Names
It’s like the authorship of the Iliad,’ said Mr. Cardan. ‘The author of that poem is either Homer or, if not Homer, somebody else of the same name.’
Aldous Huxley, Those Barren Leaves, Part V, Chapter 4.
PART
I - INTRODUCTION
What do names mean? When I say to you:
(1) Socrates is coming to dinner
What does ‘Socrates’ mean, if anything? What do I mean by ‘Socrates’? What do you understand by ‘Socrates’? And how do you understand it? These are the intuitive questions which I aim to examine in this essay. They are also the questions which, in one form or another, have been at the core of a philosophical debate which has raged from the publication of Mill’s ‘A System of Logic’ in 1843 to the present day.
To have any chance of reaching a satisfactory answer we need to know what issues the pre-theoretical formulation above conceals; in other words, we need to know what the questions themselves really mean. Unless we know what we’re asking, we’re unlikely to reach any very interesting answers. In Part II we’ll take some steps towards disentangling the different issues which a successful theory of names will have to tackle, as well as looking at the puzzles which have acted as an acid test for successive theories.
Only once we know what the questions are can we start to look at the range of answers that have been put forward. In Part III we’ll examine the main approaches to come out of the philosophical debate on names, from the description-based solutions of Frege and Russell to the direct-reference-flavoured theories of Kripke and Recanati, and we’ll see how they measure up to the data.
Finally, in Part IV I hope to draw together the analyses outlined in Part III, and, taking account of the problems they face, propose a direction in which a more comprehensive solution might lie. Setting our questions within a semantic framework along the lines of that suggested by Wilson and Sperber (1990), I hope to show that we get new answers because our theoretical framework encourages us to ask new questions.
Along the way we’ll come across certain old questions time and again: are names really part of language? If so, should we treat them just as we treat other referring expressions, or are names somehow different? If they’re not a part of language, does that mean they’re without meaning? These questions are either implicitly or explicitly answered in all the approaches we’ll be looking at in Part III and we’ll also address them in the solution suggested in Part IV.
PART II - THE QUESTIONS
What data should a theory of names be able to account for? At first pass both the data and the account might seem unproblematic: a theory of names should be able to tell us what names mean and clearly they mean no more nor less than the person or thing they stand for. No problem with the data, no need for a complicated theory. There is, however, a range of puzzles which would appear to weigh heavily against this simple account and which any satisfactory theory of names will have to account for. The puzzles can broadly be classed into two groups: those which result from the use of coextensive names and those which result from the use of empty names.
Coextensive names, names which refer to the same thing, present us with the most famous puzzle of them all: the puzzle of identity statements. The problem is this: how can ‘a = b’ if true, differ in meaning from ‘a = a’? As a boy, Peter lived next door to a family called the Blairs: John, Jane and their son Eric. Now, some years later, Peter is invited by his childhood friend Mary to a reading by the famous author George Orwell. At the reading Peter thinks Orwell looks slightly familiar and tells Mary so. She replies:
(2) George Orwell is Eric Blair
Peter is astounded to discover that his childhood friend is now one of Britain’s most successful novelists. The puzzle, then, is this: presumably Peter would not have been astounded had Mary, instead of (2), uttered (3) or (4):
(3) George Orwell is George Orwell
(4) Eric Blair is Eric Blair
Yet if, as our simple theory holds, names simply label things, (2), (3) and (4) should all be just as interesting to Peter as each other, since they would mean precisely the same thing as each other. Let us call the thing referred to by the name ‘Eric Blair’ x; clearly ‘George Orwell’ will also refer to x, since Eric Blair and George Orwell are one and the same person. (2), (3) and (4) would, then, under our simple theory, all have the propositional form ‘x is x’, not, on the face of it, a very informative statement, since it is necessarily true that all things are themselves. But, you may answer, that’s just how the identity predicate works: unlike other predicates it has the effect of creating a metalinguistic proposition. So what (2) really means is:
(5) The referent of the name ‘George Orwell’ is identical to the referent of the name ‘Eric Blair’
Our puzzle is solved, our simple theory reprieved. However, as Salmon (1986) points out, this puzzle is not, in fact, particular to the identity predicate at all. He offers the sentences:
(6) Hesperus is a planet if Phosphorus is
(7) Phosphorus is a planet if Phosphorus is
to illustrate the point. If we want to explain the problem away as an anomaly of particular predicates, we’re going to find ourselves left with a very long list.
The second puzzle thrown up by coextensive names relates to what Quine (1960) calls referential opacity. Under the metaphysical principle known as Leibniz’s Law, if x and y are the same object then they have the same properties. So, given that ‘George Orwell’ and ‘Eric Blair’ are names for the same person, if:
(8) George Orwell writes well
is true, then so is:
(9) Eric Blair writes well
So far so good. But there are certain contexts in which Leibniz’s law seems not to work. Consider the position of Peter when Mary invites him to the Orwell reading. He may well be very keen to go because he believes (8) to be true, i.e.:
(10) Peter believes George Orwell writes well
Now, if our simple theory were right, Leibniz’s Law would apply and wherever ‘George Orwell’ went, so would ‘Eric Blair’. Applied to (10) this gives us:
(11) Peter believes Eric Blair writes well
This proposition may be true, but, on the face of it, there is a clear intuition that it also may not: just because (10) happens to be true, there seems to be no reason to believe that (11) must be true; there in fact appears to be no logical relation between (10) and (11) at all. Faced with the unappealing alternative of ditching Leibniz’s Law, this looks like another good reason to abandon our simple theory.
Before abandoning it entirely, however, let’s look once more at what the simple theory says: names mean no more nor less than the person or thing they stand for. What, then, if there is nothing for a name to stand for? After the reading, Peter approaches George Orwell, who is sitting at a table signing book copies, and, having introduced himself, says:
(12) Winston Smith is a role-model of mine
Intuitively the meaning of (12) seems unproblematic. Yet under our simple theory it’s hard to see what this utterance can mean: Peter has predicated ‘being-a-role-model-of-mine’ of Winston Smith; Winston Smith, as a fictional character, does not exist; so Peter has predicated ‘being-a-role-model-of-mine’ of nothing. I hope the example in (12) has made it clear that this is not, in any sense, a non-standard use of names: we talk of fictional characters every day without difficulty.
Even more problematic is a special case of this puzzle: the puzzle of the true negative existential. Being in a particularly bad mood that evening, Orwell turns to Peter and says:
(13)Winston Smith doesn’t exist
Now if (13) is true, by the very fact that it’s true the name ‘Winston Smith’ must lack a referent. And if ‘Winston Smith’ lacks a referent there is nothing for not existing to be predicated of, in which case it’s hard to see how (13) could possibly be true. So, under the simple theory, we end up with the thorny problem that no sentence of the form ‘N does not exist’ can ever be true: if it’s false, it’s false and if it’s true, it’s false. Given the strong intuition that, in the actual world, (13) is true, this can’t be right.
In Part I, I promised to disentangle the different questions a theory of names must be able to answer, and this is where the disentangling starts. What, firstly, is the question that the puzzles above might shed light on? Since they appear to call into question our simple theory, they are likely to shed light on whatever question our simple theory was designed to answer, and it seems to me that that question is what do names mean? or, in other words, what is the semantic content of names? Now this may seem obvious, but it’s crucial that, from the outset, we separate this question from a related but distinct question: what is the nature of the connection between a name and its referent? As we’ll see in Part III, for some approaches the same theoretical machinery can answer both questions, but this is not so for all approaches: our simple theory, for instance, takes a clear position on what names mean but is silent on the question of how they mean it. Given this, we’d be well advised to keep the two questions as separate as possible. Let’s consider for a moment what this second question might mean, or rather, what the problem which lies behind it is. As George Orwell is sitting signing books, Peter and Mary bump into an old friend of theirs, Bob, who has turned up late and missed the reading. Bob asks Peter ‘which one’s George Orwell?’ and Peter, pointing at Orwell behind the table, says:
(14) That’s George Orwell
In this situation there seems a clear and unproblematic link between name and referent, between ‘George Orwell’ and George Orwell: Peter’s act of ostension. Bob just needs to look in the direction Peter is pointing and he will know who the referent of ‘George Orwell’ is. Bob then says to Mary:
(15) I think George Orwell writes better than Shakespeare
Mary, who has an inexplicable gap in her cultural knowledge, replies ‘which one’s Shakespeare?’ What does Bob answer? He can’t resort to ostension as Peter did: Shakespeare is, presumably, not in the room. So how can he set up a connection between his use of ‘Shakespeare’ and Shakespeare himself? It’s this problem that lies behind our second question.
While we’re in the business of disentangling, there are a couple more distinctions we might like to draw. Over the last thirty years the importance of separating out metaphysical and epistemological considerations has become increasingly clear. To take an example that we’ll come back to later, Kripke (1971) argues that the significance of identity statements like (2) has been misunderstood. The informativeness of these statements, says Kripke, comes not from the fact that they could have been otherwise; they are, in fact, necessarily true. Their informativeness comes, rather, from the fact that they are not true a priori: we cannot know they are true without looking at how the world is. This will be explained more clearly, I hope, in Part III; for now, we need merely be aware that there is a potentially significant distinction to be drawn here.
Another distinction that we should bear in mind is between the semantics of names and the role that pragmatic inference plays for a hearer attempting to determine the intended reference of a name; or, to put it another way, the difference between what a name means and what a speaker means when she uses a name. Again we’ll be seeing much more of this later, but just as a quick illustration, consider the following scenario: you’ve invited two friends to dinner, Mark and Michelle, who have never met each other before. You think the evening’s going rather well, until Mark leaves the room and Michelle says:
(16) I don’t think Max likes me. He’s been ignoring everything I say
In the context, you will clearly interpret (16) with Mark as the referent of ‘Max’. But what does this tell you? Does it offer a perhaps surprising insight into the semantics of names or is the effect purely pragmatic? I don’t wish to prejudge the answer to this question, merely to raise the semantics/pragmatics distinction as a significant issue, and one that our simple theory does not even begin to tackle.
But enough of questions, now it’s time for a few answers.
PART III - THE ANSWERS
The current debate over names starts from the position taken by J. S. Mill (1843). For Mill, referring expressions, or in his (rather confusing) terminology names, may carry two semantic attributes: denotation and connotation. A name denotes anything to which it can be truly applied, so, for instance, ‘man’ denotes the class of all men, whereas a definite description like ‘the king who succeeded William the Conqueror’ will denote only one individual.
At the same time as denoting a subject, some names will also connote, or imply, an attribute: as we’ve seen, ‘man’ denotes all men, but it does so because of properties that all men have in common and it is these properties which the name ‘man’ connotes. Once we know the connotation of man, we’ll be able to know what falls within its denotation. Now, for Mill, proper names differ in one crucial respect from other singular names: they lack connotation and, since Mill equates connotation with ‘meaning’, they also lack meaning:
The only names of objects which connote nothing are proper names; and these have, strictly speaking, no signification.... A proper name is but 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. (pp 36-37)
Now as far as names are concerned (I return to the use of ‘names’ to mean proper names) this view is very closely allied to the simple theory we outlined above: names mean no more nor less than the object which they are the name of. I hope it is already abundantly clear what sort of difficulties such an account is going to run up against: it can deal with none of the puzzles outlined in Part II. Moreover, while it tackles the question of what names mean - they mean nothing - it has nothing to say about what sort of connection links name and referent. This account, then, can’t explain the data; it’s significant more as a starting point for the debate than as an analysis in its own right. As we shall see later, however, it has much in common with some of the most recent views on names.
The first major challenge to accounts such as Mill’s came from Gottlob Frege, most notably in his 1892 paper, On Sense and Reference (Über Sinn und Bedeutung). In order to understand Frege’s views on the reference of proper names, however, we first need to look briefly at his general account of reference. It’s perhaps surprising that Frege’s account of reference has much in common with Mill’s: just as Mill believed most referring expressions to have a two-tiered semantics (denotation and connotation), so Frege believed that all expressions have a two-tiered semantics. Frege (1892) starts from a consideration of the type of identity statement we looked at in Part II: if ‘a=b’ is true, how can it differ in meaning from ‘a=a’? He points out that the informativeness of statements such as ‘a=b’ tells us two things: firstly, that such statements cannot just express a relation between the denotations of ‘a’ and ‘b’ (that would leave us with ‘a = a’ all over again); and secondly, they cannot just express a relation between the expressions ‘a’ and ‘b’. What does this mean? You and I are walking along a beach when I pick up a pebble and say:
(17) I’m going to call this pebble ‘Amy’
A few moments later I say:
(18) I’m also going to call this pebble ‘Bertha’
Improbable maybe, but, I hope you agree, not impossible. I can then make the true statement:
(19) Amy is Bertha
Frege’s point is that, even though (19) is an identity statement of the form we are now familiar with, it is uninformative: by the way the example is set up, we must already know that (19) is true, since the only thing we know about ‘Amy’ is that it refers to a particular object, which is also, concerning the same object, the only thing we know about ‘Bertha’.
Now what does this tell us? Well it seems to suggest that there must be some semantic element lying between name and referent, and it is this semantic element which Frege termed the sense (Sinn). If ‘a=b’ is informative, it is because ‘a’ and ‘b’ express different senses of the same referent. To see how the difference between sense and reference works, let’s look at the prototypical Fregean referring expression: the definite description. Boasting at a drinks party some weeks later, Peter says:
(20) The author of Animal Farm is the person I lived next to when I was small
Now, as we already know, if an utterance of (20) is to be true, ‘the author of Animal Farm’ and ‘the person I lived next door to when I was small’ must have the same referent. However each description is a complex semantic structure made up from the semantic contents of its individual parts, and it is through this complex semantic structure, through its sense, that each description presents its referent. Now what role, precisely, do Fregean senses play? Firstly, they constitute the semantic content of an expression: they answer the question what does that mean? But secondly, because there is a one-to-one sense to referent mapping, i.e. one sense can correspond to just one referent, sense determines reference: the referent of ‘the author of Animal Farm’ is whoever wrote animal farm.
Now, as I mentioned earlier, for Frege all expressions have a two-tiered semantics i.e. have both sense and reference. As we’ve seen, descriptions have something like a concept as their sense and an object as their reference. Sentences have a thought, or proposition, as their sense and a truth value as their reference. But what about names? It seems clear that names must have an object as their reference. But what could possibly constitute the sense of a name? After all, names are clearly not, on the face of it, complex semantic expressions like definite descriptions. Furthermore names are not bestowed because of specific properties: I am not called ‘George’ because I am particularly georgish.
So what might the sense of a name be? The notion that Frege originated was that names are, essentially, disguised descriptions. Let’s flesh this proposal out a little, by reconsidering (15), repeated here as (21):
(21) I think George Orwell writes better than Shakespeare
What, in this utterance, does Bill mean by ‘Shakespeare’? Under Frege’s analysis, Bill will have some description in mind, let us say ‘the English playwright and poet who wrote Hamlet’, such that the proposition he has expressed by his utterance would be unchanged if we substituted description for name. We then proceed to analyse the meaning of Bill’s utterance using the template we’ve already established for descriptions. At first sight this idea might seem unappealing: names, after all, just don’t look like descriptions. But before we dismiss it, let’s look at some of the advantages it offers. Firstly, it tallies with the intuition that we don’t really know what a name, say ‘N’, means, if we have no answer to the question ‘who is N?’. Strawson (1959, p 20) says:
...it is no good using a name for a particular unless one knows who or what is referred to by the use of the name. A name is worthless without a backing of descriptions which can be produced on demand to explain the application.
It also seems to have a pretty good go at solving our puzzles; it was, after all, largely designed to do so. For instance, the identity statement in (2), repeated here as (22)
(22) George Orwell is Eric Blair
can now be paraphrased as something like (20), or, more accurately:
(23) Whoever wrote Animal Farm is identical with whoever I lived next door to when I was small
Without going through them one by one, we can see that the same applies to all the other puzzles: since propositions containing names are no longer dependent on a particular individual, but, rather, on whoever fits a description, there’s no problem with, for instance, Peter holding different beliefs about Eric Blair and George Orwell’s writing skills, and so on. This analysis also has the appealing result of dealing with the semantic content of names and the connection between name and referent with one theoretical tool: the Fregean sense performs both functions.
Starting from a very different point, Bertrand Russell arrived at conclusions on names very similar to those of Frege. Russell approached the question from the viewpoint of his theory of descriptions, originally laid out in Russell (1905). Although there is no space here to go into the details of the theory of descriptions, we will need to grasp the basic principles in order to understand Russell’s view on names. For Russell there are two fundamentally different ways in which we may know an object: knowledge by acquaintance and knowledge by description. Russell (1911) sketches the difference: ‘I say that I am acquainted with an object when I have a direct cognitive relation to that object, i.e. when I am directly aware of the object itself’ (p. 108); ‘I shall say that an object is “known by description” when we know that it is “the so-and-so,” i.e. when we know that there is one object, and no more, having a certain property’ (p. 113).
Now the basic principle which underlies the theory of descriptions is that ‘every proposition which we can understand must be composed wholly of constituents with which we are acquainted’ (Russell 1911, p 117). Why should this be so? In the ‘realist’ semantic terms of the theory of descriptions, propositional constituents are real objects; as Neale (1990, p. 15) puts it, ‘every meaningful item of language stands for something real. The meaning of an expression is simply that entity for which it stands.’ Given this semantic background, Russell was determined to do away with the possibility that we might understand any proposition which contained a non-existent constituent, a possibility which seemed, to him, to lead down the road to logical absurdity and an infinite world of non-existent but real objects.
The implication of all this is that if we know an object by description, that object cannot be a constituent of any proposition which we are capable of understanding. Descriptions, then, cannot introduce propositional constituents. At first sight this may seem highly counterintuitive: ‘the author of Animal Farm’ would, on the face of, seem self-evidently to introduce a constituent to the proposition expressed by an utterance of (20), for instance. Russell (1905) provides the solution: descriptions and, indeed, what Russell calls ‘denoting phrases’ in general, do not, in themselves, have meaning: ‘a denoting phrase is essentially part of a sentence, and does not, like most single words, have any significance on its own account.’ Under Russell’s analysis, descriptions must be unpacked before they can be understood. ‘The F is P’ will, roughly speaking, be unpacked as:
There is at least one thing which is F
There is at most one thing which is F
That one thing is P
Whichever object satisfies these conditions will be, in Russell’s terminology, the denotation of ‘the F’. To give a slightly clearer example, the sentence:
(24) The boy who lived next door is famous
unpacks as something like:
(25) There is one and only one boy who lived next door and he’s famous
Whichever object is the one and only boy who lived next door is, then, the denotation of ‘the boy who lived next door’.
So what about names? On the one hand Russell held names, or rather what he termed ‘logically proper names’, to be the only denoting phrases which could serve to introduce a constituent directly into a proposition. On the other hand, he was faced with a problem: it is clearly possible to understand a proposition containing a name which lacks a referent; we only need to look at (12) to see an example of this. This led Russell inexorably to the same conclusion as Frege: names, as commonly used, are disguised descriptions. As Russell (1911) puts it, ‘the thought in the mind of a person using a proper name correctly can generally only be expressed explicitly if we replace the proper name by a description’ (p. 114). This allows Russell to unpack names in just the same way as he does descriptions (it also forces him dramatically to restrict the class of ‘logically proper names’, ultimately to ‘I’ and ‘this’ (Russell 1911)).
As with Frege’s account, the Russellian analysis treats propositions derived from sentences containing names as essentially object independent and, as with Frege’s account, Russell’s solution fares well against our puzzles (given that, as with Frege’s account, it was designed specifically with the puzzles in mind). Russell’s solution to the problem of referential opacity is particularly interesting, illustrating as it does the central role that scope plays in the theory of descriptions. Let’s look again at (10) and (11), repeated here as (26) and (27):
(26) Peter believes that George Orwell writes well
(27) Peter believes that Eric Blair writes well
Taking (26) first, we’ll need to replace ‘George Orwell’ with a description, ‘the author of Animal Farm’ say, giving us:
(28) Peter believes the author of Animal Farm writes well
Now, according to Russell there are two ways we can unpack (28), corresponding to (29) and (30):
(29) There is one and only one author of Animal Farm and Peter believes that one writes well
(30) Peter believes (there is one and only one author of Animal Farm and that one writes well)
In (29) the description takes wide scope, and has what Russell calls its primary occurrence; in (30) it takes narrow scope and has its secondary occurrence. If in propositional attitude contexts (‘A believes that B; hopes that B; regrets that B’ etc.) descriptions, and thus names, are read as having their secondary occurrence, then it is clear why substitutivity should fail: the content of the description forms part of the embedded proposition, and so part of the belief, hope etc. If descriptions take their primary occurrence, on the other hand, we should find Leibniz’s Law working fine, as in (8) and (9). It will be worth bearing the Russellian notion of scope in mind, as we shall have more to say about it and about how it compares to other analyses later.
Russell’s analysis shares with Frege’s the advantage of accounting for the connection between name and referent in the absence of ostension: a name is connected to its referent via its associated description; take the name, find the associated description and whoever satisfies it is the referent. Equally, for both Frege and Russell names are a part of language: they have descriptive content just as other referring expressions and they can be accounted for with the same theoretical machinery as other referring expressions. As we’ll be seeing later, this isn’t so for more recent theories.
Given that the Frege-Russell analysis of names as disguised descriptions seems to work so well in solving our puzzles and in accounting for the data, why should we need to look any further? Why not just accept that they’ve found the answer and go home? At the beginning of the 1970s a group of philosophers, led by Kripke, Donnellan and Kaplan, began to challenge the description-theoretic picture, pointing out that there were good reasons why this just couldn’t be the answer. Later we’ll be seeing the alternative analyses they proposed, but first let’s look at the problems which led them to challenge the Frege-Russell orthodoxy (for detailed examinations of most of these problems, see Donnellan (1970) and Kripke (1980)).
Both Frege and Russell accepted that, under their analyses, the particular description associated with a name on a given use was a matter for the speaker. As Donnellan (1970, p 340) points out, this leads to some very counterintuitive results:
It seems to me, though evidently not to them (Frege and Russell), 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’.
But, even if we accept that the same sentence expresses different propositions in the mouths of different speakers, we’re still faced with some seemingly insuperable difficulties. Let’s consider how we might choose which description to associate with a name; take the name ‘George Orwell’, for instance. Now we might choose to associate ‘George Orwell’ with the description ‘the author of 1984’, in which case, according to Frege-Russell, whoever wrote 1984 will be the referent of ‘George Orwell’. Now this just can’t be so. Consider the sentence:
(31)
George Orwell didn’t write 1984
None of us will have any problem imagining a situation in which it was discovered that (31) was true, i.e. that 1984 was written by someone other than George Orwell. Yet if ‘George Orwell’ were to mean the same as ‘whoever wrote 1984’, (31) would be a contradiction. It would mean:
(32)
Whoever wrote 1984 didn’t write 1984
The fact that George Orwell did, in fact (or so I believe at the moment), write 1984, is entirely contingent; it forms no part of what we mean by ‘George Orwell’. By parity of reasoning I hope it’s clear that pretty well any other single description we might care to choose would fall at the same hurdle (although, as we’ll see later, there is one description which might have a chance).
Let’s, for a moment, stay in the imaginary world in which George Orwell did not, as we currently believe, write 1984; let’s say (I agree improbably) that it was, in fact, written a hundred years earlier by Charles Dickens. This fact is not, however, common knowledge: most people still believe Orwell to have been the author. In this imaginary world, Peter says:
(33) I met George Orwell last night
Now if Peter’s use of ‘George Orwell’ is backed by ‘the author of 1984’, who is he referring to when he utters (33)? Orwell? No, it appears he is referring to Dickens, since Dickens is, unbeknownst to Peter, the denotation of ‘whoever wrote 1984’. This seems, at best, a highly counterintuitive result.
The problems for description theories don’t end there: I know very little about Turkmenistan and Tajikistan; I’d go so far as to say that all I know about them is that they’re both former republics of the Soviet Union. Given that that’s all I know, under Frege-Russell any use I make of either name can only be backed by the description ‘a former republic of the Soviet Union’. Now the denotation of this description includes, obviously enough, both Turkmenistan and Tajikistan along with a host of other countries, in which case, the reference of my ‘Turkmenistan’ is the same as the reference of my ‘Tajikistan’. But when I utter:
(34) Tajikistan has a population of over one million
what am I referring to? It seems clear that I am, in fact, referring to Tajikistan: my utterance will be true or false dependent upon the population of Tajikistan and will be entirely unaffected by the population of Turkmenistan. Perhaps we might like to argue that, in fact, the description I associate with Tajikistan looks more like ‘the former republic of the Soviet Union which is not Turkmenistan, Georgia, .......Ukraine’ taking in, along the way, the names of all the former republics of the Soviet Union that I know of; what, in essence we’d be trying to do, then, is boost our description from indefinite to definite. Sadly this helps us not a bit. Consider the regrettably large number of former republics which I fear I may never have heard of; each of these would fall equally well within the denotation of our new description, yet, again, their population is neither here nor there as far as (34) is concerned (of course only former republics of which I am unaware will fall within the denotation: as soon as I become aware of a new republic, X say, I will, under the above argument, associate a description with X which will include ‘not Tajikistan’).
In order to try to combat objections such as these, many writers have tried to produce a more sophisticated version of the description-theoretic approach. The first major attempt introduced the notion of descriptive clusters to the argument: the referent of a name is not whatever satisfies a particular related description, but whatever satisfies all, or enough, of a cluster or family of descriptions. Searle (1958, pp 170-1) provides one of the clearest formulations of this view:
Though proper names do not normally assert or specify any characteristics, their referring uses nonetheless presuppose that the object to which they purport to refer has certain characteristics. But which ones? Suppose we ask the users of the name “Aristotle” to state which they regard as certain essential and established facts about him. Their answers would be a set of uniquely referring descriptive statements. Now what I am arguing is that the descriptive force of “This is Aristotle” is to assert that a sufficient but so far unspecified number of these statements are true of this object.
This analysis has certain clear advantages over the traditional Frege-Russell orthodoxy: the responsibility for choosing descriptive content is shifted from speaker to linguistic community, as the above passage from Searle (1958) makes clear. No longer need we worry about different speakers expressing different propositions with the same sentence: they’ll all be using the same name to refer to the same object, as defined by their linguistic community.
It also no longer matters that we can take any individual description and point out that it expresses a merely contingent property of the referent or, indeed, does not express a property of the referent at all; after all, only ‘a sufficient number’ of descriptions need be true of the referent. All that is now necessary is that, as Searle (1958, p. 172) puts it, a referent should have the ‘logical sum, inclusive disjunction, of properties commonly attributed to him’.
This seems to work better, so what’s the problem? Both Kripke (1980) and Donnellan (1970) point out that a referent does not need to have ‘the logical sum...of properties commonly attributed to him’. He, in fact, needs none of the properties commonly attributed to him. Consider (35):
(35) Homer didn’t write either the Iliad or the Odyssey
Now (35) does away with pretty well every property which I know to be commonly attributed to Homer, so who does ‘Homer’ in (35) refer to? Does it refer to whoever wrote the Iliad and the Odyssey? No, there’s a clear intuition that ‘Homer’ still refers to Homer (this is Kripke’s rigidity, of which more soon).
Before we turn to the analysis offered by direct-reference-flavoured theories, let’s look at an interesting phenomenon pointed out by Kripke (1980). Kripke mentions that, although most names do not work as hidden descriptions (or descriptive clusters), there are some names which do work in this way: the example he gives is ‘Jack the Ripper’. It seems to me illuminating to see just how differently names such as this behave from ‘normal’ names. Compare (35) with (36):
(36) Jack the Ripper didn’t commit any of those murders
Now it’s very hard to see what (36) could mean, after all ‘Jack the Ripper’ is simply the name we give to whoever committed the murders commonly attributed to Jack the Ripper. You notice, also, that however strong may be the evidence we have that Jack the Ripper was in fact some person X, the sentence:
(37) X didn’t commit any of those murders
will always be meaningful. This seems pretty strong evidence that normal names do not, in fact, function as any sort of concealed description.
Now consider for a moment the following two sentences:
(38) George Orwell worked for the World Service
(39) The author of 1984 worked for the World Service
Who do the truth conditions of (38) and (39) depend on? In the actual world it’s clear that both depend on the same person, George Orwell, and, in fact, both sentences are true in the actual world. But imagine a hypothetical world in which, as we’ve already discussed, George Orwell didn’t in fact write 1984, Charles Dickens did. On whom, under those circumstances, would the truth conditions of the two sentences depend? Putting aside the issue of definite descriptions used referentially in the sense of Donnellan (1966), the truth conditions of (39) would now depend on Dickens and the proposition expressed by (39) would be true iff, in that hypothetical world, Dickens worked for the World Service. It seems clear, however, that the truth conditions of (38) would still depend on George Orwell, and that the proposition expressed by (38) would remain the same, and would be true in every world in which Orwell did work for the World Service. The truth conditions of (38) are singular, the proposition object dependent. It’s this feature of names which Kripke (1980) termed rigidity: names designate rigidly in that they designate the same individual in all possible worlds.
That’s all very well, you may answer, but if we’re in the business of imagining other worlds, why not imagine a world in which the person we call ‘Charles Dickens’ was called ‘George Orwell’? In that case surely the truth conditions of a sentence such as (38) would no longer depend on the same person? Well, it’s certainly true that someone uttering (38) in that world would now not be expressing a true proposition, but that is because they’d be expressing a proposition about Charles Dickens, and Dickens did not work for the World Service (although we could, of course, devise a hypothetical world in which he did). The point to bear in mind is that what we’re trying to investigate is what referring expressions refer to as we use them, not as they might be used in hypothetical worlds; they might be used any way at all in hypothetical worlds. We’re then looking to other possible worlds to see upon which individuals the truth conditions of sentences containing referring expressions as we use them will depend in those possible worlds.
This distinction between the actual world and other possible worlds is formalised in Kaplan’s (1977) distinction between context of utterance and circumstances of evaluation and his parallel distinction between character and content. Since Kaplan originated the idea of direct reference, it is worth us taking a couple of moments to familiarise ourselves with his ideas. The central notion of direct reference, as Kaplan saw it, is that there are some expressions which refer without the help of anything like a Fregean sense: they simply introduce an individual object into the proposition expressed by a sentence in which they appear. As Kaplan (1989, p 569) puts it:
The directly referential terms goes directly to its referent, directly in the sense that it does not pass through the proposition. Whatever rules, procedures or mechanisms there are that govern the search for the referent, they are irrelevant to the propositional component.
The next question is how, according to Kaplan, this direct reference works. Just as for Mill and Frege, Kaplan sees referring expressions as having a two tiered semantics; whereas for Mill it was denotation and connotation, and for Frege sense and reference, for Kaplan it is character and content. It is perhaps easiest to understand these notions if we take them in parallel with the notions of context of utterance and circumstances of evaluation. In order to establish whether an utterance of a particular sentence is true or false in a given set of circumstances of evaluation (i.e. whether it is true or false in a possible/hypothetical world) we must know what the content of that utterance is; in order to know what the content of an utterance is we will need to know both what the character of the sentence is and also the context of utterance. Character is therefore a matter of type, content of token. The distinction is not immediately easy to grasp, since, under Kaplan’s analysis, expressions tend either to have identical character and content or identical content and reference. Let’s then look at the distinction in operation:
(40) I want a chocolate ice cream
How would we set about evaluating whether an utterance of (40) was true or false? Well first we’d need to ask what its content was. Much of the content is clear, since the phrase ‘a chocolate ice cream’ has identical character and content: that is to say, in any context of utterance, ‘a chocolate ice cream’ is going to mean the same thing. The same can obviously not be said for ‘I’: ‘I’ will clearly have entirely different content according to who has uttered (40); if I’ve uttered it, then I will be the content of ‘I’, if you’ve uttered it then you will be the content of ‘I’. But how can we tell on a particular utterance who ‘I’ refers to? The answer Kaplan gives is that we look to the character of ‘I’, which will be something like ‘the speaker of this utterance’. We then put this together with the context of utterance to uncover the content of ‘I’. As we mentioned above, ‘a chocolate ice cream’ has identical character and content; central to Kaplan’s notion of direct reference is the idea that indexicals like ‘I’ have identical content and reference. This means that once content has been established by looking at the context of utterance, it is fixed once and for all and cannot be influenced by the circumstances of evaluation.
Now this is not only very closely allied to Kripke’s notion of rigid designation; it is also, interestingly, closely allied to Russell’s notion of primary occurrence which we saw earlier. When a Russellian denoting terms takes wide scope, that has the effect of fixing the reference against the context of utterance, just as happens in direct reference for Kaplan. This is not the last time we’ll have occasion to observe how closely allied descriptional and direct reference accounts are in some respects.
So where do names fit in to all this? For Kaplan, as for Frege and Russell, names are a part of language (although not necessarily part of any particular language); but, unlike the Frege-Russell analysis, Kaplan’s semantic system treats names as crucially different from other referring expressions (in this Kaplan’s approach is closer to Mill’s). As Kaplan (1977, p562) puts it: ‘in the case of proper name words, all three kinds of meaning - referent, content and character - collapse. In this, proper name words are unique’. So for Kaplan names are like indexicals in that they have a fixed character; unlike indexicals, however, they also have fixed content. But is that right? Do names really have fixed content? Consider (41):
(41) George is not here at the moment
Now it’s not hard to imagine that in different circumstances (41) could refer either to me or to, say, George Bush. Given this, doesn’t the content of ‘George’ vary according to circumstance? Not for Kaplan: under his analysis my ‘George’ and George Bush’s ‘George’ are different, albeit homonymous, names: ‘unlike indexicals like ‘I’, proper names really are ambiguous’ (Kaplan 1977, p. 562). As we shall see, more recent theories in the direct reference tradition part company with Kaplan at this point.
Before we look at those more recent theories, we might just take a moment to see how the Kripke and Kaplan accounts cope with the data we outlined in Part II. First it’s worth noticing that all we’ve looked at so far is what names mean, what their semantic content is; we’ve had nothing to say as yet about the connection between names and their referents and, unlike description theoretic approaches, direct reference approaches cannot rely on the same theoretical machinery to perform both tasks. Most direct reference theorists tend to hold to some version of the historical chain view (for examples, see Kripke (1980), Putnam (1975), Evans (1982) etc.) The basic idea is this: I decide to call my cat ‘Nathan’; I therefore connect ‘Nathan’ and Nathan through an initial arbitrary act of baptism. I then say to you ‘you must meet Nathan sometime; you’d really get on’; without ever having met Nathan, you can now use ‘Nathan’ to refer to whatever I referred to when I used ‘Nathan’. So you then say to another friend ‘George invited me to meet Nathan, that’s his cat you know’ and that friend is, in turn, connected to the ‘Nathan’ historical chain (or introduced to the ‘Nathan’ name-using practice in the terminology of Evans (1982)).
How about the puzzles, then? As I already mentioned, Kripke tackles the puzzle of identity statements containing names through appeal to the distinction between necessary truth and a priori truth: ‘a=b’, if true, is necessarily true, it is merely not true a priori. Now what does this mean? Let’s consider, as Kripke (1971) does, the time-honoured (and Frege-originated) example of:
(42) Hesperus is Phosphorus
The question we need to ask is: what might have been the case in other possible worlds with regard to the truth or falsity of this sentence? Well, it might certainly have turned out that we had, as in the actual world, given the name ‘Hesperus’ to the planet Venus as it appears in the evening; but that we had given the name ‘Phosphorus’ to some other heavenly body entirely. In that case it would seem, at first sight, as if (42) would be false and thus can only be a contingent truth in the actual world. But that runs into the problems we’ve already seen concerning ‘George Orwell’ and ‘Charles Dickens’: in that possible world ‘Phosphorus’ would not have been Phosphorus; we would be looking at the truth conditions of a different sentence because, under Kripke’s analysis, the words would not have the meaning they actually do. What is not possible, argues Kripke, is that the thing which is Phosphorus might not have been identical with the thing which is Hesperus, for they both are the planet Venus. (42) therefore expresses a necessary truth, just as ‘Hesperus is Hesperus’ does. But, as Kripke points out, this doesn’t mean that we can know (42) a priori, i.e. that we can know it without any empirical evidence. That is an entirely different matter (for further discussion, see Kripke (1979 a)).
Salmon (1994) suggests that the pragmatic approach of Kripke (1979 b) might also be pressed into service to account for the failure of substitutivity of coextensive terms in opaque contexts. The basic idea is that this puzzle cannot be semantic in origin, since it arises where there is demonstrably no semantic difference between two terms. The example Salmon offers is of Sacha who, while he knows what the words ‘ketchup’ and ‘catsup’ refer to, does not know that the words are synonyms. Under these circumstances he may, according to his experience, hold contradictory beliefs containing the two words. So, for instance:
(43) Sacha believes that ketchup is a sandwich condiment
(44) Sacha believes that catsup is not a sandwich condiment
Since ‘ketchup’ and ‘catsup’ are perfect semantic synonyms, we must look to pragmatics to solve our puzzles.
What about the puzzles concerning empty names: how does the direct reference approach deal with them? Here the answer is not so clearly worked out. Donnellan (1974) uses the historical chain approach to tackle true negative existentials in what he calls ‘discourse about actuality’, i.e. discourse about referents the existence of which is presupposed. His basic notion is that some historical chains end in ‘blocks’, points from which no referent can be identified. Take ‘Santa Claus’: a child learns of Santa Claus from his parents as if the name referred to a real referent; there is thus a historical ‘Santa’ chain leading back from child to parents. But at that point there is a block: the parents have told their child a fictional story as if it were true, so there is no historical chain leading back from the parents’ story-telling to a real referent. Donnellan’s idea is that an utterance of the form ‘N does not exist’ is true iff there is a block in the historical chain leading back from that use of ‘N’. This may, as Donnellan (p. 30) puts it, point towards ‘the outline of a solution to some problems concerning nonexistence statements’ but it does little more. There is still much work to be done filling in the details, and until that work is done direct reference, sharing much, as it does, with Mill’s analysis, would, on the face of it, seem to face a stern challenge from these puzzles.
Before moving on to Part IV we should first take a brief look at one of the latest accounts of proper names put forward within the direct reference tradition: that of Recanati (1993). The central notion of Recanati’s analysis is that names fix their reference in much the same way as indexicals. That is to say, the semantics of a name (Kaplan’s character) interacts with the context of utterance to supply the name’s content. That content is then fixed for any circumstance of evaluation. So for Recanati, unlike for Kaplan, names have distinct character and content.
What, then, is the semantics of a name, for Recanati? All directly referential expressions, under Recanati’s analysis have, as part of their semantics, a feature which he calls ‘REF’: the function of REF is to indicate that the truth conditions of any proposition in which the directly referential term occurs will be singular, that is to say the proposition will be object dependent. So what does this mean? What it essentially means is that if a term carries REF, then whatever information is used to determine the reference of the term drops out before the truth conditional stage leaving just the referent behind.
What semantic content, then, do names have beyond the REF feature? In order to understand Recanati’s answer to this question we must first return to the question of whether names are a part of language. For Recanati, the connection between name and referent is one of social convention, not one of language, so while names as a syntactic class are a part of language, individual names are, in some sense, not. In support of this view, Recanati points out that, intuitively, if I don’t know what the noun ‘tiger’ refers to, I’m less competent at speaking English than someone who does know what it refers to; if I don’t know who ‘Ralph Banilla’ refers to, on the other hand, it seems we’d be less inclined to attribute this to lack of linguistic competence, and maybe more inclined to treat it as a deficit of general knowledge. It’s also interesting to note that names do not behave, as regards translation, in the same way as other linguistic expressions. Consider:
(45) Pedro vive en un castillo
Now if we want to translate this into English, we could translate it as:
(46) Peter lives in a castle
but it would be just as natural, if not more so, to translate it as:
(47) Pedro lives in a castle
If we failed to translate any other part of the sentence, we’d end up with nonsense:
(48) Peter vive in a castle
(49) Peter lives in a castillo
and so on. Maybe not incontrovertible evidence, but at least a hint that individual names are not a part of language in quite the same way as other syntactic categories.
Returning to Recanati’s account, beyond REF names have a semantic element which points the hearer towards the object associated with the name by social convention. This boils down to something along the lines of: N means ‘the person who is called ‘N’’. Now this analysis has many echoes throughout the literature, particularly in the description theoretic account. Kent Bach (1987, p. 135), explaining what he calls his Nominal Description Theory, says: ‘on this version of the description theory, a name ‘N’ is semantically equivalent to the description ‘the bearer of ‘N’’. It seems that the direct reference account and the description-theoretic account are coming closer together. It is true that, under the description analysis, names do not introduce singular truth conditions but, as I pointed out earlier, this effect can be reproduced by manipulating scopal considerations, and this is just what Bach (1987) does. I’m not, of course, saying that there is no difference between the two positions; merely that there is less of a difference than we might at first have thought: they are essentially two forms of the same metalinguistic thesis.
Given that the account I intend to sketch has much in common with Recanati’s account, let’s now move on to Part IV.
PART IV - TOWARDS AN ALTERNATIVE ACCOUNT
Before putting forward some suggestions towards a more comprehensive account of names, let’s step back for a moment to look at what we’ve seen already. We’ve seen, firstly, that an account such as Mill’s, which holds that names are simply unmeaning labels, is going to be unable to deal with the puzzles we laid out in Part II. Next we saw that description-based accounts such as Frege’s and Russell’s are able to account for our puzzles, since on a description-theoretic analysis names serve not to pick out an individual but whoever satisfies the associated description. We also saw, however, that descriptive names appear to fall foul of the objections raised by direct reference theorists, particularly Kripke’s notion of rigidity. Finally we’ve looked at how a variety of approaches within the direct reference tradition account for rigidity but we’ve also seen that solutions to all the puzzles in Part II have yet, within direct reference theory, to be fully and satisfactorily worked out.
The obvious question, then, is whether there is any way that we can take the best from the description-theoretic tradition (i.e. its ability to handle the puzzles) and combine it with the best of direct reference (i.e. its compatibility with rigidity).
One of the major difficulties for traditional description theory is that, as Kripke (1980) points out, any descriptive properties associated with a name must be necessarily true of the name’s referent; and, as we have seen, necessarily true properties are thin on the ground. What, then, if we could associate properties, something like a Fregean sense, with a referent without those properties becoming necessary? After all, no-one would deny that we do associate properties with referents: we all believe that George Orwell did write 1984, even if, given sufficient evidence, we could abandon this belief. I believe that the anatomy of concepts suggested by Sperber and Wilson (1995, p. 86) gives us just the machinery we need to do this:
Formally we assume that each concept consists of a label, or address .... The information that may be stored in memory at a certain conceptual address falls into three distinct types: logical, encyclopaedic and lexical.
The picture, then, looks something like this:
CONCEPTUAL
ADDRESS
LEXICAL LOGICAL ENCYCLOPAEDIC
ENTRY ENTRY ENTRY
How might this picture help us then? The key thing, from our point of view, is that, under this analysis, descriptive information, stored in the encyclopaedic entry, is associated with a concept without being part of that concept. We can shuffle descriptive information around, lose some, gain some and so on, while the concept itself remains stable. Let’s suppose, then, that, at the propositional level names introduce not descriptions, as Frege would have it, not objects, as Kaplan would have it, but individual concepts; that is, concepts of individual objects, with an anatomy something like that illustrated above, although, as we shall see later, there is good evidence to suggest that individual concepts do not have the same sort of lexical entries as other concepts (this approach has much in common with Recanati’s views on de re concepts and de re modes of presentation).
What evidence might there be to support this? Firstly, it seems to me (though not to Russell who explicitly rejected the possibility) intuitively likely that we have individual concepts of the people we know or know of; after all, it’s hard to see what knowing (of) someone might mean if it doesn’t mean we have an idea or concept of them.
Secondly, if the connection between the use of a name and the name’s referent is mediated by an individual concept, that would provide a satisfying explanation for the puzzles outlined in Part II. Let’s have a look at how this works. Why should (2), repeated here as (50), be informative to Peter?
(50) George Orwell is Eric Blair
The answer is that, since Peter is unaware that George Orwell and Eric Blair are one and the same person, he has different individual concepts associated with the two names: he has a ‘George Orwell’ concept and a separate ‘Eric Blair’ concept. Once he has been told (50), so long as he believes the evidence for its truth is sufficiently strong, he will have to conflate his two concepts into one or add information concerning identity to both (for a similar approach, see Larson and Segal (1995)).
Much the same analysis can be applied to the failure of substitutivity in opaque contexts. Let’s consider (10) and (11) again. Why can we not replace ‘George Orwell’ in (10) with ‘Eric Blair’? The answer is that, since the embedded sentences are reported as beliefs of Peter’s, the connection between the names’ uses and their referents is not mediated by our concepts of George Orwell and Eric Blair, but by Peter’s concepts of them. And we already know that Peter has separate ‘George Orwell’ and ‘Eric Blair’ concepts, so we are not looking at a case of substitution of identicals: we are trying to substitute one thing for something entirely different. It’s worth noticing that this only applies if we read the names as falling within the scope of the belief operator; otherwise (10) and (11) do express identical de re propositions, and the connection between name and referent is mediated by the individual concepts that we hold of the two referents (if we do not know that George Orwell and Eric Blair are one and the same, we will not, of course, read (10) and (11) as expressing the same proposition on any reading, and so the question does not arise).
How about empty names? Well it seems to me that again, in the face of an individual concept analysis, the puzzle evaporates. Let’s assume that, as soon as we learn information about a new individual, we open a new individual concept (admittedly on this analysis some individual concepts will be very short-lived, but I don’t see this as too much of a problem). As soon as we read the first page of 1984, then, we open an individual concept for Winston Smith and start filing information away in the associated encyclopaedic entry (no doubt including the information that we believe him to be a fictional character and thus non-existent, although it not hard to imagine a situation in which this belief could be overridden). That clearly still leaves open the question of how individual concepts relate to the real world, but this is a different question.
How, then, does our account deal with rigidity? In order to answer this we need to look, again, at what names mean: what constitutes the semantics of a name. My view on this is closely allied to Recanati’s: it seems to me that some version of the metalinguistic thesis outlined at the end of Part III is unavoidably right (and in danger of becoming the consensus position). I believe, however, that approaching the question from a different theoretical perspective affords us a better explanation of the phenomenon of rigidity.
Within the relevance theoretic framework originated by Sperber and Wilson, there is a crucial distinction in the process of being drawn between two types of meaning: conceptual or representational meaning and procedural or computational meaning. At the heart of relevance theory is the notion that understanding utterances involves two complementary processes: linguistic decoding to retrieve a logical form and the manipulation of that logical form through pragmatic inference. Given these two types of process, Blakemore (1987) and others have looked at the idea that we should expect also to find two types of meaning: one that contributes to the retrieval of a logical form (conceptual meaning) and the other that places constraints on pragmatic inference (procedural meaning).
So procedural meaning can place constraints on inference after the propositional form has been retrieved: it can constrain implicatures. But what about explicatures? Wilson and Sperber (1990) suggest that procedural meaning may also constrain the retrieval of the proposition expressed. There are, after all, some obvious candidates, indexicals being the most obvious: it seems intuitively very plausible that the meaning of, for instance, pronouns, takes the form of computational instructions rather than conceptual content. Now this notion is getting very close to Kaplan’s notion of character, but it seems, to me, to have two significant advantages. Firstly it not only points out the distinction, it also provides an explanation for it: procedural meaning and conceptual meaning are fundamentally different kinds of meaning, corresponding to the cognitive difference between computation and representation. And secondly, as we saw earlier, Kaplan’s approach suffers from the problem that, for him, expressions tend either to have identical character and content, identical content and referent or (in the case of names) identical character content and referent. Now this seems somehow to be missing the right generalisation, drawing the lines in the wrong places. Under the relevance theoretic analysis there is no such problem: not every expression need have procedural meaning, as it need have character for Kaplan; indeed, as we’d expect, some expressions seem to have conceptual meaning, some seem to have procedural meaning and some seem to have both.
How might this notion apply to names, the