Constitutive Versus Normative Accounts of Speech and Mental Acts

Abstract

At the end of his “Promisings and other Social Acts: Their Constituents and Structure,” Kevin Mulligan briefly considers the question of the normativity of speech and mental acts. This has been a matter hotly debated in recent years; several authors (including Kathrin Glüer-Pagin and Åsa Wikforss) have contended that, properly understood, superficially looking normative notions that we deploy in characterizing such acts should be understood in a constitutive, nonnormative sense (in contrast with views such as the one recently defended by Tim Williamson about assertion, on which this is a constitutively normative act). The goal of my chapter would be to critically examine these suggestions and defend a normative account.

1 Introduction

In different places in his work (for instance, in the very last paragraph in Mulligan (1987) , throughout Sect. 3 in Mulligan (1999) , and in the talk “Against Rampant Normativism” given in Parma, 2000), Kevin Mulligan considers and suggests a view recently advocated in a more committed way by Glüer and Wikforss (2009a, pp. 48–52) .¹ This is the claim that constitutive accounts of meaning (force, content, or both; mental, linguistic, or both), i.e., accounts in terms of “internal relations,” are at odds with normative ones: A constitutive account fully explains any intuitive sense that “oughts” are in place when it comes to meaning, and is in fact incompatible with a further explanation in terms of genuinely prescriptive oughts. Glüer and Wikforss (2009, p. 48) put it like this: “there is a clear sense in which states or performances that are internally related cannot stand in normative relations ”; in a less committed way, Mulligan suggests: “perhaps the time has come to consider the view that semantic and cognitive relations are internal but not normative” (“Against Rampant Normativism,” § 6) .

Both Mulligan and Glüer and Wikforss mention Frege as a predecessor of the view they put forward. The main reason Glüer and Wikforss (2009a, p. 49 ff) provide in favor of the view appeals to the point that “oughts not only imply cans, they also imply the possibility of violation, of what we could call ‘forbidden combinations’”; they go on to show convincingly that, even though any plausible claim about internal relations in this field—say, modus ponens-shaped transitions for possessors of the conditional concept, transitions from perceptual experience to belief regarding perceptual content, etc.—should allow for the possibility of exceptions, such exceptions do not adequately count as the needed violations. Mulligan (1999, § 3) suggests this point too, in that he emphasizes that rule following manifests itself primarily in avoidance of rule breaking: Where rules “rule, the possibility that rule-breaking rules should not be ruled out.” In addition, he appeals to a disparity between deontic notions and the cognitively constitutive properties he takes to be fundamental: “The properties of being obligatory and of being allowed are properties which do not admit of degrees…[but] prima facie justification admits of degrees” (Mulligan 1999, § 3). In this note, I will only discuss the former, more worrying point, on the assumption that the latter admits the reply that it is not just internal relations of justification that are determined by meaning -constitutive facts . I will argue that, although the point is correct as far as it goes, the friend of a fully normative account of the relevant notions should not worry.

2 Norms and Meaning

Famously, Wittgenstein ’s later philosophy establishes a constitutive connection between meaning (linguistic, or mental) and norms or rules ; Kripke’s (1982) widely debated interpretation has further encouraged discussion about the role of normative reasons in characterizing languages. Kripke’s presentation appeals to rules or norms such as these:

(Plus) If one means addition by “ +,” one ought (to answer “125” if asked “68 + 57?”).

(Circ) If one means being circular by “circular,” then one ought not (to apply “circular” to o if o is not circular) and one is permitted (to apply “circular” to o if o is circular).

The ensuing discussions have evinced to what extent almost everything in this debate is highly controversial; here, I will limit myself to developing the line that I find compelling, without going in any depth into those debates.²

I have phrased the illustrative examples above by explicitly providing scope indications so that the obligation in (plus) mentions an act, answering, which I understand to involve asserting/judging; so does the one in (circular) under the interpretation I intend for applying, which I also take to be a form of assertion or judgment . This is because I agree that, in an alternative interpretation on which “to apply” just means predicating—cf. Glüer and Wikforss (2009b, § 2.1.1)—we do not have, here, genuine rules giving normative reasons for agents to act. Let me elaborate on this .

Most speech acts have representational contents, which can be shared by acts of different illocutionary types. Representational contents—propositions, in one construal of them—encode correctness conditions with respect to different possible worlds: Conditions that (putting aside necessarily true or necessarily false contents) obtain with respect to some worlds, perhaps the actual world among them, and not with respect to others. But the notion of “correctness” here at stake is an etiolated one for present purposes. I can assert p, or order p, or (in fiction-making mood) propose that you imagine p. Arguably, that p is not the case in the actual world (not just now, but atemporally, speaking sub specie æternitate) does not make the order thereby incorrect (if it was one worth giving, a good even if ultimately unsatisfied try), and certainly does not make the act of fiction-making incorrect; but it does make the assertion incorrect. Similarly, that p is (also atemporally) the case in the actual world does not make the order correct—the order might have been a stupid one, or one given without any proper authority, thus, providing no genuinely normative reason for the recipient to act; nor, again, does it thereby make the act of fiction-making correct: p might well be a totally uninteresting thing for anybody to imagine. Arguably, however, it might be enough for making the assertion correct. Hence, the correctness conditions encoded by propositions do not constitute normative reasons in the sense that interests us: reasons for a rational subject to act. They merely affect a division of representational contents into two classes (with respect to any possible world): those obtaining in, or correctly representing the world, and those not obtaining.³ That a representational act represents an obtaining proposition does not, by itself, furnish any agent with a particular normative reason to act; for such a reason to be determined we need, in addition, the illocutionary type—the “point” of the act .⁴

I take it that rules like (plus) and (circular) are intuitively “primitively compelling,” to use Peacocke’s (1987) terminology.⁵ Is there an explanation for why this is so? Such an explanation would appeal, firstly, to accounts of speech and mental acts in terms of constitutive rules, along the lines of Austin’s (1962) , Searle’s (1969) , Alston’s (2000) , or Williamson’s (1996/2000) for the specific case of assertion; secondly, to a view of meanings of natural language sentences and mental states as act potentials, as also developed by Alston (2000)—for the linguistic case—and others. I will now briefly outline both ideas.

Williamson (1996/2000) claims that the following norm or rule (the knowledge rule) is constitutive of assertion, and individuates it:

(KR) One must ((assert p) only if one knows p).

The obligation (KR) imposes is not all things considered, but prima facie; in any particular case, it can be overruled by stronger obligations imposed by other norms .⁶ Now, in the course of the debate that Williamson’s proposal has engendered, other writers have accepted the view that assertion is defined by constitutive rules but have proposed alternative norms; thus, Weiner (2005) proposes a truth rule, (TR), and Lackey (2007) a reasonableness rule, (RBR) :

(TR) One must ((assert p) only if p).

(RBR) One must ((assert p) only if it is reasonable for one to believe p).

As a first motivation for his account, Williamson (1996/2000, p. 252) mentions intuitive conversational patterns: We challenge assertions politely by asking “How do you know?” or, more aggressively, “Do you know that?” (Williamson 1996/2000, p. 252). Austin (1962, p. 138) already pointed out these patterns:

[I]t is important to notice also that statements too are liable to infelicity of this kind in other ways also parallel to contracts, promises, warnings, &c. Just as we often say, for example, ‘You cannot order me’, in the sense ‘You have not the right to order me’, which is equivalent to saying that you are not in the appropriate position to do so: so often there are things you cannot state—have no right to state—are in no position to state. You cannot now state how many people there are in the next room; if you say ‘There are fifty people in the next room’, I can only regard you as guessing or conjecturing.

As Hindriks (2007) notes, these facts about our practices of appraising assertions are by themselves insufficient to justify normative accounts. For we also evaluate assertions relative to (invoking Rawls’ well-known distinction) merely regulative norms, norms that regulate, relative to certain purposes, acts in themselves constitutively nonnormative—for instance, as witty, polite, or well phrased. Hindriks shows that norms for assertion could be merely regulative of a constitutively nonnormative practice, definable in the motivating reasons Gricean account that Bach and Harnish proposed, GA below (‘R-intending’ there is to be explicated in terms of Gricean communicative intentions). The regulative norms in question would then be derived from an ultimately moral sincerity rule such as (SR):

(GA) To assert p is to utter a sentence that means p thereby R-intending the hearer to take the utterance as a reason to think that the speaker believes p.

(SR) In situations of normal trust, one ought to be sincere.

I (2004) have argued, however, that there are considerations against nonnormative accounts such as (GA), and in favor of normative accounts stronger than those provided by (TR) or (RBR). Firstly, there are well-known objections to Gricean accounts of speaker meaning in general, of which (GA) is a special case for assertoric meaning, which strongly suggest that normative accounts are preferable (Vlach 1981; Alston 2000, Chap. 2; Green 2007, Chap. 3) . Thus, the clerk in the information booth uttering “The flight will depart on time,” or the victim saying to his torturer “I did not do it,” or any of us uttering to our neighbor in the lift “nice weather, isn’t it?,” may well lack the Gricean intentions that (GA) requires for them to assert but they are asserting all right; any normative account would capture this, for, no matter their intentions, they are still committed to knowing what they say (or having justification for it, or being truthful, whatever the proper rule is). In the second place, there are situations in which we may have overwhelming prudential or moral reasons to violate (SR), i.e., not to be sincere. If the “regulative rules” account were correct, any sense that we are violating a prima facie norm—even if all things considered we are doing the right thing—should vanish in those cases; but it does not, or at least it does not according to intuitions many of us share—exactly as it happens in analogous cases involving promises, as Rawls (1955) pointed out in his influential argument against utilitarian “regulative rules” accounts of them.

Williamson (1996/2000) provides additional justification for his specific normative proposal: First, the account explains what is wrong in a version of Moore’s paradox with “know’”instead of “believe”: A, and I do not know that A (Williamson 1996/2000, pp. 253–254). Second, mathematics provides for formal situations where the speaker’s sensitivity to the norms of assertion is highlighted; in those situations, being warranted to assert p appears to go hand in hand with knowing p. Third, an account based on TR seems at first sight preferable: given that the truth rule is satisfied whenever the knowledge rule is, but not the other way around, it provides for a practice with fewer violations of its governing rule; some evidential rule could then be explained as derived from TR, and considerations not specific to assertion. However, the truth rule does not individuate assertion; alternative speech acts like conjecturing, reminding, or swearing also involve a truth rule (Williamson 1996/2000, pp. 244–245). Moreover, reflection on lotteries (cases in which, knowing that you hold a ticket in a very large lottery, I assert “your ticket did not win” only on the basis of the high probability of the utterance’s truth) question the validity of any such alleged derivation (Williamson 1996/2000, pp. 246–252). Finally, intuitions about many cases in which we assert without knowing can be made compatible with the view. In some cases, it is reasonable for us to think that we know, even if we do not; what we do is not permissible, but it is, as we feel, exculpable. In some cases, additional values (saving someone from danger, enjoying a relaxed conversation) are at stake, allowing again for exculpation based on their contextual relative strength (Williamson 1996/2000, pp. 256–259).

Speech acts like assertion are thus normative; they are constituted by rules such as (KR). This applies both to those done by resorting to purely conventional means, if there are any, and also to those done in an indirect way, having recourse to the sort of pragmatic mechanism that Grice (1975) famously characterized as conversational implicature , as for instance when we “ask,” “who the hell would want to see a film with that plot?” A full explanation of our intuitive feeling concerning the appropriateness of (plus) or (circular) is provided by a view of what natural language s are for such as Alston’s (2000) , according to which the literal, primary meanings of sentences in natural language are speech act potentials.

Why “potentials?” This is one of the points at which important differences between the linguistic and the cognition case come up. Consider utterances of “he is hungry.” This is a declarative sentence, and it is reasonable to argue that such sentences are used by default to make assertions. If so, and according to rules such as (KR) or (TR), in uttering it with its default meaning, a speaker commits himself to knowing (or to the truth of) a certain proposition. Intuitively, to identify such a proposition we need information about the context in which the utterance is made; knowledge of English tells us that the referent of “he” should be some male made salient (“demonstrated”) by the speaker; but we need further information about how such salience is established in the context, in order to identify it. Recent debates in linguistics and the philosophy of language about the semantics/pragmatics distinction suggests that this point is widespread in natural languages. Literal and direct utterances of “it is raining” in a context commit their utterers to propositions concerning specific times and places; of “no one showed up at the party,” to propositions concerning specific domains of discourse; of “Peter is tall,” to propositions concerning specific tallness standards.

Kaplan (1989) famously articulated a distinction between character and content to properly account for indexicals such as “he”; character is the semantic property common to different uses of “he,” content its contribution to what is asserted in well-behaved particular uses. I (2006) have argued that additional examples such as those provided before should be accounted for by generalizing the idea: Semantics for natural languages is “character-semantics.” The semantics of a language as such does not identify the concrete speech acts that can be made with its sentences but merely constrains it; full determination of specific speech acts goes well beyond what is provided by the language as such. In fact, this point applies also to the identification of the specific type of speech act that is made. The declarative mood of the whole sentence conventionally indicates that a speech act in the saying family is made; but whether it is one of guessing, conjecturing, predicting, or a default one of asserting, this depends on contextual considerations of “saliency”; the same applies to the other conventional indicators of speech act type such as the interrogative or the imperative moods .⁷

Speech acts such as assertion are thus essentially normative, and the meanings of natural language lexical items consist of their contributions to speech act potentials; the fact that we are sensitive to these two points is what, on the proposal I am outlining, our intuitive acceptance of claims such as (plus) and (circular) manifests. Similar remarks apply to the cognition case but in that case distinctions such as that between character and content are largely irrelevant.

3 Weak and Strong Normativity

In the previous section, I have outlined the initial considerations pointing in the direction of a normative account of meaning . Meanings, be they linguistic or cognitive, divide up (at least in fundamental cases) into a force and a representational content. The fact that (mental or linguistic) “lexical” items make a given contribution to representational contents has implications—of the kind (plus) and (circular) illustrate—for acts involving them; and the very nature of those acts themselves is to be understood in normative terms, along the lines of KR, TR, or perhaps RBR for the case of assertion.

What the considerations by Mulligan and Glüer and Wikforss mentioned in the introduction primarily show, I think, is that this is still a weak sense of normativity , which does not coincide with the strong one we are after—the one on which norms prescribe actions or provide reasons for agents to act. There is an easy way to appreciate this. Let us consider again the three rules we saw in the previous section that have been proposed to explain assertion, in the framework posited by Williamson, KR, TR, and RBR. It is common ground among the parties to this debate that assertion is what is done by default (i.e., unless conditions in an open-ended list apply, such as those creating irony, fiction, or the presence of canceling parenthetical remarks such as “I conjecture”—which allows a conjecture to be made by the utterance of a declarative sentence, etc) by uttering declarative sentences: “In natural language, the default use of declarative sentences is to make assertions,” Williamson (1996/2000, p. 258) . This gives us an independent, causal-historical-intentional specification of the phenomenon that philosophers try to characterize: It is the act, whatever its proper characterization is, that is in fact associated with the indicative mood in natural languages as used on some occasions (the default ones), and which speakers intentionally purport to make by such means on such occasions. What is disputed is which of these norms an assertor is thereby subject to when uttering a declarative sentence in a default case. Of course, she may not be subject to any of them: After all, assertion might turn out not to be constitutively normative at all, as also indicated in the previous section, or might be characterized by a different norm altogether, or perhaps it only admits of a messy, disjunctive characterization appealing in part to some of those norms .⁸

Now, the obligations imposed by (TR) and (RBR), being constitutive of some act or other (even if it is not assertion, as causal-historical-intentionally picked out) “exist” in actuality (and in any other possible world, for that matter) as much as the obligation imposed by Williamson’s (KR). To avoid confusions, we may call “t-assertion” and “rb-assertion” the acts defined by (TR) and (RBR), respectively, so as not to prejudice the issue which one of the two, if either, is assertion, causal-intentionally characterized. Similarly, I will call henceforth “k-assertion” the one defined by (KR). In these terms, the debate confronting Williamson, Weiner, and Lackey is whether assertion is k-assertion, t-assertion, or rb-assertion, if it is any of them at all.

All these acts are normative in at least this weak sense: They are types constitutively defined by obligations or permissions. This is normativity only in a weak sense because assertors are in actuality, at most, subject to one of the obligations imposed by the three purported rules we have considered. The others are perhaps alternative speech acts , causal-historical-intentionally specified under a different label—say, the (less determinate) act of saying in the case of (TR)—or perhaps those practices are simply “not in place” in the actual world, in fact actually committing nobody. Thus, t-assertion, rb-assertion and also k-assertion are not strongly normative, in the following sense: Their (Platonic) existence does not, by itself, give any actual rational being a (normative) reason to act.

The point is usefully made relative to games, the model on which constitutive-norms accounts of assertion are based. All possible constitutive norms for games “exist” in all possible worlds, on the assumption that games are governed by constitutive norms defining them. But not all of them govern the particular transactions of a group of individuals . As in the case of assertion, it might even be an epistemic achievement of sorts to determine which rules apply to given cases—which one among several different games, governed by slightly different sets of rules, is the one in fact being played. There is a causal-historical-intentional sense in which players are playing one and the same game G, while it might be unclear what the constitutive rules defining G are. To illustrate: Imagine a group of people playing a given card game, causal-historical-intentionally identified. They call it by the same name, “Rummy”; they have sufficiently similar dispositions and expectations with respect to the rules applying to it, which nobody has in fact set out to articulate explicitly and precisely. A dispute then arises regarding a specific rule of such a game, which they all call “Melding.” Some players take the rule to be (BRM), others (HRM):

(BRM) If you have a valid group or sequence in your hand, you may lay one such combination face up on the table in front of you. You cannot meld more than one combination in a turn.

(HRM) If you have a valid group or sequence in your hand, you may lay one such combination face up on the table in front of you. You may lay as many such combinations in a turn as you want.

The description of the situation that the constitutive-rules account of games recommends goes as follows. There are two different games, which we may call BR-Rummy and HR-Rummy. If we use the term “Rummy,” the causal-historical-intentionally individuated game that our players are in fact engaged in, then the dispute concerns whether Rummy is BR-Rummy or rather HR-Rummy (or, of course, perhaps neither of them, perhaps it is in fact some other game or perhaps the matter is just indeterminate). The same applies to assertion, I think: Assuming that assertion is, like games, individuated by constitutive rules, it is unclear (up for grabs, in the philosophical arena) whether assertion is k-assertion, t-assertion or rb-assertion, something else, or whether the matter is simply indeterminate.

By themselves, in sum, kinds characterized by means of constitutive rules are not strongly normative. We may presuppose that the kinds defined by constitutive rules are Platonic entities, existing in all possible worlds. This is a reasonable presumption, useful at least for expository purposes.⁹ By way of analogy, Williamson (1996/2000, p. 239) mentions Lewis’s (1975) proposal (in the context of a not unrelated discussion, as it will transpire in the next section) to conceive in this way the languages linguists characterize in a mathematically sophisticated way. On this view, any such characterization specifies a language, which, on the Platonist presumptions we are adopting, should not be called a possible language, because it “exists” in all the worlds, including the actual one (Schiffer 1993) . But only some of those are actually used by a given population; there is a further issue about what makes it the case that a language is actually used by a population, an issue which might as well depend on contingent matters of fact.¹⁰ Similarly, among all possible games and, in general, all kinds defined by constitutive rules, only some of them are actually “in force,” giving actual people normative reasons to act—obligations and permissions.

By itself, thus, Williamson’s (KR), exactly like (TR) and (RBR), or the norms constitutive of different games are only weakly normative; only together with the assumption that k-assertion is what we in fact do when we assert (in the causal-historical-intentional sense) do they become strongly normative. By referring by “assertion” to the practice that the constitutive rule (KR) defines, we hide from ourselves the fact that, in the stronger sense, it is non-normative; this is because we pack into the characterization the fact that the practice so-defined is in fact implemented—it is the one people subject themselves to when they utter declarative sentences in default situations. But this is additional to the characterization provided by (KR).

Williamson is sensitive to the distinction I have made between the non-invidious Platonic existence of the types defined by constitutive norms, and their being in place, providing normative reasons to act to members of a given population, as his introductory clarifications make clear:

Given a game G, one can ask, “What are the rules of G?” Given an answer, one can ask the more ambitious question “What are noncircular necessary and sufficient conditions for a population to play a game with those rules?” Competent unphilosophical umpires know the answer to the former question but not to the latter. Given a language L, one can ask, “What are the rules of L?” Given an answer, one can ask, “What are noncircular necessary and sufficient conditions for a population to speak a language with those rules?” Given a speech act A, one can ask, “What are the rules of A?” Given an answer, one can ask, “What are noncircular necessary and sufficient conditions for a population to perform a speech act with those rules?” […] assertion is presented to us in the first instance as a speech act that we perform, whose rules are not obvious. In order to test the hypothesis that a given rule is a rule of assertion, we need some idea of the conditions for a population to perform a speech act with that rule; otherwise we could not tell whether we satisfy those conditions. […] Our task is like that of articulating for the first time the rules of a traditional game that we play. (Williamson, 1996/2000, pp. 239–40)

In the next section, I will outline a tentative answer to the more ambitious question of the two Williamson contemplates here, what makes it the case that a given population plays a game, speaks a language, or performs a speech act, in all three cases on the assumption of an answer to the first question such that those are abstract entities consisting of the constitutive rules.

4 Institutional Practices

Rawls (1955) distinguished between constitutive and regulative norms; his aim was to vindicate rule-utilitarianism as opposed to act-utilitarianism, by making this “logical” or conceptual distinction. Take the obligations ensuing from promises. If promising is a constitutively natural, nonnormative activity, the obligation of keeping promises can be understood as a regulative norm on utilitarian grounds, relative to the benefits it regularly confers.¹¹ Then it makes sense to consider, in the case of a particular valid promise, whether or not those benefits, and therefore the obligation, exist. Most of us find this intuitively wrong, as Rawls points out. This, he suggests, is because we think of the obligation of keeping promises as constitutive, definitional of the practice; if a valid promise has been made, then one must keep it.¹² This is a prima facie obligation, and hence could be overridden by other obligations with more force, including the utilitarian considerations; but even when it is, the prima facie promissory obligation still was in place.¹³

Even when the obligations related to promises are understood, as Rawls recommends, as constitutive or definitional, there is still a place for utilitarian considerations, he points out, now addressed to establish why the institution of promising, thus understood, should exist, be in place or implemented. Here Rawls is, I think, considering the same distinction I have made in the previous section, and pointing to the appropriate place at which, say, conventionalist claims may have a grip; later, in his main work (1972, pp. 344–350), he takes up the issue and gives a nonconventionalist account (unfortunately, called “conventionalist” in the literature on these matters, for reasons to be explained in a moment), as I will briefly explain now.

Consider again for the sake of comparison the case of games. Imagine that a group is playing the causal-historical-intentionally individuated game G but that there is an issue concerning which of the two possible sets of constitutive rules , Γ₁ and Γ₂, properly define G. As we have seen, the wide-scope obligations of which Γ₁ and Γ₂ consist “apply” to the situation, because they “exist” everywhere. This notwithstanding, as we have seen, in the strongly normative sense we have distinguished, at most one of those sets of obligations is in place in the situation; disputes frequently arise among players about such matters, as in the example of Rummy we provided before for the sake of illustration. What makes it the case that, say, Γ₁ but not Γ₂ is in place in the situation (so that perhaps Γ₂ is not at all in place in the actual world)? It is in answering questions like this (the second kind of issue that Williamson mentions in the quotation at the end of the preceding section) that serious debates between conventionalists and their opponents have a grip.

In the case of games, a conventionalist account seems very plausible to me, but how should we properly articulate it? Here, it is useful to go back to Lewis’ distinction between (abstract) languages and the language used by a given population. There is a way of thinking of abstract languages on which we do not have here just a useful analogy but one more example of the very same issue we are addressing: Alston’s (2000) view of abstract languages as speech-act potentials, on the assumption that speech acts are characterized in terms of norms . An abstract language can be thought of as a pairing of sentences and appropriate meanings; as sketched before, we can think of those meanings as a constraint on the force and propositional content expressed, with the help of context, in a concrete literal, serious utterance of the sentence. In the case of a declarative sentence such as “I am hungry,” the meaning assigned in English could be an indication of the rule to which assertors subject their act in default conditions, plus a Kaplanian character. What makes it the case that a language in which the sentence acquires this meaning is spoken by a given population, and hence that the constitutive rules partially defining it are in force in that population? Lewis’ (1975) well-known answer is that a convention of using the language exists in that population.

Conventions, according to Lewis’ (1969) account, are regularities that help to solve coordination problems, and are maintained because it is common knowledge that they have served this purpose in the past; they are arbitrary in that some other practice might have solved the coordination problem with a similar efficiency. Lewis’ is, for the case of conventions, a reductive, “regulative norm” account of the kind Rawls was criticizing for the case of promises, and I would reject it for Rawls’ sort of reasons; on the view I prefer, conventions should also be understood as non-reductively characterized by constitutive rules.

However, it is the existence of the kind of regularities that Lewis appeals to which is relevant to decide which practices defined by constitutive rules are in place, because, on the view I would like to suggest, this “being in place” of conventions supervenes on them. This is in fact the role Lewis (1975) assigns them in determining which among all possible languages—understood as abstract entities “existing” in all possible worlds—are in fact used or “exist” (in the other sense) at the actual world, as concrete tools of a given community. On the present proposal, conventions as much as languages themselves should be understood as defined by constitutive rules ; Lewis’ account explains which of the many conventions thus understood along Platonic lines, (DL) and (DR) among them, are actually in place, solving coordination problems in actual populations.

When it comes to other kinds defined by constitutive rules, like games, a conventionalist account contends on the present proposal that they exist if and only if a corresponding Lewisian convention implementing it exists. It is plausible, I think, to appeal to this kind of regularity to determine which set of constitutive rules, if any, defines G (without excluding, of course, the possibility that the issue is indeterminate, as I am sure it is in many actual cases). In accordance with Lewis’ explanation of conventions, we should examine the dispositions players of G have to act and think (including in particular, as Alston (2000, pp. 265–8) insists, their normatively more relevant dispositions, such as to feel resentment or guilt, to apologize and criticize, and so on),¹⁴ to settle the matter whether they conform their practice to the system of rules Γ₁ or rather Γ₂. On this proposal, a kind defined by constitutive rules is conventional if its existence (in the non-Platonic, invidious sense) is in fact the existence (in that very sense) of a Lewisian convention.

The conventionalist proposal about games, suggested here, is not simply the trivial one that certain means signaling that one is playing a given game (the figures on the cards that are used, the expressions players use in given moves) are conventional, in Lewis’ sense; indeed they are, in the sense just explained: They are Lewisian conventions in fact implementing constitutive-rules conventions specifying the normative commitments incurred in using the relevant expressions. But the envisaged conventionalist claim goes beyond this; it is rather that the very fact that the constitutive rules of a given game actually apply in a given situation, their “being in place,” is explained on the basis of what in fact constitutes the existence of a convention.

Rawls (1972, pp. 344–350) , following Hume , is said to provide a “conventionalist” account of the promises, which other writers such as Scanlon (1998, 2003) and Shiffrin (2008) reject. However, Rawls’ account is, I think, only “conventionalist” in the weaker sense that assumes the existence of a conventional practice of using certain expressions to indicate that a promise is being made—say, Lewisian (1975) conventions of truthfulness and trust in English regarding utterances of S, I promise, or I hereby promise that S in default situations .¹⁵ In the case of promises, this almost non-contentious form of conventionalism is not questioned, say, by Scanlon or Shiffrin, who explicitly accept the existence of conventions of promising (better put, conventions of expressing promises) thus understood; but this is merely a proper part of the stronger form of conventionalism illustrated with the case of games. In that stronger sense, Rawls’ (and Hume’s) account is not, I think, conventionalist.

The reason is this. Lewis (1969) distinguishes from conventions what he calls social contracts. These are also regularities serving coordination problems kept in place because their previous existence is common knowledge , but they differ from conventions mainly in that, while in the latter case agents do not have any motivation for free riding, they do in the case of social contracts. A related additional difference is that it is unclear whether they are arbitrary—whether there is another equally serviceable practice to solve the relevant coordination problem. In the case of social contracts, the explanation of conformity, and therefore of the preservation of the regularity, requires more than mere awareness of the existence of the practice; it requires sensitivity to moral norms , or other forms of commitments. Now, while driving on the right is a convention in Lewis’ sense, it seems to me that promising and asserting are rather social contracts. Correspondingly, the explanation that Rawls (1972, pp. 344–50) provides for the “being in place” of the institution of promising is fundamentally moral; he appeals to sensitivity to a principle of fairness, requiring us to play our specified parts in social institutions which are fair and from which we benefit.¹⁶

I thus propose to classify views on these matters along the following lines. Promises and assertions are subject to norms but we have, firstly, the divide between what we might call naturalists and institutionalists accounts of this fact. Naturalists (Hindriks (2007) for assertion, and, I guess, Scanlon (1998) and Shiffrin (2008) for promises) think that these acts are not constitutively normative; they might be characterized in terms of Gricean reflexive intentions. Although we do have conventional practices indicating that they are performed, they can occur in their absence; both when they are made by relying on conventions, and in other cases, they are subject to general moral rules , perhaps duties of sincerity (Hindriks), duties to meet created expectations (Scanlon), duties to uphold intimate relationships (Shiffrin), or to promote “authority interest” (Owens). These are the sources of the more distinctive regulative norms applying to promises and assertions .

Institutionalists reject these views, mostly on the basis of variations on Rawls’ argument against the utilitarian regulative norm account: The obligations resulting from a valid assertion or a valid promise still exist in cases in which, if we just took into consideration the moral benefits mentioned in naturalists accounts, they would not. They might be overridden by those moral considerations, so that agents do not have an all-things-considered obligation, but they still have the relevant prima facie obligation, as indicated by our intuitions regarding feelings of resentment or guilt or needs for excuses, apologies, compensation, and so on, which people do experience (Gilbert 2008, pp. 223–34) . According to institutionalists, promises and assertions are defined by specific constitutive norms, norms that exist, are in place or have been adopted and therefore are strongly normative, in fact prima facie obliging actual people in actual situations.

This “being in place” consists in the existence of regularities, in their turn accounted for by specific dispositions to think and act in rational beings. But only some institutionalists are true conventionalist. The issue turns, I have suggested, on whether the regularities in question constitute a Lewisian convention, as opposed to a social contract; and hence, on whether or not we need to appeal to what is in fact sensitivity to moral norms (even if ultimately consisting themselves of norms of prudence or rationality, depending on which metaphysics of moral norms is correct) to explain their preservation. It could also be that the correct account combines elements from conventionalism and anti-conventionalism (both of them of the institutionalist variety): Promises and assertions described in an abstract, sufficiently general way are not conventional but when we go into the specifics of the particular practices that a given community has implemented we do find conventional features. I myself would endorse an institutionalist nonconventionalist account of assertion, promises, and (paradoxically as it might sound) conventions and agreements themselves, but this is not the place to examine the issue in any depth.

5 Conclusion

Let us take stock. Meanings, we have assumed, both linguistic and cognitive, consist in the fundamental cases of judgments , assertions, intentions, or enjoinments, individuated by a force and a representational content. The fact that (mental or linguistic) “lexical” items make a given contribution to representational contents, and in general the “truth-conditions” determined by those contents, has nothing normative in itself but it has implications—of the kind (Plus) and (Circ) illustrate—for acts involving them, whose very nature is to be understood in normative terms, along the lines of KR, TR, or perhaps RBR for the case of assertion, and corresponding norms for the case of argumentative transitions such as inductive or deductive inferences among belief -like states, or transitions from experiences to beliefs, etc. Again, the fact that the constitutive nature of those forces is normative does not by itself have strongly normative implications: It does not suffice to create reasons for agents to act in specific ways. However, together with whatever psychological or sociological facts determine that a given norm is in place, they do provide such reasons.

Let us then see how this answers the Glüer and Wikforss concern . Let us suppose that some acts of judging or asserting (“applying” a—perhaps mental—“term”) are constitutive of a “term” having a certain meaning , for whatever reason.¹⁷ In the case of the logical constants, such as the conditional, Carroll’s (1895) famous paradox establishes that it cannot be judgments that a certain proposition follows from certain others, but inferential transitions relating them which are thus constitutive. It has not been sufficiently explored which norms could characterize such acts; for the sake of having something illustrative in mind, we can think of one corresponding to TR—a norm of conditional truth: One must ((infer p from the set ∑) only if (p is true on condition that all q in ∑ are also true)). Let us further assume that the relevant psychological or sociological facts establish that these norms are in place among us: TR for assertions/judgments, the norm just mentioned for deductive inferential transitions.

Now, are those assumptions compatible with the intuitively correct point that “oughts not only imply cans, they also imply the possibility of violation, of…‘forbidden combinations’”? Glüer and Wikforss are right, I think, that if applying “circular” to circular items in paradigm cases is determined as correct by the meaning of the term, and the corresponding point concerning the meaning of “if…then _,” the cases we can think of in which these “norms” are “violated” do not count as proper “forbidden combinations”; they should have to be accounted for by facts such as lack of attention , excessive complexity, etc. However, the relevant norms we have been contemplating are norms for the representational acts (judgments, assertions, or inferential transitions) in general, not just in these particular cases linked to the meaning of some terms. There could then be many other cases allowing for proper violations: Cases that are not in any way constitutive of any of the occurring terms. And I think this is quite enough to satisfy the intuition that “forbidden combinations” should be allowed. Let us go on with the assumption that it is TR that is “in place” for our assertions. Should this intuitively require that wrongly asserting that 2 + 1 does not equal 3, say, is a proper “forbidden combination”? My own intuitions at least do not require that much. It requires that there are assertions involving those terms that do constitute forbidden combinations; but, of course, there are plenty of them.

This reply assumes that some explanation of which rules are “in place” along the lines of the one in terms of Lewisian conventions or social contracts suggested in the previous section can be properly developed: In effect, it assumes that there is an account of when rules properly “guide” us, and not merely characterize regularities we follow. Such an account should be able to circumvent another objection that Glüer and Wikforss pose (2009a, p. 52 ff.) , the “dilemma of regress or idleness.” We have to distinguish genuinely following rules from merely conforming to regularities; intuitively, the former requires some kind of mental attitude of acceptance by the subjects involved, i.e., the psychological states that the Lewisian account contemplates; but the considerations motivating normative accounts of foundational matters extend to these psychological states. Does not this lead to a non-virtuous regress? It certainly points to the need to encompass a sensitivity to norms which is natural to characterize with the Wittgensteinian metaphor as “blind,” as the best recent work on the rule-following considerations has shown (cf. Boghossian 2008; Wright 2007) . The problem is how to make articulated sense of this while still making room for genuine rule-following as opposed to mere rule-conforming. I cannot do any justice to these issues here .¹⁸