Reference+Comm. 20Jan97


Frank Jackson

Many today speak of the description theory of reference in terms suitable for the grand old, refuted theories of the past: it belongs with behaviourism, phenomenalism, libertarianism, verificationism, and cartesian dualism. I have never understood why. The objections brought against the theory all seem to me either to misunderstand the theory in one way or another, or to overlook obvious candidates to be the descriptions or properties that secure reference according to the theory. To argue this in detail is the work of a book, not a paper. I will outline how to defend a description theory of reference for a fragment of spoken and written language, giving, I hope, enough detail to show what the book might look like. The paper has a rather negative cast; there is a good deal of listing of well-known objections followed by rebuttal. But that is how it has to be when you are taking on conventional wisdom, and some positive points are made in developing the rebuttals. I should also note that, although I am in the minority, there have always been defenders of the description theory, and many of the things I say have been said in one form or another, somewhere or other, by someone or other.[2]

I start with some necessary background about language before proceeding to look critically at some of the some of the most influential objections to the description theory. I close with a brief discussion of reference failure.

Language and communication

I often want to communicate my view about how things are to someone else. Perhaps I think that some course of action is dangerous and want to tell you so, or perhaps I have a view about who will best govern the country and hope to sway you to my view by telling you, or perhaps ... . There are a number of ways I might seek to achieve my end: by pointing, by holding up pictures, by engaging in charades, or, most simply, if you and I share a language, by uttering words and sentences in the language we share (and know that we share, and know that we know that we share, ...). Language--or much of language, and in any case that aspect of language we will be concerned with--is a convention-generated set of physical structures that has as a principal function making it easy to articulate, and in consequence easy to record, transmit or communicate, and debate the correctness of, how someone--you, me, the enemy, the ideal observer, ...--takes things to be.

I am going to presume this Lockean picture of language. With it as background, we can see the attraction of a description theory of reference. If we are to use physical structures to give information on how we take things to be, we need associations in the minds of transmitters and receivers of the putative information that link the various structures and the various ways things might be. We use flags to give information about divers below, diseases on ships, the nationalities of visiting dignitaries, and so on. The system depends on known associations. Flying a blue-and-white flag from a buoy would not be much use for telling about the diver below if the association between flying a blue-and-white flag and the diver below were a dark secret. In the same way, if we are to use the physical structures known as sentences to tell about how we take things to be, we must (knowingly) associate various sentences with various ways things might be. Moreover, it is plausible that sentences effect these associations by means of combinations of elements or atoms, some of which are in turn (knowingly) associated with properties. Among the ways we take things to be, are ways involving various objects being of various kinds, and we use these elements or atoms or words, to tell about this. This means that we must associate certain words with certain properties.[3] Here the term `property' does not signify a universal in the sense that figures in debates over the one and the many, but is simply a short word for a way objects might be in the wide sense that includes relational and dispositional ways things might be. This way-things-might-be need not be a particularly unified way things might be. As David Lewis points out, if I tell you that something is a fish or a fowl, I tell you something about how it is. Hence, being a fish or a fowl is, in our relaxed sense, a property--possessed by all the fish and all the fowl, and by nothing that is not one or the other--despite its evidently disjunctive nature. Also, there need not be only one property associated with a word. There may instead be a number of properties. The word may be one to use when we want to say that there is something with a goodly number of properties from a certain list of properties; the word may be, that is, a cluster term.[4]

If lots of words are associated with properties, it is useful to have a name for the relation between these words and the objects (in the wide sense that includes those scattered objects called `kinds') that have the properties associated with them. The description theory of reference--or, rather, the small fragment of it that we are going to talk about in order to indicate how a fuller defence might go--says, first, that a good name for this relation is `reference', and, second, that the words we use to give and receive information via their association with properties include certain proper nouns and noun phrases. As we might say it, using the pre-analytic or folk term `about': terms like `London', `Pluto', `water', and `inert gas' are used by speakers to talk about whatever has the properties they associate with the term in question; or, as philosophers of language might say it, a name T used by S refers to whatever has the properties that S associates with T.

There are many other relations that might well be called `reference'. An important example is the connection between thoughts and what they are about. I should, therefore, emphasise that the description theory of reference I am defending is not a theory of reference in thought. It is a theory of reference for certain terms in a public language (in the sense that the language tokens are publicly available--they are written or spoken or ...) that presupposes that we can refer in thought; it presupposes that we have thoughts about how things are, and about particular objects, which we want to communicate among ourselves using a language containing words.

It is true, of course, that many have argued that what we learn (as they see matters) from Saul Kripke and Hilary Putnam about reference in language has important messages for reference in thought. Although Kripke's focus in Naming and Necessity (1980) is on criticising the description theory of how words like `gold' and `Aristotle' refer, many hold that the points he makes lead inevitably to the conclusion that the referential properties of persons' thoughts, what their thoughts are about, the contents of their thoughts, are in part a function of their environment. To this extent, what I am saying does bear on reference in thought. If I am right that the description theory survives the Kripke-Putnam critique, we need to re-visit at least some of the arguments against narrow content.

Another relation we might well call `reference' is the connection between an adjective like `square' and one or more of: squareness, individual square things, the scattered object that is the aggregation of all the square things, and the set of square things; yet another is the association between words and properties we have just mentioned in stating the description theory of reference. We will, though, focus on the view that words like `Aristotle', `London', `gold', and `water' refer to whatever object or kind that has the properties associated with the words; that will keep us busy enough.

As a final preliminary, I should highlight why it is so plausible that we need associations, known ones, between certain words and certain properties. An association between a word and an object would not in itself enable us to find or identify the object, or, more generally, allow us to make it a subject of discussion, in the ways we clearly can by using words in sentences. Finding, identifying, and making London a subject of discussion is necessarily doing so under one or more of its guises. We access objects via their properties. I partly have in mind here the obvious point that you cannot find London, or establish, say, that it is what you and I are both talking about, independently of its properties. But I also have in mind the more general point that there must be something about an object which makes it true that a certain word directs us to that object and not some other object. John Bigelow (1988), p. 132, says that truth supervenes on being; I suppose that our point could be put by saying that aboutness or reference supervenes on nature. Of course, there is a trivial sense in which reference supervenes on nature: if `t' refers to a and not to b, a differs from b in being that to which `t' refers. But I take it that the supervenience doctrine is plausible in a non-trivial sense also. (Another way of putting essentially the same point is to say that there is no such thing as bare reference. A familiar issue in the philosophy of counterfactuals is whether a counterfactual can be barely true. My inclination is to deny that counterfactuals can be barely true, but I am much more confident in denying that reference can barely obtain.)

The reason the associations between words and properties need to be known ones parallels why we need to know the association between blue-and-white flags and divers below. Only in that case would the words be of use to us in the project of communicating how we take things to be by means of their appearance in the sentences we produce and encounter.

I am going to take the idea of a known association between physical structures like words and properties largely for granted. When you receive a letter with a set of proofs saying to use double underlining for small caps, such an association becomes known to you, and you and the publisher will then exploit this common knowledge in order to finalise the proofs. So I take it everyone knows roughly what I have in mind. Obviously, the association has a lot to do with entering agreements concerning the transmission and receipt of views about how things are, and we will pick up on this aspect briefly at one place in what follows.

We now start the sceptical tour.

Replies to objections to the description theory

The passing the buck objection

This objection has been put by Michael Devitt in a number of places. Here is a recent statement of it.

Description theories are essentially incomplete. A description theory explains the reference of a word by appealing to the application of descriptions associated with the word. So the theory explains the reference of the word by appealing to the reference of other words. How then is the reference of those other words to be explained? Perhaps we can use description theories to explain their reference too. This process cannot, however, go on forever: There must be some words whose referential properties are not parasitic on those of others. ... Description theories pass the buck. But the buck must stop somewhere.[5]

This objection misunderstands the theory, or at least misunderstands the theory in the Locke-inspired form that I accept. This description theory explains the reference of a word (or better, in view of the fact that we are concerned only with a fragment of the theory, some words including `London' and `water') as that which possesses the property or properties associated with the word. Just as we said above. It is not an essential part of the theory that we should have words, or `other words', for these properties.[6] As far as the theory is concerned, someone might use the word `Fred' to refer to whatever has a certain property despite not having a word for the property. Perhaps there is some complex shape they can recognise but for which they have no word. They believe that there is exactly one object which has this property and settle on `Fred' as the word to use for this object. The talk of associated descriptions applying in statements of the description theory is to be understood in terms of the possession of associated properties, not in terms of the application of (other) words in the language in question. (Perhaps it would have been better if the theory had been called the associated property theory of reference.)

Defenders of the language of thought sometimes say that the problem of reference for public words is essentially the same as the problem of reference for words in mentalese, and it might be suggested that we should re-phrase the buck-passing objection in terms of the language of thought. The re-phrased version might run somewhat as follows. According to the description theory, the reference of T in the mouth of S is to the thing that has the properties S associates with T; but for S to associate properties with T is for S to have words in mentalese that refer to those properties. This, the objection might conclude, makes it clear that the problem of reference for words in a public language is simply being handed across to words in the language of thought. The buck is being passed from words in public language to words in mentalese.

For the sake of the argument, I will accept that there is a language of thought. (There surely is internal representation, but it may not take the form of a language.) The real problem with the re-phrased objection is that the problem of reference for words in a public language is very different from the problem of reference for words in mentalese. The difference derives from the point that Locke put by saying that words are voluntary signs.7 A word like `water' might have referred to gold, and would have done so if we had agreed to use the word `water' in the circumstances we in fact use the word `gold' in: which words in a public language refer to which things is in part a matter of the largely implicit conventions of usage we enter into.[8] This is crucial to the plausibility of the description theory of reference. It is plausible that we follow the convention of using `water' when confronted with stuff we take to be thus and so, and this is why it is plausible that we use the word for stuff that is thus and so. But this picture only makes sense for words in a public language; it would be a nonsense to suppose that we entered into a convention to use the words of mentalese in certain circumstances. Because we do not know what the words of mentalese are, we cannot make agreements concerning them. Moreover, even if we did know what they are, we could not choose when to use them in accord with a convention of usage; changing when we `use' some word of mentalese calls for brain surgery, not a mere change in the conventions of usage.

There is, of course, an important problem of reference for the words of mentalese (if such there be), and, more generally, for how we refer in thought, but, as noted earlier, this is not the problem of reference that the Lockean description theory we are defending is concerned with. Locke's point about voluntary signs tells us that we should think of the problem of reference for words as posterior to, and properly treated as separate from, the problem of reference for thought (or so it seems to me, though the matter is contentious). Of course, all this means that I am agreeing with the claim that the description theory is `incomplete', but the argument does not show that it is incomplete as a theory of reference for names in language; it shows, instead, that it is incomplete as a theory of reference across the board.

The objection that reference is in part a matter of how things are outside the head

Immediately after the passage quoted above, Devitt goes on to say

This deep failing of description theories is brought out by Hilary Putnam's slogan ... `Meanings just ain't in the head" ... . The association of descriptions with a word is an inner state of the speaker. No such inner state can make the word refer to a particular referent. For that we must look for some relation that language and mind have to things outside themselves--we must look for an external relation.[9]

We can say straight off that there must be something wrong with this objection. It is too good. The description theory is correct for some words. We do use some words simply as abbreviations for definite descriptions, and the question as to what they refer to is nothing other than the question as to what has the property associated with the abbreviated description.[10] This is granted by critics of the description theory. As Kripke notes, you might say to yourself `By "Gödel" I shall mean the man, whoever he is, who proved the incompleteness of arithmetic', and Kripke grants that in this case `Gödel' in your mouth would refer to Schmidt if it was indeed Schmidt who proved the theorem. Kripke is explicit that his claim is that, as a matter of fact, this is not what most of us do; not that we could not have done it, and not that we never do it.[11]

Moreover, the description theory does give a major role to the world, in the sense of things outside the head, in settling what refers to what. According to it, what settles the reference of words like `London' and `water' is the combination of the properties associated with the words and a fact about the world, namely, what entity in the world has the properties. It is, therefore, false that the theory deprives the world of a role in settling what a term refers to.

It might well be objected that I have misunderstood the sense in which Devitt is insisting that the world must play a major role in settling reference. Of course, what a word refers to at a world depends, according to the description theory, on the nature of that world. The trouble is, rather, that the world does not play a role in settling the reference conditions of a word on the description theory, in the sense of what the word refers to at w for any w. But, in fact, description theorists per se are not committed to denying the actual world (including those parts outside the head) a role in settling reference conditions. Description theorists should (and do) allow that some words function like rigidified definite descriptions, and the reference conditions of rigidified descriptions depend on the nature of the actual world.

The core idea behind the description theory is that words like `London' and `water' refer to that which has the property or properties the speaker associates with them. One way to spell this out is by saying that, at any world w, T refers to x just if x has the associated property or properties in w. Another way, equally faithful to the core idea, is by saying that, at any world w, T refers to x just if x is the thing with the associated property or properties in the actual world.[12] The second way of spelling matters out is the right way if (but not only if) T is a rigidified definite description, a description like `the actually tallest person in 1991'. For the actually tallest person at any world is whoever is the tallest at the actual world. But if description theorists should (as they do) count rigidified definite descriptions as referring expressions, then it is transparent that they should allow that reference conditions may depend on how things actually are. The reference of a rigidified definite description at any world w depends in general on how things actually are.

The objection that the description theory is a species of eliminativism about reference

If the language we are concerned with has predicates for the various properties the description theory holds serve to secure the reference of the names in that language, the description theory becomes in effect the view that names, be they of things or kinds, are a style of abbreviated definite description: if the language has `F' as the predicate for the property that settles the reference of `N', the description theory amounts to the view that `N' is, in some sense, an abbreviation of `the F'. The objection is that Russell teaches us that definite descriptions are not referring expressions, that they do not refer, and so the description theory eliminates the very phenomenon it is supposed to be analysing.

I accept Russell's account (when due account is taken of various niceties to do with conversational context) but I think that it is misleading to say that it means that definite descriptions are not referring expressions. Russell famously shows us how to give the truth conditions for certain sentences containing definite descriptions in terms that do not contain the definite descriptions themselves and do not contain anything equivalent to them as semantically significant units. According to his account,

(1) The F is G


(2) There is an F which is G, and every F is identical with it.

But to think that this shows that definite descriptions are not referring expressions, that they do not refer, misunderstands the sense of `reference' in which the description theory is a theory of reference. Nothing in Russell's theory goes against the fact that the words `the tallest person alive in 1990' are quite distinct from the person who is the tallest person alive in 1990, and that there is some important relation between the words and the person which warrants a name. It is this relation that the description theorist calls `reference', or regards as one of the relations properly so-called, and it is this relation that the description theorist sees as holding also between names and the things they name. It is, of course, open to anyone to withhold the word `reference' from this relation on the ground that we can give a contextual analysis of definite descriptions--or perhaps for some other reason--but then they will need a new term for the relation between the words `the tallest man alive in 1990' and the tallest man in 1990. And the objection will become a claim about how best to label theories; it will no longer be an objection to the description theory as such.

I grant (of course) that there are cases where a `contextual' analysis certainly does show that some expression is not referential. To take a famous example, we can analyse `The average family has 1.8 children' as `The number of children in families divided by the number of families equals 1.8'. This shows that `the average family' is not a referring expression in `The average family has 1.8 children'. But it does so by showing that the truth of that sentence does not require the existence of the average family. By contrast, Russell's analysis of `The F is G' does not tell us that this latter sentence can be true when the F does not exist.

It might be wondered why I did not reply to the objection by insisting that the description theory treats names as rigidified definite descriptions, not as definite descriptions simpliciter. But in fact the difference over rigidity does not affect the crucial issue. For an equally `contextual' account in the spirit of Russell can be offered of rigidified definite descriptions, as follows.

(3) The actual F is G

is true at w iff

(4) There is an F in the actual world such that every F in the actual world is identical with it, which is G in w.

What we can say is that one who insists that definite descriptions are not referring expressions on the basis of a Russellian story about them, is committed to saying that rigidified definite descriptions are not referring expressions either; and that this is hard to believe. It means, for instance, that we cannot explain what a rigidified definite description is by saying that it is one which refers at every world to what (if anything) it refers to at the actual world--it never refers at all, on the view in question.

It might be replied to our case for counting definite descriptions, be they rigidified or not, as referring expressions, that obviously there is some relation between, on the one hand, the words `the tallest man' and `the actually tallest man' and, on the other, the tallest man himself, but it differs markedly from the relation between a proper name of the man and the man himself. The issue, the response might run, is whether there is enough in common between the cases to justify describing them alike; whether there is, as Gareth Evans puts it, `any natural semantical kind' here.[13]

One reply to this response is that, whatever their differences, definite descriptions and proper names are united in being markedly different from words like `and', `is', and `to'. But the really important point is that we do not have in this response a free-standing objection to regarding definite descriptions as referring terms: it is a good objection to treating definite descriptions as referring terms only if there is independent good reason to reject the description theory of reference. For if the description theory of reference is correct, there is manifestly an obvious natural semantic kind in common between definite descriptions and proper names: they are all associated with properties in the minds of their users, and hence share a special, worth-giving-a-name-to relation to whatever it is that possesses the associated properties.

It is sometimes said that the description theory of reference holds that names are synonymous with definite descriptions. But, in fact, the key claim in the theory is that the fundamental mechanism whereby names and definite descriptions secure reference is the same, namely, via possession of associated properties--it is only in this sense that names (in rich enough languages) can be thought of as abbreviated definite descriptions. This is consistent with the referential behaviour of names and definite descriptions differing in various contexts in ways that would warrant saying that they differ in meaning. Take, for example, the widely observed differences between the behaviour of names and definite descriptions under counterfactual assumptions: the reference of a name does not change under counterfactual assumptions, the reference of a definite description may change.[14] This does not affect the basic claim that reference is via associated properties in both cases; it simply means that, in the case of names, the reference under counterfactual assumptions is to what has the associated property in fact, whereas the reference of a definite description may be to what has the property under the counterfactual assumption. Once we see that the essential claim of the description theory of reference is not one of synonymy, but concerns instead the sharing of a fundamental mechanism, it is clear that the manifest variety among referential devices concerning their behaviour in various contexts is no reason to deny the underlying semantic unity postulated by the description theory.

The objection from ignorance and error[15]

Surely, runs this objection, there are cases where it is intuitively clear that speakers refer to something O by their use of T, where one or both of the following applies: (a) a speaker knows nothing, and maybe knows that they know nothing, that individuates O from many other things, and (b) most of what a speaker believes about O is wrong. But if reference is by possession of associated properties, then (a) these properties must be sufficient to individuate O--the user of T must not be ignorant of what individuates O, and (b) O must have these properties--the user of T cannot be in error about O having them.

The cases offered to support the claim about ignorance of individuating properties all seem to me to overlook obvious candidates to be the needed individuating properties. Putnam claims that he does not know what distinguishes beeches from elms but insists that he succeeds in referring to beeches when he says, say, that he does not know how beeches differ from elms.[16] I agree that he does refer to beeches, but point out that he does know how they differ from elms: only they are called `beeches' by the experts in his language community. Putnam responds that, because the word in French for beech is different from our word, this reply would commit description theorists to holding that a not very knowledgeable, monolingual English speaker's concept of a beech is different from that of a not very knowledgeable, monolingual French speaker. But how is this a problem? Persons' concepts of one and the same thing can and do differ, and it is hard to see why this should not count as a case.

There are, though, cases where appeal to knowledge that certain words are used by experts does not turn the trick for the description theory. Perhaps I read in the paper about someone called `Smith' who was robbed last night in Washington. Can't I refer to him even though there may well be more than one person called `Smith' robbed in Washington last night? And there are examples of meeting someone towards the end of a party. It seems that I can refer to him next morning--perhaps in some such sentence as `I wonder if I insulted that man I only vaguely remember whom I met as I was finishing the bottle'--despite his having no obvious individuating marks that I can recall. But, in fact, I do know something that individuates what I refer to in cases of these kinds. They are the causal source of the information-carrying trace that I am presented with. Out of the possibly many Smiths robbed in Washington last night, there will be one that is the causal origin of the story in the paper, in the sense that it is his being robbed that led to the passage of prose in the paper in front of me. Or, rather, there had better be if the case is to be a test case for theories of reference: if it turns out that the reporter muddled up a number of robbed Smiths, and the passage of prose can equally be thought of as sourced in any, or none, of them, we lose the intuition that there is one particular Smith that I refer to. Similarly, there will be only one person who is the right kind of causal origin of my vague, somewhat disturbing memory of the party that is prompting my morning-after reflections (and who will, most likely, also be the only person I would recognise as such were I to meet him). Hence, the description theorist can explain how I manage to refer in this kind of case by appealing to an association with the property of being a certain kind of causal origin.

Suggestions of these kinds in support of a description theory of reference are not new.[17] Their impact has, however, been lessened by the belief that they face serious problems. I will consider three of the problems most often brought up.

First, the appeal to language-use by experts suggests an implicit circularity. It looks circular to say that `beech' in the mouths of the ignorant refers to whatever `beech' in the mouths of the experts refers to, and isn't that what the `reference borrowing' suggestion really amounts to?[18] However, the circularity is not vicious. We can spell the suggestion out without explicit mention of reference. The suggestion is simply that the property the ignorant associate with the word `beech' is having the property, whatever it is, that the experts associate with the word `beech'. Hence, on the description theory, `beech' in their mouths refers to whatever has the property the experts associate with the word `beech'--which ensures, as it should, that `beech' in the mouths of the ignorant has the same reference as it has in the mouths of the experts.

Secondly, it is observed that it may be unclear who the experts are in the sense that it is unknown to the ignorant users, and yet reference still occurs.[19] Many folk refer to quarks when they use the word `quark', despite the fact that they do not know whether it is a term from physics or from biology, and so do not know which department contains the experts. When such a person asks, `Should I go to the physics or biology department to find out about quarks?', it is plausible that they are asking a question about quarks. However, these folk do know that there are some experts somewhere or other, and that these experts lie at one end of a reference-borrowing chain that has whoever they themselves borrowed the term from at the other--or at least this had better be the case, for otherwise the example is no longer one where it is plausible to say that these folk refer to quarks. But this means we can specify the property these folk associate with the word `quark'. It is having the property the group of users of the word `quark' that they are borrowing from associate with the word `quark'. And what property does this second group of users associate with the term `quark'? Either they are the experts, in which case it is property Q (whatever that is, I am not one of the experts) or they are not the experts, in which case it is having the property some third group of users associates with `quark'. And what property does this third group of users associate with the word `quark'? Either they are the experts, in which case it is property Q, or they are not the experts, in which case it is having the property some fourth group of users associate with the word `quark'. And so on. We have, that is, a recursive story akin to the familiar recursive account of a wff in logic.

The third problem raised turns on the point that the key issue is not whether there is some individuating property of the thing referred to. It is whether there is some individuating property of the thing referred to which is associated, in the relevant sense, with the word in question by the speakers (and maybe hearers) in question. The objection is that the kinds of properties we have mentioned are fancy ones that philosophers of language and unusually alert members of the folk might think of; they are not properties ordinary, folk speakers associate with words like `beech' and `quark'. A quick reply to this objection is that the folk often say things which make it clear that they are well aware of these properties and which strongly suggest that they are relying on them to secure reference. People who do not know much physics, and know that they do not know much physics, often ask questions like, Is it established for sure that quarks exist? When asked precisely what question they are asking, they answer that they are asking about the things physicists use the word `quark' for. Or think of what has happened with the spread of computer speak. It is a commonplace that people say things like `I haven't a clue what RAM is, but I know that the new PC I am buying has 32 megabytes of whatever it is that computer people use the term for'.

However, we need to say more by way of reply. Although the folk are aware of the relevant individuating properties and say things that suggest they are relying on them to secure reference, they are not much good at articulating them in detail (nor, if it comes to that, are defenders of the description theory of reference--a fair bit of hand-waving goes on). The longer reply is that this failure to articulate the relevant property or properties in detail is no objection to the description theory provided that what is meant by the expression `properties associated with a word or phrase' in statements of the description theory is understood in the right way.

Sometimes it is obvious which properties are associated with a word. Perhaps the speaker tells us loud and clear. But often the association is implicit or tacit rather than explicit. It is something we can extract in principle from speakers' patterns of word usage and what they say about various possible (and actual) cases; it is not something actually explicitly before their mind when they use the words. I know this way of putting things--familiar though it is--will ring some alarm bells. Some will want to say that if the association is in the mind, as the description theory says (and, as we noted early on, must say), it must be explicit. They think of appeal to the implicit or tacit in this context as a kind of cheat--a way of saying something and then taking it back.

However, there is a way of being implicit and yet before the mind in the relevant sense which is no great mystery. Consider the situation good logic students find themselves in before they are given the recursive definition of a wff. They cannot specify what it is to be a wff, but they can reliably classify formulae into wffs and non-wffs. But it would be a mistake to see their ability as like that of chicken sexers. Chicken sexers have (I understand) no idea which property triggers their reliable classifications (although they know, of course, that the property is correlated with the sex of the chicken and is instantiated somewhere around the bottom of the chick). By contrast, logic students can say, for any ill-formed formula, what triggers their judgement that it is ill-formed. When presented with `(p v q', they do not say that they can see that it is ill-formed but cannot say where the problem is. They know exactly where the problem is and how to fix it--add a RH bracket after the `q'. Similarly, they know what changes to a particular wff would make it ill-formed. They are in the following position: for each particular example, they can say whether or not it is a wff, and why; but they cannot give in words a story that covers all cases. The same is true for all of us in our judgements of grammaticality. We can say, for particular examples, whether and why they are or are not grammatical, but cannot give the general story in words.

Description theorists can and should say essentially the same about the sense in which speakers associate properties with certain words. If you say enough about any particular possible world, speakers can say what, if anything, words like `water', `London', `quark', and so on refer to in that possible world. (This does not mean that there is always a definite answer: sometimes saying what `London' refers to in a certain possible world will amount to saying that it is indeterminate what, if anything, it refers to in this world.) Our ability to answer questions about what various words refer to in various possible worlds, it should be emphasised, is common ground with critics of the description theory. The critics' writings are full of descriptions of possible worlds and claims about what refers, or fails to refer, to what in these possible worlds. Indeed, their impact has derived precisely from the intuitive plausibility of many of their claims about what refers, or fails to refer, to what in various possible worlds. But if speakers can say what refers to what when various possible worlds are described to them, description theorists can identify in words the property associated in their minds with, for example, the word `water': it is the disjunction of the properties that guide the speakers in each particular possible world when they say which stuff, if any, in each world counts as water. This disjunction is in their minds in the sense that they can deliver the answer for each possible world when it is described in sufficient detail, but is implicit in the sense that the pattern that brings the various disjuncts together as part of the, possibly highly complex, disjunction may be one they cannot state. This is not to say that, after reflection on their classifications in the various possible cases, perhaps aided by doing a course in the philosophy of language, they won't be able to make a good stab at stating the pattern in an illuminating way in words; something like: belonging to the kind which most of the clear, potable samples, acquaintance with which led to the introduction of the word `water' in our language, is roughly right. And if you describe a case that this formula fails to cover, you do not show that I haven't grasped the pattern, but that my stab at giving it in words was not good enough. Or maybe you show that the way you introduced the case `disturbed' the pattern; description theorists can and should allow that the associations between words and properties respond to conversational and linguistic contexts, as well as, of course, evolving over time.

We can now deal quickly with the error side of the objection. The objection is that speakers may have most of the properties of O wrong and yet still refer to O by their use of T. But what matters for successful reference is that O has the properties speakers associate with T, and this is consistent with O's lacking most of the properties speakers think it has. If O is the object which has the property or properties speakers associate with T, then they refer to O when they use T, according to the description theory; it does not matter if everything else they believe about O is mistaken.

The Twin Earth objection

There are two different ways of thinking of Twin Earth. On one, we think of it as a more or less remote part of our (the actual) possible world; on the other, we think of it as part of another possible world. The two ways raise quite different issues for a description theory of reference for the word `water'.

Suppose that Twin Earth is part of the actual world, located, say, on the other side of our sun, and that it has, as is traditional, a population that speaks something very like English, and stuff superficially very like our water which plays on Twin Earth a role very like the role water plays on Earth. This stuff is called `water' by Twin Earthians but has a quite different chemical composition from our water: it is XYZ, not H2O. The first Twin Earth objection is that `water' in our mouths does not refer to XYZ, despite the fact that XYZ has so many of the superficial and functional properties of water.

This is only an objection to the description theory if the description theory is committed to saying that `water' in our mouths refers to XYZ in the imagined case. But, as we have just been emphasising, the description theory is committed to `water' in our mouths referring to whatever has the property we associate with `water', and the test for being the property we associate with `water' is that it is the, possibly disjunctive, feature common to the possible cases we describe as water. The very fact that we resist saying that the water-like stuff on Twin Earth is water shows that it does not have the property we associate with `water'. There is, as we noted above, no way that an appeal to intuitions about possible cases can refute the description theory; at most, it can tell us that we, or someone, got the associated property or properties for some word wrong. In fact, what we learn (and this is an important thing to learn; I am not decrying the interest of Twin Earth) is that acquaintance is an important member of the properties we associate with `water' (and lots of other words). The reason XYZ is not what we refer to when we use the word `water' is that it is not the water-like kind we are acquainted with. We have never ourselves come across XYZ. But if we had, if some of the water-like stuff around us was H2O, and some XYZ, and there was no reason to think of one as `fools' water, then we would have been in the same situation with respect to `water' as we in fact are with the word `jade'--two different kinds would have been covered by the one word.

The second Twin Earth objection invites us to consider a possible world where, on Earth, XYZ is like H2O actually is, both in superficial qualities and role, and is the stuff we are acquainted with. In this version of the objection, Twin Earth is not imagined to be a planet remote from our Earth, but is imagined to be our Earth itself in a non-actual possible world where our Earth has XYZ as the stuff which is the clear odourless, potable liquid that we are acquainted with and call `water'. The objection now is that `water' in our mouths does not refer to XYZ in this world, and yet XYZ has, in this possible world, any and every property that might plausibly be held to be associated with `water', including the property of being the water-like kind we are acquainted with in this world.

The reply to this objection is that the moral is as before: we are learning something about the relevant associated properties from these reflections on how to describe possible cases, not that reference does not go by associated properties. Twin Earth, in this second version of the objection, is given to us as a counterfactual world, a world other than our actual one; and what we learn from the fact that `water' does not refer to XYZ in this counterfactual world is that the property associated with `water' is the property of being the kind that actually is the water-like kind of our acquaintance. We learn that what matters is not the property XYZ has in the counterfactual world, but the property it has, or rather fails to have, in the actual world. It is because XYZ is not the water-like kind of our acquaintance in the actual world that the word `water' does not refer to it. As you would expect if this account is correct, if Twin Earth is thought of as how our world actually is, instead of as a counterfactual world, `water' does refer to XYZ in it. Perhaps we are contemplating the possibility that some terrible mistake was made by the founders of modern chemistry, or that some fabulous fraud was perpetrated--they knew all along that it was XYZ but decided to falsify the first text books and rely on lazy copyings from them to later ones to perpetuate the fraud--then what else is there to say but something like, "If that's how things are, hard though it is to believe that the error or fraud could have remained undetected for so long, then water is XYZ"? And, in this case, it is of course XYZ, not H2O, that has the property of being the actual water-like kind of our acquaintance.

The objection from divergent sets of associated properties

I know of no statement of this objection in the terms that follow, but it is suggested by Evans (1982), p. 48, and it comes up regularly in discussion.

[20] The objection takes off from the claim that very often the properties associated with a given word by different people differ markedly: names for towns and people are particularly clear cases, it is urged. What, then, runs the objection, are the associated properties that, on the description theory, settle the reference of such words?

There are two things to say in response to this objection. The first is that cases where the properties associated with a given word differ markedly from one speaker of, say, English to another are not as common as is often claimed. We noted in our discussion of the objection from ignorance and error that description theorists distinguish the properties associated, in the sense that settles reference on their theory, with the word `Paris', from the properties that Paris is believed to have. It is obviously true that different speakers of English often attribute very different properties to Paris, but it is plausible that the properties they associate with `Paris' are very similar and include, as a prominent member, being a certain kind of causal origin of the use of the word.

The second is that there is only an objection here if there is a problem about holding that, in such cases--be they common or rare--the answer to the rhetorical question varies from speaker to speaker. It sometimes seems to be thought that there is a serious problem with allowing that different, competent speakers of the same public language might associate with a proper name like `Paris' very different sets of properties. But it is hard to see what this problem might be. After all, by suitable mutual quizzing, we could identify these differences, if we thought it worth the trouble. It would, of course, be a nuisance if the city that had the properties I associate with `Paris' was a different city from the one that has the properties you associate with `Paris'; arrangements for us to meet made using the word `Paris' would go badly astray. But this is unlikely to be the case, and, moreover, we know that this is unlikely to be the case.

Communicating how things are by the use of words turns on associations between words and properties in the minds of speakers and hearers, as we said near the beginning. In the ideal case, the properties associated by speaker and hearer will be the same, but in practice this is often far from the case. As long as we can bring the properties associated by speakers and hearers into line when it matters, no great harm results from divergences from the ideal, especially when the focus is more on getting our paths to cross than on the nature of the place where our paths cross.[21]

Sometimes, the objection from divergent sets of properties is phrased on the assumption that the set of properties associated with a name is the meaning of that name according to the description theory, and the objection is then put in some such terms as, "But, by definition, persons who share a language give words the same meaning. So the description theory is refuted by the fact that people who share a language do sometimes associate quite different properties with a word in their common language." But if that is the criterion for sharing a language, we should say that few people share a language, though very many go close enough for it not to matter that they don't. Of course, we might stipulate that the term `meaning' be used in such a way that people we normally count as sharing a language give words like `Paris' or `Aristotle' the same meaning. This would be a stipulation, for, as we normally use the term, it is a commonplace that people who count as speaking the same language may give words different meanings. On this stipulation, it is indeed the case that we could not call the set of properties associated with names like `Paris' and `Aristotle' their meanings; indeed, most likely, only the referents of the words would be eligible. But this would not be an objection to the description theory of reference, but rather the product of a rather strained use of the word `meaning'. Or consider Donnellan's way of putting what is essentially the same objection:

It seems to me ... absurd to suppose that a beginning student of philosophy, who has learned a few things about Aristotle, and his teacher, who knows a great deal, express different propositions when each says `Aristotle was the teacher of Alexander'.[22]

This is no doubt true on some meaning of the term `proposition', but not, I would argue, on the meaning where it means something like the thought about how things are expressed by the sentence. Surely, it is entirely plausible that the beginning student and the experienced teacher would use the sentence `Aristotle was the teacher of Alexander' to make different claims about how things are. And, as we said near the beginning, the description theory of reference is really a view about the role of certain words, including names like `Aristotle', in serving to say how things are--they do so via associated properties--along with the claim that the relation between the words and their associated properties is (one of the relations) properly called `reference'.

It might be objected that there must be some `common message' which people we count as competent in, say, English use the sentence `London is large' to convey to each other. Isn't there something we English speakers `share around' when we use that sentence? But description theorists can happily agree. They can point out that it is often a presupposition, and a known one, that there is a single object that is being referred to by the word `London' in the mouths of the competent. In that case, there is, by their lights, an obvious candidate to be an associated-property-in-common, namely, being the thing which has the properties associated with the word `London' by the various competent speakers of English. We take it that the various competent speakers associate their own set of individuating properties with the word `London', and so one thing we share around is the view that the object satisfying the various sets is large. (However, what we share around is not what we say that someone believes when we say that they believe that London is large. We are not saying that they believe that there is some single thing with a certain set of associated properties which is large, but rather that there is some single thing with the associated set of properties and that they believe of it that it is large. And we may or may not require, in addition, that they are acquainted with it under the name `London', or some near relative like `Londres'.)[23]

The objection from the history of science

This objection is that the history of science provides us with many cases where the description theory gets the reference of various terms wrong. There are, it is argued, many cases where advances in scientific understanding lead us to revise our views concerning the reference of some term in a way that cannot be explained in terms of the description theory. An example is the term `liquid'. The property we folk associate with the term `liquid' is flowing. This suggests that we can say, without any fear of refutation, that glass is not a liquid (at room temperature) simply on the basis that it does not flow (at room temperature), but in fact science tells us that glass is a liquid.

The quick reply to this objection is that glass is a solid and not a liquid, on the folk meanings of the terms. There is, however, more to say. The first time it is explained to us in the science classroom that glass is a liquid, our response is not that this is a piece of re-definition, justified by the fact that the new definition picks out a more theoretically interesting class; it is that glass's being a liquid is an interesting discovery of science. (Or anyway this was my reaction and, as far as I can recall, that of the others in my class.) How come, if, as I insist, glass is not a liquid on the folk meaning? The answer is that context affects the meanings we give terms, as has been widely noted. What counts as a square in geometry is very different from what counts as a square in day-to-day talk. In particular, the very fact that one is in the science classroom affects the properties associated with terms like `liquid' and `solid'. The term `solid' becomes associated, by the very act of entering the classroom, with having the scientifically interesting property that typically explains resistance to deformation and failure to flow. What you then learn in class is that this property is that of having a certain kind of crystalline structure, and that glass and tar lack this structure. They have the scientifically interesting property that typically explains flowing, and they therefore count as (super-cooled) liquids in the classroom.

A similar point applies to an example like `kinetic energy'. It is sometimes suggested that in Newtonian physics the property associated with the term is 1/2mv2, and so that the description theory cannot explain how Einstein managed to show that kinetic energy is not 1/2mv2.[24] However, when you read Newtonian physicists it is clear that their equation of kinetic energy with 1/2mv2 was the outcome of sophisticated theory. It was not a stipulation in the sense that the metre bar in Paris was. The property they associated with the term `kinetic energy' was that of playing a certain theoretical role, a role which they were convinced was played by 1/2mv2. What happened when relativity theory came along is that this role was shown to be occupied by something other than 1/2mv2, though it approximates it closely when v is not large; or, better, that nothing occupies exactly the role Newtonians gave 1/2mv2, but that which satisfies the relativistic expression occupies a role close enough for us to say that it is kinetic energy. Of course, it is open to any Newtonian physicist to insist that what they meant by `kinetic energy' was 1/2mv2, and that, in consequence, what they said when Einstein came along was not that kinetic energy had been shown to be something other that 1/2mv2, but rather that Einstein had shown that kinetic energy did not play the theoretical role they thought it did. I think these remarks would reveal their usage to be non-standard, but so what?

This concludes the sceptical tour. I now turn to the promised, brief concluding remarks on reference failure.

Two issues about reference failure

Reference failure in non-modal cases

I said (controversially) that the Russellian treatment of definite descriptions is compatible with regarding definite descriptions as referring expressions. But many insist that if `the F' in `The F is G' is properly speaking a referring expression, then when `the F' fails to refer, when there is no unique F, `The F is G' must lack a truth value. If this is right, the Russellian account is incompatible with definite descriptions being referring expressions, for on that account `The F is G' has a truth value--false, when there is no unique F. Why don't I follow this broadly Fregean perspective?

My reason is that the Lockean perspective on language marginalises the issue about whether or not `The F is G' lacks a truth value when there is no unique F. If language is a device for representing how things are, what is centre-stage when we discuss the role of `the F' in, say, `The F is G', is its contribution to how the sentence represents things as being. And we get a handle on what that contribution might be by asking about what is in common to all the ways things might be, to all the possible worlds, that make the sentence true, by virtue of the sentence starting with the words `the F'. And the answer, of course, is that the relevant commonality is that each such world contains a single F, which is G. This holds independently of what to say about the truth value of `The F is G' in the worlds where there is no unique F. What matters is how things are in the ways that would make the sentence true, not whether we should describe the sentence as false or as lacking a truth value in the other worlds.

Of course, we could stipulate that in order for `the F' to count as a referring term in `The F is G', it must be the case that `The F is G' lacks a truth value when there is no unique F, but we would then need another term to describe the status of `the F' on the Russellian view, and if I am right, the feature picked out by this other term would be the more important one in discussions of word-world relations.

Representationalism also marginalises the issue about whether sentences of the form `N is G', where `N' is an ordinary proper name which fails to refer, are false or lack a truth value. Most supporters of the description theory of reference say that it is false, but they don't have to in order to be good description theorists. What is central is the contribution of `N' to how the sentence represents things as being, and that is given by the nature of the ways things are in which the sentence is true, not those in which it fails to be true (whether by being false or by lacking a truth value).[25] However, I should emphasise that the contribution that N makes to how the sentence represents things as being needs to be understood in terms of what is sometimes called the A-intension of, sometimes the diagonal proposition expressed by, and sometimes the primary intension of, the sentence.[26] It is obvious that the properties associated with `water' and with `H2O' differ, so if we want this difference to show up in a difference between the ways, e.g., that `Water is plentiful' and `H2O is plentiful' represent things as being, we need to think in terms of the worlds such that if they were actual, they would make the sentences true. And, as we in effect noted in discussing the second version of the Twin Earth case, these worlds are not the same. The first is the set of worlds where the clear, potable stuff we are acquainted with and call `water' is plentiful; the second is the set of worlds where H2O is plentiful.

Reference failure in a modal case

There is an important difference between

(5) Sherlock Holmes might have existed


(6) Santa Claus might have existed.

I take it that (5) is false because `Sherlock Holmes' is a term for a fictional character. This means that, at most, and leaving aside problems to do with whether the Holmes stories are internally consistent, a person might have existed who did the things the Holmes stories would have represented Holmes as doing had the stories been works of history.

The situation with (6) is trickier, because `Santa Claus' is simply a name that fails to refer (like `Vulcan' I take it, and let's so assume), not a term for a fictional character. What should description theorists say about it? The obvious thing for them to say straight up is that (6) is true. `Santa Claus exists' represents things as being a certain way on their view, namely, the way consisting of the person with the property associated with `Santa Claus' existing. This way is possible. Ergo, we count (6) as true. However, as a result of reading Stuart Brock, I now think that there is a good case for saying that description theorists should hold that (6) is false, and so that description theorists can agree with Kripke (1980) about (6).[27]

Description theorists hold that names are associated with properties. They do not, though, have to hold that a given name is associated with the same property in every context. Indeed, it would be foolish of them to do so. No-one should think that the property associated with `London' when surrounded by quotation marks is the same as the property associated with it when it appears in `London is large'. However, they must insist that there are graspable regularities in the associations between names and properties, otherwise language would fail to serve its communicative function. Flags would not be much good for telling about divers below if the rules governing which flag is used to indicate this changed in unpredictable ways. The same goes for words. It would, in any case, be simplest (most `innocent') if the property associated with `Santa Claus' in all of: `Had things been thus and so, Santa Claus would have been such and such', `Santa Claus brings presents', `Santa Claus might have existed' and `Santa Claus exists', were the same property.

Moreover, we know that, in order to get the right answers in counterfactual contexts, we must associate with `Santa Claus' the property of being actually so and so. Otherwise, we could shift the reference of `Santa Claus' by making a suitable counterfactual assumption, and we can't: the name is rigid under counterfactual assumptions. Suppose, then, that the property associated with `Santa Claus' in `Santa Claus exists' and in `Santa Claus might have existed' is that of being actually C--being the actual person who is the subject of the `Santa Claus' stories and who did enough to mean that these stories have some basis in fact, as it might be. Now, because Santa Claus does not exist, description theorists must say that nothing actually is C. But then Santa Claus does not exist in any world, for nothing in any world has the property of being actually C--to be actually C is to be C in the actual world, and nothing is C in the actual world. It follows that (6) is false according to description theorists if (if) they adopt a sufficiently simple (or `innocent') hypothesis about the property associated with `Santa Claus'.

Brock sees this as a problem for description theorists. He insists that it is a datum--defeasible but nevertheless of considerable weight--that (6) is true. I see it as an interesting discovery about the simplest thing for description theorists to say. In my view, the intuition that (6) is true can be well handled by the observation we made earlier, namely, that the way `Santa Claus exists' represents things as being is indeed possible, and it is easy to conflate this fact with (6).

How can the description theorist who holds that the property associated with `Santa Claus' in `Santa Claus exists' is that of being actually C allow, given that nothing is actually C, that the way `Santa Claus exists' represents things as being is possible? By drawing on the point emphasised earlier: how a sentence represents things as being is given by the A-intension of the sentence, by the worlds such that were they actual, the sentence would be true at them; and there is a C in some such worlds. Although there is no C in the actual world, there are worlds such that if they were actual, there would be a C.

Frank Jackson 7 January 1998

Philosophy Program

Research School of Social Sciences

Australian National University

ACT 0200

Bigelow, John (1988), The Reality of Numbers, Oxford, Clarendon Press.

Braddon-Mitchell, David and Frank Jackson (1996), The Philosophy of Mind and Cognition, Oxford, Basil Blackwell.

Chalmers, David (1996), The Conscious Mind, New York, Oxford University Press.

Devitt, Michael (1996), Coming to our Senses, Cambridge, Cambridge University Press.

Devitt, Michael and Kim Sterelny (1987), Language and Reality, Oxford, Basil Blackwell.

Donnellan, Keith (1972), `Proper Names and Identifying Descriptions' in Donald Davidson and Gilbert Harman, eds, Semantics of Natural Language, Dordrecht, Reidel, pp. 356-79.

Evans, Gareth (1982), The Varieties of Reference, Oxford, Clarendon Press.

Frege, Gottlob (1984), Collected Papers on Mathematics, Logic, and Philosophy, Brian McGuiness, ed., Oxford, Basil Blackwell.

Jackson, Frank (1998), From Metaphysics to Ethics: A Defence of Conceptual Analysis, Oxford, Clarendon Press.

Kripke, Saul (1980), Naming and Necessity, Oxford, Basil Blackwell.

Kroon, Fred (1978), `Causal Descriptivism', Australasian Journal of Philosophy, 65: 1-17.

Lewis, David (1969), Convention, Cambridge, Mass., Harvard University Press.

Lewis, David (1984), `Putnam's Paradox', Australasian Journal of Philosophy, 62: 221-36.

Lewis, David (1986), On the Plurality of Worlds, Oxford, Basil Blackwell.

Lewis, David (1997), `Naming the Colours', Australasian Journal of Philosophy, 75: 325-42.

Linsky, Leonard (1977), Names and Descriptions, Chicago, Chicago University Press.

Locke, John (1690), An Essay Concerning Human Understanding.

Putnam, Hilary (1962), `The Analytic and the Synthetic', in Herbert Feigl and Grover Maxwell, eds, Minnesota Studies in the Philosophy of Science, III, Minneapolis, University of Minnesota Press.

Putnam, Hilary (1975), `The meaning of "Meaning"', in Mind, Language and Reality, Cambridge, Cambridge University Press.

Sainsbury, R.M. (1993), `Russell on Names and Communication' in A.D Irvine and G.A. Wedeking, eds, Russell and Analytic Philosophy, Toronto, University of Toronto Press.

Searle, John (1983), Intentionality, Cambridge, Cambridge University Press.

Slote, Michael (1966), `The Theory of Important Criteria', Journal of Philosophy, 63: 211-24.

Stalnaker, Robert (1978), `Assertion', in P. Cole, ed., Syntax and Semantics: Pragmatics, vol. 9, New York, Academic Press, pp. 315-32.

Strawson, P.F. (1959), Individuals, London, Methuen.

Strawson, P.F. (1971), Logico-Linguistic Papers, London, Methuen.

Wiggins, David (1975), `Identity, Designation, Essentialism and Physicalism', Philosophia, 5: 337-62. [1] This is a substantially longer, significantly re-worked version of a paper that is to appear in Philosophical Perspectives, ed. James E Tomberlin, vol. 12. I am greatly indebted to Sam Guttenplan, Richard Holton, Stuart Brock, David Braddon-Mitchell, Lloyd Humberstone, David Lewis, Jakob Hohwy, Bruce Langtry, Graham Oppy, Karen Green, and many, many discussions, especially at the ANU and Monash University.

[2] See, e.g., Locke (1690), book III, Kroon (1978), Linsky (1977), Searle (1983), Lewis (1984), Strawson (1971), Sainsbury (1993), Lewis (1997), and Braddon-Mitchell and Jackson (1996).

[3] Which, if either, is conceptually prior: the association between sentences and ways things might be, or between words and ways objects might be, can be left to one side for present purposes.

[4] Of course, in our relaxed sense of `property', having a goodly number of ... counts as a single property.

[5] Devitt (1996), p. 159. See also Devitt and Sterelny (1987).

[6] Searle makes this point in a number of places, see. e.g., Searle (1983), p. 233.

7 Locke (1690), book III, ch. 2, SS 2. I am indebted here to discussions with Monima Chadha.

[8] For how a convention might be implicit, see Lewis (1969).

[9] Devitt (1996), p. 160. Incidentally, I am not sure why Devitt says that the `association of descriptions with a word is an inner state of the speaker'--I happen to agree, but am unsure why the point is being taken as a given, in view of the popularity of externalist accounts of psychological states.

[10] If you hold that definite descriptions do not refer, the point could be made with descriptive names or rigidified definite descriptions. I argue against the view that definite descriptions do not refer below. Incidentally, Devitt, both in publications and in personal communications, makes it clear that thinks that the description theory is correct for some words, but I am not sure where he stands on the objection above that this means the argument in the quoted passage is too good.

[11] Kripke (1980), p. 91. Kripke allows, of course, that descriptions alone can fix the reference of names and mentions `Neptune' as a possible example.

[12] And, as we will note later, for a given term, one rubric might be the correct one in one context, the other in another; rigidity might be context-relative.

[13] Evans (1982), p. 57.

[14] Evans (1982), p. 57.

[15] This is, perhaps, the most common objection to the description theory. I take the name of the objection from Devitt and Sterelny (1987).

[16] Putnam (1975), p. 226. Strictly speaking, I should not be discussing this example because we are restricting ourselves to words like "London' and `water' but ... .

[17] The earliest stressing of the importance of causal links I know is Strawson (1959), ch. 1.

[18] This issue is discussed at length by Kripke (1980), lecture II. See also Devitt and Sterelny (1987), p. 50.

[19] See, e.g., Devitt and Sterelny (1987), p. 50.

[20] Evans is discussing a passage from Wiggins (1975).

[21] When the properties diverge, description theorists can distinguish speaker reference from hearer reference in the obvious way. The rough definition we gave near the beginning was, of course, of speaker reference. The point about divergences often not mattering is in Frege (1984), p. 158, n. 4. Of course, if what I said earlier is right, the divergences will be less common than often thought.

[22] Donnellan (1972), p. 316. I think what I say about this passage is essentially the same as the reply Sainsbury (1993), p. 15, puts (approvingly) in Russell's mouth. Incidentally, it seems in the quoted passage that Donnellan may be conflating the properties associated with `Aristotle' with the properties Aristotle is believed to have. But the point being made hereabouts is that divergence would not matter; it is independent of how common divergence is.

[23] See Lewis (1986), p. 33, on being `London' and `Londres'-acquainted.

[24] The example is often sourced to Putnam (1962) but it should be noted that he is using it to slightly different effect.

[25] But if I turn out to be wrong about this, I will insist that the issue of truth-value gaps is spoils to the victor.

[26] In, e.g. and respectively, Jackson (1998), Stalnaker (1978), and Chalmers (1996).

[27] Brock kindly showed me drafts of his PhD; the key points were first presented, I understand, at an AAP conference in July 1997.