CONVENTION AND LANGUAGE
ABSTRACT: This paper has three objectives. The first is to show how David Lewis' influential account of how a population is related to its language requires that speakers be 'conceptually autonomous' in a way that is incompatible with content ascriptions following from the assumption that its speakers share a language. The second objective is to sketch an alternate account of the psychological and sociological facts that relate a population to its language. The third is to suggest a modification of Lewis' account of convention that will allow one to preserve the claim that there are conventions of language.
1. CONVENTIONS AND SHARED LANGUAGES
Suppose that I am writing in English, that you are reading it, and that lots of people, including the two of us, speak it. In short, we, along with many others, share a particular language called "English." Given this supposition, the question arises, just what is it about us that makes it the case that English is the language that we share? or, to put the question more generally, just what does a group of people do in order to make a particular language their own? This question is considerably less pressing if one believes that we each speak our own idiolect and that there are no such shared languages, but anyone who is at all sympathetic with arguments for the 'social' character of language must face the question of what relates populations to the languages that they share. 
Perhaps the most influential response to this general question was provided almost thirty years ago by David Lewis. According to Lewis, "it is a platitude that language is ruled by convention," and if one understands languages as "functions from strings of sounds or of marks to sets of possible worlds," then one can provide a "description of the psychological and sociological facts" whereby a particular language is used by a population by specifying the conventions that relate them to it. 
Lewis' proposal is a particularly sophisticated member of a long tradition of conventional accounts of language. Furthermore, it shares with its predecessors an important presupposition that can also be found, for instance, in the following passage from Locke's conventional account of a population's relation to its language.
The comfort and advantage of society not being to be had without communication of thoughts, it was necessary that man should find out some external sensible signs, whereof those invisible ideas, which his thoughts are made up of, might be made known to others . . . . Thus we may conceive of how words . . . came to be made use of by men as the signs of their ideas; not by any natural connection that there is between particular articulate sounds and certain ideas . . . but by voluntary imposition, whereby such a word is made arbitrarily the mark of such an idea. 
Locke's speakers are "conceptually autonomous" -- they have their ideas whether the world or their fellows cooperate with them or not. It is merely for instrumental reasons that they speak as their fellows do. If speaking a language is just a matter of attaching sounds to independently accessible ideas, then the fact that we apply roughly the same sounds to roughly the same ideas may be unproblematically explained by a convention.
While Lewis certainly makes no appeal to Locke's theory of ideas, his account also presupposes that speakers are conceptually autonomous. Autonomous speakers can come to share a language in virtue of forming intentions such as "I will apply the word 'cat' to this idea, as long as my fellows do" or "I will apply the word 'cat' to that sort of thing, as long as my fellows do," and the plausibility of conventional accounts depends upon the antecedent availability of such conditional intentions. However, recent work by, among others, Saul Kripke, Hilary Putnam and Tyler Burge has seriously undermined the model of the autonomous speaker/thinker and, with it, Lewis' account.  The required conditional intentions are not available to non-autonomous speakers independently of their actually sharing a language, and so conventional accounts of our relation to language, as traditionally conceived, are unworkable.
2. CONCEPTUAL AUTONOMY AND SHARED LANGUAGES: THE PROBLEM
Convention was finished by 1968, when it was still commonly presupposed that each speaker had, on her own, command of the conditions of application for the terms in her language. In the following decade, however, it became clear that our 'natural' content ascriptions (ascriptions tied to the assumption that we do, in fact, share a language) suggested that the meanings of a speaker's terms and the contents of her thoughts extended beyond both her discriminatory capacities and the descriptions available to her. A speaker's discriminatory capacities often appeared to underdetermine or even misidentify what she was talking about.
A typical example of underdetermination is Putnam's use of "beech" and "elm."  While Putnam cannot distinguish beeches from elms, we still take the extensions of "beech" and "elm" in his idiolect, like ours, to be the sets of all beech and elm trees respectively. There is, as Putnam puts it, a "division of linguistic labor," and one can rely upon the knowledge of other speakers to help determine the referents of one's own terms. By far the most famous case of misidentification is Burge's discussion of Bert and his belief that he has arthritis in his thigh.  In this example Bert tells his doctor that he has arthritis in his thigh, the doctor informs him that arthritis only affects the joints, and Bert gives up his belief. In spite of the fact that Bert's beliefs may be uniquely true of something else (a set of ailments affecting the joints and thighs which we will call "tharthritis"), Bert is taken to have here corrected a mistaken belief about arthritis. Burge further argues that, had Bert's community used the term "arthritis" to pick out tharthritis, Bert would have had a set of true beliefs about tharthritis (and no arthritis beliefs at all). What Bert is talking and thinking about seems, in part, determined by how the word "arthritis" is used in his community.
These ascriptions suggest, then, that speakers are occasionally unable to completely or even accurately specify what they are talking about. Furthermore, these ascriptions follow from the assumption that individuals speak the same language as their fellows. One upshot of Kripke's, Putnam's and Burge's work is thus that the claim that we are conceptually autonomous is incompatible with the natural assumption that we share a language with our fellows.  This incompatibility suggests that we can preserve the conceptual autonomy that Lewis' account requires only at the expense of denying that we really share a language, but such a denial deprives Lewis' account of the very subject matter it purported to explain.
To see why content ascriptions that suggest that we are not conceptually autonomous pose problems for Lewis' account, we should look at Lewis' account in more detail. Lewis claims that a population, P, is related to the language, £, that it speaks through conventions of truthfulness and trust in that language. Being truthful in £ involves trying never to utter any sentence which is not true in £, and being trusting in £ involves imputing truthfulness in £ to others.  According to Lewis, the community's having conventions of truthfulness and trust in £ amounts to its satisfying the following six conditions: 
(1) There prevails in P at least a regularity of truthfulness and trust in £.
(2) The members believe that this regularity of truthfulness and trust in £ prevails among them. Each believes this because of his experience of others' past truthfulness and trust in £.
(3) The expectation of conformity gives everyone good reason why he himself should conform.
(4) There is in P a general preference for general conformity to the regularity of truthfulness and trust in £. Given that most conform, the members of P want all to conform.
(5) The regularity of truthfulness and trust in £ has alternatives.
(6) All these facts are common knowledge in P.
Presupposing that we are conceptually autonomous allows for a clear reading of the convention of truthfulness: each speaker knows the criteria of application for the terms of £ and makes an effort to use sentences of £ only when their criteria of application are satisfied. If conceptual autonomy is given up, however, it becomes less clear what is it about a speaker's behavior that makes it an instance of a convention of truthfulness in £ rather than some similar language £*. The cases of misidentification and underdetermination discussed above seem to affect the unproblematic applicability of Lewis' account in the following way.
The first condition, that there prevail in P at least a regularity of truthfulness and trust in £, clearly has problems with cases of underdetermination such as Putnam's use of the terms "beech" and "elm." While most speakers may follow a regularity of truthfulness and trust in £, their behavior may also be in accordance with regularities of truthfulness and trust in some alternate language £*. The things Putnam might say about elms (that they are deciduous trees, that they don't have maple or oak leaves, etc.) are all true in English, but the same sentences would all also be true in a language much like English where "beech" referred to elms and "elm" referred to beeches. As a result, he seems to display a regularity of truthfulness in both English and its alternate. Even more serious problems are raised by the 'misidentification' examples such as Bert's use of "arthritis." In such cases, the speaker not only displays a regularity of truthfulness in an alternate language £* (where "arthritis" refers to tharthritis), but also fails to display a regularity of truthfulness in £ itself.
In much the same way, expecting others to be truthful and trusting in £ could, if speakers really were autonomous, be understood in terms of a set of concrete expectations about how others will apply their terms and understand one's own. However, if we really do speak the same language as our fellows, such a set of expectations would woefully underdetermine, or even misidentify, the language in question. For instance, Bert's beliefs about what is the prevailing use in the community will be in line with £* and not £. The speaker with a non-standard conception of a term's extension will be much like Lewis' 'madman', who "thinks he belongs to a population which uses £, and behaves accordingly, and so can be said to use £, although in reality all the other members of this £-using population are figments of his imagination." 
3. POSSIBLE REPLIES
3. 1. INCORPORATING THE DIVISION OF LINGUISTIC LABOR
One way one might try to reconcile Lewis' account with the suggestion that language users are not autonomous is to build our reliance upon experts into how we understand what a regularity of truthfulness and trust is. After all, the regularity of truthfulness requires only that the agents try to be truthful in £, not that they actually be so.  One way in which speakers can try to be truthful in £ is, in those cases where their knowledge underdetermines the extension of a particular term, to actually consult someone whom they take to have the relevant knowledge. If Putnam really had to distinguish a beech from an elm, he could do so -- by asking an expert. 
However, there are at least two problems with this line of response. First of all, it is considerably more plausible for the elm-cases (where the speaker is aware that he has not mastered the conditions of application for the term and knows of others who have) than it is for arthritis-cases (where the speaker thinks (incorrectly) that he has mastered the term's conditions of application, and so would not be disposed to ask an expert when asked to pick out the relevant class). A second, and more serious, problem for attempts to repair Lewis' account by appealing to our willingness to consult experts is that the phenomena which pose problems for conventional accounts are not limited to cases in which there is a sufficiently evolved expert community close at hand. Putnam points out that we take people in the 18th century to mean what we do by, say, "water," even though no one at the time would have been able to distinguish chemically distinct, but phenomenologically indistinguishable, liquids from water. This suggests that the experts in a population (if there even are any) may themselves be unable to completely or accurately specify the extension of their terms. Furthermore, for the names of many historical figures and some other historical terms, the people who knew just how the terms were properly applied may be long gone. As a result, any account which relies upon the availability and accuracy of actual experts seems inadequate to account for the phenomena discussed in section 2.
3.2. NARROW CONTENT AND LEWIS' 'INDEXICAL' ACCOUNT
Another way Lewis could try to account for the phenomena discussed in section 2 (indeed the way he did suggest when faced with them in their nascent form) is to claim that they relate to factors which are not, strictly speaking, part of the language. More specifically, Lewis suggested that Kripke's, Kaplan's and Donnellan's work on proper names could be reconciled with his account of language by being understood as merely drawing to our attention "a subtle sort of indexicality."  We can allow the extension of a proper name to depend partly upon the causal history of the speaker's use of it, while denying that the intension or meaning of the name depends upon such causal factors, by "including a causal-history-of-acquisition-of-names coordinate in our indices and letting the intensions of names for a speaker determine their extensions only relative to that coordinate."  However, even if this were a plausible account of proper names, it seems much less attractive when extended to other terms in the language.
First of all, the notion of 'indexicality' that Lewis appeals to is nothing like what one typically finds with indexicals and demonstratives. Such terms can be analyzed in terms of Kaplan's notion of character (i.e. "functions from possible contexts to contents"),  and any competent speaker can use the character of a demonstrative such as "I," "you," "him," "here" or "now" to get its content from the context of utterance. No such ability accompanies 'indexical' accounts of proper names and other terms, since there is no comparable rules for retrieving specific information from the context of utterance that distinguishes, say, "John" from "Peter."
Furthermore, Lewis' suggestion about proper names gains part of its plausibility from the intuitions that proper names are not, in some sense, 'part of the language' (they don't appear in the dictionary etc.). As a result, it may not seem counterintuitive to claim that two speakers who meant someone completely different by names such as "John" or "Peter" could still count as speaking the same (indexical) language. However, Putnam's and Burge's examples suggest that the relevant phenomena are not restricted to proper names, but extend through most of our language. As a result, two speakers could have different truth-conditions for most of their sentences ("soaking in water can cure arthritis" meaning, say, either soaking in H2O can cure arthritis or soaking in XYZ can cure tharthritis) and still count as speaking the same (indexical) language. Speakers of the same indexical language would systematically misinterpret the truth-conditions of each others sentences, so it is doubtful that sameness of indexical language captures much of the notion of sameness of language that is associated with sharing a language.
Indeed, what one gets with an 'indexical' account of the phenomena discussed in section 2 are extremely fine-grained 'narrow' languages in terms of which what I mean by say "Nixon" is shared with my Twin-Earth counterpart, but not with my neighbor, or even with my past selves.  Such 'narrow' meanings are best suited to give the semantics for languages that supervene upon the current internal physical states of speakers, and Lewis has often expressed sympathy with focusing on such 'narrow' contents when thinking about belief and language. He supports "the common presumption that the meaning of an expression for a speaker depends only on mental factors within him"  and claims that interpretation should be interested in "belief in a narrowly psychological sense: the belief that governs behavior, the belief that is 'in the head', determined by brain states and their causal roles."  Accordingly, it might seem as if Lewis would have no problem with the narrow notion of language characterized above.
Unfortunately, this narrow and indexical notion of language seems very far from the notion of language to which Lewis originally wanted to explain our relation ("functions from strings of sounds or marks to sets of possible worlds"). Lewis initially claimed that by a language, he meant "an interpreted language . what many logicians would call a language plus an interpretation for it,"  but on the indexical account, most of the 'interpretation' is done by the indexicality. That is to say, English and Twin English, while they may have much in common, are different interpreted languages. The interpretation is precisely what distinguishes them. While there is a real question of how populations are related to their interpreted languages, there is much less of a mystery about how the members of a population are related to their indexical languages, since such indexical languages are spoken by agents simply in virtue of their being in a certain internal state. Lewis can claim that the phenomena discussed in section 2 are just cases of 'indexicality' only at the expense of re-characterizing languages in such a way that there is no interesting question of how individuals are related to them. Indeed, the interesting questions would all relate to how and why this type of 'indexicality' works, and these questions are not addressed by Lewis' six clauses.
Finally, Lewis cannot simply say that it is the population's relation to its indexical (rather than its interpreted) language that he is interested in. This is because indexical languages are not shared within a population. An economist and a metallurgist who shared few beliefs about, say, uranium, would speak different 'indexical languages' with respect to that word. Conventions were supposed to explain how populations were related to their shared languages, but the indexical version of the conventional account, by presupposing an understanding of what languages are in which they are not shared, prevents the question Lewis' account was supposed to answer from even arising.
The problems with Lewis' indexical account show up in some form or another in all accounts which make use of a distinction between broad and narrow contents. If we distinguish 'broad' from 'narrow' languages, neither type will be a suitable candidate do play the part of the language £ in Lewis' six clauses. Since we do not speak our 'broad languages' autonomously, if £ is a broad language, we will not be able to independently satisfy Lewis' six clauses in the way that Lewis' account requires. This was, after all, the original point, and it was problems accounting for how a population could be related to the proper names in a broad language that first prompted Lewis to propose his indexical account. However, 'narrow languages' (such as Lewis' indexical language) will not be shared by the members of a population, so they are not suitable candidates for £ in Lewis' six clauses either. There is no question of how the members of a population are related to the narrow language that they share, since there are no such shared narrow languages.
3.3. NON-REDUCTIVE ACCOUNTS.
A final way to make Lewis' six conditions compatible with the phenomena discussed in section 2 would be to understand Lewis' theory as a non-reductive one. That is to say, we could treat Lewis' theory as an account that allows us to appeal to the abilities the speaker has in virtue of speaking the language in order to explain how she is related to it. For instance, if my being truthful in £ could only be understood as trying to apply the word "elm" to large deciduous trees with white bark and green leaves, then no unique language will be picked out. However, as an English speaker, I can form the intention to apply the word "elm" only to elms, and by focusing on my trying to fulfill this intention, clause 1 will pick out a unique language.
There is, in some sense, no doubt that a non-reductive reading of Lewis' six conditions is compatible with the phenomena discussed in section 2. Nevertheless, the non-reductive reading of Lewis' account has a number of serious problems. First of all, it seems highly unlikely that Lewis (who entitled his 'autobiographical' piece in the Companion to the Philosophy of Mind "Lewis, David: Reduction of Mind") intended his account of convention to be interpreted as a non-reductive one. Those writers who appealed to Lewis' account when trying to reduce meaning in public languages to meanings in individual languages, and ultimately speaker's intentions, were clearly correct to do so.  When Lewis speaks about a speaker's behavior, he is quite explicit about how he is interested in "raw behavior -- body movements and the like," not "behavior specified partly in terms of the agent's intentions." 
Furthermore, the non-reductive reading can only amount to an acceptable defense of Lewis' account if it can also be shown that Lewis' conditions, so understood, can do the work he originally intended them to do. The problem that Lewis initially set himself was "if £ is a possible language, and P is a population of agents, what relation must hold between £ and P in order to make it the case that £ is the language of P?"  However, the non-reductive reading of the conventional account explains how speakers form intentions and expectations corresponding to the terms in £ rather than its close relative £* in terms of the fact that they speak £ rather than £*.  What Lewis ultimately hoped to explain must be presupposed in the non-reductive account. As a result, the non reductive reading preserves Lewis' account only at the expense of depriving it of the explanatory power that Lewis clearly intended it to have.
4. TOWARDS AN ALTERNATIVE ACOUNT
If conventional accounts do not adequately describe "the psychological and sociological facts" whereby a particular language is used by a population, how are populations related to the languages that they share? What follows are a few tentative remarks on what such psychological and sociological facts might be.
4.1. SHARING A LANGUAGE
First of all, in order to specify the "mental and behavioral and cultural factors"  that allow a population to share a particular language, we need to specify just what it is to share a language in the first place. The idea that each speaker is conceptually autonomous can make a certain answer to this question seem very compelling: two autonomous speakers share a language if they speak the same language, and they speak same language if the two languages they independently speak are of exactly the same type. Two autonomous speakers thus share a language only if they speak in exactly the same way, and once sharing a language is understood this way, it follows that ('strictly speaking') no two speakers share a language.  Populations are then understood as sharing languages only to the extent that their independent idiolects overlap. 'Loosely speaking,' they can be said to 'share' a language if their idiolects are similar enough for them to communicate effectively.
Still, we need understand neither sharing a language nor speaking the same language in the way that the model of the autonomous speaker suggests. In particular, the "division of linguistic labor" metaphor suggests a way in which speakers can share a language without speaking in just the same way. I can be understood as sharing a language with other members of my community because I look for their help when deciding how my terms are correctly applied, and they do the same with me. It is in this cooperative sense of sharing that I can share a language with my fellow speakers in more than just a 'loose' sense. Two autonomous speakers, even if they spoke in exactly the same way, would not, in this sense, share a language at all. 
Once the idea of sharing a language is pried apart from the idea of speaking in exactly the same way, a correspondingly altered conception of what it is for two people to speak the same language emerges. In particular, two people (whether they speak in just the same way or not) speak the same language if they share it. Non-autonomous speakers are understood as speaking the same language not in virtue of speaking two type identical languages, but rather in virtue of both sharing a single (token identical) language. This brings with it, of course, a correspondingly different conception of what the languages we share are. The languages we share are not abstract universals (as one might expect if they were "functions from strings of sounds or of marks to sets of possible worlds"), but rather particular practices. There may, of course, be an abstract language corresponding to each such shared practice, but two distinct practices, and thus two distinct shared languages, could both be instances of the same abstract language. There is a sense in which an animal molecularly indistinguishable from, but evolutionarily unrelated to, the animals we call "tigers" would not be a tiger, and in much the same way, someone who spoke a language (call it English*) that instantiated the same function from strings of sounds to sets of possible worlds as English did, but was historically unrelated to it, would not be an English speaker. There is an obvious sense in which English and English* instantiate a single type, but there is an equally obvious sense in which speakers of the two do not take part in the same practice. We would not share a language with speakers of English* in the way that we share a language with speakers of English.
It is only when sharing a languages is understood in the cooperative practice-driven sense that the claim that we share a language is, strictly speaking, true. As a result, an explanation of how populations are related to the language they share should work with this more cooperative conception of sharing.
4.2. OUR RELATION TO LANGUAGE
Of course a story about the "psychological and sociological facts" which relate a population to such shared languages is still owed, and while Lewis suggests that our relation to language be understood in terms of a theory along the lines of Hume's discussion of the origins of justice and property,  it may be that Hume's discussion of chastity provides a more promising model of this relation. According to Hume, chastity came to be seen as a virtue because, as he puts it, "In order to induce the men to . . . undergo cheerfully all the fatigues and expenses, to which [helping to raise children] subjects them, they must believe, that the children are their own."  If males were only willing to bring up children that they knew to be their own, one could imagine how the practice of chastity could arise in a community that focused explicitly upon the instrumental value of the practice. Females behaves chastely in order to insure the help of their male counterparts, while males helps provide for the children of their female counterparts because they feel assured that these children are also their own.
Hume, however, did not explain the purported chastity of the women in his society this way. Indeed, he claimed that since we "are apt to over-look remote motives in favor of any present temptation,"  such purely 'intellectual' motives could not explain chaste behavior. A society where men and women were able to act on such motives would, in many respects, behave much as Hume's purportedly did , but the explanation of their chaste behavior would be very different from what it was for Hume's contemporaries. Hume claims, plausibly enough, that chaste behavior should be explained in terms of the presence of a belief that it is simply wrong for women not to be chaste.
Someone who believed that chastity was a virtue would strenuously deny that there would be nothing wrong with a society that found a non-chaste way to convince men to help bring up children. Even if someone's belief that chastity is a virtue produces behavior that could be rationally justified for instrumental reasons, these reasons would not be the ones that are motivationally active for that person. Consequently, if the more instrumental account were meant as a description of the "psychological and sociological facts" which explain chaste behavior, its being radically at odds with the psychology of the participants in the practice would be a serious problem. Even if some sort of explicit rationalization explained the emergence of the practice, it certainly does not explain how it is perpetuated among its practitioners.
In much the same way, while coordinating the way we speak can be justified independently in terms of pragmatic norms relating to ease of communication, coordination need not actually be produced or maintained by our recognition its instrumental value. Even if a society that satisfied Lewis' six conditions would speak much the way we do, it seems highly unlikely that all groups of actual language users will satisfy Lewis' six conditions.
For instance, the combination of Lewis' conditions 5 and 6 (that here be an alternative language, £', available to the population, and that all the conditions be common knowledge)  require that all speakers be aware that there are equally good alternatives to the language they speak. This requirement seriously undermines the psychological plausibility of Lewis' account. For instance, it is surprisingly common for people in isolated monolingual communities to think that there really isn't an alternative to the way they speak.  Names are not considered arbitrary in many communities and they are often taken to have some sort of 'internal' connection to their bearers.  Furthermore, even when we become aware of the existence of other languages, various forms of linguistic chauvinism persist (foreign languages are too nasal, too guttural, illogically constructed, spoken too quickly, inherently more difficult to learn and use, etc.). Finally, even in the absence of chauvinism, populations can still feel that other languages that they have come across are not available as an alternative for them. Consider, for instance, University of Pittsburgh Linguist Dan Everett's description of his contact with the Piraha (a tribe native to the Amazon) and their attitude towards language:
The Piraha view language as a defining characteristic of group identity in a strong sense - you speak the language of your group. Americans in their opinion are identified partially by their ability to speak other languages (since the only Americans they know are the only people they know that speak more than one language). Even so, it is difficult for them to grasp the fact that I can speak their language. They will often have conversations about me in front of me and then look astounded when I enter into the conversation - even after all the years that I have worked there. When we go to villages that we haven't worked in much (i.e. other Piraha villages), they literally look at us with their mouths open in disbelief when we address them in Piraha. They eventually answer us, but the experience is clearly unsettling for them. 
It should not be surprising to learn that the Piraha also believe that they are incapable of ever speaking any language other than their own, and remain to this day a predominantly monolingual community.
While such attitudes about language make Lewis' account psychologically implausible, they suggest an alternative explanation of how linguistic coordination is achieved similar to Hume's explanation of his population's purported relation to chastity. Linguistic coordination is often produced in a population by their belief that there simply are correct and incorrect ways of using their words, and that they all speak a single language that can be known more or less adequately. Different tracts of the language are understood as known to differing degrees by different people, and 'discovering' how the language is correctly used can thus be understood as a cooperative venture. Even the most authoritative speakers are typically viewed as being responsive to, rather than as helping to constitute, facts about how the language is correctly used. The conception of language attributed in psychological explanations of our coordination is a conception according to which languages are particulars, and thus a conception according to which they can be understood as shared by speakers who do not speak in just the same way. 
5. CONVENTIONS OF LANGUAGE
When Lewis wrote that the claim that there were conventions of language was a platitude that "only a philosopher would dream of denying,"  he did not take himself to be one of the philosophers in question. However, if conventions are understood in Lewis' sense, we may just have to deny that the platitude is true. Nevertheless, rather than giving up the platitude, one can take the preceding story about language use, and the platitude that there are conventions of language, to suggest that conventions should not be understood in quite the way that Lewis suggests. Indeed, some of Lewis' critics have suggested that all that is required for a regularity to be a convention is that a substantially different regularity be able to perform the same function equally well (i.e. a regularity counts as a convention in a population if it satisfies Lewis' conditions 1 and 5).  The claim that there are conventions of language would thus be reduced to the comparatively modest claim that there are alternative ways of speaking. On the thin account of convention, unlike Lewis', a convention would not be so-called "because of the way it persists,"  and the claim that there are conventions of language would not entail that speakers are related to their language "by convention."
The thin conception of convention seems, however, a little too thin. It allows us to preserve the claim that there are conventions of language only at the expense of counting any custom at all as a convention. Consequently, it loses sight of something that Lewis' account may focus on too much. Namely, talk of conventions brings with it associations with rational justification that need not come with talk of customs. Even if one is ultimately unwilling to accept Lewis' account, one should admit that it captures at least something of our concept of convention. (On the other hand, it would be clearly inadequate as an account of custom.) Any adequate account of convention should be able to account for the difference between conventions and non-conventional customs, and the thin account of convention is unable to do this.
Not all customs are conventions, and one plausible account of the difference between the two is that conventions are more "reflection resistant" than non-conventional customs. Conventions need not be undermined by the realization that they are arbitrary, while non-conventional customs may be. A custom or practice counts as a convention if it admits of the sort of rational reconstruction according to which it could be maintained through clauses like 1-6 even after its arbitrary nature is recognized. To admit of such a reconstruction, the practice must be in the interests of all of the participants involved, and many customs don't satisfy this requirement. Conventions would thus be a rationally justifiable subset of customs, not a group that is coextensive with them.
Once the class of conventions is distinguished from non-conventional customs in this way, we can also distinguish (as Lewis himself does) those who are party to a given convention (which involves satisfying Lewis' conditions 1-6) from those who merely conform to it (which does not).  A convention is only conventional-for a person who satisfies 1-6. In these terms, many regularities which are conventions need not be conventional-for those involved, and a convention could become a conventional-for its participants when they come to understand it as having alternatives. However, while Lewis treats a regularity as a convention only if it is conventional-for its participants, we need not. When we merely conform to a convention, it is still a convention that we are conforming to. To capture the range of practices that we take to be conventional, we need only require that they have the potential to be conventional-for a group of rational agents. 
Since a regularity is a convention only if it has the potential to become conventional-for the people who take part in it, being party to conventions is still conceptually more primitive than merely conforming to conventions. Even if a regularity's being conventional doesn't always play an essential role in explaining the preservation of that regularity, something like Lewis' six conditions could thus remain at the 'core' of our concept of convention. Whether the proper extension of a term should extend beyond its core is often a matter of how much of the term's putative extension satisfies the requirements of the core. When reflect on our concepts, we tend to focus on the core, but this does not entail that the concepts are not correctly used beyond it.  Lewis' attaching the term "convention" just to regularities which are conventional-for their participants, to the exclusion of those which are only potentially so, may stem from his apparent belief that those who conform to conventions without being party to them will consist mainly of "narrow minded and inflexible people," and, of course, "children and the feeble minded."  Once this rather optimistically rationalistic picture of human psychology is given up, the core concept of convention turns out to capture much less of the term's putative extension (in particular, it fails to capture many linguistic conventions). In light of this, it seems better to understand the term as applied correctly to certain specifiable extensions of its core as well. In particular, it should extend to those practices which have the potential to be conventional-for its participants.
In conclusion, one can hold on to the claim that speakers' are conceptually autonomous only if one is willing to treat each speaker as speaking an independent idiolect, and thus give up the claim that we share a language. Since Lewis' account of how we are related to our language requires that we be conceptually autonomous, it is not a suitable candidate for explaining how populations are related to the languages they share. Nevertheless, Lewis' account of convention can still be used to demarcate conventions from other customs, and to the extent that we can provide a rational reconstruction of the linguistic practices of communities who perceive their usage as governed by more robustly independent linguistic norms, we can still say that there are conventions of language.
I'd like to thank Robert Brandom, Joe Camp, Dan Everett, Mitch Green, Bernard Kobes, David Lewis, Ram Neta, John McDowell, Jamie Tappenden, and members of the audience at the Eastern APA for comments on earlier versions of this paper, and the Canada Council for their generous support.
1. For examples of writers who support the idea that we each have an independent idiolect, see Davidson 1986 and Chomsky 1988. For arguments for language's 'social character', see Putnam 1975, Burge 1979 and Kripke 1982. The assumption that we share a language can be made while admitting that there are differences in pronunciation etc. It may be the case that no two English speakers pronounce "arthritis" in exactly the same way, but they still all refer to arthritis by "arthritis." Of course, even at a purely semantic level, the assumption that we actually share a language is far from uncontroversial. While the assumption that we share a language must eventually be defended, the following paper will simply investigate some of the consequences of this assumption. (For a defense of the assumption, see Jackman 1996, 1998).
2. Lewis 1969, p. 1, 1969, p. 177, and 1970, p. 190.
3. Locke 1689, p. 404-5. (Book III, Chapter II.)
4. Of course, one popular response to this literature is to argue that there are, in fact, two types of content. 'Broad' content, which is determined by the external factors the Kripke, Putnam and Burge draw our attention to, and a more internalistic 'narrow' content, which would remain autonomous. Such a distinction will turn out to be, however, no help to Lewis. Our 'broad-concepts' would be non-autonomous, so Lewis' account cannot explain how they are shared. Our 'narrow-concepts,' while autonomous, are not shared, so Lewis' account is not needed to explain why they would be. (These points are discussed in further detail in section 3.2.) Unless stated otherwise, all reference to contents, meanings or languages should be understood as referring to the 'broad' variety.
5. Putnam 1975, p. 226.
6. Burge 1979.
7. Once again, if one were to distinguish broad and narrow contents, one could argue that narrow contents would be autonomous while broad contents would be shared, but the fact would remain that the contents that are autonomous are not the contents that are shared, and Lewis' account of convention requires precisely that the contents that are shared be the ones that are autonomous. For more on this, see section 3.2.
8. Lewis 1972, p 167. See also Lewis 1969, p. 177.
9. The six conditions are found in Lewis 1972, p. 167-8. (See also Lewis 1969, p. 177-180.)
10. Lewis 1972, p.182. Furthermore, the 'irregularities' in Bert's speech cannot be assimilated to deliberate untruthfulness, which represent a failure to be true-in-£ without entailing that we do not take part in a convention of truthfulness in £. Lewis allows a liar to be a user of £, even though his usage actually conforms to, say, £*, because he believes that conventions of truthfulness in £, not £*, prevail in the community (Lewis 1972, p. 182), but this is manifestly not the case with someone like Bert.
11. Lewis 1969, pp. 178-9.
12. The availability and importance of such appeals is stressed in Fodor 1994, p. 34.
13. Lewis 1972, p.173. He has in mind the works such as Kripke 1972, Donnellan 1970, and Kaplan 1968.
14. Lewis 1970, p.228 (italics Lewis').
15. Kaplan 1989.
16. Indexical languages would have to be understood in terms of something like White's notion of "partial character," which takes one from contexts of acquisition to characters, and partial characters are only shared between speakers who are (or at are least close to being) functional duplicates. (White 1991, p. 39.). For a discussion of such 'twin cases' see Putnam 1975.
17. Lewis 1970, p.228.
18. Lewis 1983b, p.121. See also Lewis 1972, pp. 177-78.
19. Lewis 1969, p. 162.
20. See, for instance, Stalnaker 1984, Appiah 1985
21. Lewis 1974, p. 114.
22. Lewis 1969, p.176.
23. There may, of course, be other ways of having the concept of, say, elm without actually speaking English (most obviously, one could have access to the concept in virtue of speaking, say, German), though the same general criticisms will hold for an account which just takes a non-reductive approach to each individual concept without taking a non-reductive approach to the language itself.
24. Lewis 1969, pp. 206-7.
25. Hence it is not surprising that those who understand speakers as conceptually autonomous can find themselves, for instance, arguing that a convention of truthfulness in a single language is a "limiting case" which is "never reached" (Lewis 1969, p.201), that "no two individuals share exactly the same language" (Chomsky 1988, 36-7), or that "no two people actually do speak the same language" (Davidson 1992, 260).
26. While one can understand languages as shared in the sense that I 'share' a haircut with everyone who has a haircut sufficiently similar to mine (though I probably never 'fully' share my haircut with anyone), one can also understand them as shared in the sense that a cooperative project is shared by its participants. If I am working on a cooperative project, the actions of the other members determine the character of the project I am working on in a way that the people who 'share' my haircut do not affect the haircut I have.
27. Lewis 1969, p. 3.
28. Hume 1888, p. 570.
29. Hume 1888, p. 571.
30. But perhaps not in all. For instance, if chastity were practiced solely for such reasons, it would not be expected of women past child-bearing age.
31. The common knowledge condition is taken very seriously by Lewis, who claims that "anyone who might be called an exception [to the common-knowledge requirement] might better be excluded from P" (Lewis 1969, p. 76-7).
32. I had, until quite recently, such an attitude towards our expressions of, say, surprise or pain. Exclaiming "oww!" after bumping one's knee seemed a perfectly 'natural' response, and while I was aware that people from other cultures spoke different languages, it had not occurred to me that they might have something like "aiiii!" as their 'natural' pain-response.
33. As Vygotsky puts it with respect to children, "The data on children's language (supported by anthropological data) strongly suggest that for a long time the word is to the child a property, rather than a symbol, of the object." (Vygotsky 1962. p.50.) Hence the importance in some cultures of not letting people know your 'true' name.
34. Dan Everett, personal communication. The Piraha is a fascinating language in its own right (it has a extensive aspect system, lacks belief predicates, can be whistled, etc.), and I would recommend looking for Prof. Everett's grammar of the language (Cambridge University Press, forthcoming).
35. The importance of such conceptions in constituting the language we speak is discussed in greater detail in Jackman 1996, 1998.
36. Lewis 1972, p. 166.
37. After criticizing the psychological implausibility of Lewis' account, Burge offers such an account of convention when he argues that a regularity is conventional if "as a matter of fact -- whatever the participants may believe -- it is within the power of the participants to have learned an incompatible regularity that would have served substantially the same social functions without demanding significantly greater effort on the part of the participants." (Burge 1975, p. 254. )
38. Lewis 1972, p. 181.
39. Lewis makes room for a distinction of this sort when he claims that "a child or idiot will conform without reason; if so, he is not party to the convention" or that "When children and the feeble-minded conform to our conventions, they may not take part in them as conventions, for they may lack any conditional preference for conformity to an alternative; or they may have the proper preferences, but not as an item of common knowledge." (Lewis 1969, pp. 51, 75.)
40. To claim that a practice would admit of a rational reconstruction along the lines of Lewis' six conditions should not be confused with Lewis' suggestion that his account might be more psychologically plausible if one understood the common knowledge requirements as potential or even counterfactual (requiring that the speakers merely not disbelieve any of the six clauses). (Lewis 1972, pp. 165-6, 181.) Like the reconstructive interpretation, the counterfactual one would adapt Lewis' account to the psychology of speakers who never considered that there could be alternatives to the way they spoke. However, the counterfactual understanding of Lewis' account is still unable to cope with linguistic chauvinists who simply deny that alternative languages are as good or learnable as their own (i.e.: they actively disbelieve 5), and linguistic traditionalists often maintain their allegiance to the 'correct' use of a word in spite of their awareness that most members of the community no longer do so. The notion of potentiality required by the reconstructive account is much weaker than that required by Lewis' counterfactual suggestion, it requires only that, if the members of the population were completely rational, they could engage in the relevant regularity by satisfying Lewis' conditions. Lewis' account, one the other hand, requires that they actually satisfy the six conditions whenever the usage of the rest of the society changes or the arbitrary nature of the practice is pointed out to them.
41. For a discussion of how concepts can have such a 'radial' structure, with a core and various extensions of it, see Lakoff 1987.
42. Lewis 1969, p. 75.
Appiah, A.: 1985. Assertion and Conditionals, London: CUP.
Burge, T.: 1975. "On Knowledge and Convention." The Philosophical Review 84 (249-255).
Burge, T.: 1979. "Individualism and the Mental." in French, Uehling & Wettstein. Midwest Studies in Philosophy IV: Studies in Metaphysics. Minneapolis: University of Minnesota Press (73-121).
Chomsky, N.: 1988. Language and the Problem of Knowledge. Cambridge: MIT.
Davidson, D. 1986.: "A Nice Derangement of Epitaphs." in LePore, E, (ed.). Truth and Interpretation. Oxford: Blackwell.
Davidson, D. 1992.: "The Second Person." in French, Uehling & Wettstein. Midwest Studies in Philosophy XVII: The Wittgenstein Legacy. Notre Dame: Notre Dame University Press (255-267).
Donnellan, K.: 1970. "Proper Names and Identifying Descriptions" Synthese 21: 335-58.
Fodor, J. 1994.: The Elm and the Expert. Cambridge: MIT.
Hume, D.: 1888, 1978. A Treatise of Human Nature. Oxford: OUP.
Jackman, H.: 1996. Semantic Norms and Temporal Externalism, Ph.D. Thesis, University of Pittsburgh.
Jackman, H.: 1998, "Individualism and Interpretation" Southwest Philosophy Review, V.14, n. 1.
Kaplan, D.: 1968. "Quantifying In." Synthese 19: 178-214.
Kaplan, D.: 1989. "Demonstratives." in Almog, Perry & Wettstein. (eds.). Themes from Kaplan. New York: OUP.
Kripke, S.: 1972. Naming and Necessity. Cambridge: Harvard.
Kripke, S.: 1982. Wittgenstein on Rules and Private Language. Cambridge: Harvard.
Lakoff, G.: 1987. Women, Fire and Dangerous Things, Chicago: The University of Chicago Press.
Lewis, D.: 1969. Convention. Cambridge: Harvard UP.
Lewis, D.: 1970. "General Semantics." In Lewis 1983a.
Lewis, D.: 1972. "Languages and Language." In Lewis 1983a.
Lewis, D.: 1974. "Radical Interpretation." In Lewis 1983a.
Lewis, D.: 1983a. Philosophical Papers, Volume 1. New York: OUP.
Lewis, D.: 1983b. "Postscript to 'Radical Interpretation.'" In Lewis 1983a.
Lewis, D.: 1994. "Lewis, David: Reduction of Mind." In Guttenplan (ed.) A Companion to the Philosophy of Mind. Cambridge: Blackwell.
Locke, J.: 1689, 1975. An Essay Concerning Human Understanding, New York: OUP.
Putnam, H.: 1975. "The Meaning of 'Meaning.'" reprinted in his Mind Language and Reality. New York: CUP.
Stalnaker, R.: 1984. Inquiry. Cambridge: MIT.
Vygotsky, L. S.: 1962 Thought and Language. Cambridge: MIT.
White, S. L.: 1991. The Unity of the Self. Cambridge: MIT.
Department of Philosophy
University of Toledo
Toledo, OH 43606