I. Ignorance: once again

Let us begin with a usual (yet perplexing) line of argument.

P1. I don’t know that I am not a (handless) brain in a vat (BV)-even if I believe it.

P2. If I don’t know that I am not a (handless) brain in a vat, then I don’t know that I have hands (Hands).

C. So, I don’t know that I have hands.

We have a paradox. We start with seemingly true premises, use a valid logical form and arrive at what seems to be a false conclusion. Something has to give. As noted by ^{DeRose (1999, p. 184)}, whenever we deal with a paradox we confront an easy and a not so easy task. The easy part involves choosing some proposition as false or somehow inadequate. The not so easy one involves explaining to ourselves how we came to believe something false or inadequate in the first place.

Suppose we want to give a linguistic analysis of the paradox. As repeatedly noted, we could adopt something akin to a Moorean dogmatism and deny the conclusion. Whenever I do have hands, C’s negation always, or nearly always, expresses a true proposition. In addition, P1 always, or nearly always, expresses a false proposition: I do know that I am not a brain in a vat. But how could I know such a thing? After all, if I were a BV, I wouldn’t believe it. So my belief that I am not a BV seems totally unjustified. Well, I know that I have hands. And I know that if I have hands I am not a (handless) BV. So, by employing what seems to be a reasonable inference principle I know that I am not a (handless) BV.

Epistemic Closure Principle (Closure, for short):

I know that p.

I know that p implies q.

___________________________

I know that q.²

Note that this line of argument constitutes something of a paradox on its own. How could we gain such extraordinary knowledge simply by reassuring ourselves that we have hands?³ Moreover, given that under Mooreanism P1 is always false, we still lack an adequate response to the not so easy part of the paradox.⁴ Given these problems, one could follow ^{Dretske (2005)} and use these considerations to encourage a denial of Closure.

But Closure sounds plausible. Moreover, a denial of Closure commits us to accept what DeRose calls abominable conjunctions, like “I know that I have hands, but I don’t know that I am not a (handless) BV” (1999 p. 201). This sounds plainly wrong. Furthermore, consider the following argument:

P3. I have hands.

P4. If I have hands, then I am not a (handless) BV.

C. So, I am not a (handless) BV.

^{Veber (2013 pp. 1186-1187)} observed that this argument doesn’t rely on Closure. It’s just a modus ponens, and ^{Dretske (2005 p. 18)} accepts it as sound. But on Dretske’s view, though I know that the premises are true and I know that the argument is valid, I don’t know that the conclusion is true. This gives place to the following abomination: “I know that if an argument is valid and its premises are true then its conclusion must be true as well. And I know that this argument is valid and that its premises are true. But I don’t know that its conclusion is true” (^{Veber, 2013, p. 1187}). So denying Closure doesn’t look as a good idea after all. One could, of course, capitulate to skepticism. However, if possible, we should try to avoid that path.

In this scenario, contextualism appears.⁵ Suppose the truth values of KAS vary with the context of utterance. More specifically, suppose this is so because the content of a KAS varies according to the epistemic standard present at its context of use. Then we could accept Closure, explain the plausibility of AI, and retain our everyday knowledge anyway. But how? Well, if contextualism is true, then the fact that AI’s sentences may in some contexts express true propositions doesn’t preclude those same sentences from expressing false propositions in others. So the soundness of the skeptical argument in very demanding contexts need not force us to accept skepticism. After all, in non-skeptical contexts C’s negation may be perfectly true, given the non-skeptical or low standards there prevailing. This gives a promising (dis)solution of the paradox. It allows us to meet the easy part of the challenge by accepting C’s truth but, at the same time, it gives us a way to deny skepticism: C’s truth is non-toxic because it is compatible with its falsity in non-skeptical contexts. Moreover, we may fulfill the not so easy task by pointing to our own ignorance of contextualist’s semantics. Ignoring, as we did, the context sensitivity of KAS we were taken to deny C’s conclusion, wrongly believing its acceptance to be fatal. But it isn’t. We get to deny skepticism, accept Closure, and explain the paradox.

This is a nice dissolution of the paradox. Or it would be if contextualism were true. However, problems remain. Crucially, the (dis)solution requires subscribing to an error theory. Speakers are said to be mistaken about KAS’s semantics. But, as ^{Schiffer (1996)} famously pointed out, contextualists are committed to a particularly contentious kind of semantic mistake which undermines the hypothesis credibility. The problem is not to postulate semantic ignorance or confusion-not even to postulate some kind of mistake caused by them. That’s perfectly natural and presumably any linguistic dissolution of the paradox would need to take a step in that direction. The problem lies with the particular type of mistake postulated by contextualists.

We should follow Greenough and Kindermann and distinguish between semantic error and semantic ignorance. I may ignore the meaning of some word without making any mistake about it-I may simply refrain from using it. We can contrast this to what happens in malapropisms. So “semantic error involves more than mere ignorance; it typically involves some false (implicit) belief about some semantic property of an expression. Such a false (implicit) belief will typically be made manifest by some misuse of the expression in question” (^{Greenough & Kindermann, 2017, p. 306}). Contextualists claim that we wrongly take AI’s conclusion to be false (and wrongly believe AI to threaten our everyday knowledge) because we ignore the indexical nature of KAS. So, according to Schiffer, “[AI] strikes us as presenting a profound paradox merely because we’re ignorant of what it’s really saying, and this because we don’t appreciate the indexical nature of knowledge sentences” (1996, p. 325).

In turn, ignorance leads to error. Ignoring KAS’s indexicality, we tend to give them an invariant content across contexts. So we find ourselves persuaded by an argument whose conclusion we are unwilling to accept. But it proves difficult to reject some of its premises either. This painful situation should improve once we learn about contextualism. Semantic error is a crucial aspect of the story.

However, as ^{Schiffer (1996)} pointed out, the variety of error attribution posited by contextualists is problematic. Here’s Greenough and Kindermann exposition of the argument:

This form of the objection assumes something like this:

Content Articulation Thesis: Semantic competence with regard to an indexical sentence S requires that speakers be able to specify what is said by S (in some context c) by articulating a natural language sentence that literally expresses the proposition originally asserted (in c) (^{Greenough & Kindermann, 2017, p. 311}).

But even if one is suspicious about the thesis, we may modify Schiffer’s objection in the form of a general challenge to indexical contextualism.

[…] if “Knows that” is context-sensitive in the way gradable adjectives or “hidden indexicals” are (substitute a given Contextualist’s favorite model of context-sensitivity), then how come speakers are much better at clarifying what they mean using gradable adjectives etc. than they are with knowledge ascriptions? How come they have prepositional phrases such as “in London” for the articulation of location or “for a basketball player” for the articulation of a comparison class to a use of “tall” more readily at hand than they have any phrases articulating epistemic standards? (^{Greenough & Kindermann, 2017, p. 312}).⁶

II. Austin’s forgotten insight

Perhaps contextualism is not a good idea after all. However, the general contextualist strategy of dissolving the paradox remains attractive. Can we still defend a similar solution without embracing contextualism? As ^{Stroud (2008)} noted, the origins of the contextualist position go back to Austin. In Other Minds he makes a careful consideration and description of our everyday knowledge ascription practice. Stroud rightly detects the anti-skeptical potential of this procedure. Here is a reminder of his, rather extended, dialectic:

The best grounds for accepting Contextualism come from how knowledge-attributing […] sentences are used in ordinary, non-philosophical talk: What ordinary speakers will count as ‘knowledge’ in some non-philosophical contexts they will deny is such in others. As J. L. Austin observed, in many ordinary settings we are easy, and say things such as (Austin’s example), ‘I know he is in, because his hat is in the hall.’ But, even with no philosophers in sight, at other times speakers get tough and will not claim to know that the owner was present based on the same evidence; as Austin notes, ‘The presence of the hat, which would serve as proof of the owner’s presence in many circumstances, could only through laxity be adduced as a proof in a court of law’ [1976, p. 108]. We needn’t invoke anything as unusual as a high-stakes court case to find such variation-as I’m sure Austin realized. A wide variety of different standards for knowledge are actually used in different ordinary contexts (^{DeRose, 2011, p. 47}).

Stroud stresses the anti-skeptical potential of this (roughly) Austinian approach to KAS. However, he finds it wanting because:

(E) Austin never talked of the truth conditions of KAS limiting himself to an analysis of their conditions of assertability. ^{Stroud (2008, p. 56)} cites various passages from ^{Austin (1976)} that confirm his remark. More recently, Keith DeRose also laments Austin’s omission in this regard. As we saw, he gives Austin credit for paying attention to the existence of both low-standards and high-standards conversations in ordinary talk. However, he explicitly (and, in my view, correctly) refuses to call Austin a contextualist.

[…] though Austin is something of a granddaddy of Contextualism, in that he put on display some of the linguistic phenomena contextualists appeal to, he himself should not be classified as a contextualist. All grant that in some way our use of ‘knows’ is governed by different standards in different contexts. The question is: Are the varying standards we can discern those that govern the truth-conditions of the sentences, or merely their warranted assertability conditions? Insofar as I can discern a clear answer to this in Austin, it is (a peculiarly aggressive), ‘I don’t care’ (^{DeRose, 2011, p. 48}).

All in all, DeRose is simply puzzled about Austin’s peculiar omission: “The real controversy, then, is over a crucial issue about which Austin seems (unwisely, I believe) quite unconcerned” (2011, p. 48). Lawlor also notes Austin’s omission:

Austin does not provide an account of what it is for a knowledge claim to be true […]. In asking about the truth conditions of knowledge claims, I might seem here to be trespassing against the Austinian theory of meaning as use (2014, pp. 54-55).

As long as it is even intelligible to suppose that there is a logical gap between the fulfilment of the conditions for appropriately making and assessing assertions of knowledge on the one hand, and the fulfilment of the conditions for the truth of those assertions on the other, evidence from usage or from our practice will not establish a conclusion about the conditions of knowledge (2008, p. 64)

If, on behalf of Austin, I had to object some step in Stroud’s reasoning, I would point to (F). Austin explicitly rejects the idea that KAS have truth conditions. So what Stroud, DeRose and Lawlor take to be obvious, Austin takes to be obviously false.

To suppose that ‘I know’ is a descriptive phrase, is only one example of the descriptive fallacy, so common in philosophy. Even if some language is now purely descriptive, language was not in origin so, and much of it is still not so. Utterance of obvious ritual phrases, in the appropriate circumstances, is not describing the action we are doing, but doing it (1976, p. 103).

This is a classical topic of Austinian analysis, much repeated in textbooks and introductory classes. But unfortunately philosophers tend to forget this fact. Moreover, according to Austin, some parts of our language appear superficially descriptive when, in reality, they just serve a performative function. Here Austin explicitly endorses such a position regarding KAS.

The alleged factivity of “knows” is nowadays seen as axiomatic. “I know that p, but p is false” is standardly considered an outright contradiction. The same goes with the presumed descriptive character of KAS: whatever their correct semantic/pragmatic treatment turns out to be, it should consist in some set of truth conditions. It should give us a list of the (perhaps contextual) conditions under which a KAS is true or false. But, clearly, Austin would reject all of this as ill conceived. However, somehow paradoxically, he may also be seen as the precursor of Contextualism. Of course, there is a deep and important methodological similarity between Austin and contemporary contextualists: they both take the detailed description of the actual use of KAS as philosophically crucial. But there’s a big and crucial difference we shouldn’t forget: Austin explicitly rejected the assumption that KAS’s main linguistic function is descriptive. He thought that there lied the cause of our predicament (so, contrary to DeRose’s comment, in my view Austin did care about the truth conditions/assertability conditions divide). I propose to take his words at face value and see where they lead.

A telling case in this story is Lawlor’s position on KAS. In Assurance, she develops what she takes to be an Austinian account of KAS. At least in its speech act aspect, her work is similar to mine -and to my previous (2014b) account. However, surprisingly, Lawlor is a contextualist (2014, p. 61). Being an Austinian, it could be expected that she’d pointed to (F) as the main suspect in Stroud’s anti -Austinian reasoning. But she chooses a more standard (and perfectly respectable) move instead. With DeRose, she notes that we should: (H’) provide what’s missing, that is, a contextualist, truth-conditional semantics for KAS. This in turn should deliver some kind of contextualist treatment of the paradox that effectively disarms it by controlling its damage. In fact, ^{Lawlor (2014, pp. 122-128)} points to the alleged falsity of unrestricted closure. Prima facie this is a negative aspect of her semantics. She is committed to accept abominable conjunctions and all the other abominable results explored in ^{Veber (2013)}-a result avoided by my own AA.

Let’s take stock. It appears as if Stroud failed to engage with Austin’s more radical and characteristic insight: KAS are not descriptive. They lack truth conditions. Knowledge is not factive. And the philosophical distortion that’s causing all the trouble is precisely the factualist assumption to the contrary. Of course, this is not to say that Stroud just failed to note Austin’s position. Rather, he is trying to avoid what he takes to be an obviously false assumption. And so he chooses to ignore it. Stroud’s attitude was since then the usual one. This paper aims to change this situation by showing how we can articulate an Austinian account of KAS that enjoys all the benefits of the contextualist (dis)solution of the skeptical paradox but none of its problems.

III. Giving one’s word that p

What do I do when I say “I know that such and such”? At first sight we may suppose that I am simply asserting that I know that such and such, for instance, that I know that the book is on the shelf. If we assume that I do know that p (that the proposition is true), then I am simply constating that I am in a certain kind of special and peculiar state-that I have a peculiar relation with the proposition that p. This state of mine is supposed to be both stronger and of an altogether different class from the mere state of believing that p. Moreover, there are some conditions jointly sufficient and individually necessary that must be satisfied for my assertion’s content to be true. At the very least, my belief has to be true. And, to use a now common metaphor, the strength of my epistemic position with respect to p has to be peculiarly strong-if we wish, we could give some substance to this by way of a relevant alternatives theory.

All of this I take to be standard. However, according to Austin, “I know” has a performative rather than a descriptive use, i.e., in saying “I know that the book is on the shelf” I am performing, or attempting to perform, some particular and peculiar kind of action instead of merely describing some particular, and peculiar, cognitive state of mine. In this regard, according to Austin, “I know that p” would be more similar to “I promise you that p” than to “I assert that p.”

Austin contrasts what one does by saying ‘S is P’ with what one does by saying ‘I promise that S is P’. He then proposes to treat ‘I know’ as analogous to ‘I promise’.

But now, when I say ‘I promise’, a new plunge is taken: I have not merely announced my intention, but, by using this formula (performing this ritual), I have bound myself to others, and staked my reputation, in a new way. Similarly, saying ‘I know’ is taken a new plunge. But it is not saying ‘I have performed a specially striking feat of cognition, superior, in the same scale as believing and being sure, even to being merely quite sure’: for there is nothing in that scale superior to being quite sure. Just as promising is not something superior, in the same scale as hoping and intending, even to merely fully intending: for there is nothing in that scale superior to fully intending. When I say ‘I know’, I give others my word: I give others my authority for saying that ‘S is P’ (1976, p. 67).

I want to stress an obvious and important aspect of this account. Once we drop the usual assumption that KAS are descriptive, we should stop thinking in terms of truth-conditions. We are now thinking in terms of actions and, as ^{Kent Bach (2011)} stresses, actions are either successful or unsuccessful . By saying ‘I know that p’ I pretend to give you my word for the truth of p. I pretend to assure you that p. Let’s follow Lawlor’s terminology and speak of an assurance to refer to the kind of act one aims to perform by saying ‘I know that p’. We may then provisionally say that an assurance is successful if the addressee takes the speaker’s word with regard to p. From that moment on she takes p for granted in her endeavors.⁸

This is Austin’s insight. How far could we go with it? At this point, we face two obvious problems: (a) as later argued by Austin (1962), the distinction between constating and performing is false: assertion is no less an action than promising or advising; (b) Austin’s proposal may be said to ignore an obvious semantic condition of KAS: knowledge is factive. Of course, there’s a rapid response available: if KAS lack truth conditions, that’s merely an appearance (after all, they lack truth conditions). But we should expect more. Even granting, for the sake of the argument, that KAS lack truth conditions, we would like to know what explains this appearance. I attend to these concerns in sections IV and V respectively.

IV. An Austinian account of knowledge ascriptions⁹

According to Austin, in uttering “I know that p” I am attempting to perform an action. I am not merely trying to describe some peculiar and special cognitive state of mine. Moreover, if successful, the action in question is similar to a promise; following Lawlor’s terminology, we called it an assurance. The first problem faced by this proto-Austinian account is that it rests on a false distinction, that is, the one between describing and doing. As Austin himself later noticed (1971), stating and asserting are, of course, actions in themselves. If we’d like to develop an AA, we should work with Austin’s mature speech act theory and distinguish between the locutionary/illocutionary/perlocutionary acts one may perform in using a KAS.¹⁰

We may discern a stratification of speech acts. By saying:

Anna may, at once, successfully produce an English grammatical sentence, assert that the light is turning red, and frighten John. We would traditionally label each of these acts as locutionary, illocutionary and perlocutionary. Crucially, in order to successfully perform a perlocutionary act one has to successfully perform both a locutionary and an illocutionary act-in general, any given superior act performance presupposes the performance of all the lower level acts. Speech acts are stratified (cfr. ^{Alston, 2000, p. 18}).

I will follow Alston’s terminology and speak of the content of an act to refer to anything that (i) the speaker wants to communicate and that (ii) the hearer must grasp in order to understand what the speaker is saying. In this sense, content “includes illocutionary force as well as propositional content” (2000, p. 15). We may make explicit the content of Anna’s utterance by way of some, but not all, oratio obliqua reports. Let’s call those reports that manage to convey some utterance’s content “proper.” By uttering L, Anna:

Reports (i) and (ii) are improper because they deal exclusively with acoustic features of the utterance. Reports (v)-(vi) have to do with the consequences of John’s utterance and not with its content. Reports (iii) and (iv) are proper reports.

We should not mistake content used in this sense with propositional content (or proposition). We should distinguish between the illocutionary act’s propositional content and its illocutionary force. An utterance’s content is more than its mere propositional content. Anna’s utterance’s propositional content is that the light was turning red. This same proposition may have been attached to a different illocutionary force, thus generating a different content.

So, what’s the content of a KAS? Consider an utterance of (1) by John and the corresponding report candidates (2)-(5):

Unfortunately, we can’t simply rely on the oratio obliqua test to find the content of (1). We should not be surprised about this situation. After all, if Austin is right, KAS may superficially appear descriptive without being so. We should not expect the speaker’s intuitions to settle the matter. Moreover, the procedure does not pretend to offer a completely precise description of content. Communicative intentions vary as circumstances vary. However, we may discard some options. (4) is uninformative. It does not help us find (1)’s precise illocutionary force. So I take (3) and (5) as clues to develop an AA. I put (2) aside simply because (at least if one reads it in a semi-technical sense) it doesn’t respect Austin’s forgotten insight.

In what follows it will prove helpful to distinguish between speech act semantics and pragmatics. “Force” may be used in two different ways. In a pragmatic sense, it involves the actual way in which some propositional content is used in some particular occasion. In this sense, force determines the kind of linguistic act actually performed, in some particular circumstances, by uttering some sentence token (is it an assertion? A warning? A threat?). Semantically, we can take force to involve the potential ways in which some propositional content may be interpreted, abstracting from its use in certain particular circumstances. At this level linguistic types have potential conventional force attached to them. In developing the Austinian account, we would mainly be working at this semantic level-which we shouldn’t equate with the truth-conditional, propositional content level.

However, to help us get there we started by asking a pragmatic question. What do we do when we utter a particular KAS? This question is itself problematic. One may utter a particular KAS in a particular context of utterance with particular, even idiosyncratic aims. If, ignoring everything besides the utterance, we hear someone saying “I know that you did it” what should we believe she was doing? She could be making a report, or a warning-she could be regretting, thanking, or threatening her addressee, or whatnot.

This is a general phenomenon.¹¹ Suppose Mary utters:

In so doing, she could, of course, be asserting that someone should be here by now. But she could also be commanding someone else to get him, or assigning blame for his lateness, or threatening, or lodging a formal complaint, and so on. So it is natural to ask: “What makes it the case that an utterance constitutes an illocutionary act of a given kind?” (^{Fogal et al., 2018, p. 1}).

Austin himself may be said to give a conventionalist response: in order to successfully perform an illocutionary act, one has to follow a conventional procedure “whose performance is a matter of behaving in accordance with a collection of ‘felicity conditions’” (2018, p. 2). The violation of these conditions may preclude the realization of the act (as when someone who lacks the appropriate social status attempts to marry a couple, or declare war). Conventionalism starts with some highly ritualized procedures involving language (as in christening a ship or declaring war), and then extends to other, more implicit and not highly ritualized linguistic acts that are said to involve analogous (though implicit)-linguistic-conventions.

Prima facie, the postulation of conventions is attractive. If they are in place, we see how we manage to efficiently communicate with others by means of language. We may do all sorts of things by uttering L, but surely asserting that the light is turning red is an immediate and obvious candidate. To quote Lepore and Stone: “If we find a coordination problem that people reliably solve, then we have good reason to think that they solve it through some kind of convention […]” (2016, p. 96).¹²

However, this view struggles with semantic under-determination phenomena. In some contexts we will naturally judge that something is missing if the listener takes (7) as a (macabre) question, or if she takes (6) merely as an assertion.

Of course, at some level communication may be said to succeed. After all, it is not altogether wrong to resort to the macabre interpretation in (7). However, some sort of mutual understanding is clearly missing. One may wish to narrow the meaning of “illocutionary force” in such a way that the performance of an illocutionary act is always strictly connected to rules and conventions, and treat all other effects as perlocutionary ones.¹³ But this possibility is neither compulsory nor very natural. We should recognize that it is not at all uncommon to find cases in which the semantic conventions one may naturally attribute to sentence types underdetermines the content the speaker wishes to convey, in some particular context, by uttering a token of that semantic type. Conventionalism will be hard pressed to explain this data. So it is natural to turn to speaker’s intentions in order to answer our pragmatic question.

According to intentionalism, a view first promoted by Grice, one succeeds in performing an illocutionary act if one manages to successfully communicate one’s communicative intentions to one’s addressee. I can’t run a marathon simply by publicly announcing my intention to do so. However, it seems natural to say that I may communicate my intention to run a marathon simply by publicly announcing my intention to do so.¹⁴ Thus, it is natural to think that in the case of illocutionary acts their success consist in the speaker’s recognition of communicative intentions. The precise nature of communicative intentions is a matter of debate. However, the main idea is simple: “[…] performing a communicative act is a matter of producing an utterance intending both (a) for one’s addressee to have a specified response, and (b) for one’s addressee to recognize that this response is intended” (^{Fogal et al., 2018, p. 4}). If I give my addressee an order, let’s say to give me a beer, I expect her (a) to believe that I want her to bring me a beer, and (b) to realize that I intend her to recognize my intention. Unfortunately, illocutionary success is not the same as perlocutionary success. She may perfectly understand my order and refuse to comply with it. This is a nice result of intentionalism: it sheds light on the Austinian distinction between the illocutionary and the perlocutionary level.¹⁵

Attractive as it is, intentionalism has limitations. One can worry that the view promotes some kind of Humpty Dumptyism (cfr. ^{Fogal et al., 2018, p. 5}). After all, any utterance whatsoever may be used in highly idiosyncratic ways, perhaps motivated by some grotesque linguistic error made by the speaker. Suppose that I wrongly believe that the imperative “Bring me a beer!” is generally useful for asking whether there is any beer on the fridge. Let’s further assume that, at least in this particular occasion, my audience is (somehow mysteriously) capable of understanding my intentions. They reply: “We don’t have any.” Intentionalism should count this as a case of successful communication. One could resist this conclusion. Mind reading is not linguistic communication. Perhaps we could identify it with communication in general, but something seems to be missing, namely: the conventional nature of the activity. However, intentionalists could bite the bullet. After all, I did manage to communicate with my audience. It appears as if I did manage to ask for some information. So one could argue that my illocutionary act token is an interrogative, not an order.

Suppose we accept the intentionalist stance in this regard. We find it compelling and so we are willing to accept this somewhat unintuitive results. Still, we should not forget that our question is not the only important question to ask. Firstly, we are dealing with an (a) ontological question: what makes it the case that some utterance counts as an illocutionary act of a given kind? But we should also ask an (b) epistemological question, namely: how do hearers recognize that a particular utterance counts as a particular kind of illocution? (cfr. ^{Harris, 2016 p. 176}; ^{Dummett, 2003, p. 201}). Moreover: we are dealing with a (c) pragmatical question (what makes it the case that a particular utterance constitutes an illocutionary act of a given kind?) as opposed to a (d) semantical one: what, if any, illocutionary properties attach to sentence types in virtue of their form? We focused on the particular illocutionary act one may perform by uttering some particular sentence in some particular context. But we may also be interested in the properties that govern utterance types, regardless of the particular context in which they are used. In our imagined case, I somehow managed to ask a question, if only by some idiosyncratic and strange procedure. But now imagine a community of brilliant mind readers. Either they don’t have or they don’t need any code for communication. They rely exclusively on their mind reading powers for interpretation. We are not such a community. Although we may adopt an intentionalist stance regarding (a) and (c), we apparently shouldn’t leave conventions entirely out of the picture if we wish to respond to (b) and (d).

Of course, we could still be intentionalists about (a) and (c). It seems that we need both intentions and conventions in explaining the data. This may seem like a trivial observation. If it is, then, perhaps, we are on the right track. The first step in this direction involves the postulation of some kind of conventional force that attaches to sentence types and helps us hearers determine the content of particular utterances in particular situations. But any such proposal has to deal with the challenge of offering a distinction between the standard/non-standard uses of a type. ^{Davidson (1991, pp. 110-111)} famously challenged us to offer a distinction between standard/non-standard uses of the declarative.

I wish to follow this line of thought and inquire about the conventional and semantical dimension of force. We may start by noticing the difference between sentences 8 to 10:¹⁶

8 is a declarative sentence, 9 is an interrogative sentence, 10 is an imperative sentence (cfr. ^{Huddleston & Pullum, 2002, p. 853}).¹⁷ The grammatical difference, concerning word order and the inflexion of the verb, may be said to involve a meaning difference between the sentences. We could say that sentences 8-10 have different moods. What about force?

According to Dummett, a theory of meaning includes not only a theory of reference and sense, but also a theory of force (2004, p. 40). A theory of force is closely related to a theory of mood, because mood is, according to Dummett, a prima facie sign of force (^{van der Schaar, 2007, p. 62}).

This view is attractive if one wishes to answer our epistemological question (b) above. Even assuming, as we are, that intentionalism is correct at an ontological level, we still need to explain how we routinely manage to assign forces to utterances, with great efficiency, given that we lack any supernatural mind reading faculty. Contrast the case where I, somehow miraculously, managed to ask a question regarding our beer supply by uttering an imperative sentence with the more mundane one in which I simply utter an interrogative sentence. By so doing, I greatly increased my chance of success. This suggests that conventions are part of the story.

However, any such postulation needs to consider Davidson’s challenge. Here we may follow van der Schaar’s suggestion and distinguish between (i) the actual/potential force of an expression, and (ii) the standard/non-standard uses of the declarative. Moreover, in thinking about the data, we should not forget the hearer’s perspective.

Suppose that someone says to you ‘I see you blushing with shame’, whereupon you try to hide your face, and he then adds ‘I didn’t mean it. I was only joking.’ You rightly took his utterance in the first instance as an assertion. By taking the point of view of the hearer/reader as crucial and by asking for a sign that cancels assertive force it is possible to relate declarative mood to assertive force: without counter indications an utterance of a declarative sentence is understood as having assertive force (^{van der Schaar, 2007, p. 65}).

Here is a datum: without counter-indications, an utterance of a declarative will be interpreted as an assertion. Of course, this presumption may later be canceled. This doesn’t eliminate it as a presumption. We may take this at face value and say that, conventionally, a declarative sentence is standardly used to perform assertion acts. In this regard, van der Schaar introduces the notion of an assertion candidate: “The assertion-candidate need not be actually asserted; it is what can be asserted. The assertion candidate differs from the assertion made in that it has no [actual] force” (2007, p. 65).¹⁸ In van der Schaar’s example, the hearer immediately took the speaker to be asserting that p. This is so because declarative sentences are conventionally used to perform assertion acts. The linguistic type is associated with an assertion candidate. This is semantic information, not a pragmatical affair.

Max Kölbel holds a related view-note that he doesn’t use “content” in Alston’s sense, but as equivalent to propositional content:

The purpose of a conventional account of assertion is […] best explained by viewing it as the proposal that there is literal force in addition to literal content and literal sub-content. Most philosophers of language make a distinction between the literal, encoded meaning of utterance types, and the non-literal meaning which tokens (or tokenings) of those types can have on particular occasions of use […]. However, most philosophers of language employ such a distinction only at the level of the content, or sub-content, of utterances, not at the level of illocutionary force or speech acts. Most theorists have no room for a notion of literal assertion, i.e., a performance that counts as an assertion in virtue of the linguistic meaning of the utterance type used. Instead, these theorists employ a notion of assertion according to which asserting is a matter of having certain communicative intentions, with the result that they see themselves forced to deny the existence of illocutionary force indicators, i.e., context-invariant meaning at the level of speech acts (2010, p. 109).¹⁹

Conventionalism stresses the similarity between our assertion practice and other conventional activities, such as scoring a goal. The mere existence of such kind of actions depends on the existence of certain norms and social conventions. Without football rules, goals wouldn’t exist.²⁰

Kölbel distinguishes two main components of social conventional activities: implementation conventions and constitutive norms. For every type of conventional act C there must be a non-conventional act-type N whose performance counts as the realization of C. Striking a ball over some line in a court counts as scoring a goal. Signing some papers counts as buying a home. Implementation conventions specify which kind of procedure N counts as performing C. But these conventions, by themselves, lack any normative dimension. Constitutive norms provide C’s normative profile. Once I buy a house, I acquire certain ownership rights. Once I get married, I must honor certain obligations.

Going through the procedure of uttering ‘The train is late’, in normal circumstances, counts as asserting that the train is late. The normative significance of this action consists in it bringing about certain rights and obligations concerning the asserter and his or her audience. For example, the assertion might introduce the conversational background information that the train is late, with the result that this may be presupposed in subsequent utterances (2010, pp. 111-112).

Kölbel is pointing to two important things: (i) we need to take conventions seriously, and (ii) we must recognize that speech acts have a normative dimension.

In fact, these are arguably two crucial aspects of Austin’s own account of illocutionary acts. According to Dörge, we may define Austin’s account of illocutionary acts as follows:

An act A is an illocutionary act iff the following conditions are satisfied:

Following ^{Strawson (1971)} and ^{Bach (2011)}, I am taking Grice’s communicative intentions as relevant for the elucidation of illocutionary acts. Whether or not this turn is completely faithful to Austin’s original conception is debatable. Notwithstanding, there is obviously an interesting and promising connection between Gricean communicative intentions and Austin’s securing of uptake. In any case, I take the Gricean turn to be needed. Now, in defending a conventional account of assertion, Kölbel is pointing to other crucial aspect of Austin’s conception of speech acts: its conventionality. So here we have another reason to take conventions into account: doing so will honor Austin’s account of illocutionary acts. And so it will prove useful in developing our AA.²²

We should thank Kölbel (and Austin!) for stressing both the conventionality and the normative dimension of speech acts. However, he fails to distinguish between assertion/assertion-candidate. So he is committed to a view according to which the content of a speech act is always strictly connected to its implementation conventions. We are avoiding that route motivated by the under-determination phenomena. I will, then, follow Kölbel’s conventionalism without fully embracing it. Instead, I will promote an integration of conventions and intentions that takes van der Schaar’s distinction between actual and potential force into account.

As we saw, if we rely exclusively on conventions we face unnecessary complications in responding to under-determination arguments and the ontological question. If, on the other hand, we rely exclusively on communicative intentions we lack a complete satisfying response to the epistemological question-and we neglect a crucial aspect of the Austinian conception of illocutionary acts. But, why choose? We should clearly distinguish between:

By so doing, we may recognize the importance of both conventions and intentions in linguistic communication. So, I am promoting an integrated view: ontological intentionalism and epistemological conventionalism.

Here is an account of assertion that follows Kölbel’s conventionalist suggestions. Suppose I intend to assert that the book is on the shelf (that p). How do I manage to do it? (i) I perform (an instance of) a non-conventional act-type N whose performance counts as the realization of an assertion-act (implementation convention). (ii) Ipso facto I acquire certain rights and obligations conventionally attached to this kind of conventional act-Austin’s conventional effects (constitutive norm). For instance, I utter:

By so doing, I (i) assert that p and (ii) acquire certain rights and obligations.

Here is an account of assertion that assumes the general correctness of ontological intentionalism and epistemological conventionalism. Suppose I intend to assert that the book is on the shelf (that p). How do I generally and efficiently manage to do it? (i) I perform (an instance of) a non-conventional act-type N whose performance counts, prima facie, as the realization of an assertion-act (implementation convention)-N’s syntactical properties make it an assertion candidate. (ii) I am allowed (I have the right) to, literally and sincerely, assert that p iff I believe that p. (iii) In normal conditions, I count as having asserted that p because I manage to make my audience (a) believe that I believe that p, and (b) recognize that I intend them to believe that I believe that p. (iv) Ipso facto I acquire certain rights and obligations/commitments associated with this conventional act-type-for instance, I acquire the Brandomian obligation to defend or retract my assertion if challenged (^{Brandom, 1983}).²³

Let’s now say that we hear John utter an instance of a Moore-paradoxical sentence:

In this case the account will predict a general and irreparable illocutionary failure. The reason is that this sentence will be, literally and sincerely, assertable only if John believes that p and believes that he doesn’t believes that p. So we are simply perplexed as to what he is trying to communicate. The utterance of such paradoxical sentences will be generally useless to perform any illocutionary act. There’s a practical absurdity in uttering them, similar to the absurdity we see if someone tries to open a bottle of wine by smashing it into a wall.²⁴

This account of assertion postulates the existence of some conventions/rules/affirmability/use conditions that allow a speaker to effectively and efficiently communicate its intentions. But to develop an AA we need to find some kind of illocutionary difference between an utterance of 11 and an utterance of:

Here I propose to follow ^{Gomez-Torrente’s (2012} and ²⁰⁰⁵⁾ work on the analogy between Wittgenstein’s rule following considerations-interpreted à la^{Kripke (2007)}-and some Wittgensteinian considerations on the nature of epistemic justification (^{Wittgenstein, 1974}).²⁵ What’s crucial for our purposes are the affirmability conditions there explored. Let’s say that a speaker who, literally and sincerely, uses a KAS aims to assure its audience that p. We may then postulate the following use conditions for the performance of the illocutionary act of assurance.

I am allowed to literally and sincerely assure that p iff:

Notice that this approach is consistent with the provisional content delivered by the oratio obliqua test. By using a KAS, I am expressing my strong conviction that p (I am giving others my authority for saying that S is P).²⁶ The assurance is successful only if the addressee believes that the speaker believes with great certainty that p (and does so by virtue of recognizing the speaker’s intention to communicate this to her).

What about implementation conditions? We may say that to utter a first person KAS is to perform (an instance of) a non-conventional act-type N whose performance counts, prima facie, as the realization of an assurance-act. N’s syntactical properties make it an assurance candidate. What about the conventional effect of the act? What normative consequences follow from it? We may tentatively say that the assurer is now held strongly responsible/committed for it being the case that p.²⁷

V. If I know, I can’t be wrong

“I know that p, but p is false” sounds like a contradiction in terms. How could we explain this fact assuming AA? Notice that a speaker S may assert the second conjunct only if she believes that p is false. But the following principle sounds plausible: if S believes that p is false, then S doesn’t believe that p. Now, according to AA, S can literally and sincerely use the first conjunct to assure that p only if she believes that p. So we are perplexed about S communicative intentions. Does she believe that p? This sentence produces an irreparable illocutionary failure. No wonder it sounds awful.

VI. An Austinian dissolution of the skeptical argument from ignorance

I may be more or less certain that p. This may be represented by the (subjective) probability r I assign to p. Moreover, at a conversational context C I may be more or less inclined to express my certainty that p and to commit myself to p. Perhaps I strongly believe that p. But, should I assure that p? According to AA, this is a practical decision. I should aim for a wise decision, avoiding both temerity and shyness. Let’s make this explicit in the use condition for assurances:

I am allowed to, literally and sincerely, assure that p, in context C, iff:

Here r is an idealization representing a wise practical decision. According to AA, there’s no objective value one should honor. But we may idealize a little and say that at any given context the assurer acts as if there was such a thing.

Let’s now say that a speaker who, literally and sincerely, uses a first-person KAS denial aims to un-assure its audience that p. We may then postulate the following use conditions for the performance of the illocutionary act of un-assurance:

I am allowed to, literally and sincerely, un-assure that p, in context C, iff:

This helps us reflect about P1’s use conditions. What about P2?

“I don’t know that p” has certain implementation conditions. Prima facie, it is used to un-assure that p. However, in the context of a conditional construction this un-assurance candidate presumption gets canceled. This is analogous to what happens when an assertion’s candidate presumption is cancelled by an “if”. In “If it rains we will not play tennis”, the assertive force presumption gets cancelled.²⁸ Either I make a conditional assertion or I assert a conditional proposition (^{Stalnaker, 2012}). Now, it doesn’t make sense to interpret P2 as un-assuring a conditional proposition. Instead, we should analyze it as a conditional un-assurance.

Notice the difference between these sentences:

In contrast with (13), it is natural to interpret (14) as a conditional bet. But what do I do when I perform a conditional bet? By betting I commit myself to a certain course of action. So, if performing a speech act involves endorsing a commitment, performing a conditional speech act involves endorsing a conditional commitment. By uttering (14) I endorse a conditional commitment to bet in case it rains. Analogously, by uttering P2 I endorse a conditional commitment to un-assure that I have hands in case I have also un-assured that I am not a BV.²⁹ This is as it should be. Any context in which I (literally and sincerely) un-assured that not-BV is a context in which P (no-BV) < r. So, P(BV) ≥ (1-r). But given that BV implies not-Hands, P(not-Hands) ≥ P(BV) ≥ (1-r).³⁰ Thus P(Hands) < r. So P2 is a very sensible thing to say.

Moreover, notice that, according to AA, Closure is a reasonable inference-in a Stalnakerian sense of making anyone who commits to its premises thereby committed to its conclusion (cfr. ^{Stalnaker, 2009, p. 65})-. Any context in which I (literally and sincerely) assured that p is a context in which I believe that p and P(p) ≥ r. Suppose that in such a context I (literally and sincerely) assured that p logically implies q (given that I am certain of it). Then I am committed to assure that q. After all, P(q) ≥ P(p).

AI is also a reasonable inference. Anyone who wishes to remain uncommitted to his non-vat condition must remain uncommitted to his having hands, on pain of irrationality. If you concede the first premise to the skeptic, you are in (light) trouble. But, of course, you don’t have to. It may be perfectly sensible to express your (great) certainty that you have hands and that you are not a BV. AI is a powerful argument. But it doesn’t preclude you from assuring all sorts of things, and with clear conscience.

VII. Third-person ascriptions

So far, I have restricted my attention to first-person knowledge ascriptions. However, it could be doubted whether anything like the AA could be extended to third-person cases. According to AA, when a speaker uses a KAS she aims to assure her audience that p. But this is not what seems to be happening in a third-person ascription like:

Apparently, contra Austin, Mary is simply reporting that John has some particular kind of belief. She seems to be asserting that John is in certain kind of special and peculiar cognitive state.

Moreover, skeptical arguments from ignorance seem just as compelling in their third-personal form. Consider this third-person version of AI (3P-AI):

P5. John doesn’t know that he’s not a (handless) brain in a vat-even if he believes it.

P6. If John doesn’t know that he’s not a (handless) brain in a vat, then he doesn’t know that he has hands.

C. So, John doesn’t know that he has hands.

It is not obvious how we could extend the AA to cover this version of the paradox.

Furthermore, it seems that the factivity objection reappears. Even granting the correctness of the pragmatic explanation provided in V, there’s still a difference between (16) and (17):

While (16) sounds odd, (17) is perfectly fine. But, some may argue, here the Austinian strategy is useless because we can’t appeal to the fact that when we assure others that p we thereby represent ourselves as believing the claim in question.

We may find an answer to these worries by following Gómez Torrente’s work (2012 and 2005). Again, what’s crucial for our purposes are the affirmability conditions there explored. Let’s say that a speaker who, literally and sincerely, uses a third-person KAS (like 15) aims to certify that someone’s belief that p is trustworthy. This is very natural from an Austinian or Kripkensteinian perspective. From this point of view, we should focus in KAS’s use conditions and function. By certifying that John’s belief that p is trustworthy Mary is giving her word for it. She is indicating others that John’s grounds for believing that p are good enough for her. She is indicating others that, ultimately, John bases his belief in other beliefs that provide her great confidence that p. So, by using (15) she is expressing her strong conviction that p. By so doing she is enhancing social coordination.

We may postulate the following use conditions for the performance of the illocutionary act of certifying.

I am allowed to literally and sincerely certify that X’s belief that p is trustworthy iff:

The certification is successful only if the addressee believes that both the speaker and the third person involved in the ascription believe that p based on some beliefs that provide the speaker with great certainty that p (and does so by virtue of recognizing the speaker’s intention to communicate this to her).

We may also provide use-conditions for a KAS denial in the obvious way:

I am allowed to literally and sincerely un-certify that X’s belief that p is trustworthy iff:

By un-certifying John’s belief that p I indicate others that I don’t trust John’s belief that p. I inform others that, ultimately, he bases his belief in some beliefs that don’t provide me great confidence that p. This is useful information.

We may now provide a response to the worries voiced at the beginning of this section. Our treatment of the paradox will be analogous to the one offered for its first-person version. I will follow my earlier idealization and stipulate r to be certain threshold of confidence that the speaker finds sufficient in some context C for assuring that p. Again, according to AA, strictly speaking, there’s no such an objective value one should honor. In the end, the speaker must make a wise practical decision. But we may idealize a little and say that at any given context the assurer acts as if there was such a thing.

Given this, and assuming for simplicity that we are now considering contexts in which both John and I believe that John is not a BV (that not-BV John; not-p for short) then if, at C, I have successfully used P5 to un- certify John’s belief that not-p, then, at C, I have also expressed my doubts regarding John’s grounds for his belief in not-p. This means that the subjective probability I assign to not-p is less than r. This also means that I wouldn’t assure myself that not-p. What about P6? By uttering P6 I endorse a conditional commitment to un-certify that John is handed (that q) in case I have also un-certified that not-p. This is as it should be. Any context in which I (literally and sincerely) un-assured that not-p is a context in which P (not-p) < r. So, P (p) ≥ (1-r). But given that p implies not-q (BV implies not-Hands), P (not-q) ≥ P (p) ≥ (1-r). Thus P (q) < r. So P6 is a very sensible thing to say.

Moreover, (3P-AI) is a reasonable inference-in the Stalnakerian sense defined in section VI. If I express my commitment to its premises, I must also commit myself to its conclusion. Not only that: if we express our commitment to its premises, we must commit to its conclusion. If we want to remain uncommitted to John’s non-vat condition, we can’t still remain committed to his having hands, on pain of irrationality. If we concede the first premise to some skeptic, we are in (light) trouble. But, of course, we don’t have to. (3P-AI) is a powerful argument. But it doesn’t preclude us from assuring all sorts of things, and with clear conscience.

What about the worry that AA is incapable of explaining our factivity intuitions regarding third-person ascriptions? The objector complained that the Austinian strategy is useless in this case because we can’t appeal to the fact that when we assure others that p we thereby represent ourselves as believing the claim in question. But this is simply not true. It is perfectly natural to appeal exactly to this fact in order to explain the intuitive difference between (16) and (17). According to the AA, there’s a crucial difference between these cases. While (16)’s use-conditions demand the speaker to believe that the book is on the shelf (17)’s use-conditions don’t. This explains why (16) may generate an irreparable illocutionary failure while (17) is perfectly fine.

VIII. Conclusion

There’s something about epistemic contextualism. It’s motivated by ordinary practice and it has some serious anti-skeptical potential. At the same time, it has been the target of some strong objections. One may associate the position with Austin. But it is at best dubious whether he was indeed a contextualist. He appears to have suggested a much more radical position, one still ignored in the literature. Here I sketched an Austinian account of KAS. I responded to an important objection based on the alleged factivity of knowledge and showed how we could use the proposal to deal with the sceptical paradox. It would certainly be premature to conclude that the proposed account is correct based on the considerations so far discussed. My aim was merely to suggest that an Austinian account of knowledge ascriptions deserves some serious consideration and discussion.