Are “Attributive” Uses of Deﬁnite Descriptions Really Attributive?

In this essay I argue that given Donnellan’s formulation of the attributive uses of deﬁnite descriptions, as well as Kripke’s [6] and Salmon’s [10] generalized accounts, most uses of deﬁnite descriptions that are taken to be attributive turn out not to be so. In building up to my main thesis, I ﬁrst consider certain problematic cases of uses of def-inite descriptions that do not neatly ﬁt into any category. I then argue that, in general, a complete deﬁnite description we use is complex, in which there is an embedded singular term that is used referentially. From this I conclude that an attributive use of a deﬁnite description is an extremely rare linguistic phenomenon, much less frequent than what Donnellan, Kripke, and Salmon have presupposed; so much so that the standard examples given by Donnellan of the attributive use of deﬁnite descriptions do not qualify as attributive.


Abstract
In this essay I argue that given Donnellan's formulation of the attributive uses of definite descriptions, as well as Kripke's [6] and Salmon's [10] generalized accounts, most uses of definite descriptions that are taken to be attributive turn out not to be so. In building up to my main thesis, I first consider certain problematic cases of uses of definite descriptions that do not neatly fit into any category. I then argue that, in general, a complete definite description we use is complex, in which there is an embedded singular term that is used referentially. From this I conclude that an attributive use of a definite description is an extremely rare linguistic phenomenon, much less frequent than what Donnellan, Kripke, and Salmon have presupposed; so much so that the standard examples given by Donnellan of the attributive use of definite descriptions do not qualify as attributive.
Donnellan characterizes an attributive use of a definite description as a use in which a speaker "wishes to assert something about whatever or whoever fits that description." [2, p. 285] I shall * Some of the ideas discussed in this paper were developed in a seminar I took with Nathan Salmon in 1993, though only recently have I come to realize the significance of the issue. The discussions I have had with Nathan Salmon on Donnellan's distinction in that seminar and other issues in the philosophy of language the years to follow it were among the most fulfilling experiences I have had in my philosophical career. I would like to thank Stephen Voss and Tomis Kapitan for useful comments and discussion on the topic, as well as two anonymous referees of KRITERION for their helpful suggestions.
argue that given this formulation, most uses of descriptions that we have thought to be attributive will turn out not to be so.
We may wish to capture this conception of the attributive use in terms of a certain sort of semantic intention, as Kripke suggests: "In a given idiolect, the semantic referent of a designator (without indexicals) is given by a general intention of the speaker to refer to a certain object whenever the designator is used. The speaker's referent is given by a specific intention, on a given occasion, to refer to a certain object [. . . ] My hypothesis is that Donnellan's referentialattributive distinction should be generalized in this light [. . . ] In one case (the "simple" case), his specific intention is simply to refer to the semantic referent; that is, his specific intention is simply his general semantic intention [. . . ] Alternatively-the "complex" case-he has a specific intention, which is distinct from his general intention, but which he believes, as a matter of fact, to determine the same object as the one determined by his general intention." [6,[173][174] More recently, Salmon, criticizing Kripke for having left out the epistemic condition of "having an object in mind" for the referential use, generalizes "the attributive use" as follows: "Let us distinguish between generalized referential and generalized attributive uses as follows. In a g-attributive use of a singular term, the speaker has a primary, identifying, purely semantic intention of the form By my use of this term, I intend to refer to α , where α is a definite description. This intention is general, as opposed to singular; it is a de dicto intention. Further, the speaker does not have in addition a supplementary primary linguistic intention of the form By my use of this term, I intend to refer to β that is not purely semantic in nature, or where β is a directly referential, Millian term (e.g. a name) for an individual person or object that the speaker 'has in mind', in the relevant sense. Here there is no potential conflict (from the point of view of pure semantics) with the primary de dicto linguistic intention, and speaker reference is therefore governed by that purely semantic intention" [10, p. 258-259].
Though there is disagreement between Kripke and Salmon on how to extend Donnellan's distinction to apply to the uses of proper names, it appears that their accounts are essentially similar with respect to the attributive uses of definite descriptions. 1 There is however good textual evidence that Donnellan did not wish to be too strict about this characterization of the attributive use.
Here is a passage in which Donnellan talks about a possible attributive use where the speaker does not have the intention to talk about exactly whatever or whoever fits that description: "In one example of the attributive use in my paper, a person upon finding the body of his friend Smith exclaims, "Smith's murderer is insane." In the example, the speaker had no particular person in mind as Smith's murderer [. . . ] suppose that while Smith did die of natural causes, he has indeed been assaulted before death and that the evidence that led the speaker to attribute insanity to "Smith's murderer" is still good evidence that his assailant is insane. In a sense the speaker has scored a "near miss." [. . . ] A 1 There is still no consensus on what exactly the distinction between the referential and the attributive uses of terms is supposed to be. Donnellan does not provide us with any strict criteria, which leaves the door open to different but incompatible interpretations. As I will try to show below there are examples of uses of definite descriptions that turn out to be "referential" for one philosopher, "attributive" for another, and "neither referential not attributive" for some other philosopher. I believe that this apparent inconsistency is due to the fact that there are two separate distinctions that are inherent in Donnellan's discussion of the matter. Furthermore Donnellan has never given us the conditions for "having an object in mind" which seems to be an essential feature of a referential use. For a detailed discussion of both of these points see my [4]. near miss occurs with an attributive use when nothing exactly fits the description used, but some individual or other does fit a description in some sense close in meaning to the one used. Only in the referential use can a speaker have "missed by a mile," because only that use involves a particular entity that the description either fits neatly, just misses, or wildly misses. Once this is seen, taking near misses into account does not blur the distinction. If anything, it helps one to see what the distinction is." 2 [3, p. 210] Neither Kripke nor Salmon, it seems, has been concerned by the fact that their formulations do not admit of "near misses" for attributive uses. 3 But what is more important for our purpose is Donnellan's position, which also follows from Salmon's account, that "only in the case of a referential use can one "miss wildly"". 4 Prima facie this sounds quite plausible, but as we shall see there are cases of "wild misses" that could not be classified as referential uses, and have strong attributive flavor.
The least important of such cases is when the speaker has a slip of the tongue: given the tragic 2 The distinction between "near" and "wild" misses is originally due to MacKay [7]. As the quote from Donnellan above reveals it certainly is not a very precise distinction, and I do not intend to sharpen it. Vague and imprecise as it may be, I believe that the examples that MacKay and Donnellan have given do appeal to our intuitions.
3 Under Salmon's account, such cases turn out to be neither referential nor attributive, though under Kripke's account they turn out to be referential uses. That is because of the fact that Kripke leaves out the "having in mind" condition for the referential use, as Salmon points out. See the last footnote to his [10]. On the other hand, to my knowledge Searle is the only one who makes an issue of the so-called near misses in the attributive use, by arguing that under Donnellan's characterization of the referential use in his [3], such near misses turn out to be referential uses, and concluding that this characterization of the referential use "isn't quite right even on Donnellan's own terms." [11, p. 191] 4 MacKay [7] argues that in those cases when a speaker uses a definite description referentially and scores a wild miss, the speaker does not succeed in referring to the entity he has in mind, though he may nonetheless make it known to his audience what it is that he is talking about. Here I have agreed with Donnellan, that such referential uses are cases in which the speaker could succeed in referring. situation the speaker says "Sam's murderer is insane", intending to refer to Smith's murderer (attributively) and not Sam's. This does seem to be a wild miss, and thus cannot be classified as an attributive use, though I doubt that such "slips of the tongue" have any philosophical significance. 5 There are, however, more interesting cases. Consider, for instance, a modified version of one of Donnellan's examples [2, p. 290-291]. A man comes across a group of people who refer to someone they take to be their leader as "the king". The man has heard from a reliable source that the person they refer to is in fact a usurper, but he has no inkling who this person is; all he knows is that the person referred to by the group as "the king", whoever he may be, is in fact not the actual king. We may imagine him saying: "I would like to meet the king." Now what should we say about the man's use of the term "the king"? It does not seem to be a referential use, for he does not have anyone in mind; he simply wishes to say something about whoever these people refer to as "the king". But is it then an attributive use? No doubt there is a temptation to say so, though the man rather then simply scoring a near miss seems to have missed by a mile. Such a use is simply not an attributive use under any of the accounts given above. 6 In this example, the speaker in a sense uses the term "the king" deferentially, to refer to whoever it is these people wrongly take to be their king. There may be other forms of deference that could have similar consequences, as suggested by Salmon: "[. . . ] suppose that the investigating detective is completely convinced that Johnson was murdered by the same culprit, so far still unidentified, who com-mitted the recent, very similar murder of Smith. The homicide department has no suspects, no witnesses, and no leads in either case; the detective's firm belief is based entirely on the common MO. When the detective uses the phrase 'Smith's murderer' at the scene of the later crime, he primarily means: the guy, whoever he is, who murdered Johnson. The only thing preventing the use from being bona-fide referential is the exact nature of the user's cognitive access to the individual. In this respect, mock referential uses are more attributive than referential. But in other respects, they are so much like genuine referential uses that they ought to have been included in previous discussions of the referential use [. . . ]" [10, p. 255] And later Salmon concludes that such uses are neither referential nor attributive, which in fact naturally follows from his generalized formulation of the two uses. Neale concludes of a similar example that "by Donnellan's own criterion, we have here an attributive use [. . . ] and a referential interpretation is out of the question" [9, p. 203], though it is not clear what "Donnellan's own criterion" is supposed to be. On the other hand, Kripke would have to conclude that such uses are referential, given that they fit his "complex case". So we seem to have uses of definite descriptions that turn out to be attributive for Neale, referential for Kripke, and neither referential nor attributive for Salmon. Now it may be suggested that such uses of definite descriptions are rare, involving a form of deferential intention, and could be left out of the picture. However this is only the tip of the iceberg. In fact, as we shall see now, in using a definite description in quite an ordinary context, in which there is no slip of the tongue, nor any irony or deference involved, a speaker may nonetheless score a wild miss, and his use of the definite description will not seem to neatly fit into any category. What I have in mind are the uses of "complex definite descriptions", i.e. those with at least one embedded singular term.
Consider this time a modified and extended version of Donnellan's famous Smith case: Jones, who is actually innocent, is on trial for the murder of Smith. One day as he is being brought to court, someone in the crowd opens fire and kills Jones. The murderer (of Jones) manages to escape. The investigator then makes the following announcement: "the police have not yet been able to find the gun that killed Smith's murderer". The use of the complex description "the gun that killed Smith's murderer" by the investigator is certainly not referential, given that he has no gun in mind. But is it then an attributive use? In using the embedded description "Smith's murderer", the investigator has the intention to refer to someone he has in mind, namely Jones, whom he mistakenly believes to have murdered Smith. In using the larger description "the gun that killed Smith's murderer" the investigator's primary intention is to talk about the gun that killed the person he has in mind, and not whatever fits the description. So it cannot be an attributive use either. We could get more complicated cases if we considered definite descriptions that have more than one singular term embedded in them. In such cases speakers may have various kinds of intentions regarding the use of each embedded term. In using such a complex definite description, if the speaker intends to refer to a particular object that he has in mind, then no matter how long the definite description is, we could conclude that such a use is referential; however if there is no object the speaker has in mind and no primary intention to refer to whatever or whoever fits that description, then we seem to be at a loss about what to say.
In fact even Donnellan's own example of the Smith case, which has served as a paradigmatic example of the attributive use, is susceptible to the same kind of consideration. To see this, all we need to do is imagine that the speaker misidentifies the person whose dead body he comes across when he says "Smith's murderer is insane", where in fact the body is that of Brown, and not Smith. If the speaker's primary intention is to attribute insanity to the murderer of the person he has in mind, whose dead body he has observed, then there is a good sense in which he wishes to talk about Brown's murderer, and it is irrelevant whether Smith has also been murdered. The use of the description "Smith's murderer" in this particular case is not a referential use, since the speaker has no particular murderer in mind, but the speaker also does not intend to refer to whoever fits the description. Again the use of the description does seem to have a very strong attributive flavor, but the speaker has missed by a mile. 7 If we take what Donnellan says at face value, or adopt Salmon's generalized version of attributive uses, then we must admit that such uses are neither referential nor attributive. Prima facie this may not seem to be a problem for the distinction is not supposed to be exhaustive, as both Donnellan and Kripke point out. There are, no doubt, certain contexts in which a speaker, in using a definite description in an utterance, may have no intention to refer, and may even believe that the description has no referent. 8 But clearly in the above cases the speaker does use the singular term as a designator, in its Fregean customary mode, so saying that they are neither referential nor attributive seems to be problematic. It is not clear what we should do with such examples, and it does not seem that Donnellan's texts are much help. What is worse is that complex definite descriptions are not a rare variety in language. In fact apart from the use of the so-called incomplete descriptions, almost all definite descriptions used in languages are complex; leaving them out of the picture would considerably limit the application of Donnellan's distinction, thus reducing its theoretical significance. Now the problem seems to occur only in those contexts in which a speaker uses an embedded singular term in a complex definite description referentially for an object that is not the semantic referent of that embedded term. So it may be suggested that when the investigator uses the description "the gun that killed Smith's murderer", and the person he has in mind whom he wishes to refer to by using the embedded term "Smith's murderer", is in fact Smith's murderer, then the use of the larger description can be unproblematically classified as attributive. But is that so? After all, the investigator's primary intention in such a context is not to make an assertion about the semantic referent of the term "the gun that killed Smith's murderer"; rather, he intends to talk about the gun that killed the person he has in mind, namely Jones, whom he believes to be Smith's murderer. In fact if we were to point out to him that Jones may not in fact be Smith's murderer, he could then withdraw his description, and plausibly claim that what he intended to talk about was the gun that killed Jones. He may even add that he would not be interested in the gun that may have killed Smith's murderer, if it turns out that Jones did not kill Smith.
Similarly, in Donnellan's original example, when the detective uses the description "Smith's murcould perhaps be said that the speaker uses the description "the tenth planet" neither referentially nor attributively. Alternatively, following Frege, it may also be argued that the speaker uses the description intending to refer not to its customary referent but its indirect referent, namely its sense. Whether that would allow us to extend Donnellan's distinction so that it applies to such contexts of use is an issue to be explored.
derer" attributively, it would seem to be wrong to claim that his primary intention is to talk about whoever murdered Smith. Rather he wishes to talk about whoever murdered the person he has in mind, whose dead body he sees, whom he takes to be Smith. Again if were to point out to him that it may not be Smith's dead body that he has observed, he might well say that his primary intention is to talk about the murderer of the man whose body he sees, even if that man turns out not to be Smith. So the detective does not simply "wish to assert something about whatever or whoever fits that description", nor is it the case that "his specific intention is simply to refer to the semantic referent", nor would it be true to say that "the speaker has a primary, identifying, purely semantic intention" that he could express by saying "By my use of this term, I intend to refer to Smith's murderer".
A moral to be drawn from all this is that a definite description is rarely used with the primary intention to refer to "whatever or whoever fits that description", for most definite descriptions are complex and have embedded singular terms in them that are used referentially. There are, of course, incomplete definite descriptions that appear to be simple. However we may be using them as directly referential terms-in which case they are not definite descriptions-, or the context may complete them turning them into complex descriptions. The only genuinely simple definite descriptions (i.e. ones which have no embedded singular terms within them) seem to be singular superlatives such as "the shortest spy", or descriptions that uniquely pick out an object by its general properties, such as "the planet with rings around it". Even for such terms, it may well be argued that they are incomplete; at least for most such terms we have hidden singular terms that make reference to a location (the world, the solar system etc.), and to an instance in time (now). A complete simple definite description, such as "the shortest spy, anywhere, any time", "the even prime number" etc., is, no doubt, an extreme rarity of the species. Once we admit that the complete definite descriptions we use are almost always complex, it would seem that they must have at least one embedded singular term that is itself neither a description, nor an abbreviation for one. Or else the completion of a definite description would lead to an infinite regress, or would end up containing a simple definite description. Given the rarity of the latter, it would follow that in almost all complete definite descriptions we use, there must be at least one directly referential term, such as a demonstrative, a pronoun, or a proper name. If such non-descriptive terms can be used attributively, in the generalized Kripkean sense, namely with the intention to refer to their semantic referent, then we may perhaps have uses of complex definite descriptions that are used attributively in accordance with Donnellan's characterization. For instance, in Donnellan's example, we could suppose that the detective, rather than using the name "Smith", more cautiously utters "his murderer is insane", by pointing to the dead body before him. In such a case, it may be argued that his primary intention is to refer to whoever his murderer is, this time not allowing for speaker's referent to diverge form the semantic referent of the term.
But even then, the speaker may be mistaken about the gender of the dead person. And if he even more cautiously uses "the murderer of that person", it may turn out that the dead body belongs to an android. Only if he uses, "its murderer" could we perhaps conclude that his primary intention is to refer to the semantic referent. That would, of course, require us to take ordinary uses of certain pronouns and demonstratives, that have no descriptive content, as being attributive. 9 In the case of proper names, things are even more difficult. To modify Donnellan's example slightly, imagine that the detective has been told by his assistant that a certain Smith has been brutally murdered. Before arriving at the murder site, he exclaims: "Smith's murderer is insane!" It may perhaps now be said that the detective's use of the name "Smith" is attributive, and in effect we may wish to hold that his primary intention is to 9 One would expect that, on Donnellan's account, ordinary uses of demonstratives would be paradigm cases of referential uses. It was David Kaplan, in one of his workshops, who suggested that they could be classified as attributive, though to my knowledge this has not appeared in print. Interestingly, Donnellan, who was in the audience, did not object, though he did not approve of the position either. refer to whoever satisfies the description "Smith's murderer". But even then, we could imagine the assistant being wrong about the identity of the murdered person. Ordinary uses of proper names seem to be paradigm cases of referential uses, as Donnellan suggests when he likens a referential use of a definite description to the use of a proper name [2, p. 282]. 10 Even if we suppose that simple demonstratives, pronouns, and proper names could have attributive uses in ordinary contexts, it seems clear to me that most of the complex definite descriptions we use are ones in which there are embedded singular terms used referentially; so if the use of such complex descriptions can not be classified as attributive, we would end up holding that an attributive use of a definite description is a much less frequent linguistic phenomenon then has been presupposed by Donnellan and others; so much so that Donnellan's standard examples of the attributive use will cease to qualify as attributive. Most descriptions we thought were being used attributively would fall into a strange no-man's-land, neither referential nor attributive. Perhaps Donnellan would be happy with this result, given that his main purpose in making his distinction was to show that Russell's semantic theory gives an account only of attributive uses of definite descriptions. So if definite descriptions are used attributively much less frequently than has commonly been assumed, then this is, Donnellan may claim, so much the worse for Russell's theory. Russell, on the other hand, might react by pointing out that the conclusion we have reached is simply another way of stating his basic thesis that we construct descriptions by making use of objects of acquaintance, and then adding that the words we use may not always capture the description we have in mind.
Either way, our discussion seems to reveal that a speaker's intentions concerning the use of a definite description are, in general, too complicated for the referential/attributive distinction to capture.