The Theory of Meaning by Russell Dale; Chapter 5

Chapter 5
Conventions and Meaning

5.1 Introduction

David Lewis offers the following theory of the actual-language relation in his paper "Languages and Language":

[L]A language L is used by a population P if and only if there prevails in P a convention of truthfulness and trust in L, sustained by an interest in communication.<1>

Lewis defines the notion of a convention as follows:

...a regularity R, in action or in action and belief, is a convention in a population P if and only if, within P, the following six conditions hold. (Or at least they almost hold. A few exceptions to the "everyone"s can be tolerated.)

(1) Everyone conforms to R.

(2) Everyone believes that the others conform to R.

(3) This belief that the others conform to R gives everyone a good and decisive reason to conform to R himself. ...

(4) There is a general preference for general conformity to R rather than slightly-less-than-general conformity - in particular, rather than conformity by all but any one. ...

(5) R is not the only possible regularity meeting the last two conditions. ...

(6) ...the various facts listed in conditions (1) to (5) are matters of common (or mutual) knowledge: they are known to everyone, it is known to everyone that they are known to everyone, and so on.<2>

The notion of truthfulness in a language used in [L] is glossed by Lewis in the following way:

To be truthful in L is to act in a certain way: to try never to utter any sentences of L that are not true in L. Thus it is to avoid uttering any sentence of L unless one believes it to be true in L.<3>

And Lewis tells us the following about the notion of trust in a language:

To be trusting in L is to form beliefs in a certain way: to impute truthfulness in L to others and thus to tend to respond to another'sutterance of any sentence of L by coming to believe that the uttered sentence is true in L.<4>

To begin my discussion of Lewis's theory I will talk about a certain well-known objection to it which I will argue is not really directly an objection after all. But to see clearly that this well-known objection is not directly an objection sets the stage for stating a somewhat more direct objection which I will raise. Then I will object to Lewis's theory for other reasons as well. Finally I will discuss briefly a theory of the actual-language relation due to Brian Loar which deals with some of the same issues that Lewis's theory has to face, if only slightly more successfully.

5.2 Truthfulness-By-Silence

Towards the end of "Languages and Language" Lewis considers a possible objection that someone might raise against [L]. For convenience, I will refer to this objection as [O]:

[O]Objection: Let L be the language of P; that is, the language that ought to count as the most inclusive language used by P. (Assume that P is linguistically homogeneous.) Let L+ be obtained by adding garbage to L; some extra sentences, very long and difficult to pronounce, and hence never uttered in P, with arbitrarily chosen meanings in L+. Then it seems that L+ is a language used by P, which is absurd.

A sentence never uttered at all is a fortiori never uttered untruthfully. So truthfulness-as-usual in L plus truthfulness-by-silence on the garbage sentences constitutes a kind of truthfulness in L+; and the expectation thereof constitutes trust in L+. Therefore we have a prevailing regularity of truthfulness and trust in L+. This regularity qualifies as a convention in P sustained by an interest in communication.<5>

Lewis responds to [O] by claiming that the notion of trust in a language that he employs in [L] precludes [O] from presenting a counter-example to [L]. I will notcomment on this claim of Lewis's right now. Rather, I would like to consider the following comments that Lewis makes toward the end of his reply to [O]:

The above objection was originally made, by Stephen Schiffer, against my former view that conventions of language are conventions of truthfulness. I am inclined to think that it succeeds as a counter-example to that view. I agree that L+ is not used by P, in any reasonable sense, but I have not seen any way to avoid conceding that L+ is a possible language - it might really be used - and that there does prevail in P a convention of truthfulness in L+, sustained by an interest in communication.<6>

So I take it that Lewis believes that [O] is a counter-example to a theory - I'll call it "[L']" - which Lewis used to hold:

[L']A language L is used by a population P if and only if there prevails in P a convention of truthfulness in L, sustained by an interest in communication.<7>

[O] and arguments descending from [O] have been taken seriously as objections to both [L'] and [L] by some philosophers, including David Lewis.<8> In the next subsection, 5.2.1, I will argue that [O] fails as an objection to [L']. Afterwards, in 5.2.2, I will make some comments on certain of Lewis's views relevant to the argument in 5.2.1. In 5.2.3 I will indicate why [O] must also fail as an objection to [L]; this will be straightforward given the argument from 5.2.1. In 5.2.4 I will argue that a certain assumption of Lewis's that I allowed for the sake of argument in 5.2.1 is also false. In section 5.3 I will discuss how so many smart people could have come to believe there was an objection to a theory where there was none.

5.2.1 Truthfulness-By-Silence and the Theory [L']

Let us first consider [O] only as an objection to [L'].We are asked in [O] to suppose the following, which I will label [S] for convenience:

[S](i) a population P uses a language L; and (ii) the language L+ is a proper superset of L such that each L+ sentence not also in L is never uttered by a member of P and has some arbitrary meaning.

[O] claims that it follows from [S] that there is a regularity in P of truthfulness in L+ and that this regularity "qualifies as a convention in P sustained by an interest in communication." For the time being, let us suppose that [O] is correct about its following from [S] that there is a regularity in P of truthfulness in L+.<9> I want to examine the further claim made by [O] that this regularity qualifies as a convention. I will argue that it does not.

Recall the six conditions that must hold, according to Lewis, for a regularity to be a convention in a population. The first condition requires that everyone in the population conform to the regularity. Since we are supposing that there is a regularity in P of truthfulness in L+, we are, in fact, supposing that this first condition is met. But are the other five conditions met?

Each of the second through sixth conditions requires that everyone in the population have certain propositional attitudes with respect to truthfulness in L+. In particular, the second condition requires that everyone in the population believes that everyone else in the population conforms to the regularity. So, if there is a convention among the members of P of truthfulness in L+, then everyone in P must believe that everyone else in P is truthful in L+. I will now show that it does not follow from [S] that the members of P have even one of the propositional attitudes required for there to be aconvention in P of truthfulness in L+. I will construct my argument by focusing on whether it follows from [S] that everyone in P must believe that everyone else in P is truthful in L+. I will show that it does not. It will be clear how to use the argument that shows this to show in addition that none of the other propositional attitudes necessary for there to be a convention in P of truthfulness in L+, according to Lewis's analysis of convention, can be said to obtain in P given only [S].

It is helpful to unpack [S] using the definition of convention that Lewis provides<10> :

[S.1]Everyone in P is truthful in L.

[S.2]Everyone in P believes that the others are truthful in L.

[S.3]This belief that the others are truthful in L gives everyone a good and decisive reason to be truthful in L himself or herself.

[S.4]There is a general preference for general conformity to truthfulness in L rather than slightly-less-than-general conformity - in particular, rather than conformity by all but any one.

[S.5]Truthfulness in L is not the only possible regularity meeting the last two conditions.

[S.6][S.1] - [S.5] are matters of common (or mutual) knowledge.

[S.7]L+ is a proper superset of L such that each L+ sentence not also in L is never uttered by a member of P and has some arbitrary meaning.

If [O] is an objection to [L'], then [C] must follow from [S.1] - [S.7]:

[C]Everyone in P believes that the others are truthful in L+.

It is clear that if [C] follows from [S.1] - [S.7] it will have to be because it follows from the conjunction of [S.2] and [S.7]. But [C] does not follow from this conjunction.

Consider the following rough schema [A] where I intend "t" and "t'" to stand for expressions:[A](a) Everyone believes that the others ... t ...

(b) t stands in the such-and-such relation to t'

Is any instance of [A] a valid inference? I think there are at most three possible types of cases where there might be valid instances of [A], but the inference to [C] from [S.2] and [S.7] does not have the form of any of these possible types. Let's look at the types of cases and see why they won't help [O].

1. Consider cases where the expressions substituted for "t" and "t'" in [A] are synonymous. Perhaps such cases could be valid instances of [A]. But even if they are, this doesn't help [O] since the expressions "L" and "L+" are not synonymous - they are not even coreferential.

2. Consider cases where the expressions substituted for "t" and "t'" in [A] stand in some special sort of 'sense relation' other than synonymy, e.g., like that which holds between "bachelor" and "male" such that the inference in [B] is valid, supposing that there is a reading on which it is valid<11> :

[B]Everyone believes that the others saw a bachelor

Everyone believes that the others saw a male

Perhaps such cases would represent valid instances of [A]. If [O] is to be helped by there being instances of [A] that are valid in virtue of such a sense relation between terms, then the sense of "L" will have to be such that it 'contains', so to speak, the sense of "L+". But keep in mind that the terms "L" and "L+" are singular terms that each refer to a specific set. If there are to be senses, that is, I take it, concepts associated with these terms that might stand in the relevant kind of 'sense relation', still, I suppose that the concepts will have each to pick out uniquely the appropriate set: the sense or concept associated withthe term "L" will have to pick out uniquely L and the sense or concept associated with the term "L+" will have to pick out uniquely L+. But, then, what is required is that there are two senses or concepts, call them "" and "+" such that L is the unique and L+ is the unique +, and such that all 's are +'s, that is, such that the sense of "L" 'contains' the sense of "L+". But then L would have to have +. But since L+, by hypothesis, is the unique +, it will follow that L is identical to L+. But L cannot be identical to L+ since by definition L+ is a proper superset of L. So there cannot be two properties and + that will do the work required to provide a valid instance of [A] that will help [O].

3. Consider next cases where we read as de re occurrences the terms substituted for the "t" and "t'" of the schema [A]. The first two types of cases that were considered above were supposing a de dicto reading of the propositional attitudes in instances of [A]. When an expression in the "that"-clause complement of a propositional attitude verb occurs de re, the logical form of the sentence in question is such that the expression takes wide scope with respect to the propositional attitude verb. Thus, when "Cicero" occurs de re in "John believes that Cicero is wise", the logical form of the sentence is conveniently paraphrased as "Cicero is such that John believes that he is wise". We might say that the de re expression binds a position within the "that"-clause, but it has become free of the intensionality of the "that"-clause. One manifestation of this freedom is, famously, that substitution of coreferentials salva veritate becomes possible with respect to such expressions.

There are clear cases of valid instances of [A] when the propositional attitudes aretaken de re with respect to the terms substituted for "t" and "t'". For example:

Everyone believes that the others think Cicero was cool

Cicero is Tully

Everyone believes that the others think Tully was cool

Reading the occurrences of "Cicero" and "Tully" de re, this is clearly a valid inference. This, however, is no help for [O] since L is not identical to L+ as Cicero is to Tully.

At this point it should be noted that it looks like every straightforward way of construing the inference from [S.2] and [S.7] to [C] has failed to show that inference to be valid. But there is a non-straightforward way of construing the inference that should be considered.

Consider the following argument:

[D]Everyone believes that the others like some item in L

L is a proper subset of L+

Everyone believes that the others like some item in L+

If we read only the positions filled by "L" and "L+" here as de re positions, this argument is not valid. If we paraphrase the argument to accent the effect of the de re occurrences of "L" and "L+" the invalidity of [D] becomes plain<12> :

[E](a) L is such that everyone believes that the others like some item in it

(b) L is a proper subset of L+

A counter-example to this argument is the situation where everyone mistakenly believes both (i) that no member of L is also a member of L+ and (ii) that some people dislike every member of L+.

The following would be a valid construal of [D]:

[F](a) Some item of L is such that everyone believes that everyone else likes it

(b) L is a proper subset of L+(c) Some item of L+ is such that everyone believes that everyone else likes it

Here, the expressions "some item of L" and "some item of L+" are read de re, that is, with wide scope with respect to the propositional attitude verb. It doesn't matter whether anyone or even everyone has mistaken beliefs about the relation that holds between L and L+. If both L is a subset of L+ and some member of L is such that everyone believes of it that everyone else likes it, then some member of L+ is such that everyone believes of it that everyone else likes it.

Can this help [O]? The validity of [F] turns on taking the quantifier expressions "some item of L" and "some item of L+" as de re occurrences and giving them wide scope over the propositional attitude verb. But [S.2] and [C] don't have quantifier phrases containing the terms "L" and "L+". They contain just the singular terms "L" and "L+", and the invalidity of [E] shows that the inference from [S.2] and [S.7] to [C] is invalid, strictly speaking, when "L" and "L+" alone are taken to occur de re. But perhaps we can paraphrase [S.2] and [C] so that the occurrences of the singular terms "L" and "L+" are replaced somehow with quantifier expressions that contain those terms. And perhaps then we would be able to construe [S] as requiring an argument akin to [F] above which is valid. Okay, what would the appropriate paraphrase of [S.2] and [C] be?

I suggest the following as a first approximation<13> :

[G](a) Everyone believes that the others are truthful<14> in each member of L

(b) L is a proper subset of L+

We can paraphrase this for clarity by giving the appropriate quantifier phrases wide scope:

[H](a) Each member of L is such that everyone believes that the others are truthful in it(b) L is a proper subset of L+

It is obvious that [H], and therefore [G], is invalid. Suppose that everyone is such that if they live in the U.S., then everyone believes that everyone else likes them. It doesn't follow from this and the fact that the set of persons who live in the U.S. is a proper subset of the set of persons who live in North America that everyone is such that if they live in North America, then everyone believes that everyone else likes them. Suppose that Pierre lives in Canada and that Scott doesn't believe that everyone else likes Pierre. Then, everyone is not such that if they live in North America, then everyone believes that everyone else likes them.

I see no way of coming up with another paraphrase that will do the required job here. And there does not seem to be some other clear strategy for construing the sentences in question so that the inference in question is seen to be valid. I am thereby convinced that [C] simply doesn't follow from [S.2] and [S.7], and that it therefore does not follow from [S]. I conclude further that [O] is not a valid objection to [L'].

There are two further points that are interesting to take note of here.

1. Given only [S], the members of P cannot be said to have even one of the propositional attitudes that are required for there to be, according to Lewis, a convention in P of truthfulness in L+. The reasoning that shows this will be the same as above: the three straightforward ways of construing the appropriate inferences all fail, and the non-straightforward path of replacing the singular terms "L" and "L+" with quantifier phrases that contain those singular terms and reading those quantifier phrases de re also fails;since there are no alternative construals, we must conclude that the inferences are simply invalid. I won't write out the details demonstrating the failure of each such inference since, I believe, the claim here is obvious enough.

2. Nothing really important seems to have happened in my argument because of the intentionality of the attitudes in question. The real problem with the inference from [S.2] and [S.7] to [C] seems to be the same as the problem with the inference in [I]:

[I]Everyone who owns something owns more than Socrates.

Everyone who owns something is someone who is kind to dogs.

Everyone who is kind to dogs owns more than Socrates.

This can be symbolized as follows:

[J](x)(xF Rxs)

(x)(xG Rxs)

And [J] is clearly invalid. Ultimately what made the inference from [S.2] and [S.7] to [C] fail is just what makes the inference in [J] fail, and that has nothing to do with any intentional features surrounding the relation-term "R". It's simply an invalid inference.

5.2.2. Lewis on Attitude Ascriptions In Sensu Diviso

Lewis himself seems to have held a view about the attitudes required for conventions of language that might have led him to what I have been talking about as the non-straightforward construal of the inference from [S.2] and [S.7] to [C]<15> , had he tried to work out the details of the inference. In Lewis (1969) he says the following:

Recall what L is: a certain function whose arguments are pairs of a sentence (a finite sequence of types of sounds or of marks) and a possible occasion of its utterance (a pair of a possible world and a spatiotemporal location therein) and whose values are sets of interpretations (pairs of acode number and a set of possible worlds). Now it is incredible that any ordinary user of L has a concept of any such complicated entity. So how can he be party to any convention regarding L? How can we have expectations and preferences regarding truthfulness in L?

My answer is that he can have them in sensu diviso; to do that, he does not need to share our complicated concept of a possible language. ... All he has to do is to come up with the right expectations and preferences regarding particular instances of what we - but not he - would call truthfulness in L.<16>

The notion of reading a sentence in sensu diviso can be traced to Aristotle, though Lewis seems to have appropriated the notion from Abelard.<17> It applies to many types of sentences, not just attitude ascriptions. But with respect to attitude ascriptions the notion amounts to nothing more than the notion of reading the ascription de re with respect to one or more terms in the ascription's "that"-clause. Earlier in Lewis (1969) when Lewis first introduces his use of the notion of reading ascriptions in sensu diviso he makes the following remarks:

Generality in sensu diviso is problematic because expectation and the like apply fundamentally to states of affairs. If I expect that each driver will keep right, I do expect a state of affairs: each driver will keep right. But if I expect, of each driver, that he will keep right, what states of affairs do I expect? "He will keep right" does not specify any state of affairs until the pronoun has been replaced by some sort of description - verbal, pictorial, or otherwise - of the person in question.<18>

What Lewis is noting about ascriptions taken in sensu diviso is that even when a term in the "that"-clause complement of a verb of propositional attitude is given wide scope, the person to whom the attitude is ascribed is still required, somehow, to have some sort of description - or picture, or other representation. Suppose that I expect that each driver will keep right. Reading this ascription de dicto, it would seem reasonable to suppose that, put roughly, I need to have some way of representing to myself the general idea ofeach driver, but it need not be that I have for each driver a representation of that driver. But if we read the ascription de re, or in sensu diviso, with respect to the phrase "each driver", then we are imagining that for each driver I have some way of representing that driver to myself and I expect that that driver - as represented by my representation - will keep right. I no longer need to represent to myself the general idea of each driver as I seemed to when the ascription was taken de dicto, but now I need to have as many representations as there are drivers. For present purposes there is no need to worry about just what Lewis thought was problematic here. But we can note the following point.

In the first passage quoted above from Lewis (1969) Lewis seems to be trying to prevent his views from entailing that the average citizen needs to have a representation of the complicated entity that he has defined a language to be. He does this by saying that the attitudes that are required according to his theory of the actual-language relation, [L'], be understood in sensu diviso. But saying this by itself cannot help Lewis avoid the consequence he is trying to avoid since it will amount only to the requirement that any singular term referring to a language be read de re when it occurs in a "that"-clause of one of the conditions required for there to be a convention of truthfulness in the language in question. But that doesn't mean that a representation of the language is no longer required somehow. If I believe that Asa is truthful in L, de re, that just means that L is such that I believe that Asa is truthful in it. But, then, L is such that I believe what? "That Asa is truthful in it" isn't a state of affairs until, more-or-less, I have some way of representing to myself what it is. Thus, if the notion of reading an ascription in sensu diviso is to help Lewis to avoid the need for every language user to represent the wholeof the language they use, he is, I believe, going to need to replace references to language via singular terms with references to the sentences of languages via quantifier phrases. Thus, I should be taken as believing not that Asa is truthful in L, but that Asa is truthful in each member of L, where "each member of L" is to be read de re. Thus, what is the case is that each member of L is such that I believe that Asa is truthful in it. I still need representations - in fact now I need as many representations as there are sentences in L, and that may be an awful lot - but I don't need a representation of the whole language, and that seems to be what Lewis was after in the first passage that I quoted above from Lewis (1969).<19>

So I take it that there is at least some reason for thinking that the straightforward ways of construing the inference from [S.2] and [S.7] to [C] may not have been as important to consider as the non-straightforward way that I considered since Lewis himself seems, arguably anyway, to want to be understood as referring to languages in attitude ascriptions via quantifier phrases read de re<20> . In any event, however, the relevant inferences just don't go through.

5.2.3 Truthfulness-By-Silence and the Theory [L]

Up to now I have only considered Lewis's old theory from Lewis (1969) - what I have called [L']. How do things stand with respect to [O] as an objection to the theory of Lewis (1975) - what I have called [L]? If there is a convention in a population P of truthfulness and trust in the language L, sustained by an interest in communication<21> , and if L+ is a proper superset of L, does it follow that there is a convention in P oftruthfulness and trust in L+? No. And this is so independently of the fancier considerations that Lewis adduces in his reply to the objection [O] in Lewis (1975). For if there is a convention in P of truthfulness and trust in L, then there must be at least a convention of truthfulness in L. And if there is to be a convention in P of truthfulness and trust in L+, then there must be at least a convention of truthfulness in L+. It has already been shown that a convention of truthfulness in L does not entail that there is a convention of truthfulness in L+. So if [O] is going to be an objection to [L], it will have to be that it follows from [S.7] and there being a convention in P of truthfulness and trust in L that there is a convention in P of truthfulness in L+. But there is nothing in the statement of the conditions that must be met for there to be a convention of trust in L that helps us get truthfulness in L+ given truthfulness in L. So, it follows that [O] cannot be an objection to [L] any more than it was to [L'].

5.2.4Truthfulness in L+ Given Truthfulness in L, and Truthfulness and Trust in L+ Given Truthfulness and Trust in L

There is still one more point that should be made for this discussion to be complete. Up until now I have assumed that given the description of L+ stated in [S.7] and the supposition that there is a convention in P of truthfulness in L, it follows that there is at least a regularity of truthfulness in L+, even if that regularity doesn't amount to a convention of truthfulness in L+. But even that much is mistaken.

In essence, I took as a working hypothesis that [N] follows from [K] and [S.7]:

[K]Everyone in P is truthful in L.

[N]Everyone in P is truthful in L+.But the notion of being truthful in a language is defined by Lewis, as we saw above:

So we can rewrite [K] and [N] as [K'] and [N'] respectively:

[K']Everyone in P tries never to utter any sentences of L that are not true in L.

[N']Everyone in P tries never to utter any sentences of L+ that are not true in L+.

The three "straightforward" ways of construing the inference from [K'] and [S.7] to [N'] all fail. Trying a non-straightforward construal leads us to the paraphrases [K*] and [N*]:

[K*]Each sentence of L is such that everyone in P tries never to utter it unless it is true in L.

[N*]Each sentence of L+ is such that everyone in P tries never to utter it unless it is true in L+.

But [N*] does not follow from [K*] given [S.7]. It follows that the assumption is false that there is a regularity of truthfulness in L+ given a regularity of truthfulness in L.

Notice, again, that the failure of the inference actually has nothing to do with any intentional features of the verb "to try". The problem is more basic: the inference is simply invalid.

Finally, it should be clear, then, that it will not follow from [S.7] and there being a regularity of truthfulness and trust in L that there is a regularity of truthfulness and trust in L+.

5.3 Truthfulness-by-Silence Versus CMTs

There is no problem directly of truthfulness-by-silence for either Lewis's theory of the actual-language relation in Lewis (1969), or for his theory of the actual-language relation Lewis (1975), even though Lewis himself as well as others have thought that there is such a problem.

There remains the question, why do so many philosophers believe there is such a problem when there is none? My answer is that things can be set up so that, though there is no truthfulness-by-silence problem directly for Lewis's theory, there is a supposition, vaguely motivated by certain epistemological concerns, which can be appended to Lewis's theory and for which there are truthfulness-by-silence style problems. Philosophers who have objected to Lewis's theory by raising truthfulness-by-silence style problems, I believe, were really objecting to the epistemologically motivated supposition. Let me describe the epistemological concerns and the supposition that I have in mind here.

Lewis wanted, as was mentioned above, to avoid attributing to language users all sorts of complicated knowledge about set theoretic stuff. People apparently don't have that sort of knowledge, and Lewis rightly wanted to avoid his theory entailing that they do. To do this he felt obliged to construe the attitudes his theory employs in the non-straightforward way that I discuss above. But construing the relevant attitudes in that way, as we saw, results in Lewis's theory requiring enormous numbers of propositional attitudes of each language user, since languages can be enormously large. But having so many attitudes doesn't seem at all plausible unless there is some relatively small basis forthem. But there are two ways that a finite basis can be given, it would seem, for the appropriate attitudes. On the one hand, it can be supposed that some finite axiomatization is available for each usable language that would entail all the necessary meaning facts about the language: this would amount to the supposition that there is a CMT for each usable language. On the other hand, Lewis can suppose that given the sentences that actually get uttered by a population, the rest of the facts about the language the population uses are fixed: this is the supposition that given the actual usage<23> of a population, the language that the population uses is determined.

It seems that Lewis objected to the CMT route, not because he objected to CMTs in themselves, but because, again, this would require that ordinary language users had complex knowledge of set-theoretic stuff that they apparently don't have. So, the only option available to Lewis is to suppose that actual usage can be an adequate basis for pinning down the entire language that a population uses. The truthfulness-by-silence problems are objections to this supposition. They show that this supposition is wildly implausible. The point of the truthfulness-by-silence problems here seems to be a special case of the general worries now often attributed to Kripkenstein.

This, anyway, is how I reconstruct things so that it is not simply out of the blue that philosophers have taken seriously truthfulness-by-silence style objections to Lewis's theories of the actual-language relation. Of course, this reconstruction is not altogether charitable to Lewis since it ascribes to him a view that is pretty implausible, namely, the view that actual usage can determine an indefinitely large language. But perhaps this seems less uncharitable if we suppose that being caught on the horns of the dilemma,CMTs or actual usage, Lewis grabbed at actual usage and hoped for the best.

But the story doesn't just end here.

Lewis's theory<24> of the actual-language relation commits him to holding that language users have enormous numbers of propositional attitudes. This supposition is implausible without some story about how all these attitudes can be founded in some appropriately small basis. As I have already suggested, this amounts, it would seem, to asserting the disjunction: either all those attitudes hold because some small subset of them hold, or there are CMTs for large languages. But the former disjunct here is utterly implausible, as was discussed above: small sets, by themselves, don't uniquely determine big sets in the appropriate way. So, it seems that for Lewis's theory to be at all plausible is for all usable languages to enjoy CMTs. But, as was argued in the last chapter, even if there were CMTs for all usable languages, standing in some sort of cognitive relation to a CMT does not constitute a necessary condition for meaning,<25> and this is true, even if it turns out that all usable languages enjoy CMTs.

So, it seems that Lewis's theory is unacceptable because it seems to require as a necessary condition for language use that language users stand in some sort of cognitive relation to CMTs. This, we have seen is not a necessary condition for language use.

In addition to this, of course, it should be repeated that cognitive relations or not, it is not even entirely clear that there are going to be CMTs for all usable languages. As was argued above in chapter 4, there don't seem to be any good arguments that natural languages in fact enjoy CMTs. So, without independent arguments for CMTs, I would take a theory committed to CMTs to be somewhat tenuous at best.Independent of all this there are other reasons to be doubtful of Lewis's theory of the actual-language relation. I will discuss these now.

5.4 Conventions and Interest in Communication

In this section I will briefly discuss two sorts of objections to Lewis's theory of the actual-language relation that I think are important. The first has to do with Lewis's notion of convention and the second with the "interest in communication" clause of his theory.

1. The first sort of objection that I want to discuss here addresses itself to Lewis's theory of the actual-language relation by way of addressing itself to Lewis's analysis of the notion of convention. There are actually a few such objections but I will only mention those I think most telling.

One of the objections of the sort I have in mind is due to Tyler Burge. Lewis's analysis of the notion of convention requires that it be common knowledge among the parties to a convention that there is an alternative regularity that could have been adopted as a convention which would have served just the same, or near enough, purposes as the convention which they in fact have adopted. But Burge argues rather cogently that not only is this requirement of common knowledge of alternative regularities too strong, but the necessity of alternative regularities is, common knowledge of them or not, too strong:<26>

...[C]onsider the...example of the sentimental hat tippers. The convention of tipping one's hat to a passing stranger becomes a national trademark. The citizens are sentimentally attached to this mode of greeting and its associations with their culture, to the extent that each would rather fightfor the traditional greeting, or give up greeting strangers altogether, than switch to another one, even if the others were to switch. ...[T]his traditionalism does not seem to affect the conventionality of the actual practice.<27>

I think this sort of story works rather well as a counterexample to Lewis's account of the notion of convention.

Stephen Schiffer offers the following gloss of the notion of convention:

There prevails in [population] G a convention to do an act (or activity) of type X when (or only when) . . . iff it is mutual knowledge* amongst the members of G that

(1)there is a precedent in G for doing X, or an agreement or stipulation that one will do X, when (or only when) . . .;

(2)on the basis (in part) of (1), almost everyone in G expects almost everyone in G to do X when (or only when) . . .;

(3)because of (2), almost everyone in G does X when (or only when) . . .<28>

This account of the notion of convention differs from Lewis's in a number of obvious regards. Among these is the lack of Lewis's requirement that more-or-less equally attractive alternatives to regularities be available to the members of a population for the existence of a conventional regularity to prevail among them.<29> Schiffer's account of convention, therefore, would seem to escape Burge's counterexamples to Lewis's account. But Schiffer's account of convention shares other problems with Lewis's which I will get to in a bit.

Grandy, in his review of Lewis (1969), after noting Burge's arguments also suggests a counterexample to the view that language is a matter of convention in the sense of Lewis's analysis:

It is clear that...ordinary use applies 'convention' to regularities that do not fit Lewis's analysis. For example, suppose that, in Orwell's 1984society, when the Newspeak language was introduced almost everyone detested that language and would have much preferred not to speak it, but, because of the power of the state, everyone spoke the language out of fear of the consequences. There is a common sense of 'convention' in which it is a convention that these people speak the language, but they clearly do not fit Lewis's conditions.<30>

Of course, it is possible to suppose that Grandy's Orwellian society does not really use Newspeak by convention, regardless of the loose sense in which we suppose that any use of language is conventional. This supposition would protect Lewis's analysis of convention. But it would seem that this is not a supposition open to the theorist who wants to understand meaning in terms of the notion of convention since the case does seem a genuine case of language-use and therefore of meaning. A theorist who denied that the language-use here was a matter of convention would have to allow that some cases of expression-meaning are non-conventional. Since Lewis wants to hold that meaning is a matter of convention and is gearing his analysis of convention to just that sort of thesis, the only alternatives available to him are to deny that the Newspeak language is really being used or to amend his analysis of convention.

Grandy's case, I believe, is also a counterexample to Schiffer's view of convention. For since the members of Grandy's Orwellian society speak Newspeak not because they expect everyone else to do so but because they are afraid not to, it cannot be that any of the members of the society know that the others speak Newspeak because they expect the others to. We can even add to Grandy's story that everyone knows that the others speak Newspeak because they are afraid not to. So Schiffer's condition (3) is not satisfied: the members of the society don't mutually know* that the others speak Newspeak because they expect everyone else to. So Schiffer must either hold that Grandy's Orwellians don'tuse Newspeak, or that some cases of language use are non-conventional, or that his analysis of convention is amiss. The first two alternatives don't seem available directly to Schiffer, in Schiffer (1972) anyway, so he seems to need to revise his analysis of convention.

The real problem, in my view, for both Lewis and Schiffer with respect to Grandy's case is the strength of the common knowledge condition.<31> For this requires that each person who is party to a convention believe about all the others who are party to the convention that they participate in the convention for certain reasons. But, at least with respect to language-use, it seems that the sorts of reasons we can attribute to others for speaking our language can vary quite a bit, and yet the others still speak our language. If common knowledge of the motives of others in being party to a conventional regularity is necessary for conventional regularities, it seems clear that language-use cannot be a matter of conventional regularity. Believe anything you want about why I speak English and you will still speak the same language as me if you are an English-speaker too.

Now this doesn't mean that common knowledge is not required for conventions. Lewis and Schiffer both have reasons for motivating their respective common knowledge conditions.<32> So, perhaps the moral to draw is that the conditions for something being a convention are, on reflection, such that we shouldn't want to say that language-use is really a matter of convention. I will not attempt a view on this matter here at all. I just point it out.

I believe there are other good counterexamples to Lewis's analysis of the notion as well,<33> and my point in mentioning all this is to help diminish the sense that we shouldreally want to make free use of the notion of convention in a theory of meaning. The use of the notion of convention will still remain something of an explanatory debt even if a satisfying theory of meaning availing itself of that notion were provided. Granted that it is a platitude that language is conventional, still, this need not entail that a theory of meaning need use a notion of convention. The use of such a notion is not wholly unproblematic and it might be hoped that a theory of meaning could be constructed using other notions and that, when this is done, the conventionality of language can be explained somehow less formally in terms of this theory.

Thus, the absence of a completely satisfying analysis of the notion of convention might make a theorist slightly less eager to use the notion of convention in constructing a theory of meaning.

2. Another important objection to Lewis's theory of the actual-language relation is due to Stephen Schiffer. Lewis's theory of the actual-language relation has it that a population uses a language just in case there prevails among them a convention of truthfulness and trust in the language sustained by an interest in communication. It is this "sustained by an interest in communication" clause - I will call it the communication-clause for convenience - that the present objection deals with.

Schiffer claims that the communication-clause was added by Lewis to avoid a certain sort of counterexample.<34> Without the communication-clause Lewis's theory would simply hold that a convention of truthfulness and trust in a language were necessary and sufficient for a population to use the language. To be truthful in a language is to try not to utter the language's sentences unless they are true in the language and to be trustingin L is to suppose that others are truthful in L. But, in that case every convention can be made out to be one of truthfulness and trust in a language. For, suppose that there is a convention to drive on the right side of the road among a certain group of people P. The act of driving paired with the proposition that I am driving on the right (ignore the indexicality here) will be a language, call it D, with one sentence. Since we are supposing that there is a convention among the members of P to drive on the right side of the road, then each member of P will avoid driving unless they are driving on the right, or, in other words, they will try to drive only if they are driving on the right. To say that they avoid uttering the single D-sentence unless it is true in D is just another way of saying that they avoid driving unless they drive on the right. And that is to say that the members of P are truthful in D. And since they each expect the others adhere to the convention of driving on the right, they are also trusting in D. So, there would seem to be a convention of truthfulness and trust in D. But it also seems that merely having a convention of driving on the right is not to speak a language. Thus, something is badly lacking in the theory that says that a convention of truthfulness and trust in a language is sufficient for using the language in the relevant sense.

Lewis, according to Schiffer, added the communication-clause in order to avoid this sort of counterexample. But as Schiffer also points out, the communication-clause is not sufficient to avoid this sort of counterexample completely. For, suppose that the members of our group P drive on the right because they believe that if they don't their ability to communicate with each other will suffer somehow, perhaps because they have some complex astrological beliefs that correlate driving on the right with communication. In that case their convention seems sustained by an interest in communication, but they still don't speak a language. The notion of an interest in communication is apparently too vague or general to be of good use to Lewis. And it is hard to imagine how any rewording of the communication-clause alone can block counterexamples of the present sort. Schiffer claims the following about this matter:

Talk of trying not to utter a sentence of L unless it is true in L will perforce give way, I submit, to talk of not uttering a sentence unless one means the proposition L().

This seems right. The sort of convention Lewis is after seems to be a convention of speaker-meaning of some sort. It is hard to see how a theory that doesn't avail itself of some notion of speaker-meaning will manage to get things right.

5.5 Loar's Actual-Language Relation

I would like to discuss briefly another account of the actual-language relation that is due to Brian Loar.<35> Loar's theory uses a notion of convention like Lewis's<36> , but unlike Lewis, Loar uses a notion of speaker-meaning. So Loar's theory can be seen, perhaps, as a theory that tries to answer the problem, just discussed, that Schiffer raises for Lewis's theory with regard to Lewis's communication-clause.

Basically, Loar's theory is, in a slightly simplified paraphrase:

P uses L just in case there prevails among the members of P a convention such that for any sentence and proposition such that =L(), if a member of P seriously utters , then that person means .<37>

The notion of a serious utterance here is meant to filter out cases where a speaker utters a sentence non-literally or for some non-communicative purpose, and other such cases. I am not entirely sure that such a notion can be appropriately spelled out, but I will accept it in the present discussion to make other points.

This theory clearly avoids the counterexample Schiffer raised against Lewis's theory with regard to Lewis's communication-clause. But it retains, pretty much, the same spirit of Lewis's early theory, [L']. Recall that Lewis says that to be truthful in a language L is "to avoid uttering any sentence of L unless one believes it to be true in L". That is, roughly, Lewis's theory has it that language-use is a matter of there being a convention such that one tries to make it a necessary condition for uttering a sentence that one believes it expresses in the language a true proposition. Loar's theory has it that language-use is a matter of there being a convention such that a necessary condition for uttering a sentence is that one means the proposition it expresses in the language. So, there is something of a parallel between the two theories: they both require conventional regularities describable in terms of material conditionals stating a necessary condition for uttering a sentence of a language.

Loar, in Loar (1976) at any rate,<38> feels, like Lewis, that it is best to avoid requiring language users to have some sort of representation of the whole of the language that they use. We can take that to mean that Loar wants to avoid requiring language users to stand in some sort of cognitive relation to a CMT for their language. Thus, Loar suggests, though I state this in my terms, that the reference to the language within the scope of the propositional attitudes that make up the convention he describes be read as a de re quantifier expression which deal with each of the sentences of the language individually. So, effectively, Loar wants his theory to be read as follows:P uses L just in case for each sentence and proposition such that =L(), there prevails among the members of P a convention such that if a member of P seriously utters , then that person means .

Since Loar eschews the idea of language-users standing in cognitive relations to CMTs, he suggests that sense can be made of the enormous number of attitudes required by this theory by founding the infinitude of attitudes in some large but finite corpus of actually uttered sentences. His basic idea is that there will be some large but finite set of actually uttered sentences of a language used by a population and that this set will determine, somehow, grammatical rules that in turn determine the rest of the language. This, of course, is a requirement that there be a CMT for the language used, but Loar's scruples about CMTs were about ordinary language-users knowing such theories, not about natural languages enjoying them. In any event, Loar requires that a CMT for a language be determined by some suitably large finite corpus of actual utterances by members of the population in question.<39> The idea here is, I think, highly implausible for reasons Schiffer spells out in response to a suggestion by Lewis which is essentially the same as the suggestion presently under consideration.<40> The problem, in a nut-shell, is just the one that you would expect: a finite corpus will determine infinitely many incompatible CMTs and there are no good standards for picking out one of these. In other words, a finite corpus will determine infinitely many languages. I think the arguments against this proposal are compelling and I won't dwell on it further here.

Loar thinks CMTs will be required in a theory of meaning for another reason as well. He raises an important counterexample to the above theory.<41> It is plausible that no one would utter the English sentence 'Apollo set the evening sky ablaze' with theintention of trying to get an audience to believe that, literally, Apollo set the evening sky ablaze. For Loar, that means that it is plausible that it is unlikely that anyone would utter that English sentence to mean that Apollo set the evening sky ablaze. But, still, it is possible, and let's suppose that it is true, that English speakers are such that were anyone to utter that sentence, what they would be trying to get their audience to believe is something like that there is an intensely beautiful sunset this evening. Consider the language English' which is such that 'Apollo set the evening sky ablaze' is paired with the proposition that there is an intensely beautiful sunset this evening. Notice that on present assumptions, if Loar's theory is correct, then English-speakers actually use English' and not English. For English-speakers are not such that utterance of 'Apollo set the evening sky ablaze' would be a sufficient condition for meaning that Apollo set the evening sky ablaze, but they are such that it would be a sufficient condition for meaning that there is an intensely beautiful sunset this evening. Thus, it appears that Loar's theory fails to provide a sufficient condition for expression-meaning.

Loar's diagnosis of the problem here is that his conditions for literal meaning are not sensitive to the contributions to meaning made by syntactic structures and simple lexical items employed in sentences. 'Apollo set the evening sky ablaze' does literally mean among English-speakers that Apollo set the evening sky ablaze, and it seems to mean this among us independently of what actual communicative intentions we might have on any actual occasion of its utterance: it seems to mean what it does because the words and syntactic structures that make it up contribute to its meaning in ways that are set in advance of any actual utterance of it. Loar suggests, on the basis of this sort ofdiagnosis, that the theory of meaning will have to evoke a grammar for a language, that is, a CMT for the language, somehow.<42>

But there is another diagnosis for the problem here. It could be that while Lewis's notion of a convention of truthfulness sustained by an interest in communication is too broad a notion for the theory of meaning, the notion of speaker-meaning understood as a sort of intention to get an audience to have a certain belief is too narrow. Perhaps what is needed is a notion of speaker-meaning which will allow that in cases where one speaks metaphorically or non-literally by exploiting features of the literal meanings of sentences, as in the 'Apollo...' case one says the literal meaning though one suggests or implicates what one really wants to get the listener to believe. The standard architecture of Gricean analyses of speaker-meaning isn't generally understood to allow for such a notion, though it is an intuitive enough one. One can say one thing and yet not mean it. Perhaps, then, this sort of notion of saying is better suited to a theory of expression-meaning. I make this point here only to leave it alone for the time being. But I will return to it in my final chapter.

5.6 Summary

Lewis wants to avoid requiring language users to know lots of set-theoretic stuff. To do this he seems compelled to read the attitudes towards languages that his theory of the actual-language relation requires as de re attitudes towards each of the sentences of these languages. But this ends up requiring that language users need to have an enormous number of propositional attitudes. This new requirement only seems plausible if eitherthere are CMTs for usable languages or the large numbers of attitudes could be had somehow by only having some small subset of them. But the second of these disjuncts is wildly implausible and truthfulness-by-silence problems aim at showing this. Most of this chapter was devoted to making this point. But it follows from all of this that the more serious point is that a Lewis-style theory will not be able to avoid commitment to language-users having cognitive relations to CMTs. But since it was already seen in chapter 4 above that having a cognitive relation to a CMTs is not necessary condition for using a language, it was concluded that, whatever else Lewis's theory does, it does not state a necessary condition for using a language.

This point being made, I went on to mention two further issues with respect to Lewis's theory of the actual-language relation. First, I noted that there are problems with Lewis's and Schiffer's analyses of the notion of convention and that to use this notion will incur some explanatory debt. And second, I noted that it seems unlikely that a theory in style like Lewis's theory can succeed without deploying some notion of speaker-meaning.

Finally, I discussed Brian Loar's theory of the actual-language relation which seems to address the second of the two issues just mentioned. But, I suggested that the notion of speaker-meaning it suggests turns out to be too strong and that perhaps a weaker notion will be required.

In the next chapter I will discuss a type of theory that tries to do without CSTs and without the use of the notion of convention, but it also tries to work without directly tying expression-meaning to notions of communicative intention.Notes