Semantics and the Dual-Aspect Use of Definite Descriptions* Michael O'Rourke Department of Philosophy University of Idaho Moscow, Idaho 83844 morourke@uidaho.edu Forthcoming in Pacific Philosophical Quarterly 3/19/98 Abstract Many philosophers of language have held that a truth-conditional semantic account can explain the data motivating the distinction between referential and attributive uses of definite descriptions, but I believe this is a mistake. I argue that these data also motivate what I call “dual-aspect” uses as a distinct but closely related type. After establishing that an account of the distinction must also explain dual-aspect uses, I argue that the truth-conditional Semantic Model of the distinction cannot. Thus, the Semantic Model cannot explain the data for which it is developed and so fails as an account of the referential/attributive distinction. 1 It is a commonplace in the philosophy of language that a definite description can be used in at least two ways: (a) to talk about whatever satisfies a particular descriptive condition (viz., the attributive use), and (b) to refer to a particular item in the world (viz., the referential use). However, while the distinction is taken as fact, it remains an open question whether it should be treated semantically or pragmatically; that is, whether it should be regarded as sanctioned by the principles of language or as a creative exploitation of linguistic resources. In this essay, I wish to add a few considerations in support of the pragmatic treatment. A condition of adequacy on any general theory of description use is that it account for all of the relevant data in a principled way, explaining the nature of the different types of use and the relationships between them. Among these types would be what I will call dual-aspect (DA) uses, which are similar in important respects to referential and attributive uses but different from each. Consider, for example, (1) Here is the next President of the United States as it might have been uttered by John McCain from the podium of the Republican National Convention. The particular descriptive condition is crucial to the point McCain would wish to make, but at the same time he would be using the description to refer to a particular man: this use is not just attributive and not just referential, but has elements of both. Philosophers such as Saul Kripke1 and Amichai Kronfeld2 have recognized these uses but their importance for the theory of descriptions has not been appreciated.3 Many have held that a semantic theory, cast in terms of truth conditions, is sufficient to account for the data that motivate the referential/attributive distinction. I believe that this is not the case and that an examination of DA uses can help demonstrate this point. I begin by arguing that the data motivating the referential/attributive distinction also motivate recognition of the DA use as a distinct type. After establishing that an account of the referential/attributive distinction must explain the DA use, I argue that the truthconditional Semantic Model of the referential/attributive distinction is unable to supply a principled explanation of it. There seem to be two natural ways to develop the Semantic Model: truth-conditional accommodation of the DA use as a third type alongside referential and attributive uses, and pragmatic 2 accommodation of it by extension from a truth-conditional account of referential and attributive uses. The first approach yields a pure truth-conditional account of all three uses, while the second yields a hybrid account that combines a pragmatic explanation of the DA use with a truth-conditional account of the referential and attributive uses. In each case, reliance on truth conditions precludes the Semantic Model from explaining the intuitive character of the DA use. As a result, we must conclude that the truth-conditional Semantic Model fails to explain the range of data for which it is developed. I close with a few remarks about the remaining explanatory alternatives. I. Dual-Aspect Uses of Definite Descriptions A speaker uses a definite description attributively to talk about whatever satisfies a certain descriptive condition. It is commonly used this way when the condition represents the speaker's only epistemic link to the item it denotes, but it is also used in this way when the speaker is more interested in the information it conveys than in the particular object satisfying it. When using it referentially, a speaker intends primarily to talk about a specific item, and she (typically) believes that the descriptive condition represents one way of thinking of the item. With the referential use, unlike the attributive, the speaker is not committed to this specific descriptive condition — another description, name, or demonstrative identifying the item might have worked just as well. Intuitive motivation for this distinction derives from the role objects play vis-a-vis the use of descriptions: is a given use dependent on a particular object, or is it independent of any one object and dependent instead on the form of language used? In his initial characterization of this distinction, Donnellan introduces the notion of essentiality to an utterance as a way of developing the relevant sense of dependence at work here.4 In the referential case, the description use is dependent on a particular item, viz., the intended referent, and so that item is essential to the utterance, whereas the form of language and its associated descriptive condition are inessential, being "merely one tool" for the communicative task. In the attributive case, on the other hand, the description is used independently of any particular item and so no one item is 3 essential to the utterance. Here it isn’t an intended referent but the descriptive condition expressed by the description that is essential to the utterance. We can make the notion of essentiality to an utterance more precise with the help of two related conditions. First, there is the Counterfactual Condition on essentiality: an item, be it an intended referent or a descriptive condition, is essential to an utterance only if it persists across counterfactual circumstances wherein that utterance is successful. Second, there is the Report Condition on essentiality: an item is essential to an utterance only if it must explicitly figure into a report for that report to correctly represent the utterance.5 These conditions capture our intuitions about essentiality by focusing on properties instantiated by an aspect of an utterance that could not be changed without undermining the success of the utterance or a report of it. Such an aspect must be seen as constituting a part of the “essence” of the utterance. In the attributive case, the aspect will be the descriptive condition; in the referential case, it will be the intended referent. With essentiality so understood,6 it becomes clear that there is a third type of description use that is relevant here but has yet to be discussed: the case where both a particular item and a descriptive condition are essential to the utterance. Call this a dual-aspect (DA) use. For an utterance involving such a use there will be an intended referent, and both that item and the descriptive condition will satisfy the Counterfactual and Report Conditions. The success of a typical utterance involving a DA use will turn on whether the listener identifies the intended referent and conceptualizes that referent as the unique item satisfying the descriptive condition. Such uses exist. Consider the following examples: (A) The utterance of (1) described above. John McCain intends the audience to think of Dole as the next President; in this case, another expression, such as "the former Senate Majority Leader from Kansas" or "Bob Dole" would not serve the speaker's purpose. But even so, he would not have used that description if he did not believe that it would enable his audience to identify Dole. Thus, both the descriptive condition and the intended referent are persistent across counterfactual circumstances. Further, given McCain’s intentions, both must figure into a report if that report is to correctly represent 4 the utterance. Thus, both are essential, and so we have a DA use. (B) After extensively investigating Smith's murder, a certain detective is convinced that Jones bludgeoned Smith to death with the bust of Heidegger he keeps in his study. While discussing Smith’s demise with Jones in his study, the detective walks over to the bust, picks it up, tests its considerable weight in full view of Jones, and then remarks: (2) The murder weapon is heavy. We can assume that the detective intends for Jones to identify the bust and think of it as the murder weapon. Given this intention, both satisfy the Counterfactual Condition and the Report Condition, and so both are essential. Therefore, this is also a DA use. In each of these examples, the intended referent is essential, and so the description use qualifies as referential. In pure referential cases, the description itself is important since it is the spoken way of getting to the referent, but it does not satisfy the Counterfactual Condition or the Report Condition. However, that is not true in the examples above. In each of these, the descriptive condition conveys information about the intended referent that the audience must include in their interpretation if the utterance is to be successful. As a result, the descriptive conditions in these examples satisfy both conditions. Thus, both cases involve DA uses of definite descriptions. II. Explaining Dual-Aspect Uses Let’s take stock. A definite description is conventionally associated with a descriptive condition, and on a particular occasion of use it might also be associated with an item that is the speaker's intended referent. When interpreting the significance of the description, one must ask two questions: (a) is the descriptive condition essential to the utterance? and (b) is an intended referent essential to the utterance? If the answer is yes to the first but not the second, the use is attributive; if it is yes to the second but not the first, the use is referential; if it is yes to both, we have a DA use. The DA use is distinct from the referential and attributive 5 uses, but it is not different in kind: all three belong in the logical space that is marked off by the notion of what is essential to an utterance. Therefore, an explanation of the referential and attributive uses should also account for DA uses. I have provided a preliminary sketch of essentiality with the help of the Counterfactual and Report Conditions; however, both of these depend on the more fundamental notion of utterance success, and so we should be able to go deeper into the essence of essentiality, so to speak, by analyzing this notion. Our pretheoretical intuitions about essentiality key on the success of utterances: to say that an item is essential to an utterance is to say that if the utterance is to be successful, then its success must be a function of the item in question. But in what does this success consist? And what is it for the success to be a function of the item in question? Two general models of utterance success that answer these questions have gained ascendancy: the Semantic Model and the Pragmatic Model. According to the former, utterance success consists in the expression of a true proposition. This proposition is the truth condition of the utterance and is conventionally associated with the sentence uttered. The truth value assigned to the utterance will be a function of the items that are constituents of this proposition.7 According to the Pragmatic Model, it consists in the achievement of the communicative ends the speaker had in uttering the sentence, where this is a function of those items that figure into the specification of the intended means to those ends. I will focus on the Semantic Model of utterance success and the corresponding conception of essentiality. The foundational assumption is that utterances express propositions that are conventionally associated with the sentences uttered as their truth conditions, and that the success of utterances (i.e., their truth) is dependent on these propositions. An item will be essential to an utterance if it is a constituent of the proposition on which the success of that utterance depends, and so we can represent the semantic notion of essentiality formally as follows:8 an item a is essential to an utterance U of a sentence S iff a is a constituent of the truth condition assigned to S on that occasion of use by the conventions of the language.9 With this conception of essentiality in hand, we can develop the Semantic Model of the different uses 6 of definite descriptions identified above. There are two ways to do this. First, we could attempt to account for each of the uses in truth-conditionally — call this the Full Truth-Conditional Strategy. In pursuing this strategy, we might isolate the distinctive truth conditions for referential, attributive, and DA utterances, respectively, noting the different constituent(s) contributed by the definite descriptions in each case. Call this the Direct Approach. A less obvious and more indirect way would be to consider a wider range of data, including uses that are not clearly explicable in terms of essentiality, and then derive the truth-conditional differences between the three uses by implication from the account that results.10 Call this the Indirect Approach. We consider these approaches in §III. Second, we could reject the assumption that all the uses are to be individuated in truth-conditional terms, arguing instead that only the referential and attributive uses are truth-conditional and that the DA use is derivative — call this the Hybrid Strategy. According to this strategy, the Semantic Model should include only the referential and attributive uses, and so it is only the distinction between these that should be drawn with the semantic notion of essentiality. The DA use, on the other hand, is seen a conversational phenomenon whose essentiality characteristics are to be explained in terms of a pragmatic notion of essentiality. This strategy will be the focus in §IV. Figure 1 presents the models and the strategic options available in connection with the Semantic Model. In sum, then, there two general semantic strategies for explaining the character of the three uses of definite descriptions we have focused on: the Full Truth-Conditional Strategy and the semantic/pragmatic 7 Hybrid Strategy. In what follows, I consider these in turn, arguing that they are unable to account for the essentiality of the DA use in a way that is consistent with their semantic commitments. Given that these are the only viable strategies available, the Semantic Model cannot accommodate the DA use of definite descriptions. Because it cannot explain all of the relevant data, we must conclude that it lacks full generality, and so we must reject it as a general theoretical account. III. The Semantic Model, Part I: The Full Truth-Conditional Strategy In this section, I develop the Full Truth-Conditional Strategy (FTC), according to which the DA use is a distinct semantic phenomenon that belongs to the same semantic category as the referential and attributive uses. There are two approaches one can take while developing this strategy, the Direct Approach and the Indirect Approach, and I consider each in turn. If the FTC is to be successful, one of these approaches must yield a principled, truth-conditional distinction between each type of use. I argue that the FTC fails because neither approach can support a semantically principled distinction between the referential and DA uses. III.1 The Direct Approach Direct development of the FTC consists in specifying the type of truth condition that is correlated with each type of use. To this end, consider an utterance of a sentence of the form (3) The is Ψ where the predicate is Ψ contributes a property to the truth condition and contains no singular term as a part. According to the FTC, if the is used attributively, the property of being uniquely is essential to the utterance and is thus a constituent of the truth condition assigned by the conventions of the language to the utterance. This truth condition is a general proposition that is quantificational in character. If it is used referentially, the intended referent is essential to the utterance and so is a constituent of the truth condition assigned to the utterance by the conventions of the language (assuming the satisfaction of a constraint to be introduced shortly). If the intended referent is a constituent, the truth condition will be a singular proposition. 8 Thus, we have two truth conditions conventionally associated with (3), one general that I regiment as: (4) <x:’!x & Ψ’x> (where '’!( )' stands for the concept expression '( ) is uniquely ') and the other singular, or object-dependent, that I regiment as: (5) < a: Ψ’a > where a is the intended referent.11 (Variables with the prime range over propositional constituents and those without it range over sentence constituents.) According to this strategy, the referential/attributive distinction identifies two different interpretation types that can be assigned to a single type of linguistic expression in conformity with linguistic convention, and so the distinction is best thought of as an ambiguity. A given definite description the can be interpreted either as a quantifier phrase or as a singular referring term. In both cases, the term introduces an item into the discourse with the help of a single descriptive condition, being the . This condition is associated with the description across interpretations by linguistic convention, but its role changes: it forms a part of the truth condition in the attributive case but only serves to identify the truth conditional constituent in the referential case.12 Turning to DA uses, recall that both the intended referent and the descriptive condition are essential to the utterance. Therefore, both the item the speaker intended to refer to with the description and the property of being uniquely are constituents of the truth condition assigned to the utterance by linguistic convention. For an utterance of (3) containing a DA use of the , where the speaker used this with the intention of referring to an item a, I will regiment the truth condition as: (6) < a: ’!(a) & Ψ’(a) > On this reading, (3) is logically equivalent to an utterance of (7) That is uniquely and (it is) Ψ.13 In this case, the description functions both as a referring term in subject position, introducing an item for predication, and as a term that predicates a property of that item. Notice that while a definite description in 9 subject position contributes its descriptive condition to (6), the truth condition is nevertheless singular since it contains a as a constituent. These are the truth conditions, but an equally significant part of the FTC is the relation between truth conditions and utterances. Central to the Semantic Model is the idea that linguistic convention accounts for the association of an utterance with its truth condition, and it is in virtue of this that the utterance is semantically meaningful. Thus, with respect to each type of use, the correlation between an utterance embedding that use and its truth condition is conventional in character. When a speaker uses a definite description attributively, she relies on the conventionally associated quantificational apparatus to denote an item for predication; in the referential and DA cases, she isolates an item independently (via perception, perhaps) and intends to refer to that item, but in doing this her linguistic performance must be constrained by linguistic convention if it is to count as semantically well-formed. Specifically, in using the definite description, the speaker should intend to refer to the item it denotes. If the speaker does not intend to refer to the denotation, we have what we will call a misfit utterance. Misfit utterances rest on mistakes, or violations of linguistic convention, and so their capacity to express a determinate proposition is undermined. Allowing the intended referent to be a constituent of the truth condition in such cases would be to give in to an unconstrained Humpty Dumpty theory of meaning, and this is inconsistent with the conventionalism that is at the heart of the Semantic Model. With misfit utterances, the descriptive condition goes in one direction and speaker intention goes in another. The Semantic Model cannot go with the descriptive condition because these are referential uses, but it cannot go with speaker intention because of its conventionalism. Therefore, the Semantic Model cannot countenance a determinate truth condition for these utterances and so has nothing to say about them.14 This yields an a priori constraint on the semantic admissibility of utterances: Misfit Constraint: if an utterance contains a definite description used to refer to an item it does not denote (i.e., if it is a misfit), then it falls outside the explanatory reach of the Semantic Model. Given the Semantic Model’s fundamental commitment to linguistic convention, the Misfit Constraint forms an 10 important part of it and so also the FTC account we have developed here; however, as we will see, it is the Misfit Constraint that renders the FTC account unable to explain DA uses. With the FTC account in place, we now return to the DA truth condition. The extension to this truth condition from the other two is formally unproblematic, and given that it enables the FTC account to explain DA uses it seems theoretically well-motivated, but I will now argue that this appearance is misleading for two reasons. First, while it is formally straightforward, the extension misrepresents the role played by the descriptive condition in the utterance. By assigning truth conditions similar in form to (6), the account ascribes a predicative role to the descriptive condition; this, however, is a mistaken characterization of DA uses. Consider the DA use in the utterance of (2) above. The descriptive condition — being the murder weapon — is used to refer to the bust of Heidegger, to identify it for Jones as the item about which he wishes to make a claim. Thus, it brings the bust to Jones’ attention as the subject of predication but is not itself predicated of the item. Rather, this mode of presentation constrains how we are to regard the subsequent predication: we are to conceive the object in terms of it as we subsume the object under the predicated property in our interpretation of the utterance. If it were possible to introduce an item and predicate a property of it with a single definite description, you could express a complete proposition simply by uttering a definite description, but this is absurd. If we treat the descriptive condition as predicative, we lose the difference between the role it plays in presenting the subject of the claim and the predicative role played by the predicated property, and this is a sign that something has gone awry in our attempt to explain the semantics of the utterance.15 It is not clear how one could modify the truth condition so as to represent the truth-conditional role of the descriptive condition while retaining the singularity necessary to represent its referential role in DA cases,16 but even if this could be done, there would remain a second and deeper difficulty. If the truth conditions of utterances embedding DA uses can be explained by an account that includes only (4) and (5), it would be theoretically otiose to introduce (6). To justify extending the account to include (6), there must be admissible utterances where (a) the DA truth condition diverges from the attributive and referential truth conditions by 11 assigning a different truth value to the utterance in some circumstance of evaluation, and (b) it is intuitively correct to assign the DA truth condition to the utterance in that circumstance.17 Identification of such utterances would establish the existence of cases where the original account would fail to supply the correct truth condition; on the other hand, failure here would establish the semantic adequacy of the original account. To begin, note that (6) is similar to (5) and different from (4) in that both (5) and (6) are singular while (4) is general. Thus, both (5) and (6) will diverge from (4). The question then becomes whether we can locate a circumstance of evaluation with respect to which (5) and (6) diverge. The propositional constituents contributed by the predicate and the referential intention will be the same in both, but (6) contains a constituent that is not present in (5), viz., '’!(a)', and this additional element is the lone source for potential truthconditional divergence. It is also clear where this divergence will be found: in utterances where the item a does not satisfy the concept expression ( ) is uniquely but does satisfy the concept expression ( ) is Ψ . In such a case, the utterance of (3) will be true interpreted relative to (5) but false if interpreted relative to (6). However, these are misfit utterances, and so they are ruled out in advance by the Misfit Constraint. As a result, these utterances are inadmissible and so we cannot invoke them to satisfy conditions (a) and (b). Unfortunately for the Direct Approach, there are no other alternatives. Thus, for all utterances covered by this account, the truth value of utterance U , which embeds a DA use of definite description d will be the same as an utterance in which the definite description d is used referentially in every circumstance of evaluation. Therefore, (5) accounts for the truth-functional behavior of U and so there is no need to introduce (6) as its truth condition. From this and considerations of parsimony, it follows that proponents of the Semantic Model must give DA uses a referential reading, and so deny that they qualify as distinct uses. This argument assumes that the Misfit Constraint is fundamental to the Semantic Model. If we were to reject this assumption, we could gain the data necessary to establish the truth-conditional divergence between the DA and referential uses and thereby justify extending the account to include (6). However, as we have seen, rejection of this assumption is not an option for proponents of the Semantic Model. According to this 12 model, the meaning of an utterance is its truth condition, and the association of an utterance with a truth condition is determined by conventions concerning the use and interpretation of sentences within a linguistic community. If misfit utterances are admitted as semantically legitimate, then community-wide convention is set aside in favor of speaker intentions, and so it becomes possible to evaluate utterances for truth relative to propositions that are not conventionally associated with the sentences uttered. Semantic accommodation of misfit utterances is a renunciation of the priority of linguistic convention, and since linguistic convention is the foundation of the view, misfit utterances cannot be accommodated. Instead, proponents of the Semantic Model who favor the FTC strategy must at this point reject the claim that referential and DA uses diverge. Thus, the FTC developed in this way does not explain the essentiality characteristics of the DA use, and so we must continue our search. III.2 The Indirect Approach The version of the Semantic Model under consideration — the FTC account — explains the essentiality characteristics of the three uses in terms of the truth conditions with which they are correlated. We have attempted to identify these truth conditions directly by following intuitions about essentiality, but that has failed. However, there is an alternate approach available that is consistent with the commitment to three distinct truth conditions. While the data we used support intuitions about essentiality, they may not be sufficient to support a semantic distinction between the uses. It may be that a semantic distinction can be drawn in this place only if one considers a wider range of data, including bound variable uses of definite descriptions and other uses that do not readily lend themselves to characterization in terms of essentiality. By throwing open the range of considerations, we may reveal unappreciated characteristics that support semantic generalizations containing the three uses as distinct cases. If so, the truth conditional distinction between the uses would be implied as a corollary to a more general semantic theory of description use. George Wilson supplies a model of a semantic theory constructed along these lines in Wilson (1991). He argues that the distinction between referential and attributive uses is semantic in character, but he rejects the 13 idea that this distinction can be drawn on the basis of the standard range of data usually invoked in these discussions. Instead, he argues that it is a “more complicated range of considerations ... [that] license[s] a semantical distinction between pronominal and attributive descriptions,” and referential uses “reappear as a special case of the former.”18 Drawing from definite descriptions used as bound variables in quantificational constructions and as unbound anaphora in cross-sentential contexts, Wilson adduces evidence in support of the claim that definite descriptions can function like pronouns, picking up their referents from other expressions in stretches of discourse. As he sees it, this reveals a natural distinction between definite descriptions used as quantifiers and those used pronominally. Since these roles are semantically distinct, this distinction qualifies as a semantic one. With this distinction in place, Wilson observes that the analogy between pronominal uses of definite descriptions and pronouns could be extended even further: since descriptions have a pronominal function and pronouns have a deictic use, we should expect descriptions to have a deictic use as well. Thus, the referential use is explained as a specific type of pronominal use and so is semantically distinct from the attributive use. Wilson’s argument for a semantic interpretation of the referential/attributive distinction is not based directly on an examination of essentiality but indirectly on an analogy with pronouns; nevertheless, it supports our intuitive interpretation of the essentiality considerations by implying that the referential use is correlated with a singular truth condition and the attributive use with a general one. The suggestion here is that perhaps a similar indirect argument could be constructed that establishes the semantic legitimacy of the DA use. In constructing this argument, we would rely on factors in addition to essentiality to support semantic generalizations that would imply the referential/attributive/dual-aspect distinction. Taking Wilson’s account as a model, we would proceed in two stages: (1) establish that there are semantically principled generalizations involving definite descriptions that are grounded in a wider range of data, and (2) locate the DA use as a distinct type within those generalizations. The results of our investigation of the Direct Approach supply a constraint on such an argument: the referential and DA uses must not be lumped together as species in the same semantic genus unless something over and above truth conditions is 14 used to distinguish them. Call this the Independence Constraint (IC). How might we proceed? Following Wilson once again, we could begin by taking attributive uses to constitute a single semantic category; then, we could establish that the remaining occurrences of definite descriptions cluster together into categories that are analogous in salient respects to semantic categories comprising occurrences of other expressions; finally, we could locate the referential and DA uses within these categories as distinct semantic types. Note that both the referential and DA uses involve reference to a particular item, a fact that renders both different in kind from the quantificational attributive use; given this, assignment of either of these to the same category as the attributive use would appear to be out of the question. When one makes DA use of a definite description, one employs it as a singular referring expression while making essential use of the descriptive condition. If we pursue the indirect strategy, we will categorize the DA use of definite descriptions semantically by analogy with another type of expression, and the candidates are three: pronouns, demonstratives, and proper names. First, we might try to account for the DA use by extending Wilson’s account of the pronominal use of definite descriptions. To justify this extension, we would have to identify a use of pronouns that depends essentially on both the associated descriptive condition and the intended referent. The problem, however, is that there is no conventionally distinct DA use of pronouns that could support an analogy, and any attempt to distinguish such a use truth-conditionally from the referential use would would appear to run afoul of the IC. Second, we might try to exploit an analogy with demonstrative phrases that are associated with descriptive conditions, such as, “that bird over yonder.” McKinsey has developed this alternative, arguing that we should treat definite descriptions used referentially as demonstrative phrases.19 But when we turn to DA uses, we run into the same difficulty we encountered with pronouns: there is no conventionally distinct DA use of demonstrative phrases that could support an analogy, and the IC rules out any truth-conditional attempt to distinguish such a use from the referential use of demonstrative phrases. The category of proper names is the only grammatical category remaining as a candidate for this sort of argument, but it is even less clear how one could account for DA uses of definite descriptions by analogy with 15 it because names, unlike definite descriptions, do not explicitly display their descriptive conditions. While these alternatives appear to be non-starters, that does not prove that the indirect approach is a failure. There might be unnoticed aspects of the use of these expressions that could resuscitate one of these alternatives, or there might be associations with other linguistic categories that could be exploited. However, it is not clear what these would be. More fundamentally, though, the Semantic Model as we have represented it privileges truth conditions, and our inability to distinguish referential and DA truth conditions likely dooms any attempt, be it direct or indirect, to establish the semantic autonomy of the DA use. But all is not lost. There remains a semantic alternative that we have not investigated: use the Semantic Model of essentiality to distinguish attributive and referential uses but account for the DA use in pragmatic terms. It is to that alternative that we will now turn. IV. The Semantic Model, Part II: The Hybrid Strategy Our attempts to explain the DA use in terms of truth conditions have both gone for nought, but we are not forced thereby to renounce our commitment to the Semantic Model of essentiality. We could develop this model according to the Hybrid Strategy by taking referential and attributive uses to be differentiated truth conditionally in conformity with the Semantic Model and then taking DA uses to be conversational constructions that combine referential and attributive elements. From this perspective, the attributive and referential uses form autonomous semantic categories, grounded in their distinctive truth conditional forms (4) and (5), respectively, and the DA use is explained as a pragmatic phenomenon that is dependent on those two categories. This is not out of line with plausible interpretations that we can give the relevant data. Returning to our examples, it is not unnatural to interpret (1) as an attributive use — the temporal index “next” could be taken to signal this — that is accompanied by a referential implicature, given that the audience is intended to identify Dole as the person who has the predicated property. Conversely, (2) would appear to be a referential use that is made with the further intention to convey pragmatically that the referent satisfies the descriptive 16 condition, thus generating an attributive implicature. These interpretations of the data support the Hybrid Strategy, but there would appear to be a problem. We have used truth conditions to explain the essentiality considerations that motivate introduction of the DA use. If the DA use is a pragmatic phenomenon, though, we will be unable to account for it in terms of this notion of essentiality. However, there is a notion of essentiality that can be derived from the pragmatic standard of utterance success described in §II, and this is what we will use to explain the essentiality characteristics of the DA use. If we take an implicature to be a proposition, we can think of this notion as follows: an item a is pragmatically essential to an utterance U of a sentence S iff a is a constituent of those propositions that the speaker intends to implicate with U. The communicative intentions involved here would be those necessary to generate the implications by producing U in its specific context.20 Thus, we will explain attributive and referential uses in terms of the semantic notion of essentiality and the DA use in terms of a pragmatic notion of essentiality that is grounded in the communicative intentions of the speaker. Above, when we revisited the examples to evince support for the Hybrid Strategy, we noted that we can arrive at DA uses by pragmatic implication from either referential or attributive uses. This is as it should be, given that both referential and attributive uses are associated with distinct semantic categories and so with distinct truth conditions. It will be natural to use a definite description referentially in certain circumstances — when the referent is obvious to the conversational participants, as with (2)—and in others it will be natural to use it attributively, as with (1). In the former circumstances, the utterances will have referential truth conditions and in the latter they will have attributive truth conditions. In each of these circumstances it would appear that nothing precludes the speaker from pragmatically conveying the element necessary for a DA use: the (pragmatic) essentiality of the descriptive condition in the referential case (e.g., being the next President) and the (pragmatic) essentiality of an intended referent in the attributive (e.g., the bust of Heidegger). Thus both referential and attributive truth conditions should be able to play the role of what is said in pragmatic implication. 17 The role of the implicatures21 will be played by propositions that are inductively implicated from the truth conditions of the utterance, the utterance context, and the Cooperative Principle and attendant maxims.22 Our examples indicate that for a particular utterance embedding a DA use, either the referential or attributive truth condition will serve as the truth condition; call the truth condition from which the implicature is generated in a particular case the implicature’s platform. If an attributive proposition of form (4) is the platform, a speaker who makes DA use of the description will implicate that a certain item a is uniquely and so is Ψ. For example, in the case of (1) above, the implicature will be that this person — viz., Dole — is both here and the next President of the United States. This implicature can be regimented as (6) above. If a referential proposition of form (5) is the platform, a DA use of the description will generate an implicature that the item a is uniquely . In (2) above, the detective makes the property of being the murder weapon essential by implicating that the bust of Heidegger, his intended referent, is the unique item that has it. This I regiment as: (8) <a: ’!(a)> We adopt (8) instead of (6) in this case because this is what will be necessary to make the speaker conform to Grice’s Cooperative Principle. In the DA case, the speaker is not fully cooperative at the level of what is said, and so the interpreter brings the speaker in line with the Cooperative Principle by supplementing what is said with an additional proposition. Since we retain the truth condition here, there would be no reason to duplicate the meaning it expresses at the level of what is implicated, and this is precisely what would happen if we took (6) as the implicature. However, the combination of (5) and (8) is equivalent to (6) in that both place the same constraints on understanding: to grasp either (5) and (8) or (6), one must identify the intended referent as the unique satisfier of the associated descriptive condition. Before proceeding, we should note that these implicatures appear to be what Grice called “generalized conversational implicatures.”23 As we have seen, DA implicatures are made possible by the conventional association of both referential and attributive truth conditions of forms (4) and (5) with utterances of sentences that contain definite descriptions in subject position.24 Given this, there is certainly “room for the idea that an 18 implicature of this sort is normally carried by saying that p,”25 even if in many cases it is not implicated, perhaps because it is canceled by features of the context of utterance. This would seem to undermine the idea that these are particularized conversational implicatures, but it doesn’t seem strong enough to qualify them as conventional implicatures. If we compare the DA implicature to one that is paradigmatically conventional — the implicature associated with ‘but’ — there are enough differences to cast doubt on this classification. Most prominent among these is that it seems very unnatural to read ‘but’ without the implicature, but this is not true in the case of definite descriptions. It is not at all uncommon for us to read definite descriptions as simply attributive or simply referential; indeed, the DA use would seem to be the least typical reading. In light of this and similar considerations, it would appear that the implicature generated by the DA use should be classified as a generalized conversational one. At this point, things look good: we have retained our commitment to the distinction between the three uses, drawn in terms of essentiality; further the Semantic Model of essentiality explains the character of referential and attributive uses and so continues to play a fundamental role in the account. Nevertheless, things are not as good as they look, as will become clear when we examine whether the purported implicatures pass the tests that signal the presence of conversational implicatures: the Cancellation Test, the Non-Detachibility Test, and the Derivability Test. First, these implicatures do not form a part of the truth condition of an utterance, and so we should be able to explicitly cancel the implicatures without contradiction. Second, since they are derived in a discourse context from the truth condition of the utterance and the maxims, these implicatures should be generated by utterance of any sentence in that context that has the same truth condition; thus, we should be able to generate the implicature by uttering sentences in that context that are derived by substituting synonymous expressions in for the definite description. If we can, then we say that the implicature exhibits a high degree of non-detachibility, and conversational implicatures should exhibit a high degree of this. Finally, we must be able to derive the implicated proposition from the truth condition and the maxims in the context of the utterance.26 19 When there is a DA use of a definite description and the attributive proposition serves as the platform, the implicature generated satisfies these tests. Since McCain could have said, without contradiction, (9) Here is the next President of the United States, and I’m not talking about Dole, the implicature that Dole is the intended referent is cancelable. It also nondetachable, as evinced by the fact that the implicature would have been generated had he said, (10) Here is the chief executive of the United States to follow the current one. Admittedly, this is somewhat odd and would likely generate additional implicatures, but a proposition of form (6) would be a part of the pragmatic package. Finally, we can derive the implicature from the attributive platform. In uttering (1), McCain asserts the general proposition <x: H(x) & P!(x)>,27 and there is no reason to suspect his commitment to the Cooperative Principle. However, it would be a violation of the Maxim of Relation to assert in this context only that one and only one person will be the next President of the United States and that person is here, so he must be implicating something more. Turning to the Maxim of Quality, he must believe he has adequate evidence for his assertion, and given that he believes Dole will be the next President of the United States, he must believe that Dole is here. Since the context of utterance ensures that all of this is common knowledge and McCain has done nothing to suspend this, he must intend for his audience to think that Dole will be the next President and that he is here. Therefore, this is what he implicates.28 Implicatures generated on top of referential platforms, are a different story, however. Consider the following attempt to cancel the implicature associated with (2), the form of which is given by (8): (11) The murder weapon is heavy, and the murder weapon is not the murder weapon. Here, the first occurrence of the description is a DA use with a referential platform, which is to say that it contributes only its intended referent to the truth condition. Let us assume that the second occurrence is referential, and co-referential with the first, and the third is attributive. On this reading, (11) would have as its truth condition a proposition of the form: (12) <a: Ψ’(a) & ~’(a)>, 20 which is not a contradiction. However, something has gone wrong with (11), and this can be demonstrated if we enlist the help of the Misfit Constraint. The second conjunct of the sentence reads like an explicit contradiction, but since the first and second occurrences of the description in this conjunct contribute different items to the truth condition, contradiction is avoided. Thus, this conjunct is similar to an utterance of “This is not this”, where each occurrence of the demonstrative mentions a different item. However, there is an important difference between this sentence and (11): the two occurrences of the description in the second conjunct of (11) cannot mention different items. For (11) to pass the Misfit Constraint, the referential occurrences of the description must refer to the item denoted by the descriptive condition. Given this limitation, the second conjunct — and so (11) — can never be true, since it must deny the property of being the unique weapon that murdered Smith of the only weapon that has that property. Thus, while (11) is not necessarily a contradiction from a truth conditional perspective, it implies one when coupled with the Misfit Constraint. One might suggest at this point that this is an isolated case that arises because of the role played by the description in the sentence, and that if we stipulate that it not be used in the subject position in the second conjunct, we could cancel the implicature without difficulty. For example, we could use a sentence like (13) The murder weapon is heavy, and this is not the murder weapon. where a demonstrative is used to refer to the bust instead of the description. However, in addition to being ad hoc, (13) does not avoid the contradiction problem. The first occurrence of the description in (13) must conform to the Misfit Constraint, and so the descriptive condition must in that case denote the intended referent. However, this places a constraint on our interpretation of the second conjunct: if ‘this’ is to be coreferential with the description in the first conjunct, it must refer to the very item denoted by the second occurrence of the description. The second conjunct will never be true, and thus once again we have a contradiction. Any sentence that cancels the implicature by expressing (12) will imply a contradiction when combined with the Misfit Constraint. Since the Misfit Constraint is fundamental to the Semantic Model, we 21 must find fault with our attempts to cancel the implicature. Despite appearances, they must be contradictions, and so the implicature fails the Cancellation Test. This implicature also fails the Nondetachability Test. If ‘the murder weapon’ is given a DA use with a referential platform, it makes the same contribution to the truth condition that ‘this’ would have had it been used to refer to the bust. However, if the detective had uttered the sentence, “This is heavy”, no implicature would have been generated, and so it can be detached. Finally, there is reason to believe that the implicature is not derivable. On the surface, it appears that we can derive it in the following fashion. In uttering (2), the detective asserts the singular proposition <a: H(a)>, where a is the weapon used to murder Smith and ‘H(x)’ stands for ‘x is heavy’. We can assume that the detective has not been uncooperative. However, (2) would violate the Maxim of Quantity if all the detective wanted to do was refer to the bust he held in his hand, since a demonstrative such as ‘this’ would be suitable for that purpose, and so he must be implicating something as well. We get at the implicature by way of the Maxim of Relation: the description used would be irrelevant and unduly inflammatory unless the detective believed that the bust in his hand was the murder weapon. In those circumstances, all of this is clear to both parties and the detective does nothing to suggest that any of it is mistaken. Therefore, the detective intends Jones to think of the bust as the murder weapon and so this is what he implicates. However, despite appearances, this story provides no solace to the adherent of the Hybrid Strategy. There is a problem: the detective intends Jones to use the descriptive condition ‘being the murder weapon’ in identifying the truth condition of the utterance, and so it would be redundant for him to intend that Jones use the truth condition to calculate that condition as the crucial constituent of the implicature. It would appear that the process of derivation is itself short-circuited by the logical properties of the DA use. A proposition of form (8) will figure into Jones’ interpretation of (2), but it will figure into the process of interpretation before the truth condition and so need not be derived from the truth condition. According to the Hybrid Strategy, an implicature of form (8) is generated by an utterance that embeds 22 a DA use of a definite description with a referential platform. However, the purported implicature fails the Cancellation, Nondetachability and Derivability Tests, and so we must conclude that it is not an implicature after all. Since the DA use requires that this proposition be added to a referential truth condition, we can conclude that the referential truth condition cannot serve as a platform for generating DA implicatures. Therefore, only the attributive truth condition can serve as the platform when DA use is made of a description. This result is unacceptable, though. At the beginning of this section, we demonstrated how our intuitions regarding (1) and (2) could be explained in pragmatic terms given the semantic distinction between referential and attributive uses. However, the theoretical development of this demonstration — the Hybrid Strategy — permits us to derive only one of these explanations. We cannot derive the DA use from the referential truth-condition, and so we cannot explain (2) in a way that conforms to our intuitive reading. As a result, the Semantic Model developed according to this strategy cannot account for all of the DA data, given their essentiality characteristics. This is an important result. The Hybrid Strategy was the final development of the Semantic Model, and its failure means that no version of the model can account for the essentiality characteristics of the DA use. Thus, it would appear that we must opt for the Pragmatic Model over the Semantic Model. There may be a way around this conclusion, however. We have just seen that the Hybrid Strategy does account for the DA use if the attributive truth condition serves as the platform. Given this, we could accommodate all DA uses if we simply stipulate that the attributive truth condition serve as the platform in those cases — call this the Attributive Alternative (AA). Upon consideration, though, this proves unhelpful. We have assumed that the attributive proposition (4) and the referential proposition (5) can both serve as truth conditions in different circumstances. Since both can serve as truth conditions, both should serve as implicature platforms: a speaker should be able to flout the conversational maxims with an utterance that has either (4) or (5) as its truth condition and in so doing generate conversational implicatures. Further, since there appear to be no general conversational restrictions that preclude the speaker from flouting these maxims with 23 either the definite description the or the predicate is Ψ , we should expect (4) and (5) to serve as platforms for implicatures grounded either in the description or the predicate. Call this the Implicature Platform Property (IPP) — as truth conditions, both (4) and (5) should have IPP. Consider (5). AA will close (5) off from implicatures involving the associated descriptive condition being the . Are there other implicatures with (5) as the platform that are grounded in the description? Given the argument advanced above for the conclusion that (5) cannot serve as the platform for implication of (8), it would appear not. In cases where (5) serves as the platform for an implicature that is grounded in the definite description, the descriptive condition will typically figure into the implicature generated. (E.g., “The late student should consider remaining quiet”, said while looking right at the late and loquacious student.) In such a case, it will be possible to use a less descriptive expression, such as a demonstrative or a name, to refer to the item in question. Given this, the speaker’s use of a definite descriptive to refer to the item will be interpreted as a violation of the Maxim of Relevance unless the descriptive condition was implicated, and so the descriptive condition associated with the definite description will work its way into the implicature. However, as we have seen, the referential truth condition cannot serve as the platform for an implicature that involves its descriptive condition. Thus, the argument advanced above generalizes: (5) cannot be an implicature platform in cases where it should be since it cannot serve as a platform for implicatures involving the embedded definite description. The referential proposition (5) does not have IPP, a property that we would expect it to have if it were a full-fledged truth condition, whereas the attributive proposition (4) does have IPP and seems wellestablished in the role of truth condition. Above we were led to consider AA as a way of buttressing our commitment to the Semantic Model. However, investigation into AA has revealed a problem that undermines its ability to serve in this role: there is an incongruity between (4) and (5) — i.e., (4) has IPP and (5) does not — that suggests (4) is more fundamentally truth-conditional than (5), a result that does not agree with the foundational assumptions of the model. In light of this, AA does not appear promising. Therefore, since this was the last hope for the Semantic 24 Model, we seem to have good reason to conclude that it cannot account for the essentiality considerations, at least until we are given a defense of AA or some suitable substitute. I take it that we thus have reason to embrace the Pragmatic Model. V. Conclusion Referential and attributive uses are introduced initially in terms of essentiality, and that notion can be used to highlight a third, related type of use. These three uses are most clearly delineated relative to this notion, and while the notion itself may not be robust enough to support a complete theory of description use, it does serve as a desideratum for any such account. Two general explanations of essentiality stand out: the Semantic Model, with its emphasis on truth conditions, and the Pragmatic Model, with its emphasis on speaker’s intentions. Above, we used this distinction to chart the range of possible accounts. Keeping this distinction in mind, we can chart this same range in a different way, one that is related more closely to the DA use. In particular, we can divide them on the basis of how they answer the question, “Does the DA use belong to the same category as referential and attributive uses?” We could develop the affirmative answer to this question either semantically or pragmatically, depending on our view of essentiality. If we commit to the semantic conception of essentiality, we embrace the Full Truth-Conditional Strategy and account for all three uses in terms of distinct truth conditions assignable to each. As we observed, we might take either a direct or an indirect approach in our search for such an account. If we commit to the Pragmatic Model, we would argue that the proper way to explain the uses is to disengage them from truth conditions altogether and establish that they belong to the same pragmatic category, such as speaker’s meaning, perhaps.29 If we were to respond in the negative, we would also have a choice 25 between semantic and pragmatic elaborations. If semantic, we embrace the Hybrid Strategy and explain the referential and attributive uses truth-conditionally and the DA use as an implicature generated from referential or attributive platforms. If pragmatic, we might allow that the attributive use is indeed grounded in the quantificational truth condition, but we would argue that the crucial essentiality differences between it and the other uses are to be represented in terms of implicatures from an attributive platform.30 Figure 2 represents the range of alternatives. The results of the preceding three sections establish that the available semantic alternatives do not work. If we treat the DA use as an autonomous semantic category and attempt to explain it in terms of truth conditions, we discover that its autonomy is inconsistent with semantic conventionality. This conventionality cannot be compromised without renouncing the Semantic Model altogether and so we must deny that the DA use is semantically autonomous. The semantic 26 alternative is to account for the DA use pragmatically. This alternative presupposes that the referential and attributive uses would belong to one logical category and the DA use a second and derivative one; however, we find that the attributive truth condition is more primitive than either of the others, a result that threatens the Semantic Model. Thus, we are left with the Pragmatic Model. The three uses must be explained in terms of the intentions of speakers and not in terms of conditions on the truth of utterances. There remain two options from which to choose: an account according to which all the uses belong to the same pragmatic category, and one where the attributive use is explained semantically and the others pragmatically. I have no definitive argument at this point, but I am inclined toward the former option for two reasons. First, uniform classification of the uses would agree with the uniform use of essentiality to distinguish the uses in the first place. Second, identification of use in a particular case is made by reference to the intentions of the speaker and not to the conditions on the truth of the utterance; if we make the plausible assumption that identification is cued by fundamental differences between the uses, then we should disconnect explanation of the uses from truth conditional considerations altogether and develop an account that is grounded in those intentions, thereby bringing explanation in line with identification. VII. References Bach, Kent. 1981. "Referential/attributive." Synthese, 49, 219-244. Barwise, Jon and John Perry. 1983. Situations and Attitudes. Cambridge, MA.: The MIT Press. Bezuidenhout, Anne. 1997. “The communication of de re thoughts.” Nous, 31, 197-225. Burge, Tyler. 1974. “Demonstrative constructions, reference, and truth.” The Journal of Philosophy, 27 71, 205-223. __________. 1977. “Belief de re.” The Journal of Philosophy, 74, 338-362. Casteneda, Hector-Neri. 1967. “Indicators and quasi-indicators.” American Philosophical Quarterly, 4, 85-100. __________. 1979. "On the philosophical foundations of the theory of communication." In P. A. French, T. E. Uehling, Jr., and H. K. Wettstein (Eds.), Contemporary Perspectives in the Philosophy of Language, 125-146. Minneapolis: University of Minnesota Press. Davidson, Donald. 1984. Inquiries into Truth and Interpretation. Oxford: Clarendon Press. Devitt, Michael. 1981. "Donnellan's distinction." In P. A. French, T. E. Uehling, Jr., and H. K. Wettstein (Eds.), Midwest Studies in Philosophy VI, 511-524. Minneapolis: University of Minnesota Press. Donnellan, Keith. 1966. "Reference and definite descriptions." The Philosophical Review, 75, 281-304. __________. 1978. "Speaker reference, descriptions, and anaphora." In P. A. French, T. E. Uehling, Jr., and H. K. Wettstein (Eds.), Contemporary Perspectives in The Philosophy of Language, 28-44. Minneapolis: University of Minnesota Press. Evans, Gareth. 1982. The Varieties of Reference. New York: Oxford University Press. Frege, Gottlob. 1892. "On sense and reference." in P. Geach and M. Black (eds.), Translations from the Philosophical Writings of Gottlob Frege, 56-78. Oxford: Basil Blackwell. Grice, H. Paul. 1969. "Vacuous names." In D. Davidson and J. Hintikka (Eds.), Words and Objections, 118-145. Dordrecht: D. Reidel Publishing Company. __________. 1989. Studies in the Way of Words. Cambridge, MA.: Harvard University Press. 28 Kripke, Saul. 1979. "Speaker's reference and semantic reference." In P. A. French, T. E. Uehling, Jr., and H. K. Wettstein (Eds.), Contemporary Perspectives in the Philosophy of Language, 6-27. Minneapolis: University of Minnesota Press. Kronfeld, Amichai. 1990. Reference and Computation. Cambridge: Cambridge University Press. Loar, Brian. 1972. “Reference and propositional attitudes.” Philosophical Review, 81, 43-62. Lockwood, Michael. 1975. "On predicating proper names." The Philosophical Review, 84, 471-498. McKinsey, Michael. 1979. “The ambiguity of definite descriptions.” Theoria, 45, 78-89. Neale, Stephen. 1990. Descriptions. Cambridge, Mass.: The MIT Press. Perry, John. 1993. The Problem of the Essential Indexical. Oxford: Oxford University Press. Recanati, François. 1993. Direct Reference. Oxford: Blackwell. Rouchota, Villy. 1992. “On the referential/attributive distinction.” Lingua, 87, 137-167. Schiffer, Stephen. 1978. “The basis of reference.” Erkenntnis, 13, 171-206. Searle, John R. 1979. "Referential and attributive." Monist, 62, 190-208. Simons, Mandy 1996. “Pronouns and definite descriptions: A critique of Wilson.” The Journal of Philosophy, 93, 408-420. Wettstein, Howard K. 1991. Has Semantics Rested on a Mistake?. Stanford: Stanford University Press. Wiggins, David. 1975. "Identity, designation, essentialism, and physicalism." Philosophia, 5, 1-30. Wilson, George. 1978. "On definite and indefinite descriptions." The Philosophical Review, 87, 4876. __________. 1991. "Reference and pronominal descriptions." The Journal of Philosophy, 88, 359387. 29 VI. Notes * I am grateful to the many people who have helped me with this paper over the years. Special thanks are due to Bruce Glymour, Paul Kjellberg, Frank Menetrez, John Perry, David Shier, and Corey Washington. 1 Kripke (1979), p. 25 (fn. 28). Kripke contends that one could use a description to talk about whatever item fits the description and about a particular item believed to be the one fitting the description at the same time. He calls these "mixed motive" cases. 2 Kronfeld (1990), p. 48. 3 Other philosophers have also addressed this type of linguistic phenomenon. See Brian Loar’s discussion of singular terms that can make a “dual contribution” to the truth conditions of sentences that embed them in Loar (1972), pp. 52-57 (cf. Burge 1977, pp. 341-342); the distinction between purely indexical, descriptive, and “indeterminate” statements in Castaneda (1967), p. 94; and the discussion of this type of phenomenon in connection with metaphor in Davidson (1984), pp. 249-250. 4 Donnellan (1966), p. 285. See also Bach (1981) for additional discussion of this notion. 5 What it is for an item to “explicitly figure into a report” will vary depending on the character of the item. If it is the descriptive condition, then a description that expresses a condition sufficiently similar to the one used by the speaker must be used in the report (where “sufficient similarity” will be interest and context relative); if it is the referent, then the referent must be mentioned in the report. For a related discussion that focuses on DA uses of singular terms in belief reports, see Loar (1972), pp. 52-57. 30 6 There are a wide variety of ways to use descriptions — as identifiers, explicit quantifiers, bound variables, function indicators, and several more — and a general account of description use should have something to say about all of these. (For detailed accounts of these, see Barwise and Perry (1983), pp. 158159, Neale (1990), ch. 2, Wilson (1991), Rouchota (1992), and Simons (1996).) It is important to note here that I do not argue for essentiality as the cornerstone of a general account; rather, I wish only to claim that essentiality serves as a central desideratum for any adequate account of description use. It may not be robust enough to serve as the foundation for a general theoretical account, but such a theory had better explain it. 7 On the Semantic Model, semantic characteristics are taken to be truth-conditional characteristics, and so it is in terms of these that it explains the referential/attributive/dual-aspect distinction. If we were to reject this identification and take utterance semantics to include more than just truth conditions, we would have additional semantic models. However, distinctions between the truth condition of an utterance and other parts of its complete semantic value (e.g., its semantic content) are typically made with the help of pragmatic considerations. In light of this, I will regard these non-truth conditional accounts as alternatives under the Pragmatic Model. See Recanati (1993) and Bezuidenhout (1997) for helpful discussion of these alternatives. 8 Cf., Wilson (1978), Searle (1979), Devitt (1981), Barwise and Perry (1983), Wilson (1991), Wettstein (1991), Chs. 4 & 5. For critical discussion of this thesis in the context of the referential/attributive distinction, see Bach (1981) and Neale (1990), p. 65. 9 For a similar idea, see Barwise and Perry (1983), p. 146. 10 For an example of this strategy, see Wilson (1991). 11 For a related discussion, see Perry (1993), Ch. 13. Cf. Neale (1990), pp. 49-50, n. 1. 31 12 Cf. Wilson (1991), pp. 377-379. For other formulations of the ambiguity view, see Devitt (1981), Barwise and Perry (1983), and Recanati (1993). 13 Cf. Wilson (1978), pp. 68-71, for a related discussion of this type of truth condition. 14 Of course, this does not mean that we cannot recover a determinate proposition here, since in many cases, pragmatic tools will help us do just that. For related discussion of misfit utterances, see Wettstein (1991), pp. 36-38, and Wilson (1991), pp. 361, 366, and 369, both of whom regard these utterances as semantically illegitimate. Others who share this opinion include Lockwood (1975), Wiggins (1975), Kripke (1978), Casteneda (1979), Bach (1981), Evans (1982), and Recanati (1993). 15 To deny that the descriptive condition of a definite description in subject position can play a predicative role in a truth condition is not to deny that it plays a truth conditional role; after all, it is a constituent of the truth condition and so it plays a role in the determination of the truth values of utterances. 16 One candidate might be the quasi-singular proposition, introduced in Schiffer (1978) and developed in Recanati (1993). However, Recanati points out that the descriptive element — the “mode of presentation” — has the property of “truth-conditional irrelevance” (p. 39). This proposition is introduced in connection with utterances containing referential terms, though, and so one could regard the mode of presentation as relevant when these propositions are used to model utterances embedding DA uses. However, it’s not clear that this wouldn’t simply be a notational variant of (6) and so subject to criticism on the same grounds. 17 These conditions are modeled on conditions set down in Evans (1982), p. 322. 18 Wilson (1991), p. 360. 19 McKinsey (1979), p. 83. See also Burge (1974), p. 215-216. 32 20 For discussion of these intentions, see Grice (1989), pp. 30-31, and Neale (1990), pp. 82, 89-90 21 In what follows, I use the pragmatic framework developed in Chs. 2, 3, 4, 7, and 13 in Grice (1989). 22 See Grice (1989), pp. 24ff. 23 Ibid., pp. 37-40. 24 For an account of these conventional associations, see Perry (1993), Ch. 13, esp. pp. 290-296. See also Frege (1892), p. 64, where he mentions "the step from the level of thoughts to the level of reference" in discussing judgments. 25 Grice (1989), p. 37. 26 Ibid., pp. 31 & 39-40. 27 ‘P!(x)’ stands for ‘x is the next President of the United States’ and ‘H(x)’ stands for ‘x is here.’ 28 See Neale (1990) and Kronfeld (1993) for additional derivations of this sort. 29 Cf. Kripke (1979), p. 15. 30 Cf. Donnellan (1966), p. 293. 33