Shifty Contextualism About Epistemics

1.1. How the Hybrid Theory Explains Both Disagreement and Reasonableness

The hybrid theory explains both of the central desiderata that we want to explain. It explains how two agents can disagree while both being reasonable. Say that Ignorant Ian and Well-Informed Wendy are looking for some keys.

Modal bases are, let’s say, sets of propositions. Ian is talking about m₁, and Wendy is talking about m₂.

The hybrid theory then makes these two predictions.

Then Wendy and Ian disagree in what they accept. They are both accepting propositions that can’t both be true. The proposition that m₁ is correct and the proposition that m₂ is correct can’t both be true, since it can’t be that both contain all truths. That’s the simple account of disagreement. How can the two agents nonetheless be reasonable? Reasonable belief is also a hybrid state, consisting of acceptance of some modal base as epistemically correct, plus reasonable belief about that modal base. It crucially does not require reasonable belief about what’s epistemically correct. So Ian is reasonable because he has reasonable beliefs about what is compatible with his own modal base, the one that leaves open whether the keys are in the drawer.

Now the hybrid theorist is expanding the kind of disagreements that the contextualist can explain. She is suggesting that our two agents can be disagreeing about what’s epistemically correct. Importantly, though, she still allows for the sort of disagreement that more traditional contextualists can explain. Suppose that two mathematicians are disagreeing about what follows from a set of axioms. One might assert that p must be true, while the other asserts that p might be false. In this case, they would be disagreeing about what’s compatible with some modal base: the set of axioms. Moreover, we can evaluate the reasonableness of the different beliefs about what’s compatible with the axioms. We might deny that a mathematician is reasonable in believing that p must be true if we don’t think that the mathematician has a proof of p. Traditional contextualists have an elegant explanation of this kind of case, and the hybrid theory retains that elegant explanation.

The hybrid account of disagreement generalizes straightforwardly to all cases of disagreement, including eavesdroppers. Imagine that Ian is talking with Wendy. Eavesdropping Evelyn isn’t part of the conversation, but she knows where they are.

Many contextualists struggle to explain this exchange, but the hybrid theorist explains it crisply. The hybrid theorist takes disagreement to be disagreement about what’s epistemically correct. But she thinks of epistemic correctness in a striking and distinctive way. It’s a comparative notion: some modal base is more correct than some group of bases. Which ones? Any modal base whatsoever. Let’s illustrate with Ian. He’s thinking about some modal base m₁. In accepting that m₁ is epistemically correct, he’s taking it to be more epistemically correct than any other modal base whatsoever. Now m₁ is incomplete. It doesn’t say where they keys are. Imagine that Evelyn does know where the keys are: say, that they’re in the drawer. Then she accepts a modal base m₂ that includes that proposition, as epistemically correct. That’s why they disagree. Evelyn accepts that m₂ is more correct than m₁, and Ian accepts that m₁ is more correct than m₂.

This proposal works only because it takes Ian to be doing something very odd. He’s accepting a proposition that he’s in a position to know is false. Ian knows that either the keys are in the drawer or they’re not. He also knows that if the keys are in the drawer, m₁ is not epistemically correct, because m₁ plus the proposition about the keys is more correct. And he knows that some other modal base if more correct if they’re not in the drawer. So Ian is in a position to know that it’s false that m₁ is epistemically correct.

In general, though, it can perfectly appropriate to accept propositions that you can know to be false. Here are two examples. First, you can accept that Hamiltonian mechanics is true and relativist mechanics is false to simplify your calculations in an engineering context. In doing that, you’re accepting a proposition that you know to be false. Second, some religious people will insist that God loves his children, while also insisting that God isn’t male. Since his presuppo ses that the salient individual is male, they’re accepting a proposition that they don’t believe. In this case, they’re accepting it because they regard it as the least unsatisfactory option. Other possessive pronouns (in English) presuppose that the individual is female, or that the group has several members.

The central feature of my account of epistemics is that epistemics pattern with the use of Hamiltonian mechanics, or gendered language for God. In both cases, your practical interests can make it perfectly appropriate to accept propositions you know to be false. The practical interest in the epistemic case is to reason from the largest amount of information that you have, without relying on anything that your information leaves open. The latter goal is why ordinary agents only accept an incomplete modal base as epistemically correct.

1.2. Contrasting the Hybrid Theorist with Other Contextualists

The hybrid theorist strictly improves on the approaches that other contextualists have developed. To see why, imagine that Wendy told us (1).

On the hybrid suggestion, uses of (1) communicate these three propositions.

• Ian is heading towards that drawer, because he (reasonably believes that m₁ is compatible with the keys being there and accepts that m₁ is epistemically correct).
• But I, Wendy, know that m₂ is incompatible with the keys being there and accept that m₂ is epistemically correct.
• So Ian is wrong in (believing that m₁ is compatible with the keys being there and accepting that m₁ is epistemically correct).

Different modal bases are incorporated into the complements of Ian’s knowledge reports than are incorporated into the complements of Wendy’s knowledge reports. Why? Because the two agents take different modal bases to be epistemically correct.

This suggestion contrasts with an important contextualist suggestion from Janice Dowell (2011; 2013). She holds that, when one person assertively utters an epistemic and her interlocutor assertively utters its negation, the two parties are disagreeing because they’re expressing propositions about the very same modal base. Dowell gives a powerful defense of this proposal, suggesting that speaker intentions determine the modal base incorporated into the propositions expressed, and that speakers can intend to talk about the same modal base even if they don’t know exactly what modal base they’re talking about. We can simplify the comparison with Dowell by pretending that speakers intend to talk about whatever modal base is epistemically correct in the present context. Dowell thinks that they’re talking about whatever base is in fact epistemically correct, while the hybrid theory takes them to be talking about different particular modal bases which they accept to be epistemically correct.

Does Dowell’s view differ substantively from the hybrid theory? Yes. The two make different predictions about the reasonableness of judgments about what might be. The hybrid theorist predicts that it’s really easy to have reasonable beliefs about what might be. Reasonable belief depends only on reasonable belief about what’s compatible with a set of propositions that you’re already thinking about. Dowell does not make it as easy to have reasonable beliefs about what might be. Since Ian and Wendy disagree, the two attitude reports should express propositions about the very same modal base. But it’s hard to see how Ian could have reasonable beliefs about that modal base, unless he is sensitive to the information that Wendy takes to bear on the question. And it’s unclear how he could be sensitive to that information. He isn’t aware of what it is!

We can appreciate another distinctive feature of the hybrid picture by comparing it with a proposal by Kai von Fintel and Anthony Gillies (2011). Properties like epistemic correctness play no role in their proposal. They instead propose that utterances of epistemic modals communicate a cloud of propositions, propositions about several different bodies of information. One of those propositions describes only the speaker ’s body of information; another describes the body of information of the entire group. They suggest that it’s appropriate to assert an epistemic modal when you accept one of those propositions—typically, the one about your body of information. But your assertion puts the other propositions in play. So other speakers can react to your utterance by reacting to some of the other propositions put in play. This proposal explains why it’s easy for a speaker to appropriately assert an epistemic. It’s appropriate as long as her own information is compatible with the prejacent. But it also explains why her hearers can object. They’re objecting to the other propositions put in play.

Their proposal doesn’t generalize to the attitude reports under discussion.

What proposition does Ian believe? It seems like he only believes the proposition about his own body of information. (Ian may have no beliefs about any other salient bodies of information; he may suspend judgment on them.) But then it’s unclear what Ian is mistaken in believing. He is not mistaken in believing the proposition about his own body of information. That proposition is true. But that proposition may be the only relevant proposition he believes. Since ^⌜Ian is wrong in believing S^⌝ entails that Ian believes S, it’s hard to see how von Fintel and Gillies can explain this case.[3] The hybrid theory does much better.[4]

The new contextualist account makes it much easier to see how someone can have reasonable but mistaken beliefs about what might be. However, there is a danger that the account makes it too easy to have mistaken beliefs about what might be. Virtually all ‘might’ beliefs involve accepting an incomplete modal base as epistemically correct. (Ordinary agents aren’t omniscient, and mostly know that they aren’t.) But we saw in the last section that no incomplete modal base could be epistemically correct. So the account seems to predict that all ordinary ‘might’ beliefs are mistaken. And this prediction would be very costly. For one thing, I don’t regard my own ‘might’ beliefs as mistaken, whatever this account might predict.[5]

Fortunately, though, the proffered account does not make this prediction. Distinguish two pictures of a mistake in acceptance. On a belief-centric picture, you can appropriately say ^⌜it would be a mistake to accept p^⌝ if you yourself believe ¬p. On an acceptance-centric picture, you can appropriately say ^⌜it would be a mistake to accept p^⌝ if you yourself accept ¬p. The acceptance-centric picture turns out to disarm the problem in the previous paragraph, while also being significantly more plausible than the belief-centric alternative.

I’ll develop the difference between the belief-centric picture and the acceptance-centric picture with an example with gendered religious language. Suppose that Bob and Carl both use gendered pronouns for God, because they regard those pronouns as the least inadequate option. But they believe that God is neither male nor female. Now since his presupposes that the salient individual is male, an utterance of (2) communicates the proposition (+).

The belief-centric picture predicts that Bob can appropriately say (2) even if Bob agrees that God infuses grace in God’s children. After all, Bob doesn’t believe that God is male. (He believes that God is neither male nor female.) Since he doesn’t believe the proposition Carl accepts, the belief-central account predicts that (2) is appropriate. And that prediction is incorrect. Bob is ex hypothesi perfectly happy to use gendered pronouns for God—he doesn’t disagree with Carl about his use of pronouns. In order for (2) to be appropriate, Bob and Carl have to disagree about the infusion of grace. The acceptance-centric picture, by contrast, predicts that (2) is inappropriate. That picture grounds mistakes in acceptance in disagreement in acceptance. Since Bob and Carl both accept that God is male, Bob can appropriately say (2) only if he disagrees with Carl about whether God infuses grace in God’s children.

And our initial problem about ordinary ‘might’ beliefs does not arise on the acceptance-centric conception. I grant that we should believe that no incomplete modal base is epistemically correct. (The last page gave a compelling argument for that point.) But that belief doesn’t require us to accept that no incomplete modal base is epistemically correct. On the contrary, we have very good reason to accept incomplete modal bases as epistemically correct. And in diagnosing mistakes in acceptance, I reason from the propositions that I myself accept. Consider someone who is less informed than me. The modal base that they accept to be correct includes fewer propositions than mine does. So I can reasonably take them to accept something false, since, given the propositions that I myself accept, one of the propositions that they accept is false. Reasoning from what I accept also explains why I won’t regard myself as mistaken about what I accept. Given the propositions that I myself accept, all the propositions that I accept are true.[6]

Now there are lots of fruitful and interesting questions to explore about the acceptance-centric conception, and we’ll take up some of them as the paper goes on. The important point for now is that that conception has the right structure for a Goldilocks solution to the contextualist’s problem: not too few mistaken beliefs about what might be, and not too many, either.

2.1. A Central but Underappreciated Question

My systematic account answers a central but underappreciated question about the epistemic use of modals. This question is genuinely underappreciated. So I will begin by arguing for the importance of this question, and arguing for it in some depth.

This central question is about attitudes towards epistemics. It’s an instance of a more general question about attitudes towards modals, like (3).

(3) can be used to talk about the demands of morality, the demands of prudence, or the demands of some particular set of laws. It’s standard to explain this point by supposing that she must keep this promise has one more argument place than its surface syntax suggests. The moral use of (3) communicates a proposition with that argument place saturated with something that represents the demands of morality, the prudential use with something that represents the demands of prudence, and so on.

My central question arises after we explain what all the uses of (3) have in common. It asks what distinguishes the different uses of (3). What distinguishes the proposition the moral use communicates from the proposition the legal use communicates? A Naive Thought is that they’re distinguished because the propositions believed have different truth-conditions.

And this Naive Thought is pretty natural. Jill’s moral belief is different from her belief about the laws, because the two propositions have different modal profiles. The moral proposition is true because of what the moral facts are, and the legal proposition is true because of what the legal facts are. And the moral and legal facts don’t always line up.

Unfortunately, though, the Naive Thought doesn’t work in full generality. It is possible for the laws of some society to coincide exactly with the laws of morality, across all of modal space, even while someone gets confused, and thinks that the laws don’t coincide. Imagine that some set of laws does coincide with morality in just that way. It’s still possible for Jill to think that she’s morally required to keep this promise, without thinking that she’s legally required to. In that case, the moral use of (3) asserts something true, and the legal use asserts something false. But the Naive Thought can’t explain this fact. In this case, the moral proposition is true at a world w iff the legal proposition is true at w. There isn’t any difference in modal profile for the Naive Thought to latch onto to explain how the moral but not the legal use of (3) asserts something true.

Let’s make this point more concrete. We might model the demands of morality with orderings on worlds: w1 is ranked higher than w2, which is tied with w3, which are both above... And I’m imagining a case where the very same ordering that models the demands of morality also models the laws of some society. (The society designed its laws so that they correspond with morality throughout all of modal space.) Call that ordering o₁. Given the Naive Thought, the moral and legal uses of (3) both communicate that Jill believes that o₁ requires her keeping this promise. And that won’t work, as it stands. The moral use asserts something true and the legal use asserts something false.

As a result, we need to draw on richer resources than the Naive Thought uses. We need to explain how Jane can be thinking of o₁ in different ways, as the demands of morality, or the demands of positive law. I will call the different ways we think about the orderings modes of presentation. You might think of these modes of presentation as properties that we predicate of orderings. Or you might think of them as the concepts that we use to cognize the orderings. Or you might think of them as grist for Stalnakerian diagonalization.[7] Then we can say that the moral use of (3) asserts that Jane believes that o₁ (thought of as the demands of morality) requires her keeping this promise, which can be true even if she doesn’t think that o₁ (thought of as the demands of the law) requires her keeping this promise.

Let’s return to epistemic modals. I’ve just given an argument that it’s necessary to posit modes of presentation to distinguish the moral use of modals from the legal use. Once we’re familiar with this kind of argument, we’re in a position to appreciate how to generalize it for the epistemic use. The argument requires two different uses of modals: in the initial case, moral and legal uses. So I need a use of modals that pairs with the epistemic use, to power my argument. I will use what I’ll call the circumstantial use of modals. We can say the tree can grow to be ten feet tall to indicate something about its nature, that it’s compatible with its nature that it grows to be that tall. Given this use, we can construct the same sort of case constructed earlier. It’s possible for someone’s information to coincide exactly with the facts that are relevant for the circumstantial use. Say that John finds himself in that situation. There is still a difference between epistemic belief ascriptions and circumstantial belief ascriptions. (4a) can be appropriate even if (4b) isn’t.

John is aware of the information that he has. He knows that that information is compatible with the tree growing to be ten feet tall. (4a) is then perfectly appropriate. But he doesn’t realize that that information coincides with the facts about the nature of the tree. In fact, he suspends judgment about the nature of the tree. Then (4b) isn’t appropriate.

The best way to explain this contrast is to posit a mode of presentation for the circumstantial use, and a mode of presentation for the epistemic use. (4a) and (4b) both communicate propositions about the same body of information— the same modal base, to use a phrase from Kratzer.[8] But (4b) is inappropriate even though (4a) is appropriate, because appropriate use of (4b) requires John to think of that body of information as being about the nature of the tree. The explanation is exactly like the explanation of other instances of Frege’s puzzle. John thinks that Hesperus is bright can be appropriately assertable even if John thinks that Phosphorus is bright isn’t. Why? The use of Phosphorus suggests a different mode of presentation than the use Hesperus does. So we should posit a mode of presentation to distinguish the epistemic use of modals.[9]

Appreciating this point puts us in a better position to appreciate the hybrid theory from Section 1. That theory was using a particular account of the mode of presentation that distinguishes the epistemic use. It appealed to a modal base being epistemically correct—that is, as containing all truths. It uses that property as the mode of presentation for the epistemic use. (4a) is appropriate because the modal base that John accepts to be epistemically correct is compatible with the tree growing that tall. Now remember that the overarching goal of this section is to show that the hybrid theory is highly principled, and not ad hoc. I’ll pursue that goal by showing that the hybrid theory is a simple and natural answer to what I will call the Central Question.

Features of the hybrid account that initially look ad hoc do not look the same way when we see them as answering this question.

The appeal to epistemic correctness is a natural idea about the mode of presentation that distinguishes the epistemic use. And we should be willing to entertain this idea, once we see that we do need some such mode. (The conclusion will briefly consider possible modes of presentation that distinguish the epistemic use.)

2.2. Motivating the Rest of the Hybrid Theory

The rest of the hybrid theory follows from a very simple conjecture about our Central Question. It holds that the mode of presentation of the epistemic use is the property being-epistemically-correct, and that that property is part of a presupposition of a use of an epistemic.

What are presuppositions? I won’t say much here. I’ll just understand the presuppositions of a sentence as information that the sentence (i) is somehow associated with, but (ii) is interpreted as being part of the background assumptions in the conversation. It’s easier to appreciate this notion from particular examples.

Each of these sentences are somehow associated with two propositions, with one proposition understood as foregrounded and at-issue, and the other as backgrounded and not-at-issue. (i) is associated with the backgrounded proposition that someone did steal the tarts. You wouldn’t use (i) unless you thought that someone stole the tarts, but the main point of using (i) is to communicate that Tom didn’t steal the tarts. The other examples are similar—(ii) is associated with the backgrounded proposition that someone else was in NY last night, and (iii) is associated with the backgrounded proposition that John used to beat his dog.[10]

The hybrid theorist conjectures that the property being-epistemically-correct is part of the presupposed, not-at-issue content associated with bare epistemic utterances.

This conjecture is a simple empirical hypothesis. It’s definitely not the only viable hypothesis about epistemic modals under attitude reports, or the only way to answer our Central Question, not even when we appeal to epistemic correctness. But if this empirical hypothesis is right, the hybrid account of epistemic modals from Section 1 follows immediately.

The hybrid theory follows immediately because of a highly general fact about presuppositions under attitude reports: that acceptance is the attitude appropriate for presuppositions of the complement. I will spend some time belaboring this general fact, because it’s so important for evaluating the hybrid theory. Consider some uncontroversial examples of presupposition triggers, like a cleft sentence like it wasn’t Tom who took out the trash. If you assertively utter that sentence, you presuppose that someone took out the trash, while asserting that Tom didn’t take out the trash. In general, an utterance with the form ^⌜it wasn’t F who G- ed^⌝ presupposes that someone G-ed. An interesting pattern emerges when cleft sentences are embedded in attitude ascriptions.

(6a) can be true even if John doesn’t hope that someone took out the trash. (Maybe he knows that someone took out the trash, and regrets that the trash was taken out.) The other examples are the same. (6b) can be true even if John doesn’t fear that someone took out the trash; and (6c) can be true even if John doesn’t conjecture that someone took out the trash. In each case, the sentences are appropriate only if John has some attitude towards the proposition that someone took out the trash. In general, ^⌜A V-es that S^⌝ can be true even if A does not V the presuppositions of S’s complement.[11]

We should note another important fact about these examples. It’s normally odd to use (6a), (6b), or (6c) unless John is in some sense aware that someone took out the trash. The best explanation of this oddity is that attitude reports usually communicate that the matrix subject accepts the presupposition of the complement. That is, the matrix subject has to be treating the presupposition of the complement as true for some reason. (Robert Stalnaker 2002 has influentially emphasized this point.) Now you might wonder if I’ve failed to notice a stronger generalization: that that matrix subject needs to believe the presupposition of the complement. This alternative is not liberal enough. Suppose that we’re trying to figure out who would have taken out the trash, if someone did take out the trash. You can temporarily assume that someone did take it out, to simplify the conversation that follows. Once you’ve made that assumption, (6d) is perfectly appropriate.

In this case, though, you don’t believe that someone took out the trash. You’re just accepting it as a way to figure out who would have taken out the trash, if someone did. (6d) is appropriate even in that case. So we should follow Stalnaker in being more liberal about the attitude that the matrix subject has towards the presupposition of the complement. The matrix subject need not believe the presupposition of the complement in order for (6d) to be appropriate.[12]

To summarize:

• you can accept a proposition without believing it, or knowing it,
• the matrix subject of an attitude report ^⌜A V-es that S^⌝ need not also V presuppositions of the complement,
• but many attitude reports do communicate that the matrix subject accepts the presuppositions of the complement.

I take these generalizations to be constraints on a full theory of presupposition; any such theory will vindicate these three generalizations. It is unfortunately difficult to explain why these generalizations hold.[13] But the important point for my purposes is that these generalizations are true, not why they are true.

The goal of this section was to motivate the hybrid theory which associates attitude reports like (7) with the following pair of attitudes.

(7) Jane thinks that the keys might be in the kitchen.

• Jane thinks that m₁ is compatible with the keys being in the kitchen, and
• Jane accepts that m₁ is epistemically correct.

Two features of the hybrid view were particularly apt to look ad hoc: the appeal to epistemic correctness, and the appeal to acceptance. We’ve now seen the simple conjecture that motivates these two features. Epistemic correctness is introduced as the mode of presentation that distinguishes beliefs about what epistemically might be, from other kinds of modal beliefs. And the hybrid theorist is someone who conjectures that that mode of presentation is a presupposition of utterances with epistemic modals. The hybrid theory is not ad hoc. Instead, it is one the simplest and most conservative ways of extending a basically Kratzerian understanding of modals to answer a question that everyone needs to answer.

2.3. Consequences of this Explanation of the Hybrid Theory

This presuppositional explanation of the hybrid theory has an important advantage. To appreciate this advantage, let’s recall one reason the hybrid theory initially looked implausible. That theory takes ordinary talk about what might be to involve acceptance of a proposition that the speakers can know to be false. Asserting that the keys might be in the drawer means asserting that your own body of information m₁ is compatible with the keys being in the drawer, while also accepting that m₁ is epistemically correct. In accepting that m₁ is epistemically correct, you’re accepting that it’s more correct than any modal base whatsoever. Now you’re not omniscient. So you’re in a position to know that the proposition that you accept is false. Why? Let m₁ be the modal base that you accept to be epistemically correct. Since you’re not omniscient, m₁ is incomplete. So there is some way of extending it to be more complete: there’s some p where adding either p or ¬p would make it more complete. If p is true, m₁ plus p is more epistemically correct. And if ¬p is true, m₁ plus ¬p is more epistemically correct. m₁ can’t be epistemically correct; in order for it to be, it would have to be more correct than every other modal base.

This feature of the hybrid account cleanly explains eavesdroppers. It’s always reasonable for a better-informed eavesdropper to disagree with you, because they reject the proposition you accept: that m₁ is epistemically correct. But this feature of the hybrid account might look implausible. Ordinary discourse about what might be just doesn’t involve this sort of commitment! I appealed to two analogies to make the idea more plausible. One was about religious language: religious people who use a gendered pronoun to talk about God, even though they don’t believe that God is male. I suggested that they’re accepting that God is male, for conversational purposes. Now I take pronouns like he and him to trigger the presupposition that the salient individual is male.

An utterance of God loves his children presupposes that God is male.[14] It’s nonetheless intelligible for a religious person to make that utterance, even if they don’t believe this presupposition. They only need to accept it for their conversational purposes.

The hybrid theorist takes the epistemic use of modals to involve accepting a proposition you don’t believe. And the analogy with gendered pronouns is quite tight. The two phenomena have exactly the same linguistic explanation: they are both due to the fact that you only need to accept presuppositions of your utterance. Though this picture of the epistemic use is initially surprising, it ends up having a lot going for it. The epistemic use involves temporarily treating a body of information as the only truths. That’s why any eavesdropper will disagree with you. They don’t treat that body of information as the only truths. But you yourself have great practical reasons for treating that body of information as the only truths. You have practical reason for not relying on any more complete body of information: any such body of information will contain a proposition that you’re not sure about.

My presuppositional explanation of the hybrid theory makes the theory significantly more plausible. It makes the theory much more simple and elegant. And it shows that natural language already has the resources that the hybrid theorist uses in explaining the data: the analogy with gendered pronouns is exact. But the presuppositional explanation gives rise to a further question. I’ve emphasized that the hybrid theory can explain disagreement about what might be. And given the presuppositional explanation of the hybrid theory, disagreement about what might be is usually disagreement in presupposition. We’re disagreeing about which modal base is epistemically correct. But you might wonder if disagreement in presupposition licenses the full range of disagreement markers that are in fact licensed. After all, it’s very odd to use certain locutions if you want to reject the presuppositions of an utterance.

No and that’s wrong seem appropriate only as objections to what (8) asserts (that Bill didn’t steal the tarts), not as objections to the presupposition. Epistemics pattern quite differently. No and that’s wrong are both perfectly appropriate.

Someone who assertively utters (9′) disagrees with someone who assertively uttered (8′). And that disagreement is possible even if the two agents agree about what’s compatible with the salient bodies of information. The hybrid theorist takes them to be disagreeing in presupposition—the second speaker is rejecting the presupposition that the first speaker has made. But it doesn’t seem like presuppositions can bear the weight that the hybrid theorist needs. You can’t say no or that’s false if you reject the presupposition.

This worry confuses a feature of some presupposition triggers with a feature of all presupposition triggers. We should distinguish two subclasses of presupposition triggers: those whose appropriateness depends on their already being satisfied in the initial context, and those whose appropriateness doesn’t. It-clefts fall in the first subclass. It’s inappropriate to say (8) totally out of the blue, if we don’t already accept that someone stole the tarts.[15] But other presupposition triggers fall into the other class, and can be used even if the initial context doesn’t satisfy the presupposition. For example, you can use start to introduce new information: I can say that John started smoking even if my hearers are ignorant about whether he smoked in the past.[16] Crucially, though, you can use our locutions to object to the presupposition that start triggers.

This difference illustrates a general pattern. Locutions like that’s wrong can target the presupposition trigger iff the presupposition trigger can be felicitously used to introduce new information. It’s not yet clear why the two phenomena go together— why you can use that’s wrong to reject presuppositions if the presupposition trigger can be used to introduce new information. But the generalization does seem robust.[17]

Now the hybrid theorist’s presupposition about epistemic correctness can definitely be used for new information: the hearer may not realize that some new modal base is epistemically correct until the speaker says her bit. It’s thus unsurprising that you can use locutions like that’s wrong to reject the presupposition.

2.4. General Problems about Agreement

Let’s close by considering a problem about agreement. Suppose that I know only that the keys aren’t in the kitchen or garage, and you know only that they aren’t in the dining room. Then it’s true that

As a first pass, the hybrid theory seems to take (10) to communicate that

But there is no modal base mx such that you and I both accept that mx is epistemically correct. If mx contains the proposition that the keys aren’t in the dining room, I won’t accept that mx is correct. (I don’t have a high enough credence in that proposition.) Similarly if mx contains the proposition that the keys aren’t in the garage. Then you won’t accept mx as correct. (You don’t have a high enough credence in that proposition.) The hybrid theory doesn’t seem capable of explaining why (10) is true.

This problem parallels a central problem about anaphora in attitude reports. Solutions to that central problem are also solutions to this problem. Suppose that I think that Miley is the only kitten who killed a mouse, and I hope that Miley didn’t torture it first. You think that Tabby is the only kitten who killed that mouse, and hope that Tabby didn’t torture it first.

This example involves linked attitudes about the same object—agreement plus hope. It’s natural to think that the two attitudes are linked because they are both singular attitudes about the same object.

But this suggestion is mistaken for exactly the same reason that the initial hybrid gloss on (10) is mistaken. No assignment of values to variables verifies the constraints in (11lf). You don’t think that Tabby did the killing, and I don’t think that Miley did it.

We need to draw on some other formal tool to explain how (11) is true. One option is to appeal to the sort of unanchored discourse referents that Irene Heim (1982) and Hans Kamp (1981) have influentially pioneered. Functionally, those discourse referents are denoting concepts that play a similar role to Fregean individual concepts. They allow individuals to think and talk about individuals without requiring the individuals to be able to think singularly about those individuals. Where i is a term for an unanchored discourse referent ^⌜A V₁-es that i is F and A V₂-es that i is G^⌝ is true iff, in every maximal consistent extension of the propositions that A V₁-es and the propositions that A V₂-es, there is some individual that is both F and G. (11) is then true. Let F abbreviate “kill the mouse” and G abbreviate “torture the mouse”. I think that Miley F-ed and I hope that Miley didn’t G. Every relevant maximal consistent extension of my attitudes includes the two embedded propositions. At every such extension, then, there’s an object that F-ed and didn’t G: Miley. You think that Tabby F-ed and you hope that Tabby didn’t G. So at every maximal extension of your attitudes, there’s an object that F-ed and didn’t G: Tabby. Appealing to unanchored discourse crisply explains how (11) is true.

Unanchored discourse referents can crisply explain why (10) is true.

Where i is a term for an unanchored discourse referent:

Then (10) is true. Every consistent extension of my beliefs includes that mm is compatible with the keys being in the bedroom and mm is epistemically correct. So every consistent extension verifies the constraints on the discourse referent—(10) is true of me. And every consistent extension of your beliefs includes that my is compatible with the keys being in the bedroom and my is epistemically correct. So every consistent extension verifies the constraints on the discourse referent—(10) is true of you too.

This explanation preserves the two central advantages of the hybrid account. It gets disagreement about what might be exactly right. If my modal base includes the negation of the prejacent and your modal base doesn’t, we disagree. And in that case, (10) would be false. (There is no maximal extension of my attitudes where the modal base that is epistemically correct is compatible with the keys being in the bedroom.) At the same time, this explanation also preserves the central epistemic virtues of the hybrid theory. The rationality of attitudes towards unanchored discourse referents can bubble up from the rationality of the corresponding singular thoughts. Suppose that I’m rational in believing that Miley killed the mouse, and you’re rational in believing that Tabby did it. Then (11′) is true.

In other words, rational attitudes about particular objects (Miley and Tabby) ground rational attitudes about discourse referents. The appeal to unanchored discourse referents is fully compatible with the central epistemic virtues that the hybrid theorist claims.

Now there are many other ways of theorizing about intensional anaphora in examples like (11).[18] But we should expect any adequate theory of intensional anaphora to solve the problem for the hybrid theorist. Given the hybrid theory, the present problem just is a problem about intensional anaphora.[19]

3.1. Expressivist Contextualism

It’s also illuminating to contrast the hybrid view with mixed views that draw on both contextualist and expressivist resources. They also take general difficulties about epistemic expressions as a license for optimism for the legitimacy of what they’re doing. Work by Stephen Finlay and Gunnar Björnsson (2010) illustrate this general strategy.[23] They start by noting our practical reasons for using the most complete body of information available. The fuller body is less likely to lead us astray. So when Wendy says that Ian is wrong to think that the keys might be in the kitchen, she is recommending against relying on the information that Ian has. This account of disagreement is different than any account from previous sections. It’s not disagreement about what’s compatible with a modal base. But neither is it disagreement about which modal base has some further property, like being epistemically correct. Both sorts of disagreements are disagreements about which propositions are true. By contrast, Finlay and Bjornsson are thinking of the disagreement as practical disagreement: disagreement about what to do, rather than what to believe or what to accept.

This kind of proposal is viable only given an answer to the general problems that dog expressivists, including the Frege-Geach problem. Suppose I know that the cat isn’t hungry. I can appropriately assert that

In asserting (12), I need not be asserting that Ian is wrong about what is compatible with his information. (He and I can agree about that point, even as I insist on (12).) I can use (12) to assert that he’s wrong in having the whole complex of attitudes that the theory associates with bare utterances of the cat might be hungry. This is the basic point that the expressivist contextualist needs to explain.

And the expressivist contextualist cannot explain this point, unless she has an answer to the Frege-Geach problem. To illustrate the problem, let’s switch to a more complicated case. Suppose that some axioms of set theory T are provably compatible with some proposition p. Susan the set theorist knows about the proof. To express that T is compatible with p, she assertively utters p might be true. Now she and I agree that other mathematical considerations rule out p. (She’s just using p might be true to express a fact about set-theory, in a context where it’s clear that she’s talking about set theory.) Doug the dogmatist disagrees with us. He does not acknowledge the other mathematical considerations that rule out p. He says p might be true, because he also knows that p is provably compatible with T. But he insists that T are the only acknowledged facts that bear on p. I’m willing to criticize Doug in a way I’m not willing to criticize Susan. Contrast this pair.

I’m willing to assert (14) but I’m not willing to assert (13). I’m asserting something different about Susan than I am about Doug. The challenge is to explain what that difference is.

It is easy to answer this challenge given a solution to the Frege-Geach problem. Expressivist contextualists associate an utterance U with a proposition P and some kind of commitment A.[24] U is appropriately assertable for me iff I have commitment A and I satisfy the norm of assertion for P. (p might be true is associated, say, with the proposition that mT is compatible with p and a preference for relying on mT .) Given the knowledge norm for assertion, say, it’s appropriate to assert that sentence if you know that proposition and have that preference. Then we can give a crisp account of why (14) asserts more. (14) asserts that Doug has a preference for relying only on mT , and (13) asserts nothing of the sort.

I’ll accept (**), because I think it’s wrong to prefer relying only on mT . And I’ll reject (*), because Susan and I both believe P.

Crucially, though, this answer works only given a solution to the Frege-Geach problem. In particular, the second conjunct of (**) is interpretable only given a solution to the Frege-Geach problem. It requires us to interpret commitments like A under a negation operator like it’s wrong that. At this point, then, I’ve given one illustration of what it would take to answer my challenge. In fact, though, I think that this is the only way for an expressivist contextualist to answer this challenge. My challenge is to articulate what proposition is asserted by utterance of (13) and (14). I don’t care how that happens. (I’ll allow my opponent as much semantic blindness as she wants.) I just want to know what the difference is between what’s asserted by utterances of (13) and utterances of (14).

3.2. The Severity of this Problem

The only way I see to answer this challenge without answering the Frege-Geach problem is to appeal to metalinguistic reasoning.[25] But we should trust that she can do that only when we understand what exact propositions she takes uses of (13) and (14) to assert. Let’s start with one natural thought.

These propositions do capture the intuitive differences. I don’t believe (+). After all, one of the contexts where Susan would assertively utter it might be that p is a context where she’s not talking about the modal base she prefers. She’s just talking about the axioms of set theory. And she acknowledges mathematical truths that go beyond those axioms. Moreover, I do believe (++). Any context where Doug is talking about the modal base that he prefers is a context where his preference is wrong. (There are other truths beside those axioms!) And we can evaluate these propositions without a solution to the Frege-Geach problem. Evaluating them just requires evaluating the unembedded preferences.

But this kind of proposal cannot work. It forces the wrong predictions about context sensitive vocabulary.

(+++) is true iff Susan has incorrect beliefs about herself. Her beliefs about me are totally irrelevant. Those truth-conditions are definitely wrong. (15) is about Susan’s beliefs about me. It’s not just that (15) has a reading that is about me: it’s that (15)’s only readings are about me. None of them are about Susan. This attempt at a concrete, testable account of the proposition asserted by any of (12) through (14) definitely fails. And it’s hard to see an attempt that does better.

The general problem is that allowing for metalinguistic readings of the complement predicts too many readings of sentences like (15). It’s really hard to fix a theory that overgenerates. (For one thing, you can’t appeal to pragmatic processes that add in additional content. The problem is that there’s already too much content floating around.) I conclude that whatever is asserted by (12) through (14) can’t be a metalinguistic proposition. But the kind of metalinguistic proposal is the only apparent way to explain the difference between (13) and (14), without needing an answer to the Frege-Geach problem. It’s reasonable for us to conclude that the expressivist contextualist does need an answer to the Frege-Geach problem.

Now it is possible to draw on other heterodox resources to explain disagreement. I’ve been focusing on contextualists who draw on expressivist resources to explain disagreement. But it’s also possible for a contextualist to draw on relativist resources to explain disagreement. It’s helpful to see the recent suggestion by Stephen Finlay (2016) in this vein. On his proposal, (14) expresses the false proposition (++++):

(This proposition is false, because p is in fact compatible with the propositions in mT .) Finlay suggests that ordinary speakers are willing to assert (14), because they think that there is some proposition that Doug is disposed to assert, and that is false relative to our information. It’s helpful to see that proposition as the same sort of proposition that a relativist like John MacFarlane (2014) takes bare epistemic utterances to assert: a proposition that can be assessed as true relative to one point and false relative to another. One important difference between Finlay and MacFarlane is that MacFarlane thinks that these propositions exist and are genuinely assessment sensitive.

Finlay does not. He thinks that the commitment to these propositions is an error in ordinary epistemic discourse. Finlay’s proposal raises a number of issues. But one important issue is that he is assuming that ordinary speakers are committed to assessing some propositions as true relative to their own body of information. And some critics of relativism doubt that it’s even intelligible to do that. (Scott Soames (2011) is one example.) This criticism applies as much to Finlay as to MacFarlane. Those critics are challenging the intelligibility of the relativist explanation, not its correctness. And Finlay needs the explanation to be intelligible, even if he doesn’t need it to be correct.

3.3. How the Hybrid Theorist Gets this Data Right

I’ve noted that expressivists and relativists both take the epistemic use of modals to be an important license for optimism on their behalf. They think that some sort of heterodox explanation has to be on the right track, because no traditional explanation will do. Expressivist contextualists might take themselves to have the same license for optimism. In order to explain disagreement and reasonable belief, you have to draw on some sort of heterodox resource. There might be interesting technical questions about how best to do that, but those technical questions shouldn’t shake our confidence that it’s possible.

The hybrid theory from Section 1 undercuts this license for optimism. So it makes the technical questions much more pressing. It eliminates the license for optimism that natural language works in a way that guarantees that those technical questions have a solution. The hybrid theory cleanly explains the data that expressivist contextualists don’t explain. Why? She takes what the agent believes and what he accepts to both embed under wrongly.

The relevant uses of (13) assert less.

So we’ll reject (13), in the story I gave, because Susan is right about whether m₁ is compatible with p. Doug is right about that point, too. But he’s wrong in a further way. He accepts m₁ as epistemically correct. Now I haven’t explained why (13) and (14) have such different readings. But I have answered the challenge that I’ve issued to the expressivist contextualist. I’ve explained how uses of (13) can assert something different from what uses of (14) would assert. This answer is available precisely because the hybrid theorist is so conservative. She doesn’t draw on any heterodox resources (expressivist or relativist), so she doesn’t need to solve the problems that dog those approaches. But I just don’t see how expressivists or expressivist contextualists can clear even this low bar, absent a solution to the Frege-Geach problem.

And this lesson is important for other philosophical questions, too. Finlay and Bjornsson are interested in epistemic expressions in part because they accept a broadly relativist conception of moral belief. Finlay, for example, defends an end-relational semantics for moral sentences, where they express propositions about our ends.[26] He allows that different agents may have different ends even when they’re talking about what they morally ought to do. When I say you ought to keep this promise and you disagree, we may well be talking about different ends. And we may be making true claims about our different ends. But we’ll continue to disagree morally even when we agree about what would promote various ends. So he needs some additional account of moral disagreement that goes beyond the propositions that are communicated. He fills that need with a broadly expressivist account that parallels his treatment of epistemics.

In fact, Finlay and Bjornsson are explicit in taking their account of epistemic expressions as a license for optimism in defense of their relativism. They write: “since contextualism invokes the very same sort of pragmatic considerations (in particular an appeal to privileged contexts) to account for both sorts of context-insensitive assessments, no significant extra cost is added to defend standard relativity” (Björnsson & Finlay 2010: 36). The hybrid theorist undercuts this license for optimism. You don’t need expressivist resources to defend the basic contextualist idea. There is no guarantee that the Frege-Geach problem can be solved. So a fortiori there’s no guarantee that Finlay and Bjornsson’s metaethical relativism is defensible.