Propositions. Semantic and Ontological Issues. (Grazer Philosophische Studien 72) (Grazer Philosophische Studien)

PROPOSITIONS

SEMANTIC AND ONTOLOGICAL ISSUES

Grazer Philosophische Studien INTERNATIONALE ZEITSCHRIFT FÜR ANALYTISCHE PHILOSOPHIE

GEGRÜNDET VON Rudolf Haller HERAUSGEGEBEN VON Johannes L. Brandl Marian David Leopold Stubenberg

VOL 72 - 2006

Amsterdam - New York, NY 2006

PROPOSITIONS

SEMANTIC AND ONTOLOGICAL ISSUES

Edited by

MASSIMILIANO CARRARA AND

ELISABETTA SACCHI

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-13: 978-90-420-2194-5 ISSN: 0165-9227 © Editions Rodopi B.V., Amsterdam - New York, NY 2006 Printed in The Netherlands

TABLE OF CONTENTS

Massimiliano CARRARA & Elisabetta SACCHI: Propositions. An Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1

Kevin MULLIGAN: Ascent, Propositions and Other Formal Objects

29

Eva PICARDI: Colouring, Multiple Propositions, and Assertoric Content . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

49

Elisabetta SACCHI: Fregean Propositions and Their Graspability . . .

73

Pierdaniele GIARETTA: Numbers, Reference and Russellian Propositions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

95

Marco SANTAMBROGIO: On the Sameness of the Thoughts. Substitutional Quantifiers, Tense, and Belief . . . . . . . . . . . . . . . . . . .

111

Andrea IACONA: True In a Sense . . . . . . . . . . . . . . . . . . . . . . . . . .

141

Marina SBISÀ: Speech Acts Without Propositions? . . . . . . . . . . . . . .

155

Alberto VOLTOLINI: How to Get a Non-Intensionalist, Propositional, Moderately Realist Truthconditional Account of Internal Metafictional Sentences . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

179

Enrico MARTINO: Fictional Propositions and the Unprovability of Consistency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

201

Vittorio MORATO: Propositions and Necessary Existence . . . . . . . .

211

Ernesto NAPOLI: Negation . . . . . . . . . . . . . . . . . . . . . . . . . . . .

233

This page intentionally left blank

Grazer Philosophische Studien 72 (2006), 1–27.

PROPOSITIONS AN INTRODUCTION Massimiliano CARRARA and Elisabetta SACCHI University of Padua (Italy) 0. The notion of proposition is pivotal in many areas of philosophy and, in particular, in the philosophy of language for the close connection that links it to other notions such as reference, meaning, truth and understanding.1 Even though the debate on propositions and the claim that there are things like them can be traced back to the time of Plato, it is from such figures as Bolzano, Frege, Moore and the early Russell that the current debate originates. According to the traditional picture a proposition is: A. Expressed in a given context by an utterance of a declarative sentence. Propositions, according to (A) are the objects of the acts of assertion. B. Primarily true or false. Propositions, according to (B), are what enters into logical relations like entailment, compatibility and incompatibility. C. Something people bear cognitive relations to. Propositions, according to (C) are the referents of the that-clauses in the propositional attitude reports. This threefold characterisation offers the possibility of providing a functional characterization of the notion of proposition as that entity, whatever it may be, which plays the roles specified in (A)–(C).2 To these points, the traditional picture of propositions adds two more traits,3 according to which propositions are: 1. For a general discussion of the notion of proposition see the introductory essay of Pitcher 1964. See also David 1994, 3–51, Anderson 1995, Loux 1998, cap. 4, and Iacona 2002. 2. The question as to whether this functional characterization is enough to clarify the ontological status of propositions is up for discussion. As we shall see, many realists think that a more robust ontological characterization is required. 3. For the point of this integration see, e.g. Bealer 1993.

D. mind-and-language-independent abstract objects; E. public objects in the sense of being, at least in principle, accessible by many.4 Given the crucial role that propositions have traditionally been taken to play, it is no surprise that they have been the focus of much discussion in metaphysics, philosophy of language/mind and philosophy of logic. In what follows we shall mainly focus on two issues which have been central to the philosophical debate in this area, namely: i.

Are propositions dispensable within a semantic theory? In particular, which arguments can be given for or against their dispensability? ii. If propositions are not dispensable, which conception is most suited to adequately account for them? Regarding (i), those who reject the dispensability claim qualify as Realistis about propositions and their answers get labelled as “ontologically serious”.5 On the other hand, those who accept the dispensability claim are called Nominalists. Nominalists typically claim that any appeal to propositions is explanatorily redundant in so far as it is possible to account for all the relevant phenomena without actually resorting to them. On this ground they claim that a theory that introduces propositions among its theoretical posits violates Ockham’s razor because it multiplies entities beyond necessity. 1. Nominalism and realism come in many different varieties. Let us start with realism and provide a sketchy presentation of the main positions in 4. Emblematic on this point is the case of Frege for whom the publicity requirement (which was grounded on the objectivity of propositions — thoughts in his terminology) was a fundamental prerequisite for successful communication and for the very possibility of a common science. This point is central to his fight against the psychologistic conception of thoughts/propositions as mental entities. As he says “A thought does not belong specially to the person who thinks it, as does an idea to the person who has it: everyone who grasps it encounters it in the same way. Otherwise two people would never attach the same thought to the same sentence, but each would have his own thought … It would be quite impossible for the assertions of different people to contradict one another … So a dispute about the truth of something would be futile…Nor must we say that one person might communicate his thought to another…It would be quite impossible for a thought to be so communicated that it should pass out of the private world of one person into that of another” (Frege 1969, in Beaney, ed., 233–234). 5. On this qualification see Alston 1996.

2

the debate that fall under this label.6 According to realists, propositions are entities of a very special kind which, while being susceptible to a functional characterisation, are in principle independent from their functional roles and therefore in need of ontological clarification. There are many different ways in which realists conceive of propositions. According to some varieties, propositions can be explanatorily reduced to some more primitive notions. One such reductive proposal identifies propositions with states of affairs conceived as complex entities designated by gerundial phrases like: (1) 2 + 3 being 5 (2) Gold being malleable (3) Mount Everest being 8846 metres high One famous representative of this position is Chisholm who conceives of propositions as “abstract entities which exist necessarily and which are such that some but not all of them occur, take place or obtain”.7 For Chisholm, even though any proposition is a state of affairs, not any state of affairs qualifies as a proposition. In his view, a proposition is a state of affairs that is necessarily such that either it obtains at all times or it obtains at no time (Chisholm 1981, 126). According to another variety, propositions are complex abstract entities whose constituents are typically the semantic values of the expressions occurring in the sentences that express them. As to how the semantic values of sub-sentential expressions are conceived, it is customary to distinguish between the Fregean and the Russellian variety. The latter, which is a form of naïve realism, takes the constituents of a proposition to be particular individuals, properties and relations. According to Russell,8 the proposition expressed by the sentence: (4) Mont Blanc is more than 4000 meters high has as its constituents the particular individual which is the referent of the expression ‘Mont Blanc’,9 and the property of being more than 4000 6. We shall deal specifically with realism in sections 4–5. 7. Chisholm 1976, 114. Traditionally, a state of affairs that obtains is called a “fact”. 8. See, e.g. Russell 1904. 9. Russell in a reply to Frege says “I believe that in spite of all its snowfields Mont Blanc itself is a component part of what is actually asserted in the proposition ‘Mont Blanc is more

3

meters hight which is the semantic value of the corresponding predicate in the sentence in precisely that order. The Fregean sub-variety, which incorporates a sophisticated form of realism, is characterized by the claim that a more fine grained individuation of propositions is required in order for them to play the role specified in (C), namely: to be the objects of the propositional attitudes. For Frege a proposition/thought is the sense [Sinn] of a sentence, that is a complex structured entity constituted by the “modes of presentations” of the Bedeutungen of the subsentential expressions.10 According to a third variety of realism, propositions are unstructured entities not to be analysed in terms of other entities, be they objects, properties, states of affairs, senses of sentences or what have you. A supporter of the “primitive-entity theory” is Bealer who takes propositions to be sui generis un-analysable Platonic entities.11 Unlike Bealer, most of those who espouse the unstructured model opt for some form of (explanatory) reductionism and claim that the entities playing the theoretical role of propositions have a radically different nature from that traditionally ascribed to them. The general proposal is to identify propositions with set-theoretic or other mathematical constructions, sets of possible worlds or functions from possible worlds to truth-values. An example of this last variety is Montague’s which identifies a proposition with the set of possible worlds in which the proposition is true.12 The less-committed varieties of realism takes propositions to be either mere modes or aspects which, while real, do not enjoy a separate existence, but exist only in the mental or illocutionary acts on which they ultimately depend for their very existence,13 or as pleonastic entities. We shall conclude this quick survey of the main varieties of realism by considering Schiffer’s pleonasticism as put forward by Schiffer (2003).14 than 4,000 metres high” (in Gabriel 1980, 169). 10. The locus classicus of Frege’s position is Frege 1892. The claim that for Frege a proposition/thought is the sense of a sentence needs qualification. A thought for Frege is not necessarily the sense of some actual sentence — for he admits the existence of thoughts which have never been /will never be expressed; nonetheless he insists that a thought is in principle expressible and therefore can be qualified as the sense of a possible sentence. 11. See Bealer 1998. 12. See Montague 1974. See also Stalnaker 1984. 13. A contemporary advocate of this kind of realism is for example Alston 1996. In this view — which is a variant of Aristotle’s stance on universals — the proposition that Mount Blanc is more than 4000 meters high exists in all those concrete acts, be they mental or linguistic, whose content is precisely that Mont Blanc is more than 4000 meters. 14. As for the correct way of setting Schiffer’s position the ideas diverge. Before 2003 he was

4

Schiffer’s argument for the idea of propositions as pleonastic entities is in line with what he says about other entities that he takes to be pleonastic such as ficta and properties for example. According to Schiffer, a pleonastic entity is one that is deposited in our ontology by transformations that take us from sentences in which no reference to that entity is made to sentences in which explicit reference to such an entity is made. These inferential moves, which he calls “something-from-nothing transformations”, are for example the following: (!) Felix is a cat ? The property of being a cat is exemplified by Felix and (*) Felix is a cat ?The proposition that Felix is a cat is true which license the introduction of properties and propositions respectively. The qualification ‘pleonastic’ which Schiffer uses is slightly confusing. Pleonastic entities are not pleonastic in the sense of being redundant, nor are they pleonastic in the sense of the pleonastic it. Rather, they are pleonastic, he claims, because they are introduced by “something from nothing transformations”15, and because the statement that there are such entities does not commit one, he claims, to new entities one was not already committed to. This second condition is explained in terms of the notion of conservative extension. A pleonastic entity is one which falls under a himself willing to qualify his position as a form of conceptualism, i.e. of the idea that propositions are creations of our conceptual or linguistic practices. But in 2003 he repudiates conceptualism by claiming (referring to properties, but the same point applies to propositions) “conceptualism about properties is false: nothing we do creates properties. At the same time, I believe that the theory of pleonastic properties is a conceptualist manqué theory: it is a theory that should relieve the need to be a conceptualist, a theory the conceptualist would have accepted had she thought of it” (Schiffer 2003, 66–67). Of course whether pleonasticism can be taken as a variety, however weak, of realism and not, instead, as a disguised form of nominalism, is an open question. Here we have confined ourselves in setting this position where his author wishes it to be placed. 15. As he says: “‘Pleonastic’ entities are entities whose existence is secured by somethingfrom-nothing transformations (I call these things ‘pleonastic’ entities because something-fromnothing transformations often take us to pleonastic equivalents of the statements from which they are inferred).” (Schiffer 2003, 51)

5

pleonastic concept, where a given concept F qualifies as pleonastic relative to a given consistent theory T which does not contain F only if the theory T1, obtained by adding F to T together with the relevant corresponding “something from nothing transformation”, is a conservative extension of T (equivalently, T1 is consistent and includes T ). What Schiffer means by conservative extension is the following: if we add a concept to a given theory T, say the concept proposition, and this addition does not affect the truth-value of T sentences then, according to pleonasticism, the inserted new concept should not cause much worry: nothing of what can be expressed thanks to the conceptual enrichment conflicts with T’s ascertained truths. Schiffer’s conclusion is that such entities as property and proposition are brought about by hypostatizing uses of natural language and that introducing terms which denote them or even quantifying over them is utterly innocent from an ontological point of view. In this sense a pleonastic entity is one that comes for free.16 2. Having seen some of the main varieties of realism, we shall now consider some varieties of nominalism, starting from the most robust one championed by Quine. As regards the ontology of propositions, Quine adopts an eliminativistic stance,17 which is in line with his general constraints on ontological respectability. For Quine — and many philosophers agree with him — identity criteria are required for ontological respectability; the only ontologically acceptable entities are those that are associated with clearly determined identity criteria. Propositions in his view do not satisfy this requirement and therefore they are entia non grata in the ontology. Let us see how Quine’s argument for that conclusion goes. In his view, a given entity a is admissible in the ontology if it is possible to express the truth-conditions of sentences with this form: a is identical with b.18 So, for example, sets and concrete material objects are ontologically admissible, because they possess identity criteria that enable one to pro16. Whether this is really so is of course a moot point. 17. See, for example, Quine 1960, 200–209. 18. For Quine identity criteria qualify truth-conditions of identity sentences in a peculiar way. See, for example, Quine 1953.

6

vide the truth-conditions of identity statements about them. By contrast, properties, attributes, and propositions, do not satisfy the stated requirement and therefore they are not admissible. The inadmissibility of these entities is a consequence of Quine’s general thesis that intensional notions concerning meaning are radically vague. If intensional notions concerning meaning are vague, the truth-conditions framed in intensional terms will be inadequate. Therefore, they are not admissible. This is precisely the upshot of Quine’s eliminativistic argument that is provided in four steps, each one of which represents a premise: (First premise) No entity can be introduced unless the truth-conditions of the corresponding identity statement can be provided; (Second premise) two propositions are identical if and only if the sentences that express them are synonymous. This is the only available identity criterion for propositions; (Third premise) there is not enough “behavioural” evidence for synonymy; (Fourth premise) no other kind of evidence is available for synonymy; (Conclusion) there is no criterion of identity for propositions. Therefore propositions must be eliminated from the ontology. Besides Quine’s position, there are weaker nominalistic positions to consider which adopt some form of reductionism. They are strategies which try to explain the apparent commitment to propositions by reducing talk about propositions to talk about (what are taken to be) less problematic entities. The reductive nominalists about propositions typically claim that it is possible to account for the functional roles individuated by points (A)–(C) above without actually resorting to propositions. They claim that what is pivotal is to identify some entity different from a proposition that can play the truth-bearer role; if something different from a proposition can act as a truth-bearer then, in their view, no appeal to propositions is required, because all the other roles are in a sense parasitic on that one. But what kind of entity different from a proposition could play a truth-bearer role? According to many nominalists, a good candidate is the sentence itself and, in the light of this conviction, they criticize the realist’s contention that sentences cannot do. The typical realist argument to that effect makes use of indexical sentences. According to the realist, given that an indexical sentence — such as for example

7

(5) I am Italian — expresses a truth in some contexts (those in which the utterer is Italian) and a falsehood in all the others (all those in which the utterer is not Italian), one cannot construe sentences as truth-bearers on pain of admitting the undesired result that one and the same sentence can be both true and false. To this argument the nominalist replies that all it shows is at most that sentences do not have truth-values absolutely but only relatively to the contexts of utterance. But this fact is no obstacle in his view to the claim that propositions are dispensable in favour of sentences.19 The nominalist’s complete strategy, once provided with a meta-linguistic solution to the truth-bearer issue, is to make use of this result in order to deal with the other issues. As regards propositional attitudes, for example, the nominalist’s suggestion is to treat them as relations which link cognitive subjects to sentences. A serious problem for any approach which treats propositional attitudes talk as language-bound is to explain how it is possible for subjects who speak different languages to share some attitudes. For, within this picture, believing that Rome is the capital of Italy, for example, turns out to be one thing for English speakers and quite another for speakers of different idioms.20 To this difficulty, many nominalists have originally reacted by making use of Sellar’s convention of “dot quotation”: a device which applies to a term ‘T’ and creates a common noun (‘˜T˜’) which is true of all those linguistic expressions which are functionally equivalent to the quoted one.21 As regards propositional attitudes, Sellars claims that even though no public utterance is involved in these cases, still some tokening of linguistic expressions must occur. To that end he resorts to the idea of thinking as inner speech forerunning in this way the “language of thought hypothesis” 19. There are two main varieties of sententialism: according to one of them propositions are to be identified with sentences of natural language; according to another variety, they are to be identified with “mentalese” sentences that is with sentential strings in the “Language of thought” that is that symbolic medium postulated by Fodor’s version of representationalism. For an example of the first variety see Carnap 1956. As for the latter see Fodor 1978, 1998 and also Richard 1990. 20. An extended discussion of this kind of difficulty can be found in Fodor 1978 where the critical target is Carnap. 21. On this technique see Sellars 1963. Dot quotation is a device expressively introduced in order to cut across language barriers so as to solve the main difficulties that plague most varieties of meta-linguistic strategy.

8

that is a popular view nowadays within the philosophy of mind. According to Sellars’ picture, to claim that: (6) Paul believes that Rome is the capital of Italy is to be analysed in the following way: (7) Paul tokens (or is disposed to token) a Mentalese ˜Rome is the capital of Italy˜. Even though the meta-linguistic strategy here presented is a widespread and popular move among reductive nominalism, it is important to emphasize that not all nominalists actually follow this path. An important exception to the meta-linguistic strategy which is worth considering is Prior’s (1971). According to Prior, talk apparently about propositions is not about linguistic entities, but rather about familiar concrete objects. Like many other reductive nominalists, Prior holds that the basic contexts in which propositions talk is invoked are those in which truth and falsehood are attributed, such as for example in: (8) That Rome is the capital of Italy is a true proposition. His suggestion is to explain away the apparent reference to propositions in such sentences by making use of Ramsey’s redundancy theory of truth (Ramsey 1927) according to which to assert that p is true is simply to assert p from which it follows that the truth and false predicates are eliminable.22 So, for example, to assert (8) is simply to assert: (8c) Rome is the capital of Italy. In Prior’s view, propositional-attitudes talk does not need to bring in propositions. To think differently is in his view to misunderstand the logical form of propositional-attitude ascription sentences. The error of the realist in his view is to take ‘believes that’ as expressing a relation that requires an object as its relatum. On Prior’s suggested reading, instead, ‘believes that’ should be taken as a functional operator that takes a sentence 22. The obvious problem with such accounts is that they work only where we have a thatclause that incorporates a complete declarative sentence.

9

(‘p’) as argument and gives a monadic or one-place complex psychological predicate as value (‘… believes that p’), a predicate which applies to cognitive subjects and which, taking as argument the name of the believer, gives as value either the true or the false. 3. We shall conclude our sketchy presentation of nominalism by considering two recent positions which have received considerable attention within the philosophical community, namely: Yablo’s fictionalism, and Hofweber’s account. These positions are strongly influenced by an approach to the ontological issues that is very different from Quine’s and which is customarily attributed to Carnap.23 Carnap rejects Quine’s criterion of ontological commitment and the very idea that the notion of ontological commitment of a theory T makes any plausible sense. In Carnap’s view, a claim such as for example (9) The Ps exist if taken as a genuine ontological claim — or, as he says, as an “external” claim — is senseless. The only legitimate ontological claims are those that he labels “internal”, where by “internal” he means a claim framed in a linguistic framework considered as an interpreted formal language. The internal claims are taken to be legitimate because they are assessed as true or as false according to the rules which are constitutive of the framework and which specify either the methods of observation (in the case of empirical sciences such as physics or biology) or the methods of proof (in the case of formal sciences such as logic or mathematics). So, for example it makes sense in his view to ask: (10) Is there a prime number greater than a hundred?24 By contrast, to ask: (11) Are there numbers?25 23. See in particular Carnap 1950. 24. Carnap 1956, 208–209. Carnap argues that the answer to this internal question is found “by logical analysis based on the rules for the new expressions. Therefore the answers are here analytic, i.e. logically true” (Carnap 1956, 209). 25. Again, see Carnap 1956, 209.

10

taken as a question concerning the extra-theoretical existence of numbers — is in Carnap’s view an external question concerning the existence and correctness of the framework itself and, as such, it cannot be meaningfully addressed. Carnap’s approach to the ontological issues has influenced some recent positions on propositions within the nominalistic landscape. One such position is fictionalism (of which Yablo’s is an important representative, see for example Yablo 1998). For a fictionalist in an inference such as (*): (*)

Felix is a cat ? The proposition that Felix is a cat is true

we specify the non-literal or fictional use of the expression ‘the proposition that …’ which shows up in the conclusion. The inferred sentence is not to be taken literally, according to the fictionalist, and, therefore, it is not committed to undesired entities such as propositions. Talk of propositions can be taken to have a metaphorical nature.26 The idea is the following. As the sentence (12) Felix is a sponge, in its metaphorical use, does not literally commit one to sponges (in the sense that its assertibility at time t is independent of the existence of sponges at that time), so the conclusion of (*) does not commit one to the existence of propositions. Such a conclusion, according to the fictionalist, if literally taken, is false (a false conclusion of what, literally speaking, turns out to be an invalid inference). As it turns out, fictionalism can be taken as a revised version of Carnap’s approach with the dichotomy internal/external replaced by the dichotomy literal/not literal. The influence of Carnap’s views on ontology is even more manifest in Hofweber’s approach to propositions.27 For this approach, an inference such as (*) is valid, but this validity is merely a result of a syntactic transformation or reshuffling of the syntactic material in order to emphasize some aspects of the information carried by the premise. In particular, Hofweber believes that the new singular term, which appears in the conclusion of the 26. See, e.g. Yablo 1998 and 2000. 27. Hofweber 2005. See also Hofweber 2006.

11

inference (*), may simply be taken as the effect of a focus construction (on the relevance of this linguistic aspects for ontology see Hoffweber 2005, 264–265). (*) is just the result of our linguistic competence and not of some alleged capacity of establishing, by means of conceptual analysis, the ontological commitments of natural language. This approach, however promising it seems to be, is not without problems. One problem is the following. Take inference (+) having as a premise the conclusion of (*), namely: (+)

The proposition that Felix is a cat is true

  ? There exists at least one proposition, i.e. the proposition that Felix is a cat, that is true This inference seems to be valid. And yet no purely syntactic phenomenon may be called upon in this case in order to avoid the problematic ontological commitment. Hofweber’s claim here is that the validity of the inference (+) is due to a kind of “semantic underspecification” which is typical of natural language quantifiers.28 His suggestion is strongly influenced by Carnap’s approach: he claims that quantifiers are ambiguous between an external and an internal reading. The locution: (13) There is at least one … in its external reading — or domain-conditioned reading — is an objectual, ontologically committing quantifier. This reading of the quantifier is the usual model theoretic one. However, he claims, there is another reading of the quantifier, a reading quite independent of issues of metaphysics or philosophy. In its second, internal reading, the quantifier has a certain inferential role. This reading can be motivated by a situation in which we want to communicate some information. Hofweber provides the following example: You are supposed to write a psychological profile of Fred, and you learn the most valuable information that Fred admires Nixon very much. This is most useful to you since it allows you to make a number of conclusions about what kind of things Fred values, and what kind of person he is. However, the next 28. An expression is semantically under-specified if and only if it may contribute in at least two ways to the truth conditions of the sentence to which it belongs.

12

day, when you sit down to write up the profile you just can’t remember who it was that Fred admires. All you remember is that whoever it is, this person is also admired by many Republicans. This is still very useful information, and you can communicate it to someone else as follows: … There is someone Fred admires very much and that person is also admired by many Republicans. Who is that, again?29

Hofweber observes that the situation is completely general. It does not really matter who it is that is admired in the example. The quantifier, in this reading has the role of a place-holder for the part of information that is forgotten. It does not matter what the original term was, whether or not, for example, it refers or fails to refer to some entities, or whether it has a completely different function. Quantifiers in the internal reading are not directly related to the ontological issues. “The truth of … ‘Something is F’ with the quantifier in its internal use does not by itself settle the ontological question about Fs” (Hofweber 2005, 274). So, the inference resulting from the combination of (*) and (+), i.e. Felix is a cat The proposition that Felix is a cat is true  ? There exists at least one proposition, i.e. the proposition that Felix is a cat, that is true is, in each of the two constituting passages, trivially valid and not ontologically committing. “In the first step” — Hofweber argues — “we make the transition from an ordinary, metaphysically innocent statement, to another one which is truth conditionally equivalent to it, but which uses a focus construction to bring out a certain aspect of the information communicated. This inference is valid, … and is indeed trivially valid … In the second step we make a trivial quantifier inference from a sentence that contains a (syntactically) singular term to one that contains a particular quantifier. This inference exploits the inferential role of the reading, and according to it such a transition is always valid, no matter what the semantic function of the singular term is” (Hofweber 2005, 275). In the internal reading of the quantifier even the second step of the inference is trivially valid. Before analysing in the next section the main varieties of realistic positions about propositions sketched in section 1, it is worth reminding the 29. Hofweber (2005, 31–32).

13

reader that not all nominalists have tried to dispense with propositions by accounting in a different way for the phenomena put forward by the realist. Some philosophers have tried to challenge the theoretical need for propositions by calling into question the idea that those phenomena have somehow to be accommodated. This challenge, which is much more serious and devastating than the previous one, originated in the philosophy of language and the philosophy of mind. Many points of the traditional picture were challenged. Among them the following: that for a mental act to have content is simply for the subject of the act to stand in some appropriate cognitive relation to an abstract entity; that this entity is immediately available to the cognitive subject; and that it is what is primarily true or false. An example of this theoretical move can be recognized in Putnam’s though-experiment in “The meaning of ‘meaning’” (in Putnam 1975) where the philosopher invites us to imagine a distant planet (Twin Earth) which resembles Earth in almost every respect apart from the fact that the colourless, tasteless, odourless liquid stuff that fills Twin Earth oceans, rivers and so on is not H2O but a different stuff, XYZ. As for the first point, Putnam claims that what is relevant for determining the content of a mental act is also the physical-social environment in which the cognitive subject is placed. As for the second point, he claims that the real content a subject enjoys is not introspectively available. And, finally, he claims that what is available in this way is the same for any given individual and his twin on Twin Earth and yet the real content of their acts can differ in truth-value. 4. A widely discussed issue is whether the nominalistic arguments do force the dispensability stance or whether this stance is ultimately unacceptable because it is inadequate in accounting for many things that cannot be accounted for without making use of the notion of proposition. All the arguments that realists have put forward for the indispensability stance are precisely to that effect. For each of the functional requirements (A)–(C) listed at the beginning, the realist provides an argument (or a series of them) to the effect that there is an entity that plays that functional role (being the referent of a that-clause, being the primary bearer of a truthvalue and so on). In the next step the realist claims that no entity different from a proposition (be it a linguistic entity, or a mental entity such as an individual mental act or whatever may accompany one such) can play the relevant role. Finally he provides some considerations in support of the unity thesis: the thesis that there is one and only one entity that conjunctively

14

plays all the functional roles stated. As an example, let us consider how the argument goes for the first requirement. The realist’s idea is that one cannot explain the linguistic activity of making statements without resorting to propositions. In making a statement, such as for example that: (14) Leopardi is famous we utter words which constitute a sentence of a given language, English say, and in so doing we perform two referential acts: that of picking up an individual (Giacomo Leopardi) and that of picking up a universal or a property (the property of being famous). According to the realist, it is not possible to adequately explain what is implied in the linguistic activity of making statements without appealing at some points to propositions. The reason why an account that appeals only to the sentence uttered and/or the referential acts performed would not do is that the same could be stated by using a different sentence in the same language, or a sentence in another language for example. So, moving from considerations concerning the linguistic activity of making statements, philosophers have postulated the existence of an entity — a “statement” (in the sense of what is stated and not in the sense of the act of stating) — satisfying the following three conditions: (i) it is the object of the acts of making statements; (ii) it is the referent of that clauses such as Paul asserted/stated/said that so-and-so; (iii) it is essentially true or false. Analogously, moving from considerations concerning that-clauses such as: (15) Paul believes/hopes/desires that so-and-so they have arrived at the conclusion that entities must exist — called “thoughts” this time — which function as the objects of various kind of acts of thinking and which are essentially true or false. The next step is to claim that we do not have two different kinds of entities here, namely statements and thoughts, but just one. The label “proposition” is then used to cover both cases even though propositions do not perfectly coincide, from an ontological point of view, either with statements (because propositions are independent from the language whereas statements seem to exhibit some form of dependence on the acts of stating) or with thoughts (because propositions are independent

15

from mind whereas thoughts seem to depend in some way on the acts of thinking). Even though all realists agree on the existence of propositions, they differ as regards how propositions are to be conceived. In section 1 we have considered some differences having to do with whether propositions are taken as structured entities or not and with whether they are taken as primitive or not. The possible combinations of these two parameters determine a matrix of four possibilities (Primitive and Structured, Non Primitive and Structured; Primitive and Non Structured; and finally Non Primitive and Non Structured). Here we shall mention in passing two other respects where realists could differ and which determine areas of disagreement among them. One area of disagreement has to do with whether propositions that are contingently true or false can vary their truth-value through time. Another point of disagreement is whether the notion of proposition can be identified with that of sentential meaning or not. The thesis that propositions are the meanings of declarative sentences has been championed for example by Church (1956). Within this picture a proposition is what is common to all the sentences that are synonymous with the sentence which expresses it. This equivalence however is not universally accepted. One source of resistance has to do with the phenomenon of indexicality. A sentence such as: (16) I am tired expresses a different proposition in different contexts. And yet, at the risk of wrongly collapsing indexicality into ambiguity, it has a meaning that stays fixed from one context to the other. 5. Let us now move to the second issue we listed at the beginning: (ii) If propositions turn out not to be dispensable, which model is most suited to adequately account for them? In addressing this question we shall mainly confine our attention to three models of the realistic framework, namely: (a) the set-theoretic model according to which a proposition is a set of circumstances of evaluation, namely the set of (actual or possible)

16

circumstances in which the sentence is true. (b) the Russellian model according to which a proposition is a complex entity whose structure mirrors, at least in part, that of the sentence and whose components are the semantic values of the sub-sentential expressions, namely: individuals, properties and relations. (c) the Fregean model according to which a proposition is a complex entity whose structure mirrors, at least in part, that of the sentence and whose components are the senses of the sub-sentential expressions. Let us consider these models in turn and illustrate their main virtues and vices. The first model originated in the ’60s in the course of devising the theory for systems of modal logic known as possible worlds semantics. According to possible worlds semantics, linguistic expressions are assigned extensions (appropriate to their syntactic category) at possible worlds. What determines the extension of an expression at a given world is a function, called the expression’s intension. Individual constants are assigned functions from possible worlds to individuals; n-places predicate constants are assigned functions from possible worlds to sets of n-tuples and, finally, sentences are assigned functions from possible worlds to truth-values. Given that the intension of a sentence is strictly connected with truth and falsity (being something which for any world as argument gives as value a truth value), it proved very natural to identify a proposition with a function from possible worlds to truth-values (or, analogously, with the set of possible worlds in which the proposition is true). This identification was greeted with great favour because, amongst other things, it promised to reduce talk apparently about problematic entities to talk about entities that were taken as less problematic (and anyway already required in modal logic). But the unstructured model, however useful it has proved within modeltheoretic semantics, has met with several problems. In particular, it does not seem to individuate propositions finely enough in so far as it ends up treating all logical equivalent sentences as expressing the same proposition with the consequent ascription of logical omniscience (as commonly treated) to any subject to whom a grasp of a logical truth can be ascribed. Similar problems arise also in those set-theoretic models that replace possible worlds with more sophisticated theoretical constructions. Indeed, as one of the harshest critics of the unstructured model of propositions has shown, namely Soames (1987), the view continues to be riddled with serious difficulties even when one puts forward finer grained accounts by

17

introducing metaphysically impossible worlds, inconsistent worlds and incomplete worlds. What is needed to overcome the above-mentioned difficulty is an account of propositions that “keeps track of ” the semantic values of the sub-sentential expressions, and no set-theoretic account seems to have the theoretical resources to satisfy this requirement. Another point of dissatisfaction with the unstructured model was that according to many people it obstructed the understanding of the real nature and functioning of singular terms. The ’70s witnessed an enormous amount of criticism against the traditional semantic picture which, from Kripke (1980) onwards, became labelled as the “descriptivist theory of names”. Kripke himself had replaced the traditional picture with a new one according to which proper names do not function as “flaccid” referential devices (a paradigm of which are definite descriptions used attributively), which designate different individuals in different possible worlds, but as “rigid” designators, that is as expressions which designate the same individual in every possible world. According to another leading figure of the new semantic paradigm, namely Kaplan (1977), the thesis that proper names are rigid designators was not radical enough to capture what is proper to them, namely: that, unlike descriptions, proper names refer directly without the mediation of anything like a Fregean Sinn or an intension. What gives grounds to these remarks is the idea that even though a definite description can behave as a rigid designator (consider for example ‘the successor of 3’) this is not enough to qualify such an expression as a genuine singular term. What is proper to the latter is its being a directly referential device and even though every directly referential expression is a rigid designator the reverse does not hold. In the light of this, Kaplan claimed that a problem with the set theoretic model of propositions was that it tended to blur the distinction between directly referential expressions and rigid definite descriptions. In his view, to make the distinction clear one has to give up on the set theoretic model of proposition in favour of (a given version of ) the structured model.30 30. Of course not any structured model will do but only one for which the singular term’s contribution to the proposition expressed is exhausted by the expression’s referent. A model which satisfies this requirement qualifies as Russellian. Examples of structured models which do not satisfy this requirement are all those which adopt some variety of the “structured intensions approach” which has among its advocates people such as Lewis (1970) and Cresswell (1985). Even though these philosophers are motivated by roughly the same considerations that motivate the Russellians — finding a finer grained account of propositions so as to distinguish between propositions true in all the same worlds for example — they differ from them in that they take the contribution of the singular term to the proposition expressed to be an intension rather

18

To say that propositions are structured is to say that they are complex entities having parts or constituents bound together in a certain way by some “structure-inducing bond”31. By itself this is a purely metaphysical thesis about the nature of propositions that does not imply anything specific about the kind of relation in which propositions stand to sentences. Nonetheless the generally adopted view is that the sentences confer structure to the propositions in the sense that what binds together the constituents of a proposition is a structure-inducing bond that renders the structure of the proposition isomorphic to the structure of the sentence expressing it. The idea behind this view is that propositions are structured like the sentences that express them, with a proposition’s constituents corresponding to (certain) constituents of the sentence expressing it. Unlike its main competing account, the structured account is able to allow for distinct necessarily equivalent propositions. The reason why the propositions expressed by: (17) Bachelors are unmarried and (18) Brothers are male siblings are different, while being necessarily equivalent, is that they contain different constituents (i.e. the former contains the semantic value of ‘bachelor’, the latter does not). An advantage32 provided by the structured accounts than an extension. The proposition in this picture turns out to be a complex entity obtained by composing the intensions of the sub-sentential expressions occurring in the formulation in the way indicated by the sentential structure. Another crucial difference is that in an approach such as Cresswell’s, for example, the finer grained entities associated with that-clauses are not taken to be the primary bearers of truth values (this role is ascribed to intensions conceived as sets of worlds). Cresswell’s is therefore an approach which does not satisfy the “unity thesis”, namely: the idea that the requirements stated at the beginning can be satisfied by one and only one notion of proposition. 31. What exactly this bond is and how it succeeds in inducing structure is far from clear. Within these approaches the notion of the structure of a proposition seems to reduce to the linear order of the proposition’s constituents kept together by their being elements of the same ordered n-tuple. In this sense one could say that the theory of propositions as structured entities not only does not solve but neither does it address Russell’s worry about the unity of the proposition. We thank Ernesto Napoli for this observation. 32. It is worth noting that it is possible to have this advantage without endorsing the idea

19

of proposition is that they enable to see what the semantic values of the sub-sentential expressions are.33 The thesis that a proposition mirrors the structure of the sentence does not imply, according to its supporters, that there must be a uniform one-to-one correspondence between words in the sentences and constituents in the proposition. This kind of correspondence may fail for many different reasons. According to some people,34 it may fail because the proposition can have what are called “unarticulated constituents” that is: constituents not contributed by any word figuring as a syntactic constituent of the sentence. Another reason why it may fail is that there are words — a case in point is that of pleonastic expressions — which, being devoid of semantic value, are unable to make any contribution to the proposition expressed. So, what the «mirroring» thesis does imply, according to its supporters, is the weaker claim that most constituents of the propositions are semantic values of words or phrases in the sentence. But what sort of structured entities are propositions or, what sort of things are their constituents? According to the most favoured model, propositions are complex entities constituted out of elements such as individuals, properties and relations.35 This model, which is called neo-Russellian, is contrasted with what is called the Fregean model that takes the constituents of a proposition to be either the senses of the sub-sentential expressions36 or, in the neo-Fregean model, a complex of senses and denotations.37 Nowadays many people think that the neo-Russellian model ought of propositions as complex entities having parts. This is so with Bealer’s “algebraic” view (Bealer 1993). Bealer associates with each proposition a “decomposition tree” which highlights which logical operations on which entities (individuals, properties, relations) a given proposition is the result of. The algebraic approach makes it possible to recover the semantic values of expressions in the sentence from the proposition expressed together with the decomposition tree associated with the sentence. But the account provided is not a structured one, because in Bealer’s view, the semantic values of the words in the sentence which figure in the decomposition tree are not part of the proposition which he conceives as a metaphysically simple entity. Not all algebraic accounts of propositions qualify as non-structured however. Examples of structured algebraic accounts are provided by Zalta (1983 and 1988) and Menzel (1993). Zalta 1988 provides an axiomatized account of propositions as zero-place relations in which propositions are fine grained enough to account for differences among necessary equivalent propositions. 33. This comes out from the fact that those values are the constituents of the proposition and that the constituents are visible in the proposition. 34. See for example Crimmins 1992. 35. See, e.g., Kaplan 1977, Salmon 1986, Soames 2002. 36. Searle 1958. 37. Properly speaking, there are two varieties of the neo-Fregean model both of which originates out of Evans’ seminal work (Evans 1982). According to one variety, championed by McDowell (1990, 1994), only senses can be constituents of propositions, even though some of

20

to be preferred because, in their view, it is the only one that is compatible with the actual functioning of referential expressions that direct reference semantics correctly accounts for.38 The widespread impression that direct reference actually forces a Russellian conception of propositions strongly depends on the way in which the phenomenon is characterised. Indeed, if one says that a directly referential expression is one whose only contribution to the proposition expressed is provided by its reference, it follows — by definition so to say — that the resulting proposition is a Russellian one. This characterisation of the phenomenon, in turn, strongly depends on certain intuitions concerning truth-makers (i.e. what makes a given proposition true or false in various circumstances).39 The question as to whether these are the intuitions that have to guide us in providing a characterisation of the phenomenon of direct reference and not, for example, the more basic ones concerning the relation between a name and its bearer is utterly legitimate.40 Even though it was Kaplan who first claimed that directly referential semantics required a Russellian conception of propositions, it was Salmon (1986) and Soames (1987) who, by taking the structured proposition account much more seriously, laid out one of its most articulated versions. Actually, in his 1987 work Soames provides both a formal theory of structured propositions and a definition of truth relative to a circumstance for structured propositions. However promising, the Neo-Russellian model is not without difficulties. Most of them originate from its commitment to the thesis that names, as directly referential devices, contribute to the proposition expressed only them (namely: singular senses) may be object-dependent (dependent for their existence on the existence of the object they are about). According to another variety, championed by Peacocke (1992) singular object dependent senses can have objects among their constituents. 38. Direct reference semantics is one of the two parts of the anti-Fregean paradigm in semantics, namely that part which aims at answering the question concerning what the meaning of certain classes of expressions is. The other part, the causal picture, is meant to be an answer to the question concerning what accounts for the expressions’ having the meaning that they do. 39. According to these intuitions, what matters in counterfactually assessing sentences in which directly referential expressions occur is the individual that those expressions designate in the actual world and not, for example, the (maybe different) individuals who in the various worlds satisfy a given descriptive condition associated with the name. 40. In our view, a characterisation of directly referential expressions as those which do not conform to the descriptive paradigm and whose semantic contribution to the proposition expressed depends on those expressions‘ having a referent would be more appropriate because it would not compromise, from the very beginning, the question as to which model, between the neo-Russellian or the neo-Fregean, is more suited to account for singular propositions.

21

with their referents. One such difficulty is that the suggested account does not seem to be sufficiently fine-grained to account for propositional attitudes.41 Another is that it seems unfit to justify the possibility of meaningful sentences containing non-denoting expressions. These difficulties, in their turn, seem to favour the Fregean model.42 The problems raised by propositional attitudes are particularly ticklish for those who adopt a Russellian theory of propositions. For, if one holds that propositional attitudes express a relation between an individual and a Russellian proposition, it follows that two belief-sentences differing at most in having two occurrences of different co-referential expressions, such as for example: (19) Paul believes that Carl Hempel once taught at Princeton University and (20) Paul believes that Peter Hempel once taught at Princeton University do express the same relation between an individual and a proposition given that the proposition expressed in the two cases by the embedded sentences (21) Carl Hempel once taught at Princeton University and (22) Peter Hempel once taught at Princeton University is the same. But this seems to violate our intuitions that the two sentences (19 and 20) may sometimes have different truth-values. Faced with this difficulty, the direct reference theorist may react in one 41. Soames 1987, 2002; Salmon and Soames (eds.) 1988; Richard 1990. 42. Indeed they have been precisely these two issues which have motivated what is called “bi-dimensionalism”, which is one of the most articulated attempt to relaunch descriptivism after Kripke’s celebrated attack (in Kripke 1980). The most important representatives of this new brand of descriptivism are Stalnaker, Jackson and Chalmers. Bidimensionalism is the critical target of Soames’ recent book Reference and Descriptions (Soames 2005).

22

of two ways:43 he may maintain that the intuition according to which the two embedded sentences are not equivalent is ungrounded, or he may try to show that, appearances notwithstanding, the theory does not have this consequence. A typical answer of those who choose the first horn is to say that to take the two embedded sentences as not equivalent is actually to confuse two different aspects, i.e. what is semantically expressed by the sentence and what is pragmatically implied. According to the other option, instead, the two embedded sentences may actually have different truthvalues, but what they express is not just a singular proposition having only individuals and properties as its constituents, but something richer. The general idea is that the proposition expressed is an amalgam of semantic content plus an extra-element. On what this extra-element amounts to the positions diverge: according to some people it amounts to linguistic material (a case in point is provided by Richard’s “linguistically enhanced propositions”44); others say it amounts to mental representations;45 still others say it is a “mode of presentation” like a Fregean sense.46 In one way or another, all these positions seem to imply a rejection of the thesis of the unity of proposition and a consequent adoption of a “revisionistic” stance. Assessing whether this revisionism is the only available option for all those who reject eliminativism or if, by contrast, it is possible to save the thesis of the unity of the proposition is a crucial issue.47 A position that adopts the unity thesis is Frege’s. That position, however, (or its descriptivist version at least) has been claimed to be based on a wrong picture of the way in which referential expressions function. The issue as to whether the adoption of a direct reference semantics is really inconsistent, as many have claimed, with a broadly Fregean conception of propositions or whether this is a bias which can ultimately be rejected is, in our view, one of the most stimulating challenge in the current philosophical debate in this area. 43. Indeed, there is a third reaction, famously suggested by Kripke (1979), to the effect that belief attributions mark off so muddled an area of our linguistic practice that no definitive verdict for or against a given semantic theory should be based on that ground. 44. See in particular Richard 1990. 45. See in particular Fodor 1987. 46. This is so in Salmon’s “guise theory”. See Salmon 1986 47. The question as to whether there is a unitary notion of proposition conceived as “what is said” and whether this notion falls within the province of semantics or within that of pragmatics is one of the core themes in the ongoing debate between literalism/minimalism and contextualism. See e.g. Recanati (2004), Cappelen and Lepore (2005), Stanley (2005).

23

6. According to some metaphysicians, the identity of a thing is given by its origin. In this case, it may prove useful to know that the present issue originated out of a conference entitled: “Propositions: semantics and ontological issues”, which was held in Padua (Italy) on 29–30 April, 2004. Though most of the papers here collected derive from corresponding talks delivered at the conference — and conversely most of the conference talks have become papers in this issue — this is not an issue of proceedings. For some of the contributors to the issue did not give a talk in the conference, even though some of them played an active role in making it possible. We are indebted to many people for different reasons. First of all we would like to thank all those who gave their contribution either to the conference or to the issue. Secondly, we would like to thank Giuseppe Micheli, chair of the Department of Philosophy of the University of Padua, for his support to this project. Our thanks also to Mirca Gallo, the administrative secretary of the Department, for her kind collaboration. Marina Canton and Vittorio Morato were very helpful both before and during the conference. Gillian Davies made some corrections and gave some advice to improve the English of some papers. Pierdaniele Giaretta and Ernesto Napoli read and commented earlier versions of this introduction. Many thanks to all of them. Thanks also to Thomas Binder for the accurate camera-ready work. And last, but not least, thanks to Maria E. Reicher — managing editor of the journal — and to the editors of the GPS Johannes Brandl and Marian David for having supported this longstanding enterprise. This issue has been funded with a “Progetto d’ateneo giovani ricercatori (Bando 2002. Progetto Codice CPDGo28322, macroarea 6): Proposizioni. Aspetti semantici ed ontologici” (coordinated by Elisabetta Sacchi: 2003–04) sponsored by the University of Padua (Italy).

REFERENCES Alston, W. P. 1996. A Realistic Conception of Truth. Itacha and London: Cornell University Press. Anderson, C. A. 1995. “Proposition, State of Affairs”. In: J. Kim and E. Sosa eds., A Companion to Metaphysics, Oxford: Blackwell, 419–421. Barwise, J., Perry, J. 1983. Situations and Attitudes. Cambridge Mass.: MIT Press.

24

Bealer, G. 1993. “Universals”. The Journal of Philosophy 90, 5–32. — 1998. “Propositions”. Mind 107, 1–32. Beaney, M. 1997. The Frege Reader. Oxford: Blackwell. Cappelen, H., Lepore, E. 2005. Insensitive Semantics. Oxford: Blackwell. Carnap, R. 1950. “Empiricism, Semantics, and Ontology”. Revue internationale de Philosophie 4, 20–40, repr. in R. Carnap, Meaning and Necessity. Chicago and London: University of Chicago Press, 1956, 205–221. — 1956. Meaning and Necessity. Chicago and London: University of Chicago Press. Chisholm, R. 1976. Person and Object. La Salle (Illinois): Open Court. — 1981. The First Person. An Essay on Reference and Intentionality. Brighton: The Harvester Press Limited. Church, A. 1956. “Propositions and Sentences”. In: I. M. Bochenski et al. eds., The Problem of Universals. Notre Dame, Indiana: University of Notre Dame Press, 3–12. Cresswell, M. J. 1985. Structured Meanings. Cambridge Mass.: MIT Press. Crimmins, M. 1992. Talk About Beliefs. Cambridge, Mass: MIT Press. David, M. 1994. Correspondence and Disquotation: An Essay on the Nature of Truth. Oxford: Oxford University Press. Evans, G. 1982. The Varieties of Reference. Oxford: Clarendon Press. Fodor, J. 1978. “Propositional Attitudes”. The Monist 61, 501–523. Repr. in J. Fodor,. Representations. Brighton: Harvester, 1981, 177–203. — 1987. Psychosemantics: The Problem of Meaning in the Philosophy of Mind. Cambridge Mass.: MIT Press. — 1998. Concepts. Where Cognitive Science Went Wrong. Oxford: Oxford University Press. Frege, G. 1892. “Über Sinn und Bedeutung”. Zeitschrift für Philosophie und philosophische Kritik 100, 25–50. Translated as “On Sense and reference”, in P. Geach and M. Black, eds., Translations from the Philosophical Writings of Gottlob Frege. Oxford: Blackwell, 1970, 56–78. — 1969. Nachgelassene Schriften. In: H. Hermes (et al), ed., Hamburg: Meiner, 139–61. Engl. Translation by P. Long and R. White eds., Posthumous Writings. Oxford: Blackwell, 1979, 128–49. Partially repr. in M. Beaney. The Frege Reader. Oxford: Blackwell, 1997, 227–250. Gabriel, G. eds. 1980, Gottlob Frege: Philosophical and Mathematical Correspondence, Chicago: University of Chicago Press. Hale, B. 1987. Abstract Objects. Oxford: Blackwell. Hofweber, T. 2005. “A Puzzle about Ontology”. Nous 39, 256–283. — 2006. “Inexpressible Properties and Propositions”. D. Zimmerman ed., Oxford Studies in Metaphysics, vol. 2. Oxford: Oxford University Press, 155–206.

25

Iacona, A. 2002. Propositions. Genova: Name. Jubien, M. 2001. “Propositions and the Objects of thought”, Philosophical Studies 104, 47–62. Kaplan, D. 1977. “Demonstratives”. In: J. Almog, H. Wettstein, J. Perry, eds., Themes from Kaplan. New York: Oxford University Press, 1989, 481–563. King, J. 1996. “Structured Propositions and Sentence Structure”. Journal of Philosophical Logic, 25, 495–521. Kripke, S. 1979. “A Puzzle about Belief ”. In: N. Salmon and S. Soames, eds., Propositions and Attitude. Oxford: Oxford University Press, 1988, 102–148. — 1980. Naming and Necessity. Oxford: Blackwell. Lemmon, E. 1966. “Sentences, statements, and propositions”. In: B. Williams and A. Montefiore, eds., British Analytical Philosophy. London: Routledge, 87–107. Lewis, D. 1970. “General Semantics”. Synthese, 22, 18–67. Also in: D. Davidson and G. Harman, eds., Semantics of Natural Language. Dordrecht: Reidel, 1972, 169–218. Loux, M. 1998. Metaphysics. A Contemporary Introduction. London and New York: Routledge. McDowell, J. 1990. “Peacocke and Evans on Demonstrative Content”. Mind 99, 255–266. — 1994. Mind and World. Cambridge, Mass.: Harvard University Press. Menzel, C. 1993. “The Proper Treatment of Predication in Fine-Grained Intensional Logic”. Philosophical Perspectives 7, Language and Logic. Atascadero, CA: Ridgeview Publishing Company, 61–87. Montague, R. 1974. Formal Philosophy. (R. Thomason editor). New Haven: Yale University Press. Peacocke, C. 1992. A Study of Concepts. Cambridge, Mass.: MIT Press. Pitcher, G. 1964. Truth. Englewood Cliffs, NJ: Prentice Hall. Prior, A. 1971. Objects of Thoughts. (P. Geach and A. Kenny editors) Oxford: Oxford University Press. Putnam, H. 1975. “The Meaning of ‘Meaning’”. In: H. Putnam, Philosophical Papers, vol II, Mind, Language and Reality. Cambridge: Cambridge University Press, 239–297. Quine, W. V. O. 1953. From a Logical Point of View. Cambridge, Mass.: Harvard University Press. — 1960. Word and Object. Cambridge, Mass.: MIT Press. Ramsey, F. P. 1927. “Facts and Propositions”. In: D. H. Mellor ed., Ramsey: Philosophical Papers. Cambridge: Cambridge University Press, 1990, 34–51. Recanati, F. 2004. Literal Meaning. Cambridge: Cambridge University Press. Richard, M. 1990. Propositional Attitudes. Cambridge: Cambridge University Press.

26

Russell, B. 1904. Selections from the Frege-Russell Correspondence, Excerpt from Russell to Frege, 12 December 1904, reprinted in N. Salmon and S. Soames, eds., Propositions and Attitudes, Oxford: Oxford University Press, 1988, 56–57. — 1905. “On Denoting”. Mind, 14, 479–93. Reprinted in Russell 1956. Logic and Knowledge. London: Allen and Unwin, 41–56. — 1919. “On Propositions: What They Are and How they Mean”. Proceedings of the Aristotelian Society, supp. Vol. 2, 1–43. Reprinted in Russell 1956. — 1956. Logic and Kowledge. R. C. Marsh, ed. London: Allen and Unwin. Salmon, N. 1986. Frege’s Puzzle. Cambridge Mass.: MIT Press. Searle, J. 1958. “Proper Names”. Mind 67, 166–173. Sellars, W. 1963. “Abstract entities”. Review of Metaphysics 16, 625–671. Repr. in: W. Sellars, Philosophical Perspectives. Springfield (Illinois): Charles C. Thomas, 1967. Schiffer, S. 2003. The Things We Mean. Oxford: Clarendon Press. Soames, S. 1987. “Direct Reference, Propositional Attitudes, and Semantic Content”. Philosophical Topics 15, 47–88. — 2002. Beyond Rigidity. Oxford: Oxford University Press. — 2005. Reference and Descriptions: The Case Against Two-Dimensionalism. Princeton: Princeton University Press. Stanley, J. 2005. “Semantics in Context”. In: G. Preyer, G. Peter, eds., Contextualism in Philosophy. Oxford: Oxford University Press, 221–254. Stalnaker, R. 1976. “Propositions”. In: A. F. MacKay and D. D. Merrill, eds., Issues in the Philosophy of Language. New Haven and London: Yale University Press. — 1984. Inquiry. Cambridge, Mass.: MIT Press. Yablo, S. 1998. “Does Ontology Rest on a Mistake?”. Proceedings of the Aristotelian Society, Supp. 72, 229–261. — 2000. “A Paradox of Existence”. In: A. Everett, T. Hofweber eds., Empty Names, Fiction and the Puzzles of Non-Existence. Stanford: CSLI, 275–312. Zalta, E. 1983. Abstract Objects: An Introduction to Axiomatic Metaphysics. Dordrecht: D. Reidel. — 1988. Intensional Logic and the Metaphysics of Intentionality. Cambridge, Mass.: MIT Press.

27

This page intentionally left blank

Grazer Philosophische Studien 72 (2006), 29–48.

ASCENT, PROPOSITIONS AND OTHER FORMAL OBJECTS Kevin MULLIGAN University of Geneva Summary Consider “Sam is sad” and “Sam exemplifies the property of being sad”. The second sentence mentions a property and predicates the relation of exemplification. It belongs to a large class of sentences which mention such formal objects as propositions, states of affairs, facts, concepts and sets and predicate formal properties such as the truth of propositions, the obtaining of states of affairs and relations such as falling under concepts and being members of sets. The first sentence belongs to a distinct class of sentences in which only non-formal objects are mentioned and only non-formal properties and relations are predicated. We can, it seems, infer validly from the first sentence to the second. They are also equivalent. And Sam exemplifies the property of sadness because Sam is sad. What is the relation between inference, equivalence and explanation in the case of our two sentences and in analogous cases? What right have we to assume that there are formal objects?

§ 1 Introduction On the one hand, there are all these humble, familiar objects — Sam, Erna, his kiss, their collision, her shape, his dreams, her salary, the snow and its whiteness, butter and its smell, tables, chairs and their respective positions. On the other hand, there are, or seem to be, a handful of much less familiar, much less humble, even sublime objects — propositions, states of affairs, facts, concepts and sets. Call these formal objects. Formal objects seem to have formal properties and to stand in formal relations: some propositions are true, some states of affairs obtain, objects exemplify properties, fall under concepts, and are members of sets. How do we get from one type of object to the other, from non-formal objects to formal objects? How do we pass from talk about one type of object to talk about the other type? These are anthropological questions.

There is also a non-anthopological question — How do non-formal objects relate to formal objects? Quine (1960, 271) called the transition from the use of an expression to mention thereof “semantic ascent”. I shall use the term “ascent” to refer to three types of transition: nominalisation, inferential transitions from mention of humble objects only to mention of formal objects, and explanatory transitions from explanations which mention only humble objects to explananda which mention formal objects. My goal is to understand what is involved in such transitions and the relations between them. § 2 Ascent, Semantics and Syntax Nominalisations, inferential transitions and explanatory transitions may also be classified by reference to their results. We may then distinguish between ordinary ascent, formal ascent and what I shall call material ascent. Examples of ordinary ascent are the transitions from (1) Sam is sad to (2) (3) (4) (5)

That Sam is sad Sam’s sadness Sam’s being sad Sad Sam.

The singular terms, (2)–(5), are nominalisations of (1). Nominalisation also takes us in a similar way from (6) Sam resembles Erna to (7) The resemblance between Sam and Erna. The transition from the nominal part of (8) The Channel Islands are wonderful

30

to the first nominal part of (9) The Channel Islands, the Alps and the Pyrenees, are the three most popular tourist destinations takes us from a name used in one way to the same name used in another way. Among the main examples of formal ascent are the transitions which take us from (1) or parts of (1) to (10) (11) (12) (13) (14) (15) (16) (17) (18) (19) (20) (21)

The proposition that Sam is sad The state of affairs that Sam is sad The fact (circumstance) that Sam is sad The property of sadness (being sad) The class of the sad The concept of sadness The extension of the concept of sadness The content of the concept of sadness The name “Sam” The predicate “is sad” The object of “Sam” The sentence “Sam is sad”,

and from (22) Sam, Maria and Tom to (23) The set (group, class, manifold, plurality) Sam, Maria, Tom. Are the results of formal ascent the results of nominalisation? If so, they are the results of nominalisations which differ from the results of ordinary ascent. For (10)–(21) all mention formal objects. There are transitions from (1) to other sentences which not only mention formal objects but also employ formal predicates, relational and monadic: (24) The proposition that Sam is sad is true

31

(25) The state of affairs that Sam is sad obtains (26) Sam exemplifies the property of sadness (27) Sam belongs to the class of the sad. The third type of ascent is material ascent: the transition from (28) Orange lies between yellow and red to (29) The colour orange lies between the colour red and the colour yellow, and from (30) Modesty is more important than chastity to (31) The virtue of modesty is more important than the virtue of chastity. By contrast, the transition from (32) 3 lies between 2 and 4 to (33) The number 3 lies between the number 2 and the number 4 is an example of formal ascent from a sentence which mentions three formal objects, 3, 2 and 4, to a sentence which mentions numbers. As we shall see, this example has one feature in common with material ascent. What is the logical form of expressions such as The proposition that Sam is sad The state of affairs that Sam is sad The fact (circumstance) that Sam is sad The property of sadness (being sad)?

32

They certainly look like definite descriptions. But they are not ordinary (humble) definite descriptions. According to one interesting view, The property of being sad has the form (34) x Property (x) & x = being sad.1 If this view is plausible, it can presumably be generalised: x Proposition (x) & x = that Sam is sad x State of affairs (x) & x = that Sam is sad. One objection to this view goes back to Frege’s claim that questions like “How many objects are there on the table?” and answers to this question such as “There are three objects on the table” are ill-formed. Similarly, it may be felt that bare quantification over formal entities of all sorts is illformed. The questions “Is there a property, a proposition, a state of affairs?” are as ill-formed as “There is a property, a state of affairs, a proposition”. As Husserl (2002, 131) points out, “Sentential expressions in subject position are ambiguous”. Indeed, when nominalisations of sentences which preserve sentential structure, such as “that Sam is sad”, flank the identity predicate they must be understood as elliptic. The following That snow is white = Tarski’s favourite thought That vixens are female foxes = Lewy’s favourite proposition That there is a chair over there = the fact both José and Ludwig alluded to in their discussions of primitive certainty might be elliptic for, respectively, The thought that that snow is white = Tarski’s favourite thought The proposition that vixens are female foxes = Lewy’s favourite proposition The fact that there is a chair over there = the fact both José and Ludwig alluded to in their discussions of primitive certainty. 1. Cf. Schnieder (2004, 38), Künne (2003, 10), Neale (1990, 116 n. 55).

33

There are no true identities of the form That p = that q because instances thereof are ill-formed. Instances of “that p” do, however, seem to function as names which take monadic predicates: That snow is white is surprising/certain/probable/possible. But a proper account of such apparent names presupposes an account of the relation between predicates such as “ — surprising/certain/probable/ possible”, on the one hand, and the functorial expressions “It is surprising/ certain/probable/possible that —”, on the other hand. Such an account itself presupposes an account of functorial expressions, in particular of the distinction between “pure” functorial expressions, which combine only with sentences, and “hybrid” functorial expression, which combine with a name and at least one sentence. I shall not attempt to pursue these questions here. Suppose that expressions of the form “that p” which flank the identity predicate are elliptic. What is the relation between the formal part (italicised) and the non-formal part of the descriptions in truths such as The proposition that vixens are female foxes = Lewy’s favourite proposition The state of affairs that Tully is bald = the state of affairs that Cicero is bald and in falsities such as The proposition that Tully is bald = the proposition that Cicero is bald? I suggest that each of the two parts is unsaturated or in need of completion. Instances of “that p” in such contexts require saturation, for example, by formal expressions such as “the proposition/state of affairs/probability/ certainty …” or by expressions of the same type as “Sam’s certainty/belief …” or by expressions such as “the dogma”. These expressions in their turn require completion by instances of “that p”. Whether or not this is right, and whether or not bare quantification

34

over propositions and other formal objects is acceptable, some account must be given of the fact that grasp of the expressions The property of being sad The proposition that Sam is sad involves understanding both the formal and the non-formal part of these descriptions and the relation between these. In this connexion many philosophers have suggested that what I have called the formal part of such descriptions is specified by or categorises the non-formal part and have argued that there is a relation of apposition between the formal and the non-formal part of our descriptions.2 Is this correct? In fact, it it is very plausible for descriptions which are the result of what I called above material ascent and for descriptions which are the result of one kind of formal ascent. But the view is wrong about the type of formal ascent we are most interested in here. Consider first the result of material ascent. There is sortal specification inside the three main nominal parts of The colour orange lies between the colour red and the colour yellow. The property of being a colour is a material or non-formal property. Orange is not a formal object. And orange is an instance of the material type or kind Colour, if such things exist. Using a terminology employed by Husserl and Wittgenstein, we may say that the sense of “orange” is a materialisation or specification, as opposed to a formalisation, of the sense of “colour”. And what holds of orange and colour holds, too, of chastity and virtue, of chagrin and emotion etc. Then there is the result of one kind of formal ascent, which has already been mentioned: The number 3 lies between the number 2 and the number 4. The property of being a number is a formal property and 2 is a formal object. 2 is an instance of the formal kind Number, if such things exist. 2. Wiggins calls Material object, Event, Number, Fictional entity “sortal-schemas” — Wiggins (1967, Appendix 5.4, 63); cf. Stevenson (1975), Wolterstorff (1970, 70ff.) , Levinson (1978, 9ff.), Wiggins (1984, 320), Teichmann (1992, 67ff.), and especially Schnieder (2004, ch. 1), Künne (2003, 258ff.). Jespersen (1937, 23) gives as an example of apposition, “The word ‘love’”.

35

The sense of “2” is a materialisation or specification of the sense of “number”. But consider now the following descriptions, all the result of a distinct kind of formal ascent The proposition that Sam is sad The state of affairs that Sam is sad The fact (circumstance) that Sam is sad The property of being sad. The senses of the italicised parts of these expressions are not specifications or materialisations of, nor are they categorised by, the parts which precede them. Here there are no determinable-determinate relations. The senses of “the proposition”, “the state of affairs” etc. are formalisations of the senses of the expressions which follow them. The relation between “the proposition” and “that Sam is sad” is like that between a sentential variable and an English sentence and like that between a nominal variable and a French proper name. A sentential variable is not specified by and does not categorise any English sentence, rather it is a formalisation of the sentence. In the same way, “the proposition” in “the proposition that Sam is sad” is a formalisation of “that Sam is sad”, and “the state of affairs” in “the state of affairs that Sam is sad” is a different formalisation of “that Sam is sad”. § 3 Ascent, Explanation, Inference and Intentionality One might well think that there are valid inferences corresponding to many of the nominalisations already mentioned in which our starting point is a sentence and the result of nominalisation is another sentence. For example, Sam is a dog Sam has the property of being a dog Schiffer calls inferences like this something-from-nothing transitions. And it seems that there are inferences of the same kind from (1) in § 1 to each of (24)–(27) in § 1. There are, of course, also valid inferences going in the other direction

36

Sam has the property of being a dog Sam is a dog And there are true equivalences such as Sam is a dog iff Sam has the property of being a dog Sam is sad iff the proposition that Sam is sad is true. Using the terminology already introduced, we may say that a valid something-from-nothing inference takes us from sentences mentioning only humble, non-formal objects to sentences which mention formal objects or which employ formal predicates such as “is true”, “obtains”, “exemplifies”. We can put more flesh on the idea of “something-from-nothing” by considering the relation between something-from-nothing inferences, on the one hand, and explanations, on the other hand. Consider (1) Sam is sad (2) The proposition that Sam is sad is true (3) If (1), then (2) because (1) (4) Sam exemplifies the property of being sad (5) If (1), then (4) because (1) (6) The state of affairs that Sam is sad obtains (7) If (1), then (6) because (1) etc. (3), (5) and (7) assert the explanatory priority of (1), which mentions only Sam, with respect to each of (2), (4) and (6), which mention formal objects or employ formal predicates. The question then arises as to whether there are ties of explanatory priority between sentences all of which mention formal objects or ascribe formal predicates. Many philosophers have found the following plausible: (8) If (4) & Sam belongs to the class of the sad, then (Sam belongs to the class of the sad because (4)) (9) If (2) & (6), then ((2) because (6)). What is the force of the “because” in (3), (5), (7), (8) and (9)?

37

There are many types of “because”, certainly more than four. There is: the because of the exasperated adult p because p! — which is always false, the causal because Sam had a heart attack because he was terrified the because of theoretical reduction This is a water molecule because it consists of two hydrogen atoms and one oxygen atom the because of subjective reasons for actions (beliefs, desires, emotions) Sally slapped Sam because she believed him to be a sexist the because of objective reasons for actions (beliefs, desires, emotions) Sally slapped Sam because he is a sexist and the normative because This is intrinsically valuable because it is a state of pleasure. There is also what I call the “essential because”, which is the “because” employed in (3), (5) and (7)–(9). Inspection reveals that the “because” employed in (3), (5), (7) and (9) is not any of the different types of “because” distinguished so far. But what exactly does the essential because amount to? Some illumination is provided by distinguishing yet another type of because and by considering its relation to the essential because. There is a “because” of essence. One example is If x endures/occurs/obtains/is alive/enjoys intentional existence/istzum-Tode …, then x endures/occurs/obtains/is alive/enjoys intentional existence/ist-zum-Tode … becauseessence of the essence of x. Or, in slogan form, the modes of being of objects are determined by the essences, natures, kinds or types of objects (a truth denied by existential-

38

ists). Another example is: If x and y are numerically distinct, then x and y are numerically distinct becauseessence of the essences of x and y. In each case, the “because” of essence is followed by a sentence which mentions the essence(s), nature(s) or kind(s) of object(s) mentioned in the sentence which precedes “because”. In other cases, as we shall see, the “because” of essence is followed by a sentence which mentions the essence or nature of something which is ascribed by the sentence preceding “because”.3 Now the essential “because” requires the “because” of essence. For example, If the proposition that p is true becauseessential the state of affairs that p obtains, then ((the proposition that p is true becauseessential the state of affairs that p obtains) becauseessence of the essence of truth and of propositions).4 Let us now return to the ties of explanatory priority between (1), on the one hand, and (2), (4) and (6), on the other hand. Clearly, (10) If the proposition that Sam is sad is true becauseessential Sam is sad then ((the proposition that Sam is sad is true becauseessential Sam is sad) becauseessence of the essence of Sam) is false, and (11) If the proposition that Sam is sad is true becauseessential Sam is sad then ((the proposition that Sam is sad is true becauseessential Sam is sad) becauseessence of the essence of propositions and truth)

3. On the “because” of theoretical reduction, cf. Künne (2003, 154). The distinction between the “because” of subjective reasons and that of objective reasons goes back to Bolzano and is currently much discussed. The normative “because” is employed by Husserl and by Fine (in unpublished work). The esential “because” is related to what Künne (2003, 154, 229) calls the because of conceptual explanation. The because of essence belongs to the same family as Fine’s “x makes p true in virtue of the essence of x”. 4. On this view, cf. Mulligan (2006a), on its history, cf. Mulligan (2006).

39

is true, as is (12) If Sam exemplifies the property of being F becauseessential Sam is sad then ((Sam exemplifies the property of being sad becauseessential Sam is sad) becauseessence of the essence of exemplification and properties). We have considered the ties of explanatory priority between humble sentences, on the one hand, and sentences mentioning formal entities, on the other hand. We have also considered explanations of these ties. But, as (10) reminds us, neither Sam nor his essence gives us any reason to infer from Sam is sad to Sam exemplifies the property of being sad. Schiffer says that such a valid inference is “conceptually valid” (Schiffer 2003, 2; cf. below). But suppose the inference whose credentials we are examining is Sam is sad Sam falls under the concept of sadness. This looks like a good something-from-nothing inference. But then it is not clear what is meant by calling such an inference “conceptually valid”. Suppose we say that valid inferences are either formally or non-formally valid.5 If the following is a paradigm example of a valid inference which is not formally valid This is red This is coloured 5. Valid inferences which are not formally valid, like the following example, are often called “materially valid” inferences.

40

and if a formally valid inference is an inference which is valid in virtue, in part, of the logical form of its premisses, are valid something-from-nothing inferences formally valid or non-formally valid? I suggest that they are valid inferences which are not formally valid, the conclusions of which are formalisations of their premisses. In other words, Sam is sad Sam exemplifies the property of being sad, Sam is sad The proposition that Sam is sad is true, and Sam is sad The state of affairs that Sam is sad obtains belong to the same family as Sam is sad p, Sam is sad Is sad (a), and Sam is sad F (Sam). (The difference between the two branches of the family is that in one case, but not in the other, the conclusions are formulated entirely in English). The same cannot be said of the inference from “Orange lies between red

41

and yellow” to “The colour orange lies between the colour red and the colour yellow”, nor of the inference from “3 lies between 2 and 4” to “The number 3 lies between the number 2 and the number 4”. But even if this characterisation of something-from-nothing inferences is correct, it still does not provide any answer to the question: What right have we to ascend? What right have we to formalise where the result of nominalisation mentions formal objects or employs formal predicates? In this connexion, Schiffer raises an important question. Can we imagine speakers of English who lack the concepts expressed by “property”, “proposition”, “is true” etc., who lack the expressions for these concepts? The answer to these questions is surely: yes. Indeed a community could speak something resembling English and lack many or most devices of nominalizations (cf. Schiffer 2003, 52). There are good reasons for thinking that speakers who use the truth predicate (as opposed to the truth functor) must apply it to propositions but perhaps we can imagine a community which does without the truth predicate and the truth functor. Similarly, we can perhaps imagine linguistic communities which function perfectly well without such formal concepts as the concepts of class, value and ought. Whether or not such communities are possible, there is no doubt that some philosophers have thought that, since there are neither propositions nor states of affairs, we cannot refer to them and that sentences dominated by axiological and deontic functors or predicates have no truth-values (“anticognitivism”). On this view, formal terms are semantically valueless. How, then, if at all, can reference to formal objects and predications employing formal predicates be justified? As far as I can see, the only justification is to be found in the theory of intentionality. The relevant chapter of the theory of intentionality is one that is somewhat neglected, the theory of the correctness conditions for different mental acts, states and activities. It is easy to see that the correctness conditions for some types of acts and states require that, in these conditions, we employ formal predicates. Consider desire (willing, not wishing): (13) x desires to F (14) x correctly desires to F (15) x ought to F. The correctness condition of desire is:

42

(16) If (14), then (15) Furthermore, (17) If (14), then (14) because (15). “Ought”, like “is true” and “obtains”, is a formal predicate; “ought” (here) takes an action-verb to make a predicate. Similarly, (18) (19) (20) (21) (22)

x prefers y to z x correctly prefers y to z y is better than z If (19), then (20) If (19), then (19) because (20).

“Better”, like “exemplifies”, is a formal relational expression. Is there any type of intentionality which requires us to admit states of affairs? Elswehere I have argued that coming to know that p, which has no conditions of correctness because it is already correct, is a relational state the second term of which is an obtaining state of affairs, that is, a fact.6 If this is correct and if the independent view that beliefs and convictions that p are reactions to knowledge that p or to apparent knowledge that p is plausible, then we have a good reason to accept (23) If x correctly believes that p, then the state of affairs that p obtains as well as (24) If x correctly believes that p, then p. There is also a good reason for thinking that specification of the correctness condition for beliefs requires us to mention propositions, that (25) If x correctly believes that p, then the proposition that p is true. Namely, 6. Mulligan (2006b).

43

(26) If x correctly believes that p, then (the content of x’s belief that p is an instance of the proposition that p & the proposition that p is true). (26) makes use of the idea that there are token propositional contents which instantiate types, an idea defended by the young Husserl.7 Even if it is true that that there are examples of formal and material ascent which are valid inferences, a theory of ascent should provide a principled way of ruling out invalid inferences. Schiffer discusses one part of this problem. He thinks that the inference from the claim that someone has used a proper name, “n”, in a make-believe way, to the claim that that person has created a fictional character, n, is an example of a something-from-nothing transformation (see § 4 below). He points out the following problem. Suppose a “wishdate” is defined as a person whose existence supervenes on a wish for a date. It does not follow from the definition that there are wishdates if there are wishes for dates. This would only follow if there were really wishdates. He says that there is a “crucial difference between the concept of a wishdate and the concept of a fictional entity (or of any other kind of pleonastic entity)” (Schiffer 2003, 54; my emphasis). Let us put on one side the topic of fictional entities and consider ascent from sentences mentioning humble objects only to sentences mentioning formal objects such as propositions and properties and to sentences mentioning material kinds such as virtues, emotions and colours. Here are two clearly invalid inferences Sam is sad The property Sam is sad, Orange lies between red and yellow The emotion orange lies between the emotion red and the emotion yellow. The second inference is invalid simply because orange is not an instance of the kind Emotion. In the case of the first inference, and of the large family to which it belongs, no such simple explanation is available. 7. (26) has the interesting consequence that “correct” is not a normative predicate if “true” is not a normative predicate.

44

§ 4 Appendix —Transformations: Husserl and Schiffer Two philosophers who have grappled with what I have called the problem of ascent, a hundred years apart, are Husserl and Schiffer. Much of what I have said above about explanatory ties is either also said by Husserl or is close to things he says.8 Let me therefore conclude by briefly considering some of the relations between what Schiffer and Husserl say about what I have called ascent. Husserl calls the different types of ascent “modifications”, meaningmodifications. His account of modification seems to have been influenced by Bolzano’s theory of “redundant” (“überfüllte”) ideas or concepts and by some remarks of Brentano. Husserl attaches great importance to the theory of modifications because modifications provide the “fundamental conceptual material ” for logic and formal ontology (Husserl 1950, § 119; my emphasis). Schiffer attaches almost the same degree of importance to ascent and argues for the existence of pleonastic entities, of which pleonastic propositions are one sub-category: Pleonastic entities are entities whose existence is secured by something-fromnothing transformations, these being conceptually valid inferences that take one from a statement in which no reference is made to a thing of a certain kind to a statement in which there is a reference to a thing of that kind. For example, the property of being a dog is a pleonastic entity. From the statement Lassie is a dog, whose only singular term is ‘Lassie’, we can validly infer its pleonastic equivalent Lassie has the property of being a dog, which contains the new singular term ‘the property of being a dog’, whose referent is the property of being a dog. (Schiffer 2003, 2)

Modification, in particular the operation of nominalization, the “law of “nominalization” (Husserl 1950, § 119), involves a type of meaningchange, Husserl says: It naturally happens … that certain meaning-changes belong to the grammatical normal stock-in-trade of every language. (LI IV § 11 cf. tr. 513)

These are what Schiffer calls our “hypostatising practices”. The relevant “meaning-changes”, Husserl says, involve transformations: 8. Cf. Mulligan (2004) for details.

45

… we are here dealing with alterations in meaning or, more precisely, alterations in acts of meaning which are rooted in the ideal nature of the meaning-realm itself. They have their roots in meaning-modifications in a certain other sense of “meaning” which abstracts from expressions, but which is not unlike that of arithmetical talk of “transforming” arithmetical patterns. In the realm of meaning there are a priori laws allowing meanings to be transformed into new meanings while preserving an essential kernel. (LI IV § 11, cf. tr. 515)

Indeed Husserl understands “transformation” and the possibility of transformations in a very strong way: The relevant possibilities are not to be understood in an empirical-psychological, biological sense but as expressing a peculiar relation of essence grounded in the phenomenological content of the experiences. (LI V § 35, tr. 629)

The relevant sense of “possibility” is that in which there is a possibility which is a priori grounded in the essence of a geometrical figure that “one” can turn it about in space, distort it into certain other figures etc. (LI V § 36, cf. tr. 633)

Husserl and Schiffer introduce the type of transformation they are interested in by considering fiction. Schiffer calls the inference from 1 Joyce wrote a novel in which he used ‘Buck Mulligan’ in the pretending way characteristic of fiction to 2 Joyce created the fictional character Buck Mulligan an example of a something-from-nothing transformation (Schiffer 2003, 51). Husserl, like Meinong and Schiffer, thinks that proper names in novels are examples of make-believe reference. Husserl begins with the case of fiction in order to introduce the idea that certain predicates (“is a fiction”, “is true”, “exists”) can only combine with expressions the meanings of which are modified meanings: All expressions to which “modifying” rather than “determining” predicates attach, function abnormally in the above described or some similar sense: the normal sense of our utterance is to be replaced by another … so that its

46

apparent subject (on a normal interpretation) is replaced by some sort of idea of itself, a logical idea [meaning] or an empirical-psychological idea … E.g. The centaur is a fiction of the poets With a little circumlocution we can instead say Our ideas (i.e subjective presentations with the meaning-content centaur) are poetic fictions. The predicates is, is not [exists, does not exist], is true, is false modify meaning. They do not express properties of the apparent subjects, but properties of the corresponding subject-meaning. E.g. that 2 u 2 = 5 is false means that the thought is a false thought, the proposition is a false proposition. (LI IV §11, cf. tr. 514–5)9

BIBLIOGRAPHY Husserl, E. 1950. Ideen zu einer reinen Phänomenologie und phänomenologischen Philosophie I (Husserliana III). Ed. by W. Biemel, The Hague: Martinus Nijhoff. — 1975. Logische Untersuchungen (1900–01, 1913). Ed. by E. Holstein, Martinus Nijhoff: The Hague; English transl.: Logical Investigations. Transl. by J. Findlay, London: Routledge and Kegan Paul, 1973 (abbr. as LU, LI). — 2002. Urteilstheorie. Vorlesung 1905 (Husserliana Materialien V). Dordrecht: Kluwer. Jespersen, O. 1937. Analytic Syntax. Copenhagen: Levin & Munksgaard. Künne, W. 2003. Conceptions of Truth. Oxford: Clarendon Press. Mulligan, K. 2004. “Essence and Modality. The Quintessence of Husserl’s Theory”. In: M. Siebel, M. Textor, eds. Semantik und Ontologie. Beiträge zur philosophischen Forschung. Frankfurt: Ontos Verlag, 387–418. — 2006. “Wahrheit und Wahrmachen in 1921”. forthcoming. — 2006a. “Two Dogmas of Truth-Making”. forthcoming. — 2006b. “Facts, Formal Objects and Ontology”. In: A. Bottani and R. Davies, eds. Modes of Existence. Papers in Ontology and Philosophical Logic. Frankfurt: Ontos Verlag. Levinson, J. 1978. “Properties and Related Entities”. Philosophy and Phenomenological Research, 39, 1–22. Neale, S. 1990. Descriptions. Cambridge (MA): MIT Press. 9. Thanks to Paolo Bonardi and Barry Smith for comments.

47

Quine, W. 1960. Word and Object. New York: John Wiley & Sons. Schiffer, S. 2003. The Things We Mean. Oxford: OUP. Schnieder, B. 2004. Substanzen und (ihre) Eigenschaften. Berlin: de Gruyter. Stevenson, L. 1975. “A Formal Theory of Sortal Quantification”. Notre Dame Journal of Formal Logic, 16, 185–207. Teichmann, R. 1992. Abstract Entities. London: Macmillan. Wiggins, D. 1967. Identity and Spatio-Temporal Continuity. Oxford: Basil Blackwell. — 1984. “The Sense and Reference of Predicates: A Running Repair to Frege’s Doctrine and a Plea for the Copula”. The Philosophical Quarterly, 34, 311–328. Wolterstorff, N. 1970. On Universals. Chicago/London: The University of Chicago Press.

48

Grazer Philosophische Studien 72 (2006), 49–71.

COLOURING, MULTIPLE PROPOSITIONS, AND ASSERTORIC CONTENT Eva PICARDI University of Bologna Summary The paper argues that colouring is a conventional ingredient of literal meaning characterized by a considerable degree of semantic under-determination and a high degree of context-sensitivity. The positive, though tentative, suggestion made in the paper is that whereas in the case of words such as “but” and “damn” we are dealing with words lacking in specificity, in the case of pejoratives in general, and racist jargon in particular, we are dealing with words that express concepts that purport to describe the world as being in a certain way. The circumstance that in certain contexts of utterance colouring can be cancelled out, does not show that it forms a detachable part of a word’s literal meaning. It only shows that to account for the interplay between context, literal meaning and assertoric content is much trickier than meets the eye.

The question I should like to discuss in this paper is whether the apparatus of multiple propositions is an adequate tool for handling the linguistic phenomena that fall under the label of “expressive content”, and that Frege subsumed under the miscellaneous category of “colouring”. Before tackling my chosen topic, some preliminary remarks concerning current debates on the semantics/pragmatics divide are in order, for it is in that context that the multiple propositions approach was originally put forward. 1. Assertoric content and sentence meaning Michael Dummett, commenting on Wittgenstein’s picture theory, has written: A sentence cannot be a fact because it states just one thing; and the hearer, if he understands the language, must know just what it is that it states. A diagram

is not a fact, it is an object; and there are many facts about a diagram. … But a sentence is not like this : though a sentence may imply many things, it says just one thing, and if you understand it at all, you must know what it says; you cannot, just by studying the sentence more closely, elicit new things that it says, that you had not noticed before. (Dummett 1981, 37)

Here Dummett speaks of “a sentence”, but the context makes it clear that he is thinking of a token of a sentence-type produced in the process of making a statement; sentence-types, too, are bearers of meaning and hence the object of understanding. But to understand the meaning of a sentencetype and to understand which assertion has been made on a given occasion in uttering a token of it are very different things. It suffices to think of sentences with indexicals or of ambiguous sentences to appreciate the difference between grasp of the meaning of a sentence-type and grasp of the assertion made in uttering it on a certain occasion. Dummett (1993) has used the contrast between a dispositional and an occurrent or episodic sense of the verb “to understand” to illustrate the difference. That “a sentence says just one thing, and if you understand it at all, you must know what it says” is not a minority view. The problem is rather to spell out what that one thing is that a sentence says on a certain occasion of its use. That there is a core of “what a sentence literally and in the strict sense says”, was the guiding assumption underlying Grice’s approach to communication: his main concern over the years was precisely that of finding the appropriate way of drawing the line between saying in the literal sense, and communicating by dint of conventional and conversational implicatures. As I have shown elsewhere in detail (Picardi 2001), in this Grice was partly anticipated by Frege, who pursued, however, a different goal. In order to account for the workings of deductive inference Frege held that we should first identify that part of the content of a sentence (what at the time of his Begriffsschrift he called “a judgeable content” and later split into “thought” and “truth-value”) that is responsible for its inferential potential. Semantic content had to be analysed in such a way as to isolate the thought which is explicitly expressed by a given (declarative) sentence, and set it apart from the thoughts which are semantically presupposed by it and those that are merely “hinted”, without being explicitly expressed. A number of such “hints” may be traced to syntactic structure (e.g. the “if …, then …” construction), others may be supplied by prosody, intonation, gestures, facial expression, etc. A thought was identified by Frege with that part of a sentence’s Sinn that determines truth-conditions. Since Frege’s main aim was to explain how sentences, in virtue of their semantic 50

value (truth or falsehood) and logical form, contribute to truth-preserving (valid) inferences, he tended to throw all those features of sentence meaning that are irrelevant from the point of view of deductive inference into the dustbin of “psychology”. However, though he evinced no interest in working out a systematic account of linguistic communication, his writings are full of interesting insights. I am thinking here of his remarks on assertoric force and on colouring. Moreover, he was well aware that the context of utterance may play a role in determining which thought a sentence expresses on a given occasion of its use. In On Sense and Meaning he goes as far as to suggest that sometimes there a more simple thoughts expressed than sentences.1 During the last decade or so many have called into question the traditional picture of the division of labour between semantics and pragmatics. Contextualists and Relevance theorists of various hues (e.g. Travis 2002, Carston 2002, Recanati 2004) have urged that we give up the apparatus that Grice had developed to spell out what a sentence says in his favoured sense of “saying” under contextually specified circumstances. In fact, Bach (1999) argued that the times are ripe to give up the motto “One Sentence/ One Proposition”, on which much model-theoretic semantics was built. The Multiple Propositions approach was born in this context2. Also in recent discussions in epistemology the question as to which proposition a sentence expresses in a given context has been an object of some dispute. There are debates going on as to whether a sentence such as, e.g., “I know that this is called ‘slab’”, makes different assertions depending on whether it is uttered under ordinary circumstances or in the course of a philosophical discussion bearing on scepticism. In On Certainty Wittgenstein seems to favour such a reading;3 but many nowadays argue in favour of the literal (“invariant”) reading, and explain the different significance of different utterances of the same sentence by adverting to different epistemic standards that hold in ordinary discourse and in philosophical discourse and with respect to which knowledge attributions should be assessed. A good starting point for approaching the topic of multiple propositions is the literature on propositional attitudes (and, in particular, Salmon 1986, Richard 1997). A standard criticism of Frege’s theory of proper names is that the notion of cognitive significance (Erkenntniswert) to which he helped himself in order to account for informative identity statements 1. Cf. Neale (2001), Picardi (2001) and (2007). 2. For a linguistically motivated approach to multi-dimensionalism cf. Potts (2005). 3. See On Certainty, §§ 564–87, and passim.

51

and for belief-ascriptions involving sentences with different co-referential proper names, is guilty of blurring the distinction between the proposition semantically expressed by a sentence and the proposition(s) communicated by means of it in a given context, or between semantically imparted information and pragmatically imparted information. The leading idea here is that in singular propositions proper names behave rigidly, and not as Frege surmised. Sense plays no role in determining reference, but provides at most a mode of presentation of it. Semantics — so the argument goes — has to do with the proposition expressed, while pragmatics has to do with the proposition(s) communicated or the information otherwise imparted in a given context in the performance of linguistics acts. Modes of presentations may, on occasion, play a role in the pragmatics of communication but are semantically idle. The notion of “proposition expressed” admits of a variety of construals, ranging from Russellian propositions to structured propositions, sets of possible worlds, Fregean thoughts, truth-conditionally evaluable contents, etc. Pragmatics deals with the proposition(s) “communicated”, which may vary from context to context, depending on the audience directed intentions of the speaker. Grice’s work is supposed to provide the tools for handling communication and speaker’s meaning. Grice’s account is naturally wedded to a truth-conditional theory of meaning, and, as I pointed out elsewhere (Picardi 2001), Grice and Frege share a number of controversial assumptions as regards the demarcation of semantic content. Grice’s notion of the content of a “saying” is naturally construed as coinciding with truth-conditional semantic content; within the realm of pragmatics falls the study of whatever else is either conventionally implied by the use of certain words or conversationally conveyed by the speaker when addressing a given audience on a specific occasion. However, Grice’s notion of the content of a saying, like Frege’s notion of a thought, does not coincide with what the sentence says in virtue of the literal (standard, conventional) meaning of its component words and the way they are put together. A sentence’s truth-conditions are not delivered by disquotation alone: the sentence has first to be purged of colouring (Frege) and conventional implicatures (Grice). If the proposition expressed by a sentence is identified with the truth-conditionally evaluable content of the latter, then the proposition expressed rests on, but it not to be identified with, the sentence’s literal meaning. These reminders should not be taken as implying that the notion of a proposition is hopelessly ambiguous. Only I wish to emphasize that no

52

answer is forthcoming to the question “What proposition does a sentence express?” that relies solely on intuition (be it metaphysical, semantic or epistemological). The notion of literal (standard, conventional) meaning will have to be clarified first. An answer to these questions can be given only in the framework of a theory of meaning, which attempts a systematic reconstruction of the skills and pieces of knowledge which speakers deploy when using language in thought and communication. Consider, for instance, the distinction between assertoric content and ingredient sense put forward by Michael Dummett (1991) in the framework of such a reconstruction. Should we identify a proposition with that which is conveyed by an assertion of a given sentence in isolation or with that which is conveyed by the sentence’s ingredient sense, i.e. its contributions as a constituent of complex sentences in which it occurs unasserted? It is only on the assumption that a sentence’s semantic contribution is uniform that we can ignore this distinction. This assumption, characteristic of truth-conditional semantics, requires justification. The compositionality of meaning does not mandate semantic uniformity. Indeed, as I pointed out elsewhere (Picardi 2001), there are as many different construals of the notion of compositionality as there are accounts of meaning. Dummett (1991) has argued that assertoric content is the right candidate for elucidating the notion of a proposition as that which can be justified, judged, challenged. It is on this notion that a theory of meaning should concentrate its attention: the notion of ingredient sense is needed to account for the way the sentence contributes to the truth-conditions of compound sentences, and it is on this notion that, when constructing a semantics for a language, we should concentrate. Brandom (1994 and 2000) develops Dummett’s suggestion, and describes the issue in terms of inferential potential; he holds that the inferential potential of a sentence that features as the antecedent of a conditional or embedded in the scope of a modal operator may be richer than its inferential potential as an isolated assertion. Stanley (1997) has applied this distinction to modal discourse, following by and large Dummett’s lead. In this paper I will tentatively argue in favour of the view that, depending on the context of utterance, colouring can contribute to assertoric content. If colouring can make a difference to the content of an assertoric utterance, in such a way as to affect either its truth-value or its assertibility conditions, then it should be reckoned also as part of semantic content. Much depends on how we characterize both the content of an assertoric utterance and the act of assertion itself, i.e. the commitments that a speaker undertakes

53

who puts forward a sentence with assertoric force, and the epistemic state (e.g. reliably acquired knowledge, high degree of subjective probability, justified belief, blind acceptance) that entitles him to perform as he does. My plea against semantic uniformity as a non-negotiable constraint on theories of meaning has much in common with Dummett’s, but is motivated by considerations having to do with meaning under-determination and context-sensitivity. Colouring is a feature of (word’s and sentence’s) meaning with a high degree of context-sensitivity, and this is why we can’t expect its semantic contribution to be uniform across contexts of utterance and sentential embeddings. More tentatively still, I will argue that we should not view context as providing a completion or a saturation of a propositional matrix, in such a way as to turn something incomplete into something complete, possibly a full-fledged proposition.4 The parallel I want to draw here is with vagueness: much as our understanding of sentences with vague predicates does not require that we fill in detailed information so as to eliminate vagueness, similarly our understanding of a sentence lacking in specificity (be it due to the presence of conventional implicatures, colouring or expressive content) should not be construed as directed at an elliptical proposition that awaits to be spelled out in full. This is, I believe, one of the most important lessons to be learned from Wittgenstein’s Philosophical Investigations.5 A sketchy drawing is not a defective drawing: it depends on the application that we want to make of it if we consider it as wanting under this or that respect. And yet, the model of completion seems to be a common feature of the accounts put forward both by proponents of a Multiple Propositions approach (such as e.g. Bach 1999, Neale 2001, Potts 2005, and, to some extent, Kaplan 1999) and by proponents of Radical Contextualism (such as Recanati 2005 and Carston 2002). 4. In his writings Frege often appealed to the metaphors of saturation and completion. Originally the metaphor of saturation was used to illustrate the mathematical notion of a function, concepts being treated as functions from objects to truth-values. In his first Logical Investigation, on the other hand, Frege appeals to the idea of completion to illustrate how information extracted from the context of utterance contributes to turn into a thought a sentence with token-reflexive expressions, which, as it stands, fails to express a thought. These two models are very different, and should not be conflated. The process of “completion” can bring about a variety of “thoughts” or none, whereas there is only one sentence that is picked out by saturating the argument place of a first-order function with a singular term. And something similar holds at the level of ontology and semantic evaluation. 5. To a discussion of the Builders’ language-games is devoted Picardi (2007a). As Wittgenstein (1953) points out, a mistake that we should not make in this context is to consider “Slab!” as elliptical for “Bring me a slab!”.

54

2. Sentences-types and literal meaning As is well known, the distinction between type and token was first appealed to in order to explain how it comes that numerically different utterances of the same sentence can be used to make different assertions. Sentences with token-reflexive expressions such as “I”, “here” and “now”, are the standard examples employed to illustrate the phenomenon of context-dependence at the level of semantic evaluation. However, they are only the tip of the iceberg. The phenomena of vagueness, semantic indeterminacy and underdeterminacy , ambiguity, metaphor, provide equally vivid instances of cases where a gap between sentence meaning (construed as literal meaning) and assertoric content is to be expected. However, to account for the contribution that context (external circumstances together with speakers’ intentions) makes to what a sentence says on a given occasion has proved a formidably difficult task. Indeed, one may wonder whether the distinction between sentence-type and sentence-token is at all required when doing semantics and pragmatics. It is important that we realize that these are two different questions: one may argue that it is impossible to do semantics without the notion of sentencetype and word-type, while acknowledging that the type/token distinction is inadequate for accounting for context-sensitivity in general, and for that which is conveyed in the performance of a linguistic act, in particular. An uncompromising application of Frege’s Context Principle along Wittgensteinian and Davidsonian lines seems to require that speech-act meaning (as Cappelen and Lepore (2005) call it) is given priority over sentence-type meaning. In fact, radical Contextualism suggests that we can dispense with the notion of sentence-type altogether: words undoubtedly have meanings, and they come first in the order of understanding, but it is the concrete utterance that is the primary bearer of meaning, for it is on the actual use made by the sentence on a certain occasion that we must depend if we want to understand what is being said. According to some philosophers, sentence-types are “abstracted” from concrete utterances, not discerned in them (cf. Recanati 2005); others (e.g. Carston 2002) view sentence-types as matrices waiting for (relevant) completion. Presumably, understanding a matrix and understanding a sentence-type require different skills and back-ground knowledge. For all I know, Strawson (1950) and Austin were among the first to employ the type/token distinction in order to characterize this aspect of language use. Strawson (1950) employed that distinction in the course of

55

his criticism of Russell’s theory of descriptions: contra Russell he held that sentence-types are meaningful or meaningless, no matter how the world happens to be. The assertion made in uttering a token of a meaningful sentence type, on the other hand, may turn out to be defective (i.e. either void or neither true nor false) because of a radical reference failure. Austin argued more or less along the same lines. Within the locutionary act, the act of saying something, Austin distinguishes the phonetic act (the act of uttering certain noises), the phatic act (the act of uttering words of a certain type “belonging to and as belonging to, a certain vocabulary, conforming and as conforming to, a certain grammar”), and the rhetic act (the act of “using those vocables with a certain more-or-less definite sense and reference” Austin (1962, 95). Thus “He said ‘Get out’” reports a phatic act, whereas “He told me to get out” reports a rhetic act. The rheme, as Austin conceives of it, is a unit of speech, whereas the pheme is a unit of language. The typical fault of the rheme “is is to be vague or void or obscure, &c.” (Austin 1962, 98). The illocutionary act has to do with the force that speakers attach to their utterances: force, on Austin’s account, is not to be construed as part of the meaning (sense and reference) of what is said. Now, the type/token distinction should not be confused with the distinction between Universals and Particulars. Nominalists such as Quine can make use of this distinction as far as spoken and written words are concerned without embracing meaning realism, i.e. the idea that there is a meaning that all tokens of a word-type share. Quine not only denies any reality whatsoever to propositions (conceived say, as, universals capable of indefinitely many instantiations by means of sentences of the same language or of different languages); he also denies that the notion of proposition is a useful tool for doing semantics, for he holds that we can’t give a tenable account of analyticity and synonymy. One might endorse a form of instrumentalism concerning the role of abstract entities in semantics, without endorsing a commitment to propositions as self-subsistent entities. Rudolf Carnap (1950), for one, tried to occupy such a position. The type/token distinction is surely appropriate to the phonetic act; however, most of us (including Austin), tend to apply it to the phatic and the rhetic acts as well. The transition from the statement that two tokens of the same sentence had been uttered to the statement that the respective speakers stand to each other in the relation of same-saying is exceedingly natural. However, in making that transition those of us who do not subscribe to Quinean nominalistic strictures, take it for granted

56

that the sentence employed is endowed with meaning, prior to its being used in a certain context in the performance of a speech-act. The notion of same-saying rests implicitly on the notion of literal meaning, and, the latter, in its turn, is often elucidated by appeal to rules and conventions concerning sentence-types. Austin, for one, took it for granted that there are conventions of grammar and lexicon, in virtue of which a sentence’s standard or literal meaning is determined. As is well known, Davidson launched a powerful attack against this conception of literal meaning. One of the reasons why after 1986 Davidson prefers to speak of first meaning rather than of literal meaning is because he thinks that the latter notion evokes that of standard meaning, possibly governed by shared conventions of meaning and illocutionary force to which in speaking we try to conform to. Of course the conformity is not for its own sake, but in order to achieve the end of communication, i.e. understanding, Davidson believes that to explain how this end is actually achieved (or in principle achievable) nothing is gained by appealing to shared meanings, possibly governed by implicit rules or explicit conventions. Speaker and Interpreter do share a lot: they share a world, an environment, a form of life, interests, beliefs, expectations, perhaps standards of rationality. Only they do not share prepackaged meanings, rules, conventions, and if the latter are conceived as essential features of a language, then they do not share a language. Yet, they a understand each other most of the times. In the course of an article devoted to literary language Donald Davidson writes: Words, Frege emphasised, have meaning only in the context of a sentence. … It is at the sentential level that language connects with the interests and intentions language serves, and this is also the level at which the evidence for interpretation emerges. But just as words have meaning only in the context of a sentence, a sentence has meaning only in the context of use, as part, in some sense, of a particular language. There would be no saying what language a sentence belonged to if were not for the actual utterances or writings, not, perhaps of that very sentence, but of other sentences appropriately related to it. So in the end the sole source of linguistic meaning is the intentional production of tokens of sentences. If such acts did not have meanings, nothing would. There is no harm is assigning meaning to sentences, but this must always be a meaning derived from concrete occasions on which sentences are put to use. (Davidson 1993 and 2005, 170. My emphasis)

Like Davidson, I believe that no harm is done in admitting that sentences have meaning, but unlike him, I believe (a) that this move commits us

57

to the recognition of semantic entities, i.e. word-meanings and sentence meanings, und thus involves a radical departure from Quine’s picture of language; and (b) that the order of explanatory priority suggested by Davidson conceals an ambiguity. The meaning of words can’t be derived just from the concrete occasions in which sentences are put to use, for in order to understand what a newly uttered sentence means we have to resort to our knowledge of the meanings of its component words. It is only at the very early stages of the enterprise of radical interpretation (and radical translation) that the order of priority suggested by Davidson (and Quine) is appropriate; but as soon as we apply our provisional truth-theory to new utterances, we avail ourselves of the clauses of satisfaction and designation concerning individual words. Such clauses are revisable, but are clauses concerning word-meaning nonetheless. In Frege’s terminology, our grasp of the thought expressed by the utterance of a given sentence is made possible, at least in part, by the grasp of the meanings of its component words and of the pattern discernable in it. By “pattern” I here mean, following Frege, one of the various decompositions into function and argument to which a sentence lends itself. However, as I said at the beginning, grasp of sentence meaning is often insufficient to determine the content of the assertion made by means of it — i.e. the particular thought conveyed on a certain occasion made in uttering the sentence. This is invariably the case when token-reflexive words are explicitly involved, but, as I said, the phenomenon is much more widespread. Indeed, one may wonder to what extent one is entitled to apply the type/token distinction to describe language use, if a gap is generally to be expected between sentence meaning and utterance meaning. In the course of a discussion of Husserl’s conception of ideal meanings as universals of sorts, Dummett (1993) remarks: The relation between a sense or meaning as a constituent of a proposition that can be grasped by different people and expressed repeatedly and as involved in a particular act of thinking is not that of type to token, or of universal to particular: the individual act exploits the general meaning. Mathematicians do not, for example, first have various thoughts involving the concept of integration, and then arrive at the meaning of the integral sign, or at the general concept, by noticing what is in common between all these individual thoughts. … The linguistic meaning comes first. We learn how to use the words of our language, and proceed so to use them: we do not impart a ready-made meaning to our words, and subsequently partition those imparted meanings into equivalence types. In at least the overwhelming majority of cases, we

58

could not have the thought unless we had first been given the word, or some synonymous word; once given the word, we need do nothing more in order to mean by it what it means than to use it. (Dummett 1993, 66–67)

Dummett is here formulating a criticism of Husserl’s conception of ideal meanings as universals prior to language, but somehow instantiated in concrete utterances. Frege’s conception of a thought, as inhabitant of a third realm, shares all the difficulties of Husserl’s conception. But Frege gave many different characterizations of what he called “a thought” that do not rest on the picture of the third realm. Frege did not only characterize a thought as the truth-evaluable core that synonymous sentences of the same language have in common, but also as that which should be preserved under translation of sentence of a given language into different languages. He also characterized the thought expressed by a sentence as that which sentences put forward with different illocutionary forces have in common, and, more to the point, as that which a sentence expresses when it occurs as the antecedent of a conditional, stripped of any force whatsoever. He also characterized a thought as that which is named by a that-clause, introduced by verbs of saying and of propositional attitudes. However, his theory of indirect sense and meaning made the connection between thoughts and thinkers somewhat opaque. “A thought” is also that which results when one completes a sentence with indexicals with the information extracted from the context of utterance. What Frege called “a thought”, contemporary philosophers often call “a proposition”: one of the issues currently under discussion is whether there is one notion of proposition which can fulfil all the various duties that Frege assigned to thoughts. Elsewhere (Picardi 1990) I denied this to be the case. Frege’s main argument in favour of semantic Platonism rests on his claim that in order to do justice to what he considered a fact, i.e. that different people can share the same thought when uttering (tokens) of the same sentence (type) we have to acknowledge the non-psychological reality of the thoughts and of their ingredient senses. But, as Dummett (1993) points out, the notions of grasping and sharing the grasp of the same thought can be given a very natural reading, without resorting to the fiction of a third realm of entities outside time and space. What is shared is, above all, a language. It is in the course of speaking a language that we acquire a set of abilities for responding appropriately when presented with sentences belonging to a language in whose practice of use we have been trained.

59

3. Colouring in context In On Sense and Meaning Frege mentions three ways in which words may differ. Either the difference concerns the Vorstellung (idea or representation) or it concerns the sense but not the reference, or concerns the reference as well: With respect to the first level, it must be noted that, on account of the uncertain connection between ideas (Vorstellungen) and words, a difference may hold for one person, which another does not find. The difference between a translation and the original text should properly not overstep the first level. To the possible difference here belong also the colouring [Färbung] and shading [Beleuchtung] which poetic eloquence seeks to give to the sense. Such colouring and shading are not objective, and must be evoked by each hearer or reader according to the hints of the poet or the speaker. (Frege 1892, 31, and 1984, 161)

Frege failed to describe accurately the phenomenon to which he had called attention and that Dummett(1973, 1991) calls “tone”. Inevitably, perhaps, for Frege’s category of colouring is a miscellaneous one, and he was more interested in brushing aside this aspect of meaning than in accounting for it. It is a great merit of Grice to have stressed the conventional character of this feature of the meaning of words such as “but”, “even”, “although”, “yet”, “still”, “therefore”. If colouring and shading are conventional features of word-meaning and/or utterance-meaning under which respect can they be said to differ from sense? Frege might have agreed that the ingredient of colouring is to some extent conventional, while maintaining that it is not capable of affecting the truth or falsity of the thought expressed by a sentence on any given occasion of its use. But then he would have had to offer an explanation of how it comes about that a conventional ingredient of word meaning or sentence meaning is, as a rule, unable to make a difference as regards the assertoric content conveyed by a given utterance of the sentence in question. Frege held that colouring, though stylistically important, does not affect the truth of the thought expressed by a sentence, considered in isolation. He formulates this thesis by saying that the assertoric force with which we utter a given sentence does not extend to colouring, for colouring is not part of the thought expressed: “What is called mood, atmosphere, illumination in a poem, what is portrayed by intonation and rhythm, does not belong to the thought. (Frege 1984, 357). As I have argued elsewhere (Picardi 2007) colouring differs from sense 60

only (a) in degree of specificity and (b) in context-sensitivity. It is only by relying on the (shaky) distinction between pragmatic inappropriateness and falsity that Grice could maintain that the presence of conventional implicatures makes no difference to the content of the assertion made in any given case. But there is no reason to expect that, as a rule, the meaning contribution of what, typographically, counts as a token of the same word type be uniform across contexts of use, as almost mechanical conceptions of principles of compositionality demand. Alternatively, one might argue that the structure of meanings types is far more complex than the structure of their linguistic counterparts, and this explains the apparent mismatch between sentential structure and semantic interpretation without appeal to (robust) context-sensitivity. Sometimes semantic interpretation requires that parameters be taken into account that are not overtly linked to the sentence’s component words.6 Needless so say, not all that colouring contributes to the overall significance of an utterance is of semantic significance. However, colouring is a feature of the overall significance of a word or sentence that the speaker expects his audience to be able to apprehend and to respond to. As many philosophers — most notably Kent Bach (1999), Tim Williamson (2005), Stephen Neale (2001) and myself (Picardi 2001)—have pointed out, Frege’s treatment of the pair “and” and “but” closely resembles Grice’s suggestion how we should conceive of the import of the word “but”. According to Frege (1918), in using this word the speaker is, as it were, stepping back to consider what he is saying, and dropping a hint of contrast to his audience. The idea of contrast carried by the “but” is considered by Grice as generating a conventional implicature, and not as an integral part of what the sentence “literally and in the strict sense” says. In the case of “but” and “on the other hand”, Grice’s suggestion is that we conceive of the contribution of these words to what is said in uttering, e.g. the sentence “She is poor, but she is honest”, as a higher order (“non-central”) speech act, parasitic on a ground-floor or central speech act (Grice 1989, 361–2). In the case of “but” we have a speech act of contrasting grafted onto the main speech act of asserting — better, perhaps, onto the two separate assertions “She is poor” and “She is honest”. Grice’s basic idea is that to the truth of what is said only that which is conveyed in the ground-floor speech act matters, while the falsity of the proposition conveyed by the higher-order speech act renders the speaker’s statement inappropriate, but does not falsify it. In this 6. King and Stanley (2005) seem to endorse a view along these lines.

61

he fully concurs with Frege. One of the main assumptions on which Frege’s and Grice’s accounts rest is that colouring and conventional implicatures are features of word meaning and/or sentence meaning that do not display context-sensitivity. And they were both wrong in the assumption they shared. What they both overlooked — and in the case of Frege this blind-spot is particularly striking — is that colouring is not, as a rule, cancelled out in reported speech and in sentential embeddings. In fact, as Bach (1999) has pointed out, there are contexts in which substituting “but” with “and” gives rise to syntactically unnatural readings. 4. Expressive content Frege applied the notion of colouring not only to words for logical particles, but also to common nouns. In his Introduction to Logic of 1897 Frege gives the example of the two words “Hund” and “Köter” translatable into English as “dog” and “cur”, and says that “That dog howled the whole night” and “That cur howled the whole night” express the same thought and differ only in colouring (Frege 1969, p. 152, and 1979, 184). He simply takes it for granted that the words “dog” and “cur” don’t just happen to be co-referential, but stand to each other in a relation which is stronger than extensional equivalence, synonymy, in fact. However, for one thing, we don’t have the shade of an argument that proves that these two words are intersubstitutable salva veritate in all extensional contexts. To his neighbour’s utterance “That cur howled all night”, the owner of the dog may retorts “That dog is not a cur”, but plainly he is not asserting that his dog is not a dog. Possibly, all curs are dogs, but not all dogs are curs. All that Frege is entitled to say is that there are contexts of utterance in which the difference in meaning between “cur” and “dog” makes no difference to truth-conditions of what is said, whereas there are other contexts in which the difference is salient.7 I may refuse to endorse an assertion because its wording suggests a picture of reality that I do not share. In the course of his discussion of the dog/cur example Frege makes an interesting suggestion, very similar in spirit, though not in application, to the one made by David Kaplan (1999), on how to handle the derogatory ingredient attaching to the word “damn”. In the given context the choice 7. See Williamson (2003) for a discussion of Frege’s notion of colouring and Grice’s notion of conventional implicature.

62

of “cur” instead of “dog” has the value of an exclamation, and, one may add, could be rendered syntactically by means of an exclamation mark, much as assertoric force is rendered by means of a vertical stroke. Frege held that assertoric force only shows itself with the help of a suitable notation, but is not located in any part of speech in particular. Its scope is the whole utterance, not a particular segment of it. The function of an interjection mark, encapsulated, as it were, in the meaning of “cur”, in the specific utterance is to disclose the attitude of the speaker towards the matter at issue. It presents the dog as ugly or unpleasant from the speaker’s perspective; however, as Frege remarks, the dog itself may very well be a handsome representative of its race. But this circumstance does not render the use of the word “cur” incorrect, for in uttering as he did, the speaker might have wished to disclose his attitude of dislike or fear of dogs in general, not of this dog in particular. In short, Frege’s suggestion seems to be that we construe colour as a higher-order utterance modifier. While assertoric force attaches to the thought expressed by the uttered sentence, higher-order utterance modifiers do not. Considered from this perspective, Frege’s contention that assertoric force does not extend to colouring (‘it makes no difference to the thought whether I use the word “horse” or “steed” or “nag” or “prad”’) is more convincing, not for the reason he offers (that colouring is of purely psychological significance), but because we construe the utterance as conveying either a sequence of propositions in accordance with a multi-dimensional model, or a higher-order utterance modifier according to Potts’ suggestion. As I said, Kent Bach (1999) has argued that the times are ripe to abandon the tenet on which much philosophy of language has been built, i.e. the Grundgedanke encapsulated in the slogan “One Sentence-One Proposition”. In suggesting that we give up that Grundgedanke Bach is not belabouring the obvious point that different utterances of the same sentence can make different assertions under different circumstances. He offers a picture of what goes on in communication according to which there is a minimal proposition (often incomplete and not truth-evaluable as it stands), which may trigger a sequence of propositions: which proposition is actually expressed and gets semantically evaluated depends on the context of utterance, as determined both by speakers’ intentions and external circumstances. That context contributes to content in a bottom-up manner in sentences with indexicals is agreed upon by all: what is the object of dispute is whether context and speaker-intentions also

63

contribute in a top down manner to the content of what is said.8 I don’t know whether Bach deems his proposal applicable to expressive content, for, as I said, his suggestion was made in the course of a criticism of Grice’s notion of conventional implicature. The advantage of Bach’s suggestion with respect to the higher-order modifiers view is that it accounts for the circumstance that the contribution of expressive meaning to assertoric content is somewhat unstable. If I report on the speaker who said “That cur howled all night” with the words “He said that that dog howled all night”, I am certainly leaving out part of what he said (and not just conventionally implied). Whether my leaving out this piece of information renders my report wrong or simply inaccurate depends on what was the main point of the utterance on the given occasion. And this, in its turn, depends also on the audience I am addressing and on of the focus of the present conversation: in the course of an investigation that aims at discovering the culprit of evil deeds against dogs in a certain neighbourhood it may be useful to give a literal report of what the people involved say concerning dogs. In a different context, the report may be less accurate, if, for instance, our interlocutor simply wants to find out what a notoriously nagging neighbour was complaining about on a certain occasion. Plainly, there is a lot that a speaker means to convey when he uses a sentence that we should not regard as being part of what is said by the sentence he chooses to use on a given occasion; however it is unclear to what extent what the speaker means to convey (either conventionally or conversationally) can be told apart from the assertoric content of his utterance. This is not a trifling question, for assertions are the natural bearers of truth and falsehood, and hence the notion of assertoric content is one of the central notions of semantics. In a number of unpublished papers, David Kaplan has suggested that we improve on Frege’s account of colouring, by introducing the notion of “Truth plus” or “Truth with an attitude”, and split the semantic content of such utterances into expressive content and descriptive content. I don’t think that this terminological choice of Kaplan is felicitous, for it reminds one too much of the evaluative/descriptive distinction, which is itself very controversial. Kaplan’s leading idea is this: wherever there is semantic content, there must be a “logic” of such content. Kaplan (1999) suggests that a 8. The distinction alludes to a narrow and a broad way of construing the contribution of context: under the narrow reading speakers’ intention do not contribute to semantic content, but may be relevant to speech-act content (cf. Recanati 2004 and Cappelen and Lepore 2005 for the use of this distinction).

64

full notion of validity should take into account also the content supplied by expressives: the circumstance that the conclusion of a deductive inference contains more expressive content than that enshrined in its premises is a symptom that we are dealing with an invalid form of argument. Kaplan’s proposal is in harmony, or at any rate suggests, a close connection with the multidimensional semantics put forward by Bach (1999). Christopher Potts in his recent book, The Logic of Conventional Implicatures, endorses Bach’s multidimensional approach and Kaplan’s work on expressive content, but rescues the notion of conventional implicature by changing its field of application. “But”, “even”, “unless” and “therefore” are no longer paradigmatic examples of words carrying conventional implicatures. The syntactical devices that fall within the area of Potts’s study are mainly utterance modifiers, commas, supplements, epithets, honorifics, expressive adjectives such as “damn”, words, i.e., apt to convey expressive content. Potts’ account is predicated on the assumption that at-issue content (ground level semantic content) and higher level expressive content are impermeable to each other.9 While I welcome the demise of the notion of conventional implicature as far as words such as “but”, “although”, “therefore”, and a host of others, are concerned, I doubt that the machinery of multiple propositions is the correct answer to the problems posed by colouring and expressive content, and I surmise that it is insufficient for dealing with pejoratives proper, such as those exemplified by racist jargon. It seems to me that pejoratives affect the at-issue level (the level of “what is said”, in the Gricean terminology), and have a bearing on assertoric content. Generally, the point of describing or referring to matters in a certain fashion is to prepare the grounds for drawing the consequences that follow from such a description. The fact that, if a hearer reports my statement “It is late, but I am used to working at night” by saying that I have said that it is late and that I am used to working at night, I am not going to correct him, is due to the circumstance that the word “but” 9. In particular, according to Potts (2005), Conventional Implicatures are characterized by the following features: a) Lexicality, i.e. expressive content is part of the meaning of certain linguistic items; b) Entailment, i.e. such words can have consequences for the notion of Kaplanvalidity; c) Speaker orientation.; d) Independence, i.e. expressive meaning carried by a lexical item plays no role in determining truth-condition. The feature of Speaker-orientation, however, may be tricky when it comes to reporting what other people have said, and may affect the connection of what Kaplan calls “semantics from below,”, to which expressive content belongs, and “semantics from above”, that accounts for the semantics of utterances at the ground floor, without employing at the meta-level the content displayed at the ground level. However, if Kaplan is right, Independence is not generally to be expected.

65

conveys a very generic contrast and that it is hard to see how substituting “but” with “and” can make a substantive difference to the assertoric content of the sentence uttered. And much the same can be said of words such as “damn” or other expletives. It is lack of specificity of words such as “but” and “damn” that is at the origin of our uncertainty as to whether the added ingredients can make a difference to the correctness or incorrectness of the assertion. Such expletives convey a highly unspecific attitude on the part of the speaker, which only the context may help to fill in. And often there is no need to fill in anything. Thus, far from expressing a multiplicity of propositions, such utterances display lack of specificity that only the context of utterance can compensate. But what context provides is not a full-fledged proposition (or a number of them), that can be in principle spelled out in full. In this I concur with Mark Sainsbury’s (2002) diagnosis of related phenomena. To avoid misunderstandings: I do not consider “lack of specificity” as a shortcoming of natural languages. However, lack of specificity, like vagueness, is a phenomenon that we need to understand better. “Under-determination” is another word often used in these contexts, but that too carries the tacit suggestion that determinacy is unconditionally either desirable or available. I think that we should free ourselves of such preconceptions. Multi-dimeansionalism in semantics is predicated on two assumptions: (a) that content, in order to be semantically evaluable, must have propositional status; and that, (b), in order to achieve this status it must undergo enrichment and saturation (Recanati 2005). I find these two assumptions very controversial. In his seminal essay on metaphor Davidson (1979), building on Wittgenstein’s notion of aspect-seeing, offered a host of convincing arguments to the effect that what a metaphorical statement suggests is not a proposition, possibly endowed with metaphorical meaning, but a way of looking at things.10 But one may subscribe to Davidson’s approach to metaphor, without endorsing his view that metaphor belongs to the province of pragmatics and concerns only language use. The times are ripe for broadening our conception of semantics: perhaps also contents devoid of a propositional format, such as metaphorical assertions, are semantically assessable as correct or incorrect, as they stand. Sentences lacking in specificity, too, can be used to make assertions, which, depending on the standards in force in a given context and the prevailing interests of the people involved, can be assessed as correct or incorrect. 10. For a discussion of this issue see Picardi (2006).

66

However, when it comes to pejoratives proper, i.e. racist jargon and politically incorrect language, things are different. If we followed Frege’s lead we would have to say that sentences containing the words “black” and “nigger” differ from each other only in a hint that the speaker drops to a hearer and are synonyms from the point of view of the thought expressed. But this can’t be right. Someone who says “Niggers and fags are not fit for this job” is not merely dropping a hint to his hearer: he referring to blacks and homosexuals in such a way as to convey that, say, the colour of the skin or the sexual preference as such are sufficient to ground a bias in ways of thinking and acting. But how are we to handle the concepts expressed by words such as “fag”, “nigger” or “terrone”? For it is not enough to have come to the conclusion that these words convey concepts different from the ones conveyed by their colourless relatives (provided that the latter are available in the language — an empirically ungrounded assumption!). Would it not be better to say that the words like “nigger” or “fag” resemble the connective “tonk” invented by Prior (1961) and discussed by Belnap(1962), in that they fail to specify concepts, much as the connective “tonk” fails to specify a function? One may object that, unlike “tonk”, pejoratives do not give rise to logical contradictions and that their employment is, as a rule, crowned with referential success. But referential success does not prove that a concept is in good standing; in fact, referential success can be achieved also through semantically defective means, as Donnellan (1966) has shown long ago. I have mentioned pejoratives, since Tim Williamson (2003) has suggested that we extend Frege’s (and Grice’s) treatment of “but” and “and” to account for the meaning and reference of pairs of words such as “German” and “Boche”. Pejoratives are legions: suffice it to mention the pairs “Marrano” and either “Jew” or “Muslim” (depending on the context), as used in XVI Century Spain, or “meridionale” and “terrone” (as used in contemporary Italy to refer to people born in the South of Italy). Williamson’s suggestion is made in a context where he opposes the proposal put forward by Dummett (1973: 453–455) and elaborated by Brandom (2000) and Boghossian (2003) concerning the meaning of pejoratives.11 Now, it is agreed on all hand that the use of such words conveys a commitment: the disagreement is about the form in which such commitments relate to the meaning of words and utterances. Whereas Inferentialists con11. Dummett (1973, 453-5); Brandom (2000).

67

strue the pejorative ingredient attaching to words such as “Boche” in terms of inferential (material) commitments undertaken by a speaker who uses this kind of word, and consider such inferences as meaning-constitutive, Williamson suggests that the derogatory element is not part of what the sentences in which they occur strictly speaking say, but resembles Fregean tone, construed as a conventional implicature (in Grice’s sense), which, as such, has no effect on truth-conditions. However, as I said, Grice’s notion of conventional implicature does not strike me as an useful tool for tackling the semantics of pejoratives and politically incorrect talk in general. I believe that we are dealing with concepts that strike our moral sensibility as “defective”. Dummett diagnoses this defect as ensuing from a disharmony between grounds and consequences of assertions. However, independently of the diagnosis, concepts expressed by pejoratives are concept nonetheless. As Frege urged long ago, a concept under which nothing falls, unlike an empty name, is a respectable concept. It so happens that the complex properties indicated by racist words are not instantiated, or, more soberly, that such words, when used as predicates, are false for any human as argument. Whether they are necessarily false (disharmony), or whether they just happen to be uniformly false, in spite of the fact that they are used with conviction and referential success by a considerable number of people, is a question that I will leave to the reader’s moral judgement. To sum up: (a) in this section I have expressed my scepticism concerning the employment of the apparatus of multiple propositions suggested by Potts (2005) and, to some extent, by Kaplan (1999), for handling expressive content, and (b), I have tried to show that Williamson’s proposal to apply Frege’s notion of colouring — construed as a conventional implicature along Gricean lines — to the semantics of pejoratives is unsatisfactory. My positive, though tentative, suggestion is that whereas in the case of words such as “but” and “damn” we are dealing with words lacking in specificity, in the case of pejoratives in general, and racist jargon in particular, we are dealing with words that express concepts that purport to describe the world as being in a certain way. The circumstance that in certain contexts of utterance colouring can be cancelled out, does not show that it forms a detachable part of a word’s literal meaning. It only shows that to account for the interplay between context, literal meaning and assertoric content is much trickier than meets the eye.

68

REFERENCES Austin, J. L. 1962. How to Do Things with Words. Oxford: Clarendon Press. Belnap, N. 1962. Tonk, Plonk and Plink. Now in: P. F. Strawson, ed. Philosophical Logic. Oxford: OUP, 1967, 132–136. Bach, K. 1999. “The Myth of Conventional Implicature”. Linguistics and Philosophy, 22, 327–366. Brandom, R. 1994. Making It Explicit: Reasoning, Representing, and Discursive Commitment. Cambridge, Mass.: Harvard University Press. — 2000. Articulating Reasons. Cambridge, Mass.: Harvard University Press. Cappelen, H., Lepore, E. 2005. Insensitive Semantics, Oxford: Blackwell. Carnap, R. 1950. Empiricism, Semantics and Ontology. Reprinted as Appendix A in Carnap, R. 1956. Meaning and Necessity. Chicago: University of Chicago Press, 205–221. Carston, R. 2002. Thoughts and Utterances. The Pragmatics of Explicit Communication. Oxford: Blackwell. Davidson, D. 1979. “What Metaphors Mean”. Now in Davidson, D. 1984. Inquiries into Truth and Interpretation. Oxford: OUP, 246–264. — 1986. “A Nice Derangement of Epitaphs”. Now in Davidson 2005. Truth, Language and History, Oxford: OUP, 89–108. — 1993. “Locating Literary Language”. Now in Davidson, D. 2005. Truth, Language and History. Oxford: OUP, 167–18. Donnellan, K. 1966. “Reference and Definite Descriptions”. The Philosophical Review, 75, 281–304. Dummett, M, 1973. Frege. Philosophy of Language, London: Duckworth. — 1981. “Frege and Wittgenstein”. In: I. Block, ed. Perspectives on the Philosophy of Wittgenstein. Oxford: Blackwell, 31–42. Now in Dummett, M. 1991. Frege and Other Philosophers. Oxford: OUP. — 1991. The Logical Basis of Metaphysics. London: Duckworth. — 1993. Origins of Analytical Philosophy. London: Duckworth. Frege, G. 1892. “Über Sinn und Bedeutung”. Zeitschrift für Philosophie und philosophische Kritik, 100, 25–50. — 1918. “Der Gedanke. Eine logische Untersuchung”. In: Beiträge zur Philosophie des Deutschen Idealismus, I, 58–77. Engl. transl. in M. Beaney ed., The Frege Reader. Blackwell: Oxford 1997, 325–345. — 1923. “Logische Untersuchungen. Dritter Teil: Gedankengefüge”. Beiträge zur Philosophie des Deutschen Idealismus, 3, 36–51. — 1969. Nachgelassene Schriften, ed. by H. Hermes (et al), Hamburg: Meiner. Engl. transl. by P. Long and R. White eds. Posthumous Writings. Oxford: Blackwell, 1979.

69

— 1984. Collected Papers on Mathematics, Logic, and Philosophy, edited by Brian McGuinness, Oxford, Blackwell. Engl. transl. of G. Frege, Kleine Schriften, a cura di I. Angelelli, Hildesheim: Olms, 1967, 21990. Grice, P. H. 1967. “Logic and Conversation”. In: Grice, P.H. 1989. Studies in the Way of Words. Cambridge Mass.: Harvard University Press. — 1989. “Retrospective Epilogue”. In: Grice, P.H. 1989. Studies in the Way of Words. Cambridge Mass.: Harvard University Press, 339–385. King, J. C., Stanley, J. 2005. “Semantics, Pragmatics, and the Role of Semantic Content”. In: Z. Szabò (ed.), Semantics vs. Pragmatics. Oxford: OUP, 357–382. Neale, S. 2001. “Implicature and Colouring”. In: G. Cosenza, ed. Grice’s Heritage Turnhout: Brepols, 138–184. Kaplan, D. 1999. What is meaning? Exploration in the theory of meaning as use. Draft 1, unpublished. — 1999. The meaning of ouch and oops. Exploration in the theory of meaning as use. Draft 2, unpublished. Picardi , E. 1990. “‘Über Sinn und Bedeutung’. Un’esposizione elementare. Parte II”. Lingua e Stile, 25, 159–199. Now in Picardi, E. 1994. La chimica dei concetti. Bologna: Il Mulino. — 2001. “Compositionality”. In: G. Cosenza, ed. Grice’s Heritage. Turnhout: Brepols, 52–72. — 2006. “Individualismo semantico e significato letterale”. In: R. Calcaterra ed. Le ragioni del conoscere e dell’agire. Scritti in onore di Rosaria Egidi, Milano: Franco Angeli, 392–402. — 2007 (forthcoming). “On Sense, Tone, and Accompanying Thoughts”. In: Schilpp & Hahn eds. The Philosophy of Michael Dummett. LaSalle: Open Court. — 2007a (forthcoming). “Concepts and Primitive Language-Games”. In: E. Zamuner, D. Levy, eds. Wittgenstein’s Enduring Argument. London: Routledge. Potts, C. 2005. The Logic of Conventional Implicatures. Oxford: OUP. Prior, A. 1961. The Runabout Inference Ticket. Now in P. F. Strawson, ed. Philosophical Logic. Oxford: OUP, 1967, 129–131. Recanati, F. 2004. Literal Meaning. Cambridge: Cambridge University Press. — 2005. “Literalism and Contextualism: Some Varieties”. In: G. Preyer, G. Peter, eds. Contextualism in Philosophy, Oxford: OUP, 171–196. Richard, M. 1997. “Propositional Attitudes”. In: B. Hale and C. Wright, eds. A Companion to the Philosophy of Language, Oxford: Blackwell, 197–226. Sainsbury, M. 2002. “Two Ways to Smoke a Cigarette”. In: E. Borg, ed. Meaning and Representation. Oxford: Blackwell, 94–114. Salmon, N. 1986. Frege’s Puzzle, Cambridge, Mass.: MIT Press.

70

Salmon, N. and Soames, S., eds. 1988. Propositions and Attitudes. Oxford: Oxford University Press. Soames, S. 2002. Beyond Rigidity. Oxford: OUP. — 2005. “Naming and Asserting”. In: Z. G. Szabò, ed. Semantics versus Pragmatics. Oxford: OUP, 356–382. Stanley, J. 1997. “Rigidity and Content”. In R. Heck, ed. Language, Thought and Logic. Essays in Honour of Michael Dummett. Oxford: Clarendon Press, 131–156. Strawson, P. F. 1950. “On Referring”. Now in P. F. Strawson 1971. Logico-Linguistic Papers. London: Methuen, 1–27. Travis, C. 1989. The Uses of Sense: Wittgenstein’s Philosophy of Language. Oxford: OUP. — 2000. Unshadowed Thought. Cambridge, Mass.: Harvard University Press. Williamson, T. 2003. “Understanding and Inference”. Proceedings of the Aristotelian Society, 249–293. Wittgenstein, L. 1953. Philosophische Untersuchungen/ Philosophical Investigations. Edited by G. E. M. Anscombe and R. Rhees; transl. by G. E. M. Anscombe, Oxford: Blackwell . — 1969. On Certainty. Edited by G. E. M. Anscombe and G. H. von Wright, transl. by D. Paul and G. E. M. Anscombe, Oxford: Blackwell.

71

This page intentionally left blank

Grazer Philosophische Studien 72 (2006), 73–94.

FREGEAN PROPOSITIONS AND THEIR GRASPABILITY1 Elisabetta SACCHI University of Padua, Italy Summary According to Frege a proposition — or, in his terms, a thought — is an abstract structured entity constituted by senses which satisfies, at least, the three following properties: it can be semantically assessed as true or as false, it is the object of so called propositional attitudes and it can be grasped. What Frege meant by ‘grasping’ is the peculiar way in which we can have epistemic access to propositions. The possibility for propositions to be grasped is put by Frege as a warrant for their existence: to challenge their graspability would amount to jeopardise their ontological reality. But is it true, as Frege uncritically maintained, that the “graspability requirement” is satisfied as far as propositions (as he conceived them) are concerned? This is the topic of the present work. A negative answer to the above mentioned question has been given in recent time by the representatives of what has come to be labelled the “cognitive turn” in analytical philosophy. People such as Fodor and Johnson-Laird patently denied the possibility for propositions, conceived à la Frege, to be accessed by the grasping relation. What grounds their position is, to put it roughly, the following train of thought: in order for something to be the target of the grasping relation it must enter the mind. Nothing which is different from a mental entity can enter the mind. Therefore, what can be grasped must be mental. The upshot of this move implies, among other things, the rejection of that radical anti-psychologism which was characteristic of the forefathers of the analytical tradition. In our work we shall try to resist their conclusion by showing that it is not necessary to zero the distinction between propositions and mental entities in order to provide an adequate account of the grasping relation. What one has to give up, 1. Earlier drafts of this paper were presented at various meetings: at ECAP V in Lisbon (August 2005) and at Genoa’s meeting on “Mental Processes: Representing and Inferring” (October 2005). I would like to thank all those who contributed with their criticisms, comments and suggestions to improve this work. In particular, I would like to thank M. Beaney who was my discussant in Genoa and W. Künne who carefully read a previous version. I am also indebted to the participants of our weekly seminar in Padua. My thanks also to A. Coliva, P. Giaretta, E. Picardi, and A. Voltolini who read earlier versions of this paper.

instead, is only Frege’s late Platonism of the “third realm” which, in our view, is a wholly unnecessary and dispensable accretion of his picture. For, as we shall show, if Platonism is in place it is difficult to provide an account of the grasping relation which makes no use of the “representationalist hypothesis” — i.e. of the hypothesis that ideas mediate our access to whatever can be given to us. But representationalism, once in place, makes the theoretical role of the notion of sense dispensable or purely additional.

Introduction According to Frege, a thought is an abstract structured entity constituted by senses which satisfies, at least, the three following properties: it can be semantically assessed as true or as false, it is the object of so called propositional attitudes and it can be grasped. What Frege meant by ‘grasping’ is the peculiar way in which cognitive subjects such as we are can have access to thoughts. The possibility for thoughts to be grasped is put by Frege as a requirement for their existence in such a way that challenging their graspability would amount to jeopardise their ontological reality. But is it true, as Frege maintained, that the “graspability requirement” is satisfied? Moreover, do Frege’s texts contain any clear indications as to how a theory of grasping should be devised? This is the topic of the paper. In addressing this topic we shall claim (1) that the model of grasping which is most compatible with Frege’s overall assumptions, in particular with the peculiar mixture of Cartesianism about the mind and Platonism about thoughts which Frege adopts, is some version of “Representationalism” — i. e. of the hypothesis that mental representations are what mediate our access to thoughts/contents; but (2) this very model does not fit well with the role the notion of Sinn plays within Frege’s theory or with the thesis of the non-mental character of thoughts which Frege and the Fregeans in general defend. We hasten to say from the very beginning that what motivates our investigation is not so much an exegetical interest in how the issue of grasping is dealt with in Frege but rather a theoretical concern. We believe that there is a tension between Frege’s theory of thought [Gedanke], on the one hand and the direction towards which Frege’s cursory remarks on thinking seem to point, on the other, and that this tension risks to overthrow two pivotal features of Frege’s philosophy of thought, namely: its anti-psychologism and the idea that it is via senses that anything can be given to us.

74

The idea that Representationalism is in tension with the idea of the non-mental character of contents has been defended in recent time by the supporters of the so called “cognitive turn” in analytical philosophy. People such as Fodor and Field, patently denied the possibility for contents, conceived à la Frege, to be accessed by the grasping relation unless this relation is mediated by some mental representation in the brain. The idea that thinking is a process of manipulation of mental representations according to their formal properties is characteristic of the computationalrepresentational model which these authors adopt. The gist of their position is that if thinking has a computational-representational nature, then thoughts and their constituents, the concepts, must be mental. This move implies, among other things, the rejection of that radical anti-psychologism which was characteristic of the forefathers of the analytical tradition. In this paper, we shall try to show that it is not necessary to zero the distinction between thoughts and mental entities in order to provide an account of the grasping relation. What one has to give up, instead, is rather Frege’s pernicious mixture of Cartesianism cum Platonism. For, as we shall claim, if these two assumptions are in place it is difficult to provide an account of grasping which makes no use of the “representationalist hypothesis”. But Representationalism, once in place, and fully developed, makes the theoretical role of the notion of sense dispensable or purely additional. We shall proceed as follows. In the first part, we shall present Frege’s remarks on thinking trying to work out what an explicit answer to the question of grasping thoughts could plausibly look like in Frege. As we shall see, Frege does not provide us with anything resembling (not even slightly) a picture of grasping. Nonetheless he says something, and what he says, however gappy and muddy it may be, imposes some constraints which rule out certain pictures and legitimate others. What we shall try to do is to consider what kind of picture fits better with Frege’s overall assumptions. The paper’s main aim is to show that this picture (which we shall qualify as a variety of Representationalism) engenders an internal tension in Frege’s theory. This tension, which stays somewhat hidden in Frege, comes out clearly within the framework of the computational-representational theories of the mind. Paradoxically, by bringing Representationalism to its full maturity, people such as Fodor, however distant from Frege they may be, seem to fill the gap Frege left open by providing the theory of thinking which he lacked. But, appearances notwithstanding, the cognitivist philosopher does not play Frege’s game. For — and this is bad news for the Fregean — if thinking is as the cognitivists depict it, then

75

thoughts and their constituents, the concepts, cannot be as the Fregean wants them. In the third section we shall consider how this tension can be dispelled. 1. Frege on grasping thoughts Let us start by providing some general outlines of Frege’s theory of thought. The theory of thought that Frege worked out in the course of his speculation is the heart of his philosophy; it is strictly connected with both his foundational project in the philosophy of mathematics known as “logicism” and with his mature semantic proposals centred around the notions of Sinn and Bedeutung. The thought [der Gedanke] is for Frege the real subject-matter of the philosopher whose task, as he conceives it, is to purge it from all those trappings which contaminate its purity. Among them Frege includes all those elements such as, e.g., representations [Vorstellungen], mental images, ideas and the likes which, in his view, normally go with thinking [Denken], but are extraneous to thought. During all of his philosophical speculations Frege never stops insisting on the importance of separating what is logical from what is psychological. The injunction to make such a separation figures at the beginning of his Grundlagen as one of Frege’s guiding principles and it certainly represents the leitmotiv of all his philosophy. The conflation of what is logical with what is psychological is, according to Frege, a potential source of confusions which, in his view, generate unacceptable consequences. According to Frege, logic and the theory of thought are strictly linked to each other. In the light of this, it comes as no surprise that Frege’s most extended treatments on the topic of thoughts are to be found in the several sketches for logic handbooks which he wrote during his life-time. Among them, the one which offers the most detailed and complete account of his theory is no doubt the First Logical Investigation: The Thought (1918– 1919). It is in this work that Frege puts forward his Platonic ontological doctrine which conceives of thoughts as self-subsistent entities existing outside space and time in an independent realm distinct from both the inner world of the mental and the external world of perceptible things. In agreement with the general anti-psychologistic orientation of his theorizing Frege felt himself exempt from dealing with the issue of how thinking concretely works. It is in this spirit that, in the posthumous fragment Logic he says:

76

This process is perhaps the most mysterious of all. But just because it is mental in character we do not need to concern ourselves with it in logic. It is enough for us that we can grasp thoughts and recognize them to be true; how this takes place is a question in its own right. (Frege, 1969, in Beaney, ed. 1997, 246)

Actually, Frege would have happily kept on ignoring the issue if not for a general objection which threatened the very possibility of grasping thoughts. According to this objection, one cannot obtain information about something not belonging to the inner world except by sense-perception.2 This claim, if maintained, would threaten the graspability of thoughts because, as abstract objects, they are not in the reach of sense perception. As a consequence, it would also threaten their reality. The reason, as Frege claims, is that the reality of thoughts depends on their being grasped, because it is only in so far as they are grasped that they can act on us and have in this way some degree of actuality. If it were not so, thoughts would be wholly inactive and therefore unactual. It is precisely in answering what has come to be labelled the “sensualistic objection” that Frege discusses the issue of grasping.3 And here is Frege’s answer: Sense-perception indeed is often thought to be the most certain, even the sole, source of knowledge about everything that does not belong to the inner world. But with what right? Sense-perception has as necessary constituents our sense-impressions and these are part of the inner world […] Sense-impressions alone do not reveal the external world to us. Perhaps there is a being that has only sense-impressions without seeing or touching things. To have visual impressions is not to see things […] Having visual impressions is certainly necessary for seeing things, but not sufficient. What must still be added is not anything sensible. And yet this is just what opens up the real world for us; for without this non-sensible something everyone would remain shut up in his inner world. So, perhaps, since the decisive factor lies in the non-sensible, something non-sensible, even without the co-operation of sense impressions, could also lead us out of the inner world and enable us to grasp thoughts. (Frege, 1918–1919, in Beaney, ed., 1997, 342–343. Emphasis added) 2. See, e.g. Frege (1918/9, 75), in Beaney ed. (1997, 342). 3. A comment is required. In what follows we are not respecting the precise order in which the quoted passages actually figure in Frege’s texts. For example, the passage which in our view is most revelatory of Frege’s “picture” of thinking actually occurs immediately before the “sensualistic objection”. Since this objection is in our view pivotal in bringing Frege to the topic of grasping we have preferred to put it beforehand. In any case, nothing of what we will say hinges specifically on this temporal inversion.

77

This passage raises many questions, among which the following: What exactly is Frege’s aim here? Does Frege have in mind a particular ‘non-sensible something’? Moreover, is he assuming that one and the same element is implicated in sense-perception and in thinking? Many different interpretations of this passage have been provided. According to Dummett (1988, 1991) and to Carl (1994) the non-sensible component which must be added to turn a mere sense impression into a real perception clearly belongs to the third realm. Picardi (1996) leaves it open whether the non-sensible element in question is to be construed as a thought-like ingredient, or as the faculty of reason tout court. The “faculty line” is defended by Malzkorn (2001) who claims that Frege in that passage had in mind a certain faculty to process sense-impressions of actual things and thus turn them into sense-perceptions of those things. But such a faculty, he maintains against Picardi, would be a cognitive faculty and it would not need to be the Kantian faculty of reason. A different interpretation has been provided by Stuhlmann-Laeisz (1995) who claims that Frege had in mind a particular non-sensible something, but not the same in both cases. In the case of perception, Frege thought that what was to be added to a person’s sense-impression of an actual thing in order to have a real perception was the thing itself. In the case of thoughts Frege used the word ‘non-sensible’ to refer to a means to grasp thoughts. Because of this equivocation he failed, according to them, to establish the analogy between perceiving and thinking.4 As for us, we think it plausible to claim that Frege’s main aim in that passage was simply to discard the sensualistic objection by showing that the difference between the ways in which a thing and a thought are given is not so big as to challenge the possibility for thoughts to be given to us. Anyway, since it is not our aim to discuss Frege’s suggestions about sense-perception let us leave matters here. What interests us is Frege’s conception of grasping thoughts and as far as this topic is concerned we think it fair to say that Frege’s passage, however revealing it may prove to be as 4. A more assuming reading of the above quoted passage which in our view is highly plausible is Dummett’s. If Dummett is right in claiming that for Frege perception involves at least the grasp of a complete thought, one can say that in that passage Frege offers a sort of “transcendental” argument for the graspability of thoughts, something along the following lines: (i) Let us assume, with the objector, that we perceive external things; (ii) The possibility of perceiving things requires the possibility for thoughts to be grasped; (iii) It must be possible to grasp thoughts, otherwise the very possibility of perceiving things would be threatened, contrary to the objector’s assumptions.

78

regards perception, does not say anything specific about thinking, apart from the fact that there is a close analogy between it and perception. Granted that Frege did not provide us with anything resembling, not even slightly, a theory of thinking, let us consider what kind of theory could fill the gap which Frege left open. It goes without saying that not any theory would do; what would do is rather one which is compatible with Frege’s overall picture, in particular: with his conception of thoughts on the one hand and of the thinking subject on the other. The reason for this requirement is simple. Thinking, whithin this picture, is conceived to be a process which relates a thinking subject to a thought. Consequently, how the process is conceived depends very much on how the two elements in the relations are conceived. As we know, Frege adhered to a strong Platonism about thoughts which treats them as self-subsistent entities existing outside space and time and in complete independence from both the inner world of the mental and the external world of actual things. As regards the mind, or more perspicuously the thinking subject, Frege adhered to a version of Cartesianism which conceives of the subject not as immediately in touch with the (external) world, but as “shut up” in his inner world (this point comes out neatly in that passage about perception we previously quoted).5 Let us take Platonism about thoughts and Cartesianism about the mind as our two assumptions and consider what kind of constraints they put on a theory of thinking. From these two assumptions it follows that thinking is a process which has (i) to lead a thinking subject out of his inner world (from the Cartesian assumption) and (ii) to bring him into contact with something, a thought, which is a wholly independent and self-subsistent entity (from the Platonist assumption). Now, since the subject is conceived as “shut up” in his inner world, what could do the job of “bringing him out” should be an element of that very world. But mind you, not any such; what is required is something having the capacity to point beyond itself. What could do the trick? Plausibly, something having a “representational nature”. From this to the conclusion 5. It is worth emphasizing that the label ‘Cartesianism’ is not to be applied to Frege in the full historical sense in such a way as to imply substance dualism for example.

79

that what enables the subject to make contact with a thought is a mental representation there is but a short step. On the ground of these remarks we claim that the model of thinking which in our view is most compatible with Frege’s overall picture, in particular with his peculiar mixture of Platonism cum Cartesianism, is some version of “Representationalism”6, that is, of the idea that mental representations are what mediate our access to contents/thoughts. Representationalism is also the picture of thinking shared by both the empiricist-rationalist tradition which Frege inherits; therefore, it would not be very strange if it turned out, in confirmation of our thesis, that Frege’s cursory remarks on thinking which we find in his work point in that very direction.7 Let us consider some of them. In a passage from his posthumous Logic (1897), after having said that the process of grasping thoughts is perhaps the most mysterious of all, Frege adds in a footnote: I should say that this question is still far from being grasped in all its difficulty. People are usually quite content to smuggle thinking in through a back door 6. It is not easy to clarify what is meant by “Representationalism”. Our use of this notion is strictly connected with that of Luntley (1999) according to whom Representationalism, which he takes to be the commonest version of intentional realism “is the thesis that our possession of content consists in our possession of entities/states of some specified kind called ‘representations’, where these are characterizable independently of that which they represent” (Luntley, 1999, 7). 7. With this we do not want to claim that Representationalism was the only answer which was open to the Fregean. Of course other options were available. Frege, for example, could have taken the “acquaintance-line of answer” and say that thoughts are objects which are immediately given to us. (For more on this line see below). Another possibility was to appeal to language and claim that it is via sentences that our grasp of thoughts proceeds. Incidentally, it is curious (to say the least) that Frege does not appeal to language in The Thought in dealing with the issue of grasping. For — and there is plenty of textual evidence on this point — he always emphasized that, for us men, it is only via language that we can have access to thought. (A very telling passage on this point can be found in the Sources of knowledge of Mathematics in which Frege claims: “The connection of a thought with one particular sentence is not a necessary one; but that a thought of which we are conscious is connected in our mind with some sentence or other is for us men necessary. But that does not lie in the nature of the thought but in our own nature. There is no contradiction in supposing there to exist beings that can grasp the same thought as we do without needing to clad it in a form that can be perceived by the sense. But still for us men there is this necessity”; Beaney, ed. (1997, 369). A possible explanation could be that after the discovery of the contradiction in his Grundgesetze, Frege’s lack of confidence in language increased exponentially. Granting therefore that other options besides Representationalism were open to Frege, what we claim is that Representationalism represents the most plausible answer for Frege at the time he wrote The Thought given the two above mentioned assumptions.

80

in the imagination, so that they don’t themselves know how it really got in. (Frege, 1969, in Beaney, ed. 1997, 246)

Coming back to this topic in the Thought he says more. Before raising what we have called the “sensualistic objection” Frege says: We do not have a thought as we have, say, a sense impression, but we also do not see a thought as we see, say, a star. So it is advisable to choose a special expression; the word ‘grasp’ suggests itself for the purpose. To the graspability of thoughts there must then correspond a special mental capacity, the power of thinking. In thinking we do not produce thoughts, we grasp them […] The grasp of a thought presupposes someone who grasps it, who thinks. He is the owner of the thinking, not of the thought. Although the thought does not belong with the contents of the thinker’s consciousness, there must be something in his consciousness that is aimed at the thought. But this should not be confused with the thought itself. Similarly Algol itself is different from the idea someone has of Algol. (Frege, 1918–1919, in Beaney, ed. 1997, 341–342. Emphasis added)

Some key points on thinking which come out from Frege’s quoted passages are the following: i. thinking is a (mental) process through which a thought is grasped; ii. it is a process in which what is mental comes into contact with what is not mental; iii. to the graspability of thought there corresponds a special mental capacity, the power of thinking; iv. the role of this capacity is to relate a person’s consciousness to an independently existing thought; v. grasping a thought is analogous to perceiving an object; vi. there must be something in the consciousness which “aims at” the thought, and this is different from the thought itself. Whether in (vi) Frege was thinking about the same “non sensible” something we considered earlier in connection with sense perception is not clear. We think it plausible to claim that the two elements, if not identical, are strictly connected in the sense that they stand in a close relation to each other.8 But now, unlike what happened before in the case of the 8. One could claim for example that by “the non-sensible something” Frege meant a cognitive faculty of which the “element in the consciousness which aims at the thought” is an essential ingredient. An interpretation of this kind has been suggested for example by Malzkorn (2001, 47–50).

81

“non sensible something”, we have more clues for the identification of the element at stake: we know that it is in the consciousness and therefore that it belongs to the second realm; we also know that it is different from the thought itself and that it must be capable of pointing beyond itself. In our view, what best satisfies all these requirements is a mental representation. That it is a mental representation that could do the trick is also backed by Frege’s numerous remarks, scattered throughout his works, on the indispensability of the representational element in thinking.9 In the fragment Logic in the section entitled Separating a thought from its trappings Frege writes: In human beings it is natural for thinking to be intermingled with having images and feelings. Logic has the task of isolating what is logical, not, to be sure, so that we should think without having images, which is no doubt impossible, but so that we should consciously distinguish the logical from what is attached to it in the way of ideas and feelings. (Frege, 1969, in Beaney, ed. 1997, 243)10

To conclude. Frege didn’t have in our view anything original and specific to say on the topic of thinking. His only worry was to dispel any doubt as to the graspability of thoughts. What he positively said is (partly) taken from the philosophical tradition which he inherited. Frege took on the representational model of the mind from empiricism, all the more because that model fitted neatly with (some of ) his other assumptions.11 Unfor9. It should be added, for intellectual honesty, that those passages do not clearly back any particular picture of thinking in preference to others. Part of the responsibility lies with Frege’s complex (somewhat ambiguous) use of the notion of representation by which he meant many (different) things, such as: the contents of consciousness, the empiricists Ideas, the sense data, the experiential/qualitative states, maybe also the ways of associations (subject to laws). In the light of these considerations we think it wrong to depict Frege as a forerunner of the Fodorian hypothesis of the Language of Thought. (For an interpretation along these lines see, e.g. Malzkorn (2001)). The reason is that in Frege images/ideas/representations seem to be confined to play a “supporting” role in thinking. The idea that they should be taken as real vehicles seems extraneous to Frege’s theoretical framework. I am indebted to Eva Picardi for this remark. 10. Another revealing passage can be found, for example, in Frege’s Review of Husserl’s Philosophy of Arithmetic. Here Frege says: “One and the same thought can be grasped by many men. The constituents of the thought, and a fortiori things themselves, must be distinguished from the ideas that accompany in some mind the act of grasping thought” (Frege, 1894, 318). 11. Of course it did not fit with all of them. In the next section we shall consider two respects under which the representational model conflicts with Frege’s theory. Here we want to call attention to another one to which we shall not come back later, namely the idea that Representationalism conflicts with Frege’s requirement of the objectivity of thoughts. How

82

tunately, Frege did not realise that that very model was a real menace for his own theory of thoughts.12 2. The cognitive philosopher’s provocation It will be the protagonists of the so called “cognitive turn” in philosophy who, aware of this unbearable tension, will remove it by getting rid of the idea of the non-mental character of the contents of thought. People such as Fodor, will challenge Frege by claiming that there is no point in devising a theory of thought if one does not work out, at the same time, a theory of thinking capable of explaining how thoughts can actually be grasped. Unfortunately, Fodor claims, Frege does not provide us with such a theory. As Fodor says commenting on this point in his 1978 paper on Propositional Attitudes (in which, incidentally he uses the word ‘proposition’ to mean roughly what Frege meant by thought): I don’t see how an organism can stand in an (interesting epistemic) relation to a proposition except by standing in a (causal/functional) relation to some token of a formula which expresses the proposition […] Frege says that one apprehends (what I am calling) propositions, but I can find no doctrine about what apprehension comes to beyond the remark (in The Thought) that it is not sense perception because its objects are abstract and it’s not introspection because its objects aren’t mental […]. As for me, I want a mechanism for the relation between organism and propositions, and the only one I can think of is mediation by internal representations. (Fodor, 1978, 520)13 could one grant that thoughts are objective if that through which they are grasped varies from one thinker to the other? A possible rejoinder on the part of a supporter of the representational model is to claim that the similarity of our respective cognitive equipments is enough for the satisfaction of the objectivity requirement. It goes without saying that this kind of reply would not have pleased Frege! 12. On this point our (exegetical) position differs from that of other commentators. In a sense, the interpretation we have suggested is neither new nor rare. Malzkorn (2001), for example, patently endorses the “representationalist line” according to which what aims at the thought in the thinker’s consciousness is the thinker’s idea of the thought. But while he claims that this interpretation offers the remarkable opportunity of removing the obscurity which affects the relation of grasping thoughts (by reducing it to two other relations — namely: (i) the relation of having (ideas) and (ii) the relation of being the content […] of (an idea) — neither of which seems to be as obscure as the former), we believe instead that this interpretation introduces a strong tension within Frege’s theory which risks to overthrow some of its main tenets, in particular its anti-mentalism, and the idea that it is via senses that anything can be given to us. 13. Fodor is here granting Frege two points. First, that the thought is the immediate relatum

83

Several points need to be made here. First: Fodor points out that we need a theory of thinking which explains how subjects can stand in relation to thoughts (propositions in his terminology). Second: It is very likely, in Fodor’s lights, that such a theory should make use of mental representations. Fodor’s reasons for this are the following. We use propositions/ thoughts to identify the objects of the attitudes. Propositional attitudes have two main properties: they are causally efficacious and semantically evaluable. One way (the best one in Fodor’s view) to account for these two properties is in terms of mental representations. Mental representations, as real inner states of an organism, can account for propositional attitudes’ causal efficacy and, as elements which represent things and facts, can account for semantic evaluability. Third point: the best theory of thinking which accomodates the requirements stated is the “Representational Theory of the Mind” (RTM).14 This theory has two parts, one concerning thinking and the other concerning thought. According to the first part (the Representational Theory of Thinking) thinking, and mental processes in general, consist of causal sequences of tokenings of mental representations. According to the second part (the Representational Theory of Thought) to have a propositional attitude A towards the proposition that P is to stand in a particular psychological relation R to a mental representation M which means that P.15 One could say that Fodor, with his computational-representational theory of the mind, fills in the gap which Frege left open. Moreover, his is a version of Representationalism and, on the ground of what we said, it could prove congenial to Frege.16 Can one conclude on this ground that, of the act of thinking and, second, that the relation in question qualifies as “epistemic”. Fodor’s conclusion depends strongly on these two points. Of course, if one rejects one or both of them, one could question the conclusion he draws. 14. See, e.g. Fodor (1987). 15. As it is known, Fodor’s RTM is one out of three theoretical ingredients which make up the peculiar version of Representationalism which he endorses. The other two ingredients are: “The language of thoughts Hypothesis” and the “Computational Theory of the mind”. According to the former, mental representations belong to a symbolic system which is such that the representations of the system have a combinatorial syntax and semantics. According to the latter the operations on representations (which constitute the domain of mental processes) are causally sensitive only to the syntactic/formal structure of the representations defined by this combinatorial syntax. What completes the picture is Fodor’s adherence to a functionalist version of materialism according to which mental representations are functionally characterizable entities realized by the physical properties of the cognitive subject. 16. For, up to this point, Fodor seems compatible with Frege in so far as the distinction between mental representations and contents (thoughts/propositions) is preserved. But, as we

84

by irony of fate, Fodor provides a vindication of Frege’s gappy remarks on thinking?17 Unfortunately for the Fregean the answer is negative. If thinking is as the RCTM (Representational Computational Theory of the Mind) depicts it, then thoughts and their constituents, the concepts, must be mental. Not only so. If it is true, as Fodor (and the Representationalist in general) claims, that what we are immediately related to in thinking are mental representations, i.e. that the mental representations are what mediate our relation to whatever can be given to us (and therefore also to contents/thoughts/propositions — or whatever your favourite label for this kind of abstract entity is), it follows — given Fodor’s idea that mental representations partly individuate the contents of thought — that to introduce Fregean senses is to introduce unnecessary circumvolutions devoid of any explanatory role. Our point, to put it in a nutshell is this: if one believes that there must be something in the mind which mediates our access to thoughts/senses, then along this line the theoretical necessity of thoughts/ senses decays. Their dispensability becomes an unavoidable step once one takes them to be mere intermediaries devoid of any real explanatory role. This conclusion is explicitly drawn by Fodor in his book Concepts in which he tries to challenge the Fregean on his own ground, by showing that we do not need senses, conceived à la Frege, not even to account for cognitive value.18 In Fodor’s view, the Fregean stance commits itself to the following three theses about modes of presentation (MOPs): 1. MOPs are senses; for an expression to mean what it does is for the expression to have the MOP that it does. 2. Since MOPs can distinguish concepts, they explain how it is possible to entertain one, but not the other, of two shall see below, this is not so. The reason is that, according to Fodor, contents of thought are partly individuated by the “vehicles” of their constituents (the concepts), i.e. by mental representations. The idea that what is mental enters into the individuation conditions of what is “logical” is of course incompatible with the general antipsychologistic orientation of Frege’s theorizing. 17. What grounds this supposition is not so much the fact that Fodor provides a theory of thinking, nor that the theory he provides is probably the best developed version of Representationalism, as the fact that his theory seems expressly designated to address the issue of grasping contents conceived as abstract entities (that this is so comes out neatly from the passage we quoted from Fodor (1978)). From this to the conclusion that Fodor really vindicates Frege there is a big step. We shall not endorse that conclusion. What we shall claim is that, despite appearances, Fodor does not ultimately address Frege’s problem because that problem is not suited to be addressed by the “mechanism talk” in which Fodor’s answer is framed. 18. I have dealt with this topic in Sacchi (2003) to which I send the reader for a more detailed discussion.

85

co-referential concepts; […] 3. MOPs are abstract objects; hence they are non-mental. (Fodor, 1998, 15–16)

Fodor’s claim is that the conjunction of these three theses cannot be satisfied, since in his view 2 — which is the thesis that anyone who is sympathetic to the Fregean program has to maintain — excludes both 1 and 3, and thus we had better reject the latter ones. The argument he presents for the inconsistency of the claims 2 and 3 (which is the one which is more relevant for our present purpose) is the following. It is apparent, he says, that MOPs can distinguish concepts provided they can individuate them. In turn, in order for MOPs to individuate concepts they must stand to concepts in a one-to-one relation and this requires that for any MOP there be exactly one way to grasp it, otherwise each way of grasping the MOP would correspond to a different way of thinking of the referent, and consequently to a different concept of the referent. But what can legitimate the assumption that there is only one way of thinking of a referent corresponding to each mode of presentation of the referent? Fodor’s answer is that, unless we are prepared to abandon Frege’s antipsychologistic claim, nothing but sheer stipulation can legitimate that assumption. Frege’s architecture, according to Fodor, has thus the following problem: it cannot prevent different mental particulars from constituting graspings of the same MOP. So, for example, both a water mental particular and an H2O mental particular may constitute graspings of the water MOP. But in this case, Fodor claims, it follows that Frege cannot use senses to distinguish de dicto water beliefs from de dicto H2O beliefs and so cannot explain potential differences in behaviour based on these different beliefs. Fodor’s conclusion is that if one wants to pursue Frege’s program of individuating contents in terms of MOPs then one has to reject Frege’s claim that MOPs are abstract objects and to conceive of them as mental particulars instead. To put it in Fodor’s words: The Frege programme needs something that is both in the head and of the right kind to distinguish coreferential concepts, and the Mates cases suggest that whatever is able to distinguish coreferential concept is apt for syntactic individuation. Put all this together and it does rather suggest that modes of presentation are syntactically structured mental particulars. (Fodor, 1998, 38)19 19. “Mates cases” are the ones that Fodor discusses in order to show the incompatibility of claim 1 (MOPs are senses) and 2 (MOPs can distinguish concepts). The of argument he presents in support of the claim that 2 excludes 1 is the following: a) MOPs are whatever passes Frege’s

86

Within Fodor’s picture there is no work left for senses to do; senses become utterly useless: mere intermediaries devoid of any explanatory role. The extent of Fodor’s critique is remarkable. His aim is to show that nothing can do what Frege’s senses are supposed to do, i.e.: (1) to determine meaning; (2) to account for differences in cognitive value; and (3) to individuate thoughts. As far as point 1 is concerned, he adopts an informational account according to which what bestows meaning on mental representations is something about their causal-cum-nomological relations to the things that fall under them. Since meaning is information within this picture, coextensive concepts turn out to be synonymous. In turn, the conjunction of “Informational Semantics” with “Logical Atomism” — the metaphysical picture of concepts that Fodor endorses, according to which (primitive) concepts are unstructured atoms — prevents Fodor from providing an inferential role account of what distinguishes coextensive concepts. His suggestion is thus to distinguish between meaning individuation and concept individuation. The result is that concepts (and therefore the contents they make up) are only partly individuated by their meaning. What determines them fully is a meaning together with a mental representation (a MOP). To sum up, there is no single notion which plays the role of a sense: role 1 is played by causal-cum-nomological relations; role 2 is played by mental representations; and role 3 is played by mental representations together with their meanings. Is it thus necessary, as Fodor’s train of thought seems to show, to give up on the non mental nature of contents in order to provide an account “substitution test” — two concepts are identical if and only if it is not possible to think either without thinking the other; b) Senses do not to pass Frege’s substitution test; c) senses cannot be MOPs. Fodor’s evidence in support of point b) is the following: since senses are what fixes the meaning (the semantic value) of expressions (for thesis 1), senses must be what synonyms have in common. So, if senses were MOPs — whatever the “substitution test” tests for — it shouldn’t be possible for someone to think one of two synonymous concepts without thinking the other. And yet this is possible. To illustrate this point he presents two kinds of example: the “bachelor/unmarried man” one and the “Jackson/Pollock” one. The former (which he adapts from Mates (1962)) goes like this: since ‘bachelor’ and ‘unmarried man’ are synonyms, they must be identical and it ought not to be possible to think either without thinking the other. And yet it is possible for someone (Fred) to wonder whether John understands that bachelors are unmarried men even though Fred does not wonder whether John understands that unmarried men are unmarried men. The latter goes like this: suppose you were told that Jackson was a painter and that Pollock was a painter and that you believe this. It looks like, he says, that fixes the senses of the names as both having the same sense — viz. a painter. And yet it is perfectly conceptually coherent for you to wonder whether Jackson and Pollock were the same painter (even though you do not wonder whether Jackson was Jackson or Pollock was Pollock).

87

of the grasping relation? Pace Fodor, we think that this conclusion can be resisted. In our view, what one has to give up instead is rather Frege’s pernicious mixture of Cartesianism cum Platonism. For, if these two assumptions are in place it is difficult, as we have tried to show, to avoid the “representationalist hypothesis”. But Representationalism, once in place, and fully developed, makes the theoretical role of the notion of sense dispensable or purely additional. 3. Thoughts and their metaphysics The idea that the mental representations are what mediate our relation to senses clashes with Frege’s idea that it is through senses that anything can be given to us. If it is true, as Frege claims, that anything is given to us via a sense, then, and on pain of a vicious regress, there should be nothing which mediates our relation to senses,20 which amounts to saying that senses should turn out to be the only objects which are immediately and unproblematically given to us.21 To put it in other words, we can say that given the role that senses play in Frege’s theory, the possibility for senses/ 20. What we claim is that even though referring to a sense (in indirect speech for example) may proceed via a sense, grasping/apprehending a sense should turn out to be as immediate and direct as possible. Otherwise the account would be circular. The risk of circularity could be avoided by conceiving of senses as a special kind of objects, namely: as the only kind of objects which are given in an immediate and not “perspectival” way (in this case the distinction between what is given and the way in which it is given disappears). In suggesting this particular way out from the risk of circularity we differ from other authors. Bell (1987) for example, claims that what generates the problem is the conception of thoughts as objects. In his words “a vicious infinity of such acts of grasping is generated immediately if we maintain, with Frege, that this sense is, in its turn, merely an object we have in mind; for in this case the sense would likewise have to be grasped via the sense of some other expression, which being itself an object, would have to be grasped via the sense of some expression […] and so on” (1987, 46–47). His proposal is to give up on the transitive model of thinking in favour of an intransitive model in which the thought no more figures as the object of the act but as an act-modifier. We disagree with Bell on this point. In our view there is no need to abandon the conception of thoughts as objects in order to acknowledge their peculiarity (thoughts are objects even though they are not the immediate objects of thought). Frege’s mistake in our view was not so much in believing that thoughts were objects (as Bell claims), nor was it in believing that all objects are self-subsistent and independent (as Dummett, 1991, does), as in believing that the (ontological) dependence on the part of an entity goes hand in hand with its subjectivity. I have dealt with this topic in Sacchi (2005) in particular chp. 3. 21. Not in the sense that they are immediate objects of thought, but in the sense that they are objects of acquaintance. This is in our view the distinguishing mark of the sense.

88

thoughts to be given (to be epistemically accessed) should be warranted by the very theoretical role that the notion plays. If it is not so, if the graspability of thoughts turns out to be a problem, this means that one has gone astray in delineating the metaphysics of thoughts/senses. In our view, this is precisely what happened with the late Frege. His Platonism, by isolating thoughts from both the mind and the world threatens the very possibility for thoughts to be grasped. A better metaphysics should avoid turning what is proper to thoughts, namely their being thought, into a problem.22 The possibility for thoughts/senses to be given turns out to be warranted within a given picture only in so far as they figure as ways of giving of something. But in order for thoughts to so figure, they have to be characterized in our view not as self-subsistent entities, as Frege did with his mythology of the Third Realm, but as ontologically dependent entities.23 Let us expand a little bit on this point. As it has been said, what jeopardises the possibility for thoughts to be grasped is, in our view, a certain Platonic image of their nature which, together with a Cartesian conception of the thinking subject, concurs to convey the image of the mind as “out of touch” with its contents. The proposal we would like to suggest amounts to giving up on the very idea of thoughts as self-subsistent and independent entities (which, in our view, makes the issue at stake so intractable) and to conceive of them as intimately related, by their very nature, to both the mind and the world, where this relatedness serves to account (respectively) for the graspability and the directedness of thoughts. This double dependence faithfully mirrors, in our view, the peculiar “hybrid nature” of thoughts, the fact that a thought looks like a sort of two-faced Janus with one face interfacing with the mind and the other with the world. We take it to be one of the most important (and hardest) tasks of the theorist 22. Of course, one could claim that what is wrong is not so much the kind of metaphysics suggested but, in the first place, the very idea that one needs a metaphysics in connection with thoughts. This attitude, which is congenial to the neo-Fregean approach in the theory of thought, goes hand in hand with the idea that grasping a thought (in the dispositional sense, at least) is a complex ability which manifests itself in the overt disposition, on the part of the subject, to engage in several inferential practices implying the concepts which constitute the given thought. We have nothing to object to this theoretical framework (which we by and large accept), but it leaves utterly in the dark the ontological status of thoughts. The idea that engaging in such an enquiry is, by itself, wrong because it promotes a misleading picture of thoughts as queer entities is the result of an unmotivated bias. 23. Our rationale for this point is that what grounds the thought’s ability to direct to something is the internal relation of ontological dependence between it and the object it is about. For more on this point see Sacchi (2005) in particular 3.5.

89

of thought to account for this intrinsic complexity in a way that does not result in a “schizophrenic” conception of thoughts. Incidentally, this is precisely what in our view happens with so-called “dual” or “two-factors” theories of content which are so popular nowadays.24 By contrast, ours is not a schizofrenic conception of thoughts. The notion of thought in our picture is the notion of a unitary entity which happens to exhibit a double order of ontological dependence relations. But what does ontological dependence come to? Ontological dependence is a technical and complex topic which has been introduced in the modern debate by Husserl (with his work on the notion of “foundation” in the Third Logical Investigation) and his pupil Ingarden, and which has been brought to the attention of the analytical community by the works of Mulligan, Simons and Smith.25 Since it is not our aim here to enter specifically into this proposal, we shall skip all technicalities and try to be as intuitive as possible. A metaphor which could prove useful in order to understand the notion at play is that of “unsaturatedness” which Frege uses in connection with concepts. Ontological dependence can be taken as the analogue at the ontological level of the unsaturatedness of concepts at the logical level. In other words, ontological dependence is a “need of completion” in order to exist. But what kind of completion are thoughts in need of exactly? Our tentative answer is the following: on the one hand, thoughts need the external world (in a wide sense which includes not only the physical but also the social environment in which our cognitive/conceptual/linguistic practices take place) and, on the other hand, they need creatures having a certain kind of cognitive equipment. Within this picture, thoughts are the product of our higher order cognitive interactions with the world, no more no less than our sensations are the products of our sensualistic interactions with the world. As a particular sensation could not exist if certain sentient beings capable of reacting in a given way to it did not exist, so thoughts in general would not exist if creatures endowed with representational/inferential abilities did not exist. Thoughts, as sensations, are a product of our “subjectivity”, that is of the peculiar way of “being in the world” which is peculiar of us. From this it does not follow that the 24. See, e.g. McGinn (1982). 25. Nowadays there are three main theoretical approaches to ontological dependence: (i) the Modal-Existential Approach (Mulligan, Smith (1986); Simons (1982), (1987); Mulligan, Simons, Smith (1984)); (ii) the purely Essentialist Approach (Fine (1995); Lowe (1994)); (iii) the Essentialist-Existential Approach; to which a fourth has recently been added, namely the “Foundational” one. On this last approach see, e.g. Correia (2005).

90

way of being of thoughts is identical to that of sensations. The idea that whatever is a product of our subjectivity is for this very reason subjective (in the pejorative sense of being epistemically private, incommunicable and so on) represents, in our view, a radical mistake which pervades most analytical philosophy. Let us briefly sketch the intuitive ground of our proposal. Let us consider a very simple thought such as for example the one that could be expressed by uttering the sentence ‘This tree is covered with green leaves’ in a given context. We believe it quite sensible to claim that if there were no tree in the contextually relevant place, then that particular thought would not exist. That there be a tree in the contextually relevant place is a necessary condition for the existence of the thought. But this condition is not sufficient in itself. The relevant tree could exist in a possible world devoid of any thinking creatures and, in such a circumstance, one could hardly maintain, in our view, that the thought would be available. Up to this point the two kinds of dependence — thought-mind, thought-world — seem to be of the same kind, each one representing a necessary but not sufficient condition for the existence of a given thought. Notwithstanding this similarity, we claim that the two kinds of relation are different. To illustrate this point let us come back again to our perceptual demonstrative thought. The thought in question, even though it would not exist if no thinking creature existed, does not change in shifting from one thinking creature to another. In this sense we claim that thoughts, while depending on the mental for their existence do not depend on them for their identity (i.e. for their being what they are).26 Things are different with the other kind of dependence relation. In this case, the thought depends on the world (namely: on the things and properties in the world it is about) not only for its existence, but also for its identity. If the demonstrated tree got replaced by a different (although qualitative identical) one, the thought would be about a different object and therefore, according to this picture which qualifies as externalist, it would be a different thought. To be about that particular tree is essential for the thought’s being the particular thought it is. What we claim is that this double dependence which we have presented in connection with a particular kind of thought could and should be extended to all thoughts. Moreover, we claim that the difference we have tried to 26. We take identity-dependence as a form of ontological dependence. As for the relation between the the two senses of the ontological dependence relation we claim that the existential sense is implied by the individuative sense but not vice versa. We have touched on this topic in the context of a general discussion of externalism in Sacchi (2006).

91

highlight between the way in which thoughts depend on the mind and the world (by means of the dichotomy “existential dependence” vs “identity dependence”) provides a means to safeguard a moderate anti-psychologism in theory of thought. What comes out is the idea that it is possible to account for grasping not, as Fodor does, by telling a story about the mechanism capable of relating the subject’s brain to a thought, but by providing a (metaphysical) picture of thought which has the graspability of thoughts as one of its corollaries.27 But to abandon the mechanism talk requires not only the rejection of Platonism, but also of Cartesianism, even that of the materialistic brand which is so popular nowadays among cognitive philosophers.28 For with Cartesianism in place it is difficult to resist the idea that to account for how thoughts can be grasped, or, for that matter, how things can be reached in perception, is to account for how the gulf between the inner world and the outer world can be bridged. A better story would have to rethink the role of subjectivity in such a way as to avoid the idea of a radical separation between mind and world. Within this non Cartesian picture it should turn out that if it is true that the world is given to us in thought, then what comes to view in this giving is not only the world but also that 27. Another way in which the issue of grasping could be dealt with is to treat thoughts on a par with other abstract objects and apply to them a variant of a general position on abstract objects which emerges from certain passages in Frege’s Grundlagen. Let us consider for example the following biconditionals: (i) the direction of a is identical with the direction of b iff a is parallel to b; (ii) the number of Fs is identical with the number of Gs iff there is a one-one mapping from the Fs to the Gs. These biconditionals are taken to show that the ontologies of directions and numbers are not problematic. Moreover, they seem to provide a plausible answer to the issue of grasping: to grasp a given number or a given direction, for example, is to understand the senses of sentences in which terms for them occur. The question now is whether this position can be extended to senses. In a sense it cannot. For, if one were to claim on the ground of the following biconditional (iii) the sense of ‘A’ is identical with the sense of ‘B’ iff ‘A’ is equivalent to ‘B’ that to grasp a thought/sense is to understand the senses of sentences in which terms for them occur, then a circularity or infinite regress would immediately arise. But if one abandons the idea of providing an analysis of the “incriminated” notion and opts for the idea of a quasi-analysis (in the sense of Carnap), one could claim that (iii) can be replaced by (iii*) the sense of ‘A’ is identical with the sense of ‘B’ iff the expression ‘A’ can be replaced salva veritate in all contexts of type C with the expression ‘B’. And now, the circularity can be avoided by claiming that to grasp the sense (thought) is to understand the relevant equivalence relation. I am indebted to M. Beaney for these comments. 28. We do not want to claim that RCTM as a theory of cognitive processes is on the wrong track. What we claim is that it cannot constitute an answer to the issue of grasping thoughts and that to think otherwise is a mistake.

92

through which the world is given. And this is precisely the thought which is in view for us only in so far as it puts the world in view.

REFERENCES Bell, D. 1987. “Thoughts”. Notre Dame Journal of Formal Logic 28, 1, 36–50. Carl, W. 1994. Frege’s Theory of Sense and Reference. Cambridge: Cambridge University Press. Correia, F. 2005. Existential Dependence. München: Philosophia Verlag. Dummett, M. 1981. The Interpretation of Frege’s Philosophy. London: Duckworth. — 1988. “The Origins of Analytical Philosophy”. Lingua e Stile 23 (1 and 2), 3–49 and 171–210. — 1991. Frege and Other Philosophers. Oxford: Clarendon Press. Engel, P. 1996. Philosophie et psychologie. Paris: Gallimard. Evans, G. 1982. The Varieties of Reference. Oxford: OUP. Field, H. 1981. “Mental Representation”. In: N. Block, ed. Readings in Philosophy of Psychology. Cambridge Mass.: Harvard University Press, 78–114. Fine, K. 1995. “Ontological Dependence”. Proceedings of the Aristotelian Society 45, 269–290. Fodor, J. 1978. “Propositional Attitudes”. The Monist 61, 501–523. Reprinted in J. Fodor 1981. Representations. Cambridge Mass.: MIT Press. — 1987. Psychosemantics: The Problem of Meaning in the Philosophy of Mind. Cambridge Mass.: MIT Press. — 1998. Concepts. Where Cognitive Science Went Wrong. Oxford: Oxford University Press. Frege, G. 1884. Die Grundlagen der Arithmetik. Transl. in: J. L. Austin ed., 1980. The Foundations of Arithmetic. Oxford: Basil Blackwell. — 1892. “Über Sinn und Bedeutung”. Transl. in: P. Geach and M. Black, eds. 1966. Translations from the Philosophical Writings of Gottlob Frege. Oxford: Basil Blackwell. — 1918–1919. “Der Gedanke. Eine logische Untersuchung”. Transl. in: P. Strawson, ed. 1967. Philosophical Logic. Oxford: Oxford University Press, 17–38 and transl. in: M. Beany ed. 1997. The Frege Reader. Oxford: Blackwell, 325– 345. — 1969. Nachgelassene Schriften. Hamburg: Felix Steiner Verlag. Part. transl. in: M. Beany ed. 1997. The Frege Reader, 227–250. Grimm, R. H., Merill, D. D. eds. 1988. Contents of Thought. Tucson: The University of Arizona Press.

93

Lowe, J. 1994. “Ontological Dependency”. Philosophical Papers 23, 31–48. Luntley, M. 1999. Contemporary Philosophy of Thought, Oxford: Blackwell. Malzkorn, W. 2001. “How Do We ‘Grasp’ a Thought, Mr. Frege?” In: A. Newen, U. Nortmann, R. Stuhlmann-Laeisz, eds. 2001. Building on Frege: New Essays about Sense, Content, and Concepts, Stanford: CSLI Publications. Mates, B. 1950. “Synonymity”. University of California Publications in Philosophy 25, 210–226. Repr. in: L. Linsky, ed. 1952. Semantics and the Philosophy of Language. Illinois: University of Illinois Press, 111–136. McDowell, J., 1998. “Having The World in View: Sellars, Kant and Intentionality”. The Journal of Philosophy 45, 431–490. McGinn, C., 1982. “The Structure of Content”. In: A. Woodfield, ed. Thought and Object, Oxford: Clarendon Press, 207–257. Mulligan, K., Simons, P., Smith, B. 1984. “Truth-makers”. Philosophy and Phenomenological Research 44, 287–321. Mulligan, K., Smith, B. 1986. “A Relational Theory of the Act”. Topoi 5, 115– 130. Picardi, E. 1996. “Frege’s Anti-Psychologism”. In: M. Schirn, ed. Frege: Importance and Legacy. Berlin, New York: De Gruyter, 307–329. Sacchi, E. (and Coliva, A.) 2001. Singular Thoughts: Demonstrative Thoughts and I-Thoughts. Macerata: Quodlibet. Sacchi, E. 2003. “Fodor e la psicologizzazione dei sensi fregeani”. In: M. Carrara, G. De Anna, S. Magrin, eds. Linguaggio, mente e mondo: Saggi di filosofia del linguaggio, filosofia della mente e metafisica. Padova: Il Poligrafo, 81–119. — 2004. “I pensieri e il ‘Regno di Mezzo’”. Rivista di Estetica 26, 239–255. — 2005. Pensieri e rappresentazioni mentali: Frege e il cognitivismo contemporaneo. Roma: Carocci. — 2006. Cognitive Externalism and the World-Involving Character of Thoughts. Bologna: Department of Philosophy and CLUEB, (preprint 28). Schirn, M. ed. 1996. Frege, Importance and Legacy, Berlin-New York: De Gruyter. Simons, P., 1982. “The Formalisation of Husserl’s Theory of Wholes and Parts”. In: B. Smith, ed. Parts and Moments. Munich: Philosophia, 113–159. — 1987. Parts, A Study in Ontology. Oxford: Oxford University Press. Stuhlmann-Laeisz, R. 1995. Gottlob Frege “Logische Untersuchungen”. Darstellung und Interpretation. Darmstadt: Wissenschaftliche Buchgesellschaft. Weiner, J. 1999. Frege. Oxford: Oxford University Press.

94

Grazer Philosophische Studien 72 (2006), 95–110.

NUMBERS, REFERENCE AND RUSSELLIAN PROPOSITIONS Pierdaniele GIARETTA University of Verona

Summary Stewart Shapiro and John Myhill tried to reproduce some features of the intuitionistic mathematics within certain formal intensional theories of classical mathematics. Basically they introduced a knowledge operator and restricted the ways of referring to numbers and to finite hereditary sets. The restrictions are very interesting, both because they allow us to keep substitutivity of identicals notwithstanding the presence of an epistemic operator and, especially, because such restrictions allow us to see, by contrast, which ways of reference are not compatible with the simultaneous maintenance of substitutivity of identicals and the classical notions of truth and knowledge. In this paper the difference between the restricted and the unrestricted kind of reference is put in relation with Russell’s ideas on naming and it is argued that the latter as well is compatible with a certain Russellian conception of the understanding of sentences. Then it is discussed whether and how numbers could be conceived as objects of acquaintance. Finally a general question about the notion of logical form is raised.

Stewart Shapiro and John Myhill tried to reproduce some features of the intuitionistic mathematics within certain formal intensional theories of mathematics. Basically they introduced a knowledge operator and restricted the ways of referring to numbers and to finite hereditary sets. The restrictions are very interesting, both because they allow us to keep substitutivity of identicals notwithstanding the presence of an epistemic operator and, especially, because such restrictions seem to correspond to the intuitive requirements of the ideal accessibility of the objects referred to and of their perspicuous representability by means of the terms satisfying the restrictive conditions. It is quite natural to connect such requirements with Russell’s ideas

on naming. In this connection an often neglected aspect of Russell’s view is here emphasised. The paper turns then to examine the relationship of such requirements with Russell’s well known way of conceiving sentence understanding. It is argued that perspicuous representability is not implied by the main principles which are usually taken as characteristic of that conception. This might suggest that perspicuous representability is not a requirement which needs to be satisfied when a Russellian proposition is expressed. The other requirement, i. e. the accessibility of what we are speaking of, seems to be a weakening of the well known constraint of acquaintance with the constituents of a proposition. However, it is quite difficult to specify the sense and the role of the ideal accessibility of numbers, of hereditary sets and, in general, of mathematical entities, at least when such accessibility is understood as a sort of potential Russellian acquaintance. Some remarks are made on this problem. Finally a general question about the notion of logical form is raised. A Russellian proposition has a form which — as we shall argue — cannot be conceived as the most detailed representation of the positions of the constituents of the proposition, at least if the notion of position in a proposition is thought of by analogy with that of position in a sentence. The paper will not state any definite general thesis. Its main purpose is to show some interrelations among some notions and to sketch some analyses which can be useful for a better understanding of the nature of reference and propositions, especially when mathematical entities are involved and logical aspects are taken into account. 1. Knowledge and substitutivity of identicals Following Shapiro (1985) let us identify substitution of identicals with the principle: SI x = y o(I(x) lI(y)) It follows that, if “b” and “c” are two terms, then, if “b = c” is true, “b” and “c” are intersubstitutable. However, that does not hold in general. Indeed, using Shapiro’s example, “the-number-of-planets = 3 u3” and “Hegel knew that 3 u3 = 9” are both true, yet one cannot state that Hegel knew that the-number-of-planets = 9. It is natural to conclude that, if K is an epis-

96

temic operator, substitution of identicals cannot be allowed within the scope of K. It is important to observe that this failure does not depend on assuming that K represents explicit aware knowledge. The principle fails even if knowledge is conceived as the knowledge that can be achieved on the basis of available information. Indeed, with reference to the above example, we could imagine Hegel knew that there is a finite number of planets but could not know which one it is, because he lacked some relevant pieces of empirical information. Such a situation cannot occur with respect to the usual number theory, since all arithmetical terms — i.e. the terms built by means of the usual arithmetic operators —are ideally computable, in the sense that the value of each term is recursively computable and representable by a specific numeral. Indeed, to reproduce the failure of substitutivity of identicals in a number theory provided with the ideal knowledge operator K, Shapiro extends the usual theory and adds axioms for introducing functions such as1 the least w such that A(v,w) is true, if there is such a w f(v) = 0 otherwise Interestingly, when terms, abbreviated as [A]v, for such functions are introduced,2 substitutivity of identicals fails in an even more dramatic way, since it allows deriving that for every closed formula A A o K(A). Shapiro’s proof 3 is based on the special features of the terms abbreviated as [B]v, where B is of the form (x = x) š (y = 1) š A. It proceeds as follows: 1. A 2. [B]0 = 1

premise from axiom F3 4

1. Shapiro (1985, 34-35). 2. The meaning of the terms [A]v is implicitly specified by the axioms F3 having the following form: [xyA(x,y)]v = w l (A(v,w) š z(A(v,z) o z ≥ w)) › (w = 0 š z(¬A(v,z))) 3. Shapiro (1985, 35). 4. See note 2.

97

3. K(1 = 1) lK([B]0 = 1) 4. K([B]0 = 1) 5. K(A)

unrestricted SI from 3 and K(1 = 1) from axiom F3 5

Myhill (1985) obtains the same result within a set theory supplemented with an epistemic operator B having features similar to K’s. Let us keep using K, instead of B, and mention only some of the features of his system. Myhill’s system has two kinds of variables: small ones, ranging over explicitly listed hereditary finite sets (including explicitly given non-negative integers) and large ones ranging over all sets (however given).6 Correspondingly there are: 1) terms of the first kind, including small variables, 0, all expressions of the form {t l }, (t l ‰ t 2) and ft l , where t l and t 2 are terms of the first kind and f is a recursive function symbol; 2) terms of the second kind, among which we have large variables and all expressions of the form {X |I(X )}, where I(X ) is a formula not containing K. Atomic formulas are expressions of the form (t l = t 2) or (t l  t 2), where t l and t 2 are terms of either kind. The axioms for hereditary finite sets are meant to describe the “construction” of such sets from the empty set by means of the singleton and the union operations. Myhill’s system includes, or allows deriving, all usual axioms of set theory. Let us mention only the axioms for identity, using D, E, J to represent variables of either kind: 1. D = E, D = E o E = D, D = E š E = J o D = J 5. [xy(x = x š y= 1 š A)]v = w l (v = v š w = 1 š A š z(v = v š z = 1 š A o z ≥ w)) › (w = 0 š z ¬(v = v š z = 1 š A)) is an F3 axiom, whence [xy(x = x š y= 1 š A)]0 = 1 l (0 = 0 š 1 = 1 š A š z(0 = 0 š z = 1 š A o z t 1)) › (1 = 0 š z ™(0 =0 š z = 1 š A)). So we have K([xy(x = x š y = 1 š A)]0 = 1) l (0 = 0 š 1 = 1 š A š z (0 = 0 š z = 1 š A o z t 1)) › (1 = 0 š z ™(0 = 0 š z = 1 š A))), then K([xy(x = x š y = 1 š A)]0 = 1 l K((0 = 0 š 1 = 1 š A š z (0 = 0 š z = 1 š A o z t 1)) › (1 = 0 š z ™(0 = 0 š z = 1 š A))) by the deductive closure of K. 4 is K([xy (x = x š y = 1 š A)]0 = 1), so we have K((0 = 0 š 1 = 1 š A š z (0 = 0 š z = 1 š A o z t 1)) › (1 = 0 š z ™(0 = 0 š z = 1 š A))) hence, again by the deductive closure of K, K(A). 6. Myhill (1985, 48).

98

2. I(D) š D = E o I(E), 3. I(x) š x = y o I(y) 4. I(D) š K(D = E) o I(E)

where I contains no K. for arbitrary I. for arbitrary I.

Myhill observes that certain strengthenings allow to prove A \ l K\ for every sentence \. One of the strengthenings consists in eliminating the restriction to formulas containing no K in 2. That makes it possible to exploit the peculiar feature of the term {XZ|\}, where Z represents the non negative integers. This term is such that: Z if \ {XZ|\} = 0 if ¬\ The proof proceeds as follows. If the restriction to formulas containing no K in 2 is eliminated, it is possible to put K(X = Z ) for I(Z ) and to get: A X = Z š K(X = X ) o K(X = Z )

(1)

whence it follows A X = Z o K(X = Z )

(2)

Then, replacing X in (2) by {XZ|\} and Z by Z, where \ is any sentence not containing K: A {XZ|\} = Z o K({XZ|\} = Z)

(3)

Since A \ l {XZ|\} = Z

(4)

99

we also have A \ l K({XZ|\} = Z)

(5)7

From 4, by the rule A I / A KI, we have A K(\ l {XZ|\} = Z)

(6)

and by logical closure of K A K\ l K({XZ|\} = Z)

(7)

So, from (5) and (7) A \ l K\ 2. Description and provability Shapiro’s and Myhill’s proofs share three important features. Both appear to depend on the composition of certain descriptive terms, on the logical closure of the K operator and on the application of unrestricted substitutivity of identicals. For each of these three features, a question naturally arises about its indispensability. However, logical closure is not a problem, since we state what an ideal knower can know on the basis of available information. So let us focus on the other two questions. Concerning the use of descriptive terms, we consider one of Myhill’s remarks very relevant, even if he did not make it in relation to the above proof. The full explicit formulation of this remark also shows the indispensable role of the substitutivity of identicals. Myhill observes that the theorem on eliminability of definite descriptions does not hold for his theory containing the K-operator. According to this theorem, if an existence and uniqueness condition (x)(!y) I(x,y)

7. (3) and (4) imply \ o K({XZ|\} = Z). K({XZ|\} = Z) o \ follows from (4) and K({XZ|\} = Z) o {XZ|\} = Z.

100

is provable, a constant f can be added to the language and the axiom I(x,fx) can be assumed, without increasing the deductive power of the system as concerns the formulas not containing f. That is usually formulated by saying that the enlarged system is a conservative extension of the original one. Myhill shows that, if this theorem held for his system of set theory provided with the operator K, then the epistemic decidability of every sentence could be proved, i.e. for every I closed A KI › K™I The proof starts by observing that A (x) ((x = 0 š I) › (x = {0} š ™I))

(1)

The formula (x = 0 š I) › (x = {0} š ™I) satisfies the uniqueness condition. Its value is 0, if I, or {0}, if ¬I, and, in both cases, it corresponds to a constant recursive function even if we were unable to identify which one. Let us suppose that a term c of the first kind can be added to the language8 and the exemplification of (1) by c can be assumed as an axiom: A (c = 0 š I) › (c = {0} š ™I)

(2)

The principles of Myhill’s intensional set theory allow the following derivation: A K(c = 0) › K(c ≠ 0) A K(c = 0 o I) š K(c z 0 o ™I)

(3)9 (4)10

8. Myhill does not take the term he introduces as a term of the first kind. He does not qualify it in any way, but his proof seems to require that it be supposed of the first kind. 9. Outline of the proof (not given by Myhill): 1) starting from c = 0 šI and K(c = c) we get K(c = 0) by substitutivity axiom 3 and so K(c = 0) › K(c ≠ 0); 2) starting from c = {0} š ¬I and K(c = c) we get K(c = {0}) by the substitutivity axiom 3, then by an axiom expressing induction on the formation of the hereditary finite sets K(z = {0}) o K(z ≠ 0), so K(c ≠ 0) and K(c = 0) › K(c ≠ 0). 10. Outline of the proof of (4) (not given by Myhill): from (2), assuming c = 0, we get ¬(c = {0} š™I), so c = 0 šI, hence I; from (2), assuming c ≠ 0, we get ¬(c = 0 šI), so c = {0} š™I,

101

A KI › K™I

(5)11

Because of a specific desired feature of Myhill’s system,12 (5) cannot hold for an undecidable sentence. Independently of it, we can intuitively see that there are no sufficient informal reasons to accept (5) as valid. The classical point of view on knowledge and truth does not allow accepting it, even if the K operator is taken to express provability on the basis of all available capacities and information: certain true sentences might still be impossible to prove.13 The given formal derivation shows that the addition to the language of descriptive terms — including constants and terms built by means of functional letters — and the corresponding inclusion of the appropriate formulas among the axioms can be such that new formulas, which do not contain any one of the added symbols, can be proved as theorems. Myhill observes that the addition of the new constant to the language and the inclusion of (2) among the axioms do not satisfy the conditions stated by him for the introduction of new symbols. Indeed, recursive function symbols can be introduced, provided that, besides the usual existence and uniqueness provability conditions, the condition A (xy)(I(x,y) o KI(x,y)) is satisfied as well, where I(x,y) does not contain K. The corresponding condition for c would be: A (y)(((y = 0 šI)› (y = {0} š™I)) o K((y = 0 šI ›(y = {0} š™I))) but such condition is not forthcoming. Let us observe that c is not formally a definite description, even if intuitively it could be taken as the abbreviation of a definite description. Hence, what the counterexample formally shows is that the unwanted result can be proved without resorting to any proper definite description. The main hence ™I. The two deductions prove, respectively, c = 0 oI and c ≠ 0 o™I. So K(c = 0 o I) and K(c ≠ 0 o™I). 11. (5) follows from (3) and (4) by the logical closure of K. 12. A KI› K™I implies A I or A ™I. 13. We are aware that the appeal to an intuitive not definite notion of provability can be considered rather problematic. However, the same point can be raised with reference to a definite notion of provability.

102

difference with respect to the former derivation of \l K\ concerns the kind of substitutivity which is applied. In the former proof, unrestricted substitutivity, concerning terms of the second kind, is applied. In the latter proof, an admitted principle of substitutivity for terms of the first kind is appealed to. It allows unrestricted substitutivity of certain terms referring to hereditary finite sets. Let us notice that each one of these sets can be taken as an object which an ideal knower can become acquainted with. So they appear to be objects satisfying Russell’s requirement that one be acquainted with the named entity. However, this feature is not enough to justify the unrestricted substitutivity principle of terms of the first kind. To justify it, we need to consider the way in which the values of variables of the first kind are thought of and referred to. They are hereditary finite sets, which can be ideally listed element for element and are ideally representable in a corresponding way.14 Moreover, the terms for them — i.e. the terms of the same kind as the variables introduced for them — have computable values and these are sets which can be computationally identified and represented in a way which is fully informative about their elements. New functional symbols can be introduced to build terms of the same kind only if they have values which can be identified by the ideal knower. This is the purpose of imposing the requirement that (xy)(I(x,y) o KI(x,y)) in order to allow the introduction of a functional symbol representing I. 3. From a Russellian perspective The general idea underlying the way of conceiving the terms of the first kind is that their values should be ideally accessible and perspicuously representable. These are essentially the conditions which Russell puts on naming. Myhill takes them as features of all terms of the first kind, even of the complex ones which Russell avoids because he only admits propositional functions. We can neglect this difference and observe that ideal accessibility can be taken as a sort of potential acquaintance, while perspicuous representability corresponds to the requirement that the named entity be identifiable through its name. The latter point seems to be neglected in the literature on Russell’s 14. Of course it should not be assumed that they can all be represented within a numerable language.

103

semantics. However, Russell explicitly makes it in connection with the names of particulars. He says: To understand a name you must be acquainted with the particular of which it is a name, and you must know that it is the name of that particular.15

In general we could say — generalizing Russell’s affirmation — a word is understood only if one is acquainted with the specific entity it stands for and knows that the word stands for it. The general idea, expressed in Russell’s words, seems to be that a named entity should be accessible by acquaintance and such that, for each one of its names, it has to be recognised as the entity referred to. An alternative option would be to admit names like c and, instead, to give up substitutivity of identicals (of the first kind), at least when names introduced by description occur, so to say, with a narrow scope. Such a move does not imply the rejection of a Russellian conception of the understanding of sentences, i.e. of a conception based on the understanding of the propositions in the way indicated by the following principles:16 SEP A sentence expresses one and only one proposition, and such proposition is composed by the entities referred to by the terms composing the sentence (including predicates). UoS One understands a sentence only if one understands the proposition expressed by the sentence. PoAP One understands a proposition only if one is acquainted with the entities composing the proposition. It is possible to adopt such a conception even if one allows names introduced by means of reference fixing descriptions. Of course, when the introduction of names is so liberalised, it does not seem coherent to require that one can always identify the entity which he is referring to by means of a name. Indeed, agreement with the classical point of view about reference and truth requires giving up the need of identifying the object of reference 15. Russell (1918–9, 205). 16. We claim that these principles express a Russellian way of thinking of sentences and propositions. As often done in the literature concerning philosophical logic and philosophy of language, the use of the word “Russellian” does not involve complete faithfulness to a specific, fully developed theory stated by Russell. See, for example, how “Russellian” is used in Anderson (1989).

104

by means of the referring term; yet giving up identification by means of the referring expression is not logically incompatible with PoAP. To give up PoAP other reasons are needed. Are there any? 4. Acquaintance and numbers Let us focus on numbers and wonder whether the weaker requirement of their being objects of (potential ideal) direct knowledge by an ideal knower has any justification. The following remarks might be useful to provide the answer. First. We understand a sentence containing a term like the above mentioned c because we understand the two propositions which the sentence could express.17 From this point of view, the understanding of Russellian propositions satisfying Russell’s principle of acquaintance PoAP is, at least prima facie, the necessary basis of any understanding. To evaluate whether it is really so, we should take into account more complicated cases and argue in general for the thesis that the proposition expressed, whatever it is, can be understood only if, in a way possibly independent of its formulation, one can become acquainted with the entities composing the proposition. Second. If one takes numbers as idealised sequences of bars or as certain finite hereditary sets, it appears that what matters to the mathematical content of the arithmetical sentences are the relations among the entities which are identified with numbers. For it is by virtue of such relations that those entities can be identified with numbers.18 Indeed, acquaintance can be the basis for grasping such relations because, when numbers are taken as idealised sequences of bars or as certain finite hereditary sets, their nature is given by means of some sort of constitution relation — juxtaposition or membership understood in an intuitive sense — which justifies their relations. In this connection let us remark that, from an historical point of view, this way of conceiving numbers is not that proposed by Russell. According to him, numbers are universals and arithmetical statements 17. Here “could” is obviously to be understood in an epistemic sense. 18. It is well known that in the so called structuralist view of mathematics, which Shapiro himself contributed to in Shapiro (1997) and whose roots can be found in Benaccerraf (1965), the ontology of numbers is provided only by the relations they have to each other. As it will be clear from the following discussion, we just claim that numbers cannot be taken as known without some knowledge of the relations they have to each other.

105

assert relations among them.19 It is well known that Russell’s account of numbers as universals meets various difficulties; however, it cannot be excluded that a different account of numbers as universals might work. Let us also remark that Russell conceived acquaintance as a relation “with anything of which we are directly aware, without the intermediary of any process of inference or any knowledge of truths”20. Strangely enough, it is a way of identifying an entity which does not imply the possibility of distinguishing it from another entity one is acquainted with. In Our Knowledge of the External World he says: “To know that two shades of colour are different is knowledge about them; hence acquaintance with the two shades does not in any way necessitate the knowledge that they are different”21. It is rather doubtful that such identification without discrimination is of any use to account for our understanding of mathematical statements. Moreover, even granting that a single mathematical entity has an intrinsic individuality which can be grasped independently of any other entity, it is very doubtful that what concerns the relations to other (mathematical) entities is of no relevance to providing an ontology of numbers. The very conception of numbers as universals à la Russell makes very implausible that it is in principle possible to grasp the nature of one of them independently of any other. So, on one hand, it seems that some sort of acquaintance might have a role in the understanding of meaning; on the other hand, if it is conceived as knowledge of an entity in itself, as Russell originally understood it, it is not very helpful in the understanding of mathematical entities.22 5. Logical form A general feature, not tied up to any specific content, belongs to a Russellian conception of propositions. In The Principles of Mathematics, Russell strongly emphasises that a proposition is not a mere list of entities and claims that a complete analysis of a proposition should also account for 19. Russell (1912, 105). 20. Russell (1912, 46). 21. Russell (1914, 151). 22. A doubt concerning the role of the principle of acquaintance can be raised if numbers — or, more generally, entities — which cannot be known by acquaintance are taken to exist and allowed as constituents of propositions. Such propositions could not be understood according to the principle of acquaintance. If there is no other sense in which such propositions are understandable, their very role appears to be rather unclear.

106

the relations among its constituents. In the book Russell is not able to point out such relations. The only clear way in which in the end he could conceive the combination of entities like the object a, the relation R and the object b would have been “the relating of a and b by R which only exists if the fact that aRb does”23. In the manuscript Theory of Knowledge Russell resorts to the form, not in relation to propositions, but in relation to complexes that he identifies with facts. He says It is obvious … that when all the constituents of a complex have been enumerated, there remains something which may be called the “form” of the complex, which is the way in which the constituents are combined in the complex.24

Attributing a form also to the propositions conceived as ontologically structured entities which are true or false is surely compatible with the Principles of Mathematics conception. If there is a form providing complete information about the positions of the components of the proposition, so that xRx but not xRy is completely informative about the positions of the components in the proposition expressed by “a = a”, it would appear natural to speak of the more specific form as the form of the proposition. If the form of a proposition is so conceived, the intuitive principle: UoF A proposition is understood only if its form is grasped raises an obvious problem. Let us take, for example, the proposition expressed by the sentence a = b. If a = b, then the form to be grasped to understand the proposition expressed should be xRx. Hence knowledge of the reflexivity of identity would allow one to know that the proposition a = b is true and consequently the sentence “a = b” is also true. In other words, the mere understanding of an identity true sentence would be enough to state that it is true. Of course, this would justify the substitutivity of identicals within an epistemic context. Since this is problematic — to say the least —, if a form should be grasped in the understanding of a proposition, it cannot be the most detailed representation of the positions of the ingredients of the proposition within the proposition. Does this problem arise from the original Russellian conception of logical form? There might be reasons to return a negative answer, but, as we will see, that conception is not so definite, clear, and unobjectionable. 23. Linsky (1993, 200). 24. Russell (1913, 98).

107

Even if Russell seems to think that one needs to be acquainted with the logical form in order to know how the components of the proposition — except the logical form, if this is itself a component — are related to each other, some of Russell’s students are willing to claim that the main motivation is provided by the goal of making nonsensical judgements impossible25. Logical forms can perform such a task by determining the logical types of the constituents and the position of the subordinate relation of judgement. The positions of other constituents is not, in general, determined by them. So we cannot say that the logical form, as conceived by Russell, is the most detailed representation of the positions of the ingredients of the proposition within the proposition. There might be also a specific reason to deny that Russell’s logical form of a true identity proposition — even when it provides the most detailed representation of the positions of the ingredients of the proposition — raises the problem pointed out above. According to Russell’s Theory of Knowledge, some relations, like similarity, are such that their arguments are of the same kind and all have the same position in the complexes where these relations are “relating”. He takes all symmetrical relations as having this feature. We should infer that, in the complexes where they are “relating”, there is only an argument position and it can be occupied by two different entities, as suggested by the example of similarity. If we apply this conception to the kind of Russellian propositions we are considering and, in particular, to identity propositions, it seems to follow that the most detailed representation of the positions of the ingredients within an identity proposition should be something like xRx or R(x,x) both in the case it is true and in the case it is false; indeed, as concerns the arguments, only their positions have to be represented and these are identical in any case.26 So logical form would not allow one to discriminate true identity propositions from false ones.

25. See Griffin (1985–6). Russell’s motivations for introducing the concept of logical form are examined in detail by Bonino (2001). 26. There is an obvious difficulty in this view of logical form: the argument-position variable cannot be taken as a variable representing the entities occupying the argument-position without violating, in some cases, the non-ambiguity requirement.

108

6. Conclusions The first aim of the present paper was to show that certain ways of reference represented in Shapiro’s and Myhill’s formal intensional mathematical theories agree with the kind of epistemic constraints that Russell puts on reference. Secondly, we argued that the avoided ways of reference can be made compatible with a Russellian conception of the understanding of sentences, since those constraints do not belong to the core of such conception, nor are they justified by the classical point of view on truth and knowledge. As concerns the core thesis that the understanding of sentences requires acquaintance with the constituents of the propositions expressed, we took into account the case in which natural numbers, or mathematical entities in general, occur as propositional constituents, hence acquaintance with them is required. We saw that, if acquaintance has the features assigned to it by Russell, the question arises whether knowledge of such entities can be so conceived. There are serious reasons to doubt that Russell’s idea of acquaintance is adequate to provide knowledge of mathematical entities. It should be granted that some sort of acquaintance plays a basic role from the point of view of the knowledge of meaning, but, for the reasons we have indicated, it cannot be the original Russellian one. Last, but not least, we tackled the general issue of the role of the logical form in the understanding of sentences. Since logical form is attributed to propositions and its grasping is taken as necessary to the understanding of sentences, it cannot be conceived as the most detailed representation of the positions of the constituents of the proposition, at least if the notion of position in a proposition is thought of by analogy with that of position in a sentence. An alternative clear, precise, convincing specification of the concept of logical form is not provided by Russell. As a general conclusion, we would like to claim that (1) both the restrictions on reference and the kind of knowledge represented in Shapiro’s and Myhill’s theories can be compared in a rather natural way with the traditional Russellian ideas on reference and understanding of sentences, (2) the ideas concerning reference and those concerning sentence understanding are broadly independent of each other, (3) in order to make the ideas pertaining to sentence understanding more satisfactory, some modifications concerning the notions of acquaintance and of logical form are needed.

109

REFERENCES Anderson, C. A. 1989. “Russellian Intensional Logic”. In: J. Almog , J. Perry, and H. Wettstein, eds. Themes from Kaplan, Oxford: Oxford University Press, pp. 67–107. Benacerraf, P. 1965. “What Numbers Could not Be”. Philosophical Review, 74, 4773. Reprinted in P. Benacerraf, and H. Putnam, eds. Philosophy of Mathematics. Selected Readings, Cambridge: Cambridge University Press (second edition). Bonino, G., 2001. Russell and Wittgenstein’s Tractatus. PhD thesis, Università del Piemonte orientale. Griffin, N. 1985–6. “Wittgenstein’s Criticism of Russell’s Theory of Judgment”. Russell, 2, 132–145. Linsky, B. 1993. “Why Russell Abandoned Russellian Propositions”. In: A. D. Irvine and G. A. Wedeking, eds. Russell and Analytic Philosophy. Toronto: University of Toronto Press, 1993, 193–209. Myhill, J. 1985. “Intensional Set Theory”. In: S. Shapiro, ed. Intensional Mathematics, Amsterdam: North Holland, 47–61. Pears, D. 1977. “The Relation between Wittgenstein’s Picture Theory of Propositions and Russell’s Theories of Judgement”. The Philosophical Review, 86, 177–196. Russell, B. 1903. The Principles of Mathematics, Cambridge: Cambridge University Press. — 1910–11. “Knowledge by Acquaintance and Knowledge by Description”. Proceedings of the Aristotelian Society, 11, 108–28. Reprinted in Russell, B. Mysticism and Logic, London: George Allen and Unwin 1963, 152–67. — 1912. The Problems of Philosophy. Oxford: Oxford University Press, reprinted by Indianapolis: Hackett Pub. Co., 1991. — 1913. “Theory of Knowledge: The 1913 Manuscript”. In: E. R. Eames and K. Balckwell, eds. The collected Papeers of Bertrand Russell (vol . 7), London: Allen and Unwin, 1984. — 1914. Our Knowledge of the External World. London: Allen and Unwin, 2nd ed., 1926. — 1918–19. “The Philosophy of Logical Atomism”. The Monist, 28, 495–527; 29, 32–63, 190–222, 345–80. Reprinted in R. C. Marsh, ed. Logic and Knowledge. London: Allen and Unwin, 1956, 177–281. Shapiro, S. 1985. “Epistemic and Intuitionistic Arithmetic”. In S. Shapiro, ed. Intensional Mathematics. Amsterdam: North Holland, 11–46. — 1997. Philosophy of Mathematics, Structure and Ontology, Oxford: Oxford University Press.

110

Grazer Philosophische Studien 72 (2006), 111–140.

ON THE SAMENESS OF THOUGHTS. SUBSTITUTIONAL QUANTIFIERS, TENSE, AND BELIEF Marco SANTAMBROGIO Università di Parma, Italy Summary In order to know what a belief is, we need to know when it is appropriate to say that two subjects (or the same subject at two different times) believe(s) the same or entertain the same thought. This is not entirely straightforward. Consider for instance 1. Tom thinks that he himself is the smartest and Tim believes the same 2. In 2001, Bill believed that some action had to be taken to save the rain forest and today he believes the same. What does Tim think? That he, Tim, is the smartest, or that Tom is? And what does Bill believe today? That action had to be taken in 2001 or that it has to be taken now? Both answers are intuitively acceptable. This has to be accounted for somehow. Building on Mark Richard’s work on tense, Scott Soames1 claims that the substitutional interpretation of the quantifiers is unable to account for the intended meaning of such statements as (2) and the validity of some inferences involving them. I will show that his argument is not convincing. Not only does the substitutional interpretation fare no worse than the objectual one, but it seems to be able to avoid a problem which could be seriously damaging for any account of the sameness of thoughts based on the notion of structured proposition. In the first section, I state the problem allegedly raised by tensed belief ascriptions to the substitutional interpretation of the quantifiers. In the second, Soames’s argument is shown to be flawed. I also show that the content of the that-clause in (2) is not faithfully represented by any kind of structured proposition. Finally, I show how the substitutional interpretation can handle all such statements as (1) and (2) and the inferences involving them.

1. Soames (1999, 44 ff); Richard, (1981, 1–13; 1981a, 337–351).

1. According to the substitutional interpretation of the quantifiers, a statement such as There is something S believes — (p)(S believes that p) — is true if and only if the class of expressions substitutable for the variable p contains a particular clause that yields a true instance of the matrix S believes that p. For instance, if Cicero was bald is in the substitution class, and if S believes that Cicero was bald, then (p)(S believes that p) turns out to be true, since its substitution instance S believes that Cicero was bald is such. The objection we are about to consider is not the usual one, to the effect that the substitution class might be too meagre, so that no clause can be found in it to express the content of someone’s belief. This would result in some existential belief statements not being true, which intuitively are true. The objection that interests us is, rather, the following: from the two premises 3a. Jane expects that she will win the match 3b. Clara expects that she will win the match, it does not validly follow that 3c. There is something that both Jane and Clara expect, — i.e., (p)( Jane expects that p and Clara expects that p). However, if the clause she will win the match is in the substitution class and can be substituted for p, then (3c), in the substitutional interpretation, does follow from (3a) and (3b) taken together. A more elaborate version of the same objection is given by Scott Soames. His argument consists of two parts. The first proceeds as follows. Consider these sentences: 4a. On November 4, 1996, Al believed that Clinton was about to be elected president 4b. Al now believes everything he believed on November 4, 1996. Soames correctly points out that (4a) and (4b) together do not imply 4c. Al now believes that Clinton is about to be elected president. He assumes that the logical forms of (4a) and (4b) are respectively as fol-

112

lows (hopefully, the astute reader will be able to figure out by herself the intended meaning of the operators occurring here, which are standard in temporal semantics): 5a. (On November 4, 1996, + Past) (Al believe that Clinton be about to be elected president) 5b. For all P [if (on November 4, 1996 + Past) (Al believe P), then (Now + Present) (Al believe P)] However if (i) the quantification in [(5b)] is substitutional and (ii) the untensed matrix Clinton be about to be elected president is in the substitution class of the quantifier, then [(5a) and (5b)] entail [(5c)] [5c]. (Now + Present) (Al believe that Clinton be about to be elected president) Since [(5c)] is the logical form of [(4c)], it follows that either the quantification in [(4b)] is not substitutional or the matrix Clinton be about to be elected president is not in the substitution class of the quantifier.2

This was the first part of the argument. In rough outline — the details will be filled in subsequently — in its second part it claims that, if the untensed clause Clinton be about to be elected president is not in the substitution class of the quantifier, then substitutional quantification is unable to account for the valid inference from (4a) and (4b) to 4c. Al now believes that Clinton was about to be elected president. Before discussing the first part of the argument, let us briefly reconsider Jane’s and Clara’s case. Suppose that psychologically they are quite similar: they are so self-confident and determined that, if you ask either of them what she expects as to the result of the match, the answer will be “To win” or “That I win”. There is a lot that Jane and Clara have in common, even though they differ in other respects, as the following two statements bring out: 3ac. Jane expects to win the match 3bc. Clara expects to win the match 2. Soames (1999, 44).

113

(3ac) and (3bc) are not equivalent to (3a) and (3b), respectively, since the latter are to some extent ambiguous, among other things because the referents of the two occurrences of ‘she’ are not uniquely determined. But there is no denying that if the former two statements are true, then there is something both Clara and Jane expect, namely, to win. (3c) does follow from (3ac) and (3bc) taken together, even though it does not follow from (3a) and (3b). One might wonder: is it not strange to say that Jane and Clara have the same expectation, even though only one of them will be happy at the end of the match? How can the same circumstance, the outcome of the game, make the same expectation come out both true and not true? There is no reason for puzzlement, however. What (3ac) and (3bc) unambiguously ascribe to both Jane and Clara is an expectation de se. Generally speaking, those attitudes are de se that help those entertaining them to locate themselves with respect to space and time and the rest of the population. To some extent, all attitudes are de se since, even when we entertain ordinary de dicto attitudes, we are in a sense locating ourselves in some way, since we relate ourselves (in the appropriate dimension, e.g., of expectation or belief ) to some proposition or other. But we are especially interested in those attitudes that are irreducibly de se, in so far as no ordinary propositional object can be taken to be the content of the attitude. If, for a moment, we follow David Lewis’s account of irreducibly de se attitudes, which takes them to consist in self-ascribing properties falling short of being propositions (trivially, every proposition is also a property),3 then we must say that the contents of those expectations of Jane and Clara are the same and they consist in the property of being the winner of that particular match, which each of them ascribes to herself (in the appropriate dimension, i.e., of expectation).4 Of course, Jane and Clara also differ, in so far 3. “I say that all belief is ‘self-locating belief ’. Belief de dicto is self-locating belief with respect to logical space; belief irreducibly de se is self-locating belief at least partly with respect to ordinary time and space, or with respect to the population. I propose that any kind of self-locating belief should be understood as self ascription of properties”. (D. Lewis 1983, 33–60) 4. “Doubtless it is true in some sense that Heimson does not believe what Hume did. But there had better be also a central and important sense in which Heimson and Hume believe alike. For one thing, the predicate ‘believes that he is Hume’ applies alike to both: Heimson believes that he is Hume and Hume believes that he is Hume. Do not say that I equivocate, and that what is true is only that Heimson believes that he (Heimson) is Hume and Hume believes that he (Hume) is Hume. Everyone believes that Hume is Hume, but not everyone believes that he — he himself — is Hume. There is a genuine, univocal predicate, which appears for instance in ‘Not everyone believes that he is Hume’, and that is the predicate that applies alike to Heimson and Hume”. (D. Lewis 1983, 142).

114

as each of them expects herself , not the other, to have that property. But every attitude consists in self-ascribing some entity or other. Whenever that entity is a proposition, rather than a property, ascribing it to oneself is no different from ascribing it to any other. Lewis’s account of de se attitudes is not the only one available, of course, but it is especially clear and elegant. Later on, we shall have to consider other accounts.5 Let us now go back to the first half of Soames’ argument and to the case of Al, who believed on November 4, 1996, (the day before Clinton was in fact elected) that Clinton was about to be elected and still believes the same. In the quotation above, Soames claims that the substitutional interpretation is unable to block an invalid inference, if the untensed matrix Clinton be about to be elected president is in the substitution class of the quantifier. Note that he assumes without further argument that (4a) On November 4, 1996, Al believed that Clinton was about to be elected president means that Al at that date believed the proposition expressed by Clinton be about to be elected president relatively to that time. Moreover, he assumes that this proposition is the same as the one expressed by On November 4, 1996, Clinton be about to be elected president relatively to any time. Soames’ assumption therefore amounts to taking (4a) as being synonymous with (4ac) On November 4, 1996, Al believed that on November 4, 1996 Clinton was about to be elected president. Surely, today, in order to believe what he believed on November 4, 1996 according to (4ac), Al must believe that Clinton was then, not now, about to be elected. (4c) does not validly follow from (4ac) and (4b). But is Soames right in taking (4a) to mean the same as (4ac)? In the case of (2)

In 2001, Bill believed that some action had to be taken to save the rain forest and today he believes the same,

we pointed out that two answers could be given to the question, “What does Bill believe today?” (2) resembles (4a) closely enough to suggest that 5. A useful discussion of several theories of de se belief, together with his own proposal, can be found in Stalnaker (1999, 130–149). Another useful survey and a defence of a different theory is P. J. Markie (1988, 573–600).

115

we should be cautious in answering the corresponding question, “What does Al believe today?”. Suppose that on November 4, 1996, Al had lost track of time. He was unsure whether it was November 3 or 4 or even October 30. He had also forgotten when the elections were due to take place. However, he was sure that he would soon be witness to Clinton’s re-election, whenever it would happen. This belief of his would be an instance of de se belief, since it amounts to believing himself to be located at a time immediately preceding Clinton’s re-election — in Lewis’s scheme, to self-ascribing the property of being so located. His case would closely parallel that of Lingens the amnesiac, who, oblivious of who and where he was, nonetheless believed he was lost in some library or other — the only difference between Al and Lingens being that the latter was lost in space, whereas Al is lost in time.6 In the situation just imagined, the following statement would be true: (4as) On November 4, 1996, Al believed Clinton to be about to be elected president, — the infinitival clause conveying the idea that Al was in the belief state of someone expecting to be witness to Clinton’s re-election in the immediate future, even though he might or might not be in a position to locate the re-election in time otherwise than relatively to the current time. Today Al could easily retain the same belief ascribed to him by (4as). Like Rip van Winkle, he could have fallen asleep in 1996 to wake up today without realizing that the situation has dramatically changed. Or, more simply, he could be unaware that American presidents cannot be re-elected more than once. The difference between (4ac) and (4as) is far from insignificant. Intuitively, it amounts to this: assuming that Al, unlike Rip van Winkle, has not lost track of time and still believes today what he believed back in 1996 according to (4as), his present belief cannot be based entirely on his remembering his past beliefs. If, on the other hand, Al entertains the same belief today which is ascribed to him by (4ac), then his present belief can be entirely based on his memory of his own past beliefs, whether or not he has lost track of time. Another way of expressing the difference, which appeals to the apparatus of propositions, is the following: the proposition believed by Al, according to (4ac), is the tempo6. The character Lingens was first introduced by Frege, in his classic “The Thought”, and has been discussed ever since.

116

rally determinate one expressed by the untensed sentence ‘On November 4, 1996, Clinton be about to be elected president’. On the other hand, (4as) ascribes to Al a belief in a different proposition, namely the temporally indeterminate one expressed by the untensed ‘Clinton be about to be elected president’. The contrast between temporally determinate and temporally indeterminate propositions parallels Lewis’ distinction between propositions and properties. Now, if (4ac) holds, then (4a) also does. Similarly, if (4as) holds, then (4a) does. On the other hand, (4a) implies neither (4ac) nor (4as), since (4a) could be true even though Al, not knowing the date of the election, had no belief relating him to the temporally determined proposition expressed by On November 4, 1996, Clinton be about to be elected president; and it could also be true even if Al had no belief de se concerning Clinton’s election because, e.g., he was unable to locate himself in time with respect to the election date. It has to be kept in mind that, as Soames points out, Al could retain the belief ascribed to him by (4ac) at any later time. It goes without saying that, on November 4, 1996, Al could have both the belief ascribed to him by (4ac) and that ascribed by (4as). Moreover, intuitively (4ac) and (4as) do not imply each other, just as Lingens could very well believe that he himself was in a library without believing that Lingens was and, conversely, he could believe, like the rest of us, that Lingens was in a library without believing that he himself was. It would be wrong to suppose (4a) to be ambiguous. There are several reasons why it is not. First, the difference between (4ac) and (4as) depends on how Al locates himself in time, but there is no component of (4a) that refers to this. If (4a) were ambiguous just because it can be made true by different sorts of circumstances, then any number of other statements would count as being ambiguous, which intuitively are not. For instance, one can open a door either by brute force or gently with a key, or partly in one way, partly in the other. This, however, has no tendency to show that the statement ‘Al opened the door’ is ambiguous. Since none of its components refers to how Al opened the door, there is no room for ambiguity here. Second, most ambiguous statements are clearly seen to be such when negated. For instance, the negation of ‘Every boy loves some girl’ is incompatible with one of its readings, but perfectly compatible with the other. It is not so with (4a), since its negation does not introduce any asymmetry between (4a’) and (4a’’). If it is not the case that on November 4, 1996, Al believed that Clinton was about to be elected, then neither (4ac) nor (4as) could possibly be true. Third and most importantly, the truth conditions

117

of (4a) are unambiguously determined as follows: (4a) is true if and only if on November 4, 1996, Al (assuming sincerity, reflectiveness, linguistic competence, etc.) had a disposition to assent to the question ‘Is Clinton about to be elected?’, or to a translation of it into any other language known to him. Equivalently, (4a) is true if and only if on November 4, 1996, Al had a belief which was true if and only if on the same date Clinton was about to be elected. There is no room for ambiguity here. Finally, it has to be mentioned that the truth value of (4a) is not dependent on the context in which it is uttered in any relevant sense. How are (4a), (4ac), and (4as) related? We have already seen that the former does not imply the latter two, even though it is implied by them. But is it possible to be more specific? We saw that two cases are to be distinguished concerning Al’s state of knowledge. Al could either know that it was November 4, 1996, when the question ‘Is Clinton about to be elected?’ was posed to him, or be unaware what date it was. In the first case Al would also have a disposition to assent to the question ‘Is Clinton about to be elected on November 4, 1996?’and — November 4, 1996, being the time then current — also to ‘Is Clinton about to be elected at the present time?’. In the second case, Al would have no disposition to assent to ‘Is Clinton about to be elected on November 4, 1996?’. But he would still have a disposition to assent to ‘Is Clinton about to be elected at the present time?’, since the definite description is tantamount to ‘now’ and Al, while being ignorant of his actual location in time, believed that he would soon be witness to Clinton’s re-election. This shows that, even though (4a) does not mean the same as (dd) On November 4, 1996, Al believed that Clinton was about to be elected president at the current time, still the truth conditions of (4a) and (dd) are the same.7 But the definite description ‘the current time’ occurring in (dd) can be taken either referentially or attributively. In his classical paper,8 Donnellan characterized the role of a definite description in the referential reading as being comparable to that of a co-referring proper name or any other device of direct reference. If we agree with Donnellan, we conclude that (dd) in the referential reading is tantamount to (4ac). In the attributive reading of a definite description, 7. This clearly needs to be argued for more carefully. But it is not especially important to do it at this point. 8. Donnellan (1966, 281–304).

118

on the other hand, it must always be appropriate to supplement the description with some such clauses as ‘whatever it is’, ‘whoever he is’, etc.. Now, by asserting ‘Al believed that Clinton was about to be elected president at the current time, whatever that time was’ we represent Al as expecting Clinton soon to be elected, whether or not he was aware of the date of the election. This is tantamount to (4as). This shows that (4ac) and (4as) originate from a statement equivalent to (4a), by taking a definite description either in the referential or in the attributive reading. It has been argued — convincingly, in my opinion — that there is no real semantic ambiguity in a definite description whether it is read referentially or attributively.9 2. The first part of Soames’s argument against substitutional quantification starts from the assumption that (4a) On November 4, 1996, Al believed that Clinton was about to be elected president is tantamount to (4ac) On November 4, 1996, Al believed that on November 4, 1996, Clinton was about to be elected president. As a consequence, Soames also holds that (4a) On November 4, 1996, Al believed that Clinton was about to be elected president (4b) Al now believes everything he believed on November 4, 1996 together do not imply (4c) Al now believes that Clinton is about to be elected president. We saw above that there is every reason to doubt that (4a) means the same as (4ac). If, e.g., Al believed (de se) Clinton to be about to be elected, then 9. See Kripke (1977).

119

(4as) and (4a) would hold true, but (4ac) would not. Moreover, assuming that (4b) holds true, (4c) would hold true as well. The second part of Soames’ argument proceeds as follows. The untensed clause Clinton be about to be elected president cannot be in the substitution class of the quantifier (for, if it were, the substitutional quantifier would make (4c) true, contrary to his assumption). But then, he surmises, substitutional quantification is unable to account for the valid inference from (4ac) and (4b) to 4cc. Al now believes that Clinton was then (i.e., on November 4, 1996) about to be elected president. There is no denying that this inference does indeed hold good, even though it is not the same as that from (4a) and (4b) to (4cc). Why does substitutional quantification fail here? For the inference to be valid substitutionally, there must exist in the substitution class a clause that expresses the same proposition (namely, that on November 4, 1996, Clinton was about to be elected) both when embedded in ‘On November 4, 1996, Al believed that …’ and when embedded in ‘Al now believes that …’. It is Soames’ contention that such a clause does not exist. But what about the very clause 5. Clinton was then (i.e., on November 4, 1996) about to be elected president, that occurs embedded both in (4c) and in (4ac)? Isn’t this precisely what is needed? No, says Soames. He maintains that the logical form of (5) is 6. (Then + Past) (Clinton be about to be elected president), and, accordingly, the logical form of (4ac) is 7. (On November 4, 1996 + Past) [Al believe that (then + Past) (Clinton be about to be elected president)] But there is no reason to think that, relative to November 4, 1996, it expresses anything Al believed at that time, because for it to be true, Al had to believe that November 4, 1996, was earlier than November 4, 1996 — which is necessarily false.

120

Since there is nothing to suggest that Al believed this absurdity, there is still no explanation of the inference from [4a] and [4b] to [4cc] in the substitutional interpretation. Nor will one be forthcoming unless we can find a sentence that expresses, relative to every time, the proposition expressed by Clinton be about to be elected relative to the particular time November 4, 1996. Since it is doubtful that there is any such sentence in English, it is doubtful that the quantifier in the English [4b] is substitutional.

This concludes Soames’ objection. In my opinion, the argument is flawed. In the first place, it crucially assumes that (5) has a unique logical form, namely (6), irrespectively of its occurring unembedded, as in (5) itself, or embedded as in (4ac) and (4c). There is every reason to doubt, however, that this is so. In particular, when it is embedded in (4ac), its temporal operator seems to be not (then + Past) but, rather, (then + Present). Second, it is not at all obvious that the logical forms of the clauses belonging in the substitution class associated with the quantifiers in the substitutional interpretation are at all relevant. All that matters for a quantified sentence to be true is that its substitution instance(s) be true, as is clearly the case of (4ac), whatever the logical form(s) of the substituted clause. Third, if (5) does not do (and we have been given no reason why it does not), then some other clause might do. As a matter of fact, it is not difficult to find other such clauses: ‘On November 4, 1996, Clinton was about to be elected’, for instance, is perfectly acceptable. What is more important, it would be quite surprising if no such clause existed. For, whenever 8. There is something that Al believed in 1996 and that he still believes today is assertible, it must be appropriate to ask “What is the thing he believed then and still believes today?’, and unless a clause can be found to be substituted for P in ‘Al believes that P in 1996 and he still believes that P today’, without any syntactic change (not even a stylistic adjustment) so that the result is true, the assertion (8) has to be withdrawn. This holds quite generally: it is inappropriate to assert ‘X believed (or believes) that P and Y believes the same’, or even ‘X M’s and Y does the same’, unless ‘that P’, respectively M, can be substituted for ‘the same’ so that the result is true.10 10. It is instructive to see how Soames argues that the objectual interpretation of the quantifiers is able to account for the validity of the inference from (4a) and (4b) to (4c). It is assumed

121

In the third section, we shall see how the substitutional interpretation of the quantifiers can account in general not only for all valid inferences of the same form as that from (4ac) and (4b) to (4cc) — the only one considered by Soames — but also for those similar to the one from (4as) and (4b) to (4c). Here I want to point out that such statements as (4a) seem to raise a problem for all those accounts of de se propositional attitudes that — unlike Lewis’ account, which appeals to properties and propositions taken as sets of possible worlds — make use of structured propositions. We saw that 4a. On November 4, 1996, Al believed that Clinton was about to be elected president can be true in two different sorts of circumstances: Al could either know the date when the election was due to take place, or be ignorant of that that the proposition believed by Al in 1996 can be expressed today by 5. Clinton was then about to be elected president, — where ‘then’ refers to November 4, 1996. The logical form of (5) is 6. (Then + past) (Clinton be elected president) (Clearly, it is a crucial assumption here that the logical form of a clause such as (5) is independent from its occurring by itself or embedded in some other sentence. In the text above we have met with some reason to doubt that this assumption is true.) Now 4A. (On November 4, 1996 + Past) (Al believe that Clinton be about to be elected president) 4B. For all P [if (on November 4, 1996 + Past) (Al believe P), then (Now + Present) (Al believe P)], are the logical forms of (4a) and (4b) respectively: 4a. On November 4, 1996, Al believed that Clinton was about to be elected president. 4b. Al now believes everything he believed on November 4, 1996. From (4A) e (4B) we have: 4C. (Now + Present) [Al believes that (then + Past)(Clinton be about to be elected president)]. Which is the logical form of 4c. Al now believes that Clinton was then about to be elected president. It is Soames’ claim that we have here a proof, making use of the apparatus of temporal operators, that the objectual quantifiers can account for the inference above. The proof crucially depends on the fact that the same proposition can be expressed by different clauses at different times. This is a considerable advantage over the substitutional interpretation, which is more heavily constrained. Note, however, that we are given no proof that the proposition believed by Al in 1996 is the same that is expressible today by (5). This particular example is so simple that one can readily agree that not much of a proof is needed for this fact, but, in general, how do we know when two sentences express the same proposition? In § 3 an objection will be raised to the usual answer to this question, namely that two sentences express the same proposition if and only if no competent speaker could possibly believe either of them without believing the other as well.

122

date, while being able to locate the election relatively to the time that was then present (whether or not he knew that it was November 4, 1996). This has no tendency to show that (4a) is ambiguous. It is just silent about Al’s state of knowledge — it lacks specificity. The problem I see for structured propositions is that they are too specific to faithfully represent the content of (4a). Let us focus on a particular account of de se propositional attitudes in terms of structured propositions — that given by John Perry.11 According to Perry, a belief (any belief, and in particular a de se one) has two objects. The first is a structured Russellian proposition. For instance, in the case of the irreducibly de se belief which Lingens would express by saying ‘I am lost in a library’, the first object of his belief is a pair consisting of Lingens himself and the property of being lost in a library. It is meant to account for the belief being true, in case it is, which happens if Lingens has in fact the property. This is what is believed by Lingens. The second object of Lingens’s belief is meant to represent the way the first object is believed by him. It is an entity modelled on the notion of sentence meanings, such as a function that takes the subject as argument and delivers as value the first object, i.e., the Russellian proposition. It corresponds to, or just is, the character (in David Kaplan’s sense) of the sentence ‘I am lost in a library’. Now, (4a) ascribes to Al neither a belief de se, nor a belief in any proposition involving a particular time, since it can be true, as we saw, both in case Al is able to locate Cliton’s election absolutely, e.g., by means of a date, and in case he can only locate it with respect to the position in time where he is, when he has the belief. Still, it ascribes to Al some belief. And it is not ambiguous. How can we represent its content in terms of the two objects posited by Perry? One might suppose that the first object is the pair consisting of a point in time, November 4, 1996, and Clinton being about to be elected, and the second is the character of such a sentence as ‘Clinton is about to be elected’ (or perhaps the property of soon being witness to Clinton’s election). This, however, leaves out the possibility that Al had no de se belief (remember that, today, possibly without knowing how much time has elapsed, Al could still believe that Clinton was then about to be elected). On the other hand, if the second object is taken to be the character of such a sentence as ‘On November 4, 1996, Clinton is about to be elected’, 11. In Perry (1977, 474–97) and Perry (1979, 3–21), and elsewhere.

123

then the possibility is left out that Al only had an irreducibly de se belief. Alternatively, one might think of leaving the second object open by, e.g., existentially quantifying over the entities that can occur in the second object position. This would still be inaccurate, however, if Al had a de se kind of belief, but was not aware that the time then current was November 4, 1996, or even had a false belief about it. The closest one can come to faithfully representing the belief ascribed by (4a) is to take the second object in Perry’s scheme to be some kind of disjunction (e.g., of the characters respectively corresponding to ‘On November 4, 1996, Clinton is about to be elected’ and to ‘Clinton is now about to be elected’). But, clearly, this amounts to forfeiting the compositionality principle, since the disjunction bears little similarity to the syntactic structure of (4a). What is more important, we noted above that exactly two answers can be given to the question “What does Al believe today, if he continues to believe what he believed in 1996 according to (4a)?”, and this has to be accounted for somehow. The suggestion we are considering is unable to do so. It is to be noted that Lewis’s account does not have the same problem for two reasons: first, it does not assume that the objects of belief have to reproduce in structure the content sentence of the belief ascription; second, since propositions are, trivially, a kind of properties, it can take the object of the belief ascribed to Al by (4a) to be the property formed by the disjunction of those corresponding to the two kinds of cases distinguished above in which (4a) would be true. None of this is a problem for the substitutional interpretation of the quantifiers, which has no truck with the apparatus of propositions and no need to take (4a) as being in any sense ambiguous. In the next section we consider how it can account for all the inferences considered so far. 3. We have examined several belief ascriptions and a few inferences involving them, some valid and some not. Let us review the main ones. The inference from (10) Tom believes that he himself is the smartest, and Tim believes that he himself is the smartest, to

124

(11) There is something both Tom and Tim believe is valid. This can be accounted for as follows. First, Tom believes that he himself is the smartest is equivalent in meaning to Tom believes himself to be the smartest. If to be the smartest is in the substitution class, then p (Tom believes himself p and Tim believes himself p) — the quantifier being read substitutionally — turns out to be true. Equally, (11) follows from (12) Tom believes that he Tom is the smartest, and Tim believes that he Tom is the smartest, and is accounted for likewise, if that heTom is the smartest is in the substitution class. Note that we definitely do not want the substitution class to contain such expressions as that he is the smartest, unless it is determined which term ‘he’ is co-referential with. More generally, we shall not let into the substitution class any expression in which a variable occurs free, lest inferences such as the following turn out to be valid: (a) Al loves his girlfriend (b) Bill loves his girlfriend ? (c) There is someone both Al and Bill love Note, however, that from (a) and (b) taken together, (c) There is something both Al and Bill do — i.e., love their own girlfriend does follow, since the variable x does not occur free in O x.(x loves x’s girlfriend) which can therefore belong in the substitution class. The inferences involving Jane and Clara can be accounted for in the same way. Let us now consider the more difficult inferences involving (4a), as well as those involving (2), which is entirely similar to (4a). As we saw, no ambiguity is involved in (4a) and yet the question “What does Al believe today?” can be answered in two different ways, expressed by (4c) and (4cc) respectively. Neither (4c) nor (4cc) follows from (4a) and (4b) taken together — if only because the former two sentences are on a par and if either inference held good, then the other one would be left unaccounted for. However, both the inference from

125

(4as) On November 4, 1996, Al believed Clinton to be about to be elected, (4c) Today Al believes Clinton to be about to be elected, taken together, to (C)

There is a belief that Al had on November 4, 1996, and still has today,

and the inference from (4ac) On November 4, 1996, Al believed that on November 4, 1996, Clinton was about to be elected (4cc) Today Al believes that on November 4, 1996, Clinton was about to be elected, taken together, to the same (C), are indeed valid. The substitutional interpretation of the quantifiers can easily account for these inferences. But we also want to be sure that, more generally, it can account for all inferences that, from any two premisses ascribing the same belief to two subjects, possibly in different words, draw the conclusion that they believe the same. What is a thought and how is it expressed by an utterance? Utterances — particularly assertive utterances — always take place in some circumstance or other. They are uttered by a speaker, addressing some hearer or other, at some time and place, in some situation, etc. They can be classified in many different ways, most of which utterly uninteresting. It is useful, for instance, to classify them in terms of how they can be put on paper, as a stenographer can do. The stenographer makes use of a number of conventions, allowing him to ignore, as being irrelevant, a great many features of the utterances — pronunciation, emphasis, pauses, etc. The result is a segmentation of the utterances into sentences and words, according to rules which are undoubtedly quite difficult to spell out. Suppose we are somehow given the transcriptions of all the minimal units of utterances that can be evaluated as being either true or false. Relatively to a set of rules of transcription, we then have a partition of all utterances into equivalence classes, which might be called “sentences”. Two utterances belong to the same sentence if the same transcription can represent them both. For instance, if Jane utters I shall win and Clara utters I shall win, their utterances belong in the same sentence.

126

Partitioning the utterances into sentences, modulo the relation of admitting of the same transcription, is useful for many purposes. We have seen that, just because their utterances are similar enough to belong in the same sentence, Jane and Clara are likely to share some psychological attitudes. Of course, their utterances also differ in other respects. For instance, at most one of them can turn out to be true. There are other interesting similarities between utterances. For instance, if Jane says I am an eye-doctor and then I am an oculist, her utterances are similar in some respect, even though they are not in the same sentence. For one thing, they necessarily have the same truth value. But they resemble each other more than 2 + 2 = 4 and Every number has a unique prime decomposition do, which are also necessarily equivalent. For one thing, every competent speaker of English knows that Jane’s two utterances are necessarily equivalent, whereas the same cannot be said of the latter two mathematical statements. We have here another way of partitioning all utterances into equivalence classes. Now, how can we express the similarity between an utterance of Clinton is about to be elected made by Al on November 4, 1996, and an utterance of Clinton was then about to be elected, made by Al today (the context making it clear that he was referring to the same date)? They are similar enough to warrant our ascribing the same belief to Al, but in what respect exactly are they similar? And is the similarity holding between them an equivalence relation? An answer is suggested by Frege. Al’s two utterances bear the same relation to each other as the following two: It is sunny today, uttered yesterday and It was sunny yesterday, uttered today. About the latter pair, Frege says that they express the same thought: If someone wants to say the same today as he expressed yesterday using the word ‘today’, he must replace this word with ‘yesterday’. Although the thought is the same its verbal expression must be different so that the sense, which would otherwise be affected by the different times of utterance, is readjusted. The case is the same with words like ‘here’ and ‘there’. In all such cases the mere wording, as it is given in writing, is not the complete expression of the thought, but the knowledge of certain accompanying conditions of utterance, which are used as means of expressing the thought, are needed for its correct apprehension. The pointing of fingers, hand movements, glances, may belong here too. The same utterances containing the word ‘I’ express different thoughts in the mouths of different men, of which some may be true, others false. (The Thought)

127

This could amount to a full answer, if only we knew what a thought is. But do we know? Many philosophers believe that we do, because they hold that two utterances express the same thought if and only if no competent speaker of the language involved, who knows the relevant contexts, could possibly believe that only one of them is true. Whether or not this is a sufficient condition, it is certainly not necessary, however. For one thing, it is far from being obvious that any two utterances of, say, Paderewski has musical talent (which surely ought to express the same thought) are such that a competent speaker cannot believe that only one of them is true. For another, we have seen that in some cases (e.g., if one has lost track of time) one could express the same thought by uttering Clinton is about to be elected president in two different circumstances, without knowing whether or not the truth values of those utterances are the same, even though one is perfectly aware of both utterances and of the circumstances in which they took place (but is unaware of how much time has elapsed in between). But even supposing that the sameness of thoughts is satisfactorily accounted for by the aforementioned condition, much remains to be done. Among other things, given two utterances, which are such that no competent speaker could possibly believe one of them without believing the other, we want to know what it is in those utterances that accounts for this fact. Let us go back to Al’s two utterances about Clinton’s election and to their similarity. Frege’s quotation above, in particular his claim that the conditions accompanying an utterance are used as means of expressing the thought, gives us some clue as to how to characterize their similarity. When two speakers can avail themselves of different means for expressing their thoughts — possibly the same thought — what they need in order to communicate is a translation. The prototypical case of a translation is, of course, that between two distinct languages, but Frege’s idea here is that even the utterances of speakers of the same language who are differently located in time and space need translating, i.e., for an utterance to express the same thought as another, they must stand in the translation relation. The converse is not to be doubted: any two utterances that can properly be said to translate each other express the same thought. (What purpose could a translation possibly have, if it did not aim at preserving the thought expressed?) Putting the two things together, we have that a necessary and sufficient condition for two utterances to express the same thought is that they translate each other, i.e., are in the translation relation. It might be thought that this is not such a bold claim, after all. On the one hand, there seems to be general consensus that the notion of thought

128

can be captured by some notion of proposition, even though no general agreement has been reached so far as to which kind of propositions, of the large variety currently available, is best suited to do the job. On the other hand, it is often taken for granted that preserving the propositions expressed is all that there is to translation. From these two, seemingly commonsensical, tenets the claim above immediately follows. Things are not so simple, however, if only because it is certainly not true that proper translation consists in mapping utterances of one language onto utterances of another (or the same) language that express the same propositions. This is a philosophical prejudice and is quite at variance with the ordinary notion of translation, as was made clear long ago by Tyler Burge, Michael Dummett, and others. A new argument for the same conclusion will be given presently. If we consider the intuitive, not the philosophical or “propositional”, notion of translation, the claim above — that two utterances express the same thought if and only if they translate each other — is quite substantial and, as a matter of fact, an immediate objection can be raised to it, in particular to the conditional from left to right — i.e., that in order to express the same thought two utterances must stand in the translation relation. Here is the objection: no one would take Hier c’etait lundi, uttered today, to translate Today is Monday, uttered yesterday, even though one can readily agree with Frege that they express the same thought. (Note that, if it were successful, this objection would equally tell against the “propositional” notion of translation). I do not think that the objection is convincing. For one thing, it overrates the importance of written texts, which are not as dependent on the context as speech. For another, as far as the translation of speech is concerned, it assumes as mandatory the convention that a translator efface himself and his own point of view, as professional translators usually do. What effacing oneself consists in and why it is to be recommended in most cases, will be explained later. For the time being, the partial answer to that objection is, that even though the French utterance above is not the best possible translation of the English one, for reasons to be seen shortly, still it is a proper translation. (We are assuming that of course some translation or other must exist for the utterances in the example. Note also that we are here considering translations of utterances, not of sentence types.) If this is so, then by the same token, Clinton is (now) about to be elected, uttered in 1996, and Clinton was (then) about to be elected, uttered today (then referring to 1996), stand in the translation relation. Clearly, the truth value and the referents of

129

all subsentential expressions of the source utterance are preserved in the target utterance. Compositionality is also preserved. The tense of the verb, however, is not. We saw above that there is a sense in which also Clinton is (now) about to be elected, uttered on November 4, 1996, and Clinton is (now) about to be elected, uttered today, express the same thought and, if Al sincerely authored both utterances, it would be correct to say that he has preserved his belief. Can we say that these utterances, too, translate each other? Surely, we have as much reason to answer that they do as to hold that Clinton est en train d’être élu is a translation of either of them. In this case, both the lexical meaning of every expression type and the tense of the verb are preserved. Truth and reference, however, are not. Does this fact amount to any objection to taking the former two utterances as translating each other? It would do so, if there were any obviously better translation preserving truth and reference — a translation, that is, that is preferable no matter what the circumstances, the context, the purpose of translating, etc.. One can readily agree that it is a matter of degree how good a translation is, and a number of factors have to be weighted, but we are considering an objection to the effect that the proposed translation is not a translation at all. Now, there is no translation preserving truth and reference that clearly fares better than the proposed one in every case. If, for example, we offered Clinton était alors en train d’être élu, uttered today, as translating Clinton is (now) about to be elected, uttered in 1996, this would fare as well, or as badly, as Hier c’etait lundi uttered today as a translation of Today is Monday, uttered yesterday. We have already considered an objection to this translation, to which we answered that it is a proper translation, albeit clearly not the best one in all possible circumstances. Now, where are we? We have pairs of tensed utterances, in which indexical expressions may or may not occur, that clearly express the same thoughts. We ask ourselves, is there a translation, properly so called, that takes us from either utterance in each pair to the other? We seem to have good reasons to think that there is, but two parallel objections are raised: the first to the effect that, in some cases, tense, lexical meaning and character but not reference and truth, are preserved; the second to the effect that, in other cases, reference and truth, but not tense, lexical meaning and character, are. In no case, however, is a better translation in sight preserving every feature of the source utterance — tense, lexical meaning, character, reference and truth. It would be desirable, of course, that all these things be preserved, but the plain fact is that it is not always pos-

130

sible to do so. And yet, it would be preposterous to hold that, whenever it is not possible simultaneously to preserve each one of those features, no translation exists. It is fair to conclude, I think, that the two objections cancel each other: neither preserving tense etc., nor preserving reference etc. are strictly necessary for translation. Now, however, a different and more radical kind of objection can be raised to our project. If preserving either tense etc. or reference etc., or both, whenever it is possible to do so, were the only relevant desiderata for the notion of translation, then — one might object — why should we bother with the notion of translation at all? For one thing, it appears to be a rather vague notion. For another, all the hard and philosophically interesting work is done by the two familiar notions of propositional content and character (in Kaplan’s sense). Nothing new is likely to be forthcoming from examining translation more closely. This reaction, however, would be rash. The notion of translation is too important to be given up. Translation affords the best examples available of pairs of utterances that convey the same thought. But neither those pairs that only express the same propositional content, nor those that only share the same character, are perfectly satisfactory in the way of translation. We normally make an effort towards finding translations that preserve both content and character, if it is at all possible. This would hardly be intelligible if translation were an ancillary notion, defined entirely in terms of content and character, neither of which being more fundamental than the other. In order to realise how strong is the pull towards finding ways of preserving both content and character, let us consider some real-life translations of utterances containing indexicals. Suppose the French president visits the United States and addresses the American president in these words Je vous remercie pour votre invitation. The official translator could translate He (or even: the French president) thanks the American president for his invitation, but clearly this is not the best he can do. As we know, he is likely to translate I thank you for your invitation. By doing so, he manages to preserve both the reference of every subsentential expression and its lexical meaning. But, of course, some adjustments are to be made, because it is not really him, the translator, i.e. the actual speaker of the translated utterance, who is thanking the American president. What he is doing is pretending to be the French president, or — which amounts to the same — that the French president is speaking through him, so to say. In this way, even the truth value of the original utterance is preserved in translation, albeit only fictionally. The case mentioned above, of Aujourd’hui

131

c’est lundi, uttered yesterday, was entirely similar. Properly speaking, it is not wrong to translate it by means of today’s utterance Yesterday was Monday, but a crucial and perhaps intolerable change has occurred. A better translation is therefore Today is Monday if the pretence can be made clear, somehow, that the context of utterance today is what it was yesterday. It is the interest in making translations preserve all that can be preserved that explains this complex phenomenon, which Tyler Burge presented in these words: Translation of demonstrative expressions normally differs from indirect discourse reporting of such expressions in that the point of view of the demonstrator is assumed in the translation. For example, in translating an utterance of ‘I am hungry’ into German, we would use the expression ‘Ich bin hungrig’, where it is understood from the context (or perhaps made clear informally by the translator) that the referent of the German first-person demonstrative ‘ich’ is the English-speaker, not the translator. The translator presents the speaker’s point of view, rather than reports on it. On the other hand, in reporting the same utterance in indirect discourse we would say ‘(He said that) he was hungry’, or in German, ‘(Er sagte dass) er hungrig war’. The first-person demonstrative ‘I’ is replaced by a third-person demonstrative in order to maintain the point of view of the reporter rather than that of the original speaker.12

No explanation is given by Burge of the phenomenon, however, nor does he mention that some kind of pretence is involved in assuming the point of view of the demonstrator, but his remark that indirect discourse reports behave in translation otherwise than direct discourse does, is entirely correct. This, incidentally, reminds us that all the utterances concerning Al that we considered in the first part of the present paper were belief ascriptions, not direct discourse reports. The pretence trick was not available in those cases (since the point of view of the ascriber had to be kept fixed), and it was either the content or the character, but not both, that had to be preserved, depending on the belief ascribed. The second reason why one cannot define the translation relation in terms of the sameness of either character or content, or both, is that cases exist in which it is simply impossible to preserve either content or character in translation. This was made clear long ago by several philosophers. Let us begin with content. It is an obvious principle of translation that it preserves self-reference if and only if it does not preserve reference. As Burge put it: 12. Burge (1978, 140).

132

Translations which are commonly recognized as good ones sometimes do not preserve the reference of the expressions being translated, and sometimes this results from a concern with preserving self-reference. This perspective on translation conflicts with a standard philosophical notion, to wit, that a good translation expresses the same proposition expressed by the sentence it translates.13

Consider these sentences:14 4. This very sentence begins with a four-letter demonstrative 5. (5) makes reference to itself. Of course they can be translated literally. For instance, (4) can be translated into German as Dieser Satz fängt mit einem hinweisenden Artikel mit vier Buchstaben an, if it is somehow made clear that ‘dieser’ refers to the English (4). Clearly, however, something — possibly, something of the utmost importance for the point the author intended to make — is lost. If the author’s point was, e.g., precisely to illustrate the phenomenon of selfreference, that translation would not be correct. Gödel’s proof of the result that not every arithmetically valid sentence is provable, crucially depends on constructing a sentence that intuitively behaves like the following 6. (6) is not a theorem. Many widely, or subtly, diverging notions of proposition are presently available on the market. Without exception, they all agree in taking reference to be essential — that is, no two sentences differing as to the referent of any of their components can express the same proposition, on any notion of proposition. It follows from this, as Burge’s quotation above makes clear, that (6) and any of its acceptable translations into a language other than English, cannot express the same proposition, since each of them refers to itself, not to our (6), even if they are similarly numbered ‘(6)’. And yet, the thought expressed by them is the same, on any pre-theoretical understanding of the notion of thought. Character fares no better. Suppose we decide to translate 7. The seventh word of this sentence has three letters, 13. Burge, (1978, 141). 14. All the examples that follow are Burge’s. His point, however, is not the same as mine.

133

in such a way as to preserve both self-reference and truth. Then we have a choice (which has to be taken on the basis of other considerations) between, e.g., Das siebte Wort dieses Satzes hat vier Buchstaben and Das sechste Wort dieses Satzes hat drei Buchstaben — neither of which preserves character. It was assumed above that the thought is precisely what has to be preserved by translation proper. This is quite a safe assumption in general, but it certainly holds true in the particular case of sentence (6) and its self-referential translations. It would be unintelligible to suppose that the statements of Gödel’s theorem and of each step in its proof express different thoughts when couched in different languages. (This is not to say that every bilingual reader must understand and believe each of them in one language if and only if she understands and believes any of its translations into another. We pointed out above that the criterion for the identity of a thought cannot depend on the attitudes of the speakers, however competent, reflective, etc. The fact that, e.g., a bilingual student taking an advanced course in logic might think she understands the proof as given in her logic textbook in English, although she has doubt about it when reading the translation of the same textbook into German — surely not an inconceivable case — proves nothing.) Since there are steps in the proof15 that express different propositions in different languages, and yet express the same thought, we must conclude that propositions of any kind fail as rigorous explanations of the intuitive notion of thought. A fortiori, the translation relation cannot be defined in terms of content and character. Here is a third reason to doubt that the translation relation can be defined in terms of content and character. Besides preserving character, content, and self-reference, whenever it is needed and if it is at all possible, there is a further constraint that any translation has to satisfy: sameness and difference of expression types are to be preserved. For instance, it is not admissible to translate (from English into English) A woodchuck is a hedgehog as A woodchuck is a woodchuck (or vice versa) even though both character and content are preserved. Nor is it admissible to translate two distinct proper names, such as Germaniah and Ashkenaz in Hebrew16, occurring in some identity statements, by means of one and the same name, Germany say, even though their referents are the same. The reason for this is to be found in Kripke’s puzzle. The lesson it teaches is that, when belief 15. Or I should say: in the informal explanation of some steps, for reasons given by Burge in the same paper. 16. The example is taken from Kripke (1979, 102–48).

134

ascriptions are involved, no translation from the language of the ascribee to that of the ascriber that fails to meet that constraint can do justice to the rationality of the former.17 Now, direct reference theorists have convinced most contemporary philosophers that the contribution made by a proper name to the propositions expressed by sentences in which it occurs consists precisely in its referent. A suitably related conclusion holds for the names of natural species. If direct reference theorists are right, the proposition expressed, e.g., by Germaniah is in Europe is the same as that expressed by Ashkenaz is in Europe. But then, if preserving content and character were all that there is to the notion of translation, it would be unintelligible how and why the aforementioned constraint can hold for the notion of translation. Incidentally, if it holds true that two utterances express the same thought if and only if they translate each other, then we have a straightforward solution to a puzzle which embarrasses Lewis’ theory of de se belief. Stalnaker, who first pointed it out, stated it as follows. Lingens, still lost in the Stanford Library, meets Ortcutt. “I’ve lost my memory and don’t know who I am,” says Lingens. “Can you tell me? Who am I?” “You are my cousin, Rudolf Lingens,” replies Ortcutt. This seems to be a simple case of direct and successful communication. Lingens requested a certain piece of information; Ortcutt was able to provide it, and did. Ortcutt was sincere — he believed what he said — and Lingens believed what he was told. Furthermore, Ortcutt’s reply was direct: he did not just say something from which Lingens was able to infer the right answer to his question. He told him the answer.18

Lewis’ account of this case is quite complicated. He holds that to make an assertion is to ascribe a property to oneself. Hence the case is to be described as follows: Lingens asks which of a certain set of properties is correctly ascribed to himself. Ortcutt responds by ascribing a different property to himself. Lingens must then infer the answer to his question from Ortcutt’s assertion. Communication, if it is achieved at all, is therefore indirect, and this is not a special feature of the present example: if assertions are always self-ascriptions of properties, then people talk only about themselves. But suppose we are right that translation, as understood above, in particular of indexical statements, preserves thought. Ortcutt’s assertion addressing Lingens, “You are my cousin, Rudolf Lingens”, is 17. Again, see my (2002). 18. Stalnaker (1999, 146).

135

translated by the latter as “I am your cousin, Rudolf Lingens”. As Frege explained, “Although the thought is the same its verbal expression must be different so that the sense, which would otherwise be affected by the different times of utterance, is readjusted. The case is the same with words like ‘here’ and ‘there’”, and, we may add, with ‘you’ and ‘I’. Lingens has no difficulty in understanding these words and communication is achieved straightforwardly. Much remains to be said about this puzzle, but I cannot expand on it now. Let us take stock. Our aim was to establish if the substitutional interpretation of the quantifiers was able to account for inferences involving the notion of sameness of thought, in the absence of an ontology of propositions to which the standard, but not the substitutional, approach can appeal. Our first step was to find a way of capturing the sameness of thoughts without overloading the ontology. Following a suggestion we found in Frege, we claimed that two utterances express the same thought if and only if they stand in the translation relation. We considered some objections to this claim, to the effect that some pairs of utterances exist that express the same thoughts and yet are not in the translation relation, since they differ either in character or in content. We found that these objections are not convincing. In the course of answering them we pointed out that a popular conception of translation, according to which two utterances stand in the translation relation if and only if they express the same proposition, is untenable, no matter how propositions are understood. We have not attempted to define the notion of translation. A definition in terms of other notions, familiar in present-day semantic theory, is not likely to be forthcoming. The examples considered by Burge, of translations that do not preserve reference just in order to preserve self-reference, seem to show that what counts as an acceptable translation heavily depends on which point (or points, in the plural) the speaker is trying to make in uttering a particular statement — e.g., to make a step in a mathematical proof, or to give an example of the phenomenon of self-reference. The notion of the point of an utterance does not seem to be readily amenable to the usual semantic ones. It is not entirely clear if Donnellan’s attributive/referential distinction concerning definite descriptions can be accounted for entirely in terms of the notion of the point of an utterance. Be that as it may, Donnellan’s distinction perfectly matches a difference in the constraints to be satisfied in translating sentences where definite descriptions occur. When a definite description is used referentially, the referent is all-important — certainly

136

more important than the property of being uniquely satisfied by the description’s referent, since the speaker would still be able somehow to refer to the intended object even in cases in which the property is not uniquely satisfied, possibly due to the speaker’s being mistaken in his beliefs. Being so very important, reference clearly takes priority over other semantic features in being preserved in translation. It is the other way around with the attributive reading, since nothing much hinges on the particular referent of a definite description so used, except inasmuch as it satisfies the relevant property. For instance, in ‘Even without knowing who is the shortest spy, everyone believes that the shortest spy is a spy’ the definite description is used attributively. Presumably the point of this utterance is precisely that knowledge of reference is irrelevant here. In translating the utterance, it is crucial that the lexical meanings and syntax be preserved. Thus character takes precedence over reference. Incidentally, I tend to think that this connection between the attributive/referential readings of definite descriptions and the different constraints to be followed in translating, might be the whole story about Donnellan’s distinction. We saw that the definite description ‘the current time’ occurring in (dd) On November 4, 1996, Al believed that Clinton was about to be elected president at the current time, can be taken either referentially or attributively, without the truth conditions of the whole statement being affected. This might suggest that the distinction is not semantic in kind. However, if the same description occurs attributively in a statement uttered on November 4, 1996, then, if we intend to translate it by an utterance that expresses the same thought today, we can leave it as it is — ‘the current time’. On the other hand, if it occurs referentially, then some other expression has to be used today that, in the present context, refers to November 4, 1996. Any co-referring proper name can be used, as Donnellan pointed out in his paper. We have just seen that the inferences: (A) (4a) On November 4, 1996, Al believed that Clinton was about to be elected at the current time (4c) Today Al believes that Clinton is about to be elected at the current time ? Today Al has the same belief as he had on November 4, 1996

137

and (B) (4a) On November 4, 1996, Al believed that Clinton was about to be elected at the current time (4cc) Al believes that Clinton was about to be elected on November 4, 1996 ? Today Al has the same belief as he had on November 4, 1996 are both valid, if the definite description occurring in (4a) is understood appropriately. Now, one can proceed from the conclusions of these two inferences, ‘Today Al has the same belief as he had on November 4, 1996, to p(on November 4, 1996, Al believed that p and today Al still believes that p)’ — the existential quantifier being substitutional — if the following Conjecture holds: CONJECTURE. For every true utterance of the form ‘A believes that p and B believes the same’, there is a clause p*, such that p and p* translate each other, and moreover ‘A believes that p* and B believes that p*’ is also true. If this holds and p* is in the substitution class, then ‘(x)(A believes that x and B believes that x)’ is true. (Here ‘S believes that [’ is short for ‘S believes or believed or will believe, that [ ’). Note that the Conjecture does not claim that, if A utters that p in a given context, then B, no matter where or when he is placed, can say the same, possibly in other words. I can see no reason to believe that this holds true. More modestly, the Conjecture claims that if we, from our point of view, ascribe to A the belief that p, and to B the same belief, then we have the means of saying what it is exactly that B believes, even though A and B are differently located in space and time and, moreover, the very same clause can be used to ascribe to both A and B their belief. In other words, what holds for Al holds in general. The reader can easily verify that, even assuming that the Conjecture holds true, the substitutional approach will not mistake an invalid inference — such as that considered by Soames, from (4ac) and (4b) to (4c) — for a valid one. (Hint: note that ‘Clinton is about to be elected president at the current time’, uttered on November 4, 1996, where the definite description is used referentially, cannot be translated as ‘Clinton is about to be elected president at the current time’, uttered today, no matter how the definite description is used — referentially or attributively)

138

I do not know how the Conjecture can be proved to hold, although I feel that it does and, if this is so, the substitutional interpretation of the quantifiers has nothing to fear. On the other hand, it is not clear that all valid inferences similar to (A) and (B) above can be accounted for by the standard, objectual, approach according to which that-clauses refer to propositions in the range of bound variables. Suppose that Al, who has read a popular exposition of some classical results in metamathematics, believes that (6) above is not a theorem — (6) being (6) (6) is not a theorem. According to the standard approach, then, Al stands in the belief relation with the proposition expressed by (6). Suppose also that Bill has read the German translation of the same exposition, in which the numbering of the examples is exactly the same as in the English original, and believes what it asserts. Like Al, he also comes to believe that (6) is not a theorem — (6) now being (6) (6) ist kein Theorem. Now, do Al and Bill believe the same? Unless we are prepared to endorse a wildly radical form of linguistic relativism, it seems that we are bound to say that they do. We would not change our mind if Bill had only read the second, augmented, edition of the same exposition in English, where that sentence occurs renumbered as (8) (8) (8) is not a theorem. Now, even though (6) and (8) express the same thought, they do not express the same proposition, on any notion of proposition. It is not clear, therefore, how the standard, objectual, quantifers can help us in accounting for the valid inference that Al and Bill believe the same.19

19. I wish to thank Andrea Bianchi, Manuel Garcia Carpintero, Paolo Casalegno, Fabio Del Prete, Andrea Iacona, Max Kölbel, Alberto Voltolini and Sandro Zucchi for many useful comments.

139

REFERENCES Burge, T. 1978. “Self-reference and Translation”. In: F. Guenthner, M. GuenthnerReutter, eds. Meaning and Translation: Philosophical and Linguistic Approaches. Duckworth Donnellan, K. 1966. “Reference and Definite Descriptions”. Philosophical Review, 75, 281–304. Lewis, D. K. 1983. “Attitudes De Dicto and De Se”. Philosophical Papers, 1, 133–60. Kripke, S. 1977. “Speaker’s Reference and Semantic Reference”. In: P. A. French et al., eds. Contemporary Perspectives in the Philosophy of Language, Minneapolis: University of Minnesota Press, 6–27. — 1979. “A Puzzle about Belief ”. In: N. Salmon and S. Soames, eds. Propositions and Attitudes. Oxford: Oxford University Press, 102–48. Markie, P. J. 1988. “Multiple Propositions and ‘De Se’ Attitudes”. Philosophy and Phenomenological Research, 48 (4), 573–600. Perry, J. 1977. “Frege on Demonstratives”. Philosophical Review, 86, 474–97. — 1979. “The Problem of Essential Indexicals”. Nous, 13, 3–21. Richard, M. 1981. “Temporalism and Eternalism”. Philosophical Studies, 39, 1–13. — 1981a. “Tense, Propositions and Meanings”. Philosophical Studies, 41, 337– 351. R. Stalnaker 1999. “Indexical Belief ”. In: Context and Content, Oxford University Press, 1999, 130–149. Soames, S. 1999. Understanding Truth, Oxford: OUP 1999. Santambrogio, M. 2002, “Belief and Translation”. The Journal of philosophy, 99, 624–647.

140

Grazer Philosophische Studien 72 (2006), 141–154.

TRUE IN A SENSE Andrea IACONA Universidad Nacional Autónoma de México Summary The aim of this paper is to show that in order to make sense of the ascription of truth and falsity to the things we say it is essential to acknowledge a divergence between two basic intuitions. According to one of them it is plausible to talk of what is said as what the speaker has in mind. According to the other it is plausible to talk of what is said as the bearer of truth or falsity. The paper presents three cases in which these two intuitions seem not to coincide, and shows how this lack of coincidence can be accounted for in terms of underspecification.

1. It is quite natural to talk of the things we say as other than the sentences we utter to say them. Sentences are strings of words capable of being interpreted. To interpret a sentence is to understand the linguistic meaning of the words it contains and to specify the relevant contextual parameters, that is, the parameters in terms of which those words are linked to the context of utterance. For example, interpreting the sentence ‘it is blue’ in a certain context involves understanding the linguistic meaning of the word ‘it’ and fixing its reference relative to that context. A context is a location where a sentence is uttered: it includes a time, a place, a possible world, a conversational situation, and many other things. The linguistic meaning of a word is its conventional meaning, that is, the meaning it has by virtue of the conventions that are constitutive of the language. What is added to the linguistic meaning of a word by specifying a parameter — say, fixing its reference — is not conventional in that sense, and may vary from context to context. The fact that we interpret sentences in different ways on different occasions makes it plausible to talk of the things we say. Roughly, what is said by uttering a sentence on a certain occasion depends on how the sentence is interpreted on that occasion. The locution ‘what is said’, however, leaves room for two distinct intuitions. One is that, when a speaker assertively utters a sentence, there is

something that he or she has in mind. For example, the sentence ‘it is blue’ may be used by one speaker to say that the sea is blue, and by another speaker to say that the sky is blue. An obvious way of expressing the difference is to say that the two speakers have different things in mind on the two occasions. That is, they mean or want to assert different things. According to this intuition, what is said by uttering a sentence on a certain occasion is a matter of what interpretation can be attributed to the speaker on that occasion, where ‘can’ encompasses a qualification such as ‘correctly’ or ‘rightfully’. The other intuition is that when a speaker assertively utters a sentence, there is something to which truth or falsity can be ascribed. The something in question is naturally understood as a specification of contextual parameters such that, according to it, either the sentence describes things as they are or it describes things as they are not. For example, ‘it is blue’ turns out to be true if ‘it’ refers to the sea, while it turns out to be false if ‘it’ refers to a banana. According to this intuition, what is said by uttering a sentence on a certain occasion is a matter of what interpretation makes the sentence evaluable as true or false on that occasion. A deep-rooted and widespread inclination is to put the two intuitions together. This amounts to thinking that, when a speaker assertively utters a sentence, there is one thing that is both the interpretation that can be attributed to the speaker and a specification of contextual parameters that makes the sentence evaluable as true or false. The widely employed term ‘proposition’ is intended to refer to such a thing. It is usually taken for granted that the proposition expressed by a sentence uttered by a speaker is to be understood in terms of an interpretation that can be attributed to the speaker and involves a specification of contextual parameters that makes the sentence evaluable as true or false. The line of thought advanced in this paper goes in the opposite direction. The aim of what follows is to show that there is a divergence between the two intuitions, and that it is essential to acknowledge this divergence in order to make sense of the ascription of truth and falsity to the things we say. 2. Here are three cases in which it seems that the interpretation that can be attributed to the speaker does not amount to a specification of contextual parameters that makes the sentence evaluable as true or false. Case 1 is as follows. One morning you walk out of your door and see a construction site not too far from your garden. You ask your neighbour what’s going on, and he replies

142

(1) that is a chemical purifier factory. A chemical purifier factory can be a factory that makes chemical purifiers, or one that uses chemical purifiers in making something else. Each of the two possibilities divides into two, for a chemical purifier can purify a chemical or purify by using a chemical. This means that there are four possible readings of the compound noun ‘chemical purifier factory’. However, your neighbour is not considering one of them in particular, as he has not taken into account one of the two distinctions. He saw ‘chemical purifier factory’ explained in a book without mentioning that distinction, and for some obscure reason he thinks that it is appropriate to use it in the present situation. Since the factory under construction happens to be a chemical purifier factory on one of the two readings between which he does not distinguish, it is not clear whether (1) is true or false, as the evaluation of (1) depends on which of them is adopted. Intuitively, (1) is true in some sense and false in some other sense, but no such sense corresponds to what the neighbour has in mind1. Case 2 is as follows. Two philosophers are talking about a seminar held the day before at their department. One of them asks who was present, and the other answers (2) everybody was there meaning that all the members of the department expected to be present were actually present. The fact, however, is that in so answering he has clearly in mind five members of the department, but he doesn’t take into account a sixth member A. If asked whether A was among those expected to be at the seminar, he wouldn’t know what to say. A may be someone on sabbatical, or someone with a visiting fellowship, or someone who gets distracted very easily and tends to avoid seminars. In this case it is not clear whether the contextually restricted domain of ‘everybody’ includes A. In other words, what the second philosopher has in mind does not “decide” between an interpretation according to which A belongs to the contextually restricted domain and one according to which A does not belong to it. But the decision matters to the evaluation of (2) if A was not at the seminar the day before. For in that case (2) is true in one sense and false in another sense. 1. The example is adapted from Sainsbury (2002).

143

Case 3 is as follows. Two kids are in a stationery shop, and one of them utters the sentence (3) it is blue pointing at a bottle of ink. The ink in question happens to be disappearing ink, which writes blue initially but fades to invisibility, and neither of them knows it. In this case the description provided seems not to be finegrained enough to discriminate between normal blue ink and disappearing blue ink. There is a sense in which (3) is true. Presumably, this is how (3) would be interpreted by someone who knew that the ink in question is disappearing ink and wanted to distinguish it from black disappearing ink. There is also a sense in which (3) is false. Presumably, this is how (3) would be interpreted by someone who knew that the ink in question is disappearing ink and wanted to point out that it does not leave permanent blue marks on the paper. But neither of these two interpretations can be attributed to the kid2. Each of the three cases presented involves some form of context-dependence. This is why examples like these are sometimes used to make a point about the effects of context on truth conditions, and hence to provide evidence in support of this or that account of such effects. A widely accepted account is that on which the truth condition of a sentence depends entirely on some assignment of reference to elements of its syntactic structure or logical form. If one regards such an assignment as part of “semantics”, the view is that the effects of context on truth conditions can be described in purely semantic terms. By contrast, according to various forms of contextualism, the determination of the truth condition of a sentence as it is uttered in a certain context is in some important sense “pragmatic”, in that it does not depend simply on some assignment of reference to its expressions relative to that context. The point of this paper, however, is not about the effects of context on truth conditions, and does not rest on a particular account of such effects. Rather, it rests on an assumption that is neutral and hence capable of being shared by proponents of different accounts. The assumption is that a sentence is evaluable as true or false in a certain context just in case some parameters are specified with respect to that context. For any number n of parameters that are relevant for the evaluation of a sentence, a specification of them may be 2. The example is adapted from Travis (1999).

144

represented as an assignment of a n-tuple (of the appropriate kind) to the sentence. Thus we have that given any sentence and any context, there are n parameters such that the sentence is evaluable as true or false relative to the context just in case some n-tuple (of the appropriate kind) is assigned to it. To assume that given any context there is a set of parameters whose specification determines a truth condition relative to that context is not the same thing as to assume that there is a set of parameters such that given any context, their specification determines a truth condition relative to that context. On the first assumption, it is conceivable that different kinds of parameters need be specified on different occasions in order for the same sentence to be evaluable as true or false on each of them. Compare case 3 with one in which (3) is used on the beach to describe the colour of the sea. What the assumption requires is simply that in each of the two cases there is a set of parameters whose specification determines a truth condition for (3). This does not necessarily mean that the parameters involved in the two cases are of the same kind. It is not essential for the purposes at hand to decide how exactly truth conditions are determined by specifying parameters. It is not even required that there is a simple and straightforward way to account for such determination. The amount and assortment of parameters might be such as to make it very hard, if not impossible, to spell out how it works. Given the assumption that a sentence is evaluable as true or false in a certain context just in case some parameters are specified with respect to that context, cases 1-3 may be described as cases in which the two intuitive notions of what is said do not coincide. On the one hand, in each of them some interpretation can be attributed to the speaker. The speaker has something in mind in uttering the sentence, and what he or she has in mind involves specification of some parameter or basic set of parameters. For example, in case 1 the reference of ‘that’ is fixed. The same goes for ‘there’ in case 2 and ‘it’ in case 3. In this respect, cases 1-3 are unlike one in which the sentence ‘she is there’ is written on the board to make a point of grammar. On the other hand, however, in cases 1-3 the interpretation that can be attributed to the speaker does not amount to a specification of parameters that makes the sentence evaluable as true or false. The sentence so interpreted is not — directly or immediately — evaluable as true or false, in that its evaluation depends on the specification of parameters that the speaker does not take into account, and that are left open by the linguistic meaning of the words it contains. This is why it seems correct to say that

145

the sentence is true in some sense and false in some other sense, even if no such sense clearly corresponds to what the speaker has in mind. 3. Let the actual interpretation of a sentence uttered on a given occasion be the interpretation that can be attributed to the speaker on that occasion. Let a sufficiently specified — in short, ss — interpretation of a sentence uttered on a given occasion be an interpretation that is specific enough for the purpose of ascribing truth or falsity to the sentence on that occasion. Cases 1-3 may be accounted for in terms of underspecification, that is, they may be described as cases in which the actual interpretation is not a ss-interpretation. For example, in case 1 the actual interpretation of (1) is not a ss-interpretation. A ss-interpretation of (1) is one that provides a definite meaning for ‘chemical purifier factory’. But any such interpretation goes beyond the actual interpretation of (1), in the sense of not being uniquely determined by it. The neighbour is not to be understood as having in mind some specification according to which, say, ‘chemical purifier factory’ stands for a factory that makes stuff that is used to purify chemicals. Similarly, in case 2 the actual interpretation of (2) is not a ss-interpretation. For a ss-interpretation of (2) is one that assigns a domain of quantification to ‘everybody’. The same goes for case 3, where a ss-interpretation of (3) is one that settles the question of the permanency of the ink on the paper. In each case, ss-interpretations amount to alternative ways in which the sentence uttered can be interpreted. Intuitively, they are “senses” in which the sentence is true or false. The account may be phrased in more formal way. Let a valuation V be a function that assigns to each sentence s a n-tuple such that s is evaluable as true or false relative to it. We say that s is true on V if and only if s is true relative to the n-tuple that V assigns to s. Interpretations are sets of valuations. So the actual interpretation of s as it is used on a certain occasion is a set of valuations that is determined by the linguistic meaning of the words occurring in s and by what the speaker has in mind on that occasion. Interpretations are sets of valuations rather than single valuations in that there is a certain amount of indeterminacy in the way we speak. Take (2) as it is uttered in case 2. The philosopher uses ‘there’ to refer to a certain location, namely, the philosophy department. But his use of ‘there’ does not determine exactly which object is referred to, as it does not determine exactly where is the philosophy department. There can be two ways of delimiting the philosophy department such that according

146

to one of them a certain wall belongs to it while according to the other it doesn’t. If we call d1 and d2 the two objects obtained by adopting each of them, then it is indeterminate whether ‘there’ refers to d1 or to d2. Thus, there are at least two valuations V1 and V2 such that one of them assigns d1 to ‘there’ while the other assigns d2. The actual interpretation of (2) includes both. Let two valuations V1 and V2 overlap on s when they are alike as far as the truth or falsity of s is concerned. A ss-interpretation of s is a set of valuations that overlap on s, where the set is constrained by the linguistic meaning of the words occurring in s. Accordingly, truth relative to a ss-interpretation is defined in terms of truth on a valuation. For any ss-interpretation D of s, s is true relative to D if and only if it is true on all the valuations that belong to D. Similarly, s is false relative to D if and only if it is false on all the valuations that belong to D. Let an interpretation D be compatible with an interpretation E if and only if D is a subset of E. A normal case is one in which there is a unique ss-interpretation that is compatible with the actual interpretation, namely, the actual interpretation itself. By contrast, a case of underspecification is one in which there are at least two ss-interpretations that are compatible with the actual interpretation and differ in truth value. This is easily seen as follows. If the actual interpretation D of s is not a ss-interpretation, there are at least two valuations V1 and V2 in D that do not overlap on s. So there are at least two ss-interpretations that are subsets of D and assign opposite truth values to s, namely {V1} and {V2}. Inversely, if there are at least two ss-interpretations that are subsets of D and assign opposite truth values to s, then D is not a ss-interpretation. For given an arbitrary member V1 of one of them and an arbitrary member V2 of the other, V1 and V2 belong to D, although they do not overlap on s. 4. From § 3 it turns out that cases 1–3 may be described as cases in which there are at least two ss-interpretations that are compatible with the actual interpretation and differ in truth value. But nothing has been said so far about what happens exactly in each of them. Given that ss-interpretations are sets of valuations, the details of the description depend on what kind of functions valuations are, and what kind of parameters are involved in each case. Different ways of answering these questions amount to different ways of construing ss-interpretations. To illustrate, this section outlines one possible definition of valuation, and shows how cases 1–3 can be

147

described in accordance with it. The definition employs three standard notions: logical form, index and point of evaluation. A logical form is a linguistic representation constituted by a lexically and structurally disambiguated sequence of word types, where word types are individuated both by syntactic and by semantic properties. In addition to word types, a logical form can include open positions occupied by variables, which are covert at the level of surface grammar. The semantic properties of the word types that constitute a logical form are given in terms of functions that represent their linguistic meaning, or at least some essential trait of their linguistic meaning. The semantic properties of logical forms themselves result from the composition of those of their constituents, and are functions that represent how the truth conditions of the corresponding sentences depend on features of context3. The semantic contribution of a context to the truth condition of a sentence can be expressed in terms of an index, that is, a sequence of features of various sorts that we may “extract” from the context. Every logical form has a number of open positions occupied by elements whose value may vary from context to context. These may be either overt context-dependent expressions, such as indexicals, pronouns and demonstratives, or covert variables. Accordingly, an index is a number of values for those elements. That is, if a logical form has n open positions, then an index for it is a sequence of n contextual features. Call these features the coordinates of the index. For example, in case 2 the reference of ‘there’ must be fixed in order for (2) to be evaluable as true or false. Therefore, an index for the logical form of (2) must contain a location coordinate, such as a philosophy department. A point of evaluation is also a collection of contextual features, but not necessarily features that go together as part of one and the same context. They are so combined that they can shift independently of one another. For example, a point of evaluation may include a time and a possible world. Normally, these are the time and possible world at which the sentence is uttered, hence they belong to the set of features that the index includes as coordinates. But nothing prevents them from being different from the coordinates of the index. It may be the case that the point of evaluation includes a time that is not that at which the sentence is uttered, or a possible world that is not the actual world4. 3. The notion of logical form is essentially that spelled out in Stanley (2000). 4. The index-point apparatus is essentially that outlined in Kaplan (1989) and Lewis (1998).

148

Given these three notions, a valuation V can be defined as a function that assigns to each sentence a triple , where l is a logical form with n open positions, i is an index with n coordinates and p is a point of evaluation. The semantic property of l is a function Il from index-point pairs to the values 1 and 0. Truth on a valuation is defined as follows: s is true on V if and only if the triple assigned by V to s is such that Il(i,p) = 1. Similarly, s is false on V if and only if is such that Il(i,p) = 0. A ss-interpretation of s turns out to be a set of valuations such that any two valuations V1 and V2 in it assign to s two triples and such that Il1(i1,p1) = Il2(i2,p2). This definition of valuation leaves room for different ways of describing the forms of context-dependence involved in cases 1–3. One is to assume that in each of them, the truth condition of the sentence uttered is determined by some assignment of reference to elements of its logical form. This means that the different senses in which the sentence uttered is true or false are a matter of different ways in which the reference of some element in its logical form is fixed. For example, in case 1 one may suppose that some covert variable is associated with ‘chemical purifier factory’. This means that the logical form of (1) is something like [that]NP[is a chemical purifier factory (x)]VP where x takes values corresponding to readings of ‘chemical purifier factory’. Similarly, in case 2 one may suppose that a covert variable takes care of the quantifier domain restriction. This means that the logical form of (2) is something like [every body (x)]NP[was there]VP where the value of x gives us the relevant domain5. In case 3 one may suppose that a covert variable involved in the predication takes values corresponding to different ways in which something can “count” as being blue. This means that the logical form of (3) is something like [it]NP[is blue (x)]VP6 5. The account of quantifier domain restriction given in Stanley and Szabó (2000 ) is of this kind. 6. Szabó (2001) provides a treatment of adjectives along these lines.

149

Thus each of the three cases can be described by saying that there are at least two valuations V1 and V2 that belong to the actual interpretation, where V1 and V2 do not overlap because they have different indices, that is, indices that differ for one coordinate. This means that V1 and V2 assign to the sentence uttered two triples and such that l1 = l2 and p1 = p2 but i1 z i2, so Il1(i1,p1) z Il2(i2,p2). Therefore, there are at least two ss-interpretations that are compatible with the actual interpretation but differ in truth value. Here the responsibility for the non-overlap falls upon the index because we assumed that the truth condition of the sentence uttered is determined by some assignment of reference to elements of its logical form. But this assumption is not essential to the definition of valuation under consideration. For example, it is equally compatible with it to suppose that in case 1 ‘chemical purifier factory’ is to be disambiguated at the level of logical form. This means that different logical forms may be assigned to (1). Call two of them l1 and l2: [that]NP[is a chemical purifier factory1]VP [that]NP[is a chemical purifier factory2]VP The case may be described as involving two valuations V1 and V2 that assign to (1) two triples and such that i1 = i2 and p1 = p2 but l1 z l2, so Il1(i1,p1) z Il2(i2,p2). Similarly, in case 3 the intuition that there are two senses in which the ink can count as being blue may be captured in terms of a difference in the point of evaluation. In this case the supposition is that there are two valuations V1 and V2 that assign to (3) two triples and such that l1 = l2 and i1 = i2 but p1 z p2, so Il1(i1,p1) z Il2(i2,p2)7. More generally, all that is needed for the definition to be employed in the description of a case of underspecification is that relevantly different valuations belong to the actual interpretation. It doesn’t really matter what in the respective triples makes the difference. 5. The account outlined in § 3 rests on the thesis that truth and falsity apply to sentences relative to ss-interpretations. This can be construed in two ways, depending on how ‘what is said’ is understood. According to the first intuitive notion of what is said considered in § 1, the thesis is that the things we say are evaluable as true or false only relative to ss7. An analysis of this kind is offered in Predelli (2005).

150

interpretations of the sentences that we utter to say them. In other words, the ascription of truth or falsity to the things we say may require ways of understanding them that go beyond the way we understand them. According to the second intuitive notion of what is said considered in § 1, instead, the thesis is that the things we say are ss-interpretations of the sentences that we utter to say them. In other words, the things we are capable of saying by using sentences may go beyond our understanding of those sentences. If one accepts that truth and falsity apply to sentences relative to ssinterpretations, one can make sense of ordinary talk about assertions and their evaluation in terms of attribution of ss-interpretations to speakers. A first intuition that distinctly emerges from that talk is that there are cases in which a speaker assertively utters a sentence and we take the assertion to be correct. Let the condition of attribution be that an interpretation D can be attributed to a speaker S just in case only D is compatible with the actual interpretation manifested by S. Given this condition, the intuition is captured by saying that there are cases in which a ss-interpretation can be attributed to the speaker, and the sentence uttered is true relative to it. A second intuition that distinctly emerges from ordinary talk about assertions and their evaluation is that there are cases in which a speaker assertively utters a sentence and we take the assertion to be incorrect. Given the same condition, this is captured by saying that there are cases in which a ss-interpretation can be attributed to the speaker, and the sentence uttered is false relative to it. Cases 1–3 belong to neither of these two categories, in that they are cases in which no ss-interpretation can be attributed to the speaker. On the assumption that one asserts what one says, two descriptions of them are possible, depending on which of the two intuitive notions of what is said considered in § 1 is adopted. If the first notion is adopted, cases 1–3 are cases in which a unique thing is asserted, but no ss-interpretation relative to which it is true or false can be attributed to the speaker. In other words, the assertion is not evaluable. Instead, if the second notion is adopted, cases 1–3 are cases in which no unique thing is asserted. This can be construed in two ways. One is to say that more than one thing is asserted, in that more than one ss-interpretation is compatible with the actual interpretation. The other is to say that nothing is asserted, in that no ss-interpretation can be attributed to the speaker. Both options entail that there is no such thing as “the” assertion to be evaluated.

151

It doesn’t really matter which of the two notions of what is said is adopted to describe cases 1–3. What really matters is that cases 1–3 are problematic in that no ss-interpretation can be attributed to the speaker. They involve a problem of attribution rather than a problem of truth and falsity. Or at least, if there is a problem of truth and falsity, it is only because there is an underlying problem of attribution. Underspecification is compatible with the principle of bivalence in the obvious sense that, relative to ss-interpretations, truth and falsity are mutually exclusive and jointly exhaustive values. The definition of truth relative to a ss-interpretation entails that for any ss-interpretation of a sentence, either the valuations that belong to it are all such that the sentence turns out true or they are all such that the sentence turns out false. 6. What has been said so far entails that there are cases of underspecification. This entails in turn that there are cases in which the speaker does not have a clear idea of the truth condition of the sentence uttered. For it is evident that if the speaker has a clear idea of the truth condition of the sentence uttered, the actual interpretation is a ss-interpretation. However, the claim that there are cases of underspecification is not to be confused with the claim that there are cases in which the speaker does not have a clear idea of the truth condition of the sentence uttered. For a case of the latter kind is not ipso facto a case in which the actual interpretation is not a ss-interpretation. The actual interpretation is the interpretation that can be attributed to the speaker, and this is not quite the same thing as what the speaker has in mind. So the actual interpretation may be sufficiently specific even if what the speaker has in mind is not. The fact that a speaker may use a sentence without having a clear idea of its truth condition has been largely emphasized by “externalist” theories of meaning. According to such theories, the determination of the truth conditions of the sentences we utter involves “external” factors that are to a good extent independent of the mental contents we associate with the words occurring in those sentences. For example, it has been argued that the meaning of the word ‘water’ is fixed in such a way that it is an essential part of it that the word refers to a substance whose chemical composition is H2O. So independently of the mental contents that speakers associate with the word, the truth condition of a sentence containing it depends on its relation to H2O. For example, even if someone doesn’t know the chemical composition of water, when he is at the restaurant and says ‘I

152

would like a glass of water’, it is correct to interpret him as expressing the desire for a glass of H2O. More generally, it has been argued that in our linguistic community the meaning of some words is fixed in such a way that, even if we don’t have full mastery of those words, the sentences we may utter containing them are to be interpreted in accordance with that meaning by virtue of some form of “deference” to more competent members of the community8. The claim that there are cases of underspecification is compatible with externalism so understood, in that it allows that for some class of cases, the meaning of the words occurring in the sentence uttered, in combination with external factors, may help determine the actual interpretation independently of what the speaker has in mind. What it does not allow is that all cases in which the speaker fails to have a clear idea of the truth condition of the sentence uttered can plausibly be treated that way. Cases such as 1–3 resist the generalization. In case 1 no form of deference seems involved, so it is hard to see how an externalist conclusion about ‘chemical purifier factory’ can be justified. Usually, the arguments invoked to justify such conclusions rest on solid intuitions about reference or truth conditions. But here there is no intuition to rest on. No reading of ‘chemical purifier factory’ seems “the” reading to be attributed to the neighbour. It is easy to contrast case 1 with one that involves a clearly deferential use of (1). For example, the neighbour could have uttered (1) to express information about the factory had by an engineer working at the construction site. Note that in such a case there would be intuitions. That is, (1) would be true according to the reading of ‘chemical purifier factory’ intended by the engineer. Similar considerations hold for cases 2 and 3. It may certainly be contended that one of the competing interpretations of (2) or (3) is “the” correct interpretation. But the question is: which? Unlike the cases usually constructed to justify externalist claims, here no intuition pushes us one way or the other. In general, there seems to be no reason to think that the externalist treatment that applies to some cases should providentially apply to all cases. So there seems to be no reason to exclude that there are cases of underspecification.

8. Putnam (1975).

153

REFERENCES Kaplan, D. 1989. “Demonstratives”. In: J. Almog, J. Perry, and H. Wettstein, eds. Themes From Kaplan. Oxford: Oxford University Press. Lewis, D. 1998. “Index, context, and content”. In: Papers in Philosophical Logic. Cambridge: Cambridge University Press. Predelli, S. 2005. “Painted leaves, context, and semantic analysis”. Linguistics and Philosophy 28, 351–374. Putnam, H. 1975. “The meaning of ‘meaning’”. In: K. Gunderson, ed. Minnesota Studies in the Philosophy of Science, Vol. 8. Minneapolis: University of Minnesota Press. Sainsbury, M. 2002. “Two ways to smoke a cigarette”. In: E. Borg, ed. Meaning and Representation. Oxford: Blackwell, 94–114. Stanley, J. 2000. “Context and logical form”. Linguistics and Philosophy 23, 391– 343. Stanley, J. and Szabó, Z. G. 2000. “On quantifier domain restriction”. Mind and Language 15, 219–261. Szabó, Z. G. 2001. “Adjectives in context”. In: R. Harnish and I. Kenesei, eds. Perspectives on Semantics, Pragmatics, and Discourse. Amsterdam: John Benjamins, 119–146. Travis, C. 1999. “Sublunary intuitionism”. In: P. M. Sullivan and J. Brandle, eds. New Essays on the Philosophy of Michael Dummett, Grazer Philosophische Studien Special Issue, Amsterdam/Atlanta: Rodopi.

154

Grazer Philosophische Studien 72 (2006), 155–178.

SPEECH ACTS WITHOUT PROPOSITIONS? Marina SBISÀ Università di Trieste Summary This paper argues that understanding speech in terms of action requires dispensing with propositions. Austin’s outline of speech act theory did not give any role to propositions, which were introduced into speech act theory later on, in order to cope with criticism leveled by Strawson and Searle at Austin’s characterization of the locutionary act and his view of the truth/falsity assessment. The introduction of propositions had weakening effects on the claim that speech is action, foregrounding again the received picture of linguistic communication. I show that, in order to make sense of Austin’s characterization of the locutionary act, propositions are not needed and give some suggestions as to how one could give an account of the truth/ falsity assessment, compatible with the claim that speech is action, without resorting to propositions.

In this paper I analyse the relationship between speech act theory and propositions, which I find to be somewhat paradoxical. On the one hand, it is generally held that speech act theory needs to use some notion of propositional content, but on the other hand, introducing propositions into it has the effect of weakening its claim that speech is action, eventually making it empty. Those who believe that speech act theory is a valuable enterprise, that is, that it is worth trying to study speech as action, should therefore beware of propositions and possibly attempt to dispense with them. Starting from the beginnings of speech act theory, I will show why and how propositions were introduced and comment on the effects of this introduction on the conception of the speech act and the overall significance of speech act theory. In conclusion, I discuss how and to what extent speech act theory can dispense with propositions.

1. From the locutionary act to the propositional content Propositions were introduced into speech act theory for two main reasons. First, as an improvement or amendment of Austin’s way of drawing the distinction between the locutionary and the illocutionary act. Second, because of their role as truth-bearers. There is no mention of propositions in Austin’s initial characterization of the “locutionary act” or act of saying something “in the full sense of ‘say’” (Austin 1962, 92)1. His characterization does not amount to an explicit definition of the act of saying something, but spells out three ways in which to say something “must always be to do something”, namely: (a) the “phonetic act”, i.e. the act of uttering certain noises; (b) the “phatic act”, i.e the act of uttering certain vocables or words in a certain construction and with a certain intonation; (c) the “rhetic act”, i.e. the act of using the utterance or its constituents with a certain more or less definite “sense” and a more or less definite “reference” (which together are equivalent to “meaning”). (Austin 1962, 92–93) It should be noted that in his definition of the “phatic act”, Austin avoids making reference to sentences, possibly because not all utterances which he has been dealing with in his exploration of performatives are sentential in form. Correspondingly, in the definition of the “rhetic act” he avoids mentioning propositions. Although this omission can be seen as a merely superficial choice (due to unwillingness to limit oneself to sentential utterances, or to preference for ordinary as opposed to philosophical vocabulary), it is suspicious. If the problem were only how to take into account non-sentential utterances, the word “proposition” could have been used with some qualification. Moreover, the problem cannot be with technical terminology, since Austin, in these very pages of How to Do Things with Words, does not refrain from introducing technical terminology and even neologisms. So the omission of any mention of propositions throughout the characterization of the locutionary act is likely to be philosophically significant. Austin’s characterization of the locutionary act has been found unsatisfactory, and with reason. It is certainly sketchy and underexplained. 1. Page numbers in references to Austin (1962) belong to the 2nd ed., 1975.

156

The way in which the component acts are supposed to yield the complete locutionary act is not specified (do they sum up as parts of a whole? does each act include the preceding ones in a larger whole?). Worse (and this is admitted by Austin himself: 1962, 149) no definition or analysis is provided for the key terms “sense” and “reference”. These and other shortcomings have played some role in generating dissatisfaction and stimulating criticism and reformulation. I mention here three problematic features that have opened the way to the well-known reformulation of the Austinian distinctions by John R. Searle (1968, 1969). (1) The definition of the phatic act sounds somewhat implausible. Why should uttering sounds that are words be an act distinct from merely uttering those very sounds? Austin includes in his characterization of the phatic act the further condition that the words should be uttered “as conforming” to the rules of some language (1962, 92). But does the awareness that one is speaking a language suffice to constitute an additional act on the part of the speaker? Perhaps because of these or similar perplexities, the phonetic-phatic distinction is not picked up by Searle and both acts are replaced by the “utterance act” (Searle 1969, 25). (2) It is easy to see that, once (intended) “sense” and “reference” are established, the speaker may already be said to have “said” something in the full sense of the “locutionary act”: there is nothing else to add to the utterance act in order to get what the speaker has said. Therefore, the “rhetic” act cannot be a genuine component of the locutionary act, rather, it seems to coincide with the complete locutionary act. Authors discussing the Austinian notion of the locutionary act have therefore felt they could legitimately rely upon his remarks concerning the rhetic act (Searle 1968; see also Strawson 1973, 55). (3) When Austin illustrates the “rhetic” act, he cites as reports of such acts oratio obliqua reports such as “He told me to get out”, “He asked whether it was in Oxford or Cambridge”. This undermines the plausibility of his locutionary/illocutionary distinction, because telling someone to do something (that is, requesting or commanding him or her to do something) and asking a question are unquestionably illocutionary acts (Searle 1968, 147). Apart from non-serious utterances and performance failures, there can be no locutionary act which is not at the same time an illocutionary one. Thus, in Searle’s own theory (1969, 23), the illocutionary act becomes the only actual

157

communicative unit, whose characteristic linguistic form is the complete sentence and which is to be analysed into components, none of which corresponds to Austin’s locutionary act or rhetic act (Searle 1968, 148–151). It is at this point that Searle introduces the notion of a proposition. According to Searle (1968,156), while the rhetic act or the locutionary act cannot be genuinely abstracted from the complete illocutionary act, the propositional act, that is the act of expressing a proposition, is a genuine abstraction from it, since only certain portions of the uttered sentence actually contribute to the expression of the proposition. Therefore, the real distinction to be drawn is that between the illocutionary act and the propositional act (Searle 1968, 159). Such an act is performed in uttering a sentence, like the illocutionary act, and consists, like the illocutionary act, in the uttering of words in sentences in certain contexts, under certain conditions and with certain intentions (Searle 1969,25), but it cannot occur alone; rather, it is contained in the illocutionary act, just as a complete sentence contains referring and predicating expressions. In conclusion, for Searle (and the speech-act theoretical tradition to follow), the illocutionary act has both an illocutionary force and a propositional content to which the force applies, which is provided by the proposition expressed by the propositional act. As Searle himself notes (1969, 30, 1968,155–59), the distinction between force and propositional content is not new. In fact, it enjoys a long tradition in philosophy and has been endorsed and elaborated both in phenomenology and in analytic philosophy since the last decades of the 19th century. Frege defended it from his Begriffschrift (1879) to his later essays. Other uses of it, which are in various respects akin to Searle’s and have been influential in analytic philosophy, are to be found in Hare (1952) and Stenius (1967). By Searle’s move, the conception of the speech act is brought back to the mainstream Fregean tradition, from which Austin had attempted to escape. As the presence of similar distinctions in a larger philosophical context may suggest, the introduction of the distinction between force and propositional content in speech act theory is not motivated merely by the shortcomings of the locutionary/ illocutionary distinction. These independent motivations concern the role of the proposition as truth-bearer. Austin’s views on truth combined the defence of a version of the correspondence theory with the rejection of the redundancy thesis (accord-

158

ing to which “It is true that …” does not add any new meaning to the sentence to which it is attached) and the claim that truth evaluation is not a binary assessment, but involves a whole gamut of degrees and shades (Austin 1950; see also 1962, 140–147). All these claims were controversial and provoked a hard-hitting and lengthy debate, which saw P. F. Strawson as Austin’s main interlocutor (see Pitcher ed. 1964). As to truth-bearers in particular, Austin (1950) claimed that “that which at bottom we are always saying ‘is true’” is the statement, which in turn, in his discussion of assertion as an illocutionary act (1962, 133ff.), he claimed to be a speech act on a par with any other. At the same time, he also admitted of a connection between a speech act’s having a locutionary meaning (sense and reference) and its being assessable in the dimension of truth and falsity (e.g. 1962, 145–146). It is quite natural and in general accordance with the views on truth in Austin (1950), to suppose that this connection lies in the fact that the truth or falsity of the statement depends on its locutionary meaning: this would explain why, as Austin himself recognizes, the dimension of truth and falsity is foregrounded whenever hearers (or analysts) concentrate on the locutionary meaning of a speech act. But then the locutionary meaning, albeit a feature of the total speech act, should play a particular role in it, namely, provide the asserted content, that which is stated, that which is actually said to be true or false. Such a reading of Austin’s accounts of assertion and truth is basically present both in the above-mentioned discussion of the locutionary act by Searle (1968) and in the close scrutiny of “locutionary meaning” by Strawson (1973). Searle (1968,158) openly reproaches Austin for failing to distinguish the statement-act from the statement-object: to say that the statement is the ultimate truth-bearer, cannot mean that the statement as a speech act has a truth-value: rather, that which has truth-value is what is stated, the statement-object, the (propositional) content of the statement. Strawson emphasizes the similarities of Austin’s distinction between force and meaning to Frege’s distinction between assertion and thought or proposition (1973, 53, 56) and suggests as a possible line of development of Austin’s theory, which he deems not completely foreign to his aims, that the locutionary meaning should be identified, in the case of assertions at least, with the proposition expressed. The introduction of propositions into speech act theory was extremely successful. It fitted well the Gricean turn which characterized speech act theory in the 1970s, in which speech acts, as expressions of overt communicative intentions, became basically vehicles for the transmission of

159

propositional attitudes. It was compatible with Grice’s adoption of a technical sense of “saying”, according to which what is said by an utterance corresponds to its truth-conditional meaning. In their influential version of speech act theory, Bach and Harnish recognize the notion of propositional content as indispensable (1979, 9), consider “what is said” by an utterance as determined by sentence mood and propositional content (1979, 24), and apply the notion of propositional content at two different stages of their schema for speech act understanding, namely, as the propositional content conveyed by the sentence uttered at the locutionary level and as the propositional content to which the actual illocutionary force of the utterance (to be inferentially derived) applies (1979, 11, 66). The introduction of propositions has also proved compatible with views of speech acts that had developed independently, such as William Alston’s (Alston 1964, 2000), developing, but at the same time counterbalancing his earlier claim that the meaning of a sentence is its illocutionary-act potential, Alston (2000) stresses contentfulness as the main defining feature of the illocutionary act. 2. Proposition and action What is the point of having a speech act theory? It is, I believe, to introduce action into our consideration of speech. So, it makes sense to have a speech act theory insofar as it helps us view speech as action. By action, we might mean mere activity (that is an active, as opposed to passive, condition of an individual organism), in which case any description of utterance processing as an active process would meet the requirement. Less trivially, speech is viewed as action when it is considered as one way in which we bring about, or change, states of affairs in the world. This is the sense of “action” in which Austin was interested, as is shown by his frequent resort to the notion of effect: effects are part of the “conventional procedures” by reference to which he analyses performatives (1962, 14), illocutionary acts are said by him to “take effect” in this same sense (1962, 117) and perlocutionary acts (1962, 101, 107–8) are basically identified by tracing certain consequential effects of the speech act back to the speaker’s responsibility. Viewing speech as action in such a strong sense enables us to notice how, thanks to speech, we relate ourselves to each other, build up social bonds, establish and change interpersonal relations by assigning or canceling rights, obligations, commitments, and legitimate expectations,

160

and even constitute new states of affairs of an institutional kind. One of the possible lines of investigation thus opened up is the exploration of the role of language in creating a level of reality which is cultural rather than merely natural: a world of states of affairs describable by assigning deonticmodal predicates to social agents. If considered as creating and modifying states of affairs of this kind, illocution enables us to analyse this level of reality in its initial state, that is, as it arises in interpersonal agreement or coordination as to how utterances are to be taken, and helps us see how widespread its influence is, often involving a mediating function with respect to other ways of acting upon one another, such as psychological influence and even coercion.2 Consideration of speech as action in this sense, especially if it is proposed in the fully general way argued for by Austin, suggests a critical reassessment of the aspiration of assertive (or representative or constative) uses of language to count as the locus of purely factual, true/false belief and possibly of objective, scientific knowledge. Even when practised in the search for truth, speech should be seen as a special way in which human subjects operate in context-bound situations that include the social dimension. If assertions too are speech acts, and more specifically, illocutionary acts having felicity conditions and effects, the level of interactionally constructed reality to which illocution gives access should play a role also in the elaboration and transmission of knowledge, at least insofar as language is involved in these processes. Is the introduction of propositions in speech act theory compatible with a full consideration of speech as action (which I will subsequently refer to as “strong speech act view”)? The intention declared by Searle (1968) and Strawson (1973) while introducing propositions was that of improving and amending Austin’s theory. But do the philosophical specificity and the potentialities of the theory survive the correction? Searle seems confident that they do. Strawson, with deep philosophical sensitivity, detects in what Austin writes the traces of a project which would be incompatible with the proposed 2. I have been exploring the heuristic powers of such an approach in several fields, from ordinary informal and semi-formal interaction (Sbisà 2001, 2002) to literary criticism (Sbisà 2003a) and socio-political analysis (Sbisà 2006) (see also the framework outlined in Sbisà (1989)). Obviously enough, the micro-sociological study of verbal interaction and linguistic anthropology have their own approaches to these very issues, not necessarily using speech-act theoretical notions of any kind: my contention is that an Austin-inspired speech act theory offers specific contributions to their study.

161

amendments. In fact, the introduction of propositions gives rise to a weakened speech act view, one in which certain claims raised by Austin cannot be made or become highly implausible and certain consequences can no longer be drawn. In what follows, I outline some main effects induced in speech act theory by the introduction of propositions. 2.1. From action to communication The introduction of propositions in speech act theory created the conditions for turning the effect of the illocutionary act from a full-fledged effect (a change in states of affairs obtaining in the world, albeit of a peculiar kind) to a merely cognitive one. Assuming that there are propositions and that language expresses them independently of its expressing force(s) as well, a speech act may very naturally be described as an act of expressing a proposition with a certain communicative intention applying to it. Speech act successfulness may very naturally be identified with the recognition of the speaker’s communicative intention on the part of the hearer, which is a merely cognitive effect (in particular, a matter of acquiring a certain belief about the speaker). As a result, the communication that the speaker is performing an action is foregrounded at the expense of the very action performed. Also, any other effect that the speech act may turn out to have beyond the recognition of the speaker’s communicative intention is considered as exceeding the limits of the illocutionary act, that is, not a matter of illocution but of perlocution; not a matter of the speech act as such, but of its extralinguistic surroundings. The main traits of this evolution are already present in Searle (1969, 47) and are reproduced, in spite of the remarkable differences between the two theories, in Bach and Harnish (1979). So, for example, the only effect of the command “Open the door” as an illocutionary act is the cognitive effect amounting to the recognition of the speaker’s communicative intention to make the hearer form the intention to open the door. Of course, it may also be said that the command has the goal of making the speaker open the door, but whether this goal is achieved is a matter of perlocution as opposed to illocution. The fact that the command puts the addressee in an intersubjectively recognized, more or less strict state of obligation to open the door has no room in this description of the command, while it is the most obvious candidate for filling in the role of its illocutionary

162

effect according to Austin’s characterization of effects of that kind. Thus, the communication that the speaker has expressed a certain communicative intention is foregrounded at the expense of the action that the speaker may be said to have performed and of its effect of assigning a certain obligation to the addressee. A further consequence of this situation is that it becomes hard to detect, in the speech act, any action at all. If the effect of the speech act, which would be essential to its being an action, is identified with the successful communication of what action it was that the speaker intended to perform (that is, e.g., an act of commanding), where is the action itself? Or is it after all identified, according to a widespread common-sense view of communication, with the transmission of a certain proposition under a certain mood (in the case of the command, as representing the state of the world that the speaker would like the addressee to bring about)? The analysis of illocutionary classes in Searle (1975a) goes decidedly in this direction. But the most consequent, extreme development of this conception is to be found outside speech act theory proper, in the view of speech acts proposed by Relevance Theory (Sperber and Wilson 1986). There, all communication is transmission of propositions (in the fairly sophisticated form of letting the hearer inferentially reconstruct the proposition or set of propositions that the speaker intends to communicate) and sentence mood or other force-indicating devices enable the hearer to grasp the direction in which the relevance of the utterance is to be sought and to infer from it a proposition that describes it as a speech act of a certain kind (e.g., “The speaker has commanded me to open the door”). 2.2 Mutual exclusiveness of expression of meaning and indication of force The introduction of propositions has enabled Searle to propose the formula f (p) as the general structure of a speech act, where “f ” stands for illocutionary force and “p” for propositional content. It would be naive to expect that sentences can neatly split into two elements or groups of elements, the one expressing propositional content, the other functioning as force indicator. But the idea underlying the formula is that the two functions

163

are mutually exclusive, and no single linguistic element can serve both. Searle himself notices that only certain portions of the uttered sentence actually contribute to the expression of the proposition (1968, 156), and this is one of his reasons for preferring the “propositional act” or act of expressing a proposition to Austin’s locutionary act in the analysis of the whole speech act. This suggests that other portions or features of the sentence act as force-indicators. Thus a divide is generated between expressing meaning and indicating force, which was not included in Austin’s proposal and is, rather, reminiscent of other versions of the force/ propositional content distinction, for example the phrastic/ neustic distinction of Hare (1952). One further consequence is that it seems that no linguistic element having force-indicating function can express meaning at the same time. So we either admit of linguistic elements and even parts of sentences that have no meaning, at least in the sense of truth-conditional meaning (since they do not contribute to the proposition expressed), or we have to drop any specifically force-indicating function and, perhaps, give a truth-conditional analysis of sentence mood following the paratactic model proposed by Davidson (1979). The problems raised by the mutual exclusiveness of expression of meaning and indication of force manifest themselves very clearly with respect to two separate issues, both internal to speech act theory: explicit performatives and indirect speech acts. As to indirect speech acts, the problem is whether or not they constitute a genuine category. For many of our everyday speech acts, intuitive force assignment and force assignment grounded in sentence mood diverge. Searle (1975b) has therefore proposed considering intuitive force assignment as “indirect”, that is, inferentially derived. His proposal has been accepted almost universally by speech act scholars. But the need for the direct/ indirect distinction stems from the idea of sentence mood as the only, or dominant, force-indicating feature of an utterance, as argued for in Searle (1975a, 20–27). However, there are several other linguistic elements and features which intuitively may be said to play a role in orienting the hearer towards a certain illocutionary force assignment: while sentence mood can be construed as not carrying any truth-conditional meaning, these appear to have some content. It would be enough to rehabilitate these other elements and features as admissible force indicators, and the distinction between direct and indirect speech acts would disappear, at least in its current form. A more liberal notion of force indicator might include also the very content of the utterance: in fact, an utterance may count

164

as an apology, or as a protest, just because of what it says, and although this is uncommon in contemporary English, there are languages in which quite direct commands or prescriptions may be made with utterances approximately equivalent to “You have to do x”. But such a notion of force indicator is not viable if indication of force and expression of truthconditional meaning are mutually exclusive.3 The problem of explicit performatives lies in their double nature as sentences made out of meaningful words and as devices for “making explicit” the force of the utterance. If we assume strict mutual exclusiveness, explicit performatives either act as as force-indicating expressions within a larger sentential unit (“I order” in “I order you to open the door”), and in this case they cannot have any truth-conditional meaning of their own (in fact “I order” does not contribute to the propositional content of the command), or they have truth-conditional meaning and contribute to expressing a proposition, but this in turn will have a force applied to it, which will most likely be that of assertion (since explicit performative utterances have the main verb in the indicative mood). Some uneasiness with this basic alternative manifests itself, beyond the several refinements of the theory and in particular the introduction of inferential routes to force assignment, in the long-lasting debate on the pragmatic analysis of explicit performative utterances (Bach and Harnish 1979, 203–233, Searle 1989, Bach and Harnish 1992, Harnish 2004). Insofar as the proposed solutions envisage explicit performatives as expressing, or more generally communicating, assumptions about what the speaker is doing, this debate also attests to the dismissal of any strong sense of action from the issue of performative utterances, which, as everyone remembers, is the very issue from which speech act theory has originated. 2.3 The locus of the truth/falsity assessment At first sight it would seem that there is no significant difference between speaking of “sense and reference”, as Austin did, and speaking of “proposition”. After all, provided the focus is on complete utterances, does not knowing the sense in which the words were used and what they apply to amount to grasping the proposition expressed by the utterance? The change 3. An argument against mutual exclusiveness of expression of meaning and indication of force is proposed by Green (2000).

165

of terminology from “sense and reference” to “proposition” or “propositional content” may appear no more than a stylistic choice.4 But it makes a difference. When we speak of the fact that words were used in a certain sense, this may mean that they are to be taken as tokens of a type which plays a certain role in the system of language, or that they were used in accordance with a certain rule; if in addition we specify, for those words which have a referential function, what their referent was (or we non-verbally identify that referent), we have certainly specified something which is related to the truth-conditions of the utterance, but we have not yet formulated those truth-conditions exhaustively, nor perhaps them alone. Nor does this presuppose that the utterance has definite truth-conditions, in which its meaning consists. Speaking of a proposition, instead, has just such a presupposition. The truth-conditions of the utterance are there, ready and available in the proposition it expresses. So if the content of a speech act is nothing more and nothing less than a proposition, it by itself determines a truth-value. According to this perspective, both “You open the door” and “Open the door!”may be said to portray a circumstance in which the addressee opens the door. As Searle has claimed in his doctrine of the “direction of fit” of illocutionary acts (1975a, 3–4), for both these speech acts there is a match between proposition and world, when it is the case that the addressee opens the door. What is affected by illocutionary force is the logical (and when pertinent) temporal priority of either element. In the case of the imperative sentence, what is relevant is not how the world is at the time of the utterance, but how it becomes at a subsequent time as a response to the words. If in the world, in an instant belonging to the relevant stretch of time, it is the case that the addressee opens the door as a response to the words he or she has heard, this makes the imperative satisfied thanks to the same kind of mechanism which, in assertion, allows for truth or falsity. It is a mechanism written in the propositional content, for those two kinds of act as for any other. And since propositional content is expressed also by non-assertive speech acts, it appears to play a major part in communication, so that the truth-conditional, cognitive, information-carrying side of language — after the efforts made by the later Wittgenstein and by Austin to challenge its central place — is again foregrounded. Once a propositional content is specified, a truly minimal force indica4. I assume here that in the context of speech act theory, “propositional content” means semantic content taking the form of a proposition. That is, a speech act has a “propositional content” insofar as the utterance by which it is performed expresses a proposition.

166

tor (indicative mood) is sufficient to yield the assertion. No role is left to play to felicity conditions, to the corresponding possible infelicities, or to illocutionary effect in Austin’s sense. Felicity conditions, as matters of pragmatic appropriateness, are viewed as inessential to the core of assertion, which is (like in Frege) the recognition of a proposition as true. So the assertion cannot be a real action — rather, it is a cognitive gesture (or its linguistic manifestation). Speaking of assertive speech acts or calling assertion a speech act become simply ways of speaking.5 One of Austin’s aims in insisting that assertions are illocutionary acts on a par with all others, and in that sense real actions, was to interpose the illocutionary dimension between meaning and truth/falsity assessment, thus contextualizing truth/falsity assessments to the “speech act in the speech situation” (1962, 148). Whether this project is worth pursuing or completely misguided, the introduction of propositions undermines it from the start. Recent semantic contextualism (Travis (2000), Carston (2002), Recanati (2004)) has picked up and developed the suggestion that truth/ falsity assessments are context dependent. But insofar as contextualism does not dispense with propositions, this context-dependency has been explained as basically a context-dependency of the proposition expressed by the assertive utterance. Once a proposition is contextually arrived at, it possesses one and the same truth-value for ever. So even when the context-dependency of truth/falsity assessments is recognized, this does not give an opportunity to reconsider assertion as action, nor to display how radical human situatedness is, but turns out to reconfirm the reassuring context-free character of truth-values and the primacy of propositions as the locus of truth-value assignment. 3. Dispensing with propositions The weakened speech act view resulting from the introduction of propositions misses significant aspects of actions done in speech, such as their power to create and modify states of affairs consisting in the attribution 5. A speech act theorist who has opposed this mainstream trend, defending the conventionality of illocutionary acts, is Eike von Savigny (1988). Also Brandom (2000) comes close to a view of assertion as action, emphasizing entitlement to assert as well as production of commitments. However, in his view — which does not itself belong to the tradition of speech act theory — entitlement does not act as a felicity condition and commitment is not produced by achieving uptake as are illocutionary effects.

167

of deontic-modal predicates to the participants in the communication process. Moreover, it cannot conceive of, nor therefore explore, the philosophical consequences of considering assertions as full-fledged actions. In order to capture these aspects and save what I have called the strong speech act view, propositions should be kept out of the picture. It might be objected that the strong speech act view may not be worth saving. And in fact, I agree that dropping speech act theory altogether would be an option. But it is not the only option, since it is not completely unreasonable to be sceptical about propositions. There are various sources of doubt about these, which are independent of their role in speech act theory. To begin with, propositions are held to be different kinds of entities depending on the theory of language in whose context they appear. They range from ordered pairs comprising objects and properties (Russellian propositions) to senses or modes of presentation associated with declarative sentences (Fregean propositions), from sets of possible worlds to sentences in the language of thought, or just referents of that-clauses. This broad heterogeneity is itself suspicious and may hide a more basic ambiguity between propositions as truth-conditions and truth-bearers respectively. Truth-conditions are not the same thing as truth-bearers: a truth-bearer is that which is true or false, while truth-conditions are the conditions at which truth-bearers possess the property of being true. Truth-conditions may be seen as the functions or rules that, conveniently applied, tell us what the truth-value of a truth-bearer is, while truth-bearers are inevitably viewed as something substantive (whether in a concrete or in an abstract mode). Yet truth-conditions are often conflated with truth-bearers in the “proposition expressed” by an utterance, as if the truth-conditions constitutive of what the utterance says were at the same time that which, in the utterance, is true or false. Besides, the relation of “expressing” usually (and commonsensically) said to connect sentences, or utterances thereof, to propositions, is itself somewhat obscure. It ranges from codified symbolic representation to some kind of similarity or some indexical, inference-triggering function (sometimes combining more than one of these natures) and hints at the access we would have, thanks to the use of language, to something belonging to an inner or an abstract realm. But after all, the proposition itself never comes out in the public sphere in the way suggested by the word “express”: what we get are still utterances, that is contextually produced tokens of linguistic types, plus (possibly) further utterances meant to clarify or paraphrase them. In consideration

168

of all this, it seems to me that the speech act theorist is under no general obligation to accept propositions at the expense of the strong speech act view. So, it is not unreasonable to explore what speech act theory can do to dispense with them.6 In order to dispense with propositions, speech act theory should first try to make sense of the various components or aspects of the “act of saying something” without their unwanted aid. It should then propose some nonproposition based account of truth/falsity assessments. I take the former step in this paper, giving only some hints in the direction of the latter. 3.1 The locutionary act again In §1.1 we saw that in his characterization of the locutionary act, Austin avoided reference to propositions. He limited himself to the remark that those speech act reports which focus on the act of saying in the full sense of saying require the reporter to specify, beyond the words uttered by the speaker, the “sense” or the “reference” of these, or possibly both. But are there reasons for continuing to consider these two elements of the locutionary act as separate, rather than putting them together to yield the proposition? To begin with, a reconsideration of what Austin was doing with his characterization of the locutionary act can show that his account makes enough sense not to demand that kind of completion. Moreover, the unity of the sentence (as made by a subject and a predicate) can perhaps be accounted for without resorting to the notion of a proposition. From a consideration of all the characterizations given by Austin of the senses in which saying is doing or gives rise to a doing, and of the way in which he used speech reports to this aim, the conclusion may be drawn that he was thinking of doings or acts not in terms of gestures, 6. The broad variety of the conceptions of proposition suggests another objection to the line taken in this paper, namely, that one or other of those conceptions might prove compatible with the declared aims of the strong speech act view. The considerations made in § 2 above directly apply only to the Fregean conception and to its mentally implemented offsprings. But under all conceptions, propositions can be used to provide speech and attitude reports with contents, with approximately the same consequences pointed out in 2.1 and 2.2. Likewise, under all conceptions, propositions share the property of determining the truth-value of utterances, with the implications pointed out in 2.3. One might wonder whether propositions keep having the same effects on speech acts when they are collocated in the framework of use or inferential-role theories of meaning. But it is easily seen that alternative definitions of meaning do not make a difference with respect to the problems at issue here, to the extent to which the meaning of the utterance of a sentence remains the proposition expressed.

169

whether physical (bodily movements) or psychological (formation of mental states), but in terms of what an agent may be ascribed responsibility for. There is a perspectival element in his characterizations: it is no mere chance that they are regularly exemplified by citing act reports (from the speech reports exemplifying the phatic and rhetic acts, to the attributions of illocutionary and perlocutionary ones). Austin’s philosophy of action never reached a final formulation, but focus on action attribution and responsibility is clear enough in his account of the perlocutionary act as well as in some of the essays published in his Philosophical Papers (1961).7 It seems that for him, an action is singled out by singling out a state of affairs or event in the world, whose occurrence the agent can be attributed responsibility for. Also in his characterization of the senses in which saying is doing, what is attributed to the speaker is responsibility for some state of affairs or event in the world, which constitutes the effect or result of the act. In the light of this, let us reconsider the features of Austin’s locutionary act already mentioned in § 1 and the problems they appeared to raise. (1) As to the phatic act, uttering sounds that are words is no different from uttering words (at least physically), and uttering words as belonging to a language is no different gesture, either physical or psychological or both, than uttering the words. But it makes a difference whether someone is attributed speech, rather than vocalizations, or utterance of words as belonging to a language, rather than as mere sound patterns. In the attribution perspective, the phatic act enjoys a degree of reality it cannot enjoy when considered from the point of view of the speaker’s physical and psychological activities. (2) The rhetic act is perhaps no genuine component of the locutionary act, but does not coincide with it either. A specification of a locutionary act should comprise both the report of the words uttered (the phatic act) and that of their (intended) sense or reference (the rhetic act) (Austin 1962, 101). We hardly ever produce such reports in everyday life and this might be why Austin did not insist so much on the locutionary act as a whole, as on its subordinate acts, for which we have two standard report forms, oractio recta (for the phatic act) and oratio obliqua (for the rhetic act), one attributing responsibility for the choice of words, 7. See “A plea for excuses” (1956), “Pretending” (1957), and “Three ways of spilling ink” (1958). The notion of responsibility that is used here to interpret Austin‘s view of action does not by itself amount to moral responsibility: cf. Yeager (2006, 83 ff).

170

intonation and the like, the other for the intended sense or reference of the words used. (3) When Searle objected to Austin’s locutionary/ illocutionary distinction that there is no locutionary act which is not also at the same time an illocutionary act (apart from non-serious utterances and failures of performance), he must have been thinking of acts as gestures, or stretches of active behaviour. But in the attribution perspective, that is no objection, since partial reports can easily be shown to be possible. It is obviously possible, and sometimes relevant, to report force without content (“He gave her an order”) or reference without either sense and force (“He was referring to the back door”). For at least certain illocutionary acts, illocutionary reports part considerably from reports of the words used, their sense and their reference (“He apologized for being late” might be a good illocutionary report of an utterance of “Sorry for being so late” but also of “There was such a traffic jam”). As to the fact that rhetic, oratio obliqua reports obligatorily account for sentence mood, there are at least differences in degree of specificity between such reports and attributions of illocutionary acts: “He told her to go ahead” may be used to report any illocutionary act having the perlocutionary goal of getting her to go ahead, but does not indicate whether it was a request or a piece of advice, an order or a permission. Even assuming that, on occasion, rhetic reports may happen to provide all elements needed for illocutionary act attribution, the recognition of sentence mood which belongs to them does not in itself amount to an acknowledgement (or “uptake”) of the illocutionary act. So, no reason internal to Austin’s distinctions necessitates the introduction of the act of expressing a proposition: nothing obliges us to consider the attribution of meaning as (intended) sense or reference as concerning the expression of a proposition. But insistence on partial reports and on refraining from introducing the proposition at the rhetic stage may seem to disregard the problem of the unity of the sentence. What, if not the proposition expressed, is to keep subject and predicate together (or if you prefer, object referred to and properties predicated of it)? One hypothesis would be to consider the bond between subject and predicate as having a pragmatic nature. As has been argued by Gibson (2004), the assignment of the functions of subject and predicate respectively to the two main components of the sentence may itself depend on what the speaker is doing with the sentence (for example, the kind of question

171

or problem he or she is replying to). In this perspective, the unity of the sentence uttered as a rheme, that is, insofar as that unity goes beyond the phonetic unity of the pheme and the syntactic unity of the output of the phatic act, appears as a top-down effect of the total speech act, inclusive of the illocutionary dimension. If the assignment of the functions of subject and predicate is taken to be action-related in this way, clearly belonging to the illocutionary level of the speech act rather than the locutionary, it also seems fair not to include the unity that comes into being thanks to such functions among the outputs of any of the partial acts belonging to the locutionary dimension. 3.2 Truth without propositions But is it possible to give an account of truth/falsity assessments without resorting to propositions? This problem has several facets and implications. Here I will sketchily address two aspects of it. (1) First, I would like to explore whether taking statements as acts to be the subject matter of truth/ falsity assessments is indeed as nonsensical or absurd as Strawson (1973) and Searle (1968) have claimed it to be. Certainly, acts cannot be called “truth-bearers”, nor can they be said to have truth-values. But this is because these ways of speaking lose their function once we focus on what Austin tentatively called “assessment of the accomplished utterance” in the dimension of correspondence with facts (1962, 140). The assessment of an action as one which, in the light of the facts, the speaker was right in performing cannot consist, even when the action is a statement, in the recognition of the truth-value possessed by a truth-bearing entity. It is simply another sort of event. As Strawson noted (1973, 65–67), in the case of statements, the Austinian assessment of the accomplished utterance focuses on some kind of relationship between the relevant portion of the world and how the utterance says it is, while in the case of non-assertive speech acts, the sense and the reference of the uttered words are involved in the assessment in a less straightforward way, and the criteria of evaluation depart from a mere matching of the type of situation described by the words with the type to which the relevant situation in the world belongs. According to Strawson, this asymmetry shows that Austin is wrong. But after all, we may say in defence of Austin, the asymmetry

172

depends on what kind of a speech act a statement is, and it is reasonable to expect from illocutionary acts of different types that they have different ways of being the right thing to do in the light of the facts. It can therefore still be claimed that evaluating the justifiedness of a warning, the righteousness of a command, or the goodness of a piece of advice, all involve the consideration of the accomplished speech act against the relevant circumstances, and are cases of assessment belonging to the same level to which the assessment of a statement as true or false belongs. The analogies that come into focus here are less common-sense ones than the analogy which is revealed once the notion of propositional content is applied, that is, the alleged identity of propositional content across illocutionary acts with different directions of fit (“You open the door”; “Open the door!”). But this does not mean that the former kind of analogy should be disregarded and replaced by the latter. (2) We should not forget that the most obvious alternative to propositions in an account of truth/falsity assessments are just sentences. The choice of sentences as truth-bearers is quite un-Austinian, but far less incompatible with the strong speech act view than the choice of propositions.8 While propositions (insofar as their job is to provide the truth-conditional content of an utterance) are basically context-free entities, sentences naturally occur in contexts. We might take sentences in context as that which is said to be true or false, provided the definition we give of context is such that the utterance of a sentence in a context may at the same time be described, from another point of view, as the performance of a speech act (notably, of an illocutionary act in the sense of the strong speech act view). The notion of context that is pertinent here is externalist and objective, that is, governed by the goals of the speech event (what speaker and audience are up to, which is in turn at least in part governed by external circumstances) but also by the objective situation, which is mind-transcendent with respect to speaker and audience, since these might both be wrong about it. A notion of context fulfilling these requirements has been argued for and developed into a semantics of assertibility in context by Gauker (1998, 2003)9. It is perhaps no mere coincidence that the 8. This does not mean that any sentence-based conception of truth, for example those of Quine and Davidson, can be made compatible with the strong speech act view. 9. Gauker takes assertibility in context, rather than truth, to be the value which is preserved in valid arguments.

173

analysis of sentential unity proposed by Gibson (2004), which we have referred to in § 3.1 above, is also basically externalist. Gauker’s objective assertibility in context is comparable to Austin’s assessment of the accomplished utterance because its conditions trace both what the conversation is up to and how circumstances really are. It should be noted that propositions appear as contents not only in truth/ falsity assessments, but also in other cases in which the act/ object distinction may be invoked and a reading of the object side in terms of (propositional) content is suggested, such as speech reports and attitude attributions. Obviously, in order to dispense with propositions completely, these cases should also receive a non-propositional treatment. Here too, an approach based on sentences in context seems to me promising. However, I cannot tackle these issues here.10 4. Concluding remarks We have seen that the notion of proposition was introduced into speech act theory because of a dissatisfaction with Austin’s characterization of the locutionary act and with the aim of amending his view of the truth/ falsity assessment. We illustrated some of the main effects of this move on the development of speech act theory: a weakening of the notion of action involved, leading back to a more traditional conception of communicative activity; the impasse and loss of the Austinian conception of the explicit performative; the pointlessness of speaking of assertion as a speech act. In consideration of the heuristic value of the strong speech act view and of the independently motivated reasonableness of questioning propositions, we turned to a tentative exploration of how to dispense with propositions in speech act theory. I attempted to show that Austin’s characterization of the locutionary act does not need to be completed or amended by the introduction of propositions. I then tried to outline some positive proposals as regards possible accounts of assertion and truth not involving propositions. Radical departure from the mainstream account of truth/falsity assessments (in terms of truth-values that propositions possess in virtue of the 10. I have explored a possible account of belief reports in terms of sentences in context in Sbisà (2003b).

174

truth-conditions they embody) may have high costs, since it seems to require a picture of meaning not relying, in the first place at least, on truth-conditions. We might therefore want to look for some compromise strategy which would enable us to make sense of talk of propositions at least in certain connections, albeit endorsing a strong speech act view. Unfortunately, this is not easy. Definitions of propositions as abstract objects will not do, because they eventually legitimate the indiscriminate use of the received conception of propositions as contents and truth-bearers in all the contexts in which we are attempting to dispense with them. Following a suggestion of Austin’s (1962, 20), we might consider propositions as “logical constructions” out of the performance of statements and other speech acts. But it is doubtful that this would do: the whole point of the idea of logical construction is to make us aware of the constructed nature of a certain alleged entity and enable us to analyse it into the elements from which it derives. Other tentative compromises might envisage propositions as final products of language processing and thus as somewhat exceptional achievements, or as entities produced by speech acts at the illocutionary level as contents of those modal-deontic states of affairs of which illocutionary effects (in a strong speech act view) can be said to consist. The former idea arises from observing the complications of utterance processing in contextualist approaches (see Carston 2002 in particular): if this is how we reach propositions, how often do we go the whole way? But at a second glance, this misgiving is hardly compatible with the basic features of that perspective. The latter idea exploits a modality/content distinction whose affinity to the act/content distinction is in need of further analysis11 and whose ontological implications would require careful scrutiny. So I believe that the best option consistent with the strong speech act view is to attempt to dispense with propositions altogether. This too may involve compromises, or perhaps some division of labour (as already suggested above), with sentence-based accounts of truth and attitudes. But these compromises should not amount to accepting an account of meaning as equivalent to truth-conditions. Austin’s characterization of the locutionary act allows for a more liberal conception of sense as linguistic meaning (as opposed to truth-conditions), in accordance with the perspective that appears to inspire all of his discussions on the meaning of words. This involves also (as has been suggested in § 2.2 above) a liberal notion of force-indicator, according to which, on the one hand, all components of 11. A thoughtful contribution to this issue is that of Oswald Ducrot (1993).

175

an utterance are considered as contributing to various aspects of its sense or of its reference, while on the other hand, the whole utterance (whether sentential or sub-sentential) acquires “force” thanks to its overall physiognomy, which depends primarily (but not exclusively) on those features to which a specific force-indicating character is attached. In taking this direction, it would be of help to have in mind some rough idea of what sense to make of talk about meaning, once we stop taking it to make reference to so-called truth-conditional content. So I end by suggesting one very informal way to see meaning as an effect of speech. Speech might make a difference in much the same way in which your room changes when you hang a new poster on the wall. Faced with a given circumstance, we find our way through it thanks to the way we perceive it and pre-verbally (or indeed also verbally) categorize it. When words are spoken in that circumstance, the performed speech act adds its own occurrence to the circumstance and the way we perceive and categorize it is likely to change accordingly.12 The circumstance might take a new place in the network of similarities and differences in our experience, possibly enabling new inferences. A Saussurean structuralist might say that language, made out of differences, makes a difference. A pragmatist would measure the effect in terms of “interpretants”, audience reactions and disposition formation (but those who do not share the metaphysics of pragmatism, myself among them, may prefer to avoid such terminology). It is not clear to me whether this picture can be developed into an account, but perhaps it can provide a background framework both for non-proposition based accounts of linguistic meaning and for actioninspired views of speech. Acknowledgements I thank the audience at the Padova conference, especially Eva Picardi and Enrico Martino, for their questions and comments. I thank Magdalen College, Oxford, for the Visiting Fellowship I was granted in Hilary Term 2006, during which I prepared the revised version of this paper. I am also grateful to Dorothy Edgington, Elizabeth Fricker, Hanjo Glock and Elisabetta Sacchi for their comments and criticism. 12. This reflection has been inspired by Christopher Gauker, personal communication, (1999).

176

REFERENCES Alston, W. P. 1964. Philosophy of Language. Englewood Cliffs: Prentice-Hall. — 2000. Illocutionary Acts and Sentence Meaning. Ithaca: Cornell University Press. Austin, J. L. 1950. “Truth”. Proceedings of the Aristotelian Society Suppl. 24. Repr. in Austin 1961, 117–133. — 1961. Philosophical Papers. Oxford: Oxford University Press, III ed. 1979. — 1962. How to Do Things With Words. Oxford: Oxford University Press, II ed. 1975. Bach, K. and Harnish, R. M. 1979. Linguistic Communication and Speech Acts. Cambridge, Mass.: MIT Press. — 1992. “How performatives really work: a reply to Searle”. Linguistics and Philosophy 15, 93–110. Berlin, I. et al. 1973. Essays on J. L. Austin. Oxford: Oxford University Press. Brandom, R. 2000. Articulating Reasons: An Introduction to Inferentialism, Cambridge, Mass.: Harvard University Press. Carston, R. 2002. Thoughts and Utterances. Oxford: Blackwell. Davidson, D. 1979. “Moods and Performances”. In: A. Margalit, ed. Meaning and Use. Dordrecht: Reidel. Repr. in Davidson, D. 1984. Truth and Interpretation. Oxford: Oxford University Press, 109–121. Ducrot, O. 1993. “A quoi sert le concept de modalité?” In: N. Dittmar and A. Reich, ed. Modality in Language Acquisition. Berlin: Gruyter, 111–130. Frege, G. 1879. Begriffsschrift. Halle: Nebert (English transl. in Frege, G. 1972. Conceptual Notation and Related Articles. ed. by T. W. Byrnum, Oxford: Oxford University Press). Gauker, C. 1998. “What is a context of utterance?” Philosophical Studies, 91, 149–172. — 2003. Words Without Meaning. Cambridge, Mass.: MIT. Gibson, M. I. 2004. From Naming to Saying: The Unity of the Proposition. Oxford: Blackwell. Green, M. S. 2000. “Illocutionary force and semantic content”. Linguistic and Philosophy 23, 435–473. Hare, R. M. 1952. The Language of Morals. Oxford: Oxford Universiy Press. Harnish, R. M. 2004. “Performatives as Constatives vs. Declarations: Some Recent Issues”. In: F. Brisard et al., eds. Seduction, Community, Speech: A Festschrift for Herman Parret. Amsterdam: John Benjamins, 43–59. Pitcher, G. ed. 1964. Truth. Englewood Cliffs, N.J.: Prentice-Hall. Recanati, F. 2004. Literal Meaning. Oxford: Oxford University Press.

177

Savigny, von E. 1988. The Social Foundations of Meaning. Berlin: Springer. Sbisà, M. 1989. Linguaggio, ragione, interazione: Per una teoria pragmatica degli atti linguistici. Bologna: Il Mulino. — 2001. “Illocutionary force and degrees of strength in language use”. Journal of Pragmatics 33, 1791–1814. — 2002. “Cognition and narrativity in speech act sequences”. In: A. Fetzer and C. Meierkord, eds. Rethinking Sequentiality, Amsterdam: John Benjamins, 71–98. — 2003a. “Subject and Gender in H. D.’s Trilogy”. In: M. Camboni, ed. H. D’s Poetry: The Meanings That Words Hide. New York: AMS Press, 91–113. — 2003b. “Belief reports: what role for contexts?” Facta Philosophica 5, 257– 76. — 2006. “Communicating citizenship in verbal interaction: principles of a speech act oriented discourse analysis”. In: H. Hausendorf and A. Bora, eds. Analysing citizenship Talk. Amsterdam: John Benjamins. Searle, J. R. 1968. “Austin on locutionary and illocutionary acts”. The Philosophical Review, 77, 405–424. Repr. in I. Berlin et al. 1973. Essays on J.L. Austin. 141–159. — 1969. Speech Acts. Cambridge: Cambridge University Press. — 1975a. “A taxonomy of illocutionary acts”. In: K. Gunderson, ed. Language, Mind and Knowledge, Minneapolis: University of Minnesota Press. Repr. in Searle 1979, 1–29. — 1975b. “Indirect speech acts”. In: P. Cole e J. L. Morgan, ed. Syntax and Semantics III: Speech Acts. New York: Academic Press. Repr. in Searle 1979, 30–57. — 1979. Expression and Meaning. Cambridge: Cambridge University Press. — 1989. “How performatives work”. Linguistics and Philosophy 12, 535–558. Sperber, D. and Wilson, D. 1986. Relevance. Oxford: Blackwell (II ed. 1995). Stenius, E. 1967. “Mood and language game”. Synthese 17, 254–280. Strawson, P. F. 1973. “Austin and ‘locutionary meaning’”. In: I. Berlin et al. Essays on J. L. Austin. 46–68. Travis, C. 2000. Unshadowed Thought. Cambridge, Mass.: Harvard University Press. Yeager, D. 2006. J. L. Austin and the Law. Lewisburg: Bucknell University Press.

178

Grazer Philosophische Studien 72 (2006), 179–199.

HOW TO GET A NONINTENSIONALIST, PROPOSITIONAL, MODERATELY REALIST TRUTHCONDITIONAL ACCOUNT OF INTERNAL METAFICTIONAL SENTENCES Alberto VOLTOLINI University of Modena and Reggio Emilia, Italy Summary In what follows, I will first try to show that both anti-realist and realist intensionalist truthconditional accounts of internal metafictional sentences (i.e., sentences of the form “in the story S, p”) are unsatisfactory. Moreover, I will claim that this does not mean that propositional truthconditional accounts of those sentences are to be dispensed with; simply, one has to provide a non-intensionalist propositional truthconditional account of those sentences. Finally, I will show that this account is fully compatible with a realist interpretation of those sentences’ truthconditions according to which at least some of those sentences commit one to fictional entities.

1. The failure of the intensionalist accounts It is rather a commonplace that sentences directly occurring in a text of fiction, let me call them fictional sentences, like: (1) The miserable girl replied belonging to Alessandro Manzoni’s text of The Betrothed, are true not only within the fiction itself, namely in the imaginary world postulated by make-believing that so and so is the case, but also outside the fiction, i.e., in reality. This is also the case with the parafictional sentences, namely sentences occurring in one such text merely indirectly, by being either appropriate reformulations of fictional sentences, like: (1c) Gertrude, the Monza’s nun, responded to Egidio’s greetings

or make-believe entailments of these sentences,1 like: (2) Gertrude had a forbidden love affair which clearly is make-believedly entailed by the part of Manzoni’s text containing (1) and recounting the vicissitudes of Gertrude, the unhappy nun living in a Monza’s convent who, in order to feel relief from her dissatisfaction, finally responded to the greetings of an ambiguous guy named Egidio and had an intimate relationship with him. In point of fact, if one does not utter the above sentences while playing a make-believe game with Manzoni’s text, but, say, during an examination on Italian literature, this would enable one to pass the exam, for those sentences are not only fictionally, but also really, true. Yet if in the same occasion one had uttered: (3) Gertrude had sexual intercourse with Renzo Tramaglino this would have made one fail the exam, for that sentence is not only fictionally, but also really false; such a tremendous state of affairs definitely does not obtain in Manzoni’s masterpiece. In order to account for the real truth of (1) (or (1c) and (2) (or for the real falsity of (3)), it is often said that, over and above the use that makes those sentences fictionally true (false), the use made within a particular make-believe game, or conniving use, there is another use of those sentences, the use made outside such a game, or the nonconniving use.2 In this use, the afore-mentioned (para)fictional sentences are, first, really true or false, and second, they are respectively elliptical for: (1*) In The Betrothed, the miserable girl responds (to Egidio’s greetings) (2*) In The Betrothed, Gertrude has a forbidden love affair (3*) In The Betrothed, Gertrude has sexual intercourse with Renzo Tramaglino 1. Under which conception of make-believe entailment is here not in question. For one such (very reasonable) conception, cf. Evans (1982, 354). Bonomi (2006) analogously distinguishes between textual and paratextual sentences. 2. I here borrow Evans’ (1982, 365–6) distinction between conniving and nonconniving uses of singular terms and extend it to sentences. Currie (1988), (1990) draws an analogous distinction between fictional and metafictional uses of names and straighforwardly extends it to sentences. The first use is also labelled as pretending use in Schiffer (1996), (2003).

180

which are intuitively true, or false, in reality itself, not within a certain fiction (as (para)fictional sentences on the contrary are when used connivingly).3 More generally, a (para)fictional sentence “p” when used nonconnivingly is elliptical for (i.e., equivalent to) a sentence of the form “in (the story) S, p”. Let me call a sentence of the latter form an internal metafictional sentence (from now on, IMS). It is a metafictional sentence,4 for it is about a certain fictional story (Manzoni’s The Betrothed, in our example), but it is also an internal metafictional sentence, for (as I will stress afterwards) its point is precisely to contribute to tell the content of a certain story, rather than speaking of fictional matters independently of what a story conveys. This is what happens in external metafictional sentences,5 as when one utters: (4) Gertrude is one of the most famous characters of Italian literature which, since it definitely does not contribute to tell the content of Manzoni’s The Betrothed — Manzoni’s masterpiece is definitely not a piece of a metafictional story, like Pirandello’s Six Characters in Search of an Author — is a prototypical example of an external metafictional sentence. At least from Lewis (1978) onwards, it is typical to account for the truth vs. the falsity of an IMS by providing an intensionalist account of its truthconditions. By oversimplifying matters a bit, a sentence of the form “in the story S, p” is true iff there is a world, the world S determines,6 in which “p” is true. Regardless of Lewis’ own intentions, this truthconditional account is often taken to imply that the truth of an IMS does not commit one to fictional entities.7 For defenders of this interpretation take 3. An alternative to this distinction between uses of (para)fictional sentences is to distinguish between what one fictionally says and what one really asserts in uttering one and the same such sentence. Cf. Evans (1982, 363–4). Yet this latter distinction does not change things so much: what one fictionally says is precisely the content one gives to a (para)fictional sentence in using it connivingly, while what one asserts in fictionally saying something by uttering one such sentence is what an internal metafictional sentence says. 4. As Recanati (2000) simply labels it. 5. Metatextual sentences for Bonomi (2006). 6. In point of fact, this is rather a set of worlds rather than a single world. Lewis (1978) is precisely focussed on finding out the right criterion that allows one to single out one such set. Yet this is irrelevant for my purposes here. 7. See e.g. Currie (1990). As is well known, Lewis thought that such sentences actually commit one to fictional entities qua possible entities. For Lewis, therefore, the de dicto reading of one such sentence entails the de re reading, providing that the underlying existential quantifier

181

one such sentence to be read in a de dicto way. According to such a reading, an IMS is true iff the designation had in the world determined by the story S by the singular term contained in the embedded sentence “p” possesses in that world the property there designated by the embedded predicative term. In such a reading, in order for an IMS to be true it must be such that its embedded singular term has a designation in the world of the story S. Yet, since that truth does not force that term to have a designation in the actual world, it does not commit one to fictional entities. Take (1*). In its de dicto reading, (1*) is true iff the individual “the miserable girl” denotes in the world determined by The Betrothed responds (to Egidio’s greetings) in that world. As in that world the above singular term indeed denotes an individual who responded (to Egidio’s greetings), namely Gertrude, the Monza’s nun,8 (1*) is true. Yet since on that reading (1*) does not force that term to have a denotation in the actual world — in point of fact, that term has no denotation in the actual world — (1*) so read does not commit one to a fictional entity. The main drawback for this position is that from the semantical point of view it implausibly entails adopting a classical descriptivist view of proper names, by taking names as synonymous with contingently empty definite descriptions (hence, as flaccid nondesignators, as one might say). For suppose one accepts a Kripkean view of proper names. According to such a view, names that fail to actually have a designation do not have a designation in any possible world either: they are rigid nondesignators.9 By itself, this move entails adopting a non-classical descriptivism, according to which names are synonymous with rigid definite descriptions.10 Yet if by pursuing Kripke’s own intentions one moreover assumes that names are directly referential devices, in the sense that they exhaust their truthconditional contribution in their actual referent, it turns out that names that fail to actually have a designation are not only rigid nondesignators (as even a definite description that cannot have a denotation, like “the round square”, is), they are meaningless expressions tout court.11 Either way, from is taken as ranging over both actual and merely possible individuals existing in possible worlds, which are taken to be genuine and primitive individuals pretty much like the actual world. Cf. e.g. (1986). 8. Of course, in order for this description to have an (unactual) Russellian denotation, it must be taken to be elliptical for something like “the miserable girl who become an unhappy nun in a Monza’s convent under the label of ‘the Signora’”. But let me put this aside. 9. Cf. Salmon (1998, 292). 10. For this move, cf. notoriously Plantinga (1978). 11. For simplicity’s sake, I am here assuming that names are not indexical expressions.

182

this perspective there is no world in which a sentence containing a proper name having no actual designation is true. Thus, contrary to fact, there is no chance for the IMS embedding it to be true. Take for instance (2). Insofar as the name “Gertrude” occurring in it is a rigid nondesignator, (2) must be necessarily false or even meaningless; but then contrary to fact (2*), which is the relevant IMS embedding it, has no chance to be true. As a consequence, the only way to avoid this implausible result would be to take the name to be synonymous with a merely contingently empty definite description.12 Yet in want of any argument in favour of this (limited) adoption of a descriptivist position on proper names, this move sounds terribly ad hoc.13 Some Kripkean antirealists have maintained that one can both stick to an intensionalist noncommittal analysis of IMSs and to the thesis that proper names are directly referential devices.14 One can have both if one assumes that locutions of the form “in the story S” work not only as a circumstance-shifting operator for the sentence they embed, as any intensional operator does, but also as a context-shifting operator for that sentence. That is, qua such operators these locutions make it the case that the truthconditions of an IMS involve not only different circumstances of evaluation, or For if one on the contrary expouses such a view, one may well say at the same time that a name may be a directly referential device, fail to give any truthconditional contribution when empty, and still have a (truth-conditionally irrelevant) linguistic meaning such as an (unfixed) character. Cf. Recanati (1993); I have myself defended such a view in Voltolini (1995), (unpublished). 12. As e.g. Currie (1990, 158–62) has actually maintained. In (2004a), Kroon extends the descriptivist view of names he appeals to in (2004b) to empty proper names like “Hamlet”: if (per impossibile) those names had a referent, it would be fixed by a rigidified description of a certain sort. Although he says that this view amounts only to a weak form of descriptivism, according to which descriptions merely determine the reference of names but are not synonymous with them, he claims that what is really asserted by a negative existential like “Hamlet does not exist” is the descriptive content Russellianly mobilized by one such rigidified description. To be sure, however, he could not extend this move to truthconditionally account for a IMS embedding “Hamlet”. For, since the fixing-reference description he has in mind is, qua rigidified description, a rigid nondesignator, Kroon’s account may enable us to explain the intuition according to which the above negative existential is a necessary truth; yet precisely for the same reason, if the relevant IMS had a parallel descriptive content, it would implausibly turn out to be a necessary falsity. To be sure, however, Kroon explicitly refrains from so extending that move (cf. (2004a, 11)). 13. For criticisms of descriptivism for empty names, cf. e.g. Braun (1993, 454), AdamsFuller-Stecker (1997), Goodman (2005). 14. For this position, cf. Evans (1982) and especially Walton (1990), as reconstructed in Recanati (2000).

183

worlds, for the embedded sentence, but also a different context relevant for the truthconditional interpretation of that very sentence. According to such a position, an IMS of the form “in the story S, p” is true iff “p” is not only true in a circumstance of evaluation different from the actual one, as is the case with any intensional operator — notably, the world determined by S — but it is true there given the truthconditional interpretation that it receives in a context different from the context relevant for the interpretation of the whole sentence; namely, a context whose worldparameter is filled precisely by the world determined by S, a fictional, or better imaginary, context. In point of fact, the sentence “p” in question is nothing but a (para)fictional sentence in its conniving use; saying that one such sentence is connivingly used is indeed tantamount to saying that that sentence is associated with one such different context in order for it to receive fictional, or better imaginary, truthconditions — i.e., to fictionally say something — and moreover be fictionally true, i.e., true in the world of that context, if things in that world, the world determined by S, go as it fictionally says. For this reason, let me call the present truthconditional account a pretense-intensionalist account.15 This account enables one to keep a both noncommittal and direct referentialist view of the embedded proper names. For although one such name will lack a designation in the context relevant for the interpretation of the whole sentence, namely a context whose world-parameter is filled by the actual world (hence, if Kripke is right, it will lack a designation in all circumstances of evaluation), it will have a designation in the context relevant for the interpretation of the embedded sentence, namely a context whose world-parameter is filled by the world determined by S. In a nutshell, for a name to have an imaginary referent (that is, to have a referent in an imaginary context; or, which is the same, to make-believedly have a referent) is not the same as for it to actually refer to a fictional individual; a name may well have the former but not the latter referent. Take again (2*). Admittedly, the name “Gertrude” refers to nothing in the context whose world-parameter is filled by the actual world. So, in (2*), which receives its semantical interpretation in that context, “Gertrude” refers to nothing as well. Yet in the fiction inaugurated by Manzoni one makes believe that it refers to something; namely, the imaginary individual 15. In point of fact, by taking the fact that (para)fictional sentences have real truthconditions as being primary over IMSs’ having truthconditions, Walton (1990) directly holds that (para)fictional sentences are really true iff they are fictionally true (for this synthetic formulation of Walton’s position, cf. Crimmins (1998)), but this is just a question of detail.

184

Manzoni describes as an unhappy nun, etc. This amounts to saying that in a shifted context, the context whose world-parameter is filled by the world determined by that fiction, the name refers to that individual. So, take (2), where “Gertrude” also occurs. Saying that (2) is connivingly used is tantamount to saying that (2) receives imaginary truth-conditions by being interpreted in the above shifted context. Since moreover “Gertrude” there refers to that imaginary individual who, in the world of that context, had a forbidden love affair, in that world, (2) is true — alternatively said, it is fictionally true. But (2) is precisely the sentence embedded in (2*). Thus, it turns out that the operator expressed by “in The Betrothed” precisely works as a context-shifter. For it makes it the case that (2*) is true iff (2), when interpreted in the shifted context, is true.16 Context-shifting operators have been judged to be controversial figures. As is well known, Kaplan has even denied that there may be such operators.17 Yet even if one accepted that there are such operators, it remains that they do a semantically problematic job. For they make the actual meaning of one sentence depend on the meaning another sentence has in another circumstance. And this sounds very implausible. Nobody questions that the truthvalues of some sentences may depend on what happens in circumstances different from the actual ones, hence on the truthvalues some other sentences receive in those circumstances. This is precisely the idea on which the intensionalist account of IMSs (as well as of intensional contexts in general) hinges. Yet how can what we mean by a certain sentence depend on what — to put it vividly — inhabitants of another world mean by another sentence? 16. In (1997), Adams-Fuller-Stecker maintain that an IMS like (2*) is true iff, using a name like “Gertrude”, by (2) one fictionally asserts or implies a gappy proposition (i.e., a structured entity failing to have an objectual constituent) to the effect that x has a forbidden love affair. By appealing to a given name use, this antirealist account seems to me better than the analogous one suggested by Braun (2005, 604). (By limiting himself to explicitly saying that a sentence like (2*) has a ‘that’-clause whose content is the above gappy proposition, Braun does not provide a convincing explanation as to why (2*) is true but a sentence like “In The Betrothed, Hamlet has a forbidden love affair”, which admittedly mobilizes the same gappy proposition, is false. Braun says that an agent may accept (2*) while rejecting the latter sentence, yet this is irrelevant as regards these sentences’ truth-value, especially since he also says that also these sentences “express the same gappy proposition” (ib.). I nevertheless say that this is the account Braun suggests for he actually has a more complex account of IMS; see fn.35 below.) Yet by appealing to fictional assertions or entailments, Adams-Fuller-Stecker implicitly recognize that sentences like (2) bear a context-shift. But in the new context (2) does express a full proposition, not a gappy one. For, as we have just seen, “Gertrude” there is fully referential. 17. Cf. Kaplan (1989, 510).

185

In order to clearly see the problem, let me first switch to a case involving predicates rather than names. Take Lewis Carroll’s Jabberwocky’s nonsense: (5) The mome raths outgrabe. Nobody doubts that, in the context whose world-parameter is filled by the world of Jabberwocky, (5) has a meaning for the alleged predicates it contains — “( ) is a mome”, “( ) raths outgrabe” — have a meaning there.18 Given such an unactual meaning, that sentence also happens to be true in the world of Jabberwocky, i.e., it is fictionally true. In point of fact, part of the pretense we play with that poem is precisely constituted by makingbelieve that (5) is both meaningful and true. But now take: (6) In Jabberwocky, the mome raths outgrabe and ask yourselves whether this sentence is true — i.e., actually true, true in the real world. I guess that the intuitive answer would be that we have no idea, for we do not know what the embedded predicates mean; and we cannot know that, for those predicates actually have no meaning. As a result, I think we would be forced to say that (6) has no actual meaning at all, regardless of the fact that (5) is meaningful in another context whose world is not the actual world. Thus, there is no dependence of the actual meaning of a sentence on the unactual meaning of another sentence; even if (5) changed its unactual meaning, (6) would remain meaningless. A pretense-intensionalist might reply that the difficulty of giving a plausible pretense-intensionalist treatment of (6) is that, unlike (2*), (6) involves not only a context-, but also a language-shift. Yet it must be retorted that this shift is precisely what (2*) would also involve in its pretense-intensionalist treatment once the pretense-intensionalist appealed to the thesis that, qua directly referential devices, referentless names are meaningless expressions tout court; as (s)he actually does. For in such a case, the independence of the actual meaning of a sentence on the unactual 18. In point of fact, the example is more complicated than this, for, as Fred Kroon has made me note, Carroll’s suggested parsing for (5) is different, “the mome raths” working as a plural description and “outgrabe” as a verb. However, nothing would basically change if we accepted that (5) has that syntax; it would still have merely fictional truthconditions when interpreted in a shifted context in which the language parameter also shifts.

186

meaning of another sentence forces one to conclude that with names like “Gertrude” we are in the same boat as with meaningless predicates, as is the case with (6). As a result, if one sticked to the pretense-intensionalist perspective, one would have that although a sentence like (2) has imaginary truthconditions insofar as “Gertrude” has an imaginary referent, its embedding sentence (2*) would be meaningless insofar as “Gertrude” actually lacks a referent.19 As the problem here lies in making the “in the story”-operators context-shifting operators, one might think that a way out of the problem is to attempt to keep the pretense-intensionalist account while freeing those operators from such a context-shifting role. This is made possible by giving the contextual shift so to say wide scope. Suppose one considers the whole IMS as imbued with pretense, a sort of second-order pretense in which one makes believe that, alongside with real individuals, there also are fictional characters that have a surrogate existence in stories only, where moreover they are pretended to do some other things. From this perspective, take an IMS like (2*). To be sure, (2*) would literally fail to have truthconditions when interpreted in our context for, as we have seen, in our context a name like “Gertrude” is referentless hence meaningless. Yet it has imaginary truthconditions, once it is interpreted in the context of a second-order pretense like the one sketched above, where a name like “Gertrude” make-believedly refers to a fictional character (or, which is the same, it refers to an imaginary fictional character). Let me call this variant of the pretense-intensionalist approach the weakened pretense-intensionalist account of IMS.20 Yet this variant also sounds unsatisfactory to me, for we want IMS to have real truthconditions (hence real truthvalues), not only imaginary ones, let alone grounded by a second-order pretense. Put alternatively, we may well figure out conniving uses even of IMSs, for instance of (2*) themselves. For instance, I might now tell a metafictional story in which I made-believe that Gertrude is a fictional character created by Manzoni who (in my story) complains that Manzoni in the Betrothed assignes her 19. To be sure, things would be different if one appealed to an indexical view of proper names (cf. fn.11). For since in such a case names would retain their linguistic meaning in passing from being interpreted in an actual context to be interpreted in an unactual context, the context shift would not also be a language shift, as is definitely the case in the ‘Jabberwocky’-example. Yet it must be recalled that it is precisely with indexicals that intensional operators apparently fail to perform a context shift, as Kaplan has pointed out (cf. fn.17). 20. This variant is defended in Recanati (2000).

187

that sad fate. In so pretending, my storytelling would enact an occurrence of (2*) that has imaginary truth-conditions: (2*) is fictionally true iff the world determined by my story is such that in The Betrothed Gertrude has a forbidden love affair. (Incidentally, we do not even have to invent such cases, for various literary corpora actually present many of them; think of those plays that contain another play, such as Hamlet.) Yet when we normally utter (2*) we are not using it in any such conniving way. But the variant of pretense-intensionalism we are considering here is forced to give a truthconditional account of IMSs that may only concern a conniving use of them.21 It is hard to see whether there may be another way of semantically combining intensionalism with antirealism on ficta; as things stand, however, it has turned out that every real attempt to combine intensionalism with antirealism on fictional entities is doomed to fail. This may sound good news for the realist. For she may say that she may rather provide an intensional realist account of IMSs: it suffices that one gives those sentences not a de dicto, but a de re reading. In such a reading, one such sentence is true iff the designation had by the embedded singular term in the actual world possesses, in the world determined by S, the property there designated by the embedded predicate. This actual designation of that term is precisely a fictional character, taken therefore as an entity belonging to the overall inventory of what there is: of that individual, the story says that it is such and such.22 Moreover, unlike at least straightforward intensional antirealism, intensional realism has no problem in tallying with direct referentialism for proper names. For it claims that a name like “Gertrude” actually directly refers to a fictional character, in the same way as “Milan” directly refers to the real concrete entity which is the capital of Lombardy (and where incidentally Manzoni locates part of his story). Granted, it will be a matter of debate to decide which nature a fictional character precisely possesses; typically, intensional realists assume 21. In (2006), Recanati claims that there is a way to ascribe real truthconditions to internal metafictional sentences containing directly referential expressions; namely, to say that one such sentence is true iff in the relevant story there is an individual referred to by one such expression having the property designated by the embedded predicate. Yet in (2000, 224–5) he refrained from endorsing this account for its descriptivist halo, namely for its providing an internal metafictional sentence with a generic real truthconditional content rather than with the singular one it should have, provided that it had any. 22. In point of fact, for the intensional realist that actual designation may also be a real concrete individual, as when we have a IMS like “in the Doyle’s stories, Baker Street has the civic number 221B”, which is about the real famous Londoner street.

188

that it is an abstract entity of some sort. Yet this is a completely different affair.23 So, take again (2*). For the intensional realist, as I have just said, the name “Gertrude” occurring there actually refers to a fictional character. So, (2*) is true iff the actual referent of that name, i.e., the fictional character in question, has a forbidden love affair in the world determined by The Betrothed. The problem with intensional realism is that the intensional realist is implausibly forced to deny that, as far as the sentence embedded in the relevant IMS is concerned, once intensionalism is given,24 there is no way of securing an embedded singular term a reference without allowing that a context shift occurs, by thus obtaining an imaginary and not a real reference for that term. Let us look at things in detail. For the intensional realist, the world a fiction determines is relevant only as a circumstance of evaluation for the embedded sentence and not as a parametrical element of a context of its shifted truthconditional interpretation, for according to her there is no such shift: that embedded sentence, with the very same truthconditional content which it actually receives, is evaluated both with respect to the world of a fiction and with respect to the actual world.25 As a result, in order to determine whether the IMS embedding that sentence is true, one must simply look at the actual designation of the embedded singular term and then check whether in the world determined by a certain fiction it possesses a certain property. Put in another way, for the intensional realist there is no difference between the way a sentence like (2*) is actually true and the way a sentence like: (7) It is possible that Alessandro Manzoni is a rockstar

23. Among intensional realists, for Meinongians like Zalta (1983), (1988), a fictional individual is like a Platonic type, a generic individual; for artifactualists like Predelli (1997), Salmon (1998), Thomasson (1999), a fictum is an artefact which has been created by the author of the story where it first appears. Both Zalta and the artifactualists precisely share the idea that a fictum is an abstract individual; yet one may reject this conviction, by taking a fictum as a concrete Meinongian nonexistent entity, and nevertheless be ready to give an IMS a de re reading, as all abstractionists think: cf. e.g. Priest (2005). 24. As well as direct referentialism: see fn.31 below. 25. Cf. Predelli (1997), Salmon (1998), Thomasson (1999). In the first case, it probably will turn out to be true, while in the second case, it probably will turn out to be false; but this is not fundamental.

189

is such. In the latter case, in order to evaluate (7) as true, one takes the actual referent of the embedded name “Alessandro Manzoni” — i.e., the great Lombard writer — as well as a certain unactual world, and then simply checks whether in that world that very individual has been a rockstar. If this is the case, then (7) is true. In other terms, saying that (7) is true amounts to saying that its embedded sentence: (8) Alessandro Manzoni is a rockstar with its actual truthconditional meaning (determined inter alia by the fact that “Alessandro Manzoni” refers to Manzoni) is true in a certain unactual world. But, the intensional realist goes on arguing, the same goes for (2*). In order to evaluate (2*) as true, one takes the actual referent of “Gertrude” — i.e., a certain fictional character — as well as the unactual world determined by The Betrothed, and then checks whether in that world that very character has had a forbidden love affair. If this is the case, as indeed it is, (2*) is true. So, saying that (2*) is true is tantamount to saying that its embedded sentence (2) with its actual truthconditional meaning (determined inter alia by the fact that “Gertrude” refers to a certain fictional character) is true in a certain unactual world determined by The Betrothed. Yet the previous reflections show that, as far as the sentence embedded in an IMS is concerned, if one endorses intensionalism then a context shift appears to be unavoidable. Take again a sentence like (2). Pace intensional realists, if we associate with a sentence like (2) an actual context for its interpretation — at least before a make-believe game involving that sentence has been performed26 — it turns out that that sentence is meaningless, for the name “Gertrude” in it refers to nothing at all, not a fictional entity either! Indeed, in order for that sentence to originally get truthconditions, which means, in order to get the imaginary truthconditions it receives for its having been involved in a make-believe game (hence, for its having been used connivingly), one must associate with it another context, an unactual context whose world-parameter is filled by the world determined by a fiction. In this context, that sentence precisely gets imaginary truthconditions insofar as that name passes there on to (directly) refer to an imaginary individual. 26. For once that game has been performed, a fictional character may well arise out of that game, hence a (para)fictional sentence like (2) may later be used both nonconnivingly and in a committal way. This is what nonintensional realists like myself believe. Cf. Voltolini (2006).

190

Intensional realists tend not to note this contextual shift for they start by considering cases in which the truthconditions of a sentence, hence the reference of the directly referential devices involved in it, do not effectively change from being associated with an actual rather than an unactual imaginary context. Take the occurrence of the very first sentence of Chapter I of Manzoni’s The Betrothed: (9) That branch of the lake of Como, which extends towards the south, is enclosed by two unbroken chains of mountains. As is shown by the fact that this very sentence might occur with the same meaning also in a tourist’s guide on Northern Italy, this sentence does not change its truthconditions in passing from being associated with the latter, actual, context, to be associated with the unactual context whose world is the world of The Betrothed. Hence, the complex demonstrative “that branch of the lake of Como” in (9) goes on referring to the very same actual thing, namely the south-eastern branch of the lake on whose shores George Clooney presently lives. Relying on cases like this, intensional realists think that a shift in context relevant for the truthconditional interpretation of a sentence never arises in fiction; cases like (2) are precisely like cases such as (9) but for the fact that both in the actual and in the unactual context (9) concerns one and the same concrete individual — a certain real physical area, the south-eastern branch of the lake of Como — whereas both in an actual and unactual context (2) concerns one and the same fictional individual — a certain character.27 But this generalization is ungrounded. In point of fact, cases like (9) do not involve a reference shift in passing from associating the sentence with an actual to associating it with an unactual context like that of Manzoni’s telling of (9), for that telling involves what Evans calls an existentially conservative game of make believe, namely a game in which, of a certain actual individual, one makes believe that that individual is so and so; of the south-eastern branch of the lake of Como, Manzoni makes believe that the events he is going to recount happen nearby. Yet cases like (2) involve a reference shift in passing from an actual to an unactual context, that mobilized by Manzoni’s telling of (2) — a shift that I might label a ‘something from nothing’-shift — for that telling involves what Evans calls an existentially creative game of make-believe, namely a game in which 27. Cf. Predelli (1997).

191

one makes believe that there is a certain individual with a certain name (“Gertrude”) who does such and such things.28 Pace intensional realists, in these cases there is no individual, not even a fictional one, of which one makes believe something.29 Thus, the two contexts — the actual and the unactual one — are also relevant for yielding a different truthconditional interpretation of a sentence like (2). For the sentence passes from failing to have truthconditions (in the actual context) to have certain imaginary truthconditions (in the unactual context). As moreover there are mixed cases like that of the following sentence (which occurs just a few sentences after (9) in Chapter I of Manzoni’s The Betrothed): (10) Along one of the narrow lanes [linking the towns around Lecco …], Don Abbondio … curate of one of the towns alluded to above, was leisurely returning home from a walk where the name “Don Abbondio” acquires an imaginary reference, whereas the name “Lecco” retains its real reference — the real town lying on the south-eastern shores of the lake of Como — it is more proper to say that, whenever fiction is involved, a shift in the context relevant for the truthconditional interpretation of the relevant sentence occurs.30 28. Cf. Evans (1982, 358). 29. In point of fact, there are cases in which one makes-believe something even of a fictional individual. These are cases of metafictional pretenses, as the one I invented before in which I made-believe something of Gertrude the Manzonian fictional character. Yet they are definitely not the norm. 30. To be sure, moreover, if one is a realist the shift in question it is even harder to note. For, as I have said at the very beginning, an IMS is nothing but a (para)fictional sentence in its nonconniving use; so, if one wants to have a committal truthconditional conception of an IMS, one must have the same conception of the (para)fictional sentence which when nonconnivingly used is elliptical for that IMS. As a result, a realist may be ready to suppose that both in fictional contexts and when used nonconnivingly in an actual context one and the same (para)fictional sentence will be about the same fictional individual, hence will have the same truthconditions. Yet this would be just another mistake. By associating one such sentence with a fictional context — or, which is the same, by using that sentence connivingly, i.e., in a certain (existentially creative) make-believe game — one will designate an imaginary individual which, appearances notwithstanding, is not the fictional individual which, if one is a realist, one ends up with designating when using that very sentence nonconnivingly. As I said before, for a name to have an imaginary referent (or, which is the same, to make-believedly have a referent, to have a referent in an imaginary context) is not the same as for it to actually refer to a fictional individual. So, properly speaking a realist should say that a double shift occurs: first, a ‘something from nothing’-shift (when one passes to use a (para)fictional sentence connivingly, i.e., in a certain make-believe game) and second, a ‘something from something’-shift (when one passes

192

Thus, I think the scrutiny of the various intensionalist positions should lead us to the following moral. On the one hand, if one wants to stick to the idea that locutions of the form “in the story S” are intensional operators, it is better to take those operators as both circumstance- and context-shifters. For fiction is indeed characterized by such a shift. This leaves not only straightforward intensional antirealism, but also intensional realism, aside.31 Yet if these two traditional kinds of intensionalists cry, non-straightforward intensional antirealists such as pretense-intensionalists, who instead acknowledge context-shift, do not laugh. To my mind, as far as the conniving use of (para)fictional sentences is concerned, allowing for context-shift is quite right. For in that use, terms have to be interpreted with respect to the imaginary context mobilized by the relevant make-believe game. Yet there is no reason apart from adopting an intensionalist reading of IMSs to say that such a context-shift effectively affects (para)fictional sentences also when they are embedded in IMSs, as strong pretense-intensionalists claim. For this provides no satisfactory truthconditional account of IMSs. In this respect, moving that shift at the beginning of the IMS, as weak pretense-intensionalist hold, fares no better. 2. An alternative propositional account As all the intensionalist truthconditional accounts of IMSs have revealed themselves to be unconvincing, it is time to look for an alternative. In this alternative, locutions of the form “in the story S” occurring in IMSs do not express intensional operators at all. to use the very same sentence nonconnivingly). In (1999, 105) a realist like Thomasson realizes that in fictional contexts “often some pretense is involved”, but she does not draw from this fact its very consequences. 31. To be sure, the intensional realist is not forced to admit that, as regards the sentence embedded in an IMS, a context shift occurs. For (s)he might endorse descriptivism with respect to the embedded singular terms figuring in such a sentence and stick to the de re reading of the IMS. Yet this move remains ad hoc, an endorsement of descriptivism only with respect to fiction. Moreover, in this approach IMSs which are intuitively true come out as false, for the description which is taken to be a candidate synonymous for the relevant embedded singular term lacks an actual denotation. “In the Holmes stories, Holmes helps the police solve many baffling crimes” sounds true. Yet “In the Holmes stories, the pipe-smoking detective befriending a doctor named ‘Watson’ helps the police solve many baffling crimes” read de re comes out false. For the description allegedly synonymous with “Holmes”, i.e. “the pipe-smoking detective befriending a doctor named ‘Watson’” lacks an actual denotation.

193

Let me start by stressing that expressions like “story” and its cognates (e.g. “fiction”) are ambiguous between at least two readings. On one reading, such an expression means the imaginary content (para)fictional sentences possess in the very process of storytelling: for instance, when I speak of the fiction involved in The Betrothed, I mean the imaginary content possessed by sentences involved in the storytelling process originally tokened by Manzoni in writing the text of The Betrothed, and which any of us tokens again whenever she makes believe that there is a guy named “Renzo Tramaglino” and a girl named “Lucia Mondella” who intended to marry each other yet their plan was stopped by an arrogant lord named “Don Rodrigo”, and so on. On such a reading, we may legitimately speak of the world of a story. This is the imaginary world where things go the way the storytellers make-believedly say they go (the world I previously qualified as determined by a fiction). Thus, it is the world that both figures as a parameter of the unactual context relevant for giving a (para)fictional sentence imaginary truthconditions and, qua circumstance of evaluation, allows all those sentences to be fictionally evaluated (to be fictionally true or false). Yet all this does not help those sentences to have real truthconditions, which they possess when they are nonconnivingly used; or, which is the same, it does not help IMSs to have such truthconditions. For this, one has to appeal to the second reading of “story” (“fiction”, etc.). On this reading, those locutions mean the real content of those very same (para)fictional sentences. This is the content those sentences have for us as inhabitants of the real world, not the content those sentences have for us as inhabitants of an imaginary world of fiction. Inhabiting an imaginary world of fiction is the role we play when we connivingly use those sentences. Yet we get out of that role when we non-connivingly use the very same sentences, or, which is the same, we mobilize their equivalent IMSs. Now, as the real content of the (para)fictional sentences, the “story” in its second reading, I take a set of propositions including both a certain subset of explicit propositions — those that a certain group of fictional sentences in their nonconniving use mobilize — and another subset of implicit propositions — the propositions that the first subset (somehow)32 entails and which would be mobilized by other sentences were they nonconnivingly used, the parafictional sentences. This propositional set may well be called a fictional world, but only in a figurative sense. For there is a 32. There are a lot of theories on how exactly this entailment has to be conceived, but also this issue is irrelevant for my purposes.

194

sense in which the set is incomplete: for some pairs of propositions which differ only in that one contains a certain property F while the other contains its complement non-F, the set does not contain either. For instance, the propositional set that constitutes the real content of The Betrothed contains neither the proposition that Gertrude is a chess player nor the proposition that Gertrude is a non chess player — neither proposition is either mobilized by the fictional sentences of Manzoni’s text or entailed by the propositions these sentences mobilize. As a result, a propositional set of this kind does not altogether work as a circumstance of evaluation for sentences. Simply, it is (an abstract) part and parcel of the actual world. It is important to focus on this reading of the word “story”. For we can take it that, when it occurs in a locution of the kind “in the story S”, the singular term involved by that locution designates one such propositional set. For instance, “The Betrothed” refers to the propositional set that constitutes the real content of Manzoni’s tale. At this point, I am able to provide an easy truthconditional account for IMSs. A sentence of the form “in the story S, p” is true iff a certain propositional set — the set designated by the relevant singular term of the form “the story S” — contains the proposition mobilized by the embedded sentence (either fictional or parafictional) “p”. If we take again (2*), we have that (2*) is true iff the set of The Betrothed contains the proposition mobilized by the sentence embedded in it, in this case the parafictional (2). Let us look in more detail at the features of this approach. First of all, like the intensional account, it is propositional: as we have just seen, propositions directly figure within the truthconditions of IMSs. Yet it is not intensional, for such propositions are not intensions: no function from possible worlds to truthvalues is involved by those truthconditions. As I have said, the “in the story S”-locution expresses no intensional operator. Rather, it is a locative expression indicating set-membership: a proposition is in that set in the sense that it is one of its members. Qua locative expression, it may sometimes remain just understood, as it often happens with locative expressions, e.g. when one says: (11) It is raining now to mean that it is now raining in Paris. This is why an IMS is equivalent to a (para)fictional sentence in its nonconniving use. As in the case of (11), the IMS makes explicit what in the (para)fictional sentence (which

195

is actually the sentence it embeds) is implicit, namely a reference to one of its truthconditional constituents, i.e., a given propositional set. This constituent remains unarticulated in the (para)fictional sentence nonconnivingly used.33 Secondly, it is a moderately realist approach. It is realist, for it allows for singular propositions made of fictional individuals to figure in those truthconditions. For instance, a sentence like (2*) is precisely true iff in the propositional set of The Betrothed there is a certain singular proposition, namely the structured entity made by the fictional character Gertrude and by the property of having a forbidden love affair.34 As this singular proposition has a fictional character among its constituents, my truthconditional approach is thereby committed to such entities.35 Yet it is moderately realist, for the relevant propositional set may contain both singular and general propositions, made only of properties but not of individuals, let alone fictional. For example, take the sentence: (12) In The Lord of the Rings, a lot of Uruk-hai were wounded. As is well known, sentences like (12) generate a deep problem for realists on fictional entities. For if one says that the truth of this sentence commits one to fictional Uruk-hai, these alleged characters are ontologically indeter33. For the notion of an unarticulated constituent (with respect to examples like (11)), cf. notoriously Perry (1986, 138). In point of fact, the location case is not exactly identical to the one we are dealing with here. For the location may well work as an evaluation point for a sentence like (11) — the expansion of (11) articulating a certain spatial location is true iff (11) itself is true in Paris (cf. Recanati (2004, 5–6)) — whereas the proposition set designated by the term “the story S” never works as such. 34. For this conception of singular propositions, cf. notoriously Kaplan (1989). A more detailed analysis would show that it is better to conceive of such singular propositions as entities in which properties are linked to ficta in an internal way, along the lines of Zalta (1989). Cf. Voltolini (2006). 35. In fn.16, I attributed to Braun (2005) an antirealist account according to which an IMS like (2*) has a content which mobilizes a gappy proposition, and raised some criticisms against it. Yet in actual fact Braun’s account is more complex. For he also remarks that names like “Gertrude” are actually indeterminate as far as their reference is concerned and must be theoretically replaced by an empty name “Gertrude1” and by another name “Gertrude2” directly referring to a fictional character. This character inter alia constitutes “character-saturated” singular propositions that contribute to constitute a story, in something very close to my second sense above of “story”. If the second name is what actually occurs in (2*) and moreover “The Betrothed” there refers to something very close to my second sense of “story”, this explains in a different way why (2*) is true. Cf. (2005, 614fn.27). The step from these (admittedly scattered) remarks to my truthconditional account of (2*) is not that big.

196

minate: how many such fictional characters are there? What distinguishes one of them from any of the other ones? Supposing that Tolkien’s text is completely silent on these matters, it is hard for the realist to address these questions.36 Yet my approach may well leave these problems aside. For according to it, (12) is true iff in the propositional set that constitutes the real content of The Lord of the Rings there is a general proposition to the effect that there are a lot of individuals which are uruk-hai and those individuals are wounded. Yet that set is not forced to contain any singular proposition made by a fictional character U and by the property of being a Uruk-hai or by that character and the property of being wounded, or by another character U’ and that property, etc. So, in my approach the truth of (12) does not commit one to any problematic fictional entity.37 Thirdly, this approach has the possible advantage of accounting for the purported analyticity of IMSs. It is often said that such sentences, when true, are trivially so; it is hard, if not meaningless, to imagine that something may lead us to revise their truth. This cognitive feature has led some to maintain that, when IMSs are true, they are analytical truths.38 Just to avoid Quine-like controversies, suppose to stick here to the sense of analyticity appealed to by Kantians according to which a sentence is analytically true iff the meaning of the predicate is contained in the meaning of the subject. Well, any sentence saying that a certain set has a certain entity as member is analytically true in this sense. For in order to assess the truth of the sentence, one simply has to focus on the conditions of individuation of the set, namely its constituent members; in other terms, one simply has to see whether the meaning of the predicate, a certain entity, is effectively contained in the meaning of the subject, the set in question. But this is precisely what makes true an IMS in my account: insofar as a 36. Parsons (1980, 191) tries to solve this problem by saying that when individuals in fiction are spoken of collectively, the only fictional character referred to is the group of them. Yet implausibility aside, this proposal seems to me untenable. Suppose that (12) continued “… yet some of them also die”. According to Parsons, we should have here two distinct fictional characters, a bigger group and a smaller group of Uruk-hai. Yet this distinctness in characters does not account for the fact that those Uruk-hai that died belong to the bigger group of the wounded ones. For ontological scepticism regarding these cases, see also Lamarque (2003, 43). 37. The case in which the embedded singular term is a definite description is complex. On the one hand, what is literally said by the IMS embedding it is a general proposition. Yet it is clear that often people use such sentences committedly — as is the case of (1*), where the name “Gertrude” might well have replaced the definite description “the miserable girl”. For how to deal with these cases, I cannot but again refer to Voltolini (2006). 38. See Bonomi (1979, 46–8); Castañeda, e.g. (1985/6, 58–9).

197

certain set is made by certain propositions and not others, the sentence is eo ipso true.39

REFERENCES Adams, F., Fuller, G., Stecker, R. 1997. “The Semantics of Fictional Names”. Pacific Philosophical Quarterly 78, 128–148. Bonomi, A. 1979. Universi di discorso. Milan: Feltrinelli. — 2006. Fictional Contexts [online]. Available from: www.filosofia.unimi. it/~bonomi/ Braun, D. 1993. “Empty Names”. Noûs 27, 449–469. — 2005. “Empty Names, Fictional Names, Mythical Names”. Noûs 39, 596– 631. Castañeda, H.-N. 1985/6. “Objects, Existence, and Reference. A Prolegomenon to Guise Theory”. Grazer Philosophische Studien 25/6, 3–59. Crimmins, M. 1998. “Hesperus and Phosphorus: Sense, Pretense, and Reference”. The Philosophical Review 107, 1–47. Currie, G. 1988. “Fictional Names”. Australasian Journal of Philosophy 66, 471– 488. — 1990. The Nature of Fiction. Cambridge: Cambridge University Press. Evans, G. 1982. The Varieties of Reference. Oxford: Clarendon Press. Goodman, J. 2005. “Defending Author-Essentialism”. Philosophy and Literature 29, 200–208. Kaplan, D. 1989. “Demonstratives”. In: J. Almog et al., eds. Themes from Kaplan. Oxford: Oxford University Press, 481–563. Kroon, F. 2004a. “Descriptivism, Pretense, and the Frege-Russell Problems”. The Philosophical Review 113, 1–30. — 2004b. “Millian Descriptivism”. Australasian Journal of Philosophy 82, 553– 576. Lamarque, P. 2003. “How to Create a Fictional Character”. In: B. Gaut, P. Livingston, eds. The Creation of Art, Cambridge: Cambridge University Press, 33–52. Lewis, D. 1978. “Truth in Fiction”. American Philosophical Quarterly 15, 37– 46. — 1986. On the Plurality of Worlds. Oxford: Blackwell. 39. I thank Fred Kroon and Elisabetta Sacchi for their important comments to previous versions of this paper.

198

Parsons, T. 1980. Nonexistent Objects. New Haven and London: Yale University Press. Perry, J. 1986. “Thought Without Representation”. Proceedings of the Aristotelian Society suppl.vol. 60, 137–151. Plantinga, A. 1978. “The Boethian Compromise”. American Philosophical Quarterly 15, 129–138. Predelli, S. 1997. “Talk about Fiction”. Erkenntnis 46, 69–77. Priest, G. 2005. Towards Non-Being: the Logic and Metaphysics of Intentionality. Oxford: Clarendon Press. Recanati, F. 1993. Direct Reference. Oxford: Basil Blackwell. — 2000. Oratio Obliqua, Oratio Recta. Cambridge (Mass.): The MIT Press. — 2004. “Indexicality and Context-shift”. Conference Paper in Workshop on Indexicals, Speech Acts and Logophors, Harvard University Press. — 2006. “Reply to Voltolini”. In: M.J. Frapolli, ed. Saying, Meaning and Referring: Essays on François Recanati’s Philosophy of Language. Houndmills: Palgrave Mcmillan (forthcoming). Salmon, N. 1998. “Nonexistence”. Noûs 32, 277–319. Schiffer, S. 1996. “Language-Created Language-Independent Entities”. Philosophical Topics 24, 149–166. — 2003. The Things We Mean. Oxford: Clarendon Press. Thomasson, A. L. 1999. Fiction and Metaphysics. Cambridge: Cambridge University Press. Voltolini, A. 1995. “Indexinames”. In: J. Hill and P. Kot’àtko, eds. Karlovy Vary Studies in Reference and Meaning, Prague: )LORVRILD-Filosofia Publications, 258–285. — 2006. How Ficta Follow Fiction. Dordrecht: Springer. — unpublished. “Indexinames (again)”. (forthcoming). Walton, K.L. 1990. Mimesis as Make-Believe. Cambridge (Mass.): Harvard University Press. Zalta, E. N. 1983. Abstract Objects. Dordrecht: Reidel. — 1988. Intensional Logic and the Metaphysics of Intentionality, Cambridge (Mass.): MIT Press. — 1989. “Singular Propositions, Abstract Constituents, and Propositional Attitudes”. In: J. Almog et al., eds. Themes from Kaplan, Oxford: Oxford University Press, 455–478.

199

This page intentionally left blank

Grazer Philosophische Studien 72 (2006), 201–210.

FICTIONAL PROPOSITIONS AND THE UNPROVABILITY OF CONSISTENCY Enrico MARTINO University of Padova Summary We introduce an epistemic version of validity and completeness of first order logic, based on the notions of ideal agent and fictional model. We then show how the perspective here considered may help to solve an epistemic puzzle arising from Gödel’s second incompleteness theorem.

1. Fictional models Let L be a first order language. According to model theory, the L-sentences are syntactical entities that become true or false propositions only with reference to a certain interpretation of L. We learn from proof theory how to draw from a given set of sentences logical consequences in full accordance with model-theoretical semantics. According to the soundness and completeness theorems, the sentences formally derivable from a set of assumptions are exactly those true in all models of such assumptions. We want to remark that the soundness and completeness theorems show that formal derivability is sound and adequate not only to the modeltheoretical notion of truth, but also to the intuitive notion of abstract deducibility, as understood by a working mathematician, whose reasoning is not driven by syntactical inference rules but by an abstract notion of truth. Abstract mathematical reasoning is not concerned with non-interpreted sentences but with genuine propositions that are supposed to be endowed with well-determined truth values. The problem arises: what has the single abstract interpretation of the working mathematician to do with the various possible interpretations of model theory? When a theory has elementary non-equivalent models, how can one deduce logical consequences of the axioms by reasoning on a single interpretation? The answer seems to be plain: the mathematician reasons about an arbitrary model of

the axioms. We believe that this answer is essentially correct but in need of clarification. We need to make the notion of arbitrariness involved here explicit. What is an arbitrary interpretation? Is it an interpretation of some special kind, well distinct from the ordinary ones? Fictional characters are sometimes conceived of as incomplete objects. Is an arbitrary interpretation to be understood as an incomplete thing in a similar sense? Perhaps it would be possible to extend the ontology of ordinary interpretations by adding incomplete interpretations, but we think that it would be highly artificial and problematic. In particular, usual mathematical reasoning is in agreement with classical bivalent logic, which seems to be inappropriate, when dealing with incomplete objects. One could think, at first sight, that reasoning about an arbitrary interpretation amounts to reasoning simultaneously on all possible interpretations. It is not so, however. Abstract reasoning cannot be taken as referring to all possible interpretations, it genuinely refers to a single interpretation. A similar question arises about arbitrary reference to individuals. We have discussed the question extensively in (Martino 2001). We will resume those considerations here, in order to extend them to arbitrary interpretations. Russell (1908) accounts for the distinction between free and quantified variables (real and apparent in his terminology) by pointing out that, in order to prove a universal proposition, which attributes a property to all individuals (in the universe of discourse), one must reason on a single arbitrary individual. He observes that, in order to establish, for instance, that in every triangle a side is minor than the sum of the others, one carries on the proof for a single triangle, not for all triangles simultaneously. But, he continues, the triangle in question is absolutely ambiguous, and ambiguous is therefore our assertion about it. In this sense the triangle at issue is a free variable. And this ambiguity is precisely what justifies the assertion that the conclusion reached for that single triangle holds for all triangles, in other words it justifies the introduction of the universal quantifier. Russell’s remarks clearly show that abstract reasoning for establishing a universal proposition must proceed through the consideration of a single arbitrary individual. In today’s formal logic the need of reasoning about single arbitrary objects is quite perspicuous in the introduction and elimination rules of quantifiers in natural deduction. Whenever a formula A(x) has been derived (with x a free variable not occurring in any assumption the formula depends upon), the introduction rule for the universal quan-

202

tifier allows the conclusion (x)A(x). Similarly for the elimination rule of the existential quantifier. What we find unsatisfactory in Russell’s explanation is his interpretation of arbitrariness in terms of ambiguity. What are ambiguous objects? Rejecting the hypothesis that ambiguity affects the nature of the object, it seems rather to affect the act of referring. It seems that referring to an arbitrary object should be understood as ambiguously referring to an ordinary object. But we have no clear understanding of what an act of ambiguously referring is. Besides, the referent of such an act would be undetermined. That would be incompatible with the basic requirement, emphasized by Russell himself, that an arbitrary object must remain the same along the whole course of the argument. This means that, if the letter “a” indicates an arbitrary object, all its occurrences in the argument must denote the same object. If the referents of such occurrences were not well-determined objects, it would be meaningless to say that they are the same. We think therefore that the ambiguity shown by Russell is to be understood in a purely epistemic sense. Referring to an arbitrary object a amounts to supposing that “a” designates an unknown, though well-determined, object. Being well-determined justifies the behavior of “a”, in the course of the reasoning, as a name designating the same object in all its occurrences. On the other hand, being unknown guaranties that all that is recognized about it holds for any other object of the domain as well. Since nobody, however, as a matter of fact, has assigned any object to “a”, this cannot be anything but a fictional name and, consequently, any statement about a is to be a fictional proposition. The notion of arbitrary reference can be made precise by introducing an ideal agent. Let us imagine that we have direct access to him and that he in turn has direct access to every object: he can choose any object at will (here we are identifying ourselves with the working mathematician carrying on the mathematical reasoning). We can explain the locution “Let a be an arbitrary object” as follows: we ask the agent to choose an object at his will (without communicating to us anything about the chosen object) and call it “a”. It is clear that the adjective “arbitrary” does not concern the nature of the chosen object, but the freedom of the act of choice. So “a” is a fictional name, whose referent is fixed by means of an imaginary act of choice. In a similar way we will understand the notion of arbitrary interpretation of a first order language L. Let us imagine an ideal mathematician 6 and an ideal agent W. 6 reasons in agreement with classical logic, never makes any

203

mistake and is free of empirical space-temporal limitations. So 6 is able to grasp any formal proof and recognize it as sound. The ideal agent W is able to fix at will any interpretation of language L. Imagine we can communicate with 6 and W. Let us order W to choose at will an interpretation I of L and give 6 the relevant pieces of information relative to the event of the choice, so that he can refer to it without ambiguity, but not to communicate to him anything concerning the specific interpretation I chosen by W. So, by referring to that act of choice, 6 can refer to I, without knowing anything about it (beyond that it is an interpretation of L). We will say that I is a fictional interpretation of L and that the L-sentences, interpreted in I, are fictional propositions. Thus 6 is able to interpret any L-sentence A in the (fictional) meta-linguistic proposition “A is true in I”. We say that an L-sentence A is epistemically valid if 6 can recognize its truth in a fictional interpretation I. Since 6 has no piece of information about the content of I, he can recognize A only if A holds in all interpretations, i.e. if it is logically valid. Vice versa, if A is logically valid, it is, by semantic completeness, formally provable. And since 6 is free of spatiotemporal limitations, he can grasp such a formal proof of A and recognize the truth of A. Thus: An L-sentence is logically valid if, and only if, it is epistemically valid. Epistemic validity is easily extendable to epistemic consequence. Let * be a decidable set of L-sentences. We say that A is an epistemic consequence of * if, with reference to a fictional interpretation I of L, 6 can recognize A under the hypothesis that all sentences in * are true in I. Call such hypothesis the *-hypothesis. This is nothing but the assumption that I is a model of *. Observe that the *-hypothesis is not to be understood as a condition on the choice of W. We cannot, in general, command W to choose a model of *, since * might be contradictory and hence devoid of models. Besides, even if * is consistent, it is not required, for our purposes, that W be able to determine a model of it. I is an absolutely arbitrary interpretation of language L, which 6 assumes to be a model of *. More formally, we say that the pair M = (I, *) is a fictional model of *. A sentence is epistemically true (for short e-true) in the fictional model M if it is an epistemic consequence of *. Of course e-truth in M may fail to coincide with ordinary truth in I, since I may fail, as a matter of fact, to be a model of *. The e-true sentences are exactly those recognizable by 6 as true under the *-hypothesis. If * is contradictory, every L-sentence

204

is e-true in any fictional model of *. Obviously all fictional models are elementary equivalent with respect to e-truth. From semantic completeness it follows immediately that A sentence A is a logical consequence of * if, and only if, it is an epistemic consequence of *, if, and only if, it is e-true in any fictional model of *. Semantic completeness plays an essential role in the equivalence between logical and epistemic consequence. That may be one of the reasons for preferring first order to higher order logic. For the latter there is no evidence that the notion of logical consequence be adequate, even in principle, to represent what is deducible through abstract reasoning. For an axiomatic first order theory we can translate the semantic completeness theorem into the following: Epistemic completeness theorem. The theorems of an axiomatic first order theory are exactly the epistemic consequences of the axioms, i.e. the sentences e-true in any fictional model of the axioms. We want to exploit the notion of epistemic consequence in order to solve a puzzle arising from the popular understanding of Gödel’s second incompleteness theorem. 2. The unprovability of consistency Gödel’s second incompleteness theorem (Gödel 1931, 1931a), in its popular version, runs as follows: (1) Popular version of Gödel’s theorem. An axiomatic consistent first order theory, endowing a suitable fragment of arithmetic, cannot prove its own consistency. This version was formulated by Gödel himself in the famous postscript to his (1931a) paper. Recall the arithmetical expression of consistency. Let L be the language of first order arithmetic and S a decidable extension (consistent or not) of Peano’s axioms PA. Start from the metalinguistic relation that correlates a sentence with a

205

(possible) proof of it. This is expressible arithmetically through a formula Prf (x, y) (read: y is a proof of x) such that (i) if n is the code of an S-proof of Ithen Prf (§I·n) is derivable from Peano’s axioms PA (where §I·is the code of I and n the numeral of n); (ii) if n is not the code of an S-proof of I, then ¬Prf (§I·n) is derivable from PA. Define the arithmetical translations of provability and consistency by putting Con =def ™P(§0 = 1·).

P (x) =def yPrf (x, y),

The precise statement of Gödel’s theorem is the following: (2) If S is consistent, then Con fails to be S-provable. (1) is the popular reading of (2). By epistemic completeness, it follows from (2) that Con fails to be an epistemic consequence of S. So, if M = (I, S) is a fictional model of S, Con fails to be e-true in M. This means that 6 is not able to recognize the truth of Con in I under the assumption that I is a model of S. Thus, according to the popular reading, we should conclude that 6 cannot recognize the consistency of S, even under the assumption that S has a model. But that is highly puzzling: 6 knows that if a theory has a model, then it is consistent! Of course, 6 doesn’t know whether I is really a model of S, but that doesn’t matter. The point is that he reasons under the assumption that I is a model of S. His consistency proof is the following Simple proof. The axioms of S are true in I (by the S-assumption), the inference rules are truth-preserving, hence every theorem of S is true in I; no contradiction can be true in I, hence no contradiction is derivable. Since the soundness of the first order logical rules counts (for a classical mathematician) as a logical law par excellence, the simple proof seems to be a perfectly logical reasoning from the axioms of S to their consistency. However, the argument to the effect that the consistency of a theory is a trivial consequence of its axioms has often been criticized on the basis

206

of a strange objection that we want to challenge. It is objected that the mere assumption of the axioms does not commit to their truth, since hypothetical reasoning can rightfully assume any set of sentences, true or false, consistent or not. So, the objection goes, the assumption of the truth of the axioms is a further assumption. We can use certain assumptions without knowing if they are true; furthermore, the objection continues, hypothetical reasoning is even a powerful means for discovering their possible falsity by investigating their consequences. In this vein it has often been argued that the reason why we can recognize the truth of Gödel’s undecidable sentence G of PA is that we, unlike any machine, are very confident that Peano’s axioms are true, because of our familiarity with arithmetical notions, so that no doubt can arise about their consistency. The formal system, in contrast, cannot exploit the truth of the axioms. We maintain that this kind of argument rests on a misunderstanding. Let us consider how a working mathematician reasons. Of course, when she tentatively proposes an axiomatic theory S, she must be well aware of the danger that the axioms may turn out to be contradictory. But, as far as she is drawing consequences from her axioms, she must pretend to know that they are true. She reasons abstractly in terms of truth in an arbitrary hypothetical model of S. In terms of our fictional model, 6 is well aware that S may be inconsistent, but, when reasoning about the fictional model M=(I, S), he pretends to know that all axioms of S are true in I and that, consequently, their consistency is granted. However, the objection at issue against the simple proof may seem to be supported by the following argument. If the statement that what is S-provable is true were not a further assumption beyond the axioms, it shouldn’t help to prove any S-unprovable L-sentence. However, the argument goes, we can formalize the simple proof as follows. Extend the language L to a language Lc by introducing the truth predicate T for L and extend S to S’ by introducing the axiom schemata (3) P(§I·) oT(§I·), (4) T(§I·) lI

(for any L-sentence I)

We get a non-conservative extension of S , within which Con turns out to be a theorem. For, from P(§0 = 1·), by using the new schemata, we get T(§0 = 1·) and then 0 = 1, whence, by reduction, Con. So, (3) and (4) certainly do express something not already implicit in the axioms of S. That seems to refute our main claim that assuming that

207

what is provable is true is not a further assumption beyond the axioms. On the basis of the deductive power of such schemata, Shapiro (1998) has argued against the deflationist conception of truth: the non-conservativity of S’ over S would show that truth is a thick notion, not a thin one as the deflationists claim. It doesn’t do to reply that, though (3) and (4) are implicitly granted by the axioms of S, they cannot be exploited in L for the reason that the truth predicate T of L, unlike P, fails to be expressible in L itself. If they were implicitly granted by the S-axioms, so would be the L-schema (5) P (§I·) oI      (reflection principle) as well, so that this would hold in all S-models and, by the completeness theorem, it would be S-derivable and Con would be an S-theorem. So, why is 6, though he knows that, under the S-assumption, what is S-provable is true in I, nevertheless is not able to recognize, under the same assumption, that (5) is true in I? The answer is plain: the arithmetical predicate P may fail to express S-provability in I, even if I is actually a model of S. In fact, I might be a non-standard model of arithmetic, where the truth of P(§I·) does not guarantee the provability of I. This is the reason why the consistency of S does not imply the e-truth of Con. In other words, what 6 cannot recognize, when reasoning about his fictional model, is not the consistency of S, but the alleged meaning of Con as the arithmetical translation of consistency. P is a faithful translation of provability only in the standard model of arithmetic. But the axioms fail to determine the standard model and therefore the meta-linguistic meaning of P. It has been often argued that the Hilbertian finitary methods of proof are formalizable in PA. Gödel (1931a) cautiously does not endorses this thesis. In the above quoted paper he writes: For a system in which all finitary (that is intuitionistically unobjectionable) forms of proof are formalized, a finitary consistency proof, such as the formalists seek, would thus altogether impossible. However, it seems questionable whether one of the system hitherto set up, say Principia mathematica, is so all-embracing (or whether there is a system so all-embracing at all).

Gödel takes for granted that Con expresses the consistency of S and therefore questions whether there exists a system in which all finitary forms of proof are formalizable. But what we want to hold is that no for-

208

mal system is adequate even to merely express its own consistency. Gödel’s arithmetization of syntactical notions shows that the syntactical structure of a formal system is isomorphic to the standard model of arithmetic. But no formal system can characterize such a model. Hilbert’s finitism is grounded on the notion of a finite string of signs, where the attribute “finite” is to be understood in an absolute sense. And this absolute sense is just what the axioms are not able to catch, because of the impossibility to banish non-standard models. At this point it should be clear that the popular version of Gödel’s theorem is misleading. It takes for granted the syntactical reading of the arithmetical predicate P and therefore identifies the unprovability of Con with the unprovability of consistency. However, as we saw, what Gödel’s theorem shows is rather that, for any L-predicate P, no system S of axioms can guarantee that P faithfully represents meta-linguistic provability. More precisely, the argument for the unprovability of Con, applied to any L-predicate P, shows that If S is a consistent extension of PA, there is no L-predicate P such that, for every model M of S and every L-sentence ), P(§I ·) is true in M if, and only if, ) is S-provable. We conclude that, despite what is commonly believed, Gödel’s theorem has nothing to say about the provability of the genuine consistency of the theory. What it points out is rather a remarkable inadequacy of the axiomatic method, strictly connected with the phenomenon of incompleteness: the impossibility to force, by means of suitable axioms, the material equivalence of an L-predicate with the genuine meta-linguistic provability predicate. The behavior of the latter is not so far from that of the truth predicate as it is usually believed. We saw that S-provability is equivalent to e-truth in a fictional model of S. And, if S is consistent, as no L-predicate, interpreted in an ordinary model M of S, can express the truth predicate of M, so no L-predicate, interpreted in a fictional model M of S, can express e-truth, i.e. S-provability, in M.

209

BIBLIOGRAPHY Gödel, K. 1986. Collected works. Vol. I. Ed. by S. Feferman, S. Oxford: OUP. — 1931. “On formally undecidable propositions”. In: K. Gödel. Collected works. Vol. I. Ed. by S. Feferman, S. Oxford: OUP, 145–195. — 1931a. “Discussion on providing a foundation for mathematics”. In: K. Gödel. Collected works. Vol. I. Ed. by S. Feferman, S. Oxford: OUP, 201–205. Martino, E. 2001. “Arbitrary reference in mathematical reasoning”. Topoi, 20, 65–77. Russell, B. 1908. “Mathematical logic as based on the theory of types”. In: J. van Heijenoort, ed. From Frege to Gödel. Cambridge, Mass: MIT Press, 150– 182. Shapiro, S. 1998. “Proof and truth: through thick and thin”. The journal of philosophy, 45, 10, 493–521.

210

Grazer Philosophische Studien 72 (2006), 211–231.

PROPOSITIONS AND NECESSARY EXISTENCE Vittorio MORATO University of Padua Summary Timothy Williamson in his article “Necessary Existents” presents a proof of the claim that everything necessarily exists using just three seemingly uncontroversial principles relating the notions of proposition with those of truth and existence. The argument, however, may be easily blocked once the distinction, introduced by R. M. Adams, between the notions of a proposition being true in a world and of (or at) a world is introduced. In this paper I defend the plausibility of the notion of a proposition’s being true of a world by rejecting two criticisms of it raised by Williamson; in the final section, I present a conception of propositions, according to which they are equivalence classes of mental representations, for which at least one of the principles comes out as false.

1. Introduction The conclusion that we, and with us everything else, are necessary existents seems to be surprisingly easy to establish. For example, if you take the simplest system of propositional modal logic, S5, the classical theory of quantification and identity, CQT, and combine them in what it seems to be the most straightforward manner, the conclusion that everything exists necessarily, xLy(x = y), is reached by an elementary derivation as the following: (1) y(x = y) Theorem of CQT (2) Ly(x = y) 1, Rule of Necessitation (3) xLy(x = y) 2, Rule of Universal Generalization1 1. In Kripke’s system of quantified modal logic, where the combination of quantification and modality is instead substantially qualified, the derivation above would not be valid: the rule of necessitation cannot be applied to an open formula. See Kripke (1963).

The same conclusion seems to be forced on us even by some principles governing quite standard interactions of the notion of proposition with those of truth and existence. T. Williamson (2000) presents a proof for the claim that everything necessarily exists using just some of these widely shared principles. 2. Williamson’s proof Williamson’s proof relies on three principles; the first is the following: Principle 1: Necessarily, if P then the proposition that P is true where P is a schematic letter and it is substitutable by any declarative sentence. Such a principle codifies the relation between something’s being the case and the truth of the proposition according to which something is the case; as Williamson writes: “for that things are so-and-so is just what it takes for the proposition that they are so-and-so to be true”.2 Principle 1 is a special case of O-conversion: where )n is a first-order formula and t1 … tn any variables, the expression O(t1 … tn )n) is a term denoting the n-place relation expressed by )n; where n = 0, O()) is a term denoting the proposition expressed by the formula for which the schematic letter ) is substituted. Principle 1 is thus the left-to-right direction of the following more general principle: L()l7(O()))) where T is the truth-predicate. One of the theoretical roles of Principle 1 is an expressive one: it enriches the resources of a given language allowing to talk directly about propositions (thus treating them as objects) and explicitly attributing to them properties such as truth and existence. The principle, however, has mainly a substantial role: it regulates the connection between what is the case and propositional truth. The second principle is the following: Principle 2: Necessarily, if the proposition that P is true, then the proposition that P exists. 2. See Williamson (2002, 234).

212

Such a principle codifies the relation between propositional truth and existence: it claims that it cannot be the case that a proposition is true and it does not exist; if the proposition did not exist, there would have been nothing to which its truth could have been suitably attributed. As R.M. Adams writes “a proposition must be in order to be true”3. Principle 2 is a special case of a more general principle that is usually called serious actualism4. Serious actualism is the view that no object could have had a property without existing (for any world w, no object has a property in a world w where it does not exist), namely that exemplification entails existence. Principle 2 is a special instance of such a view where the object is a proposition and the property is truth. From a semantical point of view, what serious actualism requires is that an object in the extension of a predicate (at a world) must fall under the range of the existential quantifier (at that world); this can be expressed by the following first-order schema5: L()ox(x = t)) where ) is atomic and t is a singular term (an individual constant or a variable) that occurs free in ). Finally, the third principle is the following: Principle 3: Necessarily, if the proposition that P exists and o is a constituent of P, then o exists. According to such a principle, the existence of its constituents is a necessary condition for the existence of a proposition; if the existence of an entity ) is a necessary condition for the existence of an entity <, then < ontologically depends on ). The principle, then, claims that propositions ontologically depend on their constituents. A conception of propositions according to which such entities are objectual and structured set-theoretic constructions is often coupled with the principle. A proposition is structured if its structure is induced by the syntactic structure of the sentence that expresses it while it is objectual if the sentence’s singular terms contribute their referents to the proposition expressed. Singular terms that contribute their referents to the proposition 3. See Adams (1981, 18). 4. See Plantinga (1983). 5. See Menzel (1991, 336).

213

expressed by the sentence containing them are directly referential terms6 while the propositions expressed by atomic sentences containing such terms are singular propositions. Intuitively, a singular proposition expressed by an atomic sentence whose subject is “John” is, if “John” is a genuine name, “directly about” John. If singular propositions enter the picture, Principle 3 could be taken as the view that Plantinga (1983, 3), calls existentialism, namely the view that “a singular proposition is ontologically dependent upon the individuals it is directly about”. As the schema above clearly suggests, this principle too could be taken as an instance of serious actualism. The principle, however, adds an emphasis on the relation of constituency and it may be read simply as an instance of the view that the existence of a whole grants the existence of its component parts (trivially because it ontologically depends on its components). Let us see now how Williamson’s proof actually works. Assume that P stands for the sentence “I do not exist”; by Principle 1 we have: Necessarily, if I do not exist, then the proposition that I do not exist is true. By Principle 2 we have: Necessarily, if the proposition that I do not exist is true, then the proposition that I do not exist exists. And finally, by Principle 3: Necessarily, if the proposition that I do not exist exists, then, I exist. Now, by the transitivity of strict implication, we obtain: Necessarily, if I do not exist, I exist, from which, by consequentia mirabilis, we can conclude that: Necessarily, I exist.

6. See, for example, Kaplan (1989) where directly referential devices are characterized just in this way.

214

Finally, given that the indexical expression “I” does not play any special role qua indexical, it could be substituted by any free variable (under an assignment of values to variables); the proof could thus be safely generalized to the conclusion that: Everything necessarily exists. The notions of existence and necessity involved in the proof allow different interpretations. There are broad conceptions of existence and necessity (for example, logical notions of existence and necessity) as much as restricted notions (for example, naturalistic notions of existence, like spatio-temporal locatedness and causal efficacy and non-logical notions of necessity, like metaphysical or physical necessity). Williamson claims to be using a broad, logical, sense of existence (see Williamson 2002, 244) and a restricted, metaphysical notion of necessity. The conclusion of the argument is thus that everything is a metaphysically necessary existent in a logical sense. It could be asked whether Williamson’s proof works only if the existence of propositions is assumed. Let us see then what happens in the case we substitute propositions with some other surrogate entities. An analogous version of the argument where utterances (sentence tokens) are used instead of propositions and where truth is taken to be a property of utterances does not seem to work; an instance of a modified version of Principle 1 as: Necessarily, if Vittorio does not exists, then the utterance “Vittorio does not exist” is true seems to be false. Utterances are typically taken to be physical entities and there is no reason to postulate a necessary connection between something’s being the case and a physical entity of any sort, specially if it is an utterance. In particular, it seems implausible to postulate a necessary connection between my possible non existence and a physical entity like the utterance “Vittorio does not exist”: why in every world in which I do not exist, should the utterance in question be uttered? It is perfectly plausible in fact to conceive of a possible world where I do not exist but where no speaker utters the sentence “Vittorio does not exist”. Equally implausible seems to be that there be a necessarily possible connection between my possible non existence and the utterance “Vittorio does not exist”: if I do not exist in a world wi, why should there be a world w where the utterance is uttered?

215

More delicate is the case of sentences (sentence types); it all depends on what is the definition of a sentence and what are the conditions under which a sentence exists in a possible world. Suppose that, quite standardly, a sentence is characterized as an equivalence class of sentence tokens. If a sentence ) exists in a possible world w if and only if, in w, there is (at some time t) at least a sentence token belonging to the appropriate equivalence class, then a problem similar to that of utterances arises again. There might be a possible world where I do not exist but the sentence “Vittorio does not exists” is never uttered because, for example, no speaker existing in w uttered my name.7 If, on the other hand, a sentence ) exists in a possible world w if and only if it is possibly uttered, namely if and only if there is at least a world wi where there is (at some t) at least a sentence token belonging to the appropriate equivalence class, the problem is that it seems implausible to assume that if it is the case that ), the sentence ‘)’ is necessarily, possibly uttered, i.e., that, among all possible worlds, there surely is at least one where ‘)’ is uttered at some time. For example, from the fact that I do not exist in a possible world w, why should follow it that surely there is at least one world where someone utters “Vittorio does not exist”? There might be an interpretation where, in every possible world of the interpretation, no speaker utters this sentence.8 In general, it seems implausible not only to suppose that there is a neccessary connection between something’s obtaining in a world and a physical entity of some sort but even a necessarily possible connection between the two, especially in case the physical entity is an utterance (on which sentences ontologically depend). The problem, however, is also that even in those lucky interpretations where 7. Those who believe that a proper name is a genuine one only if it is introduced by a speaker s standing in some kind of causal relation with the nominee will hold that, in a possible world where I do not exist, no name for me could ever be introduced. It is a widely shared view, and even among proponents of direct reference, that genuine names could also be introduced by definite descriptions. In the case under consideration, however, there is a problem: from the perspective of a possible world where I do not exist, I am a merely possible object, but all definite descriptions for alleged possible objects seem to be improper; for example, the description “the possible child of Wittgenstein” is improper because many distinct merely possible objects are, in some possible world, fathered by Wittgenstein; descriptions for merely possible objects are improper, however, even for a deeper reason than just the multiplicity of denotations but even because it seems simply indeterminate how many denotations a definite description for a possible object might have. 8. Both the arguments above essentially depends on a quite standard thesis according to which, necessarily, a set exists only if its constituents do; K. Fine (1994, 4) calls this “the standard view within modal set theory”

216

Principle 1 comes out as true (namely where there is at least a necessary possible connection between what is the case and relevant utterances), the corresponding reformulated version of Principle 3, namely: Necessarily, if the sentence “Vittorio does not exist” existsts and “Vittorio” is a semantically evaluable constituent of it, then the semantic value of “Vittorio” exist. comes out as false. Consider in fact an interpretation where Principle 1 is true and where in a world w it is the case that Vittorio does not exist and the sentence “Vittorio does not exist” exists; given the notion of sentential existence in use, our sentence exists in w even if it is not uttered in w but in some other world wi; but Principle 3 requires that wherever a sentence exists, the semantic values of its semantically evaluable components exist. Hence, under the conception of sentences as equivalence classes and the “relaxed” notion of sentential existence, at least one of the principles of Williamson’s proof comes out false. There is then not a single notion of sentential existence for which all three principles come out true together. Other conceptions of sentences might be available according to which such entities are necessarily abstract entities, that exist independently of the existence of their utterances and this because they exist already once a primitive vocabulary and syntactic rules for a language are given. This position, however, does not, by itself, exclude counter-examples like those presented above, in case the primitive vocabulary or the formation rules are, in turn, contingently existing entities or ontologically dependent on contingently existing entities.9 If we reason in strict analogy with what happens in the formal semantics for modal logics, however, the basic elements of a language may be taken as already given “before” the various possible worlds enter the play; the existence of a language (and hence of sentences) could then be seen as some sort of a “transcendental condition” of the logic and therefore as independent of any contingency represented by what is going on within worlds.

9. In the case, for example, in which letters are taken to be equivalences classes of physically realized marks.

217

3. Dealing with the proof Williamson’s proof is, basically, a reductio of the thesis that I contingently exist. In general, what the argument shows is that there is an incompatibility between the thesis that something exists contingently and the three principles. Given that I believe that the thesis that everything exists necessarily is false, my strategy will consist in assuming that contingent existence is true and showing that at least one of the principles in the proof is false and then that the argument is unsound. In the next section I will try to rebut two criticisms that Williamson makes against a certain way a proposition could be said to be true with respect to a world, a way by means of which some of the principles could be qualified and the argument blocked; in the last section I will claim that according to a quite plausible conception of propositions, compatible with Williamson’s characterization of such entities, at least one of the principles involved in the proof should be taken as false. 3.1 The truth in/truth at distinction A way of dealing with the argument may be that of postulating an ambiguity in the way a proposition can be true with respect to a world. R. Adams (1981) introduced the distinction between the notions of a proposition’s being true in a world and that of a proposition being true at a world. The idea is that a proposition may be true at a world without it being true in a world. K. Fine (1985, 163) pointed to a similar distinction when he introduced the distinction between an inner and an outer notion of propositional truth: One should distinguish between two notions of truth for propositions, the inner and the outer. According to the outer notion, a proposition is true in a possible world regardless of whether it exists in that world; according to the inner notion, a proposition is true in a possible world only if it exists in that world. We may put the distinction in terms of perspective. According to the outer notion, we can stand outside a world and compare the proposition with what goes on in the world in order to ascertain whether it is true. But according to the inner notion, we must first enter with the proposition into the world before ascertaining its truth.

The inner propositional truth corresponds to the notion of a proposition’s being true in a world, the outer propositional truth corresponds to the 218

notion of a proposition’s being true at a world. The distinction was explicitly designed to deal with cases of negative existential claims in the framework of an actualist conception of possible worlds semantics where maximal consistent sets of structured, objectual and, of course, actually existing propositions (usually called “world-stories”) play the role of possible worlds.10 Where a is an actual object, P is the proposition expressed by the formula ™x(x = a), w is a world-story and a is not a constituent of w, P is not true in w (because in w there is not the material to build up the proposition) but is true at w. This condition seems a little ad hoc, but I think that if possible worlds are viewed as representations of some sort (as it is the case with the world-stories approach and, in general, with many actualist approaches) the condition above may receive some independent support. The intuitive idea is, roughly that a representation of the non existence of an object x should be a representation that does not count x among its constituents; a representation of x’s non existence is done by omitting x from the representation in question, not by “encoding” in the representation the explicit information that x does not exist. A world-story is supposed to be exactly a representation or a description of a way the actual world could have gone differently and it is standardly assumed that there are as many world-stories as there are ways the actual world could have gone differently. This distinction could be used to qualify the first premise of Williamson’s argument. The instance of Principle 1 would become ambiguous between: (1.1) for every world w, if in w, I do not exist then the proposition that I do not exist is true in w. (1.2) for every world w, if in w, I do not exist then the proposition that I do not exist is true at w. A fan of the distinction denies (1.1) and accepts (1.2) and by this move she may be able to block the argument. On the one hand, Principle 2 cannot, in fact, be used together with (1.2): the consequent of (1.2) is about a proposition’s being true at a world while the antecedent of Principle 2 is about a proposition’s being true in a world; on the other hand, a version 10. See Adams (1981, 22).

219

of Principle 2 is false: the truth of a proposition P at a world w does not always entail the existence of P in w (in the case P is a negative existential, it never entails the existence of P in w). Williamson is perfectly aware of this distinction; he calls, as is sometimes done in the literature, the notion of truth at truth of, and he then presents two criticisms of the possibility of giving a plausible definition of the notion of a proposition’s being true of a world; the first is that such a notion, given the conception of possible worlds as maximal consistent sets of propositions, would pose a problem of circularity,11 the second is that such a notion, if it were understood in analogy with the standard notion of an open sentence’s being true of an object, would not be “faithful to our understanding of modal vocabulary”.12 As for the first. According to Williamson13, the truth of a principle like 1 needs to be assumed in order to explain the standard notion of a valid argument. The standard notion of valid argument is one according to which necessarily, if the premises of an argument are true, so is the conclusion.14 If Principle 1 is understood in terms of the notion of a proposition’s being true of a possible world, the standard notion of a valid argument becomes one according to which, for any world w, if the premises of an argument are true of w, then the conclusion is true of w. This characterization trades in the technical notion of possible world. Now, assume that, as I have done above, a possible world is defined as a maximal consistent set of propositions. A set of propositions X is consistent if and only if for any proposition P, if there is a valid argument from X to P, then there is not a valid argument from X to ™P. A set of propositions X is maximal if and only if, for any proposition P either there is a valid argument from X to P or there is a valid argument from X to ™P. The definition of a possible world as a maximal consistent set of propositions essentially uses the notion of validity. But then we are in a circle: Principle 1 grounds the notion of validity, the notion of validity is essentially used to define the notion of possible world, the notion of possible world is used in the formulation of Principle 1 in terms of the notion of a proposition’s being true of a possible world. 11. Cf. Williamson (2002, 238). 12. Cf. Williamson (2002, 238–240). 13. Cf. Williamson (2002, 236). 14. As Williamson writes, this is not the purely logical notion of validity but rather the one used, for example, in evaluating the validity of arguments whose premises are counter-factual. Cf. Williamson (2002, 236).

220

As for the second. According to Williamson15, a plausible way to understand the notion of a proposition’s being true of a world is by analogy with the notion of an open sentence’s being true of an object. For the analogy to be perspicuous, the assumption is needed that a non modal sentence always contains a hidden variable Z for possible worlds. An open sentence with only a free variable x stands in the relation true of to an object o if and only if, in the case o is assigned to x, the proposition expressed by the resulting closed sentence is true; in the same way, an open non modal formula with only a free variable Z for possibile worlds is true of a world w if and only if, in case w is assigned to Z, the proposition expressed by the resulting closed sentence is true. As the open sentence “x is a capital city” is true of London and false of Oxford, so the proposition that London is a capital city is true of the actual world and false of a possible world where London is not a capital city. The problem with this view is the following: consider the open formula “x is a capital city in Z”; at the end of the process of interpretation, once appropriate values are assigned to all the free variables, the proposition expressed by the resulting closed sentence comes out as a necessary one, while we would intuitively count what the closed sentence expresses as contingently true. In case London is assigned to x and the actual world to Z, the proposition that London is a capital city in the actual world is necessarily true: at least in a system like S 5, if it is true that London is a capital city in the actual world, it is true in every possible world that London is a capital city in the actual world; intuitively, however, we would consider the fact that London is a capital city as paradigmatically contingent. As Williamson writes “there is genuine contingency in how things are only if, once values have been assigned to all variables, the resulting proposition could still have differed in truth-value”.16 The charge on the notion of a proposition’s being true of a world is therefore that, in the only plausible way in which it could be understood (i.e. in analogy with the notion of an open sentence’s being true of an object), heavily misrepresents our ordinary modal notions by denying the possibility of genuine contingencies.17 15. Cf. Williamson (2002, 238–239). 16. Cf. Williamson (2002, 239). 17. It could seem ironic that a defender of the view that everything exists necessarily accuses someone else’s position of denying the possibility of genuine contingencies; a defender of necessary existence, however, need not be some sort of determinist, defending the view that nothing could have been otherwise: for the former, everything could have been otherwise, except the number of existing individuals. For a defender of necessary existence the domain of individuals is necessarily the same; for a determinist what are necessarily the same are the domain of individuals and the extensions of the predicates.

221

The problem I have with these two criticisms is that I do not see how the bulk of them could not be directed also against the notion of a proposition’s being true in a world. Consider the first criticism. Even in the case Principle 1 were defined in terms of the notion of a proposition’s being true in a possible world and a possible world were, as before, defined as a maximal consistent set of propositions, the same circularity would still have emerged; validity would still have been understood in terms of Principle 1; Principle 1 in terms of a proposition’s being true in a world; and a possible world, as before, in terms of validity. The circularity affects the notion of truth of no more than the notion of truth in and it seems to me rather an effect of taking Principle 1 as the ground of the notion of valid argument, of treating modal operators as quantifiers over possible worlds and of defining a possible world as a maximal and consistent set of propositions, not something specifically having to do with a way in which a proposition is true with respect to a possible world. Consider the second criticism. Even in this case, nothing specific against the notion of a proposition’s being true of a world (as opposed to the notion of a proposition’s being true in a world) seems to me to be going on. A theorist with only the standard notion of a proposition’s being true in a world could, in fact, be blamed for the same reason. If she believes that the proposition expressed by a non modal sentence like “London is a capital city” really is the proposition that London is a capital city in the actual world, then such a proposition is necessarily true. In general, propositions of the form P is true in w are non contingent, either necessarily true or necessarily false, like those of the form P is true of w. Take a “neutral” notion between a proposition’s being true in and of a world as the notion of a proposition’s being true with respect to a world. Even in this case the same criticism applies: if one believes that what P expresses is really that P is true with respect to w (where w is a world where P is true), then what is expressed is necessarily true. Williamson’s second criticism, thus, seems to me to be directed against what the notions of a proposition’s being true in a world and of a world have in common, rather than against what differentiates the latter from the former. What the two notions have in common is that they both signals those treatments of modal truth as truth relativized to a possible world (no matter how the relativization is done), they both are distinctive marks of a quantificational treatment of modal operators. Williamson’s second criticism, then, could be taken, more generally, as a criticism of all those treatments of modality that try to

222

reduce it to an extensional phenomenon or, less ambitiously, that assigns to modal sentences extensional truth-conditions that, unlike the analyzed sentence, are necessarily true. One of the assumptions that Williamson attributes to the defender of the notion of a proposition’s being true of a world is that modal sentences contain a hidden variable for possible worlds and that what is really expressed by a non modal sentence P is really P in w; in effect, if the thesis is that modal sentences are literally quantifications over possible worlds, then the corresponding non modal sentences are to be taken as open formulas with free variables for possible worlds. But a defender of the notion of a proposition’s being true in/of a world does need to assume such a thesis. I think, in fact, that there is a difference between giving the truthconditions of a modal formula using a quantificational apparatus (as it is done in standard modal semantics, to which the notions of a proposition’s being true of/in a world belong) and claiming that modal sentences are, literally, quantifications over possible worlds (as D. Lewis, for example, does18). Only in the latter case, but, I think, not in the former, is the possible world assigned to a sentence really part of what is expressed by the sentence, part of the proposition expressed; only in the latter case, but not in the former, is a non modal sentence is not completely interpreted until a possible world is assigned to it. In modal semantics what happens is that sentences (or propositions, if one buys them), already interpreted, are evaluated with respect to possible worlds; possible worlds do not enter as constituents into the proposition expressed. In Lewis’s approach, on the contrary, what happens is that ordinary modal sentences are first translated into sentences of an enriched first-order, extensional language that contain an overt quantification over possible worlds and then they are evaluated by means of standard first-order, non modal semantics. The defender of the notion of a proposition’s being true of a world would probably not accept the thesis that all non modal sentences are left uninterpreted until a possible world is assigned to them and therefore he would probably not accept the claim that non modal sentences contain hidden variables for worlds.19 To sum things up, the situation seems to me to be the following: on the one hand the fan of the notion of a proposition’s being true of a world 18. Cf. Lewis (1968) 19. After having directed his criticism against the notion of a proposition’s being true of a world, Williamson applies his second criticism mainly against Lewis’s version of modal realism; cf. Williamson (2002, 239).

223

would probably deny, given the considerations presented above, that the plausible way to understand his new notion is in strict analogy with the notion of an open formula’s being true of an object; on the other hand, if she finds the analogy useful to understand her new notion, it is not clear to me how she could avoid to understand the notion of a proposition’s being true in a world in the same way and then it is not clear to me how Williamson’s criticism to the first notion could not be applied to the cognate one. It could be argued, however, that the notion of a proposition’s being true of a world has a feature that the other notion does not have and it is just this feature that allows Williamson to criticize the former notion without criticizing the latter. When we talk of a proposition’s being true of a world we need to mention two possible worlds: the world in which a proposition p is “generated”, i.e., a world wi where all the constituents of p exist; and the world in which the proposition p is “evaluated”, i.e., a world wii of which p is not a constituent, “within” which p cannot be deduced but with respect to which p is indeed true.20 Typically, a false proposition in a world of generation wi is said to be true of a world of evaluation wii. This is not the case when we talk of something’s being true in a world, where only one world is relevant; from the point of view of a world w, a proposition true in it, is true simpliciter. As a distinctive sign of this disanalogy, it could be mentioned that the notion of truth in a world seems to be characterizable in terms of an absolute, not relativized, notion of truth, the notion of obtaining of a world and primitive modal operators: ) is true in a world w if and only if, had w obtained, ) would have been true.21 This should mark a strong disanalogy between the two notions because presumably the notion of a proposition’s being true of a world cannot be defined using the same conceptual resources.22 20. The generation/evaluation dichotomy is taken from Almog (1986, 220). 21. This way of characterizing the notion of truth in a world has been introduced by Plantinga (1974, 46); see also van Inwagen (2001, 214). 22. This might be true in general. I want here to notice, however, that a similar definition seems to me to be possible even for the notion of a proposition’s being true of a world at least for the paradigmatic case of the proposition expressed by ™x(x = a): ™x(x = a) is true of w if and only if had w obtained, a would not have existed and x(x = a) is true.

224

To this kind of treatments, however, I would react by saying that it is not entirely clear to me that to understand the characterizations above only the notion of absolute truth needs to be used and not also some sort of truth relativized to a world. After all, what we end up with are counterfactual conditionals and such conditionals are standardly analyzed in terms of a notion of truth relativized to possible worlds. In the case of the notion of truth in a world, a sketch of analysis in terms of the semantics given by Lewis in (1973) would be something like: ‘Had w obtained, then ) would have been true’ is true œ ‘)is true’ is true in all the worlds in which ‘w obtains’ is true that are most similar to the actual world.23 One could take counterfactuals as primitive, use counterfactuals to define modal operators and avoid any use of possible worlds semantics,24 i.e., refuse to characterize their semantics in term of truth relativized to a world. But the definition of truth in a world above already commits us to possible worlds that obtain: Would a position not be quite uneconomical if it is committed to possible worlds that obtain but refuses nonetheless to define the semantics of counterfactuals in terms of truth relativized to them? These are just skirmishes, however. I am ready to assume, for the sake of the argument, that the notion of truth in a world could be defined in a way that the notion of truth of a world could not. I could then try, as Williamson suggests, to understand the latter notion in analogy with the notion of an open formula’s being true of an object. The problem I have is that the purported analogy seems to me not to be perspicuous just at the point needed, especially if the notion under consideration is taken to be part of a possible-worlds semantics for modal logic. The assignment of an object to an open formula generates a closed sentence that expresses a proposition, but the “assignment” of a possible world to a proposition does not generate another proposition (or an enriched version of it).25 From this, it follows that “assigning” a possible world to a conThis, of course, does not mean that such a definition is extensible to cover all and only the cases of a proposition’s being true of a world. 23. An actualist trying to define the notion of truth in by way of counterfactuals, may endorse Lewis’s semantics but she does not need to endorse Lewis’s conception of possible worlds. 24. See Williamson (2005, 15–19). 25. If it is assumed that the entities whose truth value is relativized to worlds are interpreted

225

tingently true proposition does not generate another proposition that has the feature of necessarily having the truth value it has. The mechanism by which possible worlds are “assigned” to propositions is not a mechanism that has any effect on the content of the proposition expressed, therefore (given that it is generally assumed that modality, at least in the metaphysical case, acts on the content of a sentence), such mechanism could not change the modal profile of the proposition received as argument. Propositions are first generated and then evaluated with respect to worlds (rather than assigned to them). It is true that the evaluation process in the case of the notion of truth of a world is more complicated, but the evaluation process, whatever its nature, does not modify the content of the evaluated propositions. The same goes for sentences: well-formed sentences of a language are first interpreted and then evaluated with respect to a possible world, but the evaluation process does not change in any way the propositions they express. What is contingent and is evaluated with respect to a world (either in the plain “in”-way or in the more convoluted “of ”-way), remains contingent. 3.2 Propositions as equivalence classes of mental representations According to Williamson, propositions are characterized by the following features:26 (i) being the bearers of the truth values (ii) being expressed by that clauses (iii) possibly being the premises or the conclusions of the arguments; furthermore he says that Principle 3 is plausible if propositions are (iv) structured entities. Applying the considerations made in section 2 in the case of sentences, I now wish to sketch a conception of propositions, compatible with (i)–(iv) and plausible, I believe, in its own right, for which either Principle 1 or Principle 3 comes out false.27 The idea is to define propositions as equivalence classes of mental representations. Mental representations are treated, as it is usual nowadays, as sentence-like, information-bearing, mental entities.28 sentences and not propositions, the same point applies: in standard possible worlds semantics, the assignment of a sentence (i.e., a closed sentence or an open sentence under an assignment of values to variables) to a possible world does not change the proposition that the sentence expresses because the sentence is not taken to be an incompletely interpreted one. 26. Cf. Williamson (2002, 235). 27. The fact that the conception I am going to present is compatible, I think, with what Williamson says about propositions in his article does not imply that he would find it plausible. 28. For the idea of propositions as equivalence classes see Perry (2001); on the notion of

226

Such a conception of propositions is plausible within theories of content that are representational and causal. A representational theory of content is, roughly, a theory according to which our mental representations are syntactic items whose constituents are symbols, belonging to an alleged “language of thought”. A causal theory of content is a theory that holds that at least the primitive symbols of an agent’s language of thought are triggered by causal connections that the agent has (or has had) with the external world: we can think about an object o only if we stand in a certain relation to a mental representation that is constituted by a symbol whose semantic value is o and that has been triggered by some sort of causal relation of the agent with o; standing (or having stood) in a causal connection with an object o is a necessary and sufficient condition to generate in the head of a cognitive agent a symbol whose semantic value is o. Given a mental representation [ p], the proposition [P] is the equivalence class of all mental representations that stand to [ p] in the following relation having the same syntactic structure and being formed by components with the same semantic value. A proposition [P] exists in a world w if and only if, in w, there is at least a cognitive agent a such that a has [ p]. A cognitive agent a has a mental representation [p] only if a stands in some sort of causal relation to the semantic values of the (primitive) symbols constituting [p]. For example, a cognitive agent a has the mental representation [Socrates is a philosopher] only if a stands in a causal relation to Socrates. A proposition [P] is true in a world if and only if [P] exists and P. While, I think, it is quite clear why the conception of propositions I am presenting is compatible with (i)–(iii), it might be argued, however, that this view is in contrast with (iv) because classes are paradigmatically unstructured entities. The plausibility of a principle like 3 would then be in danger. But the structured entity by which a singular proposition is usually represented need not to be taken as the proposition itself; something like: (3.2) <Socrates, being a philosopher> may be used to represent the singular proposition expressed by the sentence “Socrates is a philosopher” because it represents what all the relevant mental representations corresponding to the tokenings of the sentence have in common; the proposition expressed by “Socrates is a philosomental representation see Fodor (1985).

227

pher” is the class of all mental representations that have the structure and the components of (3.2). Principle 3 retains its plausibility even within this conception of propositions. Of course, some small adjustments are in order: “being a constituent of a proposition P” should be now interpreted as “being an object that triggers a symbol in a mental representation”; Principle 3 would then say something along the following lines: Necessarily, if [P] exists and o is an object that triggers a symbol in any [p] [P], then o exists. The theoretical role of Principle 3, then, seems to be preserved. Its role is that of laying down a necessary condition for something to be thinkable and this is now done by signalling the ontological dependence of mental representations (and derivately of propositions) on the objects triggering the symbols that constitute them. Under this conception of propositions, Principle 1 comes out as false: in a world w where I do not exist, the proposition [VITTORIO DOES NOT EXIST] is not true because there is no such proposition: no cognitive agent stands in w in any causal relation to me, hence no cognitive agent has in w the mental representation [Vittorio does not exist]. Nothing represents in w that I do not exist in w. It is only in those worlds where I do exist and where at least a cognitive agent a stands in some causal relation to me, that a might have the mental representation [Vittorio does not exist] and the proposition [VITTORIO DOES NOT EXIST] exist; this proposition, however, would be false in such worlds. But from “Necessarily, if Vittorio exists, then the proposition that Vittorio does not exist is false” and instances of principles 2 and 3, Williamson’s desired conclusion would not follow. The somewhat restricted notion of propositional existence used above has a serious drawback however. Assuming it, we would be forced to hold, quite implausibly, that in those worlds where there are no cognitive agents, there would not be any proposition and therefore anything that is either true or false in those worlds. No proposition would exist, for example, in a possible world w made only of rocks and nothing would be either true or false in w. Yet, there seems to be lots of truths (and falsities) that we can say about it. What is true or false in a world (i.e., what propositions exist in a world), an objector might say, is independent on what is represented in such world.

228

The distinction between the notion of a proposition’s being true in or at a world, however, may be used to dispel a little the implausibility: no proposition, it might be replied, is true in w but lots of propositions are true at (or of ) w. It is we, from the vantage point of our world, that generate, through our mental representations, those propositions that are true or false at that world. It is our, actually existing propositions, that can be used to give a complete description of w. For the unconvinced, however, the only response is probably that of relaxing the problematic condition under which a proposition exists in a world: a proposition [P] exists in a world w if and only if there is a world wi and a cognitive agent a such that a has [ p] in wi. A proposition exists if one of the mental representations that belongs to it possibly exists or, in other words, the proposition [P] exists if and only if it is possible for a cognitive agent to represent that P.29 This conception seems to be more compatible than the other with a traditional view of propositions according to which the realm of propositions is the realm of what can be possibly thought. This relaxed version of propositional existence seems in effect to be able to save Principle 1 from the counter-example above. The fact that a cognitive agent a has the mental representation [Vittorio does not exist] (in a world where I do exist) now warrants the existence of the proposition [VITTORIO DOES NOT EXIST] even in those worlds where I do not exist. Hence, it would in effect be true that in all those worlds where I do not exist the proposition [VITTORIO DOES NOT EXIST] is true. The conception of propositions as equivalence classes and Principle 1 are compatible only with a relaxed notion of propositional existence. The problem, however, is that a conflict with principle 3 now seems to emerge: according to the relaxed notion of propositional existence in fact, the proposition [VITTORIO DOES NOT EXIST] exists in w even in the case I do not exist in w (and this is what is warranting the truth of Principle 1) but according to Principle 3 this proposition can exist only in the case I exist. Under the assumption that propositions are equivalence classes of mental representations, the situation is thus the following: on the one hand, if we assume a “worldly”, restricted notion of propositional existence, Principle 1 comes out false, but, on the other, if we assume a relaxed 29. This also means that the proposition [P] exists if and only if there could have been a cognitive agent that has [ p], even in the extreme case where no actual cognitive agent possibly has [ p].

229

notion of propositional existence, it is Principle 3 that may come out false. Within this view of propositions, then, there is not a single notion of propositional existence that makes all three principles come out true together (unless one already assumes the truth of necessary existence) and therefore Williamson’s argument is unsound.30

REFERENCES Adams, R. M. 1981. “Actualism and thisness”. Synthese, 49, 3–41. Almog, J. 1986. “Naming without necessity”. Journal of Philosophy, 83 (4), 210– 242. Fine, K. 1985. “Plantinga on the reduction of possibilist discourse”. In: J. E. Tomberlin and P. van Inwagen, eds. Alvin Plantinga. Dordrecht: Reidel, 145– 186. — 1994. “Essence and modality”. Philosophical Perspectives, 8, 1–16. Fodor, J. 1985. “Fodor’s guide to mental representation”. Mind, 94 (373), 76– 100 Kaplan, D. 1989. “Demonstratives”. In: J. Almog, J. Perry, and H. Wettstein, eds. Themes from Kaplan. Oxford: Oxford University Press, 481–563. Kripke, S.A. 1963. “Semantical considerations on modal logic”. Acta Philosophica Fennica, 16, 83–94. Lewis, D. K. 1968. “Counterpart theory and quantified modal logic”. Journal of Philosophy, LXV(5), 113–126, 1968. Reprinted in (Loux 1979, 110–128). — 1973. Counterfactuals. Oxford: Blackwell. Loux, M. J. ed. 1979. The Possible and the Actual. Ithaca: Cornell University Press. Menzel, C. 1991. “The true modal logic”. Journal of Philosophical Logic, 20, 331–374. Perry, J. 2001. Reference and reflexivity, Stanford: CSLI. Plantinga, A. 1974. The Nature of Necessity. Oxford: Clarendon Press. — 1983. “On existentialism”. Philosophical Studies, 44(1), 1–20. van Inwagen, P. 2001. “Two concepts of possible worlds”. In: Ontology, Identity and Modality, Cambridge, Cambridge University Press, 206–242. 30. Many thanks to Joseph Almog, Andrea Bianchi, Massimiliano Carrara, Pierdaniele Giaretta, Ernesto Napoli, Paolo Leonardi, Elisabetta Sacchi, Timothy Williamson and all the participants of a seminar at the University of Bologna where an earlier draft of this paper was presented.

230

Williamson, T. 2002. “Necessary existents”. In: A. O’Hear, ed. Logic, Thought and Language, Cambridge: Cambridge University Press, 233–251. — 2005. “Armchair philosophy, metaphysical necessity and counterfactual thinking”. Proceedings of the Aristotelian society, 105(1), 1–23.

231

This page intentionally left blank

Grazer Philosophische Studien 72 (2006), 233–252.

NEGATION1 Ernesto NAPOLI University of Urbino Summary The paper is concerned with negation in artificial and natural languages. “Negation” is an ambiguous word. It can mean three different things: An operation (negating), an operator (a sign of negation), the result of an operation. The three things, however, are intimately linked. An operation such as negation, is realized through an operator of negation, i.e. consists in adding a symbol of negation to an entity to obtain an entity of the same type; and which operation it is depends on what it applies to and on what results from its application. I argue that negation is not an operation on linguistic acts but rather an operation on the objects of linguistic acts, namely sentences. And I assume that the negation of a sentence is a sentence that contradicts it. If so, the negation of a sentence may be obtained, in case the sentence is molecular, by applying the operation of negation not to the sentence itself but to a constituent sentence. To put it in a succinct and paradoxically sounding way we could say that in order to negate a sentence it is sufficient but not necessary to negate it. However that negation applies to sentences is true only for artificial languages, in which the sign of negation is a monadic sentential connective. In natural language, negation applies to expressions other than sentences, namely words and non-sentential phrases. Still words and not sentential phrases are interesting and valuable only as ultimate or immediate constituents of sentences, as a means of saying (something that can be true or false) and the concern with negation is ultimately the concern with the negation of sentences. So the problem is what sub-sentential and non sentential expressions negation should apply to in order 1. One could wonder why a paper on negation should appear in a volume on propositions. What connection is there between the two topics? Well, affirmation and denial (affirmation of the negation) have been considered, not infrequently, as propositional attitudes. If so, it might be a contribution to the debate on propositions to argue that negation applies to expressions and even when sentential does not apply to propositions. No doubt, natural language sentences are meaningful, still their meaning is quite irrelevant for negation. The negation of a sentence is a sentence that contradicts it. Two sentences are contradictory iff their conjunction is a logical falsity, or their disjunction is a logical truth. However, logical truth and logical falsity, unlike truth and falsity, are a matter of form, i.e. are independent of meaning.

to obtain the negation of the containing sentence. The standard answer is that the negation of a natural language sentence is equivalent to the negation of its predicate. Yet, I argue, predicate negation is necessary but not sufficient, due to the existence of molecular sentences. Finally I notice that if to apply negation to an artificial sentence is to put the negation sign in front of it, to negate the predicate of a natural language sentence may or may not be to put the negation sign in front of it.

0. The ambiguity of “negation” “Negation” is an ambiguous word. It can mean three different things: An operation (negating), an operator (a sign of negation), the result of an operation. The three things, however, are intimately linked. An operation such as negation is realized through an operator, and which operation it is depends (on what it applies to and) on what results from its application. So I should need no excuse if I focus on negation as an operation. 1. Negation is not an operation on linguistic acts, nor a kind of linguistic act One could think that negation, as an operation, is a sort of linguistic act alongside assertion, in fact its opposite. Yet negation is negation of and what negation is negation of could not be an assertion. If negation is an operation on something, the something in question could not be an assertion. Hence denial could not be a kind of linguistic act, which is the result of the application of the operation of negation to an assertion. To negate (to deny) p is surely not equivalent to not to assert p (not to assert is not a linguistic act but a failure to produce a linguistic act). It is, rather, equivalent to asserting not p. This is to say that negation is an operation not on the assertion but on the object of the assertion (what is asserted). The object of an assertion is something capable of being true or false. Notice that negation need not apply to a truth value bearer (see more about this below) and that the negation of x, whatever x may be, could hardly be qualified as a negative x. In fact it is hard to make sense of the notion of a negative x, or for that matter of a positive x, no matter if x is a sentence or a sub-sentential phrase or a word.

234

If one thinks that “unhappy” is a negative expression since it contains a negation sign one should consider that “unhappy” is synonymous to “miserable” where no negation sign appears. Can two expressions designating the same property be the one negative and the other positive?2 If they can, the notion of a negative expression is pretty void, or at least completely severed from any semantic feature of the expression. For sure, if “unhappy” and “miserable” designate one and the same property their respective negativity and positivity have no correspondence in the property designated. Further, of the following sentences: “Christ is immortal”, “Christ lives for ever”, “Christ is not immortal”, “Christ is mortal”, “Christ does not live for ever”, which is to count as positive and which as negative? Again, if we were inclined to consider “Christ is immortal” negative and “Christ lives for ever” positive we could not say that by the first sentence a negative property is attributed to an individual and by the second a positive property is attributed to an individual.3 If negation is not an operation on linguistic acts of assertion, neither are operations involving other connectives, for example disjunction and conjunction. Disjunction is not an operation on linguistic acts. Disjunction is not a kind of linguistic act. As there is no assertion which is the result of the application of the operation of negation to an assertion, so there is no assertion which is the result of the application of the operation of disjunction to two assertions. A disjunctive assertion, if anything, is the assertion of a disjunction of statements (or sentences or propositions) and not a disjunction of assertions. To assert p or q is certainly not to assert p or to assert q. Obviously any rational speaker who had asserted p or had asserted q would be ready to assert p or q. Yet what he would be ready to do is not what he has done. Further any speaker (inclined to classical logic) who had asserted p or q, would not for this be ready either to assert p or to assert q. Conjunction is not an operation on linguistic acts. Conjunction is not a kind of linguistic act. A conjunctive assertion, if anything, is the assertion of a conjunction of statements (or sentences or propositions) and not a conjunction of assertions. To assert p and q is not to assert p and to assert q. For sure if anyone asserted p and q then he asserted p and asserted q. Yet it is not the case that if anyone asserted p and asserted q then he asserted p 2. See Frege (1979, 151). 3. See Frege (1984, 380).

235

and q. Obviously any rational speaker who had asserted p and had asserted q would be ready to assert p and q. Yet what he would be ready to do is not what he has done. 2. Metalinguistic negation? One could claim and some have claimed, contrary to what I have argued in the previous section, that negation can be a metalinguistic operation, that is an operation applying to linguistic acts of assertion. Consider: • • • •

It is not a car, it’s a Volkswagen. (VW commercial)4 John does not love Mary, he adores her. The wine is not good, it is excellent.5 Chris did not manage to solve the problem — it was quite easy for him.6

According to Horn, in these cases we have to do with “a speaker’s use of negation to signal his or her unwillingness to assert, or to accept another’s assertion of, a given proposition in a given way”.7 We could re-phrase the above sentences as: • I would not say that it is a car, I would rather say that it is a VW. • I would not say that John loves Mary, I would rather say that he adores her. • I would not say that the wine is good, I would rather say that it is excellent. • I would not say that Chris managed to solve the problem, I would rather say that the problem was no problem for him, he solved it easily. Yet, it is certainly too strong to talk of “unwillingness to assert”. Even the author of the commercial would have no qualms to say that a VW 4. Horn (2001, 362). 5. Horn (2001, 424). 6. Horn (2001, 368). 7. Horn (2001, 363). Unfortunately Horn adds that “metalinguistic negation focuses, not on the truth or falsity of the proposition, but on the assertability of an utterance”. Yet if it is assertible that the wine is excellent, a fortiori it is assertible that the wine is good, in so far as x is excellent implies x is good. So metalinguistic negation focuses not on the assertibility but on the felicity of an utterance. It should be noticed that if felicity is a property which can be attributed to an utterance, assertibility is not. Utterances are means of assertion and

236

is a car. The added suggestion is that a VW is not just a car, i.e. that it is not a car as any other. In “A VW is not a car as any other” the negation is a perfectly standard objectual negation. And to say that the wine is not good, it is excellent, is another way of saying that the wine is more than good. Similarly to say that John does not love Mary but adores her, is just another way of saying that John loves Mary to a possibly excessive degree. The case of “Chris did not manage to solve the problem — it was quite easy for him” is different only if to manage is not to succeed but to succeed with difficulty. But it does not seem that it could be said of one who solved a problem easily that he did not manage to solve the problem. 3. What does negation, as an operation, apply to? If negation is not an operation applying to a linguistic act (nor a kind of linguistic act resulting from the operation), what does negation as an operation apply to? To answer we have to distinguish the case of artificial languages from that of natural languages. In artificial languages negation applies exclusively to sentences (either closed or open), while in natural languages negation applies to expressions other than sentences, namely words and non-sentential phrases. Still words and non sentential phrases are interesting and valuable only as ultimate or immediate constituents of sentences, as a means of saying (something that can be true or false). The concern with negation is ultimately the concern with the negation of sentences. So the problem is what sub-sentential and non sentential expression negation should apply to in order to obtain the negation of the containing sentence. 3.1 Negation in artificial languages Having said that negation is an operation on sentences (either closed or open) and that it consists in adding the negation sign (or one of the typographical variants of the negation sign: “not”, “¬”, “−”, “a”, etc.) to a sentence, the question is: Where is the negation sign to be added (to a sentence in order to obtain its negation)? means of assertion are neither assertible not unassertible. What is assertible or non assertible is the object of the utterance (what the utterance is about) not the utterance.

237

The question hardly arises unless by “negation of a sentence” we mean not just a syntactic operation on a sentence but a syntactic operation whose result is a sentence that contradicts the sentence to which it applies.8 Unless, in other words, we conceive negation as a truth-value switch. The sign of negation is a unary sentential connective. The recipe for the syntactic negation of a sentence is unavoidably simple: put the negation sign in front of the sentence. However, if we conceive negation as a semantic switch, i.e. as an operation that applied to a sentence results in a sentence that contradicts it, matters are not that straightforward. The negation of a sentence, i.e. a sentence that contradicts the original one, may also be obtained by putting the negation sign not in front of the sentence but somewhere else in the sentence. For example, (A › aA) š aB contradicts (A › aA) š B, and (A š aA) ›aB contradicts (A š aA) ›B (the truth-value of the whole sentence is a sheer function of the truth-value of the constituent sentence B) although the extra negation sign in the first sentence cannot be seen as the result of adding a negation sign in front of the second sentence. We have to conclude that the addition of the negation sign in front of a sentence is sufficient for negating the sentence, but it is not necessary. Hence the operation of negation is not well defined, if being well defined requires the specification of necessary and sufficient conditions. This need not be a very alarming news. After all if to put a negation sign in front of a sentence is sufficient for negating the sentence we know perfectly well what to do for negating a sentence. That negation could be managed by some other operation, for example by insertion of the negation sign in a sentence, does not detract from the fact that by putting the negation sign in front of the sentence we realize the negation of the sentence. (Notice that it is easy to put down sufficient and necessary conditions for being an operation of negation of a sentence. The addition to or the insertion in a sentence of a negation sign constitutes an operation of negation of the sentence if and only if it results in a sentence that contradicts the original one. Considering that logical invalidity is as much undecidable as logical validity — in a rich enough language like full first order logic — the 8. The operation of negation need not be applied to a closed sentence. However we do not have to renounce the notion that the negation sign is a semantic switch when considering occurrences of negation that have open sentences as their scope. We have just to qualify it by saying that any assignment (to one or more free variables) under which an open sentence is true (false) is an assignment under which its negation is false (true).

238

fact of having formulated sufficient and necessary conditions for being an operation of negation may well leave us in the dark as to whether the addition or insertion of negation in a specific sentence realizes the negation of the sentence.) 3.2 Negation in natural languages If to negate a sentence of an artificial language it is sufficient to put a negation sign in front of the sentence, matters are not so straightforward in natural languages such as English. The reason is that it is not possible in general to add a negation sign in front of a sentence. If we put “not” in front of “Mary is happy” what we get is an a-grammatical sentence rather than its negation. There are exceptions. For example “Not everybody is happy” is the perfectly grammatical negation of “Everybody is happy”. We should however refrain from thinking that any sentence beginning with a quantifier (phrase) can be negated by putting “not” in front of it. “Somebody is happy” cannot be negated this way. “*Not some” “*not several”, “*not three”, “*not most”, “*not a number of ”, “*not no”, are not grammatical. Why that it is so I am unable to tell. In any event it is far from granted that in “Not everybody is happy” “not” occurs in front of the sentence, i.e. that it exemplifies “Not (everybody is happy)”. In general “not A” where “A” is a sentence is not grammatical. The grammaticality of “Not everybody is happy” may well depend on the fact that it exemplifies “(Not everybody) is happy”, i.e. on the fact that “not everybody” constitutes a syntactic unit. If so the negation would appear in front of a phrase which appears in front of the sentence rather than in front of the sentence. In fact, negation can appear before a quantifier phrase even when the quantifier phrase is not the first constituent of the sentence and the negation is (equivalent to) sentential negation, for example in: “Marc has passed no(t) more than three exams”. But in such a case it makes no sense to claim that the negation sign occur in front of the sentence. Hence it is hardly motivated to insist that the negation sign occurs in front of the sentence when the quantifier phrase is the first constituent of the sentence to be negated. It must be said, however, that when the negation of the (quantifier) phrase is equivalent to sentential negation, the negation sign and the (quantifier) phrase constitute a syntactic unit which is not a (quantifier) phrase. If the negation of the quantifier phrase is equivalent to sentential

239

negation the quantifier phrase is the predicate. But if the negation of the predicate is equivalent to sentential negation it is only because the negation and the predicate do not constitute a new predicate. Rather they contribute in a distinct way to a non phrasal constituent of the sentence, the intermediate projection Ic, in which the predicate is the complement and negation together with mood, aspect, tense and agreement is the head. (If we accept the split inflection hypothesis, agreement is a head (and node) distinct from tense, aspect, mood and negation). However this Ic constituent, being neither a word nor a phrase is, so to speak, a feeble constituent, lacking the solid features of constituents proper and would fare poorly on constituency tests such as uninterruptability, displaceability and coordinability. The Ic constituent seems to be the theoretical by-product of the principle of binary branching, the principle according to which every node splits in at most two other nodes, or if you prefer, the principle according to which every non terminal node (word) splits in exactly two other nodes. The Ic constituent is no more than an intermediate step in the projection of the inflectional head, whose maximal projection is the sentence. The inflectional head is indeed a separate constituent of the sentence, but if we have to abide by binary branching, it cannot sit on an equal footing with the subject and predicate, a branch of a ternary branching. So it has to attach itself temporarily to the predicate and become the head of an intermediate inflectional constituent. Yet there is plenty of evidence that inflection, though applying to the predicate, is no part of the predicate. It is just an accident that in English, inflection can be affixal i.e. realized by non words that cannot stand alone (“-s”) and hence have to attach to a word, the verb heading the verb phrase (“think-s”), or that tense inflection can be incorporated in the predicate via morphological variation of the V0-head or verbal head (“thought”). There are languages such as the Mauritian Creole where head verbs are never marked for tense. 9 And even in English tense inflection applies (also, as in “Marc was captured”, or exclusively, as in “The project was fine”) to functional words independent (which are not part) of the predicate, such as the auxiliaries and the copula. That auxiliaries and the copula are no part of the predicate is shown by a number of considerations. • The predicate can be fronted to the exclusion of the auxiliary as in “Telephone Mary, I should indeed”. 9. See Adger (2003, 165).

240

• The auxiliary can be separated from the predicate by an intervening adverb as in “Marc has often thought of retiring”. • The predicate can be separated from the auxiliary in so called pseudoclefted constructions as in “What the president will do is call for greater investment”. • Two predicates can be coordinated provided the auxiliary is dropped as in “Marc will go to Paris and attend the conference”. In other words inflection should be recognized as a non-phrasal constituent of the sentence alongside subject and predicate. If it has been thought that sentences are of the form subject/predicate it is only because inflection either has been erroneously considered as part of the predicate or has not been taken into account. That a sentence is not just the composition of subject and predicate is testified by the fact that the predicate has to agree (in gender, number and person) with the subject. Agreement expresses a relation between subject and predicate and so it is not a property of a phrase but a relation between occurrences of two phrases in virtue of which the two phrases come to be constituents of a sentence. (“John drink” is not a sentence since the predicate does not agree with the subject). So, to say that a sentence is of the form subject/predicate is a bit too rough. Still inflection (or inflection and agreement) are absent from infinitival sentences. The claim that sentences are of the form subject/predicate, though rough, is after all defensible in that subject and predicate are the only inevitable constituents of the sentence besides words. As we’ll see Frege argued that in a sentence such as “Every man dies” “every man” is logically the predicate since it is what the negation should apply to to get the negation of the sentence. Still “Every man” is syntactically the subject since it is what the other phrase must agree with, “All men die”. We face here a divergence between logical subject and grammatical subject. This, in a sense, is as it should be. It is just another instance of the discrepancy between syntactic and semantic notions. As testified by passivization, the syntactic role of subject does not overlap with the thematic role of agent. If in “Mary kissed Carl” the subject is the agent in “Carl was kissed by Mary” the subject is the, may be impatient, patient. After this long detour we may come back to our temporarily suspended concern, namely what it takes to negate a natural language sentence. If even “not everybody is happy” could not be seen as the result of the application of negation to a sentence, it would seem that the recipe for negating

241

a natural language sentence in no case could be: add the negation sign to the sentence. To this it could be countered that if the language is null-subject, i.e. the subject can be omitted, the negation of a sentence can be obtained by putting the negation sign in front of it. For example the negation of “sono stanco” is “non sono stanco”. Right. But the observation has no great weight. For one thing it is limited to null-subject languages, for another these languages are improperly qualified as null-subject. It is not the subject that is null but its realization (expression). As soon as the subject is voiced the negation sign ceases to appear (to be) in front of the sentence. The negation of “Io sono stanco” is “Io non sono stanco”. An apparently more promising objection to the claim that in no case negating a sentence could consist in putting the negation sign in front of the sentence is that a sentence can be negated by putting in front of it “it is not the case (true)”. In fact this recipe is of general applicability, if not of general application. However, “It is not the case (true)” does not exemplify an operator of negation which applies to a sentence but (as I argue in fn. 13) the negation of a predicate whose argument is provided by the nominalization of a sentence. If so, we can safely assume that the negation of a sentence does not consist in putting the negation operator in front of the sentence. The question becomes where the negation should be put. Obviously, if not in front of the sentence somewhere in the sentence. A sentence has two kinds of constituents: words (ultimate constituents) and phrases (immediate constituents). Hence there are two different kinds of position where an objectual negation operator can be put: in front of words or in front of phrases. 10 Phrases can be either sentences or subsentential phrases. And phrases that are sentences may have as immediate constituents phrases that are sentences. But if to negate a sentence is not to put the negation sign in front of a sentence we can exclude that the negation of a sentence consists in putting the negation sign in front of a phrase that is a sentence, i.e. we have to assume that the negation of a sentence consists in putting the negation sign in front of a word or of a sub-sentential phrase. Words are not good candidates. For sure to negate a word is not sufficient for negating a sentence, not even when the sentence contains just a 10. So we could agree with Englebretsen’s claim that “All object language negation is internal. Either terms are negated (logical contrariety) [a is nonP] or predicates are negated (predicate denial) [a is not P]. Negation never applies to a sentence as a whole. Object language negation is never external” (Englebretsen 1981, 59).

242

single word that can be negated, as is the case with “Marc is happy”, considering that the individual name “Marc”, as any other individual name, cannot be negated. And neither can “is”, being an inflectional element. “Is” and “not” go together; they are both part of the inflection head. But an inflectional element does not apply to another. They both apply to the predicate and are ultimately features of the sentence. No doubt “Mary is not Marc” is a perfectly grammatical sentence. But the negation does not apply to the individual name. The sentence is elliptical for “Mary is not identical to Marc”. Clearly “Marc is unhappy” is not the negation of “Marc is happy”. The disjunction of the two sentences may well fail to be true. Carl may be neither happy nor unhappy. Carl may feel, as most of us most of the time, in an unexciting middle condition which does not deserve any positive or positively negative qualification. If to negate a word is not sufficient for negating a sentence, granted that the negation sign does not apply to sentences, the only possibility is that the negation of the sentence is realized by applying negation to one of its immediate constituents, i.e. phrases, that are not sentences. It should be stressed that a word is not a phrase even when the phrase is one word. Any word in use, i.e. as it occurs in a sentence, is part of a phrase. The phrase, even when one word, is a structure having an empty position for a modifier. In case it is a DP, even the head position D0 may not be occupied by a determiner. There are two possibilities: either the DP is a naked partitive description (such as “water” in “water was falling down from the ceiling” or “students” in “students were standing at the entrance of the school”) or it is a proper name. In the first case, the D0 position is empty; in the second case it is instead occupied by a name N0 which is raised to D0. As it happens not only the individual name “Marc” but also the determiner phrase (proper name) containing the individual name “Marc” (whether “Marc” is the only lexical constituent as in “Marc is tired ” or not as in “old Marc is tired”) cannot be negated. Hence we can exclude individual proper names from the domain of phrases the negation of which could realize sentential negation. Also nominalization of sentences and indexicals cannot be negated and hence are not possible candidates for the role of medium of sentential negation, i.e. of the predicate. All other sub-sentential phrases, non individual proper names included, can be negated and hence can act as predicates in some sentences. For example “the property of being a horse” is a non individual proper name that acts as predicate in the sentence “Varenne has the property of being a horse”.

243

Now, considering that a sentence may well contain more than one phrase that can be negated (for example, “Every man is mortal”) or more than one occurrence of a phrase that can be negated (for example, “No one loves no one”) the question is: which phrase or which occurrence of one and the same phrase is to act as predicate? As already anticipated, Frege’s answer is that the predicate is the sub-sentential phrase the negation of which amounts to the negation of the containing sentence. In “Marc is mortal” “Marc” is the subject and “mortal” the predicate. It is an easy step to take to hold that in “Every man is mortal” “Every man” is the subject and “mortal” the predicate. Frege resorted to negation to argue that in spite of appearances in a sentence such as “Every man is mortal” the predicate is not “mortal” but it is rather “every man”. He pointed out that one gets the contradictory of “Every man is mortal” by putting the negation in front of the quantifier phrase “every man” rather than in front of the adjectival phrase “mortal”. We should beware, though, of saying that natural language quantifier phrases are second level predicates, as Frege apparently did.11 To be a quantifier phrase is to be a certain type of phrase but to be a predicate is not to be a certain kind of phrase but to be an occurrence of a certain type of phrase having a certain role in predication. Being a predicate is not a property of quantifier phrases but at most12 a property of some of their occurrences. Some, not all. In fact a sentence may well contain more than one quantifier phrase and if the sentence is of the form subject/predicate only one of the occurring quantifier phrases will be the predicate. For example in “Everybody loves everybody” the predicate, i.e. the phrase to be negated in order to negate the sentence is “everybody” in its first, rather than its second, occurrence. 11. Frege held that the existential quantifier means existence and existence (of an individual) is a second level property, namely the property of a property of being instantiated. Likewise the universal quantifier means the property of a property of being universally instantiated. Now, if in “All men are mortal” the quantifier phrase is a second level predicate then the adjectival phrase “mortal” is an apparent predicate, i.e. is the subject providing the argument. Otherwise “All men are mortal” would consist of two predicates and no subject. Frege, however, held that concepts (unsaturated entities) can be meant only by unsaturated expressions and he would indiscriminately qualify unsaturated expressions as predicates. The only reason I could think of for his holding that unsaturated expressions are predicates is that he would think of “All men are mortal” as of the form “x (Mx)” where “M” is the predicate in the open sentence “Mx”. 12. I should explain the “at most” qualification. The reason is that one could hold that quantificational sentences are not of the form subject/predicate and hence that to negate them is not to negate their predicate. If “Every man is mortal” is rendered as “x(Fx oGx)” its negation is “¬[x (Fx oGx)]”, the negation has in its scope not the quantifier but the whole sentence.

244

The negation of a quantifier phrase is equivalent to sentential negation only provided the quantifier phrase appears in apparent subject position or in apparent object position in a sentence in which in apparent subject position there is a determiner phrase rather than a quantifier phrase. 4. Is negating a natural language sentence equivalent to negating the predicate? We have seen that in the case of artificial languages putting the negation sign in front of a sentence is sufficient but not necessary for negating the sentence. In particular when the sentence is molecular, i.e. is a Boolean combination of sentences, the molecular sentence may well be negated by putting the negation sign in front of a constituent sentence. In the case of a natural language, negating the predicate is necessary but not sufficient for negating the sentence. In fact, when the sentence is molecular, to negate it more is required than the negation of the predicates of the constituent sentences. For example the negation of “Marc has gone to the cinema and Mary has remained at home” is not “Marc has not gone to the cinema and Mary has not remained at home” but it is rather “Marc has not gone to the cinema or Mary has not remained at home”. This is to say that the negation of the predicate of the constituent sentences must be accompanied by the substitution of “and” with “or”.13 13 One could object to the necessity of the negation of the predicate for negating the sentence by pointing out that a (molecular) sentence may be negated by prefixing to it “It is not the case that” or “It is not true that”. The objection can be countered by observing that by so doing a new sentence is obtained in which “the case” or “true” is the (main) predicate. That it is so can be appreciated by considering that in “It is not the case (true) that Marc has gone to the cinema and Mary has remained at home” “that” does not belong with “the case (true)” but with the following sentence. In other words the proper analysis of “It is not the case (true) that Marc has gone to the cinema and Mary has remained at home” is not “[It is not the case (true) that] [Marc has gone to the cinema and Mary has remained at home]” but rather “[It is not the case (true)] [that Marc has gone to the cinema and Mary has remained at home]”. This sentence can be rewritten as “[That Marc has gone to the cinema and Mary has remained at home] [is not the case (true)]” where is even clearer that “the case (true)” is the predicate. It could be objected that, if so, we should renounce the claim that negating the predicate is not sufficient for negating the sentence. The conclusion is a bit precipitous. For sure, “That Marc has gone to the cinema and Mary has remained at home is not the case (true)” is equivalent to the negation of “Marc has gone to the cinema and Mary has remained at home”. Yet equivalence is not identity. The first sentence is the negation of “That Marc has gone to the cinema and Mary has remained at home is the case (true)” which is equivalent and yet not identical to “Marc has gone to the cinema and Mary has remained at home”.

245

One could think that some sentences may be negated by negating the subject and hence that the negation of the predicate is not necessary for negating the sentence. Take “To be happy is common”. If we negate “To be happy” we obtain “Not to be happy is common”. “To be happy is common” and “Not to be happy is common” are such that the first is true if and only if the second is false. The negation of a phrase (unlike the negation of a word) gives in any case the complement. But a property and its complement cannot be both common, if to be common is for a property to be enjoyed by most than half of the relevant individuals (those, for example, that can either be or fail to be happy). Still it would be precipitous to conclude that “Not to be happy is common” is the negation of “To be happy is common”. A sentence is the negation of another if and only if their conjunction is contradictory, i.e. is a logical falsity. But logical falsity, as much as logical truth, is a matter of form. A sentence is contradictory, i.e. is a logical falsity, if and only if it is false under every reinterpretation. But the sentence which is the conjunction of “To be happy is common” and “Not to be happy is common” is not false under every reinterpretation. In fact if we substitute “is common” with “is a sin” in “To be happy is common” we obtain “To be happy is a sin”. But “To be happy is a sin” and “Not to be happy is a sin” are not such that the first is true iff the second is false. In fact “To be happy is a sin” and “Not to be happy is a sin” can be both true, if everything is a sin, or both false, if nothing is a sin. One could think that some sentences may be negated by negating an occurring word and hence that negating the predicate is not necessary for negating the sentence. Take “What Marc has said at the 2004 Annual Philosophy Meeting is consistent”. If we negate “consistent” we obtain “What Marc has said at the 2004 Annual Philosophy Meeting is inconsistent”. A predicate is a phrase even when one word. But affixal negation is negation of a word and not of a phrase, hence it is not negation of the predicate.14 “What Marc has said at the 2004 Annual Philosophy Meeting is inconsistent” is true iff “What Marc has said at the 2004 Annual Philosophy Meeting is consistent” is false. Still it would be precipitous to conclude that “What Marc has said at the 2004 Annual Philosophy Meeting is inconsistent” is the negation of “What 14. “Isn’t” or “hasn’t” do not exemplify word negation, i.e. of “is” or “has”. “-n’t” is a contracted form of “not”, which clearly is a word and can stand alone.

246

Marc has said at the 2004 Annual Philosophy Meeting is consistent”. If we substitute “consistent” with “pleasant” in “What Marc has said at the 2004 Annual Philosophy meeting is consistent” we obtain “What Marc has said at the 2004 Annual Philosophy Meeting is pleasant”. But “What Marc has said at the 2004 Annual Philosophy Meeting is pleasant” and “What Marc has said at the 2004 Annual Philosophy Meeting is unpleasant” are not such that the first is true iff the second is false. In fact it may well be the case that what Marc has said is neither pleasant nor unpleasant. We should beware of thinking that the fact that “What Marc has said at the 2004 Annual Philosophy Meeting is consistent” and “What Marc has said at the 2004 Annual Philosophy Meeting is inconsistent” are neither both true nor both false depends exclusively on the complementarity of “consistent” and “inconsistent”. It depends also on the fact that “consistent” and “inconsistent“ are the head of an adjectival phrase which is the predicate of the sentences in which they occur. True” and “untrue” have complementary extensions. Yet “Everything Marc has said at the 2004 Annual Philosophy Meeting is true” and “Everything Marc has said at the 2004 Annual Philosophy Meeting is untrue” may well be both false. “True” and “untrue” are the heads of an adjectival phrase which is not the predicate of the sentences in which they occur. 5. What does the negation of the predicate consist in? Up to now I have said that, in order to negate an artificial language sentence it is sufficient (though not necessary) to put the negation sign in front of it. I have also said that, since a natural language sentence cannot be negated by putting the negation sign in front of it, in order to negate the sentence it is necessary (and sufficient, if the sentence is not molecular) to negate the predicate. The question now is whether to negate the predicate consists in putting the negation sign (immediately) before the predicate. The answer seems to be that it does not. • When in the sentence to be negated the predicate is already negated the negation of the sentence consists in eliminating the negation sign rather than adding it. In a natural language sentence negation cannot be iterated. More precisely, in a natural language sentence the occurrence of two homogenous negations, i.e. applying to the same linguistic item, either a word or a phrase, is ungrammatical. “John is not not happy”

247

is ungrammatical. “John is not unhappy” and “What you said is not inconsistent” are grammatical since the two occurrences of negation apply respectively to a phrase and to a word. • It is no more than an accident that the negation of the predicate is realized by putting the negation sign before the predicate. In English it is. In Italian and other languages it is not. In English the negation sign follows the copula or the auxiliary. In Italian and other languages the negation sign precedes the copula or the auxiliary. But neither the copula nor the auxiliary is part of the phrase that follows. If “intelligente” or “molto intelligente” as they occur in “Maria è (molto) intelligente” are adjectival phrases, this is not the case for “è intelligente” and “ è molto intelligente”. The negation of “Maria è (molto) intelligente” is “Maria non è (molto) intelligente”. So in Italian to negate a sentence is not strictly to put the negation sign in front of a phrase and hence, if the predicate is a phrase (with a certain functional role in the sentence), is not strictly to put the negation sign in front of the predicate. If the negation sign and the predicate do not constitute, as I have argued, a new predicate, it is just an accident that the negation sign precedes immediately the predicate instead of preceding the copula or the auxiliary. (As it is an accident that in English inflectional elements can be incorporated into a verb, but not into an adjective, instead of being always realized by autonomous expressions such as auxiliaries). 6. The ambiguity (of scope) of non affixal negation Affixal negation always applies to the word to which it is attached, but non affixal negation is scope ambiguous, i.e. any of its occurrences may apply either to a word or to the phrase headed by that word. If the negation sign is scope ambiguous, i.e. syntactically ambiguous, a sentence in which a non affixal negation sign occurs will likewise be syntactically ambiguous, and also, as a rule, semantically ambiguous 15 if we are prone to identify the meaning of a sentence with its truth conditions. 15. Apart from special cases such as “What Marc has said at the 2004 Annual Philosophy Meeting is not consistent”. Here it makes no difference if the negation applies to a phrase or to a word. “What Marc has said at the 2004 Annual Philosophy Meeting is not consistent” is true iff “What Marc has said at the 2004 Annual Philosophy Meeting is not consistent (= inconsistent)”.

248

A case in point is provided by Chomsky. Chomsky has claimed that the sentence “John does not like mushrooms” is ambiguous, i.e. has two different readings dependent on whether the negation sign applies to the verb or to the verb phrase. I doubt that the above sentence is ambiguous and would rather think that it only means that it is not the case that John likes mushrooms. After all the negation of the verb is lexicalized with “dislike”, and if one wanted to say that John dislikes mushrooms why not say so? Still, to many it would seem that an occurrence of “John does not like mushrooms” would, more likely than not, be understood as conveying the information that John dislikes mushrooms. The explanation suggested has been negation raising (from the verb to the verb phrase) rather than ambiguity. The problem is: What pragmatic reasons could explain the raising of negation in this case? Surely politeness or understatement are not in question here. Many have thought that the raising of negation has to do with indifference to indifference, or if you prefer, with the tendency to see matters as black or white (rather than black or not black), to see the contrary in the contradictory16. Maybe the indifference to indifference, i.e. negation raising, is not the explanation in this case. Maybe the explanation is indifference to differences which make no practical difference, as the difference between indifference to or dislike of mushrooms. You will not serve mushrooms to a friend even if he does not particularly like mushrooms rather than positively disliking them. And your friend will not serve mushrooms to you even if you say to him that you do not like mushrooms and do not take the pain of being more precise. This said, it remains that, even if Chomsky’s example is questionable, the ambiguity between word and phrase negation can surely appear in cases in which the word is not liable to affixal negation. Take “Marc has not passed more than three exams”. This sentence seems to be ambiguous between: “Marc has not passed (= failed to pass) more than three exams” and “Marc has not passed more than three exams” (= “Marc has passed no more than three exams” = “Marc has passed at most three exams). That the sentence is ambiguous is testified by the possibility of diverging continuations. “Marc has not passed more than three exams. In fact he has failed five exams.” “Marc has not passed more than three exams. To be precise he has passed just two exams.” In the first case it is said that the number of exams Marc has not passed (i.e. has failed to pass) is greater 16. See Horn (2001, 332).

249

than three, in the second that the number of exams Marc has passed is equal to three or less than three. The first interpretation exemplifies verb negation, the second exemplifies verb phrase negation. The two interpretations are quite compatible considering that it may well be that Marc has failed to pass more than three exams and that Marc has passed three or less than three (= at most three) exams. Still the two interpretations are not equivalent, as can be appreciated from the fact that the second but not the first is the negation of, i.e. it is the contradictory of, “Marc has passed more than three exams”. That “Marc has not passed (= failed to pass) more than three exams” is not the contradictory of “Marc has passed more than three exams” is clear from the fact that the disjunction: Marc has failed to pass more than three exams or Marc has passed more than three exams, need not be true. For it may well be true that neither Marc has not passed (= failed) more than three exams nor Marc has passed more than three exams. That “Marc has passed no more than three exams” and “Marc has passed more than three exams” are contradictory is obvious, if we reckon that “Marc has passed no(t) more than three (= three or less than three) exams” is equivalent to “It is not the case that Marc has passed more than three exams”. I shall conclude these observations on the ambiguity of negation with a question which defeats me: Why do we speak ambiguously when we have the means not to do so? 7. Negation raising? The ambiguity of the scope of “not”, i.e. the ambiguity of negation between applying to a word or to a phrase brings us to so called negation raising, i.e. the phenomenon by which the contrary is made into the contradictory, by which the contrary appears in the semblance of the contradictory, as in the use of “Mary does not think that John is nice” to convey that Mary thinks that John is not nice. Maybe the so called phenomenon of negation raising has nothing to do with the raising of negation. May be it is but another instance of the ambiguity of negation (between “Mary does not believe that the earth is round” and “Mary does not believe (= dis-believes) that the earth is round”). For example, sentences of the form “a does not believe that p” are more often than not intended to convey the information that a believes that not

250

p”. If so “ does not believe” is equated to “disbelieves”. Similarly, a sentence such as “John does not like mushrooms” is most likely understood as saying that John dislikes mushrooms. And sentences of the form “a does not want that p” are regularly intended to convey the information that a wants that not p. If an Italian mother says to her daughter: “Non devi sposare quell’uomo”, or even: “Non sei tenuta a sposare quell’uomo”, most likely she is telling her daughter what she ought not to do rather than what she is not compelled to do. Why is it so? It does not help much to follow Quine in considering the phenomenon of negation raising as “an incidental idiomatic complication”, “a quirk of English usage”.17 The phenomenon is certainly not confined to English. And this is enough not to consider it as idiomatic, given that idiomatisms are by and large language bound. The uncharitable and un-enlightening speculation that speakers have a proclivity to confuse sentence and constituent negation is easily disposed of by considering that no speaker would reckon “A didn’t claim/isn’t certain that p” as a way of conveying that a claimed/is certain that not p.18 I have neither the intention nor the capacity to investigate the reasons or to offer an explanation of the so called negation raising practices. I just observe that these practices, being differential, i.e. not applying across the board, do not tell against the distinction, and the appreciation of the distinction on the speakers’ part, between sentential and constituent negation (or for that matter, constituents negations) in natural language. Rather they make it more difficult to detect. Let me note, as others have already done, that negation raising may have pragmatic reasons. The use of “I do not believe” to be understood as “I disbelieve (I believe that not)” may be caused by an intention of politeness conducive to forms of understatement. Where the subject of the attribution is not the speaker himself understatement may still apply. In any case “not believe” as indicating disbelief may be motivated by the fact that in the normal course of life people are more concerned with beliefs than with suspensions of judgment (obviously there are exceptions as voting polls where the undecided play a prominent role). If the clerk asks: do you want to reserve a seat? And you answer: I do not want to reserve a seat), the clerk can safely assume that you want not to reserve the seat. It makes no practical difference whether you do not 17. Quine (1960, 145–146). 18. See Horn (2001, 318).

251

want to reserve a seat (are indifferent to seat reservation or you want not to reserve a seat) or want not to reserve a seat. But when we come to claims matters are different. The focus is not restricted to claims and disclaims (claims that not). What one did not claim is as important as what he did claim or disclaim.19

REFERENCES Adger, G. 2003. Core Syntax. A Minimalist Approach. Oxford: OUP Englebretsen , G. 1981. Logical Negation. Assen: Van Gorcum. Frege, G. 1979. Logic. In: H. Hermes, F. Kambartel and F. Kaulbach, eds. Posthumous Writings. Oxford: Blackwell. — 1984. Negation. In: B. McGuinness ed. Gottlob Frege: Collected Papers on Mathematics, Logic and Philosophy, Oxford: Blackwell. Horn, L.R. 2001. A Natural History of Negation. Stanford: CSLI. Quine, W. V. O. 1960. Word and Object. Cambridge (MA): MIT.

19. I have to thank Andrea Bianchi and Paolo Casalegno for the enduring intelligence with which they have attended to the amelioration of the paper through its many versions. I also have to thank Gloria Cocchi, Caterina Donati, Lello Frascolla and Elisabetta Sacchi for having done their best to make the penultimate version into the ultimate one.

252