1.1 Eliminativist arguments and error

Perhaps the most familiar way of motivating descriptive forms of fictionalism about a region of discourse D proceeds by arguing that sentences within D are typically false, say, because the entities or properties to which they refer (for example, numbers, other abstract objects, possible words, and so on) don't exist, or because its characteristic predicates are uninstantiated or unsatisfied (nothing is right, nothing is good, and so on). It is then pointed out that this threatens to convict speakers who utter those sentences of massive error, which is generally regarded as highly implausible, especially if the linguistic practices in question seem to be successful. Fictionalism is then posited as providing an explanation for why these practices are effective independently of the (potential) falsity of the sentences in question (cf. Eklund Reference Eklund and Zalta2015). After all, if the utterances in question aren't literal, then speakers needn't be thought to use the sentences in question to assert (or otherwise illocute)Footnote ¹¹ e.g., the false propositions that they express; rather, they might even be thought to use them to assert truths.

1.2 The Oracle

This style of argument originates in a thought experiment proposed by John Burgess and Gideon Rosen in their seminal work, A Subject with No Object, and proceeds as follows:

The moral of this story is thought to become apparent when readers are asked to envision themselves as the protagonist of this fable and to reflect on the following two questions.

1. (1) Do you think that you, following your encounter with the Oracle, would persist in talking as before – that is, do you think that you would continue to use sentences that seem to commit you to the existence of abstract objects?

2. (2) If so, do you think that you would be using such sentences any differently than you were before? (Cf. Yablo Reference Yablo2000; Eklund Reference Eklund and Zalta2015.)

To the extent that one expects that they would continue to use such sentences in the same way after meeting the Oracle (and that others would, too), one has gone some measure toward granting the fictionalist under discussion the ingredients that they need to cook up the following argument in favor of their view. For if, after returning from our chat with the Oracle, we would all go on using the relevant sentences as we were before, then just as we wouldn't be using sentences that seem to commit us to abstract objects entirely seriously – that is, literally – after speaking with the Oracle, we weren't using them literally before talking with them, either. In other words, insofar as we're inclined to think that we would most likely continue to use sentences that seem to commit us to abstract objects even with the (Oracle-given) information that there aren't any, and that nothing would seem all that different about our use of such sentences, then since we wouldn't be committing ourselves to positive claims about abstract objects post-Oracle, we weren't committing ourselves to positive claims about abstract objects pre-Oracle either – even if the sentences that we use are in fact so committal. Further, it is important to note that while this particular incarnation of the Oracle argument is designed to support descriptive fictionalist views about abstract object discourse in particular, it should be clear enough as to how it can also be used so as to argue for other kinds of fictionalism once suitably amended.

1.3 Arguments from paradox

These arguments proceed by drawing our attention to the fact that questions about whether certain sentences express truths can seem, depending on how one looks at them, easy or hard to answer (cf. Yablo Reference Yablo2000).

For example, the following can seem sound:

The problem, of course, is that this argument (and many others like it) can also seem unsound. After all, it is controversial – at least in some circles – as to whether there are numbers, and other abstract objects such as properties, possible worlds, moral facts, and the like. Because of this, in certain contexts, these are provocative claims, despite the fact that in most contexts, they are banal (cf. Dorr Reference Dorr2005).Footnote ¹²

Here, too, the fictionalist can aid us in resolving the problem at hand. Typically, they'll say, when we use the sentences in question we use them non-literally so as to avoid asserting their semantic contents and instead to assert something else altogether. However, in some situations (for example, those in which we are discussing whether there are such things as numbers, or other abstract objects such as properties, possible worlds, or moral facts), we do not speak non-literally when we use such sentences. Rather, we use them literally, and assert the propositions that they semantically express.

Thus, the fictionalist can give a neat, elegant, and reasonable-sounding solution to the puzzles presently under consideration. Questions concerning whether certain sentences express truths can seem easy to answer when we are speaking non-literally. However, questions concerning whether certain sentences express truths can seem hard to answer when we are speaking literally, such as when eliminativist arguments are in view. For such arguments seriously call into question whether the propositions semantically expressed by the relevant sentences are true, focusing our attention on their literal uses.

1.4 Analogies with uncontroversially non-literal discourse

A number of fictionalists, most prominently Stephen Yablo, have attempted to support their views by comparing our use of sentences in the target discourse with uncontroversial examples of non-literal speech and arguing that they share a number of core features. Yablo in particular has drafted up a fairly substantial list of such features, noting that while none of these similarities is anything like decisive, and that not all features mentioned are common to all examples, the cumulative effect is “nothing to sneeze at” (Yablo Reference Yablo2000: 21). Although there are far more examples on Yablo's list, the most important for our purposes concern silly questions, impatience, felt gaps, and unobtrusive non-literality (cf. Yablo Reference Yablo2000).

2.1 Epistemic fictionalism: a formulation

In order to show how the arguments just canvassed can be used to defend a fictionalist view in epistemology, in what follows I supply a sketch of what such an account might look like, the resulting position being a form of content epistemic fictionalism (henceforth simply “epistemic fictionalism” for the purposes of this paper). Such an approach would have it that expressions within the relevant region of discourse – namely, knowledge-discourse – are generally used to assert (or otherwise illocute) some propositional content distinct from (and instead of) their semantic contents.Footnote ¹⁵ I then explain how the fictionalist arguments discussed above can be used to support it.

As noted in the introduction, philosophers have long sought to explain an array of kinds of variability in our use of knowledge-attributing sentences. This task, however, has proved notoriously difficult. The fictionalist's main insight is that much of this variability may be due to the fact that we often use knowledge-attributing sentences non-literally, to assert (or otherwise illocute) a variety of propositions. To see how this might help, imagine that we grant that skeptical arguments are possibly sound; that perhaps they succeed in establishing that no one knows much, if anything, at all. From this it would follow, given a standard semantics, that sentences of the form “S knows p” (and the like) usually, if not always, express false propositions. Does this mean that we are not warranted in our use of such sentences? No, says the fictionalist. For, we do not normally use a sentence of the form “S knows p” to assert the proposition that it semantically expresses. Rather, we use it non-literally in such a way so as to avoid asserting its semantic content and to assert some other content (consistent with the proposition that S does not know p) altogether.Footnote ¹⁶

What might this alternative content be? Since non-literal speech is often deeply context-sensitive, we should expect the answer to this question to be more or less context-sensitive as well, depending on the pragmatic machinery involved (given that some pragmatic phenomena tend to be more deeply context-sensitive than others). Because of this, it may be very difficult (if possible at all) to completely spell out this potential alternative content in full generality, much less in any specific case (especially if one takes seriously the possibility that the propositional content that a non-literal utterance of a sentence of the form “S knows p” is used to assert is often indeterminate across a range of (determinate) propositions, as is frequently the case with non-literal assertions).Footnote ¹⁷ Typically asserted propositions may include those along the lines of: that S truly believes p, that S is justified in believing p, that S believes p as a result of a reliable belief-forming process, that S can rule out all relevant alternatives to p, that S can assert p, that S can rely on p in practical reasoning, and so on – in other words, propositions that many epistemologists have argued are associated with knowledge-attributing sentences.Footnote ¹⁸ However, it should be emphasized that these may not even come close to exhausting the propositions that we use sentences of the form “S knows p” to assert. Indeed, the fact that there could be an expansive and elusive array of such propositions might at least partially explain many other kinds of variability in our use of knowledge-attributing sentences (including, as mentioned above, the kinds of variability that occur as a result of changing what is at stake or the salience of alternative possibilities, as well as others in addition).Footnote ¹⁹

This account has two potentially attractive features worth singling out at this time, as they are akin to just those features that are commonly thought to make fictionalist views appealing in general. First, one can endorse it in conjunction with classical invariantism regarding the semantic content of “knows.” This carries with it two distinct benefits: one can maintain (i) a simple semantics for “knows” (thus avoiding many of the problems faced by semantic explanations of variability in our use of knowledge-attributing sentences, such as contextualism) and (ii) a simple metaphysics of knowledge (thus avoiding many of the problems faced by other forms of invariantism, such as subject-sensitive invariantism). Second, because it interprets much of our ordinary knowledge-talk in terms of the more widely recognized phenomenon of non-literal speech, it need not require the introduction of any new pragmatic machinery or special rules (thus avoiding many of the problems faced by a number of other pragmatic explanations of variability in our use of knowledge-attributing sentences).Footnote ²⁰

Of course, as in the case of other types of fictionalism, whatever its problem-dissolving potential, one might think that this view has no chance. After all, knowledge-talk – just like talk about numbers, other abstract objects, possible worlds, morality, and the like – does not generally strike us as non-literal in any way. However, despite this, the form of fictionalism outlined here can be supported by the same sorts of arguments that are usually employed in the service of fictionalist views. In addition, it can be supported by another argument, not universally employable by other types of fictionalists, concerning retraction behavior. I thus take there to be excellent reason to think that an epistemic fictionalism along the lines I sketch above is likely to be at least as plausible as other forms of fictionalism, if not more so.Footnote ²¹ The plan, then, is to devote the remainder of this section to illustrating how all this can be done.Footnote ²²

2.2 Skeptical arguments and error

No matter what, skeptical arguments – like eliminativist arguments – threaten to convict speakers of having made some sort of mistake. For although it seems as if sentences of the form “S knows p” are used to express truths, a number of initially appealing skeptical arguments suggest otherwise. It looks, then, as if we must be wrong about something: either sentences of the form “S knows p” are usually, if not always, used to express falsehoods, or skeptical arguments, despite appearances to the contrary, are (at least in many cases) unsound. Epistemic fictionalism offers a novel and tenable explanation for why we want to insist on saying things like “S knows p” even if we can't refute the skeptic, and for why these practices are effective independently of the (potential) falsity of the sentences in question. Further, we can see that there is a comparison between knowledge-talk and non-literal speech worth drawing here, as we often credit speakers with engaging in non-literal speech when doing otherwise would be to charge them with possibly falling into error by making a false assertion, particularly if we think that they are (a) also in some sense or on some level aware of this and (b) clearly attempting to state a truth.

That said, epistemic fictionalism, like all other forms of fictionalism, does appear to require the controversial psychological claim that speakers are at the same time generally either in some sense or on some level unaware of (or perhaps even at times flat-out wrong about) the fact that knowledge-attributing sentences are typically used non-literally. This is where the argument from unobtrusive non-literality can come to the rescue. For this sort of thing is far from unusual. Indeed, it is plausible to think that non-literal speech normally goes (in some important way) unnoticed by us, as much of our ordinary speech exhibits this quality. To see this, one need only reflect on the variety of kinds of non-literal speech (including, but not limited to, the kinds of non-literal speech mentioned earlier), and on how frequently we engage in them. What's more, not only is non-literal speech regularly not explicitly believed to be such (nor is it the case that people regularly have second-order beliefs to the effect that they are or are not speaking non-literally), it is arguably often misidentified as literal in addition (especially when the locutions in question are sufficiently common). If this were not so, it would not be possible to debate whether a given utterance was literal or non-literal. Thus, it should not be all that surprising if it turns out that much of our knowledge-talk is non-literal as well, and that people largely lack explicit (or second-order) beliefs to the effect that it is, or even frequently have mistaken explicit (or second-order) beliefs to the effect that it isn't. So, that speakers are largely in some sense (or on some level) not cognizant of, and even at times incorrect about, the fact (if it is one) that sentences of the form “S knows p” are typically used non-literally needn't be thought to pose all that serious, or special, a problem for the fictionalist, if it were thought to pose a problem at all.

2.3 The Oracle part II

Here we are, back in line, waiting to talk to the Oracle. Finally, after years more of waiting, it is our turn to put a question to them once again. Fantastic! We humbly approach and ask the other question that has been consuming us for as long as we can remember: “O Oracle, could you please answer just one more question for us? What, if anything, is wrong with skeptical arguments?” To this the Oracle responds: “You again? I don't know why you waste my time with these questions. There is absolutely nothing wrong with skeptical arguments. Now go away and don't come back!”

Dejected, we withdraw. The Oracle isn't pleased with us, but at least we received an answer to our question. No matter what, it seems as if the Oracle has given us: (a) knowledge that skeptical arguments are sound, (b) sufficient reason to believe that skeptical arguments are sound, or (c) sufficient reason to be agnostic about whether skeptical arguments are sound. From this, we can infer that we either have (a) knowledge that many sentences of the form “S knows p” express something false, (b) sufficient reason to believe that many sentences of the form, “S knows p,” express something false, or (c) sufficient reason to be agnostic as to whether many sentences of the form, “S knows p,” express something false. Of course, one might think that skeptical arguments, if sound, show that we have no knowledge whatsoever, and in that case the (a) option above is out (indeed, it is arguably out even if we are agnostic about whether skeptical arguments succeed). That's okay, though – we don't really need it. In fact, we don't really need the (b) option either. So long as we have the (c) option, we have all that the epistemic fictionalist requires.

Let's imagine that all of this happened. We are now in the position to reflect on the following two questions once again.

1. (1) Do you think that you, following your encounter with the Oracle, would persist in talking as before – that is, do you think that you would continue to use sentences of the form “S knows p”?

2. (2) If so, do you think that you would be using such sentences any differently than you were before?Footnote ²³

If you are inclined to answer these questions with a “yes” and a “no” respectively, then the epistemic fictionalist just might be able to convince you that their view is worthy of your consideration. For if we would continue to use sentences of the form “S knows p” despite the fact that we would either (b) believe that they express falsehoods, or (c) be agnostic as to whether they express falsehoods, and nothing would seem all that different about our use of such sentences, then since we wouldn't be using the relevant sentences to commit ourselves in the way under discussion post-Oracle (after all, we would either (i) believe that many sentences of the form, “S knows p”, express false propositions post-Oracle, or (ii) be agnostic as to whether sentences of the form, “S knows p,” express false propositions post-Oracle), then we weren't using them to so commit ourselves pre-Oracle, either.

2.4 A paradox about knowledge

This problem begins by drawing our attention to the fact that questions about what sorts of things are known can seem, depending on how one looks at them, easy or hard to answer.

For instance, the following argument can seem sound:

But so can skeptical arguments that purport to show that such things are not (and indeed cannot be) known. And, once one starts thinking about those, the above argument can start to seem unsound.

Here, too, the fictionalist can aid us in resolving the problem at hand. Typically, they'll say, when we use a sentence of the form “S knows p”, we use it non-literally in such a way as to avoid asserting its semantic content (the proposition that S knows p), and instead to assert, for example, that S can rule out all relevant alternatives to p (possibly among other things). However, in some situations (for example, those in which we are discussing whether skeptical arguments are sound, or in which stakes are high or skeptical arguments or alternatives have somehow been made salient), we do not speak non-literally when we use such sentences. Rather, we use them literally, and assert the propositions that they (semantically) express.

Thus, the epistemic fictionalist can also give a neat, elegant, and reasonable-sounding solution to the puzzle under consideration. Questions concerning what we know can seem easy to answer when we are speaking non-literally. However, questions concerning what we know can seem hard to answer when we are speaking literally, such as when skeptical arguments are in view. For, such arguments seriously call into question whether the propositions semantically expressed by knowledge-attributing sentences are true, focusing our attention on their literal uses.Footnote ²⁴

2.5 More silly questions and impatient reactions

Recall our silly questions from earlier. Can an epistemic fictionalist argue that certain queries involving knowledge also strike us as somehow beside the point, or provoke impatient reactions? Once again, the answer is yes. For if a person is contemplating whether a sentence of the form “S knows p” is being used to express a truth, it seems similarly beside the point (at least, in most contexts) to ask that person whether, for example, they believe that people can rule out the possibility that they are brains in vats (even if p is a proposition about the external world), or whether they think that people can ever substantiate their claims in a non-viciously-circular fashion. Such questions often cause people to become, as Yablo might say, impatient, because they do not seem to be relevant to what is under discussion. Or, they strike people as silly, plausibly in the same kind of way that questions that treat non-literal utterances as literal do.

However, even though the sorts of questions (that is, those related to skeptical arguments) just mentioned are not usually considered to be relevant, or serious, there are contexts in which they are. How might this be explained? The proponent of epistemic fictionalism has a simple and straightforward answer to this question. These questions are relevant when the literal (semantic) content of a sentence of the form “S knows p” is what is at issue – that is, when what is in question is whether S knows p. However, when what is at issue is something else (for example, whether S can rule out all relevant alternatives to p), it is not at all clear that they are, and it quite often seems as if they are not. Because the latter sort of context is the most common, these questions strike us as irrelevant, or silly, most of the time. But because we are sometimes in the former sort of context, they do not always.

2.6 Felt gaps

Speakers also act in ways that suggest a possible perceived gap between what sentences of the form “S knows p” express and what speakers use them to express. To see this, we might observe that although it is admittedly difficult for speakers to spell out what they used a sentence of the form “S knows p” to pragmatically convey (as opposed to what it semantically expresses) it is not completely beyond their ken. Rather, the English language arguably has some devices available for this purpose. For instance, one can clarify one's knowledge attributions by saying things like: “They know, in effect,” “They know, with reasonable assurance,” “They know full well,” “They pretty much know,” “They know well enough,” “They know, for certain,” “They know, without a doubt,” “They know, assuming that q,” and “They don't literally know” (cf. Hawthorne Reference Hawthorne2004: 105).Footnote ²⁵ Further, even if the sentences targeted by the fictionalist cannot be successfully paraphrased in such a manner, once again it is possible to argue for felt gaps in another way. After all, speakers are often very surprised to learn that sentences of the form “S knows p” plausibly carry with them certain commitments that they may at the same time find much harder to accept. And this may well suggest a significant difference between what speakers are inclined to use certain sentences to convey and what the sentences do in fact express. To see the appeal of this way of thinking, try to remember the first time that you heard a skeptical argument. Did its presuppositions regarding what the sentences in question express to some degree surprise you? But did you feel the intuitive pull of the argument nonetheless? If you are inclined to answer these questions with a “yes,” and people are generally disposed to react this way to skeptical arguments, it is reasonable to think that a discrepancy between speaker meaning and expression meaning might explain this.

2.7 Retraction behavior

Another way in which knowledge-talk can be argued to bear similarity to uncontroversially non-literal speech concerns retraction behavior. It is commonly acknowledged that skeptical arguments tend to put pressure on people to “take back” or “retract” their uses of knowledge-attributing sentences. Upon being presented with such arguments, many are inclined to assent to the possibility that they spoke falsely when they previously uttered sentences of the form “S knows p.” Many also find themselves wanting to say things like “S didn't really know p” with certain kinds of emphases. Despite this, however, many seem positively disinclined to change their knowledge-attributing practices much, if at all.

These patterns are just what one should expect if epistemic fictionalism were correct. For it seems as if hearers are always – or, at least, almost always – entitled to insist on a sentence's literal interpretation, even if doing so would be extremely uncooperative (cf. Camp Reference Camp2006a: 296–7). And, in the face of such challenges, although speakers are expected to withdraw their non-literal utterances, they are not typically expected to feel as if they shouldn't have spoken non-literally, or that they should refrain from speaking non-literally in the future. This is even true of fairly clear cases of non-literal speech.Footnote ²⁶ Consider the following:

In this sort of exchange, A can't typically simply dismiss B's literalist challenge by saying, “You're wrong. I do have butterflies in my stomach,” or even, “You're wrong. I am really nervous” (cf. Camp Reference Camp2006a: 296). Rather, there seems to be some pressure on A to retract their utterance by acknowledging that what they said (that they have butterflies in their stomach) was false, even if what they asserted (roughly, that they are nervous) was true.Footnote ²⁷ What's more, A can acquiesce to this pressure without having the strong sense that they shouldn't have said what they did, and without feeling as if they should stop using this sentence (and others like it) in this way. This is not what we would expect if A believed themselves to have made a false assertion. Thus, it seems reasonable to say that what A is retracting in this case is the semantic content of their utterance, or what they said, rather than their claim, or what they asserted.

We can expect something similar to happen in this conversation:

Here, too, there seems to be some pressure on A to retract what they said by acknowledging that it at least may have been false, even if what they asserted was true. Also, A can very often acquiesce to this pressure without having the strong sense that they shouldn't have said what they did, and without feeling as if they should stop using such sentences in this manner.

Interestingly, it is worth pointing out that the fictionalist can also account for what is going on when speakers “stick with” their utterances (even after having been presented with skeptical arguments), which, as many commentators have pointed out, tends to happen from time to time (cf. Cohen Reference Cohen2000b; Hawthorne Reference Hawthorne2000; Davis Reference Davis2007, Reference Davis2015). What they are sticking with is what they asserted (what they used a sentence of the form “S knows p” to convey), not what they said (what the sentence that they used expresses). However, although epistemic fictionalism allows for this, it predicts that it should be harder to pull off, which seems the right result. For, although speakers who engage in non-literal speech typically face significant pressure to retract their utterances in the face of literalist challenges, this is not always the case. Let us again consider a fairly clear case of non-literal speech. If someone says:

although a literalist challenge along the lines of,

always seems in some sense legitimate, it can nonetheless be dismissed, rather than engaged with, if other conversational norms mandate it. For example, if one were in a hurry to finish making their point before Bill had a chance to overhear, they might respond to such a challenge by saying:

This form of argument is not generally employable by fictionalists. I expect that fictionalists about mathematical discourse, for example, would have a harder time convincing people – especially ordinary speakers – to consider retracting their uses of “2 + 2 = 4” or even “There are numbers.” I anticipate that this would also be true of ordinary speakers’ uses of sentences that appear to make reference to other abstract objects such as properties, as well as moral sentences.Footnote ²⁸ Modal fictionalists might find such an argument useful, however – at least to the extent that it is plausible to think that those who go in for possible-worlds talk would be happy to assent to the claim that there aren't really possible worlds. For now, I will consider it an open question as to whether others could as well. For the principal point of the foregoing discussion is that retraction behavior can be used to motivate fictionalist views (insofar as the relevant patterns are similar to those that we observe with respect to uncontroversial or less-controversial examples of non-literal speech). And, it promises to offer a significant measure of support for epistemic fictionalism in particular.

3.1 Assertion and semantic content

In Knowledge and Lotteries, John Hawthorne registers a variety of concerns about pragmatic explanations of variability in our use of knowledge-attributing sentences, the most relevant of which – as far as the fictionalist's proposal is concerned – will be addressed here. The first pertains to what Hawthorne describes as being “subtle issues about the nature of assertion.” Many defenders of pragmatic explanations of variability in our use of knowledge-attributing sentences maintain that although such sentences very often, if not (nearly) always, express falsehoods, we are frequently warranted in asserting them because doing so allows us to convey truths. This, however, prompts the question of why (given that we are in some sense or on some level aware that knowledge-attributing sentences may express falsehoods) we should be thought to assert their semantic contents at all.

I am inclined to agree that such considerations put pressure on pragmatic theorists – including fictionalists – to endorse a position on which the relevant semantic contents are not asserted. However, as Hawthorne points out, this in turn raises two pressing questions. First, if “knows” is typically used non-literally, then why hasn't our semantics evolved to reflect this? After all, if the customary use of a sentence s is to assert a certain propositional content, (almost) invariably, that content becomes the literal (semantic) content of s (cf. Szabó Reference Szabó2001: 305; Hawthorne Reference Hawthorne2004: 123). Second, if “knows” is generally used non-literally, from where could its alleged semantic content derive? Any pragmatic theorist who goes this route needs to tell a story about where knowledge-attributing sentences get their literal meanings, if not from facts about what they are customarily used to assert (Hawthorne Reference Hawthorne2004: 122–3). Although Hawthorne doubts that satisfactory responses to these questions can be supplied, I am more optimistic, and will provide answers to each.Footnote ²⁹

Regarding the first question, two principal considerations are relevant. The first concerns the ubiquity of skeptical arguments and the second, the frequently indeterminate nature of non-literally conveyed propositional content.

3.2 The ubiquity of skeptical arguments

As various commentators have pointed out, one doesn't need (seemingly) far-fetched scenarios involving extended dreams or brains-in-vats to cast doubt on a vast number of our everyday claims to knowledge. Much of the time, far more mundane suggestions will do.

To illustrate this, here are a few examples, originally due to Jonathan Vogel:Footnote ³⁰

In each of these cases, it seems as if A is calling into question B's claim to knowledge without invoking any sort of bizarre skeptical hypothesis. Further, such challenges are easy to construct. The formula is simple: all one has to do, when presented with a sentence that (at least appears to) attribute knowledge, is figure out one proposition that it entails that we would be intuitively disinclined to take ourselves to know.Footnote ³¹ Once this is done, one has the weapons needed to mount a skeptical attack. What's more, ordinary speakers can, and do, attack each other's (apparent) knowledge attributions in this sort of way. And though such challenges often strike people as being antagonistic, they just as often do not (at least in any straightforward fashion) seem to be completely changing the subject.Footnote ³² This would be extremely puzzling if sentences of the form “S knows p” expressed propositions with less demanding truth conditions, as such challenges would seem to miss the point almost entirely if this were so.

Another upshot of these considerations is that we often seem to act as if sentences of the form “S knows p” have more demanding truth-conditions than our use of such sentences in fully (or at least highly) cooperative (i.e., non-antagonistic) contexts might suggest. This gives us reason to suspect that the semantic contents of these sentences cannot be read in any straightforward way off facts about our standard uses of them alone, and especially not off facts about our uses of them in non-antagonistic contexts to the exclusion of all others.Footnote ³³ Rather, to paraphrase Elisabeth Camp, antagonistic contexts are not only real but also sufficiently pervasive and illuminating so as to warrant inclusion in our theoretical deliberations instead of being set aside as abnormalities (or, as I will add, distinctively philosophical oddities). Indeed, as Camp points out, it may well be the case that our linguistic behavior in antagonistic contexts is more pertinent to the task of determining semantic content than is our linguistic behavior in non-antagonistic contexts, as conversations in antagonistic contexts may reveal conversational rules (as opposed to mere conversational conventions) which can be occluded by charity in highly cooperative contexts. If so, then the raising of standards stands to be more suggestive of semantic content than the lowering of standards. As Camp writes:

Thus, in Camp's view, antagonistic contexts are theoretically revealing because “the fact that one or both parties is prepared to be only minimally cooperative, working to conversational rule, helps to reveal what the conversational rules are, in a way that is obscured when the conversational operation is maximally lubricated with convivial charitability. To shift metaphors, draining away the waters of charitability reveals underlying structure where we might otherwise see only an undifferentiated pond” (Camp Reference Camp2013: 29).

Granted, such speculations require more argument than I have provided here if they are to present as fully compelling. However, the fact that they are at minimum serious possibilities should suffice to support epistemic fictionalism as a live option. If it could be shown that our uses of expressions in antagonistic contexts really were more pertinent as regards determining semantic content, then this would do more than just substantiate the claim that epistemic fictionalism is worth pursuing further – it would actually give epistemic fictionalism an edge over its, e.g., semantic competition.Footnote ³⁴

3.3 Indeterminacy

As noted above, the propositional content that a non-literal utterance is used to assert is – in certain cases – frequently indeterminate across a range of (determinate) propositions, and because of this, it is plausible to think that, given epistemic fictionalism, a sentence of the form “S knows p” may not generally be used to assert something (more or less) determinate. Rather, the content that a sentence of the form “S knows p” is typically used to assert may usually be to some degree indeterminate over the set of propositions: {that S truly believes p, that S is justified in believing p, that S believes p as a result of a reliable belief-forming process, that S can rule out all relevant alternatives to p, that S can assert p, that S can rely on p in practical reasoning, …}. Thus, although it is held that we have some idea as to what such sentences are being used to assert, according to the epistemic fictionalist, it may be the case that we can't quite get a handle on precisely what it is.

Why should we agree, along with the fictionalist, that the relevant content may be frequently indeterminate in this way, however? Three considerations are worth noting in connection with this question. First, speakers’ intentions are often insufficiently specific to determine, with any precision, just what proposition or propositions they are asserting, especially when they are speaking non-literally (cf. Walton Reference Walton1993; Soames Reference Soames2002: 134–5, inter alia). Just what does a speaker intend to assert when they say, for instance, “Bill is a bulldozer”? Perhaps they intend to assert that Bill is inconsiderately forceful, that Bill is insensitive to obstacles, both, or neither (as they may intend to assert something else about Bill altogether). Likewise (and for example) it may be indeterminate as to whether a speaker intends to assert, in saying “S knows p,” the proposition that S can rule out all relevant alternatives to p, or almost all, or just the ones the speaker cares about, or perhaps something completely different. Further, even if assertion is, e.g., a social phenomenon, similar questions will arise pertaining to what a speaker is committed to when they use a sentence non-literally. Second, the burden of proof is on those who would maintain that some asserted content is determinate to specify what that content is, or at least give some reason for thinking that it must be so. For if one can't spell out just what a sentence is being used to assert, why should one think that it is being used to assert a determinate propositional content at all? If this challenge is not met, it seems as if we are within our rights to assume that the content that a sentence is being used to assert is not determinate.Footnote ³⁵ Third, and finally, if the contents that sentences of the form “S knows p” are used to assert were generally indeterminate, then this would explain why epistemologists have struggled to the degree that they have in their quest to specify the truth or assertability conditions of sentences of the form “S knows p,” and the fact that while many theories have proved remarkably plausible with respect to certain cases, they have proved exceedingly implausible with respect to others, for indeterminate things are, by their nature, elusive.

3.4 Blocking semantic modification

We are now in a position to give an answer to the first question posed at the beginning of this section. For it should be clear that (at least) two things might have prevented non-literal meanings from becoming literal meanings in this instance: (i) non-literal assertions of sentences of the form “S knows p” are often used to assert indeterminate propositional contents (thus making it the case that there is nothing sufficiently determinate to semantically encode) and (ii) our ability to effectively wield commonplace skeptical arguments (such as those discussed under the heading “the ubiquity of skeptical arguments” above) depends on the literal meaning of “knows” being such that sentences of the form “S knows p” have highly demanding truth conditions – perhaps so highly demanding that they are (nearly) impossible to meet.

Though I take the first of these points to require no additional argument, one might understandably wonder why the second should be endorsed. The short answer is that skeptical arguments – including those that we employ on a regular basis (and not just in the philosophy classroom) – can succeed only if a case can be made for the claim that the subject in question does not know a given proposition; however, if knowledge-attributing sentences did not have demanding truth conditions, such cases would be difficult, if not impossible, to make. For skeptical arguments can perhaps only have real bite when the literal meaning of “knows” is such that it is very difficult for sentences of the form “S knows p” to be true. Otherwise, challengers would be open to allegations of having made non-sequitur remarks, and although (as noted earlier) speakers can sometimes “stick with” their non-literal assertions even in the face of skeptical arguments, it does not seem as if they can accuse those who employ them of completely changing the subject. And while other theories on the market may be able to account for these phenomena as well, it is clear that epistemic fictionalism too offers a plausible way to explain them.Footnote ³⁶

3.5 Knowledge-talk and semantic content

This brings us to the second of Hawthorne's questions. Explaining exactly how the semantic content of an expression gets to be its semantic content is no easy work, and there are many divergent ways of going about it, all of which are controversial. It is commonly presumed, however, that what a sentence is conventionally used to assert does some pretty heavy lifting. A more precise explication of this line of thinking can be found in Soames (Reference Soames2002), which offers the following constraint on the semantic content of a sentence:

As Hawthorne notes, though on the right track, C+ is insufficiently general because it does not apply to sentences that we use in our own silent deliberations and practical reasoning. That said, an analogous constraint can be generated by invoking the notion of conscious acceptance:

Of course, as Hawthorne acknowledges, principles of this kind are obviously not incompatible with views on which the semantic contents of knowledge-attributing sentences are held to be something very different from what such sentences are typically used to convey, as proponents of such views can accept them while appealing to caveats (i) and (ii) in C+ and to the exception clause in C++. However, according to him, this presents a problem, as defenders of such accounts are left with the at best exceptionally difficult task of explaining just where they might get their semantic contents from instead. For presumably, such an explanation would have to include information about how sentences of the form “S knows p” behave in other contexts: otherwise, there would be no reason to suppose that they are being used non-literally in the contexts in question. However, it does not seem as if such information is forthcoming, as on the views under consideration, sentences of the form “S knows p” are typically used non-literally. Further, unlike expressions that are more plausibly used non-literally most of the time (such as “This line is taking forever”), sentences of the form “S knows p” do not have lexical constituents (relevant to the task at hand) that are held to have a sober use in some significant number of contexts. But, if the semantic content of “knows” rarely, if ever, plays a role in the propositions that are conveyed and believed when knowledge-attributing sentences are used, from where might their purported semantic contents derive? To the extent that one thinks that this question should be answered, and that there are no better places to look for an answer than those already canvassed by Hawthorne, one might be tempted to conclude at this point that views on which knowledge-talk is held to be largely non-literal are indefensible (Hawthorne Reference Hawthorne2004: 122–3).

I would like to question three assumptions about such views that Hawthorne appears to make. The first is that they must hold that “knows” does not have a sober use in some significant number of contexts. The second is that they must hold that the semantic content of “knows” rarely, if ever, plays a role in the propositions that are conveyed and believed when knowledge-attributing sentences are used. Finally, the third is that they must hold that the semantic content of an expression must somehow be determined by facts about speakers’ uses of it in non-antagonistic contexts, rather than from facts about their uses of it in antagonistic contexts or facts that do not directly concern use at all.

Regarding the first assumption, it seems as if “knows” quite possibly does have a sober – and yet highly, if not impossibly, demandingFootnote ³⁷ – use in a significant number of contexts on the fictionalist account proposed here. After all, we can observe on the basis of the discussion of retraction behavior in Section 2.7, and Camp's work on antagonistic contexts and their relevance in Section 3.2, that the literal, i.e., sober, use of “knows” might indeed be so demanding. We have also noted that skeptical arguments can be employed both within and outside of the philosophy classroom, and on a fairly regular basis – especially in antagonistic contexts – which gives us additional reason to suspect that sentences of the form “S knows p” may have more stringent truth-conditions than our rather permissive use of such sentences in non-antagonistic (i.e., fully, or highly, cooperative) contexts suggests. In connection with this, it is worth emphasizing that the examples provided at the outset of Section 3.2 exemplify just a few ways that skeptical arguments between people – which notably share the same basic structure as traditional skeptical arguments proposed in philosophy – might play out in ordinary conversations, especially in cases where agents are not disposed to interpret each other charitably. Nearly any everyday claim to know can be undermined in a similar fashion, as is evidenced by the fact that such arguments, if compelling, appear to suggest that we should be hesitant to attribute knowledge regarding propositions about the past, present, and future (as they clearly generalize beyond the specific situations to which they are meant to apply).Footnote ³⁸ Further, while philosophical discussions about skepticism per se are not part of everyday English-language discourse, contradicting other people's perceived claims to know – overtly or covertly (even pedantically) – in antagonistic contexts is. Certain skeptical possibilities, such as the press making a mistake, a person lying, or an unforeseen disaster occurring are frequently admitted to be knowledge-imperiling when they are raised even if they are ultimately set aside. To see this, one need look no further than ordinary discussions regarding, say, politics or personal safety, in which conspiracy theories, calls for vigilance, and other types of far-fetched, but still possibly true, allegations or suspicions abound. In places such as these, as elsewhere, the old adage, “You never know,” is a common refrain.

Moreover, it seems as if sentences of the form “S knows p” are no (or at least not a lot) worse off than “This line is taking forever” in this regard. For although “forever” can be used literally, it arguably is very rarely used in this way, unless one thinks that what was once a hyperbolic, non-literal use has been encoded into the semantics of the English language in such a way so as to make it a literal use. This, however, is not something that Hawthorne seems to think, as he appears willing to countenance the possibility that “This line is taking forever” is in fact typically used hyperbolically, and presumably, “forever,” rather than any of the other terms in this sentence, would be to blame.

Hawthorne's second assumption is also problematic. For even if sentences of the form “S knows p” are not typically used to assert their semantic contents, it does not follow from this that the semantic content of “knows” rarely, if ever, plays a role in the propositions that are conveyed and believed when such sentences are used. To see more clearly why not, let us again take a look at what an epistemic fictionalist claims typically gets asserted when sentences of the form “S knows p” are uttered with declarative force. According to them, the content that a sentence of the form “S knows p” is customarily used to assert may usually be to some degree indeterminate over the set of propositions: {that S truly believes p, that S is justified in believing p, that S believes p as a result of a reliable belief-forming process, that S can rule out all relevant alternatives to p, that S can assert p, that S can rely on p in practical reasoning, …}. Granted, none of the propositions mentioned here explicitly reference knowledge or knowing (although they tacitly might rely on some conception of them). But there may be others in the set that do. Possibilities include propositions along the lines of: that S's believing p is close enough to knowing p, that S's believing p resembles knowing p in some relevant respect, and that S can act as if S knows p, and so on. Of course, it would have to be the literal meaning of “knowing” or “knows” that they consider to be in play at this point, on pain of regress. So it seems as if, on their view, the literal meaning of “knows” actually can play a significant role in the propositions that are conveyed and believed when knowledge-attributing sentences are used.

Finally, Hawthorne's argument appears to rely on the unjustified and overly simplistic assumption that the semantic content of an expression must somehow be derived from facts about speakers’ typical uses of it in non-antagonistic contexts, rather than from facts about their uses of it in antagonistic contexts or facts that do not directly concern use at all. It is, however, far from clear as to whether this is correct. For, as discussed above, antagonistic contexts are not only real but also sufficiently pervasive and illuminating so as to warrant inclusion in our theoretical deliberations – indeed, it may even be the case that we should look primarily to expressions’ uses in antagonistic contexts in order to ascertain their semantic contents. Thus, not only is there no reason to assume that the uses of an expression that are responsible for fixing its semantic content must be its most common (or “majority”) uses, but there is reason to think that we in fact should not take an egalitarian attitude toward all uses of “knows,” and that we should prioritize certain uses – namely, those made in antagonistic contexts – over others, especially given that our retraction behavior is what it is.

Moreover, perhaps focusing solely on facts about use (whether in non-antagonistic or antagonistic contexts) is misguided from the get-go. Consider the possibility that semantic contents are grounded not in facts about speakers’ uses of expressions considered alone or as such, but rather in facts about the broader communicative or cognitive purposes that speakers have that are associated with their uses of those expressions. For if not all uses of an expression have equal weight when it comes to determining its semantic content, something must determine their relative importance. Purposes may do this: after all, some uses of an expression might be more central to the achievement of our aims than others, and because of this, one might think that those uses should be accorded greater importance as far as determining its semantic content is concerned. However, if this is right, then although it could be the case that the use-facts could not change without the semantic facts changing, since any change in the use-facts would be due to a change in the purpose-facts (as, by hypothesis, purpose-facts determine use-facts – in other words, our purposes for using an expression will determine how we use it), it may be the case that the latter (the purpose-facts), rather than the former (the use-facts), are what ultimately determine the semantic content of an expression.Footnote ³⁹

If this hypothesis is correct (and perhaps even if not, as we may need to ask the following question anyway if we are to figure out how to weigh use-facts), in order to properly answer a question of the form “What is the semantic content of the expression e?” we would first need to be able to answer the question “What is the (primary) purpose of the expression e?”Footnote ⁴⁰ In order to ascertain the semantic content of “knows,” then, we would first need to provide an answer to the question: What is the (primary) purpose of the expression “knows”? Or, in other words, what is knowledge-talk (principally) for?

One initially promising line of response to this question is that knowledge-talk plays a distinctive role in (conscious, deliberate) inquiry. Saying that it is not known whether p is true generally yields (an at least pro tanto) reason to start inquiring about whether p is true (for the first time or anew), as it introduces the idea that fruitful discussion about whether p is true is possible (provided that p isn't taken to be beyond our ken). Similarly, saying that it is known whether p is true generally yields (an at least pro tanto) reason to stop inquiring about whether p is true. For, it introduces the idea that fruitful discussion about whether p is true is not possible. We use knowledge-talk to challenge, but we also use it to give ourselves a place to rest. Although both of these functions of knowledge-talk are obviously crucial, the former arguably takes a certain sort of priority over the latter. For even though the ability to end inquiry about whether p is true (if only temporarily) is important to our proper functioning as epistemic agents, it seems as if the ability to begin inquiry about whether p is true is more fundamental. After all, one cannot end inquiry without first having begun it. Moreover, the main (if not the whole) point of ending inquiry about something seems to be to free up cognitive resources so that one might inquire about something else instead. If this much is granted, then it is possible that the primary purpose of knowledge-talk is to coordinate inquiry in a unique sort of way, either by inspiring us to begin inquiry, or end it (so that we can inquire about something else). And if the semantic content of an expression e is grounded in facts about the primary purpose of e, then the semantic content of “knows” might somehow be grounded in these facts about its distinctive role in coordinating inquiry (provided, of course, that its primary purpose is approximately as it is described here), rather than in facts about our uses of it per se. And if the more demanding “questioning-use” of “knows” is more fundamental (as opposed to the less demanding “resting-use”), then it may be better regarded as its literal meaning than the less fundamental “resting use”.

Although I cannot flesh out and defend a more detailed proposal here, it should be clear enough from all of this that it is not in any way obvious that the semantic content of an expression must somehow be derived from facts about speakers’ uses of it – rather, perhaps facts about purposes are more apt. To be sure, (at least some) facts regarding our uses of certain expressions are undeniably useful for determining (in the epistemic sense) just what the purpose of an expression is. After all, for the reasons just mentioned, at least some purpose-facts are likely to be revealed by use-facts. But it does not follow from this that facts about our uses of certain expressions determine (in the metaphysical sense) their semantic contents. Again, facts about purposes may well do that. And even if use-facts are what ultimately ground semantic content, in order to figure out how to weigh these, we might need to get into a discussion of purposes nonetheless.Footnote ⁴¹

3.6 Conclusion

I hope to have shown that the form of epistemic fictionalism outlined here is well-supported by several notable patterns of fictionalist argument. What's more, it can also be motivated by an additional argument concerning retraction behavior, and can weather some pretty tough objections.Footnote ⁴² This establishes it as a position worth pursuing further, especially since no other way of addressing the problems under discussion is even close to being widely accepted.Footnote ⁴³