[to appear in On Descriptions: Semantic and Pragmatic Perspectives,
Anne Bezuidenhout & Marga Reimer, eds., Oxford University Press]

Descriptions: Points of Reference





I. Referring and Describing

Point 1.    Names and descriptions can both be used merely to indicate what we are speaking about.

Point 2.    We generally choose the least informative sort of expression whose use will still enable the hearer to identify the individual we wish to refer to.

Point 3.    In many cases, using a description may be the only way to refer to something.

Point 4.    Using a description to refer identifies by implicitly conveying an identity.

Point 5.    The distinctive quantificational character of definite descriptions helps explain how and why they can readily be used to refer, because it plays a key role in their referential use.


II. Referring by Describing

Point 6.    With a specific use of an indefinite description, one is not referring but merely alluding to something.

Point 7.    One can describe a (singular) proposition without being in a position to grasp it.

Point 8.     One can describe (single out) an individual without actually referring to it.

Point 9.    Descriptive 'reference' is not genuine reference.

Point 10.   Various kinds of pseudo-reference are not to be confused for the real thing.


III. The Pragmatic Status of the Referential-Attributive Distinction

Point 11.   Various contrasts may seem to capture the referential-attributive distinction, but only one really does.

Point 12.   Incomplete definite descriptions pose no real threat to Russell's theory.

Point 13.   Definite descriptions do not have referential meanings.





Russell (1905) made a compelling case that descriptions, definite as well as indefinite, are devices of quantification, not referring phrases. Strawson (1950) and Donnellan (1966) pointed out that definite descriptions can be used to refer. And even indefinite descriptions can be used to refer. All this is old hat. If 'On Denoting', 'On Referring', and 'Reference and Definite Descriptions' had not provoked decades of debate, philosophers might have just thought it obvious that the mere fact that an expression can be used to refer does not show that it is inherently a referring expression, an expression that itself refers. After all, there is an obvious need for a distinction between linguistic meaning and speaker's meaning, and the distinction between linguistic reference and speaker's reference is just a special case of that (Kripke 1977, 263). Invoking this distinction does not, of course, tell us which sorts of expressions are inherently referring expressions and which are merely capable of being used to refer, but it is enough to suggest that a special reason is needed to support the claim that descriptions, which obviously can be used in nonreferring ways and which have the syntactic earmarks of quantifier phrases, nevertheless have referential readings.

        Consider the sentence, 'The discoverer of X-rays was bald'. It is one thing for a speaker to be able to use the sentence to convey the singular proposition that R–ntgen was bald and quite another for the sentence itself to express that proposition. The mere fact that descriptions can be used to refer does not provide much support for the claim that when so used they are semantically referential. After all, there are many things expressions can be used to do that have non-semantic explanations. For example, so-called rhetorical questions ('Why are you so lazy?') are statements made using interrogative sentences. No one would seriously suggest that the interrogative form has an additional declarative meaning. Indeed, it is because a rhetorical question involves the utterance of an interrogative sentence whose semantics accounts for its literal use to ask a genuine question that it has its force: it is taken as a statement and is intended by the speaker to be taken as such, precisely because the answer to the question is obvious. Various strong pragmatic arguments have challenged the claim that definite and even indefinite descriptions sometimes function semantically as referring expressions.[1] Indeed, Jeff King (2001) has recently made a compelling case that the best candidate for referential descriptions, namely demonstrative descriptions (of the form 'that F'), are quantificational, even in their paradigmatic referential use. The key points of Part I combine to provide new reason for supposing that definite descriptions are not semantically referential. The main point is that they can readily be used to refer precisely because they are quantifier phrases of a certain sort.

        Part II addresses the question of what it takes to use a description to refer. One can use a description to refer to something one believes to satisfy the description, but what if one is not in a position to think of that individual, if, as Russell would say, one 'knows it only by description'? Can one refer to it anyway? If descriptions are quantificational and propositions expressed by sentences containing them are general, how can one use such a sentence to convey a singular proposition involving whichever individual satisfies the description in question? Also, there is the intermediate case (between referential and merely quantificational uses) of the so-called specific use of indefinite descriptions. It illustrates one way to fall short of referring to an object (others will be discussed too): one indicates that one has a certain individual in mind without indicating which individual it is. In short, one merely alludes to it.

          Part III revisits Donnellan's referential-attributive distinction in light of the preceding points. I will review various ways in which it has been characterized, by Donnellan himself and by others, and try to pin down what the distinction amounts to. Also, I will take up the apparent problem for Russell's theory posed by incomplete definite descriptions, whether used referentially or attributively. These are descriptions, such as 'the table', that are not uniquely satisfied. There are too many tables (more than one is enough) for a sentence like 'the table is covered with books' to be made true in the way that Russell's theory requires. Finally, I will rebut Michael Devitt's new arguments (this volume) that descriptions have referential meanings.

        Addressing these many issues requires taking a position on what it involves to refer to a particular thing (our discussion will be limited to reference to spatio-temporal things) and on what it takes to think of one. I will assume a certain conception of singular reference, either by a linguistic expression or by a speaker.[2] When reference is made by the noun phrase itself, indicative sentences in which it occurs express singular propositions, at least when they express a proposition at all.[3] The referent of the expression is a constituent of that proposition. When a speaker refers, he uses a noun phrase to indicate which thing he is trying to convey a singular proposition about. He can use a noun phrase to refer his audience to something even if the noun phrase does not itself refer.

        I also assume a certain conception of singular thought. In my view, we can have singular thoughts about objects we are perceiving, have perceived, or have been informed of (Bach 1987/1994, ch. 1). We do so by means of non-descriptive, 'de re modes of presentation', which connect us, whether immediately or remotely, to an object. The connection is causal-historical, but the connection involves a chain of representations originating with a perception of the object. Which object one is thinking of is determined relationally, not satisfactionally. That is, the object one's thought is about is determined not by satisfying a certain description but by being in a certain relation to that very thought (token). We cannot form a singular thought about an individual we can 'think of' only under a description. So, for example, we cannot think of the first child born in the 22nd century because we are not suitably connected to that individual. Strictly speaking, we cannot think of it but merely that there will exist a unique individual of a certain sort. Our thought 'about' that child is general in character, not singular. Similarly, we cannot think of the first child born in the fourth century BC either. However, we can think of Aristotle, because we are connected to him through a long chain of communication. We can think of him even though we could not have recognized him, just as I can think of the masked man I recently saw rob my bank. Nor does being able to think of an individual require being able to identify that individual by means of a uniquely characterizing description.[4]

        So on my conception of singular thought, there must be a representational connection, however remote and many-linked, between thought and object. A more restrictive view, though not nearly as restrictive as Russell's, would limit this connection to personal acquaintance (via perception and perception-based memory), and would disallow singular thoughts about unfamiliar objects. A more liberal view, though one I would contest, would allow singular thought via uniquely identifying descriptions. In any case, although I am assuming the above conception of singular thought, the questions to be asked and the distinctions to be drawn, such as the distinction between referring to something and merely alluding to or merely singling out something, do not essentially depend on that conception (of course, how one uses these distinctions to divide cases does depend on one's conception). All that is required is the assumption that one can have singular thoughts about at least some objects one has not perceived and that only certain sorts of relations one can bear to an object put one in a position to have singular thoughts about it. What is essential to our questions is the above conception of singular reference: speaker's reference and the hearer's full understanding of it can be achieved only by way of a singular thought of the object.


I. Referring and Describing           

Descriptions can be used to refer. What does that involve? Quite a bit. What does it show the semantics of descriptions? Not much.


Point 1. Names and descriptions can both be used merely to indicate what we are speaking about.

We commonly talk about particular persons, places, or things.[5] We refer to them and ascribe properties to them. In so doing, we are able to accommodate the fact that an individual can change over time, that our conception of it can also change over time, that we can be mistaken in our conception of it, and that different people's conceptions of the same individual can differ. All this is possible if in thinking of and in referring to an individual we are not constrained to represent it as having certain properties.

        It has long been thought that proper names, along with pronouns, are the linguistic devices best suited for doing this.[6] As Mill wrote, the function of proper names is not to convey general information but rather 'to enable individuals to be made the subject of discourse'; names are 'attached to the objects themselves, and are not dependent on ... any attribute of the object' (1872, 20). According to Russell, a proper name, at least when 'used directly', serves 'merely to indicate what we are speaking about; [the name] is no part of the fact asserted ... : it is merely part of the symbolism by which we express our thought' (1919, 175).[7] Russell treated definite descriptions differently, of course, since the object a description describes 'is not part of the proposition [expressed by the sentence] in which [the description] occurs' (170). Whereas a (genuine) name introduces its referent into the proposition, a description introduces a certain quantificational structure, not its denotation. The denotation of a description is thus semantically inert -- the semantic role of a description does not depend on what, if anything, it denotes. In contrast, a proper name 'directly designat[es] an individual which is its meaning' (174).[8] Nevertheless, Russell allowed that proper names can not only be 'used as names' but also 'as descriptions', adding that 'there is nothing in the phraseology to show whether they are being used in this way or as names' (175).

        Interestingly, Russell's distinction regarding uses of names is much the same as Donnellan's famous distinction regarding uses of definite descriptions (see Part III). If the property expressed by the description's matrix (the 'F' in 'the F') enters 'essentially' into the statement made, the description is used attributively;[9] when a speaker uses a description referentially, the speaker uses it 'to enable his audience to pick out whom or what he is talking about and states something about that person or thing' (1966, 285). Donnellan's distinction clearly corresponds to Russell's. Whereas an attributive use of a definite description involves stating a general proposition, as with the use of a proper name 'as a description', a referential use involves stating a singular proposition, just as when a proper name is used 'as a name'. And just as Russell comments that 'there is nothing in the phraseology' to indicate in which way a name is being used, so Donnellan observes that 'a definite description occurring in one and the same sentence may, on different occasions of its use, function in either way' (281).

        If Russell and Donnellan are right, respectively, about proper names and definite descriptions, then expressions of both sorts can be used referentially (as a name, to indicate what we are speaking about) or attributively (as a description). This leaves open whether either sort of expression is semantically ambiguous or whether, in each case, one use corresponds to the semantics of the expression and the other use is accountable pragmatically from that use.[10] For Russell a definite description, whichever way it is used, is inherently a quantifier phrase, whereas a 'logically proper' name is a referring term.[11] Donnellan was evidently unsure whether to regard the referential-attributive distinction as indicating a semantic ambiguity or merely a pragmatic one.[12] But at least this much is clear: regardless of the semantic status of the distinction between using a description referentially (as a name) or attributively (as a description), there is a definite difference between asserting a singular proposition and asserting a general one.[13]


Point 2. We generally choose the least informative sort of expression whose use will still enable the hearer to identify the individual we wish to refer to.

Suppose you want to refer to your mother-in-law. In some circumstances, it may be enough to use the pronoun 'she'. The only semantic constraint on what 'she' can be used to refer to is that the referent be female (ships and countries excepted). So its use provides only the information that the intended referent is female. If it is to be used successfully to refer the hearer to a certain female, there must be some female that the hearer can reasonably suppose the speaker intends to be referring to. If out of the blue you said, 'She is unforgettable', intending with 'she' to refer to your beloved mother-in-law, you could not reasonably expect to be taken to be referring to her. However, if she were already salient, by being visually present and prominent or by having just been mentioned, or you made her salient in some way, say by pointing to a picture of her, then using 'she' would suffice. In other circumstances, you would have to use some more elaborate expression. For example, to distinguish her from other women in a group you could use 'that woman', with stress on 'that' and an accompanying demonstration. Or, assuming the hearer knows her by name, you could refer to her by name. Otherwise, you would have to use a definite description, say 'my beloved mother-in-law'.

        This example suggests that a speaker, in choosing an expression to use to refer the hearer to the individual he has in mind, is in effect answering the following question: given the circumstances of utterance, the history and direction of the conversation, and the mutual knowledge between me and my audience, how informative an expression do I need to use to enable them to identify the individual I have in mind? Note that informativeness here can depend not only the semantic information encoded by the expression but on the information carried by the fact that it is being used.

        Some linguists have suggested that which sort of expression one must use depends on the degree of 'givenness' (or 'familiarity' or 'accessibility') of the intended referent. Giving my own gloss (in parentheses) on the well-known scale proposed by Gundel, Hedberg, and Zacharski (1993), we can distinguish being:

+         in focus (being the unique item under discussion or current center of mutual attention)

+         activated (being an item under discussion or being an object of mutual awareness)

+         familiar (being mutually known)

+         uniquely identifiable (satisfying a definite description)[14]

Notice that in the first two cases the referent is already an object of mutual awareness, that in the third case it is at least mutually known, and that only in the last case may it be something unfamiliar to the hearer. So we should keep in mind the distinction between keeping the hearer's attention on something he is already attending to, calling his attention to something he is already familiar with, and bringing to his attention something unfamiliar to him.

        Gundel et al. plausibly claim that correlated with each status is a type of expression most appropriately used to refer to items with that status:

+         in focus: unstressed pronominals (and, in some languages, zero pronominals)

+         activated: stressed third-person pronouns and simple demonstratives

+         familiar: demonstrative phrases

+         uniquely identifiable: definite descriptions

For some reason Gundel et al. do not discuss first- and second-person pronouns and proper names. The reference of singular first- and second-person pronouns is determined by linguistic role (the speaker, the hearer), but with 'we' and plural 'you' there must be a contextually identifiable constraint on the relevant group. As for proper names, generally it is appropriate to use them only if one's audience is familiar with the name and knows its bearer by name. Otherwise an introduction is in order.

        Now Gundel et al. suggest that different degrees of givenness are not merely associated with but, as a matter of linguistic convention, are encoded by different types of referring expressions. Perhaps they suggest this because, taking their scale to concern the cognitive status of representations in the mind of the hearer, they think this status has to be linguistically marked if it is to play a cognitive role. As I see it, however, this scale concerns the mutual (between speaker and hearer) cognitive status of the intended referent. After all, in using an expression to refer the speaker aims to ensure that the hearer thinks of the very object the speaker is thinking of, and what matters is that the expression used to refer, and the fact that the speaker is using it, provide the hearer with enough information to figure out what he is intended to take the speaker to be thinking of, hence to think of it himself. This is why there is a parsimonious alternative to Gundel et al.'s conventionalist view: the different degrees of givenness associated with different types of referring expressions are not encoded at all; rather, the correlation is a by-product of the interaction between semantic information that is encoded by these expressions and general facts about rational communication.[15] On this, the null hypothesis, it is because different expressions are more or less informative that the things they can be used to refer to are less or more given or accessible. That is, the more accessible the referent is, the less information needs to be carried by the expression used to refer to it to enable the hearer to identify it.

        Notice that not only is it enough to use the least informative sort of expression needed to enable your audience to identify the individual you have in mind, it is misleading to use a more informative one. For example, in telling a story about a particular individual, it is always sufficient, once the individual is introduced, to use a personal pronoun -- provided, of course, that no other individual of the same gender has been introduced in the meantime. There are stylistic or other literary reasons to use their name or a definite description every so often, but unless it is obvious that this is the name or a description of the individual in question, it would be inferred that reference is being made to some other individual. This inference would be made on the charitable assumption that one is not being needlessly informative (and violating Grice's (1989, 26) second maxim of quantity).


Point 3. In many cases, using a description may be the only way to refer to something.

Suppose you want to refer to some thing (or someone). Suppose it is not perceptually present, has not just come up in the conversation, and is not otherwise salient. Suppose that it does not have a name or that you are unaware of its name or think your audience is unaware. Then you cannot use an indexical, a demonstrative (pronoun or phrase), or a proper name to refer to it. If you want to refer to it, what are you going to do? Unless you can find it or a picture of it to point to, you need to use a linguistic expression, some sort of singular noun phrase (what else?), to call it to your audience's attention. You must choose one that will provide your audience with enough information to figure out, partly on the supposition that you intend them to figure this out, which object you're talking about. Your only recourse is to use a description.

        This raises the question, when you use a description, how does your audience know that you are referring to something and expressing a singular proposition, rather than making a general statement and expressing an existential or a uniqueness proposition? Although the presence of a description does not signal that you are referring -- semantically, descriptions are not referring expressions -- what you are saying might not be the sort of thing that you could assert on general grounds, that is, as not based on knowledge of some particular individual (see Ludlow and Neale 1991). This will certainly be true whenever it is mutually evident which individual satisfies the description in question and what is being said regarding the individual that satisfies the description can only be supposed to be based on evidence about that individual. For example, if my wife says to me, 'The DVD player is broken', I can't not take her to be talking about the actual DVD player of ours. On the other hand, if before we decided on a DVD player she said, 'The DVD player had better not cost more than $500', clearly she would be making a general statement pertaining to whichever DVD player we buy (notice that the corresponding demonstrative description, 'that DVD player', is usable only in the latter, non-referential case). Also, its being mutually evident which individual satisfies a description will generally be sufficient for a referential use, since there will usually be no reason for the hearer not to be taken as making a singular statement about that individual. This applies especially to descriptions of occupiers of social positions or practical roles, such as 'the boss' or 'the freezer'. Moreover, if the description is incomplete, as in these cases, and there is no mutually salient or obviously distinctive completion in sight, then the hearer, at least if he is mutually familiar with the boss or freezer in question, can only take the description as being used referentially.[16] But if 'the F' is incomplete and it is obvious that the hearer is unfamiliar with the relevant F, then a (referential) use of 'the F' must be preceded by an introduction of the relevant F.


Point 4. Using a description to refer identifies by implicitly conveying an identity.

In using a noun phrase to refer to a certain individual, you aim to do two things: to get your audience to think of that individual and to take that individual to be the one you are thinking of, hence the one you are referring to. As per Point 3, if you do not have a name at your disposal for what you wish to refer to, and an indexical or a demonstrative pronoun or phrase won't do the trick, the only available sort of noun phrase left is a description. You must describe what you have in mind if you are to get your audience to think of it too and to take you to be referring to it. You do this by exploiting a presumed identity between the individual you wish to refer to (thought of by means of the mental counterpart of a singular term) and that which satisfies the description (perhaps as contextually restricted -- see Point 12). To appreciate how this works, let's take the perspective of the audience and ask what is involved in recognizing the intention behind a referential use. We should keep in mind that recognizing a referential intention is part of recognizing a communicative intention and that this involves looking for a plausible explanation for the fact that speaker said what he said, partly on the Gricean basis that he intends one to do so (see Bach & Harnish 1979, esp. 4-8, 12-15, and 89-93).

        According to Russell's theory, a sentence of the form 'The F is G' expresses a general (uniqueness) proposition. Then if you utter such a sentence but use the description referentially, what you say is a general proposition but what you mean is a singular one.[17] But how and why does the hearer takes you to be doing that? So, for example, if you uttered, 'The mailman is dangerous', I would take you to be asserting not a general proposition but a singular one (about the mailman, i.e., the neighborhood mailman, our mailman). Why would I do that? Well, I am acquainted with the mailman and presumably so are you. Besides, that a certain individual is the mailman has nothing to do with his being dangerous. To suppose that it does would be to take you to be stating something for which you have no evidence (you would be violating Grice's (1989, 27) second maxim of quality). So I have no reason to suppose, as if you were unfamiliar with the mailman, that you are making a general statement, the content of which is independent of which man is the mailman. It is unreasonable for me to suppose that you do not have in mind the particular man who delivers mail in our area and that you are not basing the belief you are expressing on information or evidence derived from him. So I have positive reason to think that you have in mind, and intend me to think you have in mind, a certain individual who satisfies the description you are using. Now for me to think you intend this requires my understanding what you are saying, namely, that the mailman is dangerous, but that is a general proposition. So in attributing to you the intention to be making a singular statement, I am in effect explaining why you said what you said, and thereby able to infer what you meant.

        If you are using the description to refer and I am taking you to be doing so, we must have ways of thinking of the individual in question, the mailman, in some other way than as the mailman. Presumably we both remember him by way of a memory image derived from seeing him. In thinking of him via that image, you take him to be the mailman and use the description 'the mailman' to identify him for me, which triggers my memory of him. We both think of him, via our respective memories of him, as being the mailman. This fits with how Mill describes the functioning of a proper name in thought as an 'unmeaning mark which we connect in our minds with the idea of the object, in order that whenever the mark meets our eyes or occurs to our thoughts, we may think of that individual object' (1972, 22). Though not 'unmeaning', a definite description can play a similar role.

        In using a description referentially, you are using it in lieu of a sign for the object. You are not making a general statement about whatever satisfies the description but a singular statement about some unnamed, undemonstrated, and otherwise unsalient individual. You are thinking of a certain individual by means of the mental counterpart of a singular term but there is no suitable linguistic singular term for you to use. In using 'the F' (in 'the F is G') you are implicitly indicating that this individual is the F and that it is this individual that you are stating to be G. What matters to your assertion is that this individual is G, not that it is (the) F. So the content of the description is inessential to what you are stating (though not to what you are saying). In effect, you are using 'the F' to instruct the hearer to think of, and take you to be talking about the individual that is the F.


Point 5. The distinctive quantificational character of definite descriptions helps explain how and why they can readily be used to refer, because it plays a key role in their referential use.

So far as I know, no one has ever argued that the mere fact that a quantifier phrase can be used to refer shows that it semantically refers, hence that it has a referential as well as a quantificational meaning. No one has ever argued that because, say, 'three thugs' might to used (say in 'Three thugs are outside') to refer to the three thugs that the speaker knows the hearer is expecting, this phrase has a referential reading. So why suppose that descriptions have such readings? Why isn't it enough simply to appeal the distinction between what a speaker says and what a speaker means and, in particular, that between linguistic reference and speaker's reference are taken into account? For one thing, as Devitt (this volume) points out, unlike other sorts of quantifier phrases that can be used to refer, definite descriptions are regularly are so used. This is one of his reasons, to be rebutted in Point 13, for attributing referential meanings to definite descriptions. However, as I will now suggest, the distinctive quantificational character of singular definite descriptions actually helps account for the fact that, despite being quantifier phrases, they can readily be used to refer. They are quantifier phrases of a distinctive kind and, moreover, they imply uniqueness (besides, as Point 3 indicated, sometimes there is no singular noun phrase of any other sort suitable or even available to use to refer to the intended object).

        Definite descriptions are quantifier phrases of a special sort. In general, sentences of the form 'Q F is G' or 'Q Fs are G' say how many Fs are G. That is, they express propositions whose truth requires that the set of Gs include a set of Fs of the size specified (precisely or vaguely), by 'Q' ('one F', 'ten Fs', 'a few Fs'), and in some cases a set of no greater size ('exactly one F', 'exactly ten Fs,' 'few Fs'). Any set of Fs of the right size will do (universal quantifiers are a degenerate case, because there is only one such set). The quantifier phrase 'Q F(s)' indicates how many Fs have to be G for the sentence to be true. Definite quantifier phrases do more than that: they indicate how many Fs there are. In 'The nine planets have elliptical orbits', 'the nine planets' indicates how many planets there are (nine) as well as how many must have elliptical orbits for the sentence to be true. In 'Both authors of Principia Mathematica were English', 'both authors of Principia Mathematica' indicates how many authors of Principia Mathematica there were (two), in 'the inventor of television died in poverty', 'the inventor of television' indicates how many inventors of television there were (one), etc. Even the vague 'the few surviving condors' and 'the many Enron employees' fall, in a vague way, into this category.[18]

        When using a definite quantifier phrase, one is not just saying how many Fs are G, one is effectively saying (or at least implying) how many Fs there are. When a sentence of the form 'Q F is G' or 'Q Fs are G' contains such a quantifier phrase, there is one particular individual or set of individuals to which the predicate ascribes a property. This individual or set of individuals does not enter into the proposition expressed by the sentence -- the proposition is general, not singular -- but this proposition is made true or false by facts involving this individual or set of individuals. Accordingly, the question can arise of which individual(s) this is. The speaker may or may not have any idea which one(s) this is, but there is a determinate one (or set) to be thought of. In this way the possibility of referring can arise without any special stage setting.

        Now singular definite descriptions are a special case, for they imply uniqueness.[19] Notice, however, that this uniqueness is not encoded in the meaning of 'the'. As Richard Sharvy observed, recognizing that singular definite descriptions have something in common with plural ones, 'the primary use of 'the' is ... to indicate totality, implication of uniqueness is a side effect' (1980, 623). That is, it is a kind of universal quantifier, and the side effect is the result of combining 'the' with a nominal in the singular. Sometimes the nominal itself implies uniqueness, as with descriptions containing superlatives ('the shortest spy') or indicative of unique positions ('the queen of England') or accomplishments ('the inventor of the Segway'). When it does not, you may need to use 'the one F', 'the only F', or even, to forestall any suspicion that there is more than one F, 'the one and only F'. Nevertheless, 'one' and 'only' are redundant in these phrases, since uniqueness is already implied by the combination of singularity and totality. Just as 'both Fs' is equivalent to (the ungrammatical) 'all two Fs', so 'the F' is equivalent to (the ungrammatical) 'all one F'.[20]

        Uniqueness is implied even if 'the F' is incomplete, that is, even if there is more than one F. This does not show that the description contains some hidden modifier that would make it complete or some phantom variable of domain restriction. The situation is much simpler. As Point 12 explains, in using an incomplete definite description, one is not using the description is a strictly literal way, even though one is using 'F' literally. What one means is distinct from what one says -- it goes beyond the semantic content of the sentence one is uttering -- since one is not making fully explicit what one means. If one is using the description attributively, what one means is expressible by some more elaborate description, e.g., of the form 'the F that is G'. If one is using it referentially, one is referring to a certain F, an F that is salient or contextually relevant. One is not using 'the F' to mean 'the salient F', 'the contextually relevant F', 'the F I have in mind', or anything of the sort. Rather, it is as if there is only one F, at least for the present purposes of the conversation. In effect, one is pretending that there is only one F, namely the F one intends the listener to think of and to take one to have in mind.

        That an object is the unique F (or the only salient or contextually relevant F) does not alone explain how 'the F' can be used to refer to it. There is also the fact that any connection between the properties being the unique F and being G is irrelevant to what you are trying to convey. Consider cases in which there is a connection, e.g., 'The next pet I get will be a cat', said long before you have decided on a pet, or 'The next president will be a Democrat', said before anyone has any idea which Democrats will even be running. In cases like these you could also have included 'whatever it is' or 'whoever he is' to indicate that you don't know which individual satisfies the description. In such cases, your audience does not have to identify some individual as the F. In using a description referentially, however, you intend your audience to rely on the fact that you uttered the description as a basis for figuring out which thing you are talking about, by way of identifying what satisfies that description. To recognize that you are doing this, the audience has to realize that the property of being F has in itself no direct bearing on what you are trying to convey. They have to recognize that you are using 'the F' to enable them to think of a certain individual and to instruct them to take you to be talking about a certain individual that is the F.

        In sum, the quantificational character of a definite description plays a role in its referential use. The speaker thinks of a certain object, takes that object to be the F, and uses 'the F' to refer to it. The speaker, on hearing 'the F', thinks of a certain object that he takes to be the F, and takes that to be what the speaker is referring him to. His thinking of that object may depend in part on the fact that it is the only plausible candidate for what the speaker has in mind in using 'the F'. Moreover, the property of being F plays an essential role in this process. Indeed, if the speaker uses 'the F' refers to a certain object that isn't F, he is speaking falsely or not literally. Or he could be exploiting the hearer's false belief that the object is F. In any case, if he uses 'the F' to refer to an object that isn't F, then even if what he means is true, what he says is false -- unless the actual F has the ascribed property. So consider this dialogue between two people who watched TV one morning in May of 2002, one of whom has confused two different gesticulating curly-haired guests:

                  A:      The inventor the Segway, that young guy we saw on the

                            Today Show this morning, is very clever.

                  B:      No, that was Jason Stanley. Dean Kamen is the inventor the

                          Segway, he's older, and he was on Good Morning America.

                  A:      Oh yeah, Dean Kamen. Well, like I said, the inventor the Segway

                            is very clever.

Both of A's uses of 'the inventor of the Segway' are referential, first to Stanley and then to Kamen. He meant two different things, even if both are true, but what he said was the same each time, a general proposition made true or false, but not about, Dean Kamen. In both cases he conveyed an identity (Point 4), first a false one, which B recognized as such, and then a true one. In both cases the quantificational character of the description was operative.


II. Referring by Describing

So far I have just assumed that to refer to something one must be in a position to have singular thoughts about it and that the propositions one attempts to communicate in the course of referring to it are singular with respect to it (I have assumed also that being in a position to have a thought about a particular (spatio-temporal) thing requires being connected to that thing, via perception, memory, or communication). The question to be taken up now concerns whether it is necessary to be in such a position in order to refer to something. Put it this way: can one communicate a singular proposition about something one is not in a position to have singular thoughts about?

        The following points address different aspects of this question and certain issues underlying it. They illustrate various ways in which uses of descriptions, indefinite as well as definite, can come close to being referential but fall short. As we have seen, in using a description 'the F' referentially (in uttering a sentence of the form 'The F is G'), one is trying to convey that a certain thing, to be identified as the F, is G. One is communicating the singular proposition that this thing is G, so that understanding one's utterance requires grasping that proposition. In order to appreciate how seemingly referential uses of definite descriptions can fall short of actually being referential, we will first take up the case of specific uses of indefinite descriptions, a phenomenon of interest in its own right.


Point 6. With a specific use of an indefinite description, one is not referring but merely alluding to something.

Indefinite descriptions can be used nonspecifically, referentially, or specifically.[21] In the very common nonspecific (or purely quantificational) use, there is no indication that the speaker has any particular thing in mind, as with a likely utterance of (1),

                (1)     A martini will cheer you up.

This is a clearly quantificational use of an indefinite description. No singular proposition is being conveyed or even in the offing.

        As for the referential use, it is relatively rare, as it is with quantificational phrases generally. Consider an utterance of (2), for example, made while watching a well-known philosopher on TV discussing how the Fed should control inflation.

                (2)     A philosopher thinks he's an economist.

It is obvious that the speaker is not making a general statement, about no philosopher in particular. He is clearly talking about the philosopher he is seeing on TV. Perhaps he is implicating that there is something incongruous about a philosopher pontificating on the economy, but it is clear that he is talking about this particular philosopher.

        What is distinctive about the specific use of an indefinite description is that the speaker communicates that he has a certain individual in mind, but he is not communicating which individual that is -- he doesn't intend you to identify it. Suppose someone says,

                (3)     A famous actress will be visiting us today.

Unless he thinks this is the sort of day for a visit by a famous actress, presumably he has a particular one in mind (he could have made this clear by including the word 'certain' (or 'particular'), as in:

                (3')     A certain famous actress will be visiting us today.

He could even explain why he is not specifying which actress it is:

                (3a)     A famous actress will be visiting us today, but I won't tell you who.

                (3b)     A famous actress will be visiting us today, but I want her to be a surprise.

In a specific use, the speaker indicates that he is in a position to refer to a certain individual, but is not actually doing so. He is not identifying or trying to enable the hearer to identify that individual -- he is merely alluding to her. He has a certain a singular proposition in mind but is not trying to convey it. So what must the hearer do in order to understand the utterance? It would seem that she must merely recognize that the speaker has some singular proposition in mind, about a certain individual of the mentioned sort, in this case a famous actress.[22]

        It might be objected that a specific use of an indefinite description is a limiting case of a referential use, not mere allusion but what might be called 'unspecified' reference. After all, can't the hearer, recognizing that the speaker has some individual in mind, at least think of that individual under the description 'the individual the speaker has in mind'? But, as Point 9 will say, descriptive 'reference' is not genuine reference. Besides, the speaker is not really referring the hearer to that individual and, in particular, does not intend her to think of the individual he has in mind under the description 'the individual you (the speaker) have in mind' or in any similar way. He is merely indicating that he has a certain unspecified individual in mind. That is, he is not referring but merely alluding to that individual.

        To appreciate why this is, consider a situation in which the speaker has one F in mind among many and proceeds to say something not true of that individual. Suppose a teacher has a particular student in mind when she utters (4),

                (4)     A student in my ethics class threatened me yesterday.

but does not specify which student (and does not expect that to be evident). Obviously the words 'a student in my ethics class' do not refer to the student she had in mind, for if some student threatened her but she is mistaken about which one, (4) would still be true. So (4) expresses a general proposition. Even so, since she does have a certain student in mind, could she be using this indefinite description be refer to that student? Even if what she said is a general proposition, is what she meant a singular proposition, about the student she had in mind? No, because the audience could understand her perfectly well without having any idea which student she has in mind. They understand merely that she has a certain student in mind, the one she is alluding to. Indicating that one has a certain unspecified individual in mind does not suffice for referring to that individual. If it did, then any way of coming to know that someone else has some individual in mind would be enough to put one in a position to think of and to refer to that individual oneself.


Point 7. One can describe a (singular) proposition without being in a position to grasp it.

Our discussion of understanding a specific use of an indefinite description illustrates that it is one thing to entertain a singular proposition and another thing merely to know that there exists a certain such proposition. Russell's famous discussion of Bismarck illustrates how this can be. He operates with a notoriously restrictive notion of acquaintance, but this is not really essential to the distinction he is drawing. I agree with Russell that we cannot have singular thoughts about individuals we 'know' only 'by description', but I will not assume that these individuals are limited to those with which we are acquainted in Russell's highly restrictive sense. They include individuals we are perceiving, have perceived, or have been informed of and remember. So although Russell's choice of example (Bismarck) would have to be changed to be made consistent with a much more liberal notion of acquaintance, I will use it to illustrate his distinction.

        Russell contrasts the situation of Bismarck himself, who 'might have used the name ['Bismarck'] directly to designate [himself] ... to ma[k]e a judgment about himself' having himself as a constituent (1917, 209), with our situation in respect to him:

when we make a statement about something known only by description, we often intend to make our statement, not in the form involving the description, but about the actual thing described. That is, when we say anything about Bismarck, we should like, if we could, to make the judgment which Bismarck alone can make, namely, the judgment of which he himself is a constituent. [But] in this we are necessarily defeated. ...What enables us to communicate in spite of the varying descriptions we employ is that we know there is a true proposition concerning the actual Bismarck and that, however we may vary the description (as long as the description is correct), the proposition described is still the same. This proposition, which is described and is known to be true, is what interests us; but we are not acquainted with the proposition itself, and do not know it, though we know it is true. (1917, 210-11)

The proposition that 'interests us' is a singular proposition, but we cannot actually entertain it -- we can know it only by description, that is, by entertaining a general (uniqueness) proposition which, if true, is made true by a fact involving Bismarck. But this general proposition does not itself involve Bismarck, and would be thinkable even if Bismarck never existed.

        The difference in type of proposition is clear from Russell's observations about the statement 'I met a man':

What do I really assert when I assert 'I met a man'? Let us assume, for the moment, that my assertion is true, and that in fact I met Jones. It is clear that what I assert is not 'I met Jones'. I may say 'I met a man, but it was not Jones'; in that case, though I lie, I do not contradict myself, as I should do if when I say I met a man I really mean that I met Jones. It is clear also that the person to whom I am speaking can understand what I say, even if he is a foreigner and has never heard of Jones.

      But we may go further: not only Jones, but no actual man, enters into my statement. This becomes obvious when the statement is false, since there is no more reason why Jones should be supposed to enter into the statement than why anyone else should. Indeed, the statement would remain significant, though it could not possibly be true, even if there were no man at all. (Russell, 1918, 167-8)

The same points apply when a definite description is being used. Imagine that Russell met Whitcomb Judson, the man who invented the zipper. Then if Russell had said, 'I met the inventor of the zipper, but it was not Whitcomb Judson', he would have spoken falsely but would not have contradicted himself. And, although what Russell said in uttering 'I met the inventor of the zipper' would have been made true only in the event that he had met Judson, someone could have understood what Russell said without ever having heard of Judson. For no actual man, Whitcomb Judson or anyone else, entered into Russell's statement. His statement would have been the same even if Sigismond Zipschitz had instead invented the zipper.


Point 8. One can describe (single out) an individual without actually referring to it.

Kaplan suggests that one can use a description to refer to something even if one is not in a position to have a singular thought about it or, as he would say, even if one is not 'en rapport' with it. He asks rhetorically, 'If pointing can be taken as a form of describing, why not take describing as a form of pointing?' (1979, 392). I will explain why not.

        First consider the following example of Kaplan's 'liberality with respect to the introduction of directly referring terms by means of "dthat",' which 'allow[s] an arbitrary definite description to give us the object' (1989a, 560).[23]

                (5)     Dthat [the first child to be born in the 22nd century] will be bald.

 'Dthat' is a directly referential term and, as Kaplan explains in his 'Afterthoughts', 'the content of the associated description is no part of the content of the dthat-term' (1989b, 579); it is 'off the record (i.e., off the content record)' (1989b, 581). So 'dthat' is not merely a rigidifier (like 'actual') but a device of direct reference.[24] It is the actual object (if there is one) uniquely satisfying the description, and not the description itself (i.e., the property expressed by its matrix), that gets into the proposition.[25]

        Not only does Kaplan's 'liberality' impose no constraint on the definite description to which 'dthat' can be applied to yield a directly referring term, it imposes no epistemological constraint on what one can 'directly refer' to.[26] However, even if we concede that any definite description can be turned into a directly referring term, so that a sentence containing the 'dthat' phrase expresses a singular proposition about the actual object (if there is one) that uniquely satisfies the description, it is far from obvious that the user of such a phrase can thereby refer to, and form singular thoughts about, that object. Kaplan seems to think this ability can be created with the stroke of a pen.

        Consider, for example, whether one can refer to the first child born in the 22nd century. Assume that nearly one hundred years from now, this description will be satisfied (uniquely). Then there is a singular proposition involving that individual, as expressed by (5).[27] Without 'dthat' (and the brackets), (5) would express a general (uniqueness) proposition, the one expressed by (5'),

                (5')     The first child to be born in the 22nd century will be bald.

Now can one use the description 'the first child to be born in the 22nd century' referentially, to refer to that child? Kaplan thinks there is nothing to prevent this, that it is a perfectly good example of pointing by means of describing. However, what enables one to form an intention to refer to the individual who happens to satisfy that description? If one is prepared to utter (5') assertively, surely one is prepared to do so without regard to who the actual such child will be -- one's grounds are general, not singular. For example, one might believe that the first child born in the 22nd century is likely to be born in China and that Chinese children born around then will all be bald, thanks to China's unrestrained use of nuclear power. But this only goes to show that one's use of the description is likely to be taken to be attributive. Unless one were known to be a powerful clairvoyant, one could not plausibly be supposed to have singular grounds for making the statement. Nevertheless, Kaplan thinks that one could intend to use the description referentially anyway, as if putting the description in brackets and preceding it with 'dthat' could not only yield a term that refers to whoever actually will be the first child born in the 22nd century but could enable a speaker to refer to that child. However, it seems that one is in the same predicament as the one Russell thought anyone other than Bismarck would be in if he wanted to refer to Bismarck.

        Would it help to have the tacit modal intention of using the description rigidly, or even to insert the word 'actual' in the description?[28] Referring to something involves expressing a singular proposition about it, but rigidifying the description or including the word 'actual' would not make its use referential. Even though the only individual whose properties are relevant to the truth or falsity of the proposition being expressed (even if that proposition is modal) is the actual F (if it exists), still that proposition is general, not singular. This proposition may in some sense be object-dependent, but it is not object-involving. The property of being the actual may enter into the proposition, but the actual F does not.

        The upshot of these observations is that one can use a description to describe or, as I will say, single out something without actually referring to it. The fact that there is something that satisfies a certain definite description does not mean that one can refer to it. If a different individual satisfied the description or you are describing a hypothetical situation in which that would be the case, you would have singled out that individual instead (see Point 9). Nevertheless, you can use the description just as though you were introducing the thing that satisfies it into the discourse. You could, for example, use pronouns to 'refer' back to it. You could say, 'The first child to be born in the 22nd century will be bald. It will be too poor to use Rogaine'. Giving it a name won't help. You could dub this child 'Newman 2' (just as Kaplan dubbed the first child to be born in the twenty-first century 'Newman 1'), but this would not enable you to refer to it or to entertain singular propositions involving it. In this, as Russell might have said, 'we are necessarily defeated'. Even though (5') does not express a singular proposition, the proposition which 'interests us' but which we cannot entertain, one can still use the sentence to describe that proposition.

        It might be objected that in calling this 'singling out' rather than 'referring' to an object, I am not making a substantive claim but am merely engaging in terminological legislation. I would reply that anyone who insists on calling this 'reference' should either show that a singular proposition is expressed or explain why expressing a general, object-independent proposition should be thought to involve reference. One possible reason is taxonomic: if we are to maintain that indexicals and demonstratives are inherently referring expressions and not merely expressions that are often used to refer, allowances must be made for the fact that we sometimes use them to do what I would describe as merely singling out an object, that is, an object that the speaker is not in a position to have singular thoughts about. For example, one could use 'he' or 'that child' to single out the first child born in the 22nd century. But the question is whether this counts as genuine reference. Indeed, one can use such expressions without even singling out an individual, as in, 'If a child eats a radioactive Mars bar, he/that child will be bald'. The mere fact that philosophers are in the habit of calling indexicals and demonstratives 'referring expressions' does not justify doing this.


Point 9. Descriptive 'reference', or singling out, is not genuine reference.

In summing up his account of the referential-attributive distinction, Donnellan concedes that there is a kind of reference, reference in a 'very weak sense', associated with the attributive use of a definite description (1966, 304). Since he is contrasting that use with the referential use, this is something of a token concession. Reference in this very weak sense is too weak to count as genuine reference, for one is 'referring' to whatever happens to satisfy the description, and one would be 'referring' to something else were it to have satisfied the description instead. This is clear in modal contexts, such as in (6):

            (6)     The next president, though probably a man, could be a woman.

The speaker is not asserting of some one possible president that he or she will probably be a man but could be a woman, say if he had a sex-change operation before her inauguration. In this context, the description is understood to fall within the scope of 'could'. The speaker is allowing for different possible presidents, some male, some female, only one of whom will actually be the next president. Surely this is not reference, not even in a very weak sense. Even so, we might wonder whether there is a weak sort of reference that still counts as genuine reference.

        Let's call this 'reference under a description' or simply 'descriptive reference'. Can one use a description to refer to an individual without having some independent, non-descriptional way of thinking of and intending to refer to a certain object? That is, can one intend to convey a singular proposition about a certain individual that one is not in a position to have singular thoughts about? As we saw in Point 8, it won't help to use a rigidified description, say of the form 'the actual F'. This guarantees that no object other than the actual F could be the one on whose properties the truth or falsity of one's statement depends, but it does not yield reference. One may be singling out the object that satisfies the description, hence describing the relevant singular proposition about that object, but one is not referring to that object or expressing that proposition.

        Here is a possible but dubious route toward counting descriptive reference as genuine reference. Philosophers have often referred to singular thoughts as de re thoughts or attitudes. However, they have also used the term 'de re' for a type of attitude ascription (contrasted with 'de dicto' ascriptions). Moreover, de re attitude ascriptions can be true even when no de re (i.e. singular) attitude is being ascribed. So, for example, suppose that even though there is no suspect in the case, the mayor has claimed that Smith's murderer is insane. Smith's murderer, on hearing this, could say, 'The mayor that I am insane', without imputing to the mayor any de re attitudes about him. It is easy to overlook the fact that not all de re attitude ascriptions are ascriptions of de re attitudes because of the structural ambiguity of the phrase 'de re attitude ascription', in which 'de re' can modify either 'attitude' or 'ascription'. Just as there is a difference between 'Tibetan history-teacher' and 'Tibetan-history teacher', so there is a difference between 'de re-attitude ascription' and 'de re attitude-ascription'.[29] A de re attitude-ascription can refer to a certain object without implying that there is a certain object the ascribee has an attitude about. The point carries over to indirect quotation. Having read that the police have said, 'Smith's murderer is insane', Smith's murderer could say to a confidante, 'The police say that I am insane'. In reporting this, Smith is referring to himself. It may seem that he is reporting that the police are referring to him too (with their use of 'Smith's murderer'), but they are not.

        It is sometimes suggested that giving a 'descriptive name' (Evans 1982, 31) to an individual whose identity is in question enables one to refer to that individual. Suppose that Smith's murderer left some vitriolic hip-hop lyrics under a rock at the scene of the crime and the media came to call him 'Rock the Rapper'. Would using this name put people in a better position to refer to Smith's murderer? It would not, not even if sentences containing that name expressed singular propositions. Using such sentences would not enable one to have singular thoughts about Smith's murderer and thereby be in a position to refer to him. Again, propositions cannot be put within one's grasp with just the stroke of a pen.[30]



Point 10. Various kinds of pseudo-reference can be confused for the real thing.

We have already seen several examples of this. Point 6 denied that specific uses of indefinite descriptions are referential. And Points 9 and 10 denied that singling out (descriptively 'referring' to) something count as genuine reference. Singling out whatever uniquely satisfies a certain definite description is not to refer to it. Here I will discuss discourse reference and briefly comment on fictional reference and failed reference, recognizing that each deserves much more extensive discussion.


Reference and discourse 'referents'

It is well-known that unbound pronouns, as well as definite descriptions, can be used anaphorically on indefinite descriptions, as in these examples:

                (7)     Mary met a man today. He/The man was bald.

                (8)     A girl went to yesterday's Giants game, and she/the girl caught a foul ball.

                (9)     If there were a unicorn there, Macdonald would have seen it/the unicorn.

                (10)    Every farmer owns a donkey. He rides it on Sundays.

In (7) and (8), the pronoun (and the definite description) can be used to refer. If the speaker is using 'a man' specifically in uttering (7), he could use 'he' (or 'the man') to refer to the man he thinks Mary met that day. If, on the other hand, the speaker is not in a position to refer to such a man, he could only use 'a man' nonspecifically and could not use 'he' (or 'the man') referringly; the most he could intend to convey is the general proposition that Mary met a man that day who was bald. Similarly, if the speaker of (8) is using 'a girl' specifically, he could use the pronoun 'she' (or 'the girl') to refer to the girl he thought went to the game, but not if he were using 'a girl' nonspecifically. However, it seems that the pronouns (and the definite descriptions) in (9) and (10) cannot be used to refer at all. In (9) 'it' is not being used to refer to an unspecified (and presumably nonexistent) unicorn, and in (10) 'it' functions quantificationally, ranging over the different donkeys owned by the different farmers.

        Despite the fact that the pronouns and the definite descriptions in cases like (7) and (8) need not, and in cases like (9) and (10) cannot, be used to refer, many semanticists have attributed 'discourse referents' to them. I am not suggesting that they seriously believe that discourse referents are real referents, but this only makes it puzzling why they use this locution. Here is how Karttunen (1976) introduced the phrase:

Let us say that the appearance of an indefinite noun phrase establishes a discourse referent just in case it justifies the occurrence of a coreferential pronoun or a definite noun phrase later in the text. ... We maintain that the problem of coreference within a discourse is a linguistic problem and can be studied independently of any general theory of extra-linguistic reference. (Karttunen 1976, 366; my emphasis)

 He explains further,

In simple sentences that do not contain certain quantifier-like expressions,[31] an indefinite NP establishes a discourse referent just in case the sentence is an affirmative assertion. By 'establishes a discourse referent' we meant that there may be a coreferential pronoun or definite noun phrase later in the discourse.[32] Indefinite NPs in Yes-No questions and commands do not establish referents. (383)

So the 'coreferential pronoun or definite noun phrase later in the discourse' can, thanks to the discourse referent 'established' by the indefinite NP, have a discourse referent even if, as in our examples, it is not used to refer.

        The notion of discourse referent has inspired a great deal of theorizing in semantics, including discourse representation theory (DRT), dynamic semantics, and the study of reference accessibility. However, as is implicitly conceded by Karttunen's definition ('[it] can be studied independently of any general theory of extra-linguistic reference'), discourse reference is no more reference than our relation to our perceptual experiences (as opposed to objects perceived) is perception. The basic problem is simply this: a chain of 'reference' isn't a chain of reference unless it is anchored in an actual ('extra-linguistic') referent.

        Despite the widespread use of the phrase 'discourse referent' in some semantic circles, so-called discourse referents are not literally referents. The pronouns in the examples like those we've just considered are not used referringly and, as King (1987) suggested, must be viewed quantificationally. They are used as surrogates for definite descriptions, descriptions which if present in place of the pronouns would not be referential. These pronouns are what Neale (1990, ch. 5) calls 'D-type pronouns'. The basic idea is that the pronoun is used elliptically for a definite description recoverable from the matrix of the antecedent indefinite description (Bach 1987/1994, 258-61). Neale develops a detailed account of how D-type pronouns work in a wide variety of cases. It is essential to this account that the descriptions implicit in the use of D-type pronouns are not construed as referential, even when they are used referentially.

        To see the significance of looking at such pronouns in this way, let us return to two of our examples. Consider (8) again:

                (8)     A girl went to yesterday's Giants game, and she caught a foul ball.

Suppose that the speaker merely heard that a girl had caught a foul ball at yesterday's game. Then he does not have any particular girl in mind and is not using 'a girl' specifically. Even so, it might be thought that 'she' refers to a certain unspecified girl (the 'discourse referent') who went to the Giants game the day before. But how could this be? The first clause of (8) expresses a general proposition, but what about the second clause? Does it express a singular proposition about a certain girl who went to yesterday's Giants game? Suppose this clause is not true, hence that no girl who went to yesterday's Giants game caught a foul ball. Then which girl would the second clause of (8) be about, and what singular proposition would it express? Answer: no girl, and no proposition. On the view that 'she' is referential in the second sentence of (8), that sentence would express a singular proposition if and only if it is true! Surely which proposition a sentence expresses, or that it expresses any, cannot depend on whether or not it is true.

        The situation is similar with a quantified sentence, as in (10),

                (10)     Every farmer owns a donkey. He rides it on Sundays.

Suppose there are farmers who own more than one donkey. In that case, what does it take for the second sentence in (10) to be true? Its truth does not require every farmer to ride on Sundays every donkey that he owns, but also it doesn't require merely that every farmer ride on Sundays one donkey that he owns. So it is not clear what it would take for the second sentence in (10) to be true when there are farmers who own more than one donkey. However, DRT and dynamic semantics have been motivated by the thought that this is clear, and they have tried to make sense of sentences like the second one in (10) in terms of the notion of discourse referent.


Fictional reference

This is far too big a topic to take up in any detail here. Any serious discussion has to distinguish fictional reference (reference in a fiction) and reference (outside the fiction) to fictional entities. Reference in a fiction does not count as genuine reference, at least if it is to fictional persons, places, and things (in a fiction, there can be genuine reference to real persons, places, and things). If Salmon (1998) is right in claiming that fictional entities are real, albeit abstract entities, then we can genuinely refer to them. Otherwise, we can only pretend to. In my view, both fictional reference and reference to fictional entities involve special sorts of speech acts, but there is nothing special about fictional language itself (Bach 1987/1994, 214-18).


Failed reference

A speaker can intend to refer but fail in the (communicative) sense that his listener does not identify the individual he intends her to identify. This is of no special interest here. More interesting is the case in which the speaker intends to refer to a certain individual but there is no such individual. This does not affect any claim being made here, so long as it is understood that in this case there is no singular proposition about that individual and no such proposition for the speaker to convey.

        In such a case the speaker can use an expression to refer but not successfully, since there is no individual being referred to. In such a situation, even if the speaker sincerely utters 'a is F' and means it literally, there is nothing he believes to be the F. Similarly, if he sincerely utters 'the F is G' and is using 'the F' referentially but he fails to have anything in mind, there is nothing he is referring to. In this case, his referential intention cannot be fulfilled, and full communication cannot be achieved. There is nothing for the hearer to identify, and no singular proposition for her to entertain. The best the hearer can do is recognize that the speaker intends to convey a singular proposition of a certain sort. The speaker has the right sort of intention, to be speaking of some particular thing, but there is no thing for him to succeed in referring to.

        A different situation would arise if the speaker merely made as if to refer to something, perhaps to deceive the hearer or perhaps to play along with the hearer's mistaken belief in the existence of something. In this case, although the speaker does not intend to refer to something, he does intend to be taken to be. He can succeed in communicating if he is taken to be referring to the individual the hearer mistakenly believes in. But since there is no such thing, there is no singular proposition to be grasped.


III. The Pragmatic Status of the Referential-Attributive Distinction

We have seen that descriptions can be used to refer not in spite of but partly because they are devices of quantification (of a certain sort) and that there are restrictions on what counts as using them to refer. Various uses of descriptions that may seem to pass as referential fall short of that. Now I will apply certain of the observations made so far in order to clarify the referential-attributive distinction and bolster the case for the pragmatic status of referential uses. In what follows, we should keep in mind Point 3, that frequently one uses a description to refer because no other sort of expression is suitable to refer to what one wishes to talk about. And, as Points 4 and 5 suggested, in using a description to refer one expects the hearer to take one to be relying on information derived from and specific to the particular object that in fact satisfies the description. On the other hand, in using a definite description attributively, one does not purport to have (or at least to be relying on) any information specific to the actual satisfier of the description One is in the same epistemic position that one would be in if something else uniquely satisfied the description.


Point 11. Various contrasts may seem to capture the referential-attributive distinction, but only one really does.

When he introduced the referential-attributive distinction, Donnellan (1966) tried to explicate it in various ways, in each case drawing an apt contrast that intuitively seems to mark the difference. These and a couple of other contrasts are worth reviewing. As we will see, suggestive as these various contrasts are, only one does the job.


A. Being what the speaker has in mind vs. being whatever fits the description

Donnellan suggests that in using a description referentially, your statement is about whatever or whoever you have in mind, whereas with an attributive use your statement is (weakly) about whatever or whoever satisfies the description. This test seems plausible, but it is not as reliable as it sounds.[33] First, it suggests that the speaker can't know who satisfies the description, when of course he can, as Donnellan himself points out. So, for example, you could utter 'Schmidt's editor is insane' and be using 'Schmidt's editor' attributively even though you could know who Schmidt's editor is. For your statement might not depend on this knowledge, say if you made it on the basis of the many strange scrawls on Smith's manuscript. Then what you are stating would not be affected if it turns out that you're mistaken about who Schmidt's editor is.

        Donnellan sometimes describes the relevant contrast as between believing something about 'someone in particular' and about 'someone or other'. However, as he notes, it is possible to think de re of the satisfier of a description even if one is using the description attributively. For example, you might be personally acquainted with Andrew Wiles and yet, in uttering 'The man who proved Fermat's Last Theorem must be a genius', use the description attributively. It is irrelevant to what you mean that you believe Andrew Wiles to be man who proved Fermat's Last Theorem or even that he actually did.

        Also, it is possible to use a description referentially without in a certain sense knowing who satisfies it. In fact, we can distinguish two sorts of knowing who, only one of which is relevant here. There is the context- or interest-relative sort of knowing who, whose slipperiness has been well documented by Boer and Lycan (1986). Not knowing who someone is in this way is compatible knowing who they are for the purpose of referential identification. You can use 'the masked man over there' to refer to a certain masked man, and your listener can know who you are referring to, without either of you knowing who (of the other sort) that man is. Suppose that neither of you could recognize him without his mask or know any distinguishing characteristics of his. Even so, you can refer to him then and there and your listener can know who you are referring to. Although neither of you have any way of identifying him beyond the circumstances of utterance, you both can identify him for purposes of reference.


B. Description inessential vs. description essential

Donnellan suggests that a description used referentially is inessential to what is said, i.e., that the referent's satisfying the description is not necessary for the truth of what is said. Presumably he suggests this because there is a particular individual in question, and because which one it is fixed independently of the property expressed by the matrix of the description. I would agree that the description (i.e., this property) is inessential to what is asserted, but I would insist that it is essential to what is said.

        In the referential case there is something that the speaker intends to refer to even if it doesn't in fact satisfy the description he is using (or, if the description is incomplete, some intended completion of it) and even if some other individual does. But this does not show that the description is inessential to what the speaker says. It shows only that the speaker, in uttering something of the form 'the F is G', could have ascribed the same property (being G) to the same object (the one he takes to be the F) even if that object did not satisfy the description. The description is essential to what he says, because clearly he would have said something different if he had used a different, non-synonymous description. For example, if he utters, 'The man drinking a martini is inane', and the man he is referring to is holding a martini but not drinking it, then what he is saying is true only if some other man is the one drinking a martini and he is inane. If advised that the man he intend to refer to hasn't touched a drop, he might even revise his utterance and say, 'The man holding a martini is inane'.

        The situation is different if the speaker uses a description he knows not to be true of the intended referent but believes that the hearer thinks it is. For example, suppose you say 'The man holding a martini is inane' when you believe that his glass is filled with tequila but assume your audience is unaware of this. In this case, in which you are exploiting your audience's ignorance, you do not mean what you say. However, you intend your audience, in order to identify the referent, to take you to mean what you say.


C. Speaker's referent vs. semantic referent

Kripke (1977) invokes this contrast in the course of arguing against the semantic significance of the referential-attributive distinction. If a description is being used referentially, the speaker's referent may be distinct from the semantic referent (what satisfies the description), whereas with an attributive use they cannot be distinct, since the semantics of the description determines the relevant individual.

        This may seem right, but Kripke's distinction is both misleading and inadequate. It is misleading because definite descriptions, being quantificational phrases, do not have semantic referents at all, at least not in the way in which genuine singular terms do (see note 8). In Russell's jargon, they merely denote (where denoting is a semantically inert property, so that what is denoted does not enter into the proposition expressed). More importantly, as we saw above, definite descriptions can be used attributively even if they are used in a nonliteral way. When so used, the speaker's 'referent' is not the semantic referent.


D. Primary aspect vs. secondary aspect

Searle (1979) suggests that the difference between the referential and the attributive use boils down to whether some description other than the one actually used is primarily relevant to the speaker's utterance. If some other one is primary, then the one actually used expresses merely a 'secondary' aspect under which the speaker is referring. In that case, he has some 'fallback' description in mind, under which the hearer is to think of the referent.

        There are two problems here. First, Searle assumes that one can think of an object only under an description and not in some de re way. He does not seriously consider the possibility that attitudes can be de re. Rather, he insists that only attitude reports can be de re and thinks that the view that attitudes themselves can be de re is based on a confusion. Perhaps he thinks this confusion arises from the structural ambiguity of the phrase 'de re belief report' (see Point 9), but he fails to show that there are no de re attitudes (singular thoughts).

        The second problem is that even if one uses a certain description with a fallback description in mind, one that expresses the primary aspect, one could still be using the former description attributively. One might use 'the F', believing that the F is the H and being prepared to fall back on 'the H'. However, this does not mean that if one had used 'the H' instead, one would have used it referentially. For example, looking at the way a certain car is parked you might utter, 'The owner of this car must have been drunk'. You have no idea who the owner is but you have assumed that the owner of that car was the driver who parked it. Although you're prepared to fall back on the description 'the driver who parked this car here', surely you were not using the original description referentially.


E. 'right' thing' vs. no 'right' thing

Donnellan suggests that 'in the referential use as opposed to the attributive, there is a right thing to be picked out by the audience and its being the right thing is not simply a function of its fitting the description' (1966, 304). Obviously the description could not be merely a device for getting one's audience to pick out the referent unless there were a right thing for them to pick out, something which one wishes to speak about independently of how one is referring to it. That's indisputable, assuming that there is something the speaker intends to refer to, for obviously the reference can succeed only if there is some such thing and only if the hearer identifies it. But there is a sense in which an attributive use also involves a 'right' thing, namely the thing that satisfies the description, though of course this is not what Donnellan means. The point of his contrast here is really that the speaker has some independent way of thinking of, and of intending the hearer to think of, the referent, as something other than the (unique) satisfier of the description being used. This seems right so far as it goes but as it stands, the contrast between there being a 'right' thing and there not being one boils down to the difference between using and not using the description referentially.


F. Particularity vs. generality

This contrast gets to the heart of the referential-attributive distinction. Taking Russell's theory of descriptions to apply only to attributive uses, Donnellan observes that with them 'we introduce an element of generality which ought to be absent if what we are doing is referring to some particular thing'. But

this lack of particularity is absent from the referential use of definite descriptions precisely because the description is here merely a device for getting one's audience to pick out or think of the thing to be spoken about, a device which may serve its function even if the description is incorrect. (1966, 304).

This seems right. After all, in using a description referentially one is using it to refer to something, and that involves conveying a singular proposition. In using a descriptive attributively, one is not referring -- one is conveying a general proposition. Not surprisingly, the referential-attributive distinction is just an instance of the distinction between using a noun phrase to refer and using it quantificationally. The essential difference between a referential and an attributive use of a definite description is that with a referential use one is expressing a singular proposition about a certain object (unless there is no actual referent), whereas with an attributive use one is expressing a general proposition. This is old news, but it is important not to confuse this with any of the other ways we have considered of trying to capture the referential-attributive distinction.

        To sum up. With a referential use, one must be thinking of the F in some de re way, hence in some other way than as the F. In using 'the F' referentially, one is, in effect, instructing the hearer to think of the F, and that requires him also to think of the F in some other (de re) way than as the F. With the attributive use, one need not be in a position to think of the F in any de re way, and even if one does, it is irrelevant. One is simply saying something about whatever is the F, which one need not be in a position to think of in any way at all (or least in any other way than as the F, insofar as that counts as thinking of it at all). In the referential case, it is this other way of thinking of the F that matters. That way of thinking determines the object being referred to, the object that is a constituent of the singular proposition the speaker is trying to convey.

        Notice that most referential uses involve incomplete definite descriptions, descriptions that are insufficiently restrictive to identify a unique individual. The property expressed by the matrix of the description serves merely to distinguish one's intended referent from anything else that, under the circumstances, one might be talking about. So if you say, 'The table is covered with books', using 'the table' is enough to distinguish the one table in the vicinity (or within the attention of you and your audience) from other objects in your presence. If there were more than one table there and it didn't stand out in some blatantly obvious way, using 'the table' would not enable your audience to figure out which individual you are talking about and would be inappropriate -- you'd have to point to the relevant table and use 'that table'.


Point 12. Incomplete definite descriptions pose no real threat to Russell's theory.

Some singular definite descriptions, either by necessity or as a matter of fact, are uniquely satisfied. Most of the ones that we ever have occasion to use are not. Even so, they imply uniqueness. As mentioned earlier, this uniqueness is not encoded in the meaning of 'the' -- 'the' indicates totality, as in plural definite descriptions. The implication of uniqueness is a side effect of combining 'the' with a nominal 'F' in the singular, to yield 'the F'. So what happens when, as is typical, we use a description that is incomplete and we believe it to be? Even if we are using 'the' literally and 'F' literally, we do not mean that there is exactly one F. Rather, we are in effect making a pretense of uniqueness, at least for current conversational purposes. This is a perfectly ordinary phenomenon and no different in kind from various other simplifying pretenses we implicitly make in everyday conversation. For example, we often pretend that the immediate environment or situation is all that is relevant to statements we make. We utter sentences like 'It's not raining now', 'I guess nobody's home', and 'I haven't eaten' as if they expressed much more restricted propositions than the much more far-reaching propositions they literally express. Similarly, we use incomplete definite descriptions, like 'the book' and 'the table', as if there is no other book and no other table than the ones we're talking about. So if, in order to tell someone where to find a certain book, you say 'The book is on the table', for the immediate purpose of the conversation the book and the table in question are the only ones that come into consideration.

        The incompleteness of ordinary definite descriptions poses no real problem for Russell's theory.[34] That a description 'the F' is incomplete does not affect the proposition expressed by any sentence in which it occurs and does not affect what a speaker says in uttering such a sentence; it can only affect what the speaker means. Russell's theory does not have to treat uses of incomplete descriptions as either elliptical for complete ones or as containing implicit but semantically relevant restrictions on their domains. In particular, if a speaker uses an incomplete description referentially, he is attempting to convey a singular proposition involving an individual that is F (assuming he is using 'F' literally). It does not matter that the description is incomplete, because the speaker does not intend the description by itself to provide the hearer with the full basis for identifying the referent. But Russell's theory, being silent about referential uses anyway, does not say that it does. Nor must Russell's theory say that 'the F' is somehow elliptical for 'the salient F', 'the aforementioned F', 'the F under discussion', 'the relevant F', 'the F I have in mind', or anything of the sort, just because it is used to refer to some such F.

        The attributive case may seem to pose more of a problem for Russell's theory. If a speaker is using an incomplete definite description attributively, he intends to convey a uniqueness proposition, just as Russell's theory would predict, but this is not the uniqueness proposition expressed by the sentence he is uttering; it is the one that would be expressed if the description were made complete by being expanded in a suitable way. One apparent problem is that different verbal means of doing this may yield the very same proposition. However, this is not a genuine problem, since what matters is which proposition the speaker intends to convey, not which words he would use to convey it (were he to do so in a fully explicit way).

        A more serious objection, made by Kripke (1977) and by Wettstein (1981), is that there are cases in which the speaker does not have a determinate uniqueness proposition in mind but nevertheless communicates successfully. For this reason, and despite treating the referential-attributive distinction pragmatically, Kripke questions whether a Russellian approach can accommodate the case of incomplete descriptions. Whereas Grice supposed that all this approach needs in order to handle a description like 'the table' is 'suitable provision for unexpressed restrictions', such as 'the table in this room' (1969, 142), Kripke 'doubt[s] that such descriptions can always be regarded as elliptical with some uniquely specifying conditions added' (1977, 255), and finds it 'somewhat tempting to assimilate such descriptions to the corresponding demonstratives (for example, "that table")' (1977, 271). However, this is a temptation to be resisted. Even if definite descriptions can sometimes be used like demonstrative descriptions, they are semantically and syntactically different. And obviously Grice's example of an incomplete description used referentially is not a demonstrative just because it contains one. But the main point is that Russell's theory does not (falsely) entail that if an incomplete description is used referentially, there must be some intended restriction on the description. What matters is the speaker's intention to use the description to refer to a certain particular individual, in the way discussed above, to some salient or contextually relevant F.

        Whereas Kripke's worry concerns referential uses of incomplete definite descriptions, Wettstein's concerns the attributive. The worry is that Russell's theory predicts that if no determinate restriction on the description is intended, the speaker's utterance has no determinate content. But it may well have determinate content, although this content is not, according to Wettstein, of the sort required by Russell's theory.[35] Now Wettstein's examples all involve complex descriptions containing expressions used referentially: 'his murderer', 'the murderer of Smith', and 'the murderer of the man on the couch'. But such examples show merely that a definite description can be relativized, and this, as Neale has pointed out (1990, 99-100), is perfectly compatible with Russell's theory. It does not apply only to 'purely qualitative' descriptions but is equally applicable to descriptions containing referential elements or even variables.

        Finally, there is a common objection, registered for example by Devitt (this volume, note 36), that it is 'implausible' that sentences containing incomplete definite descriptions do not literally express truths. This is not accurate, because for any sentence with a truth value, either it or its negation is true (besides, Devitt should not appeal, as he sometimes does, to 'intuitions about truth conditions' without saying truth conditions of what -- what speakers mean have truth conditions too). The point, presumably, is that the proposition, true or false, expressed by the sentence (according to Russell's theory) is not the one that intuitively is expressed. The Russellian should reply by saying that such sentences, because their propositional contents (or truth conditions) are what Russell's theory says they are, are typically not used in a strictly literal way (even when all their constituent words are being used literally). This view may initially seem counterintuitive, but that is only because of a misunderstanding of the kind of nonliterality at work here.[36]

        In any case, there is a psychological reply to this common objection. Yes, it seems problematic for Russell's theory to predict that a sentence like 'The table is covered with books' is false when the sentence can straightforwardly be used to make a true statement -- intuitively, one has no sense that there is any false proposition in the air. But there is an explanation for this: as soon as one hears (or reads) 'the table', because of its obvious incompleteness one immediately takes it either to be used referentially or with an implicit completion. One never actually computes the proposition expressed by the sentence in which it occurs.

        Moreover, it is a misconception to suppose, as for example Stanley and Szabo do (2000, 235-6), that the literal falsity of a sentence like 'The table is covered with books' has anything to do with how it manages to be used to make a true statement. This is not a case of Gricean quality implicature. It is not like an utterance of the obviously false 'Ken Lay cried all the way to the bank', used to convey how happy Lay was with his ill-gotten gains. What triggers the hearer's search for something other than what is said is not its obvious falsity but its lack of relevant specificity.

        To sum up, there are two basic cases. In the referential case there is an implicit narrowing of things under consideration or relevant to the conversation, and 'the F' is used to refer to the only F among them. In the attributive case there is an implicit, contextually relevant narrowing of the property expressed by 'F'; 'the F' is used to single out whichever individual uniquely has that narrowed property. In both cases the speaker intends the hearer to take such a narrowing into account in figuring out what he means. If no such narrowing is evident to the hearer, she will not figure out what the speaker means. If she recognizes the use as referential, she will wonder which F the speaker is talking about; if she recognizes the use as attributive, she will wonder what unique case of an F is in question.


Point 13. Definite descriptions do not have referential meanings.

Various reasons have been given for supposing that descriptions have referential readings, but they are commonly, though not universally, thought to have been adequately rebutted. Michael Devitt (this volume) revisits several of these reasons and offers some new ones. He is perfectly aware of the distinction between what a speaker says and what a speaker means and, in particular, of that between linguistic reference and speaker's reference. Even so, he is not convinced that special conventions or referential meanings are not needed to enable descriptions to be used referentially.

        Why am I convinced? Not because such conventions or meanings do not help explain how or why a given use of a description is taken referentially rather than attributively. After all, no ambiguity thesis has to explain how hearers manage to disambiguate the allegedly ambiguous expression. Still, it is important to appreciate what goes on in the case of definite descriptions. Whether or not there are referential conventions or meanings for them, how does a hearer figure out which way, referentially or attributively, a description is being used? Compare situations in which the following sentences might be uttered:

                  (11)       The man drinking a martini is in the corner nearest the bar.

                  (12)       The man in the corner nearest the bar is drinking a martini.

Regarding the first situation, imagine that you are talking by cell phone to a fellow spy who is at a party and you're helping him find an informant at a party, whose identity neither of you knows. You and he mutually know that the informant is there and will be drinking a martini. You have been advised where the informant is supposed to be, and you pass on this information by uttering (11). You are using the subject description attributively, merely enabling your cohort to identify the individual who satisfies it (i.e., its obvious completion). In the second situation, you are both at the party, you see a man in the corner nearest the bar and, since you and your cohort mutually know that the informant is at the party and is to be drinking a martini, you utter (12). In this case, you are using the subject description referentially. You are asserting a singular proposition about the man in the corner nearest the bar, and are enabling your cohort to identify the individual you have in mind by identifying him as the man in the corner nearest the bar.

        Notice, however, that this identification involves the quantificational meaning of the description. So if definite descriptions are ambiguous, their ambiguity is most extraordinary: one of their senses (the quantificational) plays a role when the other (the referential) is operative. An alternative account was presented in Part I: the distinctive quantificational character of definite descriptions, together with their implication of uniqueness, helps explain how, without any special stage setting, they can readily be used to refer (Point 5), in part because there may be good reason to suppose that the speaker is implicitly conveying an identity between something he is thinking of and what satisfies the description (Point 4). In effect, he is instructing the hearer to think of and take him to be talking about whichever individual is the F. Besides, even though descriptions are devices of quantification, they are often the only sort of singular noun phrase available to use to refer to what one wishes to refer to (Point 3).

        Now Devitt offers some interesting arguments in defense of the thesis that descriptions have referential meanings, operative when they are used referentially. In his view, 'the core of the referential meaning of a description token is its reference-determining relation to the particular object that the speaker has in mind in using the description'.[37] Devitt does not offer a complete account of referential meanings and, in particular, maintains 'neutrality' about the precise role of 'F' in 'the F'. In this regard it is not clear that his discussion precludes an account on which 'the F' (on its alleged referential reading) means the same as 'the particular F I am talking about/thinking of', a description that is uniquely satisfied by the unique object the speaker has in mind. If this, or anything like it, counted as a referential meaning, then Devitt's account would be vulnerable to an objection based on Neale's point (mentioned in Point 12) that a definite description containing a referring term, in this case the indexical 'I', is nevertheless amenable to Russellian treatment as a quantifier phrase.

        At any rate, I will address only those of Devitt's arguments that might adversely affect my pragmatic approach to referential uses of definite descriptions.[38] In particular, I will not be rebutting his arguments II, IV, and V.[39] His main argument for referential descriptions is Argument I, from regularity, according to which regularized uses of expressions, even if not their original uses, correspond to conventional meanings of those expressions. He bolsters this one with Argument III, which likens referential uses of definite descriptions to deictic uses of demonstrative pronouns and phrases. And he rests his case with Argument VI, which purports to show that when descriptions are exportable from opaque contexts they are referential. Since space limitations preclude giving Devitt's arguments the detailed attention they deserve, I will focus on the main one, Argument I, and respond briefly to III and VI.

        Argument I aims to rebut Kripke's (1977) and Neale's (1990, ch. 3) implicit assumption that because referential uses could be explained pragmatically, they should be so explained. In rebutting this assumption, Devitt takes his opponents to be likening referential uses to particularized conversational implicatures, whose understanding requires a fancy inference on the part of the hearer. It is not clear to me that this is Kripke's or Neale's view, but it is definitely not mine, as Devitt acknowledges (see his notes 9 and 12). In any case, Argument I does indeed challenge the analogy with particularized implicatures. It exploits the fact that not only can we use definite descriptions referentially, we regularly do so (hence no need for special stage-setting). This is not true of other sorts of quantifier phrases that can also be used to refer, but generally only in special circumstances. I agree with Devitt that this is an important difference, but why he take this as strong evidence that descriptions have referential meanings? His main point is that even if there is an historical explanation for how definite descriptions, considered as quantifier phrases, came to be used referentially, this does not show that they have not thus acquired a separate referential meaning. He likens this situation to the case of dead metaphors, such as 'kick the bucket' and 'spill the beans', which have acquired an idiomatic phrasal meaning distinct from their compositionally determined meaning. He suggests that even in cases where the phrasal meaning is inferable from the compositional meaning, this is still a distinct, conventional meaning.

        Devitt's dead metaphor analogy is dubious. To endorse the Russellian claim that definite descriptions are quantifier phrases, not referring terms, and that their attributive use is their only strictly literal use, does not commit one to the view that their attributive use came first and their referential use somehow developed later. It does not require the supposition that there was a time when definite descriptions were not used referentially. Howver, Devitt is correct to argue that the mere derivability of one use from another does not show that the second use is not a conventional use in its own right. Even so, he does not consider the possibility that the quantificational meaning of descriptions plays a role in referential uses and in how hearers recognize uses as referential.

        Moreover, if their quantificational meaning does play such a role, then Devitt's view implies that definite descriptions are semantically ambiguous in a most peculiar way: one of their meanings, the quantificational one, would be operative whenever the other meaning is. No other semantic ambiguity is like this, not even when formerly figurative uses of certain locutions become conventionalized in their own right. For example, if you hear an utterance of 'Kirk kicked the bucket', you can understand the idiom 'kick the bucket' to mean to die without having a clue about the once living metaphor (whatever that was) from which it derived. In contrast to these examples, one can't understand a referential use of a definite description without grasping its literal, quantificational meaning. At least so I have claimed. To address this claim, Devitt would need to show either that the quantificational meaning does not play a role in referential uses or, granting that it does, that this does not render his ambiguity claim problematic.

        Argument I is also vulnerable to a familiar objection. The claim that descriptions have referential as well as quantificational meanings seems committed to a massive cross-linguistic coincidence. For what could be the source of descriptions' 'systematic duplicity of meaning' (Grice 1969, 143) but the semantic ambiguity of the definite article? But if that's the source of their double use, then the thesis that definite descriptions have referential meanings misses the cross-linguistic generalization that in any language that has a definite article, definite descriptions have double uses. It would be a remarkable fact that an ambiguous word ('the' in this case) in one language should have translations in numerous other languages that are ambiguous in precisely the same way.

        In any case, Argument I is directed only to the analogy of referential uses with particularized conversational implicatures. I take referential uses to be akin to generalized implicatures. Both are cases of standardization, pragmatic regularities that obviate the need for full-blown inferences on the part of the hearer (of course Grice would insist that the complete inference be 'capable of being worked out' (1989, 31), of being calculable, even if it doesn't need to be calculated). As Grice explains, in the case of a generalized conversational implicature, 'the use of a certain form of words in an utterance would normally (in the absence of special circumstances) carry such-and-such an implicature' (1989, 37; my emphasis). The audience still has to infer what the speaker means from what he says, but the inference is attenuated, not requiring the more elaborate reasoning needed for recognizing particularized implicatures. Acknowledging the difference, Devitt laments that 'the absence of a [full-blown] Gricean derivation is, sadly, not sufficient to make a use semantic' (note 12). In this way, he concedes that Argument I does not undercut the analogy of referential uses with generalized implicatures.

        It is important to see that if Argument I were effective against this analogy, it would prove too much.[40] This regularity argument would lead to the conclusion that in sentences (13), (14), and (15), the words 'and', 'believe', and 'some' have special conventional meanings.

                (13)     It's worse to go the hospital and get sick than to get sick and go

                            to the hospital.

                (14)     I don't believe astrology is bogus, I know it is.

                (15)     Michael drank some of my wine if not all of it.

In (13) 'and' indicates a causal relation in its first occurrence and a temporal relation in its second. These uses of 'and' are quite regular, but there are (well-known) Gricean explanations for them. Yet Devitt's argument implies that these uses, being regularized, are conventionalized, hence that 'and' is ambiguous in at least three ways. Similarly, it implies that the sense of 'believe' in (15) entails 'does not know', contrary to the popular philosophical opinion that knowledge implies belief, and that the sense of 'some' in (14) entails 'not all', contrary to elementary logic. As I have long stressed (see, e.g., Bach and Harnish 1979, ch. 9, and Bach 1995), we need to observe the distinction between conventionalization and mere standardization or pragmatic regularity, as illustrated by these and numerous other examples (see Levinson 2000). Whereas conventionalization makes possible a certain sort of use, standardization merely facilitates a certain use. By turning a speaker meaning not determined solely by the linguistic meaning of the words being used into what Levinson calls a 'default meaning', it streamlines the hearer's inference to what the speaker means. A pragmatic regularity is not a matter of convention.

        In arguing that referential uses of definite descriptions are like literal deictic uses of demonstrative phrases (Argument III), Devitt simply assumes that the latter have referential meanings. A recent book by Jeff King (2001) suggests otherwise, arguing that such demonstrative phrases are quantificational. Even if he is wrong about that, King provides compelling linguistic data to show that definite descriptions are fundamentally different from demonstrative phrases in that speakers' intentions do not contribute to determining the semantic contents of uses of definite descriptions.[41] Also, there are clear pragmatic differences between definite and demonstrative phrases. For example, compare the following uses of 'that' and 'the'. In the case of (16) suppose that there are many books on the table and the speaker points to one of them.

                (16)     a. I haven't read that book.

                            b. #I haven't read the book.

Clearly the utterance of (16b) is pragmatically anomalous. As for (17), suppose that several physicists have been mentioned successively in the conversation.

                (17)     a. That physicist is a genius.

                            b. #The physicist is a genius.

In uttering (17a) the speaker is referring to the last physicist mentioned. It is not clear what he would be doing if he uttered (17b). These examples show that the conditions for using an incomplete definite description to refer are not the same as those for using a demonstrative phrase. For example, in contexts when one would use 'that F', with stress on 'that', along with a demonstration to pick out one F from others, it would be not only odd but ineffective to use 'the F'. This is not surprising, considering that singular definite descriptions imply uniqueness and demonstrative descriptions do not.

        Devitt's Argument VI is based on the claim that 'exportation works when the exported term is referential because then the opaque ascription requires that the believer be, like the speaker, en rapport with [have a singular thought about] the object of belief'. The short Russellian response to this argument is that it is analogous to the following dubious argument for an exclusive as well as an inclusive sense of 'or': whereas 'Michael drinks wine or Michael drinks beer' and 'Michael drinks wine' do not entail 'Michael does not drink beer', 'Michael is New York or Michael is in Sydney' and 'Michael is New York' does entail 'Michael is not in Sydney'. The trouble with this argument, of course, is that the second inference is just as invalid as the first, even though inferences of that sort do not lead from truth to falsity (people can't be in two places at once). Similarly, what Argument VI shows is that when a description is used referentially (and there is actually something to which the speaker is referring), exporting a description from an opaque context cannot turn a truth into a falsehood. It does not show that the description has a referential as well as a quantificational meaning. The fact that exportation sometimes works and sometimes doesn't does not show that descriptions have different meanings in the two cases, with an inference rule operative in one case and not in the other. It is just a symptomatic by-product of the difference between the two uses.[42]

        Devitt is well aware of the distinction between semantics and pragmatics and of the correlative distinction between what a speaker says and what a speaker means. He is not attributing referential meanings to definite descriptions arbitrarily, and he is right to insist that the possibility of a pragmatic explanation does not establish its correctness. Even so, it does not seem that his arguments succeed in undercutting the claim that referential uses are a matter of pragmatic regularity, not semantic convention. Moreover, there is a presumption in favor of pragmatic explanations. There is good methodological reason to appeal to independently motivated principles of rational communication, rather than to special features of particular expressions and constructions, in order to explain linguistic phenomena in as general a way as possible. In the present case, the fact that the referential use of definite descriptions is a cross-linguistic phenomenon suggests that it demands a pragmatic explanation, not a semantic one. Moreover, the denial that descriptions have referential meanings does not require crudely wielding Grice's semantic model of Occam's Razor and baldly appealing to parsimony. Rather, as I have suggested, the fact that descriptions are quantificational phrases of a certain sort helps explain their referential use. That is because their quantificational meaning plays a key role in their referential use. [43]




Abbott, B. (1999). 'Support for a Unique Theory of Definite Descriptions'. SALT 9: Proceedings from Semantic and Linguistic Theory IX. Ithaca, NY: CLC Publications, 1-15.

_____ (forthcoming). 'Definites and Indefinites', in L. Horn and G. Ward (eds.), Handbook of Pragmatics. Oxford: Blackwell.

Bach, K. (1987/1994). Thought and Reference, paperback edition, revised with postscript. Oxford: Oxford University Press.

_____ (1994). 'Conversational Impliciture'. Mind & Language, 9: 124-62.

_____ (1995). 'Standardization vs. Conventionalization'. Linguistics and Philosophy, 18: 677-86.

_____ (1997). 'Do Belief Reports Report Beliefs?', Pacific Philosophical Quarterly, 78: 215-41.

_____ (1999). 'The Semantics-Pragmatics Distinction: What It Is and Why It Matters', in K. Turner (ed.), The Semantics-Pragmatics Interface from Different Points of View. Oxford: Elsevier, 65-84.

_____ (2000). 'A Puzzle about Belief Reports', in K. Jaszczolt, The Pragmatics of Propositional Attitude Reports. Oxford: Elsevier, 99-109.

_____ (2001a). 'You Don't Say?'. Synthese, 128: 15-44.

_____ (2001b). 'Speaking Loosely: Sentence Nonliterality', in H. Wettstein (ed.), Midwest Studies in Philosophy, vol. 25. Oxford: Blackwell, 249-63.

_____ (2002a). 'Seemingly Semantic Intuitions', in J. K. Campbell, M. O'Rourke, and D. Shier, eds., Meaning and Truth. New York: Seven Bridges Press, 21-33.

_____ (2002b), 'Giorgione Was So-called Because of His Name'. Philosophical Perspectives, 16: 73-103.

Bezuidenhout, A. (1997). 'Pragmatics, Semantic Underdetermination, and the Referential-Attributive Distinction', Mind, 106: 375-410.

Birner, B. and Ward, G. (1994), 'Uniqueness, Familiarity, and the Definite Article in English', Proceedings of the Berkeley Linguistics Society, 20: 93-102.

Boer, S. E. and Lycan, W. G. (1986). Knowing Who. Cambridge, MA: MIT Press.

Chastain, C. (1975). 'Reference and Context', in K. Gunderson (ed.), Minnesota Studies in the Philosophy of Science, vol. 7, Language, Mind, and Knowledge, Minneapolis: University of Minnesota Press, 194-269.

Devitt, M. (this volume), 'The Case for Referential Descriptions'.

Donnellan, K. (1966). 'Reference and Definite Descriptions', Philosophical Review, 75: 281-304.

Graff, D. (2001). 'Description as Predicates'. Philosophical Studies, 102: 1-42.

Grice, H. P. (1969). 'Vacuous Names', in D. Davidson and J. Hintikka (eds.), Words and Objections, Dordrecht: Reidel, 118-145.

_____ (1989). 'Logic and Conversation', ch. 2 of Studies in the Way of Words. Cambridge, MA: Harvard University Press, 22-40.

Gundel, J., Hedberg, N. and Zacharski R. (1993), 'Cognitive Status and the Form of Referring Expressions in Discourse', Language, 69: 274-307.

Kaplan, D. (1979). 'Dthat', in P. French, Theodore Uehling, and H. Wettstein (eds.), Contemporary Perspectives in the Philosophy of Language. Minneapolis: University of Minnesota Press, 383-400.

_____ (1989a). 'Demonstratives', in J. Almog, J. Perry, and H. Wettstein (eds.), Themes from Kaplan. Oxford: Oxford University Press, 481-563.

_____ (1989b). 'Afterthoughts', in J. Almog, J. Perry, and H. Wettstein (eds.), Themes from Kaplan. Oxford: Oxford University Press, 565-614.

Karttunen, L. (1976). 'Discourse Referents', in J. D. McCawley (ed.), Syntax and Semantics 7. New York: Academic Press, 363-86.

King, J. (1987). 'Pronouns, Descriptions, and the Semantics of Discourse', Philosophical Studies, 51: 341-63.

_____ (1988). 'Are Indefinite Descriptions Ambiguous?', Philosophical Studies, 52: 417-40.

_____ (2001). Complex Demonstratives: A Quantificational Account. Cambridge, MA: MIT Press.

Kripke, S. (1977). 'Speaker's Reference and Semantic Reference', Midwest Studies in Philosophy, 2: 255-76.

_____ (1980). Naming and Necessity. Cambridge, MA: Harvard University Press.

Levinson, S. (2000), Presumptive Meanings: The Theory of Generalized Conversational Implicatures, Cambridge, MA: MIT Press.

Ludlow, P. and Neale, S. (1991). 'Indefinite Descriptions: In Defense of Russell'. Linguistics and Philosophy, 14: 171-202.

Ludlow, P. and Segal, G. (this volume), 'On a Unitary Semantical Analysis for Definite and Indefinite Descriptions'.

Mill, J. S. (1872). A System of Logic, definitive 8th edition. 1949 reprint. London: Longmans, Green and Company.

Neale, S. (1990), Descriptions. Cambridge, MA: MIT Press.

_____ (1993), 'Grammatical Form, Logical Form, and Incomplete Symbols', in A. D. Irvine and G. A. Wedeking (eds.), Russell and Analytic Philosophy. Toronto: Toronto University Press, 97-139.

_____ (forthcoming), 'This, That, and the Other'.

Recanati, F. (1993). Direct Reference: From Language to Thought. Oxford: Blackwell.

Russell, B. (1905). 'On Denoting'. Mind, 14: 479-93.

_____ (1917/1957). 'Knowledge by Acquaintance and Knowledge by Description'. In Mysticism and Logic, paperback edition. Garden City, NY: Doubleday Anchor, 202-24.

_____ (1918/1956). 'The Philosophy of Logical Atomism', in Logic and Knowledge. London: George Allen and Unwin, 175-281.

_____ (1919). 'Descriptions', ch. 16 of Introduction to Mathematical Philosophy. London: George Allen and Unwin, 167-80.

Salmon, N. (1991). 'The Pragmatic Fallacy'. Philosophical Studies, 63: 83-97.

_____ (1998), 'Nonexistence', Nous 32: 277-319.

Searle, J. (1979). 'Referential and Attributive', ch. 6 of Expression and Meaning. Cambridge, UK: Cambridge University Press, 137-61.

Sharvy, R. (1980). 'A More General Theory of Definite Descriptions'. Philosophical Review, 87: 607-24.

Strawson, P.F. (1950). 'On Referring'. Mind, 59: 320-44.

Szabo, Z. (2000). 'Descriptions and Uniqueness'. Philosophical Studies, 101: 29-58.

Wettstein, H. (1981). 'Demonstrative Reference and Definite Descriptions'. Philosophical Studies, 40: 241-57.




[1] Regarding definite descriptions, see Kripke 1977, Bach 1987/1994, ch. 6, Neale 1990, ch. 3, and forthcoming, and Salmon 1991; as to indefinite descriptions, see King 1988 and Ludlow and Neale 1991. The general strategy has been to seek a pragmatic explanation for referential uses. Kripke was the first to develop such an account, arguing that the distinction between linguistic reference and speaker's reference is just a 'special case' of the distinction stressed by Grice between linguistic meaning and speaker's meaning (Kripke 1977, 263). And Grice himself had insisted that despite having both an 'identificatory' and a 'non-identificatory' use, 'descriptive phrases have no relevant systematic duplicity of meaning; their meaning is given by a Russellian account' (1969, 143). However, Kripke doubted that Russell's account can accommodate the case of incomplete descriptions, such as 'the table'. And Devitt (this volume) has suggested some new reasons for supposing that descriptions have referential as well as quantificational meanings. Kripke's worries are addressed in Point 12, and Devitt's arguments in Point 13.

[2] Some observations on the term 'singular':

      Singular' is contrasted with 'plural', of course, but the singular/plural distinction has various applications. In the first instance it is the linguistic distinction as to number, and applies to nouns and noun phrases, as well as, in some languages, such as French and German, to adjectives (including predicate adjectives), which are also marked for gender. Subject-verb agreement is sensitive to the number of the subject. And pronouns anaphoric on noun phrases agree with them as to number. However, certain quantifier phrases, namely 'some F', 'each F', and 'every F', are grammatically singular (so are quantified mass terms, like 'some water' and 'much money') but semantically plural, and pronouns anaphoric on them are singular as well, at least in the King's English (in contemporary American English it is common to use plural pronouns, perhaps because they are gender-neutral).

      A singular/plural distinction may also be drawn as to reference. Philosophical discussions of reference to individuals generally concentrate on singular reference, but, even though there may be nothing in principle different about it, we should not overlook the case of plural reference, reference to more than one individual at a time. Plural reference can be made not only with a plural pronoun ('they') and with a conjunction of noun phrases, e.g., proper names ('Dick and Jane') or pronouns ('he and she') or combination thereof ('Jane and he') but even, on occasion, with quantifier phrases ('the nine justices', 'both authors').

      It should be mentioned that in philosophy the phrase 'singular term' is often used in contrast to 'general term', corresponding to the notational distinction in logic between individual constants or variables and predicate letters. Individual constants or variables are used to represent singular proper names and indexicals, not plural ones. Also, singular definite and even indefinite descriptions are sometimes labeled 'singular terms', even though they are usually represented by quantifiers.

     As for the phrase 'singular proposition', as introduced by Kaplan (1977), it applies to propositions having individuals as constituents. General propositions are quantificational. This distinction is not exclusive, however, since a proposition may be particular with respect to one argument slot and general with respect to another, as in the case of the proposition expressed by 'She has seven children'. It is interesting to note that Russell had occasion (e.g., 1918, 235) to use the phrase 'particular proposition' for propositions that have individuals as constituents. Ideally, it would be better to use Russell's phrase, not only because outside of philosophy 'singular' ordinarily contrasts not with 'general' but with 'plural' (as above), but also, and more importantly, because so-called singular propositions can involve more than one individual, as in the case of propositions expressed by 'I love you' and 'They're both angry'. Even so, I follow standard usage and stick to the usual 'singular proposition'.

[3] This qualification is meant to allow for cases in which the sentence occurs fails to express a proposition, either because the referring expression, though referring in character, does not have a referent or because the sentence is semantically incomplete. Regarding the first sort of case, I have in mind the view that if an expression lacks a semantic value, then sentences in which it occurs fail to express propositions. Examples of the second sort include sentences like 'John hasn't finished' and 'She is late', where an additional constituent is needed to yield a complete proposition (for discussion see Bach 1994, 126-33).

[4] There is the further question of whether a singular proposition comprises the complete content of a singular thought. Schiffer (1978) argued that this is not possible. In my view, de re modes of presentation are also involved (Bach 1987/1994, ch. 1). Moreover, I have argued that a belief ascription whose 'that'-clause expresses a singular proposition does not fully individuate the belief being ascribed (Bach 1997, 2000). I point out, for example, that the one 'that'-clause in the two ascriptions, 'Peter believes that Paderewski had musical talent' and 'Peter disbelieves that Paderewski had musical talent', does not fully characterize something that Peter both believes and disbelieves. And, as I say, every case is potentially a Paderewski case.

[5] I will generally use the words 'individual', 'object', and 'thing' as neutral terms for persons, places, things, and other concrete objects.

[6] For the sake of discussion and contrary to my own view, I am going along with the widespread view that proper names are referring terms (and that sentences in which they occur express singular propositions). However, elsewhere I have defended an alternative, a version of metalinguistic descriptivism I call the Nominal Description Theory of names, and offer an explanation of why it seems that proper names are referring terms (Bach 1987/1994, chs. 7 and 8, and 2002b). I am assuming also that there are singular propositions, which have particulars among their constituents. Anyone who has qualms about them can make do with object-dependent truth-conditions.

[7] Notoriously, Russell supposed that one has to be in a special, uniquely intimate relation with an object in order to think of or to refer to it, a relation which, he maintained, one cannot bear to ordinary things. Mill did not suppose this, and neither do I. My view of what it takes to be in a position to think of or refer to something is far more liberal than Russell's.

[8] Thus the relation of a description to what it denotes is fundamentally different from the relation of a name to what it refers to. This is suggested by the fact that we speak of individuals satisfying descriptions but not of them satisfying names. In Russell's terminology, descriptions 'denote' what they describe but do not refer to them. And notice Russell's use of the adverb 'directly' in characterizing how names designate their objects, just as Kaplan (1989) characterizes indexicals and demonstratives as 'directly referential'. However, given the distinction between denotation and reference, the occurrence of 'directly' in Kaplan's (1989) 'directly referential' is redundant, and 'indirectly referential' is an oxymoron. Also, when Kripke (1977) speaks of the 'semantic reference' of a definite description, he can only mean its denotation. So if we distinguish reference from denotation as two different species of what Kripke calls 'designation', then all referring expressions are rigid designators and all denoting expressions are nonrigid designators, except those that are rigid de facto, like 'the smallest prime' (Kripke 1980, 21). None of these terminological points addresses the question of which expressions are referring (as opposed to which ones, though not referring, can be used to refer).

[9] Assume here and in general that the description occurs in a simple sentence of the form 'the F is G'.

[10] If it is, then it is arguably nonliteral, though in an innocuous sort of way (see Bach 1987/1994, 69-74). For my take on the semantic-pragmatic distinction, see Bach 1999.

[11] Of course Russell held that ordinary proper names are 'disguised' or 'truncated' descriptions, in which case they too, contrary to appearances, are quantificational.

[12] Recanati (1993, ch. 15) and Bezuidenhout (1997), deny that definite descriptions are semantically ambiguous. However, they are disinclined to treat one use as literal and to explain the other pragmatically, because intuitively they find referential uses to be no less literal than attributive ones. Accordingly, they suggest that the existence of both uses is symptomatic of semantic underdetermination (or what Recanati calls, borrowing a phrase from Donnellan, 'pragmatic ambiguity'). From this it follows, implausibly, that a sentence like 'The discoverer of X-rays was bald' does not express a determinate proposition (Recanati allows that the meaning of such a sentence determines an 'external' proposition, which he distinguishes from any proposition that the sentence expresses in context, but it turns out that this is the same uniqueness proposition that is expressed on an attributive use of the description). If we wish to maintain that such a sentence does express a determinate proposition, and does so univocally, the obvious choice is a general proposition, in which case the description functions as a quantifier phrase and only its attributive use is the strictly literal one. And, as Point 5 below suggests, the referential use, though requiring a pragmatic explanation, is partly explained by the distinctive quantificational character of descriptions.

[13] The type of general proposition is what Strawson (1950) called a 'uniquely existential' proposition. I'll call it a simply 'uniqueness' proposition (once we opt, as below, for restricted quantification notation rather than that of standard first-order logic, Strawson's label no longer applies). Russell analyzed a uniqueness proposition of this sort into a complex logical form, expressible in the notation of modern first-order logic as:

              (LF-dd) ($x)(Fx & ("y)(Fy -> y=x) & Gx)

The key point is that the object that is the F does not appear in this proposition. So, for example, 'The queen of England loves roses' does not express a proposition about Elizabeth II. It means what it means whether or not she is queen of England and, indeed, whether or not England has a queen. As a quantifier phrase, 'the queen of England' does not refer to Elizabeth II (Russell would say it 'denotes' her, but for him denotation was a semantically inert relation), though of course it can be used to refer to her.

      It is commonly and justly complained that this form (LF-dd), although it captures the truth conditions of description sentences with its three conjuncts (or even the equivalent '($x)(("y)(Fy <-> y=x) & Gx)' with its two), distorts syntax beyond recognition. However, as Neale has stressed (first in his 1990), such syntactic complexity is not essential to Russell's theory of descriptions, in particular to his view that definite descriptions are 'incomplete symbols'. What makes them incomplete is not that they 'disappear upon logical analysis', making grammatical form misleading as to logical form, but that they introduce quantificational, variable-binding structure, rather than individuals, into the propositions expressed by sentences in which they occur.

      Besides, the repugnant complexity of (LF-dd) is an artifact of the notation used in first-order logic. As is often observed, even by beginning logic students, the usual symbolizations of English sentences with 'every' or 'some' also make a hash of syntax:

              (LF-un) ("x)(Fx -> Gx)

              (LF-ex) ($x)(Fx & Gx)

(LF-un) introduces a conditional and (LF-ex) a conjunction that are absent in the English. These distortions of syntax are also artifacts of the notation of first-order logic. And that notation suffers from severe expressive limitations anyway. It is well-known that quantifier phrases in general are not amenable to first-order treatment (Barwise and Cooper 1981), and Neale, following Barwise and Cooper, urges the use of restricted quantification as a uniform device for representing sentences containing any sort of quantifier phrase. In the simplest case it takes the form '[Qx: Fx] Gx'. Phrases containing standard quantifiers ('every', 'some', 'the') or non-standard ones ('few', 'most', 'three', etc.) can be rendered in like fashion: '[every x: Fx]', '[some x: Fx]', '[the x: Fx]', '[few x: Fx]', '[most x: Fx]', '[three x: Fx]', etc. Graff (2001, 32) suggests simplifying the notation with '[Q: F]x Gx', where the subscript indicates the variable bound by the quantifier. Presumably this simplified notation can perspicuously handle the case of one quantifier phrase containing another.

[14] Gundel, Hedberg, and Zacharski (1993) also include the statuses of being referential and being type-identifiable, but both are problematic. Being referential is a property of expressions, not objects of reference (strangely, at one point (p. 277) they speak of 'expressions which are referential but not uniquely identifiable'), and it emerges that the intended difference between being uniquely identifiable and merely referential turns on whether the expression being used to refer, without consideration of the rest of the sentence, suffices to enable the hearer to identify the reference. Being type-identifiable is the weakest category in Gundel et al.'s hierarachy. It does not distinguish one member of a type from another and is even a weak way of being able to be referred to. So it really doesn't belong on the list.

[15] There is an obvious problem with the view that degree of reference accessibility is encoded in expressions used to refer: most expressions that can be used to refer can also be used descriptively, in utterances made to convey not singular but general propositions. Unfortunately, the linguistic literature on accessibility and givenness tends to ignore the difference between referential and non-referential uses. If this difference were taken into account, it would be realized that unless the expressions in question were systematically ambiguous, which there is no reason to believe, it is incoherent to hold that they can encode degrees of accessibility just for when they are used to refer.

    Linguists have devoted considerable attention to the question of when it is appropriate to use a definite description and when to use an indefinite one. As they often observe (see Abbott forthcoming), when introducing something into a conversation, ordinarily one uses an indefinite description, not a definite one. There is no need to suppose, as many linguists do, that familiarity is somehow encoded by definite descriptions (that can't be right anyway, if only because definite descriptions can be used attributively, in which case the speaker need not be familiar with what satisfies the description). Rather, the requirement of familiarity, such as it is, is a simple consequence of the fact that the descriptive material in most descriptions does not apply (and is presumed not to apply) to exactly one individual. Most definite descriptions are not like 'the inventor of the zipper' or 'the queen of England'. However, once an individual is introduced as, say, 'a banker', he can subsequently be referred to as 'the banker'. For even though there are many bankers, once the introduction is made one can use 'the banker' implicitly to mean the banker being discussed or the contextually relevant banker.

[16] Why mutually salient or familiar? Obviously it is not enough for the intended referent to be salient or familiar merely to the speaker, if it is not salient or familiar to the audience and if this is not evident to the speaker (etc.). A speaker presumes that it is when using an expression that requires it to be; if his presumption is mistaken, the referent won't be evident to his audience. So, in general, when I say that something is salient or familiar, I will mean that it is mutually so. Similarly, when I speak of something's being obvious in a conversational situation, I will mean mutually obvious.

[17] Here and throughout I am assuming a distinction between saying and meaning or stating, a distinction which I have tried elsewhere to vindicate (Bach 2001). It corresponds to Austin's distinction between locutionary and illocutionary acts. This distinction is often blurred, e.g., by Donnellan (1966), whenever he suggests that in using a description referentially rather than attributively, one is saying something different, allegedly because the content of the description does not enter into what is said. See Point 11 below.

[18] Notice only definite quantifier phrases can be complete or incomplete. A definite 'Q F' is complete iff the number of Fs is in fact the number (however vaguely specified) indicated by 'Q'. So 'the ten Fs', 'both Fs', and even 'the few Fs' can be complete or incomplete. Of course, any quantifier phrase can be used with some implicit restriction intended by the speaker, but that doesn't make it incomplete. For example, you could say, 'Some students flunked', and mean that some students in your current logic class flunked (that class). But 'some students' is not incomplete, and 'some students in my current logic class' is not complete in the above sense, for it does not indicate how many students are in the class.

[19] It is commonly supposed that singular possessive descriptions also imply uniqueness, but this is far from clear. I can refer to one of my brothers as 'my brother', introduce one of my friends as 'my friend', and complain that I sprained 'my ankle', even though I have two ankles. Can these descriptions be included in the category of incomplete definite descriptions, as Neale (forthcoming) would insist, or are they genuine instances of what Abbott (forthcoming) calls the 'uniqueness problem'? Szabo (2000) and Ludlow and Segal (this volume) offer a various examples intended to show that even singular descriptions introduced by 'the' do not imply uniqueness. However, such examples have been explained away by Abbott (1999) and by Birner and Ward (1994), who also point out that such non-unique uses tend not be productive and seem to have a kind of generic character. Compare, for example, 'John went to the hospital' with 'John went to the hospital near his house' or with 'A fire truck went to the hospital'.

[20] It seems to me, contrary to Szabo (2000), that there is nothing linguistically odd about the fact that the simplex quantifiers 'both Fs' and 'the F' have complex semantic specifications. It is predictable that they are special cases, because of the unavailability of 'all two Fs' or 'all one F'.

[21] They can also be used predicatively, as when one refers to an object in one way and then describes it as a such-and-such, as when you say of the thing in your hand, 'This is a pomegranate'. It is arguable that this is not really a quantificational use but a distinctively predicative use, no different in kind from saying, 'This is red'. It is merely because that phrases containing singular common nouns require (in English) an article that one cannot say, 'This is pomegranate' (one could say the equivalent of this in Russian). Graff (2001) has argued that all uses of indefinite descriptions are fundamentally predicates and even extends her arguments to definite descriptions. Her account also covers generic uses of both definite and indefinite descriptions, as in 'The tiger has stripes' and 'A philosopher is not in it for the money'.

[22] I do not mean to suggest that 'a certain F' is always used to indicate that the speaker has a particular unspecified individual in mind. He might merely have in mind some unexpressed restriction on 'F'. For example, one might say, 'A certain contestant will go home happy', without specifying that whoever wins the contest in question will go home happy. Similarly, an utterance of the quantified 'Every author loves a certain book' could be made true if every author loves, say, the first book he wrote.

[23] Since I am discussing Kaplan, I will here use his term 'direct' to modify 'reference', although, as observed earlier (note 8), it is redundant. An expression, like a definite description, that merely denotes an object does not refer to that object, in the sense that the object is not a constituent of propositions expressed by sentences in which the expression occurs.

[24] As Kaplan explains (1989b, 579-82), certain things he had previously said, and even his formal system (in Kaplan 1989a), could have suggested that 'dthat' is an operator on definite descriptions, with the content of the associated description included in the content of the whole phrase. This would make 'dthat' a 'rigidifier' rather than a device of direct reference.

[25] To stipulate that any phrase of the form 'dthat [the F]' refers 'directly' does not, of course, mean that it is guaranteed a referent. It means only that the referent, if there is one, is a constituent of the singular proposition (if there is one) expressed by a sentence in which the phrase occurs.

[26] For example, Kaplan writes, 'Ignorance of the referent does not defeat the directly referential character of indexicals', from which he infers, 'a special form of knowledge of an object is neither required nor presupposed in order that a person may entertain as object of thought a singular proposition involving that object' (1989a, 536). He goes on to explain that he is objecting to what he calls 'Direct Acquaintance Theories of direct reference'. Here he seems to be conflating thinking of an object with referring to one, but he soon explains that he has 'only denied the relevance of these epistemological claims to the semantics of direct reference' (1989, 537).

[27] A singular proposition is not only object-involving but object-dependent, in that it would not exist if its object-constituent did not exist (at some time or other). So a singular proposition exists contingently. This does not imply that it exists only when its object-constituent exists. Singular propositions are not temporal just because they are contingent.

[28] This is related to Hintikka's suggestion, as quoted by Kaplan (1979, 384), that the referential-attributive distinction 'is only operative in contexts governed by propositional attitudes or other modal terms'. Evidently, Hintikka's idea was that if there is no modal or other operator whose scope interacts with that of the description, the distinction between what actually satisfies the description and any possible satisfier of it does not come into play.

[29] Some would deny the difference. In Searle's view, for example, the very idea of de re belief is an illusion 'arising from a confusion between reports of beliefs and features of the beliefs being reported' (1979, 157). But Searle maintains that belief about an object is always under a description, hence not really singular. He does not recognize the category of singular thoughts and the possibility of de re ways of thinking of things.

[30] It has even been suggested that giving a (descriptive) name to something whose existence is in question enables us to refer to it if it does exist. Suppose it is speculated that the perturbations in the orbit of Pluto are caused by a small but very dense planet beyond Pluto whose entire surface is very dark, making this planet invisible even to the largest telescope. Nobody knows whether there is such a planet, but astronomers take this speculation seriously and decide to call this planet 'Goofy'. Accordingly, they believe that Goofy, if it exists, affects the orbit of Pluto. Even so, it does not follow that they can have de re (singular) thoughts about it and use the name 'Goofy' to refer to it. Of course, they can single it out and describe singular propositions about it, provided it exists. Otherwise, they are just going through the motions.

[31] Karttunen is excluding negative quantificational phrases like 'no donkey' and negative cases like 'Bill didn't see a unicorn. *It/The unicorn had a gold mane'.

[32] In other words, each link in an 'anaphoric chain' (Chastain 1975) is treated as having a discourse referent, even if intuitively it does not refer. It should not be supposed, as Chastain (1975) and many others have, that when the links in the chain, the expressions anaphoric on the indefinite description, are used to refer, the indefinite description itself refers. In most such cases the indefinite will be used specifically, merely to allude to something.

[33] The difference may seem to be captured by the contrasting effects of inserting 'whoever he is' and a 'namely' phrase after the description, as in 'Schmidt's editor is insane':

            (i)      Schmidt's editor, whoever he is, is insane.

            (ii)      Schmidt's editor, namely Ross, is insane.

However, as the following considerations in the text suggest, this test is far from airtight.

[34] For more thorough defense of this point see Bach 1987/1994, 103-8, 124-6, and Neale 1990, 93-102, and forthcoming.

[35] In some cases of attributively used incomplete descriptions, there may be no intended expansion of the description at all. But what these cases illustrate is not a problem for Russell's theory but a common problem in communication. When it is indeterminate what the speaker's intention is, communication cannot really succeed, since nothing counts, strictly speaking, as understanding the speaker correctly. Practically speaking, though, this may not matter. In the case of attributively using a description without a determinate restriction, there is no problem as long as one and the same individual satisfies all the plausible candidates for the completion of the description. Then it makes no practical difference precisely which uniqueness proposition the speaker wishes to convey or which one the hearer takes him to convey.

[36] As I have previously argued, seemingly semantic intuitions confuse what speakers mean with the semantic contents of expressions (Bach 2002a), and, in particular, there is a kind of nonliterality that is compatible with each of the words in a sentence being used literally (Bach 2001b). Sentence nonliterality, as I call it, is not like metaphor, metonymy, and other figurative uses of words. Even though we may not intuitively think of this phenomenon as nonliterality, because no specific words are being used figuratively, it is a way of not being literal. For the speaker says one thing but intends to convey something distinct from that.

[37] Presumably Devitt is not equating the reference of an expression token with what a speaker refers to in using the expression, for there would be no point in calling speaker reference 'token reference' and in regarding that as semantic. Here I will not address Devitt's view that expression tokens have semantic properties of their own, and not just ones inherited from their types. However, I do think this view is problematic. Indeed, I regard token semantics as, well, token semantics (Bach 1987/1994, 85-88, and 1999). It is odd that he singles me out for 'restrict[ing] semantics to the study of types' (n. 4), for it is he who departs from the now common practice, endorsed by such prominent semanticists as Kaplan, Salmon, Soames, Richard, and Braun, of treating semantic reference as a property of types. This practice allows for referring expressions, notably 'pure' indexicals, to have references relative to contexts. I do not deny this, but only that definite descriptions, even when used to refer, have (semantic) references. Note that on this view, as Kaplan (1989a, 522-3) points out, an expression type can refer relative to a context even if not uttered/used in that context. As Kaplan argues, strange consequences result from attributing acts of utterance (or tokens) semantic properties of their own. For example, an utterance (or token) of the sentence 'I say nothing' cannot express a truth, but the sentence (type) is true relative to any context in which it is not uttered (1989b, 584). In any case, my discussion of Devitt's arguments will not rely on my doubts about token semantics.

[38] Also, I will not take up Devitt's arguments as applied to indefinite descriptions. For one thing, at least in my opinion, most of his examples of referential uses of them are really cases of specific uses (see Point 6). In these cases, the listener can't engage in 'reference-borrowing' in order himself to refer to the individual in question, for there is no reference to borrow. And once these sorts of cases are disallowed, it turns out that referential uses of indefinite descriptions are not regular but relatively rare, like referential uses of other quantifier phrases, and these, Devitt grants, do not have referential meanings.

[39] Argument II aims to neutralize Kripke's (1977) well-known pragmatic argument from Russell English (as opposed to Donnellan English), where it is stipulated that definite descriptions are strictly quantificational. But just as he insists that Kripke's argument is "unlikely to seem persuasive to someone who has not already accepted the case against referential definites." Devitt concedes that his own argument is "unlikely to seem persuasive to someone who has not already accepted the case [for referential definites]." As for his weak rigidity argument (IV), Devitt acknowledges that it does not provide independent support for the referential meaning thesis, since his opponent will treat referentially used descriptions as involved merely in the speaker's expression of weakly rigid thoughts. And his argument from incomplete descriptions (V) assumes that his opponent treats uses of incomplete descriptions as either elliptical for complete ones or as containing implicit but semantically relevant restrictions on their domains. It should be clear from Point 12 above that I do neither.

[40] Devitt also offers a clever argument that to deny that certain regularities of use involve semantic conventions is to slide down a slippery slope to the conclusion that 'there are no conventional meanings at all: it is pragmatics all the way down'. Such 'Gricean Fundamentalism' is, of course, absurd. However, the fact that Devitt has not identified where the Gricean draws the line between semantics and pragmatics does not show that there is no line to be drawn (see Bach 1999 for how I draw it). I do not deny that there are problem cases (see Bach 2001a, sec. 5, for examples), but Devitt's reductio, as well as Argument I (as I explain in the text), conflates the distinction between conventionalization and mere standardization or pragmatic regularity. Moreover, this 'Gricean fundamentalist' reductio is based on a bad analogy. In a given communication situation the hearer's job is figure out how the speaker is using a certain expression, not how it could have come to be used in that way. Whenever there is a pragmatic regularity of using that expression in a certain way and similar expressions in relevantly similar ways, this job is virtually effortless. For further discussion see the references in the following paragraph.

[41] For details see King 2001 (67-78 and 133-139). As King explains and illustrates with various examples, this contrast between definite descriptions and demonstrative phrases applies not only to their referential uses but also to their non-referential and even to their bound or what he calls 'quantification in' uses. King uses the existence of these other uses and various facts about them to support his contention that demonstrative phrases have a unified, quantificational semantics.

[42] If I were to respond to the details of Argument VI, I would argue, as I have previously (Bach 1987/1994, 195-202), that one can referentially use a description embedded in the that'-clause of a belief report without ascribing a singular thought, and one can ascribe a singular thought without using the description referentially. But I think it suffices to point out that Argument VI can be neutralized, as in the text, in a way consistent with Russellian pragmatics.

[43] For their very helpful comments, suggestions, and corrections, I would like to thank, and refer to not by description but by name, Barbara Abbott, Michael Devitt, Robin Jeshion, Jeff King, David Sosa, and Zoltan Szabo.