is one and only one sense of ‘know how’ and that the ability-entailing use simply is a special use of that unambiguous locution. N either of these suggestions are the kind of explanation that I shall pursue her e. Either of them might in principle be sustained without, at any point, appealing to considerations pertaining to the nature of the relevant knowledge and ability. For instance, suppose y ou gather conclusiv e enough linguistic evidence that ‘kno w how’ is indeed ambiguous in E nglish between the ability-entailing and the ability-neutral reading.3 The question that comes 2. For the latter two examples, it would no doubt be mor e accurate to say , respectively, that you know how one gets dr unk, and that the exper t knows how a bike is (to be) ridden. Noë (2005) insists that a key distinction is that betw een ‘knowing how to do something ’ and ‘knowing how something is done’: “To know how to do something is to have the relevant ability; but one can know how something is done, or how one does it, without knowing how to do it, that is, without having the ability.” (284) The distinction pointed out is all the more important because it tends to be blurr ed in ev eryday practice of kno wledge-how attribution, wher e the ‘how to do something’ will often be used as a ‘ho w something is done’. The account I want to propose revisits that distinction in a sense. 3. Cross-linguistic evidence drawn fr om French and Portuguese for instance points in that direction. Restricting my attention to French, there is a difference in that language between ‘savoir faire’ and ‘savoir comment faire’, two constructions that English would translate as ‘kno w how to do’. Using ‘savoir faire’ systematically entails the possession of an ability b y the subject; but not so with ‘savoir comment faire’. (Portuguese encodes a similar difference and does it by similar means: ‘saber fazer’ vs. ‘saber como fazer’.) The difference is justly pointed out by Rumfitt (2003). Purported evidence that the difference is likely to be a real semantic one comes from the fact that you can correct someone who makes use of ‘ savoir faire’ instead of ‘savoir comment faire’, as in the following dialogue: a. Ne m’aviez-vous pas dit que vous saviez faire des crêpes? (Didn’t you tell me that you knew how to cook pancakes?) b. Non, j’ai juste dit que je savais comment faire des crêpes, pas que je savais faire des crêpes! (No, I only said that I knew how to cook pancakes, not that I could cook pancakes! )

264

next is: what do y ou do with that? A t best, y ou would hav e established the existence of an ability-entailing sense of ‘kno w how’; but y ou could not, on that sole basis, conclude to the existence of an ability-like kind of knowledge: that knowledge-how is a kind of ability does not immediately follow from the claim that ‘knowing how’ sometimes entails ‘being able.’4 There is no ob vious step from the semantic claim to the epistemological conclusion. Instead of endorsing a linguistic appr oach, then, I would like to off er an epistemological account for the linguistic fact in question, an account that appeals to the kind, or kinds, of knowledge that we talk about when we talk about knowledge-how, and with the kind, or kinds, of ability that are relevant to such talk. The main lines of the account I hav e in mind can anticipatively be formulated as follo ws. There are two distinct kinds of knowledge which ‘know how’ can be used to talk about; there are also two distinct kinds of ability that a subject can hav e when it comes to performing a certain task; and the following: (1) A subject S knows how to perform a task F only if S is able to F holds for and only for one of the two kinds of knowledge and one of the two kinds of ability. In other wor ds, it is because ther e is an ability-like kind of knowledge that an ability-entailing reading of ‘know how’ is available. My primary purpose, then, will be to clarify the appropriate notions of ability and knowledge that will allow for an explanation of that sort. To this end, the first part of the present paper introduces the two aforementioned distinctions that form the basis of the view I propose regarding knowledge-how and ability . The distinctions will then be put to use in the second part, dedicated to highlighting some of the theoretical, mainly In easily conceivable contexts the second speaker would be right to correct the first in that way, and she would be right because ‘savoir comment faire des crêpes’ does not entail the possession of the ability that ‘savoir faire des crêpes’ does systematically entail. In the same vein: c. Je sais comment fair e des crêpes mais je ne suis jamais arriv é à en fair e. (I know how to cook pancakes but I’ve never managed to cook any.) is interpretable, whereas: d. Je sais faire des crêpes mais je ne suis jamais arrivé à en faire. (I can cook pancakes but I’ve never managed to cook any.) sounds like a contradictio in terminis. For a more principled case for the existence of an ambiguity in English, see Rumfitt (ibid.) 4. For example, S tanley and Williamson (2001) account for the av ailability of the ability-entailing use of constructions in terms of ‘know how’ without thereby making the relevant knowledge into an ability of any kind, on the contrar y.

265

epistemological, benefits which I think can be gained from accepting the proposed view. Par t One 1. Subjects and devices As a star ting point, an impor tant question to addr ess is what, in something like (1), can count as ‘a subject’ and what cannot. Here, we have to be careful neither to be too narr ow-minded nor too open-minded as to where we draw the line. We should av oid being too open-minded on the matter , because as a matter of fact, many things can be described as per forming a task, y et hardly deser ve the title of ‘kno wing-how subjects’. An electric coff eemachine can make coffee, but does it know how to make coffee? Note that it hardly deserves to be considered a ‘knowing subject’ at all. So, a natural thought would be that if y ou cannot call them ‘kno wing subjects’ at all, then you cannot call them ‘knowing-how subjects’ either. A no less natural thought would be that if y ou can call them ‘knowing subjects,’ then you can call them ‘knowing-how subjects’ too. Putting these thoughts together, we end up with a candidate criterion for being a knowing-how subject: a knowing-how subject is something to which you may ascribe knowledgehow and you may ascribe knowledge-how to something exactly if you may ascribe knowledge to it. That is probably what would be suggested by those who argue from the assumption that all knowledge is propositional, that is, is kno wledge-that, to the conclusion that kno wledge-how is nothing else but a species of knowledge-that. For instance, Stanley and Williamson (2001), who defend the r educibility of kno wledge-how to kno wledgethat, argue that on the one hand, since “we are often reluctant to ascribe propositional knowledge to babies” (440), it should be no mystery that it would often sound incongruous to ascribe knowledge of how to suck (or how to cr y for hours) to babies, and that, on the other hand, since “ we are not always r eluctant to ascribe pr opositional knowledge to babies ” (441), it should be no mystery either that it would sometimes not sound incongruous to ascribe knowledge of how to suck (or how to cry for hours) to babies.5 5. “For example—they say—suppose that there is some question about whether the infant

266

Things, however, are not that easy, for there are prima facie hard cases too. For instance, w e might fi nd ourselv es saying of phylogenetically ‘lower’ animals, or ev en inanimate things, that they kno w ho w to do certain things. For instance, you might say that the young bear from the local circus knows how to ride a monocy cle. Or perhaps an appr opriate context might be found in which it wouldn ’t sound outright incongr uous to say how happy you are that your brand new coffee machine knows how to make such a good coff ee (not like the old one). D oes this mean that bears and coff ee machines should count as pr oper ‘knowing-how subjects’? We might be in the pr esence of some v ery loose use of ‘kno w how’ there, especially in the second case. To settle the matter whenev er we are in doubt, we can simply ask the following question about the candidate ‘subject’: assuming that it cannot perform the relevant task, could a reason for this be that it does not know how to per form the task? F or example, if we are told that this or that person cannot ride a monocy cle, a reason for this could be that he doesn’t know how to ride a monocy cle. Same thing if we are told that unlike its older brother, the young bear from the local circus cannot ride a monocycle: the reason for this could be that he doesn’t know yet how to do so. We might even go further in that direction: if your scientific pocket calculator, unlike mine of the same model, cannot compute the gr eatest common denominator of two numbers, a reason for this could be that y ours, not having the appr opriate program that mine has, does not know how to perform the computation. Couldn’t we say so? That might not be v ery clear. On the contrar y, it is clear that if we are told that the coff ee machine next door cannot make coff ee, we would never think that the r eason for this is that it doesn’t know how to make coffee. Persons and bears can be considered as proper knowing-how subjects with respect to the task of riding a monocy cle; on the contrar y, coffee machines cannot be consider ed as pr oper knowing-how subjects with respect to the task of making coff ee. I am not suggesting that the ‘could you say she doesn ’t know how?’ test off ers the ultimate criterion for telling proper knowing-how subjects from improper ones. But at least, it helps. It is impor tant to insist that being a pr oper knowing-how subject be defined with respect to a par ticular task. F or something might w ell be a proper knowing-how subject when it comes to per forming a given task, Paul is handicapped. The doctor may note with pleasure that [Paul knows how to suck].” (Stanley & Williamson 2001, 441)

267

yet not be a kno wing-how subject when it comes to per forming another task. Consider the example of the bab y again. O n the one hand, if the baby is unable to suck or to cry at all, it can hardly be because she doesn’t know how to suck or cry; but on the other hand, if the baby is unable to put on her pyjamas all b y herself, it is because she doesn’t yet know how to do so (for this, she ’ll have to wait a couple of months, or y ears). This is why we must be careful not to be too narr ow-minded either in telling what counts as a kno wing-how subject fr om what does not. A subject never is a kno wing-how subject ‘in the absolute ’, but always r elative to a task. The term ‘ subject’ tends to conv ey the idea that what w e are talking about when we make use of it is of a human-like natur e or, at least, of a sufficiently-evolved-animal-like nature. But as we just saw, we might well speak of kno wledge-how—depending on the task w e have in mind—in the case of inanimate things, like scientifi c pocket calculators. S o, perhaps we had better stop talking about ‘ subjects’ for a while, and speak instead in terms of ‘ devices’ and of ‘kno wing-how devices’. A device can be characterised as anything that can be described as per forming a task, any task. A coff ee machine, in that sense, is a device; a tennis play er, in that sense, is a device too . Now, of all the devices that may be pr operly described as per forming a task F—the ‘ F-ing devices’ as w e may call them—, only some can pr operly be said to know ho w to F , while the others cannot. Here, because of what w e said earlier about the task-r elativity of the notion of a knowing-how subject, two cases must be considered. First, a device can be considered a knowing-how device with respect to a task that it per forms if w e could say that she kno ws or does not know how to perform it. Second, a device can be considered a knowinghow device simpliciter if there is some task, any task, that it can per form and of which w e could say that she kno ws or does not kno w ho w to perform it. This being said, let me now clarify an expression which I have been using heavily since the beginning of the paper, that of ‘performing’ a task. 2. Task and procedure A device, I said, is anything that w e could describe as performing a task, any task; and a F-ing device is one that w e could describe as per forming the task of F-ing. This prompts the question: what is it for a task F to be ‘performed’? 268

Take the following task: killing Zoé. Many have tried, always in v ain. On a windy day, however, the wind blo ws through the apple tr ee under which Zoé is sitting, causing an apple to fall and hit Zoé’s head at a high enough speed so that it kills Zoé right away. Let us stick to simple matters. What killed Zoé? The (fall of the) apple? The (blast of the) wind? The blastof-the-wind-that-caused-the-apple-to-fall, etc.? Whatever we may think to have been the proper ‘agent’ of the event consisting of the killing of Zoé, what we can say at most, if w e are to speak in terms of tasks, is that the task in question—killing Zoé—was ‘fulfilled’, not that it was ‘performed’ by anything, not even by the ‘agent.’6 So, what is it that makes for the diff erence between a task F being just ‘fulfilled’, as in the apple case, andF’s being, properly speaking, ‘performed’, as when a coffee machine makes some coffee for instance? Performing a task F is what a F-ing device does. So another way to put the question is: what is it that makes for the diff erence between a device that F-s in virtue of its being a F-ing device and one that F-s without being a F-ing device? The answer I would like to submit builds on the notion of aprocedure’. ‘ Intuitively, it is the notion of ho w a device per forms a task: to per form a task F is for a device to execute some procedure whose execution constitutes, precisely, the F-ing. But more precisely, what is a procedure? It is relatively easy to describe in what this or that particular procedure consists: you could easily describe ho w you usually go fr om your place to the mall, or ho w you usually make yourself a cup of tea, and you could describe how to do such things in a pr ecise and ev en concise manner . It is mor e diffi cult to define what a procedure to perform a task consists in, without making the definition circular at some point. A t least we may try this: a procedure for a device is a (preferably finite) sequence of possible actions that the device may be described as doing. Then, for a device to execute a procedure is for it to do all the actions making up the corresponding sequence in the order in which they appear in the sequence. And it is by executing the procedure that the device performs a task. If F-ing is the task, we may call a procedure whose execution by a device constitutes a F-ing a ‘F-ing procedure’. 6. It should be clear that nothing in the example seems to bear the mark ‘ of agency’, at least if agency is vie wed in a standar d Davidsonian perspective, for whatever killed Zoé did not do it intentionally. However, I do not intend the notion of intentionality or intentional action to play any role in distinguishing what I call ‘performing’ from others ways of ‘doing’. This should be obvious from the fact that I use ‘ perform’ to describe what a coff ee machine does. I do not intend intentionality to play any role in distinguishing knowledge-how from mere ability either, as shall become clear in the following sections, especially Section 5.

269

Note that I wr ote ‘do’ instead of ‘ performing’ when I mentioned the actions that compose the procedural sequence. This is because the sub-tasks or actions that make up a procedure for performing a given task need not themselves be tasks that a device performs, but will sometimes be tasks that only need to be fulfi lled. Suppose that the task to be per formed is that of skydiving. Then, in the sequence of actions constituting the procedure, you might want to include falling or letting yourself fall, just before opening your parachute. But falling is likely to be considered more as a fulfilment than as a performance of a task. Note also that I do not want to be understood as saying that for a giv en task there must be one and only one pr ocedure— sequence of possible actions—that a device may execute to perform the task. To make yourself a cup of tea, you can fi rst put the tea bag into your cup and then get some water boiling, but oy u can alternatively first get the water boiling and only then put the tea bag into your cup. For some devices, like human ones, there may be more than one procedure that it may execute to perform a given task (depending of course on the task). Let me pause a little longer to consider possible worries with the above characterisation of a procedure. A first worry is that, as it stands, the characterisation of what it is to ex ecute a procedure might seem incomplete, since it does not clearly ex clude the possibility that the apple fr om the above example might have executed a Zoé-killing procedure. One might indeed think of Zoé’s being killed by the apple in terms of the apple’s being involved in a sequence of actions (say, from the action of falling from the tree to the action of hitting Z oé’s head), whereby the apple would count as having per formed the killing of Z oé, that is, as a Zoé-killing device; and this does not sound right. Let me r espond to this fi rst worr y after considering a second one. The second worry is that my notion of a pr ocedure might seem quite abstract in that it leaves open the possibility that a procedure executable by a device might exist ‘outside’ the device. Just in case that would be a ‘real’ worry, let me add that for a device to execute a procedure, the procedure must be ‘integrated’ in or by the device, as we shall say (alternatively: the device must ‘integrate’ the procedure, in the sense of having the procedure ‘within it’). This is likely to sound mysterious. tI shouldn’t, however, sound more mysterious than saying that my scientific pocket calculator integrates the small program that I wrote for it and that computes the greatest common denominator of two numbers.Your scientific pocket calculator of the same model does not integrate such a program, although you could write or download one for doing the same thing. The program is what, here, may be 270

most directly identified with the procedure.7 It is because mine integrates the program/procedure in question that it performs the task of computing the greatest common denominator of two numbers. Yours can’t, because it does not integrate that program/procedure. There is no more to my use of the word ‘integrate’ than that. We can now address the fi rst worry by simply saying that the apple is not a Zoé-killing device because it does not integrate any Zoé-killing procedure (any killing procedure indeed). We have to be careful not to associate the fact that a device integrates a procedure too tightly with the idea that it has it stor ed in its memor y, its ‘procedural’ memory as we may say. Some devices do need to include something like a memory to ‘store’ the procedures for some tasks. But this need not be so for all devices and tasks. There are other ways in which a procedure can happen to be integrated in a device. O ne of them is to build the device ‘around’ the pr ocedure, so to speak. This is the case with the coffee machine, which is built around the procedure, for the sole purpose of making coffee if you prefer.8 With the notion of a procedure at our disposal, we can now proceed to define the two notions of ability that I pr opose to distinguish. 7 Program running is just a special case of pr ocedure execution. In that case, the possible actions making up the pr ocedure consist of operations possibly pr ompted by instr uctions in some programming language. 8. The present notion of procedure thus extends beyond the most ordinary sense of the term ‘procedure’ so as to overlap the notion of a process (in the sense of p‘ rocess’ in which you can say that technicians have invented a revolutionary process for keeping the fourth wheel of supermarket trolleys from going erratic). It extends beyond the usual way of characterising a pr ocedure too, in that the notion is usually characterised, in par t, in terms of the success that it makes quasi-ineluctably possible to achieve when applied. Reference to success is not included in the present definition of a procedure for the sake of generality and non-redundancy. First, for some tasks at least, one can be said to perform or to have performed a task at which one is or has been obviously unsuccessful—we can say that a steadily incompetent eye surgeon operates cataracts, even if her patients usually end up losing an ey e during the operation. Second, for other tasks, it is relatively unclear what the criterion for success is or could be. Consider or dinary driving for instance, as opposed to car racing. What is the criterion of success for a performance of this task? A suggestion might be that one’s being successful in performing a task simply equates one’s executing the procedure for the task. For example, success in ordinary driving could be said to consist in driving in accordance with the procedures stipulated by a driving instructor or in the driving code (keep left/right, stop when the r ed light is on , etc.) But I am afraid that that would make reference to success superfl uous in the v ery defi nition of a pr ocedure. Alternatively, one might suggest that context enters the picture at some point, for instance by setting the relevant standards of success when it comes to evaluating a performance of a task. In any case, however, reference to success will not be needed for the purpose of this paper . Sequences of possible actions will suffi ce. (Interesting analyses of kno wledge-how that appeal to some notion of success can be found in Carr 1981, Hawley 2003, Williams forthcoming, and Wallis 2008.)

271

3. Intrinsic and extrinsic ability Sometimes, one will know how to do something that one cannot do. But this will not always be because one lacks the ability to do the thing in question. Not all uses of ‘ can’s and ‘ cannot’s bear on matters r elated to human agency. And of those that do, only a fe w are made to ascribe or deny abilities. She cannot play the piano because there is no piano around. This is the ‘ can’ of oppor tunity, and if she cannot play the piano in this sense, it is not because she lacks the ability: she lacks the oppor tunity. He cannot drink alcohol because he is a straight Muslim. This is the ‘can’ of moral permission, and if he cannot drink alcohol in this sense, it is not because he lacks the ability: he lacks the (self-)permission. I t cannot operate a cataract because it is a dog. This is the ‘can’ of mere and simple capability, if we may call it so; and if the dog cannot operate a cataract, it is not because it lacks the ability: it is just utterly incapable of operating anything.9 Now, compare with the following, which clearly involves the notion of ability. She cannot ride a monocycle because she has never learnt how to ride one. This is the ‘can’ of ability: if she cannot ride a bicy cle, it is because she lacks the ability. Note that you could say, instead, that she doesn’t know how to ride a bicycle. That is the ‘can’ that anyone interested in the nature of knowledge-how should primarily care about. I nonetheless believe that another ‘can’ should be paid par ticular epistemological attention to . She cannot ride a bicycle because she has lost her legs in a car accident. This is not stricto sensu the ‘can’ of ability: if she cannot ride, it is not because she lacks the ability to ride, but because she lacks the ability , or better , the capacity, to exercise that ability. In both cases, w e could say that the subject is not able or that she is unable to ride a bicycle. Yet, the two kinds of ‘abilities’ are different and, I submit, necessar y for the performance of a task b y a subject. Translated into the language of pr ocedures, the idea is that in or der to per form a task, fi rst a device has to integrate a pr ocedure to that effect, and second it has to have ‘what it takes’ to execute the procedure. These two conditions are what the two notions of ability that 9. There are many other uses or senses of ‘can’ and ‘cannot’ that have no particular bearing on human agency and, in particular, nothing to do with ability or inability. Interesting analyses of various concepts of ‘can’ of interest for ability judgements can be found in Bronaugh (1968), Carr (1979), K enny (1957), R yle (1949) of course, and Taylor (1960). A possible basis for a tentative unified picture of the diff erent senses of ‘can’, including those alluded to in the body of the text, can be found in Elgesem (1997)’s modal logic of agency.

272

I would like to distinguish are meant to capture. The first notion of ability is that of ‘intrinsic ability ’, which I propose to define as follows: Intrinsic Abilit y: A device D is intrinsically able to perform a task F iff D integrates F-ing procedures, that is: iff D is a F-ing device. To illustrate, consider the scientific pocket calculator again. It is intrinsically able to compute the gr eatest common denominator of two integers because it integrates a procedure to that effect: the program I wrote for it and which is stor ed somewhere in its memor y. If I make it r un the program, it will compute the gr eatest common denominator of two giv en input integers. I f I r emove or erase its memor y, or if I simply erase that particular program from its memory, it will no longer be able—intrinsically able—to per form the task. Your calculator is of the same model as mine, but mine already has the program that I wrote for it to compute the greatest common denominator, while you still have to write or download such a program for yours. Then, mine is intrinsically able to compute the greatest common denominator because it integrates a pr ocedure to that effect; yours is intrinsically unable to do so because it does not integrate any similar procedure. I intend the notions of extrinsic and intrinsic abilities to apply to animate devices too, of course. So let me now consider a task of a totally different order and performable by a person this time: br ushing one’s teeth. You are intrinsically able to brush your teeth in virtue of your integrating a procedure to that eff ect (minimally: up and down from the red gums to the white teeth for a while, same thing with the lo wer jaw, then insistent circles for the teeth at the back, etc.). The procedure is there somewhere in your memory, and you (should) execute it several times a day. That wild man is offered a toothbrush and some toothpaste, but he has never heard of a toothbrush before. Perhaps he will quickly learn how to take care of his few dental remains. But for the moment, he is intrinsically unable to brush his teeth because, not yet having been inculcated with it, he does not integrate a procedure to that effect. In brief, intrinsic ability to F is the notion of what it takes for a device to be a F-ing device. By contrast, ‘extrinsic ability’ to F will be the notion of what it takes for a F-ing device to F. And to capture this notion, I propose the following definition:

273

Extrinsic Abilit y: A device D is extrinsically able to perform a task F iff, for each sub-taskG whose performance is required by some F-ing procedure for D, a G-ing device Dc is part of D. The LCD screen is part of the scientifi c pocket calculator. In our specific example, it is that par t of the calculator which, r elative to the task of delivering the greatest common denominator of two given numbers, is in charge of displaying the r esult. Mine has the intrinsic ability to per form the task because it has the program to that effect. Suppose, however, that its LCD screen has ceased to function after I inadv ertently dropped it. It can no longer deliv er the r esult in question (any r esult, for the same reason). It has lost its extrinsic ability to do so b y losing its screen. But it has not thereby lost its intrinsic ability to do so . If by chance it could be given a new LCD screen, it could deliver the result again. Then it would have regained its extrinsic ability. To illustrate extrinsic ability in humans, I will use the same silly albeit simple example as abo ve. When you brush your teeth, y ou have to keep visual track of what is going on and of what y ou are doing. So your optic nerve is inv olved in ex ecuting the pr ocedure for br ushing y our teeth. What the optic nerve does is, to put things roughly, convey the basic visual information (nerve impulses) fr om the ey es to the primar y visual cor tex. Conveying that information is not something thatyou perform, something you have the intrinsic ability to do; you are not properly speaking a nerveimpulse-transmitting device. It is your cortical nerve that is the device. I t has the intrinsic ability to transmit the nerve impulses to the primary visual cortex because it integrates the r elevant procedures (and is probably built ‘around’ these procedures). And because, among other things, you have that device—the cortical nerve—as a par t of y ou, you are extrinsically able to perform the more general task of brushing your teeth. If your cortical nerve got severely damaged, you could lose that ability, as that device would itself no longer be able, intrinsically able this time, to per form its task. It is impor tant to understand the connection betw een the two kinds of ability. To sum up, in order for a device to perform a task, it must have both the intrinsic and the extrinsic ability to perform the task. If F is the task, for a device D to perform it, on the one hand, it has to integrate a procedure for F-ing, that is, it has to be a F-ing device; and on the other hand, for every sub-task G that the procedure in question includes as an action to be per formed, the F-ing device D has to hav e as a par t a ‘subdevice’ Dc that is itself intrinsically able to per form G, that is, that is a 274

G-ing device. I n other wor ds, the intrinsic ability of each such ‘ smaller’ G-ing device makes for the extrinsic ability of the ‘bigger ’ F-ing device. If, for all the sub-tasks G whose performance is required by the procedure for F, a part of D is intrinsically able to F, then the device is extrinsically able to F; otherwise not.10 An important further point is that a device can be described as being ‘more able’, or as being ‘less able ’ to per form a giv en task than another device of the same type. The availability of comparative ability attributions of that sor t can be dev eloped along two lines: either in terms of gr eater intrinsically ability, or in terms of greater extrinsic ability. Intrinsic ability sets the tone. Given a task F, a F-ing device can be more intrinsically able to F than another F-ing device, if the former integrates a ‘better’ procedure than the latter. This idea of a ‘better’ procedure can be spelled out in two different ways. First, and most obviously, the procedure can be better in vir tue of the fact that it r equires exploitation of a lesser amount of resources of a given type. In the case of the pocket calculator and with respect to the task of delivering the greatest common denominator of two numbers, the pr ocedure or pr ogram it ex ecutes to that eff ect can be mor e or less demanding in ex ecution time or in memor y space, for example. S econd, and perhaps less ob viously, the pr ocedure can be better in virtue of the fact that the pr ocedure itself is shorter in terms of the number of sub-actions it requires to be done. Roughly, if the program you have downloaded for your calculator to deliver the greatest common denominator of two integers was written so as to involve fewer operations than the one I wrote for mine to the same eff ect, both calculators will be intrinsically able to per form the task, y et yours will be more intrinsically able than mine to do so. ‘Greater extrinsic ability’ of a device can then be defi ned, relative to a given task, partly in terms of the greater intrinsic ability of its sub-devices involved in the task. A device can be more extrinsically able to F if some of its sub-devices involved in F-ing are more intrinsically able to perform the relevant sub-task; or, when the amount of allocated r esources can make a difference, if the sub-device in question has at its disposal a greater amount of those resources the exploitation of which is involved in performing the 10. I am relying here on a fairly commonsense notion of a p‘ art of a device’, and am assuming no par ticularly sophisticated vie w about par t-hole relations. For the purpose of the paper , the latter is unnecessary. If we can identify a hand as being part of a person qua grasping device, or the pad and scr een as being par ts of the calculator , or the plug as being par t of the coff ee machine, etc., that shall be largely suffi cient for our purpose.

275

sub-task. Your calculator and mine can have the same small program (written in the same language) for drawing fractal fi gures, yet yours, with its faster processor, its bigger memory, its higher-resolution LCD screen, etc., can be better than mine at drawing fractals; that will be because, as w e shall say, it is more extrinsically able than mine to do so. All this holds similarly for tasks that are performable by human beings, even if the kinds of r esources involved are presumably of a v ery different nature from those exploited by inanimate devices. If you are going to brush your teeth in the dark—this is the task—, you had better have good night vision. This will crucially depend on how fast your eyes’ supply of rhodopsin gets renewed. Producing the pr otein is not something that you properly speaking perform. You have no procedure for this. It is your body, or some more specific part(s) of your body, that properly speaking performs the task of refilling your eyes’ supply of rhodopsin. I t integrates the pr ocedure, or process, for doing this. It is it that has the intrinsic ability to do that, and it thereby contributes to your extrinsic ability to see in the dark. Now, if my body, or the relevant part of my body in charge of eplenishr ing the supply of rhodopsin has a har der time than y ours doing so—for instance if it takes it longer to renew the necessary amount of the protein in your eyes, or if it takes it more of the same resources than it takes yours to do so, or if it has that kind of r esources available in a greater quantity than yours (say, if I suff er from a defi ciency in vitamin A while y ou are an inv eterate carr ot fan)—, it is likely that y ou will hav e better night vision than me, and ther efore that y ou will be mor e able, more extrinsically able, than me to br ush your teeth in the dark, as well as to perform many other tasks in the dark. And that is because the task of replenishing your eyes’ supply of rhodopsin is executed by a sub-device of yours—the relevant part(s) of your body—which happens to be more intrinsically able to perform that task than the similar sub-device of mine, or which has at its disposal a greater amount of those resources on the basis of which the supply of rhodopsin can ultimately get refilled. Hoping that the for egoing will hav e given a sense of the distinction between the two notions of ability, I now put the notion of intrinsic ability to use in or der to make a fur ther distinction, a distinction betw een two kinds of knowledge this time: ‘practical’ knowledge, and ‘theoretical’ knowledge of how to perform a task. 11 11. The terminological distinction betw een ‘practical’ and ‘ theoretical’ knowledge is borrowed from Franklin (1981).

276

4. Practical and theoretical knowledge of how to F There seems to be a close link betw een possession of an intrinsic ability by a device and ascription of kno wledge-how to that device. M y scientific pocket calculator knows how to draw a low-quality black-and-white portrait of M arilyn, while y ours does not; that is because mine has the appropriate program, while yours doesn’t. Maybe it is not clear, intuitively, that we are here properly ascribing and denying knowledge to the pocket calculators. So let us take other examples. This old bear knows how to take honey out of the pot while that y oung bear does not; that is because the first has the appropriate technique while the second doesn’t. That Briton sailor knows how to open up o ysters with a cent as the only tool while I do not; that is also because he has an appropriate technique which I lack. In all these cases, what the ones that kno w possess which the ones that don’t know lack is an integrated procedure. I therefore suggest that intrinsic ability be made into a necessar y condition for the kind of knowledge referred to in the ability-entailing sense of ‘know how’, ‘practical knowledge’ as I will call it:S has practical knowledge of how to F only if she has the intrinsic ability to F, that is: only if she integrates some F-ing procedure, or equivalently: only if she is a F-ing device. A more radical suggestion would be to make intrinsic ability into a necessary and sufficient condition, that is: (2)

A subject S has practical kno wledge of how to per form a task F if and only if she has the intrinsic ability to F.

This fur ther suggestion cannot be right. A coff ee-machine is a coff eemaking device, it has the intrinsic ability to per form the task of making coffee, but you don’t want to say that it has practical knowledge of how to make coffee. (Told that the machine cannot make coffee, you would never think of its not knowing how to make coff ee as the r eason for that.) S o, intrinsic ability cannot be all there is to practical knowledge. Then, what else is missing? One of the smallest diff erences between a coffee-machine on the one hand and a scientifi c pocket calculator, a bear and a human being on the other hand is, one may pr opose, that the former device can perform only one task whereas the latter ones can perform more than one task. But that cannot be a difference that makes a difference. For there are coffee machines that tell y ou the time, ther e are radio alarm clocks, and telephones that play music. So, again, what is missing? 277

To approach this issue, I pr opose to make a quick digr ession. Let us turn to another question: the Gettierability of knowledge-how. An issue that Stanley and Williamson address in ‘Knowing How’ (2001) has to do with the respective properties of knowledge-how and knowledgethat. In that justly famous paper they defend the r educibility of kno wledge-how to knowledge-that. But if knowledge-how is simply a species of knowledge-that, as they claim it is, then any pr operty of the latter must be a pr operty of the former . Now, knowledge-that can be G ettiered, one may presume; but what about knowledge-how? As they themselves put the question: “we can imagine cases of justified true belief that fail to be knowledge-that, because they fail to satisfy some extra condition. I t may appear difficult, however, to formulate examples that fall shor t of being kno wledge-how for a similar r eason.” (435) Their reply is that w e can think up appropriate Gettier cases for knowledge-how. The one they off er is this: Bob wants to learn how to fl y a fl ight simulator. He is instructed by Henry. Unknown to Bob , Henry is a malicious imposter who has inser ted a randomizing device in the simulator ’s controls and intends to giv e all kinds of incorrect advice. Fortunately, by sheer chance the randomizing device causes exactly the same results in the simulator as would hav e occurred without it, and by incompetence Henry gives exactly the same advice as a proper instructor would hav e done. Bob passes the course with fl ying colors. He has still not fl own a real plane. Bob has a justifi ed true belief about ho w to fly. But there is a good sense in which he does not know how to fly. ( ibid.)

I clearly do not shar e Stanley and Williamson’s intuition that “ there is a good sense in which he does not kno w how to fl y”. I hav e a diff erent intuition. They describe the situation by saying that because it is by sheer chance that his learning how to fl y did not go wrong, Bob does not now know how to fl y. For my part, I would be inclined to describe the situation by saying that given the more than hazardous way in which he learnt how to fl y a flight simulator, there is a great chance that Bob now knows how to fl y. To deny that he kno ws how to fl y would be to deny that he has learnt how to fly, which I find absurd since, as a matter of fact, he can fly and do it successfully. I find it all the more absurd since practicing on a flight simulator while listening to a fl ight instr uctor who giv es y ou just the right advice, the simulator’s controls working with exactly the good results—this is exactly what is supposed to happen in this case, even though it happens by mere luck —, is certainly among the best ways to learn how to fly! And after all, 278

the learning went well. I am not presupposing here any complicated notion of when learning does and does not go w ell. Learning how to perform a task goes well if the learner ends up being able to per form the task as an effect of the learning. And once you’ve learnt, you know. So, I confess to not sharing Stanley and Williamson’s intuition on that case. 12 What is key in the for egoing purported defusing of S tanley and Williamson’s argument is the idea of learning, where learning to per form a task is learning fr om one’s own experience of per forming the task. I t is because Bob has as a matter of fact learnt how to fly, and thus gained the intrinsic ability to do so, that he cannot now fail to know how to fly. So, could one’s gaining the intrinsic ability to perform a task by one’s learning how to perform the task be the missing link in the earlier proposal about practical knowledge? It cannot be the last word. A first reason why it cannot is that I still want to say that my scientifi c pocket calculator kno ws how to draw the portrait of Marilyn while yours does not; and that would not be possible if learning from one’s experience of performing a task were the last wor d. But let me mention a mor e serious r eason why that can ’t be the last word. It is true that in human- and animal-like knowing-how devices, practical knowledge of how to perform a task comes with personal experience of per forming (or at least tr ying to per form) the task. This is why it is actually true of human- and animal-like knowing-how devices that having practical knowledge of how to F implies having F-ed. But that needn’t be so. To see this, let me draw a parallel with what Lewis (1988) says at some point about knowing what an experience is like: Having an experience is sur ely one good way, and sur ely the only practical way, of coming to know what that experience is like. Can we say, flatly, that it is the only possible way? Probably not. There is a change that takes place in you when y ou have the experience and ther eby come to kno w what it ’s like. Perhaps the exact same change could in principle be produced in you by precise neurosurgery, very far beyond the limits of present day technique. Or it could possibly be produced in you by magic. […] For instance, the casting of a spell could do to y ou exactly what y our fi rst smell of skunk would do . We might quibble about whether a state pr oduced in this ar tificial fashion would deserve the name ‘knowing what it’s like to smell a skunk,’ but we can 12. Ted Poston (forthcoming) does not share their intuition either, and, I think, for pretty much the same reasons. He insists that “there is a good sense in which Bob does know how to fly”.

279

imagine that so far as what goes on within y ou is concerned, it would diff er not at all. (Lewis 1999, 264)

Without pronouncing on whether what Lewis says about knowing-whatit’s-like-to-have-an-experience is true or not, we may still notice that the remark he makes about it nicely transposes to (practical) kno wing-howto-perform-a-task. There are changes that take place in you when you learn how to perform a task, say kung-fu fi ghting, changes that may take long y ears of experience and practice to take place. Perhaps the exact same changes could be produced in you by expert neurosurgery, or by magic. Or perhaps some version of numerical mechanism is right and y ou could gain master y of a set of kung-fu fi ghting techniques just b y having the corr esponding programs ‘downloaded’ directly into y our computer-like brain. I see no reason why, in principle, the master y of kung-fu techniques acquir ed by a person in such ar tificial fashions would not deser ve the name ‘kno wing how to kung-fu fi ght’ since, pr ecisely, we can imagine that so far as what goes on within you is concerned, that would make no diff erence at all. (Note that a person in that position would be in a similar position as a scientifi c pocket calculator being fed with ne w programs. So, if one agrees that the former is capable of practical kno wledge, ther e should be no r eason why one would disagr ee that the latter is capable of such knowledge too!) That is why I believ e that what is key when it comes to practical knowledge is not so muchhaving an intrinsic ability ashaving got, as having acquired the intrinsic ability, regardless of how it has been gotten or acquired.13 13. The question of the acquisition of kno wledge-how is usually not thematised as such in the literature. Noë (2003) is a r ecent exception, with whom I nonetheless disagr ee when it comes to that question, pr ecisely. Noë defends that “ the possession of abilities is a matter of knowledge-how” (286), but according to him only one kind of ability is relevant to knowledgehow: ‘practical ability’, which he identifi es with the idea of a ‘ skill’. And r egarding skills, he says that as a general r ule, they “aren’t acquired all at once, in a fell swoop . They’re built up or acquired gradually and there may not be sharp lines here” (280). I nevertheless think that there is some truth in what was said earlier regarding the (counterfactual and far-fetched) possibility of knowing how to perform a task by having the adequate procedure integrated by neurosurgery, by magic, or by other fancy means, rather than b y being taught by experience and practice. In such a possibility, practical abilities, skills, could be acquired non-gradually, all at once, in a fell swoop; a subject could know how to F without having been taught how to F by experience and practice, and possibly without even knowing that she can F. Back to the real world, remember that I would like it better if my scientifi c pocket calculator could count as kno wing how to perform certain tasks.

280

What I propose, then, is that we identify possession of practical knowledge with possession of an acquired intrinsic ability (whether its acquisition was by learning from personal experience and practice, or by precise neurosurgery, or by magic, etc.): Pra ctical Kno wledge: A subject S has practical knowledge of how to perform a task F iff she has acquired the intrinsic ability to F that is: iff she has got some F-ing procedure integrated into her, or equivalently: iff she has become a F-ing device. A consequence, of course, is that an adult, a child, a bear, and even my scientific pocket calculator, all become equally eligible for practical knowledge-how. That is a good thing. At least as far as non-human animals are concerned it is just a fact that a specimen can learn how to do something which another specimen does not kno w how to do. (Just compare monocycle-riding bears in circuses with luckier bears.) Another consequence is that appeal to intentionality becomes irr elevant when it comes to practical knowledge of how to perform a task: the tasks that a device performs need not be tasks that it performs intentionally for it to have practical knowledge of how to perform them. That too is a good thing, at least for those who find it controversial to ascribe capability of intentional action to non-human animals and inanimate devices, y et do not find it too controversial to ascribe capability of knowledge-how to some of them. (The issue of kno wledge-how and intentional action will be addressed in the next section.) 14 The other relevant notion of knowledge that I would like to present is very different: Theor etical Kno wledge: A subject S has theor etical knowledge of how to perform a task F iff she has knowledge of, or about, procedures by the execution of which F-ing devices perform F that is: iff it has knowledge of procedures whose integration by such devices constitutes their intrinsic ability to F. The fi rst thing to note is that in this defi nition, I did not say that 14. Charles Wallis (forthcoming) would certainly agree on the latter point, as he defends an account of knowledge-how which, in spirit, resembles my account of practical knowledge. For instance, he says that “kno w-how requires the instantiation of an ability and of the capacities necessary for exploiting an ability—not conscious awareness of purpose or explicit knowledge of rules.” However, the specifics of our respective accounts of knowledge-how differ considerably.

281

in order to hav e theoretical knowledge of ho w to F, the subject had to have knowledge of how a device that has practical knowledge of how to F performs F, but that she had to hav e knowledge of ho w a F-ing device, one that has the intrinsic ability to F, performs F. The reason is that it is unclear what theoretical benefit would be gained from substituting ‘intrinsic ability’ with the str onger ‘practical knowledge’ in the defi nition. On the contrary, the definition as it stands contributes to making theoretical knowledge independent of practical kno wledge, in the sense that for a subject to possess theor etical knowledge of how to perform a task, ther e need not be any subject, not ev en herself, possessing the corr esponding practical knowledge. This, on my opinion, is a very good thing, for I surely want to say that people often hav e theoretical knowledge of how certain devices proceed to perform a certain task, even though those devices ar e not pr oper knowing-how devices. (I do not kno w it myself , but some people probably know how an earthworm makes tunnels.) The definition as it stands makes one ’s possession of theoretical knowledge of how to F absolutely independent of one’s, and even of anyone’s prior possession of practical knowledge of how to F: one may have theoretical knowledge of how to F even if no one has practical knowledge of how to F. (This feature of the notion of theoretical knowledge of how to perform a task shall be largely exploited in the second part of the paper.) I would probably have to amend the definition so as to make clear that it should ideally allow for the possibility that one might hav e theoretical knowledge of how to perform a task that no device will ever actually perform. We might, for instance, insist that one can gain kno wledge of how to do things in non-actual possible, say fictional, worlds, such as knowledge of how to get rid of a z ombie in a Romero movie (Aim at the head!) and even a hor de of z ombies (Run! ), or kno wledge that using an invisibility cape makes one invisible to others in a H arry Potter story. I will simply ignore this complication here, although it might be relevant elsewhere. I haven’t said anything so far about whether theor etical knowledge of how to F had better be conceived of as propositional or non-propositional. I haven’t said anything either about whether it should be ultimately analysed in terms of justifi ed tr ue belief or , instead, be consider ed a basic, unanalysable notion. Theoretical knowledge of how to perform a task can certainly be, but need not be conceived to be, propositional or analysed in terms of justifi ed true beliefs. It can indeed alternatively be conceived of as another kind of ability, besides the one that constitutes practical knowledge. A suggestion could be that while practical kno wledge of how to F 282

consists in an acquired intrinsic ability to F, theoretical knowledge of how to F consists in a further ability (acquired or not) to discriminate performances of a task from performances of other tasks, or good performances from bad performances, or good or better procedures from bad or worse procedures, etc. (This would not necessarily make phylogenetically lower animals incapable of t‘ heoretical’ knowledge: some trained police dogs can certainly discriminate ‘good’ submissive behaviour fr om ‘bad’ aggressive behaviour of potential criminals, standing still in the former case, attacking in the latter case.) Another suggestion in the vicinity would be to make theoretical knowledge of how to perform a task F into the ability to perform certain other tasks, ‘epistemic’ tasks, like forming beliefs about how to F, providing justifi cation for those beliefs, answ ering questions about how to F, reasoning about how to F, etc., and possibly to per form each such task in an enhanced manner , say, in a mor e truth-maximizing way, in a mor e accurate way, etc. (This suggestion would make it a bit mor e difficult for non-human animals to be ascribed such kno wledge, the difficulty being proportional to that of ascribing the ability to believe, justify, answer, reason, etc., to them: the lo wer the animal on the phylogenetic scale, the less epistemic abilities we will grant them, and the more difficult it will be to ascribe theoretical knowledge to them.) I confess sympathy for both suggestions which, probably, constitute the next step and the natural continuation of the present account. 15 Discussing them, however, would take me too far, I think, from the main purpose of this paper, which is not that of clearing the epistemological place from all forms of propositional knowledge, but that of making a little room for a non-propositional, ability-like form of knowledge. For the purpose of this paper, then, it shall not matter too much if, in what follows, theoretical knowledge of how to F is understood as being of a pr opositional nature and as being or inv olving true justified belief about how to perform a task. (I especially do not mind that it be thus understood in Sections 9 and 10 below.) Finally, saying that ‘kno w how’ can sometimes be used to r efer to a practical kind of kno wledge and sometimes be used to r efer to a mor e theoretical kind of knowledge might appear to some like the most trivial triviality ev er. I will not deny this. B ut fi rst, it is the kind of triviality 15. I believe both suggestions to be compatible with the vie w defended by Stephen Hetherington (2006; this v olume) on kno wledge-that and kno wledge-how, on ‘ epistemicity’ more generally speaking. It should, however, be kept in mind that I am not here addressing the issue of the r elationship between knowledge-how and kno wledge-that, but that of the r elationship between knowledge-how and ability.

283

that has been contested, for instance b y Stanley and Williamson (2001), who defend that an ability-like kind of kno wledge would simply be no knowledge at all. S econd, this fi rst part of the paper aimed at clarifying the notions we usually apply, or so I think, when we think about knowledge-how and ability. You do not usually expect a clarificatory work to be especially thought-provoking. So, if it sounds deadly trivial, it is at least to some extent exactly ho w things should be. The only possibly nontrivial part of the present account of knowledge-how so far is to be found, perhaps, in the distinction betw een, and the defi nitions of, the intrinsic and the extrinsic sorts of ability, in the identification of the practical kind of knowledge with an acquir ed ability of the fi rst sort, and the idea that having acquired the ability does not necessarily mean having acquir ed it from experience or practice. The interest of the account, then, does not lie in the account itself. It is rather in the several theoretical, epistemological benefits which can be gained from it. That there are some such benefits is what I shall argue for in the second par t of the paper.16 PART TWO Practical knowledge-how is where I intend the distinction between intrinsic and extrinsic ability to enter the picture of knowledge-how. So, let me first put forward two positive reasons why I think such a distinction ought to enter the picture of knowledge-how at that place, precisely. 5. Knowledge-how and intentional action We do digest food. So this is something we are somehow able to do. But clearly, “digesting food is not the sor t of action that one kno ws how to do” (Stanley and Williamson 2001, 414). This is right, indeed. Does that speak against the idea that knowledge-how and ability go hand-in-hand? Unlike Stanley and Williamson, I think not. Indeed, what seems to me to 16. The definition of practical knowledge does not mark any diff erence in ‘grade’ between the ‘silly’ and ‘basic’ acquired intrinsic ability to wiggle one’s ears and the high-level ‘know-how’ of a heart surgeon. Discriminating between various grades of knowledge-how is, of course, an interesting option, suggested b y Fantl (2008) for example, which the pr esent account is not intended to rule out. It is not, however, an option that shall be pursued her e, the main reason being that we can deal very well with the issues that constitute the material of the next sections with the ‘basic’ definition of practical knowledge in terms of acquired intrinsic ability.

284

transpire most clearly from the example is that it is someho w ‘irrelevant’ to the question of kno wledge-how. The real challenge, in my opinion, is to explain that irrelevance. A fairly natural way to explain it would be to insist that, for a task to be the sort of task one may know how to perform, it has to be “the sort of operation that is executed with intelligence” (Stanley & Williamson 2001, 415), and to maintain that this goes with being the sort of operation that is executed intentionally. This is why, according to Stanley and Williamson, the ‘digestion’ example is not a case of knowledge-how: Digesting food is not the sort of operation that is executed with intelligence. […] Digesting food is not something done intentionally, and that is why it is not a manifestation of knowledge-how. (Stanley & Williamson 2001, 415)

Hawley (2003) would agr ee with them, as she says, r egarding another example: We don’t describe ourselves as knowing how to produce white blood cells. At a push, we might say that someone knows how to produce white blood cells if she kno ws what sor t of diet or dr ugs might pr omote this pr ocess, but to this extent producing white blood cells is indeed an intentional action. J ust as propositional knowledge requires representation, knowledge-how requires intentional action. […] We neither know nor fail to kno w how to produce white blood cells.

Hawley and Stanley and Williamson are right that the white blood cells and the digestion examples are not cases of knowledge-how. It would be somehow ‘improper’ to think so. But is the reason for this the one they advance, that we can speak of knowledge-how only when the range of actions is that of intentional actions or actions executed with intelligence?17 About the digestion example and against S tanley and Williamson’s intended use of it as an argument against the claim that knowing how to F requires having the ability to F, Noë says: I would agree that digesting food is not the sort of action that one knows how to do. But that is because digesting is not the sor t of thing one does (inten17. Note that the idea that tr ue knowledge-how ought to be r estricted to intentional or intelligently per formable actions can not only be put for th by proponents of the claim that knowledge-how is a relation between a subject and a proposition, like Stanley and Williamson, but also b y pr oponents of the alternativ e claim that kno wledge-how is a r elation betw een a subject and an action, like Carr (1981), and ev en by proponents of the opposite claim that knowledge-how is a disposition or ability, like Ryle (1949) himself.

285

tionally or otherwise). Hannah doesn’t digest food; her digestive system does (in her or for her). H annah may have an excellent digestion, but she is not, in that case, ex cellent at digesting. D igestion is not an action that a person or animal can perform. (Noë 2003, 279)

Recall the criterion for telling proper from improper knowing-how devices: ask whether y ou could pr operly say that the F-ing device doesn ’t know how to F if y ou w ere told that it can ’t F. I submit that no one would ever say that a digestive system cannot digest food just because it doesn ’t know how to digest food, and get away with it. S o I totally agr ee with Noë that this is the r eason why w e can’t attribute kno wledge of ho w to digest food to a person. And here is how the present framework supports that explanation. A digestive system is a pr oper digesting device, the person or animal of which it is par t is not. A person or animal ’s digestive system has the intrinsic ability to digest food, but the person or animal does not have the intrinsic ability (no mor e than it has the extrinsic ability , by the way) to digest food. Only intrinsic ability could make a device qualify for (practical) knowledge-how, though. This is why the person or animal cannot be said to know how to digest food. However, the appropriate intrinsic ability has to be someho w acquired for a device to qualify for kno wledge-how. This is why the digestiv e system itself could only be said to kno w how to digest food in an improper use of ‘know how’ in the practical, abilityentailing sense: it is not a pr oper knowing-how device anyway. A similar explanation holds regarding Hawley’s white blood cells example. By having both kinds of ability enter the pictur e of practical knowledge, you can explain the intuitiv e irrelevance of such examples with r espect to the question of kno wledge-how and ability. The explanation does not need to r ely on a r estriction of the range of practically kno wable things to intentional or intelligent actions. As far as practical knowledge is concerned, it does not r equire of any action (kno wn how to be per formed) that it be intentional or intelligent. I n this, I depar t from what appears like a commonplace assumption in the literature.18 On my view, practical knowledge just is an acquired intrinsic ability, nothing more. I now turn to a further motivation for the distinction between intrinsic and extrinsic ability, a motivation related to comparative knowledge-how ascriptions. 18. A recent happy exception is Wallis (forthcoming).

286

6. Greater ability without better knowledge-how, and vice versa Sometimes, we will agree that two subjects know how to perform a task, while at the same time we will acknowledge that one of them is more able to perform the task than the other. Take this example. Both Rafael Nadal and I are tennis players; I know how to play tennis and I play pretty well (suppose); Nadal knows how to play tennis too; but for sure, he can play tennis infinitely faster, stronger, longer, infinitely better; it is not simply a matter of physical capacities: even if I could play as fast, as strong, as long as he can, he would still be technically, tactically, and strategically better; he has trained for long years with an eye to reaching his astonishing level of proficiency. Both N adal and I kno w how to play tennis, but ther e is a good sense in which he is mor e able to do so than I am. I t’s just a fact that he is. Or take this other example, borrowed from Jonathan Ellis (manuscript). As a proficient weight-lifter, I know how to lift weights (I’ve got the right techniques and master the safe positions, for example), I can lift as much as 190 pounds, but I can’t lift more; my weight-lifting partner knows how to lift weights too (he’s got the same right techniques and the same safe positions too), he can lift as much as I can, but he can do even better: he can lift up to 200 pounds. Both my w eight-lifting par tner and I kno w how to lift weights, but there is a good sense in which he is mor e able to do so than I am. In each example, each of the two depicted subjects knows how to perform the relevant task; and in both cases one is more able to perform the task than the other . There is, ho wever, a big diff erence between the two cases. In the first case, because one subject is more able than the other, we will be ready to say that the one who is betterknows better how to perform the task than the other: the diff erence in ‘ability’ between Nadal and me is the kind of diff erence that will make us say that he kno ws better how to play tennis than I do. On the contrary, in the second case, despite the fact that one subject is more able than the other, we won’t be ready to say that the one who is better kno ws better than the other: the diff erence in ‘ability’ between my weight-lifting partner and me is not the kind of difference that will make us say that he knows better how to lift weights than I do. In brief: greater ability sometimes makes for better ‘knowledge-how’, but sometimes it doesn’t. We can easily account for the diff erence betw een the two types of example in light of the defi nition of practical kno wledge-how and the 287

distinction betw een intrinsic and extrinsic ability . In the w eight-lifting case, both my w eight-lifting par tner and I hav e acquired, over months or even years of practice, the intrinsic ability to lift w eights, that is, both of us hav e integrated the pr ocedure(s) to lift w eights in the appr opriate fashion, which includes the mastery of certain techniques and positioning. So, with r espect to intrinsic ability , I am as able as he is to lift w eights, and since the ability in question is an acquired intrinsic ability, I know as much as he does how to lift weights. But with respect to extrinsic ability, there is a diff erence. He is stronger than I am, has bigger muscles than I do. The resources we exploit when we execute the procedure(s) involved in doing bench-press, for instance, are the same, but those at his disposal are obviously available in greater quantity. He is more extrinsically able than I am to lift weights. Differences in extrinsic abilities make for no difference in intrinsic ability, therefore no difference in practical knowledge. That is why it seems right to say that w e both know how to lift weights, that he is more able to lift weights than I am, but not that he kno ws better how to lift weights than I do. Things are diff erent in the tennis example. The difference there is not so much in terms of extrinsic abilities than in terms of intrinsic abilities. Even if w e suppose that I don ’t diff er too much fr om Nadal in terms of extrinsic abilities, that I could hit a smash as str ongly as he can, that I could run as fast as he can, that I could be as quick as he is when it comes to identifying the trajector y of the ball while at the net, etc., he would still play tennis better than me. This is because his long years of specialised training and practice hav e made him mor e intrinsically able than I am to play tennis, and this may be true in two respects. First, for each of the procedures whose integration is inv olved in my intrinsic ability to play tennis, the corresponding procedures that are involved in Nadal’s intrinsic ability to play tennis may be better than mine, for instance in the sense that they would be made of a smaller sequence of actions or that they would be of a lesser cost than mine in terms of execution time, energy, etc. (This may already be true at a purely technical level. For instance, whereas my procedure for smashing involves drawing a nice circle in the air with my racket, his does not.) Second, his intrinsic ability to play tennis may inv olve more procedures than mine, pr ocedures which make him a much mor e efficient player than I will ever be. (For instance, he may have integrated a procedure to wrong-foot the opponent, while I have never come further than ‘Make your opponent run’.) This is why, although we could say that both Nadal and I kno w how to play tennis, w e would judge fr om the 288

difference in ability between him and me that he nevertheless knows how to play tennis better, infinitely better than I do. By having both kinds of ability enter the pictur e of practical knowledge, you can explain the difference between those cases in which greater abilities make for better knowledge-how and those in which they don’t.19 I now turn from the benefi ts of the distinction betw een intrinsic and extrinsic ability to some of the benefi ts of making the other distinction, that betw een practical and theor etical knowledge of ho w to per form a task. 7. Knowledge-how without ability, and vice versa Among the examples that might be cited as ‘ evidence’ against the idea that ability and kno wledge-how go hand-in-hand ar e, on the one hand, those which can be described as cases in which a subject is obviously able to perform a certain task but does not kno w how she performs the task, and on the other hand, those which can be described as cases in which a subject obviously knows how to perform a task that she is ob viously not able to perform herself (in some sense of ‘kno wing how’ and some sense of ‘being able’). Consider the ultimate chicken sex teller .20 He is ultra-pr oficient. He can tell you the sex of a one-day-old chick. It has taken him long years of self-training to r each that lev el of pr oficiency. No one is a natural born chicken sex teller. But suppose that despite his long years of self-training, he still has no single clue ho w he does it; he just does it. H e wouldn’t be able to tell y ou how he does it, he doesn ’t know how he does it, and he might even not be able to teach y ou how to do it, he wouldn ’t fi nd the 19. It goes without saying that there is another way in which one may know how to F better than another, or than oneself at an earlier stage: if theor etical knowledge of how to F is added to one’s practical knowledge of how to F, one may be said to kno w how to F better than one used to, or better than another does. I n many cases, practical kno wledge of how to perform a task will be found in a subject together with a certain amount of theoretical knowledge of how to F. F-ers sometimes have a bit of expertise in F-ing. For instance, I can play tennis and play it a certain way, but at the same time I might also know about what goes on at the biomechanical level whenever I do so, and I might kno w how I ought to play tennis and about the ways in which more efficient players play tennis. The added value of theoretical knowledge in comparative knowledge-how ascriptions is alluded to by Fantl (2008) in his excellent review of the literature on knowledge-how and knowledge-that. 20. Mention of the chicken sex teller case can be found her e and ther e in the literatur e, for instance in Lewis (1996). The case is analysed by Williams (forthcoming) from the specific perspective of the debate over knowledge-how.

289

words. Consider now the case of an expert cognitive scientist who knows everything about per ceptual recognition and a gr eat deal about r elated matters, in par ticular, about ho w a man capable of telling the sex of a one-day-old chick manages to do so; he knows everything about chicken sex telling. B ut despite all that kno wledge, if y ou asked him to tell y ou the sex of this one-day-old chick, he would not be able to do it. A way to put the difference in knowledge between the two men would be this: the first knows how to tell the sex of a one-day-old chick but doesn ’t know anything about how to tell the sex of a one-day-old chick, wher eas the second knows everything about how to tell the sex of a one-day-old-chick, but he doesn’t know how to tell the sex of a one-day-old-chick. In the present framework, this difference is easily dealt with. It is interpreted in terms of a difference between practical and theoretical knowledge of how to perform a task. The chicken sex teller has practical kno wledge of how to tell the sex of a one-day-old chick, because he has the intrinsic ability to do so and has acquir ed this ability thr ough his long y ears (or few months) of self-training. H e has the pr ocedure or set of pr ocedures involved in his telling the sex of a one-day-old chick integrated some where in him. He just lacks the theoretical knowledge of how to tell the sex of a one-day-old chick, because, although he has a good example of a device that does that, that is, himself, for some reason he has never been able to ‘extract’ the pr ocedures he ex ecutes from the many occasions on which he has executed them. He surely knows how to look, and wher e to look for, the subtle visual clues that he will exploit to deliv er his v erdict on the sex of the chick. B ut he doesn’t know how he exploits the clues, that is, the pr ocedure by which he concludes fr om the clues to the v erdict. (Compare with a chemist who has somehow figured out all by herself how to uncover the pr esence of star ch in v arious aliments: does the stuff turn bluish-black—that is the clue—after I put it in contact with the appropriate iodine solution? If, and only if, y es, conclude to the pr esence of starch in the stuff—that is the v erdict. Perhaps she alr eady kne w about the r eactions between amylose and iodine and corr ectly inferred from that knowledge that starch would turn bluish-black in contact with a w ell-dosed iodine solution. The comparison, ho wever, cannot be per fect, because what is needed for the chicken sex teller to each r the relevant theoretical knowledge which he lacks is to extract, or ‘abstract’, the procedures from his practical knowledge, whereas the procedure of which the chemist has theor etical, as well as practical knowledge, was presumably extracted from antecedent pieces of theoretical knowledge.) The cognitive scientist, on the other hand, 290

has only theoretical knowledge of how to tell the sex of a newborn chick. He knows a great deal about how chicken sex tellers manage to tell the sex of a one-day-old chick, and ev en knows the details of the psy chological and neurological aspects of the pr ocedures involved in per forming that task. He knows about most aspects of what constitutes the intrinsic ability of a chicken sex teller. He just lacks practical knowledge of how to tell the sex of a chick. He has never learnt how to do so in the sense of not having himself acquired the intrinsic ability to do so . He has not integrated the procedures that can make one a chicken sex teller. Other examples that might be cited against the knowledge-how as ability hypothesis are, on the one hand, those that can be described as cases in which a subject is no longer able to perform a task which she nonetheless still knows how to per form,21 and on the other hand, those that can be described as cases in which a subject no longer kno ws how to perform a task that she is still able to perform (in some sense of ‘knowing how’ and some sense of ‘being able’). On the one hand, you have a master pianist who, unfortunately, loses her arms in a car accident. S he is no longer able to play the piano, in a very good sense of ‘being able’. But she still knows how to play the piano, in a certain sense of ‘knowing how’. On the other hand, you have a welleducated musicologist. Being well-educated, he can play the piano . As a musicologist, he knows everything there is to know about piano and piano performance; suppose he specialises in this instr ument. Unfortunately, his brain gets damaged in a car accident. As a consequence, he cannot remember anything of what he used to kno w about piano and piano performance, for the r eason that the exact par t of his brain that was in charge of the kno wledge in question was destr oyed in the car accident. But he can still play the piano. This is surprising even to him, because he cannot even name the instrument. He just sits at the piano, his hands do the rest. He no longer knows how to play the piano, in a certain sense of ‘knowing how;’ but he is still able to play the piano, in a good sense of ‘being able’. The difference betw een the two cases can easily be explained in the present framework. The master pianist loses her ability to play the piano in the car accident. What she loses is her extrinsic ability to play the piano, 21. Such counterexamples to the claim that being able toF is sufficient for knowing how to F abound in the literature. Good places to fi nd such examples are Brown (1971, 1974), Ginet (1975), Stanley and Williamson (2001), and Snowdon (2003). The master pianist example below is taken from Stanley and Williamson (2001).

291

but she retains her intrinsic ability to do so. A crucial part of the resources and of the sub-devices r equired b y an ex ecution of the complex set of procedures involved in her intrinsic ability to play the piano is no longer there. But that does not aff ect her intrinsic ability to play the piano . A long time ago having arms was a necessar y condition for her to integrate the procedures, that is, for her to learn ho w to play. And no w that she is no longer extrinsically able to play , the pr ocedures themselv es hav e remained in place. S he is still a piano-playing device. I f, by the miracles of micro- and neur osurgery, she w ere given new arms and if ev erything went well, she might even be able to play again, perhaps not as well as she used to, but still. She would have regained her full extrinsic ability to play the piano. As to the w ell-educated musicologist, he loses his kno wledge of how to play the piano in the car accident. What he loses is only his theoretical knowledge of how to play the piano, but his practical kno wledge remains in place. His extrinsic as well as his intrinsic ability to play the piano are left unaffected by the brain damage. His past knowledge of how piano-playing devices, including him for sur e, perform the task, of the procedures that such devices ex ecute to this end, is no longer ther e. But he has not thereby ceased to be a piano-playing device himself for all that. The procedures for playing the piano ar e still integrated in him, he has remained intrinsically able to play, he has remained extrinsically able to play, and this is why he can still play. The distinction between practical and theoretical knowledge-how and their respective defi nitions provide a straightfor ward explanation of different kinds of possible examples that might be used against the idea that a tight connection exists betw een knowledge-how and ability . The tight connection exists: it is betw een practical kno wledge and intrinsic ability. 8. Ways, good ways, better ways The definition given earlier of theoretical knowledge of how to perform a task is likely to be reminiscent of Stanley & Williamson (2001)’s proposed truth-conditional analysis of kno wledge-how ascriptions which, r oughly put, is this: ‘S knows how to F’ is true iff she knows, of some contextually relevant way to F, that it is a way to F.

292

I talk about p‘ rocedures’ while they talk about ways’. ‘ But there is of course more than just a terminological difference between my proposed account of knowledge-how and theirs. A ccording to them, the ability-entailing use of knowledge-how ascriptions reduces to a special case of the previous equivalence: in a context in which it gets an ability-entailing r eading, ‘S knows how to F ’ is true iff she knows, of some contextually relevant way for her to F, that it is a way for her to F and entertains the pr oposition that it is so under a practical mode of pr esentation, which they describe as a certain mode under which a subject may think about ways of doing something, and about which they add that thinking of ways under a practical mode of presentation “undoubtedly entails the possession of certain complex dispositions”. (Stanley & Williamson 2001, 429) A big difference, then, between their account and mine is that they see just one sense of ‘know how’ where I see two distinct senses of ‘kno w how’, one practical, the other theoretical.22 Talking about ‘ways,’ however, will help me formulate what I consider a benefi t of distinguishing betw een a practical and theor etical kind of knowledge of how to F. If I say that S knows how to F, I am not jut saying that she has knowledge of some, or ev en the way(s) to F. For there aren’t just ‘ways;’ there are ‘good’ and ‘bad’ ways, and ther e are ‘better’ ways than others. I f Dr. Ross knows how to operate cataracts, he knows that poking the patient’s eye balls is not a good way to operate cataracts. And he also kno ws that 22. As Stanley and Williamson (2001) r emark, a kno wledge-how attributing sentence of the form ‘S knows how to F ’, such as: a. Hannah knows how to ride a bike admits of two kinds of ‘normative’ readings, one which we may call ‘prescriptive,’ as in: b. Hannah knows how she (or one) ought to ride a bike And another one, more frequent, which we may call ‘permissive,’ as in: c. Hannah knows how she (or one) could ride a bike. They account for this fact, grosso modo, in terms of an ambiguity in ‘kno w how’ constructions between a ‘mention-all-of-the-ways’ kind of reading and a ‘mention-some-of-the-ways’ kind of reading, an ambiguity that context is in charge of r esolving: assuming that ther e is more than one way to ride a bike, accor ding to the fi rst kind of reading, Hannah knows, of the ways that are ways for her (or for one) to ride a bike, that it is a way for her (or for one) to ride a bike; according to the second kind of r eading, she kno ws, of some ways, that they ar e ways for her (or for one) to ride a bike. In most conversational contexts, the second reading prevails because “it is irrelevant to our communicativ e purpose that our audience come to kno w, of every way which is a way for x to F, that x knows that [it] is a way for x to F ”. (426) A similar explanation could also easily be formulated within the framework I propose, simply by replacing ‘ways’ with ‘procedures’ in Stanley and Williamson’s explanation of the fact about normative readings of ‘S knows how to F ’.

293

the way in which the kind and cautious D r. Green operates cataracts is better than the way in which the odious and ner vous Dr. Romano operates cataracts. We might be tempted to generalize and say that when one knows how to per form a task, one does not only kno w that this or that way is a way of performing the task; one also knows that this or that way is a good or a bad way of per forming the task, and that such-and-such a way of per forming the task is better than such-and-such other way of performing the task. However, the proposed generalization would be misleading. Consider the following case. Hannah knows how to ride a bicy cle; but as it is still new for her, when she rides a bike, she contracts ev ery single muscle of her body. That’s her way of riding a bike. I t works. But that’s a bad way, and she may not yet know that it is. Also, although she knows how to ride a bike, she may not know that the way in which her mor e relaxed friend rides a bike is better than her o wn, because she has not y et realized that you can ride a bike as effi ciently when y ou’re relaxed as when y ou’re all contracted. (We may even suppose that she doesn’t even care about how her friend rides.) So here we have a diff erence: sometimes, knowing how to F goes with knowing good and better ways from bad and worse ways; but sometimes, this is not the case. The present account easily explains the diff erence by appealing to the distinction between theoretical and practical knowledge of how to perform a task. Both D r. Green and D r. Romano are cataract-operating devices; they have integrated procedures for operating all sorts of things, including cataracts; they hav e practical kno wledge of ho w to do these things. F or sure, Dr. Ross possesses this kind of kno wledge too. But when we insist that if he knows how to operate cataracts, he must know that this or that is a good or a bad or a better way to operate cataracts, what w e have in mind is another kind of knowledge-how, not practical, but theoretical this time. We are ascribing to Dr. Ross theoretical knowledge of how cataractoperating devices proceed, knowledge of procedures whose integration by such devices constitutes their intrinsic ability to operate cataracts. And in ascribing such kno wledge to him, w e expect that on the basis of that knowledge he be able to compar e diff erent procedures and to tell which one is good, which one is bad, and which one is better . In the second case, on the contrary, the subject has almost exclusively practical knowledge of how to perform the task. Hannah has learned how to ride a bike; she has acquired an intrinsic ability to do so. Even if it is still new for her, she can ride a bike. S he knows how to ride a bike in that sense, but she doesn ’t 294

know much of or about how to ride a bike. The only bike-riding procedure she has paid attention to until now is hers, we may suppose, and this has given her too little theoretical knowledge of how to ride a bike for her to be in a position to compar e and tell whether the way she rides a bike is good, or bad, or better than her friend’s. But what is a good, or a bad, or a better way to operate a cataract? We know how to deal with good, bad, and better ways because w e already know from Section 3 how to deal with good, bad, and better procedures. Dr. Ross can tell that this or that is a bad way or pr ocedure to operate cataracts if, for instance, it inv olves too many actions, or if it takes too much time or too much place or too much of other kinds of r esources to execute it. And he can tell that D r. Greens’ way(s) or pr ocedure(s) is better than Dr. Romano’s way(s) or procedure(s) on the basis of a similar criterion: that of Dr. Green involves fewer actions, or less time, or place, or surgical instr uments, or money, etc., than that of D r. Romano. As to Hannah, she just lacks suffi cient theoretical knowledge of how to ride a bike to be able to tell the good, fr om the bad, fr om the better ways or procedures. The only way or pr ocedure she seems to kno w (and to car e) about, anyway, is her o wn way of riding a bike. I t is not v ery good, but this, she cannot tell. By drawing the proposed distinction between practical and theoretical knowledge-how, we can explain the diff erence between the two types of case. 9. Knowing how by mistake An additional benefit of the distinction between practical and theoretical knowledge of how to perform a task is that it allows us to refine our reflection, started in Section 4, on the Gettierability of knowledge-how. I concluded from that section that Stanley and Williamson had not shown that knowledge-how could be Gettiered. Now, I must add that that was only insofar as the ability-entailing, practical use of ‘know-how’ was concerned. (The use of ‘know how to fly’ in the flight simulator example was obviously ability-entailing.) Although I think that the kno wledge referred to by an ability-entailing use of ‘kno w-how’ cannot be G ettiered, I nev ertheless think that w e can conceiv e of G ettier cases for that kno wledge which is referred to by a non-ability-entailing, or theoretical, use of ‘know-how’. Hawley (2003)’s annoying smoker example appears to be one such case: Susie thinks she anno ys Joe by smoking, wher eas in fact J oe is anno yed 295

by her concomitant tapping on her cigar ette box, which she does whenever she smokes; does S usie know how to annoy Joe? Here is the answ er I submit: if we are talking about theoretical knowledge of how to annoy Joe, no; if we are talking about practical knowledge of how to annoy Joe, then yes. Let me explain. On the one hand, S usie believes that to anno y Joe what she can do is smoke cigar ettes. Her belief is tr ue: that is indeed one pr ocedure or way to anno y Joe. Her belief is justifi ed: w e may conjectur e that it is justified inductively on the basis of her repeated observation that Joe has shown signs of ner vousness every time she has smoked cigar ettes in his presence. Despite having a justifi ed true belief about ho w to anno y Joe, there is a good sense in which y ou can nevertheless say that she does not know how to anno y Joe for all that. B ut what makes y ou say that she doesn’t know how to annoy Joe? The reason you will advance for saying so will be that her justified true belief about how to annoy Joe is somehow grounded in a false belief that it is her smoking cigar ettes that anno ys Joe, while it is in fact her concomitant tapping on the cigar ette box that annoys Joe. Granted. Here we are indeed dealing with a G ettier case for knowledge-how. However, this holds only as long as the knowledge we have in mind is theoretical. For on the other hand, there is also a good sense in which you can say that Susie does know how to annoy Joe. If you want to annoy Joe, don’t ask her how you could do that. She will tell you that you can do that by smoking cigarettes, and that will be a misleading advice, especially if unlike her you are not given to tapping on y our cigarette box when you smoke. But if you ask her to annoy Joe, you can trust her for that. She’ll do that per fectly. Where did she learn ho w to do that fr om? From her repeated observation that Joe has shown signs of ner vousness every time she has smoked a cigarette in his presence. She has learnt, and the learning bore its fr uits: she no w knows how to anno y Joe. How does she do that? By smoking cigar ettes the way she does, which includes her tapping on her cigarette box. The procedure she executes to annoy Joe is simple: she lights up a cigarette and smokes it while tapping on her cigar ette box. She’s got the procedure. Of course, she has mistaken beliefs about it. But she’s got it in another sense: she has integrated the procedure, and that was by learning from her repeated experience of smoking in Joe’s presence. Susie has acquired the intrinsic ability to annoy Joe. In other words, Susie has practical knowledge of how to annoy him; this knowledge is left un-Gettiered. But she lacks the corresponding theoreti296

cal knowledge of ho w to anno y him, because the r elevant knowledge is indeed Gettiered. (I would be r eady to make a fur ther step and say that in general, Gettierability is a property of theoretical knowledge of how to F, but not a property of practical knowledge of how to F, for any task F.) This, I think, is a corr ect way to put things. I n my opinion, the simple fact that it allows us to put such a nuanced diagnosis on examples of that kind, counts as a good enough argument for something in the vicinity of the proposed distinction between practical and theoretical knowledge of how to perform a task. 10. Opacity in knowing how contexts A last benefi t of the pr esent account is that it allo ws us to r econsider an important worry with the claim that knowledge is an ability of some kind. The worry is that ascriptions of knowledge-how appear to be non-extensional, whereas ascriptions of abilities seem to be extensional. Consider the following example from Carr (1981), who is usually credited for being the first to have expressed that worry: Suppose a famous dancer was to per form before an audience, an item of his r epertoire to which he has himself giv en the [title ‘ A per formance of Improvisation No. 15’.] To the astonishment of a member of his audience who just happens to be an exper t on communications, the mo vements of the dancer turn out to resemble an accurate (movement perfect) semaphore version of Gray’s ‘Elegy’, though the dancer is quite unaware of this fact. We may describe what is seen by an audience member as [a semaphore recital of Gray’s ‘Elegy’.] (Carr 1981, 407)

The dancer can be said to be able to perform a semaphore recital of Gray’s ‘Elegy’ since she is able to bring about a per formance of Improvisation No. 15. On the contrary—or so Carr thinks—, “although we can describe the dancer as knowing how to bring about [a per formance of Improvisation No. 15,] we cannot reasonably suppose that he also kno ws how to bring about [a semaphore recital of Gray’s ‘Elegy’.]” (idem) Should we consider the diff erence between knowledge-how contexts and ability contexts with respect to extensionality as evidence that knowledge-how is not an ability of some kind? I think not. O f course, I agr ee that in a sense the dancer does not kno w how to per form a semaphor e version of Gray’s ‘Elegy’; but in my opinion, there is also a good sense in which he does know how to bring about a semaphore version of the poem 297

given that he has the ability to bring about what he calls ‘ a performance of Improvisation No. 15’. To make the last point clear er, consider another case. You have been trying to teach y our three-year old child ho w to draw a smiley for the past couple of w eeks. But for some r eason what y ou told her that y ou were tr ying to teach her was ho w to draw a happy face—say , because you thought that ‘happy face ’ would say mor e to her than ‘ smiley’. The learning bore its fruits: now she knows how to draw a happy face. This is true, she can draw a happy face. D oesn’t she thereby know how to draw a smiley? There is a sense of ‘kno w how’, I submit, in which she does. I t is the sense in which y ou could go to y our neighbour and declar e how proud you are that y our child kno ws how to draw a smiley . Of course, if you want her to sho w that she can draw a smiley , you will have to ask her to draw a happy face, or else teach her that ‘ smiley’ means the same as ‘happy face’. But that does not make an iota of diff erence to the fact that she already knows how to draw a smiley in a sense. S he knows how to draw a smiley because y ou have taught her so . Suppose that during a paediatric exam the paediatrician asked y ou to tick the right bo xes on a skill-evaluation questionnaire for your daughter. Would you tick the box next to the question ‘D oes the child know how to draw a smiley?’ (S uppose that might be impor tant for the child ’s future development.) I bet you would tick the box, and that anyone in your shoes would. I cannot see why someone would not … unless they have another sense of ‘know how’ in mind instead of the ability-entailing sense. 23 And in that other sense, she does not know how to draw a smiley giv en that she lacks knowledge that ‘smiley’ means the same as ‘happy face’. What the smiley case tends to suggest is that ability-entailing kno wledge-how ascriptions are extensional. 24 On the pr esent account, that is exactly what is to be expected if ability ascriptions in general ar e extensional. According to the present account, if there can be ability-entailing 23. Another example that suggests that ability-entailing knowing-how ascriptions are transparent is this. Suppose that a Canadian French knows how to cook what is called a ‘hambourgeois’ in the isolated community he liv es in; and suppose that he doesn ’t know that what he calls a ‘hambourgeois’ is the same thing as what people in the rest of Canada would call a ‘hamburger’. From the fact that he lacks the latter item of knowledge, does it follow that he doesn’t know how to cook a hamburger despite his knowing how to cook a hambourgeois? 24. Carr himself notes that “it appears […] that wher eas knowing how contexts are nonextensional for some substitutions of alternative descriptions of a given action, they are not for others.” (Carr 1981, 408) He does not draw the same conclusion as I do from that observation, though.

298

uses of ‘know how’, it is because there is a sense of ‘know how’ in which it refers to a kind of knowledge that is practical; and that practical knowledge of how to do something is an acquir ed intrinsic ability, therefore it is an ability. So it shouldn’t be a surprise that ability-entailing knowledge-how ascriptions are extensional. Your daughter has learnt ho w to draw a smiley, although she knows it as a ‘happy face’, and has thereby acquired the intrinsic ability to draw a smiley; she therefore has practical knowledge of how to draw a smiley. It is in this sense that she can be said to know how to draw a smiley, the ability-entailing sense. I think that something similar holds in the case of the dancer. The case of the dancer is similar to the case of y our daughter. He has learnt ho w to perform what he calls ‘ a performance of I mprovisation No. 15’ (this can be extrapolated fr om Carr’s mention that the per formance is that of an item of the dancer’s ‘repertoire’), but which is identifi ed by the expert audience as ‘a semaphore recital of G ray’s ‘Elegy’’. The learning bor e its fruits and no w he can per form that item of his r epertoire. If you want him to perform a semaphore version of Gray’s poem, you will have to ask him for his performance of Improvisation No. 15. But this doesn’t change the fact that the dancer is able to per form the former . And he kno ws how to do it in a sense. I n this sense, the ability-entailing sense, ‘kno w how’ refers to practical kno wledge of ho w to per form a semaphor e version of Gray’s ‘Elegy’, a kind of kno wledge that he surely has, given that he has learnt ho w to per form such a v ersion, although under the name ‘performance of Improvisation No. 15’. ‘Know how’, as used in its ability-entailing sense, refers to practical knowledge, that is, to some acquired intrinsic ability, therefore to an ability. If this is true, then we can expect that, like any ability ascriptions, the appr opriate ability-entailing ascription of knowledge-how to the dancer will be transpar ent or extensional. I believe that this is indeed the case. Suppose that someone comes to the expert and tells him that it would be nice if a dancer could come up with a semaphore performance of Gray’s poem, the exper t could then answ er that he knows a dancer who knows how to do that perfectly. He would be right in making that answer, because his use of ‘know how’ here would be ability-entailing. I do not want to be understood as saying that ther e is no extensionality failure of some kno wledge-how ascriptions. There are. In the smiley case, there is a sense in which your daughter does not know how to draw a smiley when she does not know that ‘smiley’ means the same as ‘happy face’; and in the dancer case, ther e is a sense in which he does not kno w 299

how to perform a semaphore version of Gray’s ‘Elegy’ when he does not know that what he calls ‘a performance of Improvisation No. 15’ is what others would call ‘a semaphore expression of Gray’s ‘Elegy’. But that is not the ability-entailing sense. The knowledge that is being denied here is not of a practical kind: it cannot be denied that y our daughter has acquir ed the ability to draw a smiley or that the dancer has acquir ed the ability to perform the semaphor e version of G ray’s ‘Elegy’. The knowledge being denied is more of the theoretical kind. The problem has to do with what your daughter and the dancer kno w about what they hav e respectively acquired the ability to do. If you ask your daughter how to draw a smiley, she won’t be able to answer you; and if you ask her to draw a smiley, she won’t be able to draw one. Likewise, if you ask the dancer how to perform the semaphore expression of Gray’s ‘Elegy’, he won’t be able to answer you either; and if you ask him to perform a semaphore recital of Gray’s ‘Elegy’, he won’t be able to do so . Perhaps we tend to make it a r equirement for possessing the theoretical kind of knowledge of how to perform a task F that the subject be able to answ er questions such as ‘How to F ?’, or that she be able to r espond to requests such as ‘P erform F ’, both things that she cannot do if she is ignorant that ‘F-ing’ means the same thing as what she can indeed perform but knows by another name, ‘G-ing’. Earlier I chose to remain neutral on whether theoretical knowledge of how to F was to be conceived of as propositional or not. Yet, there seems to be a close connection between what has just been said about theor etical knowledge of ho w to F and the behaviour of kno wing-that contexts with respect to opacity. Knowledge-that ascriptions ar e opaque. To take a commonplace example, Lois Lane kno ws that S uperman can fl y but does not kno w that Clar k Kent can fl y, although S uperman and Clar k Kent are one and the same person, and that is because she does not know that they are one and the same person. And because she does not kno w this—and in a way parallel to what we just said about the daughter and the dancer’s theoretical knowledge—, if you asked Lois Lane how to contact Superman, she might not be able to answer you, even if she could answer the question how to contact Clark Kent; and if you asked her to contact Superman, she wouldn’t be able to do it. Thus, a similar type of ignorance as that involved in the daughter and the dancer case seems to hav e similar eff ects as well on the theor etical knowledge-how as on the opacity of the knowledge-that that we can ascribe to Lois. I f theoretical knowledge of how to perform a task w ere propositional, that is, if it w ere a kind of knowledge-that, the similarities should be no surprise and that would 300

explain the non-extensionality of non-ability-entailing kno wledge-how ascriptions. Ascriptions of theoretical knowledge of how to do things ar e opaque. Ascriptions of practical knowledge of ho w to do things, that is, abilityentailing ascriptions of kno wledge-how, ar e transpar ent. This sheds a new light on cases such as Carr ’s Gray’s ‘Elegy’ example and the smiley example. Practical knowledge of how to F is an acquired intrinsic ability; it is no surprise then that like any other kind of ability ascription, abilityentailing knowing-how ascriptions are transparent. The present proposal allows us to explain away the worry concerning the purported opacity of knowing-how contexts. 11. Conclusion We started our discussion with what we presented as a ‘fact’ about knowledge-how ascriptions. The present account throws an interesting light on that fact, as we can appeal to the double distinction on which the account relies to explain the diff erence between the ability-entailing and the ability-neutral uses of ‘know how’. On the present account, there are two distinct kinds of knowledge that one can ascribe to a subject b y using ‘kno w how’, the practical and the theoretical, and there are also two distinct kinds of ability that a subject may or may not possess, the intrinsic and the extrinsic; to hav e practical knowledge of how to perform a task is to have the acquired intrinsic ability to perform the task. It is in that sense of ‘kno wing how’ and in that sense of ‘being able ’ that the following: (1) A subject S knows how to perform a task F only if S is able to F is true. Because I intend to refer to the practical kind of knowledge when I declare ‘I kno w how to wiggle my ears ’, ‘I kno w how to blo w smoke rings’, ‘the Italian cook knows how to cook an excellent risotto’, ‘The eye surgeon knows how to operate cataracts’, ‘The Belgian monk knows how to brew a per fect beer’, I then want to be understood as ascribing the corresponding acquired intrinsic ability to the corresponding subject, the kind of ability that makes me an ear-wiggler , a smoke-ring blo wer, that makes the cook a risotto-cooker, the surgeon a cataract-operator, and the monk a beer-brewer. And unless context makes my audience think other301

wise, or unless I make it explicit that they should think otherwise, it is as ascribing the ability in question that my kno wledge-how ascription shall be understood. The same schema, ho wever, is not tr ue in another sense of ‘kno wing how’ and it is not tr ue in another sense of ‘being able ’. When that is the case, it can be either because the kind of kno wledge referred to in that other sense of ‘know how’ is not the same kind of knowledge as before, or because it is not an ability of the same kind as before, or for both reasons. In other words, there are three possible ways in which the left-hand side of (1) can be tr ue while its right-hand-side is false, that is, thr ee possible ways in which a subject can be ascribed kno wledge of how to perform a task that she lacks the ‘ability’ to perform. In a fi rst kind of case, while the kno wledge being ascribed is of the practical kind, the relevant inability is of the extrinsic kind. This is what happens in the case of the master pianist, already dealt with in Section 7. On the one hand, as a master pianist, she must have acquired the intrinsic ability to play the piano, and this acquired intrinsic ability is the practical knowledge that one ascribes to her by saying that she knows how to play the piano. In this sense the antecedent of: (3) If the master pianist knows how to play the piano then she is able to play the piano is true. On the other hand, having lost her arms in the car accident, she has become extrinsically unable to play the piano, and in that sense the consequent of (3) is false. In a second kind of case, while the relevant inability is of the intrinsic kind, the kno wledge being ascribed is of the theor etical rather than the practical kind. This is what w e hav e in the case of the world ’s leading expert in the biomechanics of cy cling. She has theor etical knowledge of how to ride a bicy cle, knowing nearly everything there is to know about the procedures (the complex biomechanical pr ocesses) that ar e executed when a rider rides a bike; in this sense the antecedent of: (4) If the expert in biomechanics knows how to ride a bicycle then she is able to ride a bicycle is true. But she is not herself a bike-rider , as she has nev er even tried to ride a bike, and therefore has never had a chance to acquir e the intrinsic 302

ability to ride a bike (if she gets on a bicy cle, she will fall); and in that sense the consequent of (4) is false. In a third and final kind of case, neither is the knowledge being ascribed practical nor the relevant inability intrinsic. The case of allergy to alcohol is one such case. You are allergic to alcohol. For this reason you have never been able to drink enough to get drunk, and you are still unable to do so. The inability here is extrinsic, as in or der to get dr unk, your body must ingest a certain amount of alcohol—that is par t of the regular procedure for getting drunk—and that is a task that y our body cannot perform for the obvious reason. In that sense the consequent of: (5) If you know how to get drunk then you are able to get drunk is false. B ut on the other hand, y ou might kno w ho w those who ar e extrinsically able to get dr unk manage to get dr unk, say, by swallowing glass after glass of alcohol. The knowledge you would then have would be purely theoretical, and in this pur ely theoretical sense the antecedent of (5) would be true. The present account thus yields an explanation for the linguistic fact w e started with, an explanation that is of an epistemological or der. It might be protested that the cost of accepting the account is too great a cost, as it has the consequence that much of what makes for human specificity fades away in a sense: humans, non-human animals, and scientific pocket calculators become all equal befor e practical knowledge. I cannot see why the proposed account of practical knowledge should give rise to any such pr otest in the fi rst place. For even if we allow that some non-human animals and such things as pocket calculators ar e capable of practical kno wledge, that is, of acquir ed intrinsic ability, there might still be a range of such abilities that ar e specifi c to humans. I n that case an interesting philosophical task would be, pr ecisely, to unco ver those intrinsic abilities and to discover how and to what extent they contribute to making us so special. 25

25. I am grateful to L ucca Baptista, Yannick Chin, Paul Egré, Martin Montminy, Daniel Ramalho, Manuel Rebuschi, João Sàágua, and M ichael Sweeney, for stimulating discussion, useful input and feedback on early versions of the paper, and/or encouragements.

303

REFERENCES Bronaugh, R. 1968. ‘The Logic of Ability Judgements’, The Philosophical Quarterly 18, 122–130. Brown, D. 1971. ‘Knowing How and Knowing That, What’, in Ryle: A Collection of Critical Essays, (eds.) O. P. Wood and G. P itcher, Anchor Books, G arden City, 213–248. — 1974. ‘Reply to Brett’, The Canadian Journal of Philosophy 4, 301–303. Carr, D. 1979. ‘The Logic of Knowing How and Ability’, Mind 88, 394–409. — 1981. ‘Knowledge in Practice’, The American Philosophical Quarterly 18, 53–61. Dubuc, B. 2007. ‘The Brain from Top to Bottom’, from: . Elgesem, D. 1997. ‘The Modal Logic of Agency’, The Nordic Journal of Philosophical Logic 2, 1–46. Ellis, J. (manuscript). ‘Knowing What to Do What You Can’t Do’. Last access June 8, 2008, from: . Fantl, J. 2008. ‘Kno wing-How and Kno wing-That ’, Philosophy Compass 3, 451– 470. Franklin, R. 1981. ‘Kno wledge, B elief, and U nderstanding’, The Philosophical Quarterly, 31, 193–208. Ginet, C. 1975. Knowledge, Perception, and Memory, Reidel, Dordrecht. Hawley, K. 2003. ‘S uccess and Kno wledge-How’, The American P hilosophical Quarterly 40, 1, 19–31. Hetherington, S. 2006. ‘H ow to Kno w (that Kno wledge-That is Kno wledgeHow)’, in Epistemology Futures, (ed.) S. Hetherington, Clarendon Press, Oxford, 71–94. Kenny, A. 1957. Mental Acts, R.K.P., London. Lewis, D. 1988. ‘What Experience Teaches’, Proceedings of the Aristotelian Society 13, 29–57. Cited as r eprinted in Papers in M etaphysics and E pistemology, D. Lewis, Cambridge University Press, Cambridge, 262–290. — 1996. ‘Elusive Knowledge’, The Australasian Journal of P hilosophy 74, 549– 567. Noë, A. 2005. ‘Against Intellectualism’, Analysis 65, 278–290. Poston, T. (forthcoming). ‘Know How to Be Gettiered’, Philosophy and Phenomenological Research. Rumfitt, I. 2003. ‘Savoir Faire’, The Journal of Philosophy 100, 158–166. Ryle, G. 1949. The Concept of Mind, University of Chicago Press, Chicago. Snowdon, P. 2003. ‘Knowing How and Knowing That: A Distinction Reconsidered’, Proceedings of the Aristotelian Society 104, 1–29.

304

Stanley, J. and Williamson, T. 2001. ‘Kno wing How’, The Journal of P hilosophy 98, 411–444. Taylor, R. 1960. ‘I Can’, The Philosophical Review 69, 78–89. Wallis, C. 2008. ‘Consciousness, Context, and Kno w-How’, Synthese 160, 123– 153. Williams, J. (forthcoming). ‘Propositional Knowledge and Know-How’, Synthese.

305

Grazer Philosophische Studien 77 (2008), 307–324.

KNOWING-THAT, KNOWING-HOW, AND KNOWING PHILOSOPHICALLY Stephen HETHERINGTON University of New South Wales Summary This paper outlines how we may understand knowing-that as a kind of knowinghow-to, and thereby as an ability. (Contrast this form of analysis with the more commonly attempted reduction, of knowing-how-to to knowing-that.) The sort of ability in question has much potential complexity . In general, questioning can, but need not, be part of this complexity. However, questioning is always an element in the complexity that is philosophical knowing. The paper comments on the nature of this particular form of knowing.

The conception of knowing as being, most fundamentally, knowing-that continues to dominate epistemological analyses of knowledge. But it need not. This paper highlights a conceptual alternative, which is then applied to the potentially special case of philosophical knowledge. A person’s skill in questioning will be seen to contribute constitutively and essentially to the nature of such knowledge. 1. A standard conception of knowing Vigorous epistemological debates persist as to the natur e of what it is to know that p. Vigorous, yes; but exhaustiv e, no: some commitments are rarely, if ev er, questioned ev en within these debates. A fe w of those persisting commitments ar e clearly display ed in the follo wing standard, albeit generic, pictur e of kno wing that p, a pictur e that has become an epistemological reflex: To know that p is to be in a state of knowing that p. Most typically, this is deemed to be a state of accurately believing that p, a state possessing

precise boundaries 1 and constituted in par t b y some fur ther specifi c ‘knowledge-making’ properties.2 For example, the belief will hav e been r eliably formed; or it will be evidentially w ell suppor ted; and so for th. The belief is the single ‘ thing’ or ‘substance’ which is the kno wledge ‘within’ the kno wer. It would be knowledge by having these further properties. 2. An epistemic diaspora Section 1’s generic picture of knowing should be familiar to epistemological readers (especially Western analytic epistemologists). Even so, I wish to question a few aspects of it. If my generic counter-picture is correct, as far as it goes,3 then there is no single ‘thing within’ that is the knowledge (even once appropriate further properties are present). Nor will kno wing be a state, or at least not one as simple as a belief thatp.4 Nor also will knowing be the right kind of thing clearly to hav e precise boundaries. (So, will there be the possibility of a person’s knowledge that p and her knowledge that q blurring somewhat into each other, partially overlapping? Maybe; which sounds correct to me.) The epistemological stakes are therefore high. We begin carefully, then, by reflecting on the fact that knowing that p both expresses and generates many cognitive outcomes. Accurate believing is one of these; only one, though. I will call this grouping p’s epistemic diaspora. It includes something like the following listing of p-related cognitive phenomena:5 1. For mor e on this aspect of the usual epistemological pictur e, see H etherington (2006b). 2. In other words, a true belief that p is rarely thought to be suffi cient for knowing that p. For dissenting voices, see Sartwell (1991; 1992); Goldman (1999, 23–5); Hetherington (2001, ch. 4; 2007a). 3. Elsewhere (2006a; 2007b), I hav e presented similarly motiv ated outlines of the metaphysics of kno wing. Each includes and emphasiz es further details within what is at pr esent a developing conception of the nature of knowing. 4. This is not to say that believing that p is absolutely simple. I said ‘as simple’ — however simple that might be. O n some conceptual options av ailable for understanding the natur e of believing, see Armstrong (1973, Part I). 5. I am rendering these as success phenomena, with each including a r elated accuracy condition. I take accuracy to be the most central featur e of kno wing, in whatev er form(s) such accuracy might take. As to what form(s) it would take, note that in the case of questioning, for example, some questioning does, while some does not, respond to aspects of the pertinent truth

308

believing accurately that p, remembering accurately that p, asser ting accurately that p, questioning accurately in p-related ways, answering accurately in p-related ways, r epresenting accurately that p, sensing accurately that p, reasoning accurately in p-related ways, explaining accurately in p-related ways, hypothesizing accurately inp-related ways, acting accurately in p-related ways, and so on. Section 1’s epistemologically standar d picture plucks just one privileged member from that epistemic diaspora—usually, one’s believing accurately that p—so as to be the ‘inner stuff’ that, once augmented by some favoured properties, can constitute knowing that p. Yet already that approach is puzzling. Why must knowing that p be a belief at all (even an accurate one), out of all the members ofp’s epistemic diaspora?6 More generally, why must knowing that p be any single member of that epistemic diaspora? Even when augmented by pertinent properties (typically, some sort of epistemic justifi cation, such as reliability or good evidence), accurately believing that p is only one of manyequally epistemically relevant members of p’s epistemic diaspora. Each of these members is equally epistemically indicative of knowing that p. This should prompt us to wonder whether each is equally epistemically constitutive, too, of knowing that p. Even philosophers adopt habits of thought, often unwittingly so. And clearly epistemologists hav e become comfor table with thinking of any instance of knowing as being a case of augmented accurate believing.Ways of talking that accommodate this picture have also grown tall within the world’s epistemological garden. Thus, it now feels natural to epistemologists to regard p’s epistemic diaspora’s further members as mer e uses, say, of the knowledge that p (where the knowledge is a ‘thing’ constituted by one particular member of the diaspora). For instance, accurate questioning as to p would standardly be regarded by epistemologists only as reflecting, and being answ erable to, the independently characterisable and constituted knowledge that p (the latter being the augmented accurate belief that p). that p. (But we could, if we wish, distinguish between the accuracy aspect and the r est of each phenomenon. This would not aff ect my main points.) 6. Here, Descartes has much to answer for, with his emphasis upon knowledge as an inner, entirely pr esent, entity. B ut as I hav e argued else where (when r emarking on contemporar y epistemological methodology: 2006d ), Descartes need not still be constraining us so much as epistemologists.

309

But that approach, epistemologically recognizable though it is, nonetheless remains epistemologically arbitrar y. This is because that way of interpreting the members of p’s epistemic diaspora is easily r eversible, uniformly so thr oughout that diaspora. I n other wor ds, any of those members could be that of which, at a given time, the others are uses. For example, believing accurately that p, sensing accurately that p, reasoning accurately in p-related ways, and so on, could be understood as er flections or results of one ’s raising suffi cient, and sufficiently revealing, questions as to whether and ho w p is true. In this example, questioning accurately would be the central member ofp’s epistemic diaspora; believing accurately would not be. And once we realize the availability of that conceptual move, a further conceptual step immediately beckons. Specifically, we need not continue to single out any par ticular member of the diaspora as being the only one of which all the others would be mer e uses or r eflections. No single member need be assumed to be special in that r espect. Epistemology has developed detail after detail in its quest to understand ho w an accurate belief needs to be augmented if it is to be knowledge. (Towards that end, we talk at length of degr ees and kinds of evidence, of r eliability, and the like.) But epistemology could pr ofitably begin to inv estigate how other members of that diaspora may likewise be augmented, mutatis mutandis. For instance, questioning accurately in p-related ways can occur in structured ways that reliably focus on one after another aspect of the truth that p. This could be likely to generate accurate beliefs, sensings, r easonings, answers, actions, and so on. Along such lines, therefore, it seems that each member of p’s epistemic diaspora admits of a congruent kind of epistemic augmentation. Moreover, notice how readily the details of this augmentation, for any particular member of p’s epistemic diaspora, will direct our notice to the further members. For instance, augmentation of an accurate belief that p might be understood in terms of accurate questioning and answering, reasoning, acting, remembering, and so on.7 Let us treat believing accurately that p, then, as just one of sev eral equally constitutiv e aspects of kno wing that p—these aspects being the members of p’s epistemic diaspora. And let us tr eat knowing that p as, equally, any of those members. D oing so enables us, in a fi rst approximation, to say this: 7. Hetherington (2007b) gestures at how p-related inquiry, in particular, could function as the central member of (what I am now calling) p’s epistemic diaspora.

310

One’s knowing that p is one’s accurately believing that p, and/or one’s accurately questioning whether or about p, and/or one’s accurately answering as to whether or about p, and/or ones accurately r ecalling that p, and/or one’s accurately reasoning about p, and/or one’s accurately discussing p, and/or one’s accurately providing evidence as to p, and/or one’s accurately asserting that p, and/or one’s accurately acting as if p, and/or the like. In short, one’s knowing that p is any or all of these, at once or separately. 3. Knowledge-attribution Here is an initial objection to section 2’s exploratory proposal: If one accurately answers that p, without believing that p, then one is not knowing that p. At best, one is imparting knowledge one does not possess. (One can impar t the kno wledge to others without having it oneself.) In eff ect, one could be r elevantly akin to a S earlean Chinese room, furnishing corr ect answers without having any ‘inner sense ’ of being committed to them. Patently, that argument begs the question at hand, of whether kno wing is only ev er a kind of believing. S till, it may dir ect our attention to something worth noting, which is that, even when lacking a belief that p, one could aptly hav e knowledge that p attributed to one, such as b y the person to whom one’s answer that p has been directed. Such an attribution can be apt even when one lacks a belief that p, so long as one is answ ering accurately that p (while manifesting no signs of disbelieving that p).8 Insofar as a person can be instilling in others the kno wledge that p, such knowledge may in turn be attributed to her.9 8. Nevertheless, it is not required, in general, that the knowledge-attributor know in advance that, because one is answering that p, one is answering accurately. Knowledge could be attributed aptly to one, on the basis of one’s inquiring aptly—with there being only a consequent inference to the belief that one’s conclusion is true, and thereby an inference to its being tr ue that p. On this methodological obser vation about how to use inquir y epistemically, see Hookway (2007) and Streeter (2006, 303–4). 9. Why do we ever talk of knowledge at all? Could we have begun doing so, just to register people’s usefulness to others as informants? S ee Craig (1990) on this idea. Our Cartesian epistemological ancestry makes us think of knowing as, above all else, an inner state whose presence is at least fallibly , maybe ev en infallibly, knowable from a fi rst-person perspective if fr om any

311

Yet must this attribution then be inaccurate, simply because that person lacks a belief thatp? Unless we are independently persuaded that knowledge is only ever a kind of belief, we need not assume so. In which event, if the attribution can be accurate anyway, the presence of a belief is not necessary to the presence of knowledge. Although believing can be part of knowing, it need not be. (‘ What else could it ev er be?’ S ection 5 will answ er this question.) And this result coheres with section 2’s approach. 4. Conceptual priority What else might hinder epistemological acceptance of section 2’s general sort of pr oposal? Current epistemology’s recurring analytical sear ch for instances of conceptual priority might do so. We are familiar with the usual notion of epistemic priority . One kind of knowledge (such as of physical objects) is pr esumed to be attainable only once some independently describable kind of knowledge (such as of sensory experiences) is in place. The former knowledge is said to depend for its existence upon the latter kno wledge, which is allo wed to be able to exist (if at all) independently of the former . A non-sceptic about the dependent form of knowledge will accept that the other sort of knowledge is attainable by itself, in advance of the former kind; a sceptic about the dependent kind of knowledge will not share that acceptance. And the standard assumption (mentioned in section 1) of knowledge’s being a kind of belief—more generally, a kind of state—is another instance of ‘priority thinking’ within analytic epistemology. A single member (such as accurately believing) from within p’s epistemic diaspora is being allowed to be the kno wing that p—with the other members consigned to being mere expr essions, manifestations, or embellishments of that privileged member. Always, the privileged member is pr esent if ther e is kno wledge that p; never need any of the other members be pr esent, at least not for that purpose. Thus, knowledge that p can be constituted, in one particular way, independently of the presence of the other members of p’s epistemic diaspora; while none of those other members can even express knowledge that p unless such kno wledge is alr eady constituted independently , in itself, via the privileged member’s presence. In this sense, the belief thatp’s at all. But must that Cartesian presumption be correct? Not according to this paper. (I do not assume that one ’s knowing that p need be accompanied b y one’s either kno wing or believing that one is knowing that p, for instance.)

312

constituting knowledge that p is conceptually prior to any of the diaspora’s other members even expressing knowledge that p. Yet that form of analysis does not do justice to the constitutiv e reciprocity inherent in the situation (a phenomenon noticed in section 2). The augmented accurate belief that p would not even be knowledge that p if it could not give rise to, or be expressed by, other members of p’s epistemic diaspora. This constitutive dependence exists, even where there has standardly been presumed to be a constitutive independence. Epistemologists have, it seems, o verlooked this when seeking to understand the natur e of knowing. They have focused on the augmented belief b y itself, asking whether it is kno wledge on its o wn—and accepting that it is, once it is augmented in standard ways. Then sometimes, as an inter esting further, but extrinsic, philosophical exercise, epistemologists will ask whether the knowledge that p can generate questions, explanations, assertions, and/or other members of p’s epistemic diaspora. That methodological predilection is misleading, though. It is only because we take for granted that the knowing that p can generate these fur ther members of p’s epistemic diaspora that we may discern—in the pr esence of the augmented accurate belief that p—the presence of the knowledge that p in the fi rst place. My point is that even the augmented true belief is not already knowledge, purely in itself, independently of the actual or possible existence of the other diaspora members. I n part, it is knowledge only because it can guide accurate asser tion, generate accurate questions, gr ound accurate answers, give rise to accurate explanations, and the like. When a belief is knowledge, in par t this is due to those links. The belief would not be knowledge, ev en when tr ue and w ell justifi ed, if the accurate p-related questions, answers, assertions, representations, explanations, and so on, to which it can and will give rise were not themselves equally worthy candidates for being knowledge. In such ways, there are reciprocally constitutive links between members of p’s epistemic diaspora. These pertinent links allow us to regard all of the members of p’s epistemic diaspora as equally good candidates for kno wledge that p. This is not to say that all will actually be pr esent, being manifested, whenev er knowledge that p is present. But any of them could be, even as no one of them in particular need be.

313

5. Knowledge as an ability Section 4’s conclusion reinforces section 2’s thinking, by telling us that all members of p’s epistemic disaspora are equally good candidates for being knowledge that p. By that same token, however, no one of those members is the knowledge that p—such that the others are not. With which realization, we may well wonder, ‘Then where is the knowledge that p?’ It is at once ev erywhere and no where, both within and without the knower that p. Less cr yptically: the knowledge that p is whatev er unifies both the manifested and the potential members ofp’s epistemic diaspora. It is no single member; it is what all of them, manifested or not, most clearly have in common. Of course, in aWittgensteinian mood we might wonder whether there can be any such unifying thr ead. But that mood may pass (as it should: see Suits 2005); after which, we can understand the unifying thread as being the person’s ability to generate any of those members of that diaspora. We may say that there is no core to p’s epistemic diaspora—this collection of manifested or potential expressions of knowing—other than the ability to generate them. This ability could itself expand or contract—it could strengthen or w eaken—with a person’s conceptual and cognitiv e, maybe more generally his personal, development. One could know that p today in a different way to how one did so last year, not merely by having new evidence, but by being able to perform new p-related activities from among those featured in p’s epistemic diaspora. 10 So, by being the ability unifying the members ofp’s epistemic diaspora, the knowing that p is not some particular member of that diaspora, over and above the other members. 11 To express knowledge that p is not to express just a single member of that diaspora (with that member alr eady being the only kind of member that could ev er be kno wledge that p), by manifesting another member. To express or manifest kno wledge is to express or manifest an ability—the ability that most typically can generate 10. Correlatively, one could also know more strongly now that p than one did last year. One’s knowledge that p could be epistemically improved. In short, once knowing that p is seen to be an ability, we may also discern disparate degrees or epistemic qualities of knowledge that p. For more on this idea (but without the explanatorily underpinning conception of knowledge as an ability), see Hetherington (2001; 2005). S osa has also endorsed the idea of ther e being diff erent possible degr ees or qualities of kno wledge, even for a par ticular p. See, for example, S osa (1997; 2003; 2007, 113). 11. ‘H ow strong must the ability be, so as to constitute kno wing?’ This is the question of how much of a skill the ability must be. But in whatever form is appropriate for a given theory of knowledge, this question arises for each theory of knowledge.

314

the expression or manifestation of any given member of that diaspora. One does this by manifesting some one or more of those members. Here is a simple example. O ften, one does not kno w whether one believes that p, until after one has answered accurately that p to a question. ‘Is q true?’ ‘No … [reflecting] … it’s p, not q, that’s true.’ Only then do you realize, ‘I do believe that p. Now I see this clearly.’ And, admittedly, that could be because, all along, you already believed that p. Then again, the phenomenology of such occasions suggests that it need not be. Rather, it could be that in fact y ou lacked the belief until after answ ering the question. Is there knowledge only then, only once the belief appears? Not on my proposal. The knowledge was already present, in part as the ability to provide that answer (that correct answer, as we may take it to hav e been)—and thereby to attain the commitment to it, to which the belief amounts. (And if, as it happens, the belief is always present as part of the knowing, this does not entail that, even once augmented, it constitutes the knowing. The ability to believe—and/or assert and/or question and/or … —could be the kno wing, even as a par ticular case of believing expr esses that knowing on a particular occasion.) I end this section with a comment on kno wing’s boundaries. Exactly where to locate v arious aspects of the boundar y between knowing that p and not kno wing that p is a challenge for epistemologists in general. 12 Still, it is less of a conceptual difficulty-of-principle for my form of theory than for others. This is because, in general, the possession of an ability is not even expected to be an especially precise matter. It is normal for borderline cases to abound, inescapably so, conceptually so . Also in general, though, we do not regard this as preventing us from accurately discerning someone’s possessing, or indeed lacking, a particular ability. And, again, in making such an assessment we will not be expecting to be able to know, for every possible circumstance, whether it is an instance of that ability ’s being manifested. This should now be our attitude, also, towards knowledge-possession. Insofar as knowing is an ability, it is not a kind of thing which, in general, needs to have sharp boundaries in order to be present.

12. For more on the nature of the problem, along with a fallibilist response to it, see Hetherington (2006c).

315

6. Rylean know-how We may put section 5’s main point in Rylean terms (see Ryle 1949, ch. II; 1971). For we may conceive of that section’s proposal—an ability analysis of knowing—as a r eduction of kno wledge-that to kno wledge-how. Ryle denied that knowing-how could be reduced to knowing-that, in the sense of some prior and guiding kno wing-that being r equired as par t of any manifestation of knowledge-how. Acting intelligently—acting in a knowing-how way—does not need to be intellectualist—an application of some knowledge-that.13 Nonetheless, a contrar y reduction is suggested b y this paper ’s ability analysis of knowing that p. On this ability analysis (which I sometimes call a practicalism about the nature of knowledge), knowing that p—the philosophically traditional representation of knowledge-that—is interpreted as a kind of kno wledge-how. You know that p insofar as y ou know how to generate and answer accurate questions as to p, and/or to form accurate beliefs as to p, and/or to assert accurately that p, and the like. The knowing-that would be this multi-faceted kno wledge-how. Knowledge would be knowing: to have knowledge that p is to know that p is to be knowing that p. And this need not be one’s being in a state of accurately believing that p, or one’s formulating and/or answering accurately a question about p, etc. None of these in particular is the knowing as such. Each is simply one available manifestation of that complex ability which is your knowing that p. In Rylean terms, each stands to the knowing as an intelligent action of performing an action A stands to the ability to do A. Of course, those possible kinds of state—those potential manifestations of the ability which is the knowing that p (for some specific ‘p’)—are interlinked. To manifest one of them is to be mor e likely, other things being equal, to manifest some or all of the others. But underlying all such links is the unifying p-related skill which is the knowing that p. If this skill is a state, it is a complex dispositional one. It should be obser ved that Snowdon (2003, 8–11) argues that having an ability to do A is neither necessar y nor suffi cient for knowing how to do A. Thus (in one of his representative examples), a great chef still knows how to cook omelettes, ev en after losing his arms and ther eby no longer having the ability to cook them. But this sort of example (no matter how 13. Here, without argument, I accept this Rylean conclusion. Hetherington (2006a, sec. 2) defends Ryle’s reasoning against Stanley and Williamson (2001), as does Noë (2005).

316

effective against a general identifi cation of kno wing-how-to-do-A with an ability-to-do-A) leaves my particular analysis unharmed. In Snowdon’s cases, the knowledge-how is purely cognitive (involving only the person’s brain, say), while the ability is at least par tly non-cognitive (such as b y involving also the use of his arms). And most of the abilities I mention, as belonging to p’s epistemic diaspora, are purely cognitive in that sense. So, the sort of disconnection at the heart of Snowdon’s cases does not arise in applications of my comparatively specific analysis. 7. Knowing philosophically Section 5’s analysis of knowing posits an ability which is complex (as section 6 notes) by being disjunctive. Many possible expressions of knowing are conceived of as belonging to p’s epistemic diaspora. And each of the respective abilities to generate such expressions can be part of the overall ability which is y our knowing that p. No single one of them needs ev er to be manifested, ho wever, if y ou are to manifest y our knowing that p, let alone if y ou are to kno w that p. (Accordingly, there is scope for different people, or ev en for a single person at distinct times, to kno w that p by having varying p-related sub-abilities, and/or by possessing these to different strengths, and/or b y having diff erent numbers of them. Again, this av ailable str uctural complexity within a par ticular ability which is an instance of kno wing that p allows for the possibility of ther e being distinct degrees or str engths of kno wledge that p, even for a specifi c p.) In general, for example, even accurate questioning and answering are not required if kno wledge that p is to be pr esent (either as expr essed or as unexpressed).14 Nonetheless, this does not entail that there are no specific kinds of knowing for which some particular member of the associated epistemic diaspora is constitutively essential. And indeed there is such a pairing. It links (i) a person’s knowing philosophically that p, with (ii) her ability to raise accurate p-related questions—ones that, if answered correctly, will direct us to p (the main tr uth in question) as w ell as to r elated truths. These related truths could be about aspects of the truth that p. (See Hetherington 2005 on how this str uctural dimension can help us to understand kno wledge that p. That dimension r eveals especially w ell the gradational dimension 14. At any rate, this is a standard view. For a dissenting argument, see Schaff er (2007).

317

of knowing that p. For instance, you would know that you have a hand, in part by knowing more or fewer aspects of what it is for you to have a hand.) I cannot say with cer tainty, of course, that ther e is any philosophical knowledge. I have argued elsewhere (1992) against there being any epistemological knowledge, for a start. And although I am unsure as to whether I was right about that, the question r emains a substantive one. The confidence with which many philosophers asser t their fav oured philosophical positions (and the scorn with which some dismiss competing vie ws) could well suggest their regarding those favoured philosophical positions as constituting kno wledge. And many of us wish to gain philosophical knowledge, not merely philosophical belief, say. What would philosophical knowledge be like, though? I have one suggestion; which is that philosophical kno wledge would be a questioning kind of knowing. Philosophical practice is in part a matter of knowing how to uncover lingering philosophical questions within what would other wise seem to hav e been settled on non-philosophical grounds. One person confi dently says that there is a physical world—an ‘everyday’ claim par excellence. But someone else may ask, philosophically, what that claim means, what makes it true, and how we can know it to be true: ‘What is it to be physical? And what is a world? H ow does a world ever make thoughts true? Moreover, what evidence do we use in forming beliefs about physical aspects of a world? How good is such evidence, too? And so it goes: questions, more questions, all bearing upon a particular p. Without them, there is no philosophy ofp (so to speak) at that moment in that place. Wherever they disappear, so does any philosophy ofp. (Correlatively, whenever a putative philosopher ceases being genuinely questioning about p, he is no longer being philosophical about p. Appeals to intuition or plausibility run that risk. An odd question might be raised, sparking a new and puzzling proposal. ‘No, that’s implausible,’ comes the response; which might well halt that questioning as top. Has the questioning ceased in a philosophical way? Seemingly, no. Of course, the dismisser can proceed to return to the line of philosophical inquir y in which he was pr eviously engaged; and it might well be philosophical, either about some q distinct from p, or still about p. But this does not entail that the dismissal as such was also philosophical in its reflection on p.) So, questioning is vital to the deriv ation of any philosophical claims that at least might be philosophical kno wledge. And section 5’s form of ability analysis accommodates a congr uent conception of such kno wl318

edge. Thus, philosophical knowing would be a kind of kno wing-how. It would be a skill. Philosophy itself is nothing if not skilful. It is not merely a listing of tr uths, even ones appearing as conclusions of arguments. I n a given person’s hands, it is her kno wing how to dev elop, to r efine, and to ev aluate such arguments—r evisiting and explaining and r ephrasing, imagining possible objections, understanding actual ones, modifying or defending premises and/or conclusions, and so on. Without all of this, there is not philosophy; there is only a possibility of philosophy. But with all of that, there is philosophy because ther e is philosophical know-how. And an essential par t of such kno w-how is one ’s kno wing ho w to ask the right questions. One would know not only ho w to ask the right ones with which to begin an inquir y; one would kno w how to pose the right ones for extending an inquir y. And this is tr ue at each stage of a philosophical quest. At each stage, the ability to think of accurate questions is required.15 One must kno w how to question w ell, then to answ er well; to question again and to answ er again, still well; insightfully to question further and answer further; and so forth. That is a necessary component in philosophical knowledge-how —and thereby in philosophical knowledge at all. 16 Consequently, insofar as (fr om section 5) kno wledge-that is knowledge-how, we may link these thoughts about philosophical practice with a conception of manifested philosophical knowledge—manifested philosophical knowledge-that. An assertion, for instance, is a manifestation of philosophical knowledge only if it is a manifestation of an apt philosophical ability. And it satisfi es the latter necessar y condition only b y being par t of one’s developing, refining, or evaluating a philosophical question. This is so, even if it is a philosophical answ er to a philosophical question. For the philosophical answer would itself need to be philosophical knowledge; and so, in turn, it needs to be developing, refining, or evaluating a philosophical question. By being a philosophical asser tion, it needs to be par t 15. A necessary element in the doctoral-equivalent exams undergone by Tibetan Buddhist monks is the examinee ’s asking such questions of the examiners. O n the present Dalai Lama’s experience of this, see Goleman (2003, 29–30). 16. It might be objected that ther e is no philosophical kno wledge that p even if ther e is philosophical know-how. Frances (2005) could be vie wed as pr oviding an argument for ther e being no philosophical knowledge that p, or at least much less than is often presumed to exist. And Armstrong (2006, 160) denies that there is any philosophical knowledge, or at least much of it (1999, 79), or at least any that can be ‘rationally claimed ’ (2006, 164). I am describing what philosophical kno wledge that p is like, in par t, if there is any. (Still, section 8 will bear more upon whether there actually is any philosophical knowledge.)

319

of one’s generating—dev eloping, refining, or ev aluating—some fur ther philosophical question. Repeatedly, we are told that philosophy begins with wonder. It also lives on, whenever it does, in that state. And to wonder is to question. Having begun with questions, philosophy must continue like that. B ut this implies that to answer the final philosophical question philosophically is impossible. To answer it non-philosophically (such as scientifically) is possible, at least in philosophical principle. However, philosophical knowledge cannot be a fi nal answer. Philosophical knowledge is, in par t, the ability at least to pose, even when it is also the ability to answ er, philosophical questions —because it is, in part, the ability still to find philosophical questions even in philosophical answers. So, philosophical knowledge that p is, in par t, the ability to pose philosophical questions bearing upon the truth that p. And philosophical kno wledge that p at a par ticular time is, in par t, the ability still to pose such philosophical questions. P hilosophical knowing is thus unfinished knowing, so to speak. 8. Final philosophical knowledge That is a point about all philosophical knowledge. Now, some philosophical questions ar e often thought b y philosophers to be unansw erable in themselves—in principle incapable of ev er admitting of a fi nal answer. (Such questions include the following ones. Is scepticism about our knowledge of an external world tr ue? Do people really have free will? What is the meaning of life?) B ut my suggestion is that no questions admit of a finally definitive philosophical answer.17 Does this suggestion imply that there is no philosophical knowledge? Such knowledge would at least be a special kind of knowledge. Section 7 has indicated one constraint upon it; which is (and ob viously I cannot claim to propose this with philosophical fi nality) that any genuinely philosophical knowledge would be knowledge which essentially includes the ability both to provide some sort of answer and to continue questioning. I do not know philosophically that there is an external world, if I do not still have philosophical questions about there being an external world —what it is, what it is not. I f I no longer hav e any such questions, then at best I know non-philosophically that there is an external world. 17. See Rescher (2006) for a recent exploration of this sort of idea.

320

Of course, that need not be a failing. O n the contrar y; my pr oposal might suggest a r espect in which philosophical kno wledge, if it is ev en possible, would be inferior to at least some non-philosophical knowledge. For we are hereby warned never to expect philosophical kno wledge that p to be the fi nal word on p. If our kno wledge that p is philosophical, it is still inherently questioning. In this sense, therefore, it is dubitable and uncertain—unfinished. Even when based on a sound argument, if it is philosophical then it remains, in part, an exercise in questioning. Even as an answer to philosophical questions, it is philosophical kno wledge only if it is raising further philosophical questions. The associated knower might well be aware of these further questions; in which case (and with all else being equal), his state at that time is itself philosophical in relation to the p in question. But if he is unaware of raising extra philosophical questions, then he is not noticing the philosophical potential within his answer. And in that event his answer—in his hands, at any rate—has ceased being philosophical. Although others could adopt and use it philosophically, he is not doing so. 9. Interim philosophical knowledge Section 8 suggests that no philosophical answ er is ev er defi nitively final and knowledge. But what of non-final philosophical knowledge—interim philosophical knowledge (as we may term it)? Is it possible? Or does the fact (if it is one) that philosophical essays or books, for instance, can nev er end b y attaining defi nitive or unquestionable philosophical kno wledge imply that at no stage within their pages are they ever finding philosophical knowledge? Here, section 5’s ability analysis assists us. A philosophical essay or book —in each case, focused upon arguing philosophically thatp, say—may well end when more ideas in support of p could still have been written by the author, and indeed would have been written if not for manifest pragmatic constraints. More could still have been written, too, in the sense that the ability can be present even when no longer being manifested. And aspects of the essay or book can express that knowledge, with the knowledge itself being the ability to produce that or another such expression of it—including an extension of that expression, by writing even more. This is like wise tr ue of a philosophical component within a life as a whole. A person may have been devoting time to arguing philosophically 321

for p. When this period ends, the subsequent life may be non-philosophical as r egards p. No matter; the person might r etain the ability to raise philosophical questions about p, ev en if she pr oceeds not to act in that way . (It would be as if she has gone to sleep , not still thinking actively and philosophically about p, while nonetheless retaining the ability, once she awakes, to r esume such thinking.) A ccordingly, the fi nal moment prior to the ending of a book, essay , or component of a life might have been philosophical about p—even if it cannot have produced definitive and final philosophical kno wledge that p. It can include the person’s expressing some philosophical kno wledge that p, regardless of there being no actual further philosophical questioning by her that bears upon p. And this condition is not trivial. I t will be failed if , for example, the person’s philosophically inquiring spirit has shut down—if closed-mindedness has taken over within her—as to whether p is true. For then, realistically, the relevant ability will hav e departed. Someone who is offi cially a philosopher need not continue being philosophical about p, even as she continues to assert that p (and even if she takes herself to be doing so in her capacity as a philosopher). I f she merely asserts (in the sense that she is looking no more to fi nd further philosophical questions bearing upon whether p is true), then no knowledge that p with which she might walk away is philosophical, at least for her . If her p-related know-how is not still, in par t, an ability to question accurately about p, then ev en if she does know that p, she does not do this philosophically.18

REFERENCES Armstrong, D. M. 1973. Belief, Truth and Kno wledge. Cambridge: Cambridge University Press. — 1999. ‘A Naturalist Program: Epistemology and O ntology’, Proceedings and Addresses of the American Philosophical Association 73, 77–89. — 2006. ‘The Scope and Limits of H uman Knowledge’, Australasian Journal of Philosophy 84, 159–66. Craig, E. 1990. Knowledge and the State of Nature: An Essay in Conceptual Synthesis. Oxford: Clarendon Press. 18. A version of this paper was presented at a UNSW Symposium on Linguistics and Philosophy of Language. I appreciated the audience’s probing questions.

322

Frances, B. 2005. Scepticism Comes Alive. Oxford: Clarendon Press. Goldman, A. 1999. Knowledge in a Social World. Oxford: Clarendon Press. Goleman, D. 2003. Destructive Emotions and H ow We Can Overcome Them: A Dialogue With the Dalai Lama. London: Bloomsbury. Hetherington, S. 1992. Epistemology’s Paradox: Is a Theory of Knowledge Possible? Savage, MD: Rowman & Littlefield. — 2001. Good Knowledge, Bad Knowledge: On Two Dogmas of Epistemology. Oxford: Clarendon Press. — 2005. ‘Knowing (How It is) That P: D egrees and Q ualities of Kno wledge’, Veritas 50, 129–52. — 2006a. ‘How to Know (That Knowledge-That is Knowledge-How)’, in Epistemology Futures, (ed.) S. Hetherington. Oxford: Clarendon Press, 71–94. — 2006b. ‘Introduction: The Art of Precise Epistemology’, in Aspects of Knowing: Epistemological Essays, (ed.) S. Hetherington. Oxford: Elsevier, 1–13. — 2006c. ‘Knowledge’s Boundary Problem’, Synthese 150, 41–56. — 2006d. ‘Knowledge That Works: A Tale of Two Conceptual Models’, in Aspects of Knowing: Epistemological Essays, (ed.) S. H etherington. Oxford: Elsevier, 219–40. — 2007a. ‘Is This a World Where Knowledge Has to Include Justification?’, Philosophy and Phenomenological Research 75, 41–69. — 2007b. ‘Knowledgeable Inquiry’, in Action in Context , (ed.) A. Leist. B erlin: De Gruyter, 372–381. Hookway, C. 2007. ‘ Acting and I nquiry’, in Action in Context , (ed.). A. Leist. Berlin: De Gruyter, 351–371. Noë, A. 2005. ‘Against Intellectualism’, Analysis 65, 278–90. Rescher, N. 2006. Philosophical Dialectics: A n Essay on M etaphilosophy. Albany, NY: State University of New York Press. Ryle, G. 1949. The Concept of Mind. London: Hutchinson. — 1971. ‘Knowing How and Kno wing That’, in his Collected Papers, Vol. II. London: Hutchinson, 212–25. Sartwell, C. 1991. ‘Kno wledge is M erely True B elief ’, American P hilosophical Quarterly 28, 157-65. — 1992. ‘Why Knowledge is Merely True Belief ’, The Journal of P hilosophy 89, 167–80. Schaffer, J. 2007. ‘Knowing the Answer’, Philosophy and Phenomenological Research 75, 383–403. Snowdon, P. 2003. ‘Knowing How and Knowing That: A Distinction Reconsidered’, Proceedings of the Aristotelian Society 104, 1–29. Sosa, E. 1997. ‘H ow to R esolve the P yrrhonian Problematic: A Lesson F rom Descartes’, Philosophical Studies 85, 229–49.

323

— 2003. ‘Kno wledge, Animal and R eflective: A R eply to M ichael Williams’, Proceedings of the Aristotelian Society, Supp. Vol. 77, 113–30. — 2007. A Virtue Epistemology: Apt Belief and Reflective Knowledge, Vol. I. Oxford: Clarendon Press. Stanley, J. and Williamson, T. 2001. ‘Kno wing How’, Journal of P hilosophy 98, 411–44. Streeter, G. 2006. ‘Putting the Virtues to Work in Epistemology’, American Philosophical Quarterly 43, 299–313. Suits, B. 2005 [1978]. The Grasshopper: Games, Life and U topia. Peterborough, Ontario: Broadview Press.

324

Grazer Philosophische Studien 77 (2008), 325–339.

KNOWING THE ANSWER, UNDERSTANDING AND EPISTEMIC VALUE Duncan PRITCHARD University of Edinburgh Summary This paper principally argues for two contr oversial theses: that understanding, unlike knowledge, is distinctively valuable, and that understanding is the proper goal of inquiry.

1. One of the most central topics in contemporary epistemology concerns the issue of epistemic value; in particular, the value of knowledge. Knowledge has been the focus of much of our epistemological theorising, and this prompts the question of why . What is it about kno wledge that prompts us to regard it as distinctiv ely valuable in this way such that it is wor thy of this scrutiny? It is, I would suggest, incumbent on any epistemological theory that it can answer this question, even if the answer that is off ered adverts to some error on our part. That is, while one validatory answer to the question might proceed by offering an explanation of why knowledge is distinctively valuable, an alternative, revisionary, answer to the question might proceed by explaining why kno wledge isn’t distinctively valuable while conjoining this explanation with a further story about why we might have naturally thought it to be distinctiv ely v aluable. In what follo ws, we will begin b y exploring the pr ospects for a v alidatory response to the question. We will refer to the task of answering this question, whether in a validatory or revisionary manner, as the value problem. 2. Before we can turn to this issue, however, we need to get a better grip on what it means to say that knowledge is distinctively valuable. It certainly

means more, for example, than simply showing that knowledge is generally more valuable than mere true belief, what I have elsewhere called the primary value problem.1 After all, suppose that one argued that knowledge, unlike mere true belief, entailed that one possessed justifi cation in support of one ’s belief, and that this pr operty of belief was v aluable. O ne would thus have an explanation of why knowledge is more valuable than mere tr ue belief, but clearly no explanation at all of why kno wledge is distinctively valuable, at least giv en the fur ther thesis that ther e is mor e to knowledge than mere justified true belief. After all, if this is all there is to be said about the value of knowledge, then it prompts the question of why it is knowledge, specifically, that we focus upon in our epistemological theorising, rather than just justifi ed true belief.2 One might thus think that what is r equired is an explanation of why knowledge is more valuable than that which falls short of knowledge, what I have elsewhere termed the secondary value problem. Even a response of this sort will not ob viously suffi ce to solv e the v alue problem, however. In order to see this, suppose that one argued that kno wledge is mor e valuable, as a matter of degree, than that which falls short of knowledge. One would thus have a response to the secondary value problem. Notice, however, that this account of the value of knowledge invites a ‘continuum’ picture of epistemic value, with knowledge simply marking one point further up the continuum than that which falls short of knowledge. But this picture of epistemic value doesn’t present us with any explanation of why it is this point on the continuum that is the focus of our epistemological theorising, rather than, say, a point just befor e or just after the one that knowledge marks. In short, this picture doesn’t explain why knowledge is distinctively valuable. Elsewhere I hav e thus argued that what is r equired of a satisfactor y response to the v alue problem is a r esponse that satisfi es what I call the tertiary value problem, which is the pr oblem of sho wing that kno wledge is more valuable than that which falls shor t of kno wledge not just as a 1. The use of the word ‘generally’ here is important, since it ought not to be a constraint on an adequate answer to the v alue problem that it sho w that knowledge is always more valuable than the comparable epistemic standing, in this case tr ue belief. In Pritchard (2007e) I r efer to this as the modest reading of the v alue problem, and off er a fuller defence of this r eading. Henceforth, I shall take the modest reading of the value problem as given. 2. Of course, one might at this juncture opt to take a revisionary line on the value of knowledge, rather than a validatory line, and claim that what is really of distinctive value is just justified true belief, an epistemic standing that could easily be confused with kno wledge given that Gettier-style cases are relatively uncommon. This is just the sort of line taken by Kaplan (1985).

326

matter of degree but as a matter of kind. The idea is that as one moves up the continuum of epistemic v alue something special happens when one reaches the point that knowledge marks such that more than just a difference in the degree of value enters the picture.3 3. I suspect that many would want to r esist the rather auster e demand laid down by the ter tiary value problem, but this is not the place to explor e this issue fur ther.4 What ought to be uncontentious is that it is at least desirable to answer the value problem in such a way that one can meet the tertiary value problem. Indeed, in what follows we will simply assume that an adequate response to the secondary value problem is available—i.e., that we have an explanation of why knowledge is of more value, as a matter of degree, than that which falls shor t of kno wledge—and focus our attentions exclusively on the ter tiary value problem. As we will see, this issue has important ramifications for the ultimate topic of this paper, which is the relationship between knowing-why and understanding. 4. There is ev ery reason for thinking that ther e is no account of the v alue of knowledge available which can satisfy the ter tiary value problem. The most pr omising account in the literatur e is that off ered b y cer tain virtue epistemologists which claims that kno wledge, pr operly understood along virtue-theoretic lines, falls under a general class of entity which has non-instrumental—i.e., final—value.5 This more general entity is that of achievement. Consider an archer—let’s call him Archie—taking aim at a target and hitting that target with his arrow. Suppose now that Archie in fact has no 3. For further discussion of the primary, secondary and tertiary value problems, see Pritchard (2007c; 2007e). 4. I defend this formulation of the value problem further in Pritchard (2007c; 2007e). 5. This proposal finds its most explicit expression in the work of Greco (2002; 2007; forthcoming a; forthcoming b), though the original source for this idea is earlier work by Sosa. For Sosa’s most recent statement on this issue, see Sosa (2007, ch. 4). Interestingly, the claim is usually that knowledge, qua achievement, is intrinsically valuable, but this thesis rests on a confusion between final and intrinsic value. For more discussion of this point, see Pritchard (2007c; 2007e).

327

archery-relevant skills and simply hit the target by luck. Would this success on Archie’s part count as an achievement? Surely not, since achievements are clearly the product of skill. Interestingly, however, it does not suffice for achievement that one is both relevantly skilful and successful. For consider the following Gettier-style case. Archie has all the relevant skills and fires at the target. Moreover, he hits the target. Still, the success doesn’t count as a bona fide achievement, and the reason for this is that during the flight of the arr ow a fr eak gust of wind ble w the arr ow off -course, and then a second freak gust of wind fortuitously blew the arrow back on course again. The moral of cases like this seems to be that achievements result when an agent is successful because of their ability. It does seem to be tr ue that achiev ements are distinctiv ely v aluable. Imagine two agents, one of whom hits the target because of ability , and the other who hits the target through luck, either because he lacks the relevant skills, or else because while he has the er levant skills the relationship between his success and his ability is ‘ gettierized’ by luck inter vening in the flight of the arrow to the target. From a practical point of view, it may make no difference at all how the success was achieved. Even so, wouldn’t one prefer to be in the situation of the archer who succeeds through ability rather than through luck? More generally, wouldn’t we value success that is through ability—i.e., a bona fide achievement—differently to a success that is not through ability? To sharpen our intuitions here we can simply stipulate that the instr umental v alue of the success in this case is kept fixed, regardless of whether the success constitutes an achievement or not. Accordingly, if we grant that there is a difference in value present, then it will not be a diff erence in instr umental value, but rather a diff erence in final value. And that seems just right. For whatever additional instrumental value an achievement contributes over a corresponding success that falls short of an achievement, the fact remains that the fundamental difference in value between the two r elates to the fact that it is only achiev ements, and not successes that fall short of achievements, which are of final value; that are valuable for their own sake. 5. The impor t of this to the debate r egarding the v alue problem, and the tertiary value problem in particular, should be obvious. On the standard virtue-theoretic account of knowledge at least, knowledge is to be under328

stood as cognitive success (i.e., tr ue belief ) that is attained thr ough cognitive ability (i.e., epistemic vir tue, broadly conceived). Moreover, given Gettier-style cases in epistemology, the right kind of relationship between cognitive success and cognitiv e ability seems to be of the same sor t that one fi nds in achiev ements more generally—i.e., one ’s cognitiv e success must be because of one’s cognitive ability. If this account of the str ucture of knowledge is right, however, then knowledge is just the cognitive aspect of the more general phenomenon of achievement—i.e., knowledge just is cognitive achievement. If this thesis w ere corr ect, then it would off er an elegant r esponse to the ter tiary value problem. Why is kno wledge distinctively valuable? Because knowledge, unlike that which falls short of knowledge, is a cognitive achievement and so deser ving of fi nal value. Thus, the diff erence in value between knowledge and that which falls shor t of knowledge is not merely a matter of degr ee but of kind, just as demanded b y the ter tiary value problem. 6. Unfortunately, this account of the value of knowledge does not work. The problem does not lie with the idea that achievements are finally valuable, though there are issues to be dealt with on this score.6 Rather, the fundamental difficulty with this proposal is with the identification of knowledge with cognitive achievement. The reason for this is that there are clear-cut cases of knowledge which aren’t cases of cognitive achievement, and clearcut cases of cognitiv e achievement which ar en’t cases of kno wledge. We will take these cases in turn. Consider again the case of Archie. Suppose that Archie has all the relevant archery abilities and that he brings to bear these abilities in effecting the hitting of the target. More specifically, Gettier-style luck does not intervene between his ability and his success to undermine his achiev ement. On the account off ered above, then, this is an achiev ement on Ar chie’s part, a success that is because of ability . Suppose now, however, that Archie chose his target at random from the range of targets before him, but that, unbeknownst to him, if he had chosen any of the other targets on this 6. I discuss these issues fur ther in Pritchard (2007e). In order to simplify matters, in what follows I will take it as given that achievements are finally valuable.

329

range then he would hav e missed because of a for cefield hidden in the target that would have deflected the arrow. That is, there is luck in play in Archie’s success, in the general sense that this is a success that could v ery easily have been a failur e. Note, however, that the luck in play is not of the ‘intervening’ sort that is operative in Gettier-style cases, but is instead specifically related to the envir onment in which the success is achiev ed. What is inter esting about ‘ environmental’ luck of this sor t is that it is entirely compatible with bona fide achievements. Archie’s achievement in hitting the target because of his ability is in no way diminished because of the presence of this environmental luck. This result is significant, since knowledge is incompatible with environmental luck. In order to see this, consider the famous ‘barn façade’ case, an example that, structurally, is the epistemic analogue of the archery case just given. Our agent—let’s call him ‘Barney’—sees a barn in ideal cognitive conditions and on this basis forms the true belief that there is a barn in front of him. Moreover, nothing intervenes between Barney’s barn-detecting abilities and his cognitive success—there is, for example, no barn façade before him which is obscuring fr om vie w a genuine barn—and so w e should say that Barney is exhibiting a cognitive achievement in this case—i.e., a cognitive success that is because of cognitive ability. But all w e need no w do to cr eate a pr oblem for the ‘kno wledge as achievement’ thesis is stipulate that Barney is in fact in barn façade county , in which nearly all the barn-shaped objects are fakes which are undetectable to the naked ey e—and thus that had he looked at any other barnshaped object in the ar ea then he would hav e formed a false belief . The presence of this environmental luck does not undermine B arney’s cognitive achievement any more than it does in the case of Archie that we just looked at. Nevertheless, luck of this sort, such that Barney has an unsafe belief—i.e., a belief that could hav e easily been wr ong—is incompatible with knowledge.7 There is thus sometimes more to knowledge than a mere cognitive achievement.8 Consider now a second case. I magine our agent—let ’s call her J ennifer—steps off the train in an unfamiliar town and asks the first person she meets for directions.9 Suppose that this informant has first-hand knowledge of the area and straightforwardly communicates this to Jennifer. Wouldn’t 7. For more on safety, see Sosa (2000) and Pritchard (2002; 2005, ch. 6; 2007a; 2007b). 8. G reco (2007; forthcoming a; forthcoming b) offers a response to objections of this sort. For further discussion of this response, see Pritchard (2008a); cf. Pritchard (2007d; 2007e). 9. This example is adapted from one offered by Lackey (2007).

330

we say that Jennifer could gain kno wledge of the right way to go in this way? I think w e would. M oreover, if testimonial kno wledge cannot be gained in this manner then it seems that w e know an awful lot less than we might antecedently think we know. What is significant about such cases for our purposes, however, is that we wouldn’t treat Jennifer’s cognitive success as being because ofher cognitive abilities. If anything, we would treat it as being because of her informant’s cognitive abilities (or at least because of the combined cognitiv e abilities of Jennifer-and-her-informant), but certainly not just her cognitive abilities. Thus, her cognitiv e success does not constitute a cognitiv e achievement on her part, even though it is an instance of knowledge. Sometimes, then, there is a lot less to knowledge than a cognitive achievement as well.10 7. So given that this is the only plausible account available of the distinctive value of kno wledge—i.e., an account which can meet the ter tiary value problem—then the failure of this account means that ther e is no validatory response to the v alue problem available. Nevertheless, this does not mean that there is no adequate revisionary response available. This is the line that I take. Since paradigm cases of knowledge are almost exclusively cases of cognitive achievements, it is no surprise that we might think that knowledge is distinctively valuable even though it isn’t. I also have a further claim up my sleeve in this respect. Like Jonathan Kvanvig (2003), I hold that understanding is distinctively valuable. Unlike Kvanvig, however, I have an explanation of why this is the case. nI order to sharpen our discussion, let us focus on understanding of a specific proposition (rather than, say , a subject ar ea), such as—to borr ow an example from Brit Brogaard (2007)—understanding of why one ’s house burned down. According to Kvanvig, understanding, unlike knowledge, is entirely compatible with epistemic luck, in the sense that ther e is no difficulty in supposing that an agent might hav e understanding of this fact and y et his belief be unsafe. This claim is, however, false, and recognising this has important implications for our understanding of the er lationship between knowledge-why and understanding. 10. Notice that the claim being made here is only that Jennifer’s cognitive success does not constitute a cognitive achievement. In particular, the claim is not that it is of no credit at all to Jennifer that she has a true belief, which is the conclusion drawn by Lackey (2007).

331

8. A natural explanation of why kno wledge is distinctively valuable adverts to the role that knowledge plays in inquiry. After all, once could plausibly construe all knowledge as the answer to a question, at least if one construes what constitutes the target question in a br oad fashion. 11 At best, such an account of the value of knowledge only seems to be able to supply us with an answ er to the secondar y value problem, since ther e appears no inherent reason why simply in vir tue of being the successful pr oduct of inquiry knowledge should have specifically final value. As we will now see, however, the matter is more complex than it fi rst appears. When w e talk about the pr oduct of inquir y as being kno wledge of the answer to a question I take it that what we are interested in—at least typically at any rate—is kno wledge of why such-and-such is the case. For example, suppose I under take an inquir y to fi nd out why my house burned down. In such a case we might regard the question in play as being ‘Why did my house burn down?’ and the knowledge that is the result of such a (successful) inquiry as being knowledge of the answer to this question—viz., knowledge of why my house burned down. The standard way of accounting for knows-why is in terms of pr opositional knowledge. So I know why my house burned down because I know that it burned down because of (say) faulty wiring. 12 One might think that understanding—at least of the sor t that is our focus here—is nothing more than just knowing why. That is, to understand why my house burned down is just to know why my house burned down, and to know why my house burned do wn is just to know that it burned down because of (say) faulty wiring. 13 If that’s right, and many think so, then understanding collapses into pr opositional knowledge anyway, and thus insofar as knowledge is not distinctively valuable, then (pending some further argument at least) understanding is not distinctively valuable either. Importantly, however, as I will now argue, understanding does not reduce 11. S ee Schaffer (2005) for an excellent defence of this thesis. 12. The locus classicus for the reductionist view is Hintikka (1975). See Schaffer (forthcoming) for a modification of the reductionist position, albeit one that is very much in the spirit of standard reductionist views. 13. Consider the follo wing r emark made b y Lipton (2004, 30) and quoted in G rimm (2006, 1), for example: “Understanding is not some sort of super-knowledge, but simply more knowledge: knowledge of causes ”. The natural way to r ead this passage is as suggesting that understanding why one’s house burned down is just knowing why it burned down—i.e., knowing that it burned down because of (say) faulty wiring.

332

to propositional knowledge, and thus the way is open to argue that ther e is a fundamental epistemic standing that is distinctively valuable. Moreover, insofar as we can construe the product of inquiry as specifically providing us with understanding, rather than simply pr opositional knowledge (or something reducible to pr opositional knowledge, like kno wledge-why), then the possibility emerges that inquir y is the r oute to a distinctiv ely valuable epistemic standing. 14 9. There is a v ery simple r eason why ther e is mor e to understanding than mere knowing-why. In order to see this, we first need to think a little more about what is involved in understanding. To begin with, notice that understanding, like kno wledge, is factiv e. This point might not seem initially obvious, and the reason for this is that there are species of understanding which ar e not factiv e, at least in any straightforward sense. I might be said to understand quantum physics, for example, and yet this seems entirely consistent with the possibility that I have some false beliefs in this r egard (though note that y ou’d better not have many false beliefs, and the false beliefs you have had better not be of facts which are central to the subject matter in hand). R emember, however, that the type of understanding that we are concerned with is more localised, and concerns a specifi c proposition. Of this sor t of understanding ther e can be no doubt that understanding is factiv e. In order to illustrate this, suppose that I believe that my house has burned do wn because of an act of vandalism, when it was in fact caused by faulty wiring. Do I understand why my house burned down? Clearly not. 15 So far, then, understanding and kno wledge hav e similar pr operties. Another property they share is an incompatibility with Gettier-style epistemic luck. I magine, for example, that I arriv e home to see my house burned to the ground and a bunch of people outside my house dressed up 14. It is worth noting the order of explanation here. Whereas one might argue that knowledge is valuable because it is the pr oduct of inquir y, the thought in play her e would rather be that inquiry is valuable because it results in something distinctively valuable—i.e., understanding. 15. I further defend the claim that understanding is factiv e in Pritchard (2008b). See also Kvanvig (2003) and Grimm (2006). For a defence of the non-factive conception of understanding (bearing in mind that different things are sometimes meant by this term), see Zagzebski (1996), Riggs (forthcoming) and Elgin (forthcoming).

333

as fire officers. Suppose I ask one of them what happened, and I am told that the reason why my house burned down was that I had faulty wiring. Suppose further, however, that the person that I am speaking to is not in fact a fi re officer at all. Indeed, of all the people outside my house, none of them is a fi re officer, the fire brigade having long since left the scene. Instead, they are simply a gr oup on their way to a fancy dr ess party and are merely taking the oppor tunity to pretend to be genuine fi re officers. Luckily, however, the entir ely invented answer that my informant giv es me is true. Can I know why my house burnt down on this basis? Clearly not, since, for one thing, I manifestly don ’t know that my house burned down because of faulty wiring, giv en the G ettierization involved in my gaining a tr ue belief. But do I understand, nonetheless, why my house has burned down? Again, surely not. How can one gain an understanding of why one ’s house burned do wn by (unbeknownst to one) speaking to someone who knows nothing about what caused the fi re? I noted earlier that K vanvig holds that understanding is compatible with epistemic luck, but one might no w wonder why this is the case, given that it is so clear that it isn’t compatible with Gettier-style epistemic luck. The reason for this is that the r elationship between understanding and epistemic luck is not straightforward since, while understanding isn’t compatible with G ettier-style epistemic luck, it is compatible with the sort of envir onmental luck that w e discussed earlier . For consider no w the following scenario. Suppose I arrive home as before to find my house burned down, but that this time the person I get the information fr om regarding the r eason why this happened is genuinely a fi re officer who knows what she is talking about. S uppose, however, that when I arriv ed back at my house there was a group of people outside who all seemed to be fire officers, and that I chose one of these people at random to speak to about the cause of the fire. Imagine, though, that nearly all of the people outside my house ar e merely dressed as fi re officers on their way to the fancy dress party noted earlier, and that I just happened to choose the one genuine fire officer among them. Furthermore, had I spoken to one of the fake fi re officers then I would hav e been told a false explanation of why my house had burned down, but be none the wiser. The structure of this case should be familiar to us, since it is str ucturally analogous to the barn façade example in that one has a tr ue belief that is lucky in the sense that it is unsafe but wher e the epistemic luck involved is not of the intervening, Gettier-style variety. What is interesting about this case, ho wever, is that although one ’s knowledge of why one ’s 334

house burned down—more specifically, one’s knowledge that one’s house burned down because of faulty wiring—is undermined b y the pr esence of this luck, one’s understanding of why one’s house burned down is not undermined. That is, one can gain a genuine understanding of why one’s house burned down via the testimony of a genuine fire officer, even if there is environmental luck in play . Understanding, then, is compatible with one type of epistemic luck— viz., environmental epistemic luck—while being incompatible with another type of epistemic luck— viz., intervening, Gettier-style luck.16 10. So understanding, then, unlike kno wledge, is compatible with envir onmental epistemic luck. There is a good reason for this, in that understanding, again unlike kno wledge, is in its natur e a cognitiv e achiev ement. The necessar y mar ks of a cognitiv e achiev ement are all ther e. O ne can gain understanding and y et, like cognitive achievements more generally, not have knowledge of the target pr oposition because of the pr esence of environmental luck. Furthermore, just as one can gain knowledge without exhibiting a cognitive achievement—as in the testimonial case described above—so one can have knowledge of the target proposition and yet lack understanding. In order to see this, just consider the follo wing case. S uppose I gain an understanding of why my house has burned do wn by speaking to a genuine fi re officer. Suppose further, that I no w tell my y oung son why our house burned down. My son, unlike me, has no idea at all how faulty wiring might cause a fi re. Nevertheless, he can surely come to know why his house burned do wn by being told the r eason from someone who he knows is a reliable informant, such as his father. Manifestly, however, he does not understand why his house burned do wn, since understanding in this context clearly would require some conception of how the cause is meant to bring about the eff ect. This is y et another r eason why understanding should not be thought to be reducible to knowing-why. But it is also another reason for thinking that understanding is a cognitiv e achievement. For it reinforces the idea 16. By failing to recognise this distinction between environmental and intervening epistemic luck, Grimm (2006) ends up arguing that understanding is entirely incompatible with epistemic luck, and thus that it is simply a species of kno wledge.

335

that understanding essentially consists of cognitive success—understanding is factive, remember—that is because of cognitiv e ability. For notice that while the child might acquir e knowledge of why his house burned down from my testimony, understanding requires much more than this, which is a genuine conception of how the cause and the effect are related, something which, it turns out, knowledge-why does not demand. Moreover, unlike knowledge, understanding, like cognitive achievements more generally, is consistent with environmental epistemic luck—you don’t lose your understanding, or any other cognitiv e achievement for that matter, just because of the presence of environmental epistemic luck. If understanding is a cognitive achievement, then it is thereby of distinctive value. As we will now see, this point is important for not only how we think of epistemic value more generally, but also for how we think about the value of inquiry. 11. I noted earlier that the failure of the ‘knowledge as achievement’ thesis to answer the tertiary value problem naturally prompts a revisionary response to the value problem. Knowledge is not distinctively valuable after all, but merely seems to be since it is paradigmatically distinctiv ely valuable. We are now in a position to enhance this revisionary story, for we can further argue that there is an epistemic standing, distinct from knowing but closely related to knowing, which is distinctively valuable, and that is understanding. Given the close relationship between knowledge and understanding, the fact that understanding is distinctiv ely valuable can help to explain why we might naturally think that knowledge is distinctively valuable, even though closer inspection of this claim reveals that this is false. Moreover, I think w e can also cast light on why inquir y might be thought to be par ticularly r elevant to the pr oblem of epistemic v alue. As we have seen, if we regard the product of successful inquiry as simply knowledge, then it isn’t clear how inquiry can contribute to this debate. Given the contrast betw een understanding and kno wledge that has just been defended, however, the possibility opens up of arguing that the pr oduct of successful inquiry is not knowledge at all, but rather understanding. Indeed, I think that such a thesis has a gr eat deal of intuitive force. In order to see this, recall the case just described of my son coming to know why his house burned do wn while failing to understand why his 336

house burned down. If the product of inquiry is just knowledge, then we ought to regard my son’s coming to know why his house burned down as the product of a successful inquiry. Clearly, however, we would not regard this as the conclusion of a successful inquir y at all. I ndeed, the inquir y has stopped too soon. I think this point often gets obscured by the simple fact that when one thinks of paradigm cases of successful inquiry one imagines an agent who has both knowledge and understanding. B y focussing on cases in which only the former is possessed, however, we can come to see that the inquiry in question has not been completed successfully. Another potential reason why this point might be overlooked could be because of the existence of collaborative inquiries in which the result of the inquiry is not fully understood by all parties to the inquiry because of the different fi elds of expertise involved. Such inquiries might be thought to constitute a direct refutation of the thesis that understanding is the pr oduct of a successful inquir y. In fact, there is no tension betw een collaborative inquires and this thesis. After all, it is sur ely essential in such cases that at least some of the par ties fully understand the r esult of the inquir y if that inquiry is to be deemed successful. For example, imagine an inquiry undertaken by a Nobel Prize winner and her lab assistant which results in an important scientifi c result. For this inquir y to be successful it is vital that someone—presumably, in this case, the N obel Prize winning scientist—understands this thesis, ev en if some par ties to the inquir y do not understand it. The product of a successful inquiry, even a collaborative one, is thus still understanding, it is just that in some cases the understanding is not possessed by all parties to that inquiry. 12. I submit, then, that the right way to think about successful inquir y is as specifi cally aiming at understanding and that, since understanding is distinctively valuable, this has impor tant implications for the debate regarding epistemic value.17 17. I am grateful to F ranck Lihoreau for inviting me to write this paper . An early v ersion of this material was presented at the University of Edinburgh in 2007, and I am grateful to the audience for their feedback. Work on this topic was completed while in r eceipt of an AHR C research leave award.

337

REFERENCES Brogaard, B. 2007. ‘I Know. Therefore, I Understand’, typescript. Elgin, C. Forthcoming. ‘Is Understanding Factive?’, Epistemic Value, (eds.) A. Haddock, A. Millar & D. H. Pritchard, Oxford University Press, Oxford. Greco, J. 2002. ‘Knowledge as Credit for True Belief ’, Intellectual Virtue: Perspectives from Ethics and Epistemology, (eds.) M. D ePaul & L. Z agzebski, Oxford University Press, Oxford. — 2007. ‘The Nature of Ability and the Purpose of Knowledge’, typescript. — Forthcoming a. ‘The Value Problem’, The Value of Knowledge, (eds.) A. Haddock, A. Millar, & D. H. Pritchard, Oxford University Press, Oxford. — Forthcoming b. ‘What’s Wrong With Contextualism?’, The Philosophical Quarterly. Grimm, S. 2006. ‘Is Understanding a Species of Knowledge?’, British Journal for the Philosophy of Science 57, 1–21. Hintikka, J. 1975. ‘Different Constructions in Terms of the Basic Epistemological Verbs: A Survey of Some Problems and Proposals’, The Intensions of Intentionality and Other New Models for Modalities, 1–25. Kaplan, M. 1985. ‘It’s Not What You Know That Counts’, Journal of Philosophy 82, 350–63. Kvanvig, J. Forthcoming. ‘Responses’, The Value of Knowledge, (eds.) A. Haddock, A. Millar, & D. H. Pritchard, Oxford University Press, Oxford. Lackey, J. 2007. ‘Why We Don’t Deserve Credit for Everything We Know’, Synthese 158, 345–61. Pritchard, D. H. 2002. ‘Resurrecting the Moorean Response to the Sceptic’, International Journal of Philosophical Studies 10, 283–307. — 2005. Epistemic Luck, Oxford University Press, Oxford. — 2007a. ‘Anti-Luck Epistemology’, Synthese 158, 277–97. — 2007b. ‘Knowledge, Luck and Lotteries ’, New Waves in E pistemology, (eds.) V. F. Hendricks & D. H. Pritchard, Palgrave Macmillan, London. — 2007c. ‘Recent Work on E pistemic Value’, American Philosophical Quarterly, 44, 85–110. — 2007d. ‘Sosa On Epistemic Value’, 2nd On-Line Philosophy Conference, http:// experimentalphilosophy.typepad.com/2nd_annual_online_philoso/2007/05/ ernest_sosa.html. — 2007e. ‘The Value of Knowledge’, typescript. — 2008a. ‘Greco on Kno wledge: Virtues, Contexts, A chievements’, The Philosophical Quarterly 58, 437–47. — 2008b. ‘Knowledge, Understanding and Epistemic Value’, Epistemology (ed.) A. O’Hear, Cambridge University Press, Cambridge, UK.

338

Riggs, W. Forthcoming. ‘Getting the Meno Requirement Right’, Epistemic Value, (eds.) A. H addock, A. M illar & D. H. Pritchard, Oxford University Press, Oxford. Schaffer, J. 2005. ‘Contrastiv e Kno wledge’, Oxford Studies in E pistemology 1, 235–71. — Forthcoming. ‘Kno wing the Answ er’, Philosophy and P henomenological Research. Sosa, E. 1999. ‘H ow to D efeat Opposition to M oore’, Philosophical Perspectives 13, 141–54. — 2007. A Virtue Epistemology: Apt Belief and Reflective Knowledge, Oxford University Press, Oxford. Zagzebski, L. 1996. ‘R ecovering Understanding’, Knowledge, Truth and Duty: Essays on Epistemic Justification, Responsibility and Virtue, (ed.) M. Steup, Oxford University Press, Oxford.

339