Direct Reference and the Export Gambit
by Will Wilkinson
Copyright: Will Wilkinson
I discuss objections to the direct reference theory of meaning or “Millianism”. I contend that the main difficulty for Millians lies in explaining the apparent failure of substitutivity in opaque contexts. I discuss the most promising Millian answer to the problem: that there really is no substitution failure and that the appearance of failure stems from a pragmatic violation. I argue that although the Millian appeal to pragmatics is a less ad hoc move than some critics have charged, it fails to account for the substitution failure in self-ascribed thoughts, and thus fails to give a comprehensive explanation for apparent substitution failure, and thereby fails to solve Millianism’s biggest problem.
The theory of direct reference, 'Millianism' for short, is popular these days, though not popular with everyone. Detractors allege insuperable difficulties. A profusion of defenders, however, argue that at least the balance of alleged difficulties is illusory; there is nothing really to overcome. The alleged problems, it is argued, are dissolved by seeing that they stem not from the distinctly semantic principles proffered by Millians, but instead from confusing these with the pragmatic principles that govern polite and rational discourse--those set out by Grice, for example. These pragmatic matters, unfortunately, obscure the semantic ones. Once we've swept away the complicating pragmatic detritus and regard the genuinely semantic issues in isolation from distractions, we shall see that Millianism and its lovely austerity is free from malady, despite dire diagnoses.
Michael Devitt, however, does not share the Millian's sanguinity. He charges that the Millians fail to give a principled basis for separating the semantic from pragmatic issues as they do. Devitt's charge has force because he sets forth a plausible general semantic methodology under which the Millian problems are not readily exported to pragmatics. With his methodology in hand, Devitt goes on to charge Millians with ad hocery for their failure to provide even a sketch of a methodology in which the export gambit is well motivated.
In the following pages I shall lay out a few of the main charges against Millianism, and then seek to determine the success of the "export gambit" in dodging them. Is it possible to successfully mute the force of the charges against Millianism by removing the relevant issues from semantics and resettling them in pragmatics? Or are Devitt's objections decisive? In the course of the discussion, the question about the precise nature and limits of semantics as an area of inquiry will arise. Although I have no hope to provide a solid answer to that question, I hoped that a clear-eyed assessment of the Millian appeal to pragmatics will at least clarify the issue.
First, it will be useful to say clearly what Millianism is. The doctrine is named in honor of John Stuart Mill, who held that "proper names are not connotative" but are merely "unmeaning marks" which stand for the named object.. Contemporary Millianism is similar in character to the original. However, where Mill held that names, strictly speaking, are wholly lacking in meaning, new style Millians have no compunction identifying a name's meaning with its referent, or its role of directly referring to its referent. Contemporary Millianism does not look back to Mill for impetus, but instead finds its motivation in Donnellan’s work on referentially used descriptions, in David Kaplan's theory of demonstratives and in work by Ruth Marcus and Saul Kripke on the behavior of names in modal contexts. Millianism has moved beyond demonstratives and names, however, to encompass some general terms as well, especially natural kind terms.
It may not be immediately apparent why a doctrine ostensibly about the nature of reference should turn out to be a theory of meaning. In order to grasp why "direct reference," as Millianism is usually called, is taken as a theory of meaning, it is necessary to go back to the theory of names invented by Frege and popularized by Russell.
Frege wrestled with a number of related and well-discussed puzzles. For instance, why should it be that some identity statements containing co-referential terms should be informative--'The Morning Star is the Evening Star", for example--while others should be uninformative--"'The Morning Star is the Morning Star', for example. Or, to take a different case, why should it be that co-referential terms cannot apparently be substituted one for the other salva veritate in all contexts? For example, why should substituting 'Clark Kent' for 'Superman' in 'Lois believes Superman can fly' create a change in the statement's truth-value, considering that Clark Kent and Superman are one and the same individual.
Frege's solution to the puzzles was, in essence, to distinguish a term's sense from its reference. Frege observed that there are various ways in which an individual can be grasped by, or presented to, a thinker's mind. So, although singular terms refer to single individuals, two co-referring terms may present the referent to the thinker's mind in two different ways. Terms that present a single referent under different modes are said to differ in sense, although they share reference. Now, according to most followers of Frege (though not necessarily according to Frege himself) a term's mode of presentation was said to be a definite description, a phrase employing the definite article, such as 'the first star to appear in the morning', that uniquely picks out the referent. A term's sense, through the device of the definite description, was said to fix the term's reference. And it is this, the sense, that was said to be determinative of the term's meaning.
So the model the theory gives us (let's call it 'the classical theory') has three main parts: There is (1) a word, (2) an associated definite description which determines reference and presents the referent to the thinker in a certain way, and (3) the referent itself. The classical theory, then, is evidently a theory of indirect reference. Between word and world is a reference-determining liaison. What's more, on the classical theory, the mediating element, the description, is identified as the term's meaning.
Against this background it is possible to understand Millianism more clearly. Millianism, a theory of direct reference, cuts out word/world middlemen altogether. This direct connection between terms and their referents is captured vividly by Marcus's notion that names are simply tags for things. If the mediating element is excised, what then is left for a term's meaning to be? What else but the referent itself, or, to put it another way, the term's role in referring to its referent.
Identifying meaning with reference is not without the support of common sense. It is not unusual when asked for the meaning of a word to answer the question by pointing out one of its referents. A perfectly good way of answering the question, "What does 'wombat' mean?" is to point out a wombat.
But there is much more to motivate Millianism than the common sense appeal to ostensive definition. There is, as I have already mentioned, Donnellan’s argument that reference can be achieved without satisfying a description, and Kaplan's theory of demonstratives as directly referential expressions, the success of which gave credence to project of winning larger territories for direct reference.
Furthermore, the classical description theory has shown itself unsatisfactory in a number of ways. Kripke has argued that descriptions are unable to determine the same reference in counterfactual scenarios, which runs afoul of our conviction that a name ought to pick out its bearer regardless of what may or may not happen to be true or false of the bearer. Kripke has also argued that a name may refer successfully despite a thinker's erroneous description, or despite a thinker's ignorance of the referent's unique characteristics. Putnam has argued, with respect to natural kinds, that it is unnecessary for a thinker to be able to pick out members of a kind in order to refer to it. All one needs is the appropriate relation to experts who are able to pick out members of the kind. So, in these cases, descriptions are unnecessary for reference. And, even more fundamentally, the classical theory is incomplete. If a term's meaning is determined by a definite description, in the case of names, or a set of necessary and sufficient conditions, in the case of kind terms, a dilemma arises regarding the meanings of the terms which make up the descriptions or defining conditions. Either these terms must be given meanings through the device of descriptions, and the terms in this next order of descriptions must be given meanings in like manner, and so on ad infinitum, in which case descriptions will be insufficient for reference. Or, alternatively, some terms in the descriptions are referentially successful without the help of further descriptions, in which case descriptions are unnecessary for reference.
These arguments have made reference mediating descriptive senses seem otiose and so have cleared a path for Millianism. However, all is not made well by rejecting descriptive senses. Millians remain afflicted with the puzzles that kept Frege up nights. And not only these.
II. Problems for Millians
Having eschewed descriptive senses, Millians are burdened with the task of resolving the puzzles which gave rise to senses in the first place. Millians are left, for instance, to explain the difference in meaning between statements of the form 'a = a' and 'a = b'. This problem, "the Identity Problem," is brought regularly against Millians and gets a great deal of attention in the literature. I shall focus most of my attention, however, on the problems involved in the seeming failure of substitutivity of co-referring terms in attitudinal contexts. This is "the Opacity Problem." I shall ignore here the problem of empty names -- "the Emptiness Problem" -- and the problem of accounting for the non-triviality of true positive existence statements and the meaningfulness of true negative existence state -- "the Existence Problem" -- which is not to say they are not problems.
But before turning to the Opacity Problem, it is well to indicate that Millians are resourceful and not all the common objections against them are as damning as some suppose. Millians have, for instance, carried the burden of the Identity Problem well, and this has historically been the main objection to Millianism. William Tashchek has argued persuasively that Frege's belief that a difference between two sentences in cognitive significance must reflect a difference in meaning is ill founded. Frege's error hangs on an implausible epistemology of language, one that assumes speakers to have unimpeded and direct epistemic access to the meanings of their concepts. If one accepts the epistemology, it follows from that a difference in epistemic significance between two sentences must signal a difference in meaning. If the meanings were the same, the thinker would know it, and they would not seem different.
However, one need not, and probably ought not, accept this epistemology. First of all, semantic competence does not entail semantic knowledge; at least not propositional knowledge--knowledge-that rather than knowledge-how. It is easy and tempting to slide from the fact that a speaker is well able to use her language to the conviction that she must have a great deal of propositional knowledge about her language. One should be wary of so sliding. For competence does not generally imply propositional knowledge. And if Putnam is right, as I suspect he is, and "meaning just ain't in the head," then thinkers naturally will lack immediate and effortless access to the meanings of at least some of their concepts. These meanings will be simply unavailable to inner eye of semantic reflection, since these meanings are in part comprised of objects and relations in the external world, outside the mind. Furthermore, Devitt argues plausibly that differences in cognitive significance between sentences of the form 'a = a' and 'a = b' hangs more on mastery of the meaning of '=', the law of identity, than anything to do with the meanings of the terms. And, as should seem obvious, type-differing co-referring terms just sound different.
So the Millian has, I think, ample resources to resist arguments from felt differences in cognitive significance. And so Millianism has the resources to fend off at least some of the common objections. Can the Millians expect equal success with the Opacity problem? Let's see.
III. Substitutivity and Opacity
Consider the following argument.
(1) Kaczynski took a math course with Quine.
(2) Kaczynski is The Unabomber.
so, (3) The Unabomber took a math course with Quine.
(A1) is sound. The premises are both true. And the conclusion follows if Leibniz's Law, which, roughly put, states that whatever is also true of something is true of whatever is identical to that something (or to put it less semantically, if some object has a property, then any identical object has that property too, i.e. (AP)(Ax)(Ay)(Px & x=y -> Py) is true. In the linguistic context, Leibniz's Law cashes out as the principle that co-referring terms may be substituted one for the other in a sentence without altering its truth or falsity. This is the Principle of Substitutivity.
But now consider an argument that might have troubled Frege (supposing, for fun, that the date is prior to the discovery of the Unabomber's identity, and that Quine had remembered Kaczynski.)
(i) Quine believes that Kaczynski was an excellent student.
(ii) Kaczynski is the Unabomber.
So, (iii) Quine believes that the Unabomber was a excellent student.
This is troubling because once again the premises (we are supposing) are both true, and the argument seems to be of the same form as (A1), yet we are not inclined to assent to (iii), the conclusion. Had Quine remembered Kaczynski, he might have nevertheless remained wholly ignorant of The Unabomber. Perhaps he was too busy reading Carnap to read the news. So he would have had no Unabomber-beliefs at all. And, of course, the Unabomber's identity was shrouded in mystery. So Quine would naturally have lacked beliefs about the Unabomber's record of scholastic achievement, even were he to be aware of the terrorist. Nevertheless, as in (A1) it seems that the Principle of Substitutivity ought to apply; we ought to be able to substitute 'the Unabomber' for 'Kaczynski' without losing truth. But, it appears, we do lose truth by substituting. What has gone wrong?
Frege's own explanation, in short, is that 'Kazcynzki' is used one way in (i) and another way in (ii). In (i), 'Kazcynzki' is part of the content of a propositional attitude, it is the content of a "t-clause" (that clause), while in (ii) it is not. And this makes a difference. According to Frege, the reference of the embedded sentence in (i)--'Kaczynski was an excellent student'--is what the sense of the proposition it expresses would be if used in an extensional context. The reference in this case is not to the usual referent of a sentence, namely "the True" or "the False", since the sentence is embedded within a larger sentence within the scope of an intensional operator (in this case, 'believes'). Thus 'Kaczynski' in (i) refers to whatever descriptive mode of presentation is associated in Quine's mind with 'Kaczynski'. On the other hand, 'Kaczynski' in (ii) refers to Kaczynski himself, not some mode under which Kaczynski is made an object of thought. Since the 'Kaczynski' in (i) has a different referent from the 'Kaczynski' in (ii), what is identical to the 'Kaczynski' in (ii) cannot be substituted for the 'Kaczynski' in (i).
Now, one may accept the spirit of Frege's answer without adhering to the letter. One needn't accept his view that the usual reference of sentences are truth-values. Nor must one accept that the meaning of a term is different inside and outside attitudinal contexts, that is, one needn't surrender what Davidson has called "semantic innocence." Also one is not obliged even to accept that reference determining modes are internally represented descriptions. Devitt, for instance, has argued with great plausibility that senses may be understood as reference-determining causal designation chains rather than in-the-head descriptions (and if he is right, the main dangers of descriptivist senses discussed above are defused.)
What then is left of Frege? Well, what is crucial in Frege is the notion that the sense of a term contributes to its meaning and co-referring names may therefore have divergent meanings. Substitutivity fails in attitudinal contexts because the point of reporting someone's attitudes (one of them at least) is to faithfully represent the content of the subject's attitude, to say something true about the nature of the subject's cognitive life. But the overall thought content is determined by the meanings of the individual elements that compose the thought, that is to say, sentence meaning is compositional. Substituting an element with one meaning for an element with another meaning will change the overall content attributed to the subject, even if the switched terms are in the same line of business reference-wise. If the initial content was accurately attributed, rendering a truth, a change in a meaningful element of the content will change what is attributed with the likely effect of wringing falsehood from truth. This is what renders attitudinal contexts, to use Quine's phrase, "opaque."
How are Millians, who eschew senses, to deal with substitutivity failure in opaque contexts? One tactic is to flatly deny that there is such a thing as a genuinely opaque context. Quine himself is displeased by opaque contexts: a "lawless swarm" he called them. So he devotes considerable energy showing we can do with out them. We can take statements like,
(A) Quine beliefs that Kaczynski was an excellent student.
eliminate the 'that' clause and adopt one of the following alternatives in place of (A):
(B) Quine believes lx(x was an excellent student) of Kazcynski.
(C) Quine believes Kaczynski to have been an excellent student.
These 'transparent' reworkings of (A) bring 'Kaczynski' into "fully referential position, and thus are more lawful, permitting substitutions salva veritate. For instance, we can convert (C) to
(D) Quine believes the Unabomber to have been an excellent student.
without losing truth. Now, the strategy doesn't seem to be a comprehensive success. For example, if we are to perform Quine's operation on,
(E) Billy believes that God has a big white beard.
(F) Billy believes God to have a big white beard.
in which case we end up quantifying over, and ontologically committed to, God. This is an unhappy consequence.
Despite difficult cases, Quine did succeed in showing that it is not always clear in each individual case whether a sentence ascribing belief is best understood transparently or opaquely, whether terms inside attitudinal contexts can be exported to a fully referential position. A transparent interpretation will license substitution, and an opaque interpretation will block it.
Now, the Millian is wont to argue that is rare, if ever, that one is warranted in offering an opaque rather than transparent interpretation. However, it should be clear that a flight from opacity and neo-Fregean senses will have deep consequences. Think about argument (A2) above. If meaning just is reference, then the Millian will have some hard choices. One option is to rather uncharitably attribute to Quine a piece of unreason--the belief that Kaczynski isn't self-identical. The more popular strategy, however, is to deny flatly that substitutivity does fail. The Millian will assert that strictly and literally Quine does believe that the Unabomber was an excellent student, regardless of the intuitions of ordinary speakers, or of Quine himself. Our firm feeling that Quine doesn't really believe this is, according to the resolute Millian, something of a delusion brought on by inexplicit extra-semantic consideration interfering with our better semantic judgment.
This is a hard line. And it leaves the Millian a chore: to make good sense of the dissent of ordinary speakers on this verdict. Why should the folk be so systematically mislead about what they strictly and literally believe, hope and fear?
Here is where the Millian is inclined to dabble in linguistic geography, drawing a border between semantics and pragmatics, and setting the apparent problems for Millianism firmly on the pragmatic side of the divide. The problem, the Millian asserts, comes from mixing semantic issues from pragmatic ones. The semantic issues considered purely in isolation show Quine to have believed things about the Unabomber's mathematical prowess, show Lois Lane to believe Clark Kent to fly, and so on. The ordinary speaker cringes at these ascriptions not because they are false, but because she has integrated pragmatic principles that proscribe such ascriptions in normal contexts of discourse. These principles run so deep that speakers confuse the verdicts generated by their pragmatic convictions for evidence of semantic content, of meaning.
Along these lines, Nathan Salmon maintains, in the spirit of Grice, that there is a distinction between information that is semantically encoded and pragmatically imparted. By making the distinction, Salmon indicates that thinks he knows the difference, that he knows how to mark off semantics from pragmatics. However, if he does knows this, it would be helpful for him to explain it in greater detail. For chief among the problems of semantics is that hardly anyone has a clear idea of its proper domain. That is what Lycan is pointing out with his mocking "double indexical theory of meaning" (i.e., meaning is that property of linguistic phenomena that I’m interested in now.). The bounds of pragmatics are similarly indistinct. But Salmon apparently has confidence in placing the boundary between pragmatics and semantics just where he does. A successful defense of the demarcation would be a significant achievement in and of itself.
How might one tell the difference between semantic matters and pragmatic ones? Grice, the inspiration for the Salmon's distinction between the pragmatically imparted and semantically encoded, put forward a couple of tests to distinguish what he called, 'conversational implicatures' from the semantic entailments of sentences. Beyond Grice's tests, one might seek to determine what is and is not semantically conveyed by deploying a general semantic methodology. This is how Devitt proceeds when he accuses Millians of making ad hoc moves. So let's look at the semantics/pragmatics distinction an little closer, and then at Devitt's methodological proposal, and then let us see whether Salmon’s exportation of substitutivity problems can be given some motivation by fitting it into Grice’s program.
In a recent discussion of the semantics-pragmatics discussion, Kent Bach begins by saying that it is "easier to apply than explain." Such statements can cause a sense of despair when what we seek to know is whether a particular application of the distinction is justified. For a justification would seem to require an explanation. After a careful examination of competing formulations of the distinction, and after providing an account of the distinction himself, Bach remarks that "Whether a given phenomenon has a semantic or pragmatic explanation or, as is often the case, some combination of both, must be settled on a case-by-case basis." Again, this does not fill one with confidence. However, a skeptical attitude toward the whole enterprise of elucidating the distinction cuts both ways in the present debate. Devitt may be right that the export gambit is methodologically unmotivated, and that the Millian's boundary drawing is ad hoc, but that doesn't mean that they have drawn the line in the wrong place. One would need a principle to decide that.
Pragmatics is often dismissively characterized as the trash bin of linguistics. Whatever linguistic phenomena doesn’t seem proper to phonology, syntax or semantics are thrown into pragmatics. Somewhat less negative characterizations draw the distinction along lines of linguistic meaning vs. use, or truth-conditional vs. non-truth-conditional meaning, or context independence vs. context dependence. However, all of these formulations seem to fail for as in each case it seems as though the putatively pragmatic aspects make a relatively straightforward semantic contribution.
It seems that a more general methodological principle is needed. Devitt does not provide a principle for determining if some linguistic property is pragmatic. But he does provide a principle for determining if some property is semantic or not. According to Devitt, semantic properties are those we use to meet our semantic purposes, and our semantic purposes are to “explain other people’s behavior and to use others as guides to reality.” We ascribe meanings to others to make sense of their actions, to predict what they will do, and to be able to gather information about the world from their utterances. According to Devitt, information about modes is semantic, because the way in which one thinks of a referent is relevant to one’s behavior. Lois’s reaction to kissing Superman will certainly differ from her reaction to kissing Clark. If I see a man with burning trousers in a mirror, and I think ‘That man’s trousers are burning’ I will react much differently than I would had I thought ‘My trousers are burning’, even if “that man” is me. Thus modes matter to the explanation of behavior and so on Devitt’s methodology, information conveyed by modes is semantic information. According to Devitt, Salmon and other implicature theorists owe us an account of what our semantic purposes are, so that we can at least identify which linguistic properties are semantic and which ones are not.
Salmon seems to accept the burden. He recognizes some methodological presumption in favor of the verdicts of the ordinary speaker and common sense that substitution into attitudinal context does fail to preserve truth. Salmon admits that in the case of commonsense challenges to putative philosophical analyses
. . . philosophers are rightly skeptical of the reply that ordinary usage is incorrect and that the subject does indeed know the proposition in question [e.g., 'knowledge is justified true belief'], even though we typically say that he or she does not. Anyone maintaining this position may well be suspected of protecting and invested interest in the theory being challenged, rather than pursuing in good faith the philosopher's primary purpose of seeking truth no matter where the facts may lead. … [T]he claim that Lois Lane does, strictly speaking, believe and even know that Clark Kent is Superman (since she knows that he is Clark Kent) must not be made lightly, lest he or she who makes it be placed under the same suspicion. 
Salmon, however, does little more in defense of his claim that Lois really does know that Clark Kent is Superman than argue that there is a distinction between semantically encoded and pragmatically imparted information, and that the distinction can be brought to bear on substitutivity cases. He does not offer a principle for determining whether some bit of information is semantically encoded or not. Thus, no alternative to Devitt’s proposal is on offer and there is nothing to defuse the suspicion, of which Salmon seems well aware, that the move is ad hoc.
However, Salmon takes inspiration in his distinction from Grice’s theory of conversational implicature. It may be thought that if Salmon’s claim that information about modes of reference is pragmatically imparted can be shown to comport with Grice’s tests for determining an implicature, then Salmon could claim that he has some independent methodological motivation. So let’s look more closely at Salmon’s proposal from a Gricean perspective.
First, it is necessary to give a brief account of the Gricean picture. On the Gricean picture, the meaning of a sentence conventionally determines what is literally said by uttering the sentence. That is to say, the meaning gives the literal truth conditions of the sentence. The meaning of 'I have not had breakfast today' determines that if a speaker utters the sentence on a certain day what he thereby says is that he has had not breakfast on that day. However, by uttering this sentence, the speaker may intend to communicate the fact that he is hungry and would like to grab a bite. This, however, is not part of what is literally said, or conventionally implied, by the sentence. Rather, his hunger is conversationally implicated. Conversational implicatures are pragmatically rather than semantically determined. Grice accounts for conversational implicature by showing how they fall out of certain general principles or 'maxims' of conversation that rational participants in conversation are bound to observe. Assuming that that a speaker is following the maxims, certain utterances can contextually imply information not conveyed by the literal meaning of the sentence.
Because what is communicated by an utterance of a sentence includes pragmatic elements, the fact that it is possible to interpret a given expression differently in different contexts need not entail that the expression has more than one meaning, i.e., that it is semantically ambiguous. The felt difference in expression's meaning across different contexts of utterance can be accounted for by positing more than one literal meaning. But it can also be accounted for pragmatically, by positing a pragmatic implicature which combines in certain contexts with what is literally said.
The example most often appealed to in this connection is that of disjunctive sentences like 'P or Q'. The sentence can be given either an inclusive or exclusive interpretation. However, to avoid propagating multiple meanings of 'or' one can take it as unambiguously inclusive and account for the exclusive interpretation by arguing that in certain contexts the utterance conversationally implicates that 'P' and 'Q' are not true together.
In general, Grice argues, accounting for seeming ambiguities pragmatically is preferable to positing an actual semantic ambiguity. He erects this point into methodological principle: "Modified Occam's Razor", which states that meanings are not to be multiplied beyond necessity. Like the original Occam's Razor, Grice's version is a principle of theoretical economy. When available, it is best to give a pragmatic explanation because pragmatic explanations are based on very general and independently motivated principles, whereas explanations appealing to semantic ambiguity tend to be ad hoc and theoretically costly, like so many epicycles.
It should now be easy to see Grice's appeal to Millians. The Millian can point to the seeming ambiguity between the transparent and opaque readings of sentences like 'Quine believes that the Unabomber was an excellent student'. And, appealing to Grice's Modified Occam's Razor, the Millian can argue that we ought not posit actual semantic ambiguity, but seek to explain our sense of ambiguity in terms of some pragmatic phenomenon. So it behooves us to hold that all attitude ascription sentences are, strictly speaking, transparent. In Salmon's Gricean language, we can say that the appearance of opacity arises due to information about modes of reference pragmatically imparted in certain contexts of utterance. And so we need not understand information about modes to be semantically encoded in the belief ascription sentence. We can understand these sentences to be unambiguously transparent and subject to the principle of substitutivity. In Salmon's words:
Utterances of the locution 'a believes that S' may even typically involve a Gricean implicature to the effect that the person referred to by a believes the information that is typically pragmatically imparted by utterances of S. Even so, that is not a part of the literal content of the belief attribution.
This is, I think, a very nice dialectical move. However, it is successful only if Millians like Salmon can provide principled support to the notion that information about modes is pragmatically imparted and not semantically encoded in attitude ascriptions. As Devitt writes,
To make good his position Salmon must show that the normal purposes served by conveying information about modes are not semantic but merely pragmatic. For this he needs an account of what purposes are semantic ... He takes the semantic task to be explaining the natures of "semantic cognitive content[s]" ... What he fails to do is to tell us what purposes are served by attributing these contents. This leaves the semantic task ill-defined and leaves no basis for the claim that conveying information about modes is not attributing content.
Perhaps one way that Salmon can defend himself against Devitt is by demonstrating that information about modes is pragmatically implicated according to the tests Grice devised specifically to distinguish pragmatically conveyed information from genuine semantic entailments of sentences. Because Grice's tests may be thought to be well-motivated methodologically, and independently plausible, a passing grade on the tests might give Millians something with which to ward off Devitt's challenge.
Grice's first test is the cancellability test.
A putative conversational implicature that p is explicitly cancelable if, to the form of words the utterance of which putatively implicates that p, it is admissible to add but not p, or I do not mean to imply that p, and it is contextually cancelable if one can find situations in which the utterance of the form of words would simply not carry the implicature.
Quine showed that some seemingly opaque constructions can be read transparently without strain, which goes to show that seemingly opaque constructions are contextually cancelable according to Grice's test. And Barwise and Perry have argued that a natural seeming opaque interpretation of an attitude ascription can be cancelled by explicit denial of an intention to convey information about modes. Barwise and Perry write:
Some arguments for referential opacity seem based on a confusion between conversational implicature and semantic entailments. 'Smith believes Cicero was an orator' does not imply, but at most suggests, that Smith would check 'Cicero was an orator' true. The suggestion is clearly cancelable: 'Smith believes that Cicero was an orator, but only knows to call him "Tully."'
So it appears that the hypothesis that information about modes is pragmatically imparted passes Grice's first test. The hypothesis seems to fare reasonably well on Grice’s second test as well, non-detachability.
An implicature is non-detachable just in case there is no way of rephrasing the utterance that seems to generate the implicature in a way that does not generate the implicature without making it explicit. If it were possible to remove opacity by changing ‘believes that’ constructions to ‘believes of’ constructions, then the opacity would be “detachable”. But Kent Bach has shown that even ‘believes of’ locutions can be interpreted opaquely.
The implicature analysis of apparent substitution failure depends on something like Grice’s cooperative principle. The cooperative principle and the attendant maxims are principles of conversation. Let us grant that the implicature analysis is sufficient to explain away apparent substitution failure in conversation. Now, if it turns out that substitution failures seems to occur not only interpersonally, in conversation, but also intrapersonally, in thought, then there remains a question whether the implicature analysis is adequate to explain substitution failure in the latter case as well as the former.
We ascribe thoughts to ourselves. It is common to say things like, “I thought that I saw your car on 12th Street”, or to privately think things like, “I think that he loves me.” Since we ascribe thought to ourselves, it is interesting to consider apparent substitution-failure cases in the first-person.
Put yourself in the position of Lois Lane (as well as you can). After a stolen kiss in the copy room, suppose you have the following thought:
(a) It believe that Clark Kent loves me.
But it just happens to be the case that
(b) Clark Kent is Superman.
So, by the principle of substitutivity, you (Lois) ought to have the following thought.
(c) I believe that Superman loves me.
If you are like me, your intuition is that you do not have that thought. My offline simulation tells me that if Lois did have it, her reaction to it would likely differ radically from her reaction to (a). The idea of a “man of steel” might well be enticing in a way that the idea of a “mild mannered reporter” would not.
Yet the Millian claims that if you are Lois, then you do think (c). According to the implicature theorist, our hesitation to self-ascribe the t-clause in (c) flows from our adherence to something like the following Grice-like maxim:
In reporting a belief about an object, and especially in referring to that object, use an expression which the believer himself would use (insofar as differences in language or context permit), or at least, try to be faithful to the believer’s own point of view, unless there are reasons not to do so.
Thus, the reason I (as Lois) will not self ascribe ‘Superman loves me’ is that ‘Superman’ is not an expression I would use, or using ‘Superman’ would be unfaithful to my point of view. But this sounds strained. Why is ‘Superman’ an expression I would not use? Why is it unfaithful to my point of view? The intuitive answer is that ‘Superman’ is in this context an inept expression to ascribe simply because I don’t believe that Superman loves me, that is, because I do not hold (c) true. This suggest that the alleged implicature is not cancelable. If the implicature were cancelable, then I ought to be able to consistently maintain the following thought.
(d) I believe that Superman loves me, but I only know to call him ‘Clark Kent’.
But if I could think (d), then I would necessarily think something false. If I earnestly thought that (d), then it would be false that I only knew to call Superman ‘Clark Kent’. For thinking (d) implies knowledge of the co-reference of ‘Superman’ and ‘Clark Kent’, which is precisely what the right-hand conjunct denies. So (d) is not a thought I could think truly. Applying Barwise and Perry’s language to the present example, it ought to be the case that ’I think that Superman loves me' does not imply, but at most suggests, that I would check ‘Superman loves me’ true. But it seems that if I reported (c) to you, and then refused to check ‘Superman love me’ true, I would be guilty of an inconsistency, which runs counter to the thesis that the alleged implicature in (c) is cancelable. And indeed, if I try to explicitly cancel, by thinking (d), I run myself straight into a inconsistency, at once implying and denying that I know that ‘Clark Kent’ shares reference with ‘Superman’. So, we ought to conclude that in the case of first person ascriptions, information about modes of reference is not pragmatically imparted through implicatures. But then the Millian is left without an explanation of the apparent substitution failure.
The foregoing argument for the failure of a pragmatic account of substitution failure is analogous to an argument of Max Black’s against Davidson’s pragmatic account of metaphor. According to Davidson, metaphorical language does not involve using terms in unconventional ways, such that their use introduces a kind of ambiguity. Rather, one is acting through metaphorical speech in a way to induce to hearer to grasp some perhaps obscure similarity between the reference of the metaphorically used term and the subject. But, according to Black, this proposal faces trouble when applied to intrapersonal language use.
On the speech-act approach, it is hard to make sense of what happens when somebody expresses a thought to himself… Any clear cases of speech-acts that come readily to mind involve communication with an audience: it makes little sense to think of promising oneself something, or warning, or advising, pronouncing judgment, and so on to oneself. What then, on Davidson’s view is a soliloquizing thinker, using metaphorical language, supposed to be doing? Nudging and provoking himself to pay attention to some covert likeness? But surely he has already done so…. Is he perhaps pretending to talk to himself, as if he had not already been seized by an unobvious resemblance between the two things in question?
Similarly, it is hard to see why someone would bother to heed the cooperative principle and its maxims while conducting a monologue. The function of the maxims is to facilitate the efficient communication of information between conversation partners. Certainly I may wish, in some sense, to communicate something to myself. But the maxims are needful largely because of differences in the conversation partner’s contexts of belief. But there are no differences between my context and itself. Surely, I cannot confuse myself about the way in which I think of some object by thinking about it under more than one mode. If I know that two modes go with the same object, then I know it. So I don’t have to worry about which mode to think about it under when I am thinking to myself. If I don’t know that two modes go with the same object, then I don’t know it. And so I won’t think to think about the object I have in mind under the other mode. So it should not be surprising that it would be unnecessary to follow the maxim of faithfulness when self-ascribing thoughts. And, more generally, it should not be surprising that pragmatic principles will generally fail to account for phenomena, like metaphor or substitution failure, that occur in first-personal thought as well as in public discourse.
So I conclude that although the implicature account seems promising for accounting for apparent substitution failures in public discourse, and that on that ground, it appears to have some non ad hoc methodological basis, it fails when applied to apparent substitution failures in cases of self-ascribed thoughts. Thus, the implicature analysis fails to provide the Millian a way to uniformly handle the substitutivity problem.
Michael Devitt, Coming to our Senses
2J.S. Mill, A System of Logic, 1843, reprinted as John Stuart Mill's Philosophy of Scientific Method, edited by Ernest Nagel (Hafner Publishing Co.: New York, 1950).
3 David Kaplan, "Demonstratives," in Themes from Kaplan (New York: Oxford, 1981), edited by J. Almog, J. Perry, and H. Wettstein, pp. 481-564. Ruth Barcan Marcus, Modalities (New York: Oxford, 1993); Saul Kripke, Naming and Necessity, (Cambridge: Harvard, 1981).
4 This is the so-called modal argument against descriptions. Schematically, the argument goes like this:
(a) If 'N' meant 'the F', then 'N might not have been the F' would be false.
(b) 'N might not have been the F' is true.
(c) 'N' does not mean 'the F'.
5 These are the so-called arguments from ignorance and error. Kripke imagines a fanciful case in which a man named Schmidt discovered the incompleteness of arithmetic, but was murdered and plagiarized by Godel. Kripke argues that someone who associates 'The discover of the incompleteness theorem' with 'Godel' would pick out Godel although the description is true exclusively of Schmidt. Error. Or one might know nothing of someone, Eintein say. All the same, Kripke argues, one may successfully refer to him just in case one has become part of a chain of designation stretching back to Einstein himself. Ignorance. Kripke, p. 79-91.
6 Hilary Putnam, "The Meaning of 'Meaning' in Mind, Language and History (Cambridge: Cambridge, 1981), pp. 226-7.
7 Michael Devitt and Kim Sterelny, Language and Reality, 2nd Ed. (Cambridge: MIT Press, 1999), p. 60.
8For a recent attack on Millianism via the Emptiness Problem see Avrum Stroll, "Proper Names, Names, and Fictive Objects," Journal of Philosophy , Oct. 1998.
9William Taschek, "Content, Character, and Cognitive Significance," Philosophical Studies 52, pp, 161-189.
10Michael Devitt, Coming to our Senses (Cambridge: Cambridge, 1996), pp. 49-53.
11Putnam, ibid, p. 227. See also Michael Devitt, "Meanings just ain't in the head" in Meaning and Method: Essays in Honor of Hilary Putnam (Cambridge: Cambridge, 1990), ed. George Boolos, pp. 80-104.
12Devitt, Coming, pp. 177-8.
14Yes, Kaczynski did take a course with Quine. In an interview Quine remarked, "Just the other day, I took out my old records. I did teach Kaczynski, although I don't remember him. He tied for top, 98.9%."The W.V. Quine Homepage, <http://users.aol.com/drquine/wv-quine.html>.
15Devitt, Coming, pp. 163-170.
16Kent Bach, "The Semantics-Pragmatics Distinction: What It Is and Why It Matters," on the internet at http://userwww.sfsu.edu/~kbach/semprag.html .
19 Devitt, Coming to Our Senses, p. 57-58.
20Nathan Salmon, Frege's Puzzle (Cambridge, MA: MIT Press, 1986), p. 83.
21The following characterization leans heavily on Francois Recanati, Direct Reference: From Language to Thought, (Oxford: Blackwell, 1993), pp. 233-4.
22Paul Grice, Logic and Conversation
23Nathan Salmon, Frege's Puzzle (Cambridge, MA: MIT Press, 1986), p. 83.
24Devitt, Coming, p.182.
25Paul Grice, Studies in the Ways of Words (Cambridge, MA: Harvard University Press, 1989), p. 44.
26John Barwise and John Perry, Situations and Attitudes
27 From Recanati, Direct Reference, p. 333.
28 Donald Davidson, “What Metaphors Mean”, Critical Inquiry Vol. 5, pp 31-47
29 Max Black, “How Metaphors Work: A Reply to Donald Davidson”, Critical Inquiry Vol. 6, No. 1, pp. 138-139.
Michael Devitt, Coming to our Senses
J.S. Mill, A System of Logic, 1843, reprinted as John Stuart Mill's Philosophy of Scientific Method, edited by Ernest Nagel (Hafner Publishing Co.: New York, 1950).
 David Kaplan, "Demonstratives," in Themes from Kaplan (New York: Oxford, 1981), edited by J. Almog, J. Perry, and H. Wettstein, pp. 481-564. Ruth Barcan Marcus, Modalities (New York: Oxford, 1993); Saul Kripke, Naming and Necessity, (Cambridge: Harvard, 1981).
 This is the so-called modal argument against descriptions. Schematically, the argument goes like this:
(a) If 'N' meant 'the F', then 'N might not have been the F' would be false.
(b) 'N might not have been the F' is true.
(c) 'N' does not mean 'the F'.
 These are the so-called arguments from ignorance and error. Kripke imagines a fanciful case in which a man named Schmidt discovered the incompleteness of arithmetic, but was murdered and plagiarized by Godel. Kripke argues that someone who associates 'The discover of the incompleteness theorem' with 'Godel' would pick out Godel although the description is true exclusively of Schmidt. Error. Or one might know nothing of someone, Eintein say. All the same, Kripke argues, one may successfully refer to him just in case one has become part of a chain of designation stretching back to Einstein himself. Ignorance. Kripke, p. 79-91.
 Hilary Putnam, "The Meaning of 'Meaning' in Mind, Language and History (Cambridge: Cambridge, 1981), pp. 226-7.
 Michael Devitt and Kim Sterelny, Language and Reality, 2nd Ed. (Cambridge: MIT Press, 1999), p. 60.
For a recent attack on Millianism via the Emptiness Problem see Avrum Stroll, "Proper Names, Names, and Fictive Objects," Journal of Philosophy , Oct. 1998.
William Taschek, "Content, Character, and Cognitive Significance," Philosophical Studies 52, pp, 161-189.
Michael Devitt, Coming to our Senses (Cambridge: Cambridge, 1996), pp. 49-53.
Putnam, ibid, p. 227. See also Michael Devitt, "Meanings just ain't in the head" in Meaning and Method: Essays in Honor of Hilary Putnam (Cambridge: Cambridge, 1990), ed. George Boolos, pp. 80-104.
Devitt, Coming, pp. 177-8.
Yes, Kaczynski did take a course with Quine. In an interview Quine remarked, "Just the other day, I took out my old records. I did teach Kaczynski, although I don't remember him. He tied for top, 98.9%."The W.V. Quine Homepage, <http://users.aol.com/drquine/wv-quine.html>.
Devitt, Coming, pp. 163-170.
 Devitt, Coming to Our Senses, p. 57-58.
Nathan Salmon, Frege's Puzzle (Cambridge, MA: MIT Press, 1986), p. 83.
The following characterization leans heavily on Francois Recanati, Direct Reference: From Language to Thought, (Oxford: Blackwell, 1993), pp. 233-4.
Paul Grice, Logic and Conversation
Nathan Salmon, Frege's Puzzle (Cambridge, MA: MIT Press, 1986), p. 83.
Devitt, Coming, p.182.
Paul Grice, Studies in the Ways of Words (Cambridge, MA: Harvard University Press, 1989), p. 44.
John Barwise and John Perry, Situations and Attitudes
 From Recanati, Direct Reference, p. 333.
 Donald Davidson, “What Metaphors Mean”, Critical Inquiry Vol. 5, pp 31-47
 Max Black, “How Metaphors Work: A Reply to Donald Davidson”, Critical Inquiry Vol. 6, No. 1, pp. 138-139.