Prospective Contextualism

Prospective Contextualism
March 28, 2015
Matthew Mandelkern
Abstract
I develop a new theory of epistemic modals, according to which a bare epistemic
modal quantifies over the prospective common knowledge of the conversation in which
it is used. I develop this theory within the orthodox semantic and pragmatic framework
for analyzing discourse dynamics. My aims are threefold: to offer a new and attractive
account of epistemic modality; to propose, more generally, a new way of thinking about
negotiation about contextual parameters; and, finally, to defend the orthodox semantic
and pragmatic framework, which has come under sustained attack in the literature on
epistemic modals in recent years.
1
Introduction
The following exchange is totally unremarkable:
(1)
a.
b.
c.
d.
[Jack:] The keys might be in the garage.
[Ruth:] No, they can’t be; Julie had them last, and she always leaves them in
the house.
[Jack:] Ok, then I think they must be upstairs.
[Ruth:] Yes, they must be in Julie’s room.
Yet the canonical contextualist account of the semantics of epistemic modals has struggled
to give a plausible account of how ordinary dialogues like this one work.1 Worries about
the adequacy of the canon to account for the behavior of epistemic modals have prompted a
variety of heterodoxies,2 many of which reject not only the specifics of the canonical account
of epistemic modals, but also the orthodox semantic and pragmatic framework for analyzing
discourse dynamics within which the canonical approach is typically spelled out.
1
An epistemic modal is a natural language modal expressed by a term such as ‘might’, ‘must’, ‘may’,
‘could’, ‘possibly’, and so on, under an epistemic (in a sense to be made precise) reading, in contrast to
deontic, circumstantial (or metaphysical), or dynamic readings. I will use ‘might’ as an exemplar of an
epistemic possibility modal, and ‘must’ as an exemplar of a strong epistemic necessity modal. I will discuss
intermediate grades of epistemic modals, like ‘probably’, later on. I won’t discuss other forms of epistemic
modals, like modal adjective or adverbs, but it should be clear how to apply the strategy of this paper to
them.
2
E.g. dynamic accounts (e.g. Veltman (1996), Willer (2013)), expressivist accounts (e.g. Yalcin (2007, 2012a,
2011), Swanson (2015)), innovative forms of contextualism (e.g. von Fintel and Gillies (2011), Stalnaker
(2014)), and relativist accounts (e.g. Egan et al. (2005), MacFarlane (2011)).
1
In this paper I defend that orthodox semantic and pragmatic framework, according to
which an assertion of a sentence at a context is a proposal to add the proposition expressed
by the sentence to the common ground of the conversation. I do so by laying out and
defending a theory of epistemic modals within that framework. According to my theory,
which I call ‘prospective contextualism’, bare epistemic modals—epistemic modals that are
otherwise unrestricted by features of the context (e.g. by clauses like ‘for all S knows’ or
embedding operators)—quantify over the set of worlds compatible with the propositions
which are common knowledge after the epistemic modal claim is accepted or rejected. This
approach is a substantial revision of the canonical account of epistemic modals, but unlike
many of its competitors, it is solidly within the boundaries of the orthodox theory of discourse
dynamics.
There have been a great many proposals in recent years about the proper treatment of
epistemic modals. The present proposal seeks to advance the literature in three ways. First,
and most importantly, I argue that prospective contextualism is a better theory of epistemic
modals than its competitors: many of the best competitor views, even if they are empirically
plausible, make what seem to me to be substantial theoretical sacrifices when they modify
or reject various planks of the orthodox theory of discourse dynamics. I return to this point
in the conclusion, where I make explicit comparisons with some other theories of epistemic
modals.3 Second, the account I give aims to provide a new framework for modeling not
just epistemic modals, but also a wide range of natural language constructions that are used
to negotiate about parameters of the context. Finally, in presenting what I take to be an
attractive theory of epistemic modals (and, more generally, of contextual negotiation) within
the bounds of the orthodoxy about discourse dynamics, I aim to defend that orthodoxy from
the sustained attack it has sustained in the literature on epistemic modals in recent years.
I will begin by laying out the orthodox framework for analyzing discourse dynamics. I
then lay out the canonical account of epistemic modals and the most potent objection to
that view. I then spell out my own view, prospective contextualism. I argue that this view
gives an adequate account of the responsible production and negotiation of epistemic modal
claims, and that this account provides a helpful and general framework for thinking about
the way we use language to negotiate about parameters of the context. Finally, I show
that prospective contextualism makes empirically adequate predictions about the meaning
of embedded epistemic modals.
3
Let me briefly note here, though, the relation between the present proposal and the proposal about epistemic
modals laid out in Stalnaker (2014). Prospective contextualism is similar in motivation and empirical
predictions to Stalnaker’s view, and part of what I aim to be doing is spelling out the motivation for a view
along the lines of Stalnaker’s. However, I implement the view very differently from Stalnaker, as I discuss
in the conclusion.
2
2
The Canon and Its Discontents
I begin by sketching the orthodox theory of discourse dynamics, the canonical account of
epistemic modals, and the dilemma that has led some to reject both theories.
According to the orthodox semantic and pragmatic framework for analyzing discourse
dynamics, as I will understand it, a well-formed declarative sentence asserted at a context of
utterance (the concrete speech situation) has as its intension a proposition, which is a set of
possible worlds.4 We adopt a pragmatic framework according to which conversations track
the mutually recognized commitments of their interlocutors with a context set—the set of
possible worlds compatible with what is mutually accepted in the context at a given time.
The assertion of a sentence is a proposal to make the context set entail its intension. If the
proposal in question is accepted, then the context set is updated by intersecting the prior
context set with the intension of the sentence (except when this intersection would make
the subsequent context set empty, in which case the prior context set is first adjusted in a
minimal way so as to avoid that outcome, before adding the content by intersection).5
If epistemic modal claims—well-formed declarative sentences containing bare epistemic
modals6 —fall within the bounds of the orthodoxy, it follows that an epistemic modal claim
asserted at a context has as its intension a set of possible worlds, and that the assertion
of an epistemic modal claim is a proposal to update the context set by intersecting it with
that set of possible worlds (minimally changing the prior context set, if necessary, to avoid
yielding an empty set through this process).7
The canonical theory of the meaning of epistemic modals is a specific theory about the
meaning of epistemic modal claims, developed within the orthodox framework. According
to that theory, an epistemic modal quantifies over a non-empty set of worlds provided by
4
For simplicity I will assume that the intension is a set of possible worlds, but we could also adopt the
position that the intension is something more fine-grained which determines a set of possible worlds as its
truth conditional content.
5
This follows Stalnaker (1978), modulo the last, parenthetical clause, which I believe constitutes a conservative generalization of this approach to non-monotonic discourse situations. It is a difficult question what
counts as ‘minimal’ in this context, one I will not address here; see Yablo (2011) for discussion. But the
orthodoxy needn’t take any particular stance on this.
6
I will also use ‘epistemic modal claim’ to refer to the token assertion of such a sentence.
7
Here’s a characterization of the doctrine from Yalcin (2011) (Yalcin calls the view in question—which he
rejects—‘descriptivism’):
Descriptivism. . . is a view about the informational content of an unembedded, declarative
epistemic modal sentence. It is the view that the informational content of these sentences has
the effect of dividing the space of possible ways things might be into those which conform, and
those which fail to conform, with how things are represented as being; and moreover it is the
view that the fact these sentences effect such a division forms a crucial part of the explanation
for their communicative import. (Yalcin, 2011, p. 298)
3
context.8 More precisely, context provides a function from possible worlds to sets of worlds:
call it fc (⋅), or the modal domain function. For all worlds w, according to orthodoxy, fc (w)
represents the set of worlds epistemically accessible (in some way made clear by the context)
from w. A bare epistemic modal9 evaluated at a context c and world w quantifies over fc (w).
‘Must’ is a universal quantifier over fc (w); ‘might’ is an existential quantifier over fc (w). In
short:
(2)
Canonical Account of Epistemic Modals:
• JMight ϕKc,w = 1 iff ∃w′ ∈ fc (w) ∶ JϕKc,w = 1
′
• JMust ϕKc,w = 1 iff ∀w′ ∈ fc (w) ∶ JϕKc,w = 1
′
• fc (w) = {w′ ∶ w′ is epistemically accessible from w (in a way made clear by the context
c)}.
The principle objection to the canonical view about epistemic modals is that however we
try to flesh out the relation of epistemic accessibility that plays a crucial role in this theory,
it seems that we run into trouble.10
We can put this as a dilemma. One horn of the dilemma—solipsistic contextualism—
takes the set of epistemically accessible worlds, as determined by a speech context c, to be
the set of worlds compatible with the knowledge of the speaker of c at any world w. Thus—
using ‘ϕ’ as a sentence variable—⌜Might ϕ⌝, asserted by S, means roughly ⌜ϕ is compatible
with my evidence⌝. The problem with this approach is that it can’t make sense of the way
we disagree and agree about epistemic possibility claims. For instance, in the example we
began with, when Jack says ‘The keys might be in the garage’, Ruth can felicitously disagree
with him if she has sufficient reason to think that the keys aren’t in the garage. But this is
not well explained by solipsistic contextualism, according to which Jack is simply asserting
that it is compatible with his evidence that the keys are in the garage. Ruth isn’t disputing
that it’s compatible with Jack’s knowledge that the keys are in the garage. Yet it is still
perfectly felicitous for Ruth to reject the claim Jack makes.
The data about agreement and disagreement thus push us away from solipsistic contextualism and towards a view on which the epistemically accessible worlds from w are those
8
E.g. Moore (1962), Hacking (1967), Teller (1972), Kratzer (1977), and DeRose (1991).
From here on, assume epistemic modals are bare unless otherwise noted.
10
See Price (1983), MacFarlane (2011), and von Fintel and Gillies (2011), among others. This objection was
taken in a very different direction by MacFarlane, who argued that an extension of the dilemma I lay out
here merits relativism about epistemic modals. I will not discuss that line of argument here, partly for
reasons of space, partly because recent empirical work has undermined some of its starting assumptions:
see (Knobe and Yalcin, 2014) and (Khoo and Knobe, 2015).
9
4
compatible, not with what the speaker of c knows at w, but rather with what the group
of speakers at c know at w. Call this variant of the canon group contextualism. There are
different ways of spelling out group contextualism. On the most natural approach, the domain of quantification for epistemic modals to be the knowledge state that would result if
all the individuals in the group pooled their knowledge: in short, the strongest proposition
known by the group. (Formally: where KS,w represents the set of worlds compatible with S’s
knowledge in w, the strongest proposition known by a group I at w is defined to be ⋂ Ki,w .)
i∈I
But here we find ourselves on another horn of the dilemma. This approach makes better
sense of disagreement about ‘might’-claims, but it fails to account for how we can responsibly
assert ‘might’-claims in normal circumstances. On this approach, ⌜Might ϕ⌝ is true only if
no one in my group of interlocutors knows that ϕ is false. According to standard accounts of
the pragmatic norms governing assertion, one ought to assert a sentence ψ only if one bears
the appropriate attitude—knowledge or justified belief, and in case at least belief—towards
JψKc .11 But clearly one may assert ⌜Might ϕ⌝ while being agnostic about what information
one’s interlocutors have regarding the truth-value of ϕ: to assert ⌜Might ϕ⌝, one need only
think that it should be left open that ϕ is true (and perhaps that this fact needs to be made
salient).12 One needn’t know or even believe that one’s interlocutors do not know that ϕ is
false. (Thus in the example we began with, Jack can responsibly assert that the keys might
be in the garage even if he has no idea whether Ruth has knowledge that rules out that
possibility.)
Whichever way it tries to spell out what worlds count as epistemically accessible, then,
the canon seems to face problems. There are three possible responses to this. One response
is to take these worries to be decisive, and reject the canon, either by rejecting something
specific to the canon, or by rejecting some plank of the orthodoxy about discourse dynamics
(which the canon assumes as background). Another response is to maintain the orthodoxy
and the canon, but give up the ambition of giving a general account of what epistemic
accessibility relation delivers the domain of quantification for epistemic modals. A third
option is to hold onto the orthodoxy but to reject some part of the canon, by looking farther
afield for a characterization of the modal domain function that can capture the facts about
11
See e.g. Williamson (2000). I use JσKc to represent, where defined, the intension of ⌜σ⌝ as interpreted
at a context c for any string σ. This is shorthand for {w′ ∶ λw.JσKc,<p1 ,p2 ,...pn ,w> (w′ ) = 1}, where <
p1 , p2 , . . . pn , w > is the index of evaluation. I will often not be explicit about what the context is for a
given sentence under consideration, but will nonetheless refer to it, whatever it is, with ‘c’.
12
For simplicity in what follows I will ignore how epistemic modal claims are used to change what propositions
are contextually salient, though I think it is clear that one of the principle uses of ‘might’-claims is to make
salient the possibility that their prejacent is true. There are a number of ways of modeling this that are
compatible with my approach in this paper, e.g. by enriching the context with a partitional structure that
represents the question under discussion, as in Yalcin (2011).
5
discourse dynamics that the canon has trouble with. This is the option I’ll take.
I’ll draw two lessons from the dilemma presented just now. First, we must give an
account of what epistemic modals are used to do in a conversation that makes sense of what
we are proposing, agreeing about, and disagreeing about in negotiating about epistemic
modal claims. Second, we must give an account of the pragmatic norms governing epistemic
modal claims that explains how we are able to responsibly assert them (as well as accept
and reject them) in ordinary contexts. These two questions are obviously very closely tied
together.
3
An Observation
I’ll begin motivating my own account by sketching a response to the first of these questions.
What are epistemic modal claims used to do? Consider the example we began with.
When Jack says that the keys might be in the garage, he is proposing that he and Ruth
treat this as a live possibility. When Ruth disagrees, she proposes, by contrast, that they
rule this possibility out. When Ruth says the keys must be upstairs, she proposes that she
and Jack rule out the possibility that the keys aren’t upstairs; and this is what Jack agrees
to do if he accepts her assertion.
If we assume the orthodox framework laid out above, we can gloss this observation as
follows: epistemic modal claims are used to negotiate about what entailment and compatibility properties the context set of the conversation has. The context set, again, is a formal
tool that keeps track of a group of speakers’ joint commitments at a given juncture in the
conversation: the context set for a conversation at a given time is the set of worlds compatible with the propositions that are mutually accepted by the conversants at that time.13 A
proposition is mutually accepted by a group just in case everyone in the group accepts it,
accepts that they accept it, and so on ad infinitum (say that a proposition is presupposed
or common ground just in case it’s accepted in this way—in other words, just in case it’s
entailed by the context set).
In other words, a claim of the form ⌜Might ϕ⌝ is a proposal to ensure that the context
set is compatible with JϕKc ; and a claim of the form ⌜Must ϕ⌝ is a proposal to ensure that
the context set entails JϕKc .14
13
See Stalnaker (2002, 2014); I follow the latter formulation. Acceptance is any one of a class of propositional
attitudes that includes belief, presumption, acceptance for the sake of conversation, knowledge, etc.; see
Stalnaker (1984) for discussion.
14
This observation about ‘might’ was first made in Stalnaker (1970), I believe. The corresponding observation
about ‘must’ follows naturally if ‘must’ is the dual of ‘might’; see von Fintel and Gillies (2010) for a series
of arguments to the effect that asserting a ‘must’-claim commits a speaker to its prejacent.
6
4
Prospective Contextualism
I will take this observation about the role of epistemic modal claims in the dynamics of
typical conversations as the starting point for my view, prospective contextualism. In this
section and the next, I’ll lay out prospective contextualism and argue that it captures this
observation well, and thus gives us the right story about what epistemic modal claims are
used to do. Then I’ll show that this account also provides a good story about how we use
them to do this—about the pragmatic norms that govern the production and acceptance of
epistemic modal claims.
How can we predict that epistemic modal claims are used to negotiate about the properties of the context set? A very natural approach would be to take epistemic modals to
quantify directly over the context set of the conversation. Then a ‘might’-claim would be a
claim that its prejacent is compatible with the context set, and a ‘must’-claim would be a
claim that its prejacent is entailed by the context set.15
This view is moving in the right direction, but it makes badly wrong predictions about
truth-values. First, it makes it impossible to truly assert ⌜Must ϕ⌝ in an informative way. If
JϕKc is not already common ground in c, then ⌜Must ϕ⌝ will be false in c. So this account
predicts that ⌜Must ϕ⌝ can be truly asserted only when JϕKc is already common ground.
But this is clearly false: one can assert ⌜Must ϕ⌝ as a way of informing one’s interlocutors
of the truth of ϕ. This point alone would be sufficient reason for us to reject the view as
stated, but there’s another bad prediction that this view makes: that any ‘might’-claim that
is asserted is true. By asserting ⌜Might ϕ⌝, one makes clear that one doesn’t accept J¬ϕKc ,
and thus the context set (at the time at which one evaluates the claim) will be compatible
with JϕKc .16 The view as stated would thus, like solipsistic contextualism, struggle to make
sense of why we ever reject asserted ‘might’-claims.
We can preserve the central insight of this approach but avoid these bad results by making
the domain of quantification the prospective context set rather than the context set at the
time at which the claim in question is evaluated. We define the prospective context set as
the context set as it stands after the epistemic modal claim in question has been accepted
or rejected. This lets us avoid the prediction that ⌜Must ϕ⌝ is true only provided that JϕKc
15
Cf. Veltman (1996) for a dynamic implementation of an approach like this, and Yalcin (2007) for brief
discussion of a static view along these lines.
16
Assuming that we evaluate claims relative to a context set updated with the fact that the claim has been
made; see Stalnaker (1998) and von Fintel (2008) for discussion. Note importantly that this is not the
prospective context set in the sense in which I define it below, but rather an earlier point in the evolution
of the context set. A view according to which epistemic modals quantified over this context set would be
much simpler, insofar as it assimilated the interpretation of epistemic modal claims to the interpretation
of all other assertions; but it won’t work for the reasons given here.
7
is already common ground: rather, we predict that ⌜Must ϕ⌝ is true at c just in case JϕKc is
common ground after ⌜Must ϕ⌝ is asserted and accepted or rejected in c. Likewise, it lets
us avoid the prediction that ⌜Might ϕ⌝ is always true when asserted at c: rather, it will be
true provided that JϕKc is compatible with the common ground after ⌜Might ϕ⌝ is asserted
and accepted or rejected at c.
I will ultimately adopt something very similar to the view under consideration, and I
think it could work in its present form. But it makes somewhat unsatisfying predictions
about ‘must’-claims. In particular, it predicts that ⌜Must ϕ⌝ can be true while ϕ is false
(namely, just in case JϕKc is common ground and also false). This is a result that von Fintel
and Gillies (2010) have argued is unacceptable. The view under consideration can capture
something very close to the claim von Fintel and Gillies advocate (that ⌜Must ϕ⌝ entails
ϕ): on the view under consideration, one cannot accept ⌜Must ϕ⌝ in a context without also
accepting ϕ. It may be that that is a strong enough prediction to capture the data that von
Fintel and Gillies discuss, but there is some pressure to think it is not: if someone asserts
⌜Must ϕ⌝ at c and JϕKc turns out to be false, speakers tend to judge JMust ϕKc to have
been false as well. The data here are tricky to assess, but this is at least some reason to
think that ⌜Must ϕ⌝ really does entail ϕ. In any case, there is a relatively trivial change
we can make to the proposed semantics to predict that ⌜Must ϕ⌝ entails ϕ: we can say
that instead of quantifying over the prospective context set, epistemic modals quantify over
the prospective common knowledge of the interlocutors. We define the prospective common
knowledge in a conversation, relative to a given assertion, as the set of worlds compatible
with the propositions that are known by everyone in the conversation, known to be known,
and so on, after the assertion in question is accepted or rejected.
From a pragmatic point of view, this account will be identical to an account according
to which epistemic modals quantify over the prospective context set. This is because the
context set represents itself as representing the common knowledge of a group (even when
it doesn’t, and even when interlocutors recognize that it doesn’t). A proposition is accepted
for the sake of a conversation just in case it is being treated as known in that conversation.
This is part of what it is to accept a proposition for the sake of a conversation: as Stalnaker
puts it, ‘The common ground is what is presumed to be common knowledge’.17 And so it is
commonly accepted that JϕKc is common knowledge just in case JϕKc is commonly accepted.18
17
(Stalnaker, 2014, p. 45). This is formally suggested in the logic for common ground given in Stalnaker
(2002) by the fact that models for common ground are quasi-reflexive (i.e., if a world is accessible from
some world, then it accesses itself).
18
To guarantee this condition held, we could add an axiom to a logic that includes an operator ◻ for ‘it is
common ground that’, an operator ◊ as its dual, an operator ∎ for ‘it is common knowledge that’, and an
operator ⧫ as the dual of ∎. The axioms we’d need would be ◻ ∎ ϕ → ◻ϕ and ◻⧫ϕ → ◊ϕ.
8
Putting our view in terms of prospective common knowledge, rather than the prospective
context set, thus preserves the attractions of the latter view and also seems to make slightly
more satisfactory predictions about the truth-values of epistemic necessity claims. We can
thus characterize prospective contextualism as a theory of epistemic modals according to
which a bare epistemic modal evaluated at w quantifies over the prospective common knowledge of the speakers of c at w: in short, the modal domain function fc (⋅) takes each world
w in its domain to the prospective common knowledge of the conversation of c at w. ‘Must’
evaluated at w is a universal quantifier over fc (w); ‘might’ is an existential quantifier:
(3)
Prospective Contextualism:
• JMight ϕKc,w = 1 iff ∃w′ ∈ fc (w) ∶ JϕKc,w = 1
′
• JMust ϕKc,w = 1 iff ∀w′ ∈ fc (w) ∶ JϕKc,w = 1
′
• fc (w) = {w′ ∶ w′ is compatible with the common knowledge in the conversation of c as
it takes place at w at the time after the claim under consideration has been made and
accepted or rejected}
We will have reason to generalize this account slightly when we turn our attention to
embedded epistemic modals, where we will see that fc gets shifted in systematic ways, but
until then, I’ll let this be the official statement of the view. Note that fc will necessarily be
a partial function: I assume that it is defined only at worlds where the conversation taking
place in c is also taking place at w.
Note that I intend this theory to be interpreted squarely within the orthodox theory of
discourse dynamics. So one of the commitments of prospective contextualism is the claim
that the information an assertion of an epistemic modal claim contributes to a conversation—
the claim’s assertoric content—is just the possible worlds proposition formed by abstracting
on the world parameter of the denotation function.19 Thus the assertoric content of an
assertion of ⌜Might ϕ⌝ at a context c is the set of worlds in JMight ϕKc , and likewise the
assertoric content of an assertion of ⌜Must ϕ⌝ at a context c is the set of worlds in JMust
ϕKc , and when an epistemic modal claim is accepted at a context, the context set of that
context is updated by intersecting it with the claim’s assertoric content—with the set of
worlds which is the claim’s intension.
Is prospective contextualism a form of the canon about epistemic modals advocated by
Kratzer and others? It preserves the structure of the canon, and we could count it as a
19
In general; in cases of certain kinds of uncertainty, which won’t concern us here, the assertoric content
may be the diagonal proposition—λw.JϕKcw ,w , where cw is the concrete speech situation corresponding at
w to the actual context. See Stalnaker (1978).
9
version of the canon if we’re willing to think of the accessibility relation I’ve specified as an
epistemic accessibility relation of some sort. But in obvious ways it’s a stark departure from
the canon. I won’t worry further about how to classify it with respect to the canon; what
is of principle importance to me is that—unlike many other departures from the canon—
prospective contextualism is consistent with the orthodox theory of discourse dynamics.
5
Getting Force From Content
We are looking for an account within the orthodox framework according to which asserting
and accepting a claim of ⌜Might ϕ⌝ has the result of ensuring that the context set is compatible with JϕKc , and asserting and accepting a claim of ⌜Must ϕ⌝ has the result of ensuring
that the context set entails JϕKc . In this section I argue that prospective contextualism gives
us such an account.
For concreteness let’s look at what prospective contextualism predicts about the force of
the assertion of an epistemic possibility claim. According to prospective contextualism, a
claim of the form ⌜Might ϕ⌝ is a claim that the prospective common knowledge is compatible
with JϕKc . As I argued above, this will amount (for the purposes of assessing the force of
the claim in question) to a claim that the prospective context set is compatible with JϕKc .
According to the theory of discourse dynamics which we are assuming, accepting an assertion
of ψ updates the context set by intersecting it with JψKc . Thus if a claim of ⌜Might ϕ⌝ is
accepted, the context set will come to entail that the context set is compatible with JϕKc .
An epistemic possibility claim, on this view, is thus a proposal which, if accepted, brings
it about that the conversation’s context set (at the prospective time) entails that the conversation’s context set (at that time) is compatible with its prejacent. What does this mean?
In any conversation, it will be obvious to the conversants that the conversation is taking
place (and obvious that this is obvious to everyone), so it will be common ground that it is
taking place. It follows that each world in the common ground represents the conversation,
and thus represents the conversation’s context set. So an epistemic possibility claim, on this
view, is a proposal which, if accepted in a context c, brings it about that each world in the
context set represents the conversation of c as having a context set that is compatible with
its prejacent.
At first glance, this does not seem to be quite what we are looking for. We are looking,
rather, for an account of epistemic possibility claims according to which an epistemic possibility claim, if accepted, directly brings it about that the context set is compatible with its
prejacent; not an account on which the claim, if accepted, brings it about that the context
set entails that the context set is compatible with its prejacent. To see the problem, suppose
10
that we are looking for a certain kind of claim which, if asserted and accepted, has the result
of ensuring that the rabbit is in its box. ‘The rabbit is in its box’ is one candidate. But,
barring some extra machinery, this is clearly a bad candidate: if this claim is asserted and
accepted, then it’s ensured that the context set entails that the rabbit is in the box, but it’s
not ensured that the rabbit is in the box.
Context sets are special, though. The relationship between a context set and itself is
not like the relationship between a context set and the rabbit. If the context set entails
that the context set has a certain structural property, then in many cases, the context set
will automatically have that property. This is not magic; rather, it’s because the context
set represents what is mutually accepted, and our pragmatic framework assumes that the
underlying attitude of acceptance has a certain degree of transparency.
In particular, given the account of the structure of context sets laid out in Stalnaker
(2002), it will always be the case, first, that if it is mutually accepted that JϕKc is compatible
with what’s mutually accepted, then JϕKc is compatible with what’s mutually accepted; and,
second, that if it’s mutually accepted that it’s mutually accepted that JϕKc , then it is also
mutually accepted that JϕKc .20 Let’s use ◻ as a modal operator interpreted as ‘it is common
ground that’, and ◊ as its dual, ‘it is compatible with the common ground that’. Then
these two claims amount to the claim that the following are theorems of the logic of common
ground:
(4)
Negative Collapse: ⊧ ◻◊ϕ → ◊ϕ
and
(5)
Positive Collapse: ⊧ ◻ ◻ ϕ → ◻ϕ.
Both these claims are theorems of the logic of common ground given in Stalnaker (2002).21
20
We can make this claim a bit more general to account for epistemic modals embedded under disjunction
and conjunction. Consider a modal language that includes a modal operator ◻ interpreted ‘S presupposes
that’, where we say that S presupposes JϕKc just in case S believes JϕKc is common ground. Then, where
ψ is any purely modal sentence (any sentence such that all of the atomic sentences it contains are within
the scope of a ◻ operator in ψ) then ◻ψ → ψ will be true at the designated world of any model for speaker
presupposition. This is a property I call self-determination of speaker presupposition, and argue
for and discuss in Mandelkern (MS). Note, however, that we do not have the corresponding property if we
interpret ◻ as ‘it is common ground that’. For we can have ◻(◻η ∨ ◻χ) true at the designated world of a
model for common ground and ◻η ∨ ◻χ false at that world. However, the switch to speaker presupposition
ultimately suffices for an account of how we interpret epistemic modals.
21
Proof: That framework takes as an axiom ◻(◻ϕ → ϕ) (corresponding to quasi-reflexivity in the Kripke
model for common ground). I assume the logic for common ground is normal; then ⊢ ◻(◻ϕ → ϕ)→ (◻◻ϕ →
◻ϕ). Since ◻(◻ϕ → ϕ) is an axiom, ◻ ◻ ϕ → ◻ϕ, i.e. Positive Collapse, will thus be a theorem of our
system. Likewise the logic in question will validate Negative Collapse, thanks to the serial and transitive
constraints adopted in the framework of Stalnaker (2002). The serial constraint says that ∀w∃w′ ∶ wRw′ :
11
And both claims are fairly intuitive. Let’s think through a case in which the common ground
is constituted by belief, since such cases are typical. In that case, the first of these principles
amounts to the claim that if we believe that JϕKc is compatible with our beliefs, then JϕKc
is compatible with our beliefs. The second amounts to the claim that if we believe that we
believe JϕKc , then we believe JϕKc .22 Neither of these claims is uncontroversial as a claim
about propositional attitudes in general. There are certainly psychological phenomena that
are naturally described as cases in which someone believes she believes something, but in fact
believes the opposite. These should be captured by a semantics for ‘believes’.23 Thus the
two collapse principles should not be taken for granted without further discussion as axioms
of the logic of belief. But we should accept them as describing the attitude that underlies
common ground under the idealizing assumptions we make when doing formal pragmatics:
the transparency of the attitudes in question is a crucial assumption of the framework.
Given the validity of these two collapse principles, prospective contextualism does indeed
give us an account of epistemic modals according to which epistemic modal claims are used
to negotiate about whether the context set is compatible with (or entails) their prejacents.
Suppose a claim of the form ⌜Might ϕ⌝ is asserted at t1 and accepted at t2 . Then at t2 ,
it is common ground that JϕKc is compatible with the prospective context set. Since the
prospective context set in this case is just the context set at t2 , at t2 it is common ground
that ϕ is compatible with the context set at that time. It follows from Negative Collapse
that ϕ is compatible with the context set that time. So an epistemic possibility claim is a
that is, at every world, the common ground propositions are consistent. The transitive constraint says
that, where R is our accessibility relation, ∀w∀w′ ∀w′′ ((wRw′ ∧ w′ Rw′′ ) → wRw′′ ), and corresponds to
positive introspection for mutual acceptance. The axiom corresponding to the serial constraint is ◻ϕ → ◊ϕ.
Substituting ◊ψ for ϕ, we get ⊢ ◻◊ψ → ◊◊ψ. The axiom corresponding to the transitivity constraint is
◻ϕ → ◻ ◻ ϕ; contraposing and substituting ψ for ¬ϕ, we get ⊢ ◊◊ψ → ◊ψ. It follows by truth conditional
reasoning that ⊢ ◻◊ψ → ◊ψ.
22
Some caution is needed here. As I discuss below, there are cases in which local contexts coexist alongside
global context sets. It is not plausible that local contexts obey the same structural constraints as global
context sets; for instance, in counterfactual local contexts, the speakers themselves need not exist at the
worlds in question, and therefore obviously acceptance does not have the iterated structure required for
these collapse principles to hold. But all we need is for these principles to hold for the global context
set (and I assume that the attitudes of acceptance that underly the global context set all must have that
iterated structure—so, e.g., counterfactual supposition does not, in general, count an attitude of acceptance
for our purposes). Again, let the global context set be constituted by an attitude represented by ◻, and the
local context by an attitude represented by ∎ (e.g. counterfactual supposition), with ⧫ its dual. Then if
an epistemic modal is asserted and understood to be updating the local context, it will do so by updating
the global context set with the the proposition expressed by ◻ ∎ ϕ or ◻⧫ϕ, depending whether the modal
in question is a necessity or possibility modal. The local context, insofar as it is derivative on the global
context set, will automatically be updated with ∎ϕ, or ⧫ϕ, respectively. These inference patterns should
be carefully distinguished from the inferences from ◻ ∎ ϕ to ◻ϕ, and from ◻⧫ϕ to ◊ϕ, respectively: those
latter inferences are not in general valid.
23
One way of doing this while still validating these principles is to appeal to fragmentation; see Stalnaker
(1984).
12
proposal to ensure that the context set is compatible with its prejacent. Likewise, suppose
a claim of the form ⌜Must ϕ⌝ is asserted at t1 and accepted at t2 . Then at t2 , it is common
ground that JϕKc is common ground in the prospective context. In this case, the prospective
context is just the context at t2 , so at t2 , it is common ground that ϕ is common ground at
that time. It follows from Positive Collapse that ϕ is entailed by the context set at that
time. So an epistemic necessity claim is a proposal to ensure that the context set entails its
prejacent.
According to prospective contextualism, then, to assert an epistemic modal claim is to
make a proposal to make the context set entail (be compatible with) its prejacent; when
interlocutors discuss and dispute epistemic modal claims, they are negotiating about this
proposal. This gives us an account of the content of epistemic modals which lets us explain
what conversants do with epistemic modal claims in conversation, and in particular what
they agree and disagree about when they debate them.
We achieve this result, moreover, within the orthodox framework in which intensions are
possible-worlds propositions and updates occur by intersection (or, in cases in which such
an update would make the context set empty, by minimal adjustment of the prior context
set, and then by intersection). This shows that the orthodox framework is substantially
more flexible than it might first appear. Since according to that framework, an assertion
with content JϕKc is a proposal to intersect the context set with JϕKc , it looks prima facie
like the only semantic affect an assertion can have on the context set (ignoring pragmatic
effects connected with presupposition and implicature) is to bring it about that the context
set entails the assertion’s content. But some assertions have more structurally complex
semantic effects on the context set: namely, epistemic possibility claims bring it about
that the context set is compatible with their prejacent (but does not necessarily entail their
prejacent). The present discussion shows that we do not, however, need a new pragmatic
framework to account for this: rather, the argument of this section shows that it’s possible
to achieve this kind of structural change to a context set with the resources of the orthodox
pragmatic framework.
6
Norms of Epistemic Modal Claims
I proposed above that to address the dilemma we started with, we needed to address two
closely related questions. The first question was what epistemic modals are used to do in a
conversation. Prospective contextualism gives us an answer to this question, one that makes
sense of what we are negotiating about when we negotiate about epistemic modal claims.
The second question motivated by the dilemma that we started with was how we are
13
entitled to use epistemic modal claims—to responsibly assert, agree, and disagree about them
in ordinary contexts. In this section I’ll show that the answer prospective contextualism gives
to the first question motivates a very natural answer to the second question.
At first glance, it might seem that, while prospective contextualism gives a good answer
to the first of these questions, it in fact does very badly on the second of these questions.
According to prospective contextualism, epistemic modals quantify over the prospective
common knowledge of a conversation: but it’s fairly hard to know what will be common
knowledge in a conversation at some point in the future, and thus hard to see how speakers
could know whether a given epistemic modal claim is true or false, and thus be in a position
to make the claim in question.
This is right; but prospective contextualism has the resources to explain why, in spite
of the difficulty of knowing what the prospective common knowledge in a conversation will
amount to, epistemic modals can be responsibly asserted and accepted in ordinary circumstances. I will argue that there is no norm of the form: assert (accept) an epistemic modal
claim only if you know (justifiably believe) its content. Rather, it follows from the form of the
prospective contextualist semantics that epistemic modal claims are governed by a different
normative regime, one that makes sense of how we responsibly assert epistemic modals in
normal circumstances.
On the account I have given, in paradigm cases, the truth of an asserted epistemic modal
claim depends in part on whether it is asserted and accepted. Suppose that at the initial
context c1 , JϕKc1 is not common ground. Suppose that Rose is deciding whether to assert
⌜Must ϕ⌝ at c1 or to say nothing. Let c2 represent the prospective context: the context
which differs from the prior context only insofar as the claim proposed at the prior context
is accepted or rejected at the prospective context. Then JϕKc1 is common ground at the
prospective context c2 only if Rose asserts ⌜Must ϕ⌝ at c1 , and thus ⌜Must ϕ⌝ is true at
< c1 , wc > (where wc is the world of the context) only if Rose asserts ⌜Must ϕ⌝ at c1 . Whether
Sam’s assertion is true, then, depends in part on whether she makes it; in particular, in this
situation, if it is true, then it follows that Rose did make the assertion. Very similar reasoning
will apply not only to Rose, but also to her interlocutors; suppose that Rose does assert ⌜Must
ϕ⌝. Whether her assertion is true will depend in part on whether her interlocutors accept it;
in particular, if Sam’s assertion is true, it follows that her interlocutors accepted it.
Similar considerations apply to ‘might’-claims: whether a ‘might’-claim is true in a given
context will often depend in part on whether it is asserted and subsequently accepted at that
context. There will be different permutations of these cases, depending on the prior status
of the epistemic modal’s prejacent, but the basic idea is robust: according to prospective
contextualism, an epistemic modal claim’s truth depends in part on whether it is asserted
14
and subsequently accepted.
Now we’re in a position to see why it follows directly from the form of prospective
contextualism that there can be pragmatic norm that says one ought to assert (accept) an
epistemic modal claim only if one knows or justifiably believes it. It is a general principle of
action theory that, for propositions η and χ, if η entails χ, then in deciding whether to act
to bring about χ, one cannot take into account the truth of η: one cannot deliberate about
whether to do χ on the basis of a premise that entails that one in fact does χ.24
In deliberating about whether to assert (or accept) a sentence ψ, then, one cannot take
into account anything that entails that one in fact asserts (or accepts) ψ. In particular, where
ψ has the form of an epistemic modal claim, one cannot take into account any propositions
whose truth entails that the claim itself is asserted (accepted). But as we saw above, JψKc
often will entail that ψ was asserted and accepted, when ψ is an epistemic modal claim. To
advert to the truth of ψ in deliberating about whether to assert (accept) ψ, then, would take
for granted that ψ will be asserted and accepted: in other words, it would take for granted
the very thing that one is deliberating about.
If epistemic modals have the content ascribed to them by prospective contextualism, it
will thus not in general be possible to deliberate about whether to assert (accept) an epistemic
modal claim on the basis of whether it’s true. Assuming that norms should not ask us to do
the impossible, it follows from prospective contextualism that there is no norm of the form:
assert (accept) an epistemic modal claim only if you know (justifiably believe) its content.25
Thus the particular kind of difficulty involved in knowing the content of epistemic modal
claims, on this account, forces us to posit that the norms governing them do not advert to
agents’ epistemic relations to the claims’ contents.
What do we have instead? There are, of course, many moves in a language game that
are governed by norms which do not mention the truth-value of the move’s content. In some
cases (but not all) this is because the move’s content does not have a truth-value. Consider
in particular proposals like ‘Let’s go to the beach!’ A proposal like this is governed by a
24
This idea goes back at least to Ramsey (1978); for recent discussion, and another application to the
philosophy of language, see Ninan (2005).
25
There is some subtlety here: I am certainly not claiming that, in considering whether to assert ⌜Might ϕ⌝
(⌜Must ϕ⌝), one cannot or should not take into account the fact that one knows that ϕ might (must) be
true. Indeed, this is probably a necessary condition on responsible assertions of epistemic modal claims, as
Moore-paradoxical sentences involving epistemic modals show. But on the account of embedded epistemic
modals I develop below, JS knows must ϕKc,w can be true, while at S’s context of utterance, k, JMust
ϕKk,w =0, since as I discuss below, the truth conditions of the former depend just on S’s knowledge state,
while the truth conditions of the latter depend also on what is common knowledge at k; likewise for ‘might’.
Thus I am arguing that assertions of epistemic modals are not governed by a norm of the form ⌜S should
assert JϕKc only if she knows JϕKc ⌝, even if they are governed by a norm of the form ⌜S should assert JϕKc
only if JS knows ϕKc,w = 1⌝.
15
norm along the lines of: propose ϕ only if you know (reasonably believe) that ϕ makes a
good proposal, and clearly not by a norm which adverts to the truth of ‘Let’s go to the
beach!’.26
In the last two sections, I argued that epistemic modal claims are proposals about how
the context set should evolve. They are proposals about how to proceed with respect to a
certain problem: namely, what should count as agreed upon, and what should be left open, in
a given conversation. And so we expect that the norm for asserting and accepting epistemic
modal claims should be a version of the norm that, I take it, is in place for proposals in
general: namely, assert an epistemic modal claim only if you know (reasonably believe) it
is a good proposal for how to proceed with the problem of coordinating the context set;
likewise, accept an epistemic modal claim only if you know (reasonably believe) it is a good
proposal about how to update the context set. Since none of these norms adverts to the
truth of an epistemic modal claim, it will not be a problem that according to prospective
contextualism, the truth of an epistemic modal claim is very hard to know: the only special
doxastic relationship one needs to have towards the content of an epistemic modal claim is
reasonably believing that it describes a way the world should be, not necessarily a way that
the world is.
Indeed, according to the orthodox pragmatic framework, all assertions are proposals, and
so it’s natural to think that all assertions are governed by the norm for proposals: make the
assertion only if you know or reasonably believe that the proposal it makes is a good one.
Generally speaking, the aim of the conversation will be to make the context set maximally
informative regarding the issues that the conversants evince interest in. The conversants will
thus generally aim to make the context set accurately represent the information accessible to
them. This dictates that broadly epistemological considerations will guide the making and
acceptance of claims of all sorts. When a proposal is an update whose felicity depends only
on certain facts that are settled independently of whether the update is made, then presumably there is a specific norm that comes into play, something like: assert ϕ only if you know
JϕKc , or: assert ϕ only if you reasonably believe JϕKc . But when, for whatever reasons, the
proposal’s felicity does not depend (only) on certain facts that are settled independently of
whether the proposal is made and accepted, then one will have to advert to other considerations to determine its felicity. This is the case for a large class of proposals, epistemic modal
claims among them. But epistemic considerations will still in general guide the production
and acceptance of epistemic modal claims: typically, one should assert (accept) an epistemic
26
There are also various more objective versions of the norm, e.g. ⌜just in case ϕ makes a good proposal⌝. I
don’t mean to commit to any particular stance on the question of the correct form of the norms in question
here.
16
modal claim just in case one knows, or has good reason to believe, that the proposal the
claim makes will put the group in the best epistemic situation possible given the information
available to them at a given time. So, for instance, in the exchange we began with, when
Jack says that the keys might be in the garage, his proposal is presumably motivated by a
reasonable belief that the group should leave that option open. When Ruth disputes Jack’s
statement, she is proposing the opposite: that the group should, in fact, rule out that option,
and she better have good reason to think that the keys aren’t in the garage.
Of course, not every conversational move is made with the goal of maximizing mutually
accessible information. Consider this case, from Yablo (2011):27
(6)
a.
b.
Steve: So, are you voting for Kucinich?
Erin: I might be, I might not be.
Erin knows who she’s voting for; she just thinks it’s none of Steve’s business. Here, Erin
simply does not want the context set to reflect all the knowledge available to her; so she has
(non-epistemological) reason to leave it open in the context set who she is voting for, and
thus to assert her epistemic modal claim. This is a perfectly acceptable move: we have many
reasons to shape the context set in a certain way, not all of them epistemological. Epistemic
modal claims vary in their aims, and thus in the local norms that govern them, but they are
all proposals about how the context set ought to evolve, and thus all governed by a general
norm which governs proposals of that kind.
This discussion should help clarify something that can look misleading about the prospective contextualist framework. I have argued that the truth conditions for epistemic modals
are about the future. But it doesn’t follow that epistemic modals are predictions about the
future. That would only follow if there were some norm in place of the form: assert an
epistemic modal claim only if you believe/know it’s true. Since there is no norm of this
form, there is nothing prediction-like about epistemic modal claims; you can assert an epistemic modal even if you don’t think it will be accepted, and even if you think that it will be
rejected—provided, of course, that you think it makes a good proposal.
This completes my response to the dilemma laid out in the second section. Epistemic
modal claims are proposals about how the context set should evolve. Agents ought to make
proposals only if they have sufficient reason to think the proposals are good ones. In some
cases of assertion, this norm may entail a more determinate norm, e.g., that agents ought
to assert ϕ only if they know that ϕ is true. But, as we have seen, this is not the case
for epistemic modals; they are governed only by the more general norm. We thus have an
explanation of both what epistemic modal claims are used to do in a context, and of how
27
Cf. similar cases in von Fintel and Gillies (2008).
17
epistemic modal claims can be responsibly asserted and accepted in ordinary situations.
Note that it follows from this discussion that we might want to taxonomize speech acts
in different, incompatible ways. On the one hand, we could classify as an assertion any conversational move whose typical effect is to update the context set by intersecting it with a
content determined conventionally as a function of the specific features of the conversational
move. On this approach, assertions contrast with distinct kinds of speech acts like interrogatives and (controversially) imperatives. On the other hand, we could treat the class of
assertions as the class of speech acts governed by a certain set of norms—a set that at least
includes a norm of the form: assert ϕ only if you know (reasonably believe) ϕ. According
to prospective contextualism, epistemic modal claims are assertions on the first taxonomical
approach, but not on the second. These two taxonomic approaches thus diverge in important ways. Pragmatic theory should concern itself both with the question of how to formally
model discourse dynamics, as well as with the question of what norms govern those dynamics; both taxonomic approaches, then, are crucial to pragmatic theory. I will continue to talk
about epistemic modal claims as assertions, because I want to emphasize that we can model
epistemic modal claims as proposing intersective updates. But I also want to emphasize here
that, in spite of that, epistemic modal claims differ in important and systematic ways from
many non-modal claims with regard to the norms governing them.
7
Prospective Performatives
According to prospective contextualism, an assertion of an epistemic modal is a proposal
about how the context set should evolve that is made by asserting a claim about how the
context set will evolve. This may seem implausibly roundabout. It particular, it may look
as though the truth conditions in my account are idle wheels, and that a more perspicuous
formulation would not posit ordinary truth conditions for epistemic modal claims at all,
instead finding another, more direct way to predict that they are proposals about how the
context should evolve.
One reason to stick with the proposal I’ve given here is that it lets us preserve a uniform
formal pragmatics for modal and non-modal claims. I return to this point in the conclusion.
Another reason, which I discuss here, is that the method which I have argued epistemic
modals exploit—making a proposal about how to set some contextual parameter by making
a truth conditional claim about the future—is, in fact, widespread in natural language. In
this section, I’ll briefly explore some of these other constructions. The local point of doing
so is to show that this phenomenon is widespread, and so that it is plausible to suggest
that epistemic modals function this way, too. These constructions also strike me worthy of
18
investigation in their own right, and the larger aim of the present discussion is to suggest
one way of analyzing them.
I’ll call the class of constructions in question—the class to which I’m proposing epistemic
modals belong—‘prospective performatives’. Like epistemic modals (on my account), these
constructions have truth conditions that are about a future state of affairs, but a future
state of affairs that depends, in part or in whole, on whether the construction in question
is asserted and accepted. It follows, reasoning as in the last section, that the norms that
govern these constructions are not the specific epistemic norms that govern typical factual
assertions, but rather the more general norms that govern proposals in general.
Here’s an example of a pair of prospective performatives. Suppose that John is supposed
to be cleaning the rabbit cage, that Mark is partly in charge of John:
(7)
a.
b.
c.
Mark: What will you be doing this afternoon?
John: I’ll be playing outside.
Mark: Oh no you won’t be! You’ll be cleaning the rabbit cage.
When John says, ‘I’ll be playing outside’, he makes a claim that has straightforward truth
conditions: his claim is true in every scenario in which he is playing outside in a (contextually
specified) future time. But John’s aim in making this claim is not only to inform Mark about
the future, but also to make a proposal. Mark is partly in charge of determining what’s
permissible for John. If Mark accepts that John will be playing outside, then he implicitly
makes it permissible for John to be playing outside. Part of John’s aim in asserting that he
is going to be playing outside is thus to change the permissibility facts by making a claim
about the future. John’s assertion is perfectly responsible, even if he doesn’t know that he’s
going to go outside—indeed, even if he thinks that his bid will probably fail—as long as he
has good reason to think that the proposal he makes is a good one.
Mark’s response, likewise, is a statement about the future, with straightforward truth
conditions. But, again, part of Mark’s aim in asserting it is not to inform John about the
future, but to establish certain normative facts about what it is permissible for John to
do—to establish those facts by making a claim about the future. And the main normative
constraint governing Mark’s assertion is that it should make a proposal about the normative
facts that Mark has reason to think is a good one. Note that Mark’s assertion might be
a reasonable thing even if Mark is unsure if it is true. Suppose that John is under the
normative sway not only of Mark, but also of another authority, Jake. Mark knows that
John might appeal his decision to Jake, and that Jake might well overturn it. What Mark
seeks to do in asserting (7-c) is not to inform John about the future, but rather to establish
certain norms. And again, whether or not this is a responsible, permissible move in the
19
conversation depends not on whether Mark knows that John won’t go outside, but rather
on whether Mark reasonably thinks that John ought not to go outside.
Compare
(8)
You are going to open that door.
given as an order. The structure here is, again, that of a statement whose truth-value
is settled in part by whether it is made and accepted, and thus whose assertability and
acceptability depend primarily on considerations about whether the proposal it makes is a
good one, rather than on considerations of whether or not it is known to be true. We can
give an analogous treatment to a case discussed by Anscombe: a doctor tells a patient, in
the presence of a nurse,
(9)
The nurse will now take you to the operating theater. (Anscombe, 1963)
In this case the doctor clearly has two aims: first, to communicate information to the patient;
second, to make a request of the nurse. She achieves both these effects with one assertion,
an assertion whose truth conditions are about a future state of affairs which depends on the
uptake and behavior of her interlocutors.28 Or consider a sign next to a train track that says
(10)
The front of the train will stop here.29
This statement serves both to instruct the conductor where to stop, as well as to inform
passengers where the train will stop.
These examples attest exactly the structure that I am attributing to epistemic modal
claims: they make proposals about how to change a feature of the context (in these cases,
normative features) by making a claim with prospective truth conditions, a claim which, if
accepted, ensures that the change in question takes place. In these cases, this structure is
manifest in the compositional semantics of the constructions in question; the treatment I
have proposed is by far the most natural and simplest treatment.
On one hand, this shows that the structure I am attributing to epistemic modal claims
is attested in natural language; I take that to substantially increase the plausibility of the
proposal I am making about how epistemic modal claims work. On the other hand, precisely
the fact that this structure is so clearly manifested in the case of the examples considered
here may be reason to think that epistemic modals do not share this structure: after all,
epistemic modals are not explicitly marked as being prospective, and indeed, nothing about
28
29
See Velleman (1989) for an extended discussion of this case.
Thanks to Bob Stalnaker for this example.
20
their use forces us to attribute truth conditions to them at all. So do the examples given here
tell in favor of, or against, the thesis that epistemic modals are prospective performatives?
Let me come back to this question in a moment, after considering another class of constructions: performatives of the sort discussed in Austin (1979): ‘I hereby name you. . . ’,
‘It is hereby agreed that. . . ’, ‘I promise you. . . ’, and so on. These constructions are clearly
constructed compositionally out of lexical items which, when combined according to certain
syntactic and semantic constraints which seem to be met in these cases, produce clauses
whose intension is a set of possible worlds. On the other hand, assertions like ‘I promise
you. . . ’ seem to be doing something different from assertions like ‘It’s raining outside’. How
can we capture this within a compositional semantic framework? One approach is to postulate that at the level of logical form, performatives like this are prefixed by performative
operators which modulate their particular force.30 But it seems to me that the treatment
of these constructions as prospective performatives provides a more economical and simpler
approach. Let’s say, for instance, that the context set keeps track of what promises are made.
Then if we interpret ‘I promise you. . . ’ in a prospective way, we get a treatment which can
account both for the fact that this construction is a sentence whose intension at a context
is a set of possible worlds, and for the fact that this construction is characteristically used
to do something other than inform one’s interlocutors about the way the world is. In other
words, let JS promises x to ϕKc,w = 1 just in case it is common knowledge at the prospective
context that S is committed to x to bring about JϕKc . Then ⌜I hereby promise x to ϕ⌝ is an
assertion that has ordinary truth conditions—it’s true just in case I am publicly committed
to ϕ at the time at which it’s accepted—but it is an assertion whose content is made true
just by being asserted and acknowledged, and which has the force of changing facts about
who is committed to what, rather than conveying information about who, antecedently, is
committed to what.31 We can give similar treatments to other performatives of this kind.
All these examples vary with respect to certain normative facts: for instance, whether my
proposal that you be married, or that you close the door, is accepted depends partly on
whether I am invested with the right normative power to marry you, or get you to close the
door. But those are separable features of the examples, and parallels to the case of epistemic
modal claims are easy to draw. Indeed, we could think of epistemic modal claims as pro30
As in Lewis (1970), a strategy that has been pursued in the linguistics literature in recent years; see e.g.
Szabolcsi (1982), Krifka (2014).
31
Performatives of this kind seem to require less acceptance on the part of one’s interlocutors than do
performatives of the kind considered above, though it certainly requires minimal acknowledgement, and it
does seem possible for me to make a claim like this false by retorting that you do not have the authority
to make a promise in this case, or name the child, or whatever. These seem to be separable features of
the cases, though, features having to do with the specific, varying normative structures of the cases: the
underlying structure is analogous to the cases considered above.
21
posals that, if accepted, change the context set of the conversation thanks to the normative
power invested in the conversants to determine the nature of the context set.
In these cases, even though the prospective structure is not compositionally manifest,
I think it is very natural to treat these constructions as prospective performatives, in the
framework I’ve laid out. We can make sense of this from a compositional point of view in
a variety of ways. In general, the present tense can sometimes be interpreted as extending
into the future relative to the time of assertion.32 I propose that we take this option, for
pragmatic reasons, when interpreting assertions which are about features of the context
which depend wholly or in part on the uptake of the audience; such an interpretive strategy
allows us to use these assertions to negotiate about those features, rather than simply make
claims about how things stand with respect to them.
This brings us back to the question of whether we should treat epistemic modals as
prospective performatives. I take it that there is decisive reason to treat some constructions—
those we considered at the beginning of this section—as prospective performatives. I think
there is also strong reason, on the grounds of simplicity, to treat performatives like ‘I
promise. . . ’ as prospective performatives. Insofar as this strategy is effective for this wide
class of natural language constructions, there is reason to see how far it applies beyond that
class. Uniformity in semantic and pragmatic explanations, to the degree that it’s empirically
adequate, is pro tanto preferable on the grounds of simplicity. So a treatment of epistemic
modals as prospective performatives is worth pursuing: if this approach is empirically adequate, then whether it is ultimately correct will depend on overall theoretical considerations
of comparative simplicity and explanatory power. And there is at least some reason to think
that prospective contextualism does well on these criteria, since treating epistemic modals
as prospective performatives lets us explain their behavior in the same way we explain the
behavior of a wide class of other natural language constructions, within the context of the
orthodox truth conditional semantic and pragmatic framework.
8
Further Extensions
In the last section I argued that we find the structure of prospective contextualism recapitulated for claims that, like epistemic modals, have a certain kind of connection to what is
common ground. In this section, I show how to further extend this framework to account
for probability modals and for deontic modals.
Probability modals are modals of the form ⌜Probably ϕ⌝, ⌜There’s a good chance that ϕ⌝,
and so on. Extending our framework to capture these modals is fairly straightforward. We
32
E.g. Ogihara (2007, 402-3) notes ‘that the present tense morpheme can describe a current situation or a
future situation, whereas the past tense can only describe a past situation’.
22
need, first, a measure structure33 on the context set which is jointly accessible in the same
way the context set itself is. It doesn’t matter for our purposes what form the structure
takes;34 as long as this measure structure obeys an analogous collapse mechanism to the one
we discussed above for common ground, we can give an analogous semantics to the one we
gave for ‘might’ and ‘must’. Say that JProbably ϕKc,w = 1 just in case the measure structure of
the prospective context makes JϕKc,w sufficiently likely. Provided that ‘the measure structure
of the context set makes JϕKc,w likely’ follows from ‘it is common ground that the measure
structure of the context set makes JϕKc,w likely’,35 an assertion of ⌜Probably ϕ⌝ will have the
force of a proposal to adopt a contextual measure structure that makes JϕKc,w likely. I take
it this is exactly the force we should expect for an epistemic modal of this form: this account
makes sense of the phenomena and norms of production and uptake of claims of this form
in precisely the same way the parallel account made sense of this for ‘might’ and ‘must’.
And a collapse principle of this kind is very natural: part of what it means for a measure
structure Γ to be the measure structure of a context is, plausibly, for it to be the case that
if it’s commonly accepted that Γ is the measure structure of the context, then Γ is indeed
the measure structure of the context. We can extend this account straightforwardly to other
epistemic modals of intermediate grade.
We can make very similar moves for deontic modals.36 According to the standard theory
about deontic modals, a deontic modal claim says that some proposition JϕKc is required by
(or compatible with) some salient set of deontic commitments.37 Intuitively, deontic modal
claims are used to negotiate what those commitments are, just as epistemic modal claims
are used to negotiate what is compatible with (or entailed by) the context set. But it’s not
obvious how to get this result on the orthodox approach. We face a parallel dilemma to the
dilemma faced for epistemic modals: whose deontic commitments are in question? A moral
realist might be tempted to respond that it’s the real deontic commitments, whatever those
amount to. But this response—beyond being committal in a way we might want to avoid—
doesn’t obviously work for many uses of deontic modals, e.g. in conversations about where to
go for dinner or what shirt to wear. But the commitments in question can’t be the speaker ’s
deontic commitments, for in that case we couldn’t account for the way we argue and disagree
33
Or measure structures—I assume for simplicity there’s just one, but this assumption could easily be relaxed.
The two most prominent contenders are: a normality preorder on worlds, following, e.g., Kratzer (1981);
or a probability measure on a σ-algebra of the context set, following e.g. Yalcin (2012a). See Holliday and
Thomas Icard (2013) for discussion.
35
In symbols: where ◻ is the common ground operator and △ ϕ says that the context’s measure structure
makes ϕ probable, we need ⊧@ ◻ △ ϕ →△ ϕ to be true (i.e., ◻ △ ϕ →△ ϕ should be true at the designated
world of every model structure for common ground).
36
Cf. Stalnaker (2014).
37
I’ll leave it open just what a set of deontic commitments amounts to; most simply we could think of it as
a set of propositions.
34
23
about deontic modal claims. And, for a variety of reasons, it’s equally hard to make sense
of how we use deontic modals if it’s the group’s deontic commitments: it is not obvious,
first of all, how to spell out that notion; second, this approach, like the parallel approach
for epistemic modals, makes it hard to explain how interlocutors can know enough to make
deontic permission claims. Prospective contextualism can resolve this dilemma in the same
way it resolves the dilemma for epistemic modals. Deontic modal claims are proposals about
what deontic commitments the group ought to treat as commonly accepted.38 We derive
this force by giving deontic modals truth conditions as follows: JMustd ϕKc,w = 1 just in
case, at the prospective context, JϕKc will be entailed by the deontic commitments of the
group; JMayd ϕKc,w = 1 just in case, at the prospective context, JϕKc will be compatible
with the deontic commitments of the group. Provided that we have collapse principles
of the form ‘If it’s common ground that our deontic commitments require JϕKc , then our
deontic commitments require JϕKc ’, and ‘If it’s common ground that JϕKc is compatible with
our deontic commitments, then JϕKc is compatible with our deontic commitments’, deontic
modal claims will be claims about what deontic commitments we ought to adopt as a group.
These principles are, again, very plausible: joint deontic commitment should have roughly
analogous structure to joint acceptance. And again, the same pragmatic norms will govern
deontic modal claims: norms which require that assertion and uptake be well-advised, but
not necessarily that the claims’ content be known.
The strategy of this section is general, and can be further applied in fairly straightforward
ways. Where the truth of a certain kind of construction depends in part on whether it
is asserted and accepted, the framework of prospective contextualism can often provide a
simple account of that class of constructions. As we have seen, this is a fruitful approach
for understanding various kinds of modal claims; more obviously performative claims of the
kind I discussed in the last section; as well, perhaps, for understanding further phenomena
that have to do with negotiating contextual parameters, like resolving vagueness39 or local
lexical questions.
9
Embeddings
I’ve defended a semantics—and a companion pragmatics—for bare epistemic modals. I’ll
close by briefly addressing the question of what predictions this account makes about embedded epistemic modals. On the account I’ve given, the domain of quantification for a bare
38
In the case of epistemic modals, we built in some objectivity to the semantics by making the domain of
quantification the prospective common knowledge, rather than the prospective context set. A similar move
could be made in this case by a moral realist in at least some cases.
39
Cf. Barker (2002, 2013).
24
epistemic modal is determined by the prospective context set. When epistemic modals are
embedded, the embedding operator typically introduces a local context. I propose that in
these cases the modal domain function shifts in a predictable way: it is determined by the
local context, rather than the global context set.
Local contexts play a variety of roles in linguistic theory. For instance, local contexts
are often discussed in connection with presupposition projection: an operator introduces a
local context which comprises the set of worlds relative to which the presuppositions of the
embedded clause must be satisfied for the claim as a whole to be felicitous.40 Local contexts
may also play a role in the interpretation of some of the items embedded under the operator
that determines that local context, and I suggest that this is what happens when epistemic
modals are embedded. Let us ignore for the moment epistemic modals embedded under
attitude verbs; I’ll return to those at the end of this section. I propose that in other cases,
the modal domain function for an epistemic modal asserted at c takes each world w to the
prospective local context in the conversation of c as it takes place at w; when the modal is
unembedded, we assume that the local context is equivalent to the common knowledge.41
The approach pursued here is an extension of accounts in Kratzer (1991), Yalcin (2007),
Klinedinst and Rothschild (2012), Ninan (2014), and as cited below, among others. The
main innovation of the approach I take is adding prospectivity into the picture. This move is
attractive in embedded cases for exactly the same reasons it’s attractive in bare cases: it lets
us explain how embedded epistemic modals are responsibly used to negotiate about features
of both global and local context sets. Although I won’t go through the arguments in detail,
similar dilemmas show up for embedded epistemic modals as those we discussed above for
bare epistemic modals, and prospective contextualism provides a solution in similar ways.
Beyond this innovation, my proposals follow extant proposals in the literature fairly
closely. I leave many questions open in what follows, and it may be that some of the details
of what follows are not quite right. The main point of this section is to show that, however we
ultimately work out the details of the operators discussed below, prospective contextualism
will complement those accounts well.
40
Also discussed as derived or subordinate contexts; see e.g. Stalnaker (1970), Karttunen (1974), Heim (1982,
1983, 1992), Schlenker (2010). Note that a context, as we have been using a word, is a concrete speech
situation, and as such local contexts aren’t contexts in our terminology. I stick with this terminology
because it is standard.
41
One general objection that some have had to proposals involving local contexts is that they are ad hoc. But
Schlenker (2008, 2010) has shown how to systematically predict what local context a given operator will
introduce from a simple generalization. I will not go through the details of his account, but I mention this
because it’s important to know that the various accounts sketched below follow from a simple generalization,
plus truth conditional semantics for the operators we’re interested in, and so are not as ad hoc as they
otherwise might look.
25
9.1
Disjunction
Consider:
(11)
Either Bob is in his office, or he must be in the pub.
‘He must be in the pub’ as used in (11) is clearly not a proposal to make the context set
entail that Bob is in the pub. It rather seems to be a proposal to conclude that Bob is in
the pub provided we learn that he is not in his office. This matches the observation that the
local context introduced for the right disjunct of ‘or’ is the intersection of the global context
set with the negation of the left disjunct. To model this, let’s add a parameter to the index
of evaluation that represents the local context set.42 Let s represent the local context for
a given expression at a context c; when the expression is unembedded, we assume s is just
the global common knowledge of c. I omit reference to c in what follows for simplicity of
presentation. Then
(12)
s ,w
Or: Jϕ or ψKs,w = 1 iff JϕKs,w = 1 or JψKs∩J¬ϕK
= 1.
Then the generalization sketched above for predicting the shifted modal domain function,
fc′ , will predict that the modal domain function for epistemic modals embedded in a right
disjunct is the smallest function that meets the following condition:
(13)
∀w ∶ fc′ (w) = fc (w) ∩ J¬ϕKs
(I continue to use fc to represent the unshifted modal domain function which takes each
world to the prospective common knowledge at the conversation as it takes place at that
world.) And so, for instance, Jϕ or must ψKs,w = 1 just in case JϕKs is true at w or JψKs is
entailed by the intersection of prospective common knowledge at w with J¬ϕKs . Note that
since local contexts depend on mutual acceptance in a parallel way to global contexts,43
it will follow that if it is common ground that the local context has a certain entailment,
compatibility, or measure theoretic property, then the local context has that property. So
embedded epistemic modals can be used to change the properties of their local contexts
in just the same way that bare epistemic modals can be used to change the properties of
their global context sets. It is here that the prospectivity plays a crucial role, just as for
42
I now use JσKs to represent the intension of ⌜σ⌝ at local context s. Note, importantly, that just because the
local context set is a parameter of the index of evaluation, it does not follow that the assertoric content of
a sentence is a set of local context-world pairs. On the orthodox account, the assertoric content is always
a set of worlds; the identity of the local context set parameter is always fully determined by the context
of utterance.
43
See Footnote 22 for discussion.
26
bare epistemic modals: the local context in question need not already have the property
being ascribed to it for an embedded epistemic modal to come out true; rather, the claim
containing an embedded epistemic modal is a proposal which, if accepted, ensures that the
local context has the property in question.
9.2
Conjunction
‘And’ is commonly taken to introduce a local context for the right conjunct that is the
intersection of the global common knowledge with the left conjunct:
(14)
s ,w
And: Jϕ and ψKs,w = 1 iff JϕKs,w = 1 and JψKs∩JϕK
=1
The shifted modal domain function will thus be the smallest function meeting the condition:
(15)
∀w ∶ fc′ (w) = fc (w) ∩ JϕKs
This approach predicts that an epistemic modal embedded in a right conjunct is a proposal
to make the global context updated with the left conjunct have the property attributed to
it by the epistemic modal. This seems to be the right prediction:
(16)
Arden is at home, and Jack must be, too.
It’s a bit hard to test this prediction, since conjunctions are naturally heard as successive
assertions, but an example from Klinedinst and Rothschild (2012) makes clear that ‘and’ is
shifting a parameter that ‘must’ is sensitive to:
(17)
Either John is in Taiwan and Tim must be too, or John is in Haiti.
The ‘must’ in (17) clearly quantifies over a shifted parameter—the prospective context set
updated with John is in Taiwan—and not the global (prospective) context set.
The present proposal also immediately predicts the observations of Yalcin (2007) that ⌜ϕ
and might not ϕ⌝ is always infelicitous, even when embedded, as in
(18)
It’s raining and it might not be raining.
and
(19)
Suppose it’s raining and it might not be raining.
If the right conjunct is evaluated relative to a modal domain function whose image always
entails the left conjunct, as posited here, embeddings of sentences of this form will always
27
be false or infelicitous.44
9.3
Indicative Conditionals
Some indicative conditionals have overt epistemic modals in the consequent; for those conditionals, our semantics will be as follows, letting ‘em’ stand for any epistemic modal:
(20)
s ,w
Indicative Conditionals: JIf ϕ then em ψKs,w = 1 iff Jem ψKs∩JϕK
=1
This suggests that the shifted modal domain function is defined as the smallest function
(21)
∀w ∶ fc′ (w) = fc (w) ∩ JϕKs .45
Consider:
(22)
If Bob isn’t in his office, he must be in the pub.
On the proposed account, this is a proposal to make the local context—the intersection of
the global context set with the proposition in the antecedent—entail the consequent. In
other words, it is a proposal to conclude that Bob is in the pub, assuming he isn’t in his
office. If accepted, it encodes a constraint on the evolution of context sets: we agree that,
should we come to presuppose that Bob isn’t in his office, we will also come to presuppose
that he is in the pub.46 This seems like a good prediction, one which makes sense of what
we negotiate about with indicative conditionals and how we can responsibly produce them.
44
The nature of the infelicity will differ according to the environment in which we find conjuncts like this,
of course, and in the case of certain embeddings of this conjunction will be due to the failure of the
requirement—presumably a presupposition—that epistemic modals have non-empty domains of quantification. For different reasons, ⌜Might not ϕ and ϕ⌝ will be predicted to be infelicitous under attitude verbs,
though it is predicted to be possible in the antecedent of conditionals, though perhaps infelicitous on the
grounds that the first conjunct will contribute nothing to the conditional, and in a sense will be washed
out by the second conjunct. I’m not sure if this is the right prediction; if not then we can modify the
semantics for indicatives given here to make it closer to Yalcin’s and predict that conjunctions like this
will be infelicitous regardless of the order of the conjuncts. See Dorr and Hawthorne (2013) for discussion.
45
On this semantics, as on Yalcin’s, a conditional antecedent of the form ⌜if it might be that ϕ . . .⌝ contributes
a trivial condition on the evaluation of the consequent, a condition which is met by all worlds or none;
likewise for ⌜if it must be that ϕ . . .⌝. I’m not sure whether this is a good prediction or a bad one. If
indicative conditionals presuppose that their consequents are evaluated relative to a shifted context that is
non-empty, then asserting a conditional with one of these antecedents could have an effect on the context:
ensuring that the prospective context set is compatible with (entails) JϕKs . Then ⌜If it might be that
ϕ, then ψ⌝ would be equivalent to ⌜It might be that ϕ, so ψ⌝. This seems plausible. I think that often
when conditionals of these forms are asserted, there is an explicit or implicit relativization to a body of
information, or else the ‘might’ is intended in an objective or metaphysical sense; in those cases, of course,
there will be no threat of triviality and these comments will not apply.
46
And in general, asserting a conditional of the form ⌜If ϕ then must ψ⌝ has the same affect as asserting the
material conditional ϕ → ψ; see Stalnaker (1975) for discussion.
28
When there’s no overt epistemic modal in the consequent of the conditional, I assume the
conditional encodes a modal of some kind which is systematically related to the prospective
local context supplied by the antecedent. I won’t take a stand on how exactly this works;
we could follow Kratzer and others in assuming that these conditionals have covert or covert
necessity modals.47 We could also maintain that indicative conditionals encode an operator
of the kind posited by Stalnaker (1968), determined by the prospective common knowledge,
that selects the world among those in the prospective local context which is the most similar
to the actual world and checks for the truth of the consequent at that world.48 Either way,
accepting an indicative conditional will have the effect of encoding an agreement about the
evolution of the conversation: the conversants agree that if they come to mutually accept
the antecedent of the conditional, they will mutually accept the consequent.
9.4
Subjunctive Conditionals
One benefit of the present approach is that on this approach, embedded epistemic modals
do not necessarily have any kind of epistemic flavor.49 This feature lets us straightforwardly
extend our account of indicative conditionals to subjunctive conditionals which have an overt
epistemic modal in the consequent, like
(23)
If Bob hadn’t been in his office, he might have been in the pub.
I propose the following semantics:
(24)
Subjunctive Conditional: Jϕ > em ψKs,w = 1 iff Jem ψKMs ∩JϕK
s ,w
=1
where > is the subjunctive conditional operator and Ms is the set of worlds metaphysically
accessible, in a contextually restricted sense, from the local context s.50 Then our shifted
47
Kratzer (1986), following Lewis (1975), or, for an alternate approach, Gillies (2010). This account would
yield a prospective form of a strict conditional account of indicatives. See Williams (2008) for discussion
of how to make an account like this work, following parallel discussions about subjunctives in von Fintel
(1999, 2001), and Gillies (2007).
48
And as in Stalnaker (2014).
49
This of course raises the question of whether they are really in the same semantic class as the bare modals
we considered above. I think they are, though I won’t argue for that conclusion here; in any case the best
way to decide is to work out and compare different theories of their behavior.
50
This is obviously going to be a restricted accessibility relation (along the lines of the modal horizons of
von Fintel (2001)), not the universal accessibility relation that figures in some philosophers’ analysis of
metaphysical necessity. I am assuming here that what worlds count as metaphysically accessible in a given
context can be read off of what is common knowledge in that context. This is mainly a notational convenience, and not meant to reflect a controversial position in the metaphysics of modality. Note that facts
about salience, which surely play a role in determining the accessibility relationship for counterfactuals,
can typically be read off of facts about what’s common knowledge, since usually some question or topic is
salient in a context just in case it’s common knowledge that it’s salient. The default assumption might be
29
modal domain function is defined as the smallest function fc′ meeting the constraint that
(25)
∀w ∶ fc′ (w) = Mfc (w) ∩ JϕKs
In other words, the shifted function takes us to the worlds metaphysically accessible
from the prospective common knowledge. A subjunctive conditional with an overt epistemic
modal, then, is a proposal to adjust the context set so that the worlds that are being treated
as metaphysically accessible are all ψ-worlds if they’re ϕ-worlds. We again have a choice of
how to extend this account to subjunctives that lack an overt epistemic modal; we could say
that those subjunctives contain something with the meaning of an epistemic necessity modal
in the consequent,51 or that they encode a Stalnaker-style selection modal, determined by
the prospective common knowledge, which selects from the accessible worlds the one most
similar to the world of evaluation.
Either way we go, the prospectivity built into this account encodes the way in which
accepting or rejecting subjunctive conditionals influences the prospective context set. Accepting a subjunctive conditional whose antecedent is compatible with the prior context
set amounts to accepting the corresponding indicative or material conditional. And accepting a subjunctive conditional whose antecedent is not compatible with the prior context
set amounts to accepting a certain constraint on what worlds should henceforth count as
metaphysically accessible or closest.
9.5
Attitude Verbs
The last case of embedded epistemic modals that I’ll consider is the case of epistemic modals
embedded under propositional attitude operators. As I mentioned above, things work slightly
differently here than in the cases considered so far. Formally, the cases are similar: attitude operators introduce local contexts which embedded epistemic modals can quantify over.
There are two important differences between the local contexts introduced by attitude operators versus those introduced by the operators we’ve looked at so far. First, the local contexts
introduced by attitude operators are not derivative on the global context set; they depend on
facts about the attitude of the agent in question, not on the speakers’ presuppositions about
those facts.52 Second, because of this, there is no role for a prospective feature for epistemic
that Ms = s. This assumption, of course, will have to be relaxed when the antecedent of the conditional is
incompatible with s, as will typically be the case. I think this rightly accounts for embedded subjunctives:
φ > (ψ > χ) is naturally interpreted as (φ ∧ ψ) > χ, unless φ and ψ are contextually incompatible.
51
This semantics would again be a prospective version of a strict conditional account; see, again, von Fintel
(1999, 2001) and Gillies (2007) for discussion of how to make this approach plausible by positing some
constraints on how the set of metaphysically accessible worlds evolves.
52
Though of course when the global context set is updated with an attitude attribution of any kind, this
update immediately updates the local contexts that represent what is compatible with the agent’s attitude,
30
modals. Epistemic modals embedded under attitude operators have truth conditions that
depend straightforwardly on the state of the world, and not the state of our common knowledge. Thus the prospectivity that we built in to our system to explain how epistemic modal
claims could be used to change the context set will no longer have a role to play. In cases like
this, then, epistemic modals will be evaluated relative to a function that delivers the local
context at a world, not the prospective local context. (The existence of variation like this
suggests, as discussed above, that the prospective interpretation of the constructions in question arises pragmatically due to the two-way connection between the assertion/acceptance
of the claim and the facts that the claim describes; when that connection is severed, the
prospectivity drops out.)
Our account of modals embedded under attitude verbs obviously depends on our background semantics for attitude verbs. The most conservative approach is as follows (focusing
on ‘believes’ for concreteness):
(26)
Believes: JS believes ϕKs,w = 1 iff ∀w′ ∈ BS,w ∶ JϕKBS,w ,w = 1,
′
where BS,⋅ is a function from worlds to the set of worlds compatible with S’s beliefs in that
world. Then, where the world of evaluation for the belief attribution is w, we can define fc′ ,
the shifted modal domain function, as the constant function on worlds meeting the constraint
(27)
∀w′ ∶ fc′ (w′ ) = BS,w
And likewise, mutatis mutandis for other non-factive attitudes.53 We thus predict, e.g., that
(28)
Rose believes Bob might be in his office.
is true just in case it is compatible with Rose’s beliefs that Bob is in his office. This seems
like a plausible prediction.
This approach will get the right results. But Kratzer’s approach to conditionals suggests
an alternative that I think is worth exploring:
(29)
Believes: JS believes em ϕKs,w = 1 iff Jem ϕKBS,w ,w = 1,
Then we take our shifted modal domain function to be the smallest function fc′ that meets
the constraint
given what we mutually accept: these local contexts are derivative on the global context set. The important
divergence with other cases we’ve looked at, then, is not in how attitude attributions update the global
context set, but rather that their truth value doesn’t usually depend on the global context.
53
Factive attitude operators like ‘knows’ get a similar but slightly different treatment. But that requires
discussion beyond our scope here; see discussion in Yalcin (2012b) and Moss (2013).
31
(30)
∀w ∶ fc′ (w) = BS,w
em is, again, any epistemic modal; when there is no overt epistemic modal, we assume a
covert ‘must’ or some modal like it.
This approach loosely follows an unpublished suggestion of Kratzer’s.54 It is less parsimonious than the first approach we considered insofar as it requires positing a covert ‘must’
in many cases. But it has other benefits. On this approach, attitude ascriptions, indicative
conditionals, and subjunctive conditionals behave very much alike: the role of the distinctive
operator in each case is simply to determine the shifted context, while the quantificational
force in each case comes from an explicit or implicit modal. This approach would also explain
a fact that otherwise is completely puzzling: why these operators do not have lexically simple
duals. If ‘believes’ were a modal operator, we might expect to find another word that’s the
dual of ‘believes’. But instead, when we want to express the dual of ‘believes’, we use ‘might’
under ‘believes’. Likewise for all of the other many attitude operators, as far as I know, in
natural language, and likewise for both indicative and subjunctive conditional operators.
This is precisely what we’d expect if these operators are not quantifiers but instead have as
their sole compositional contribution the introduction of a local context. In any case, either
approach will get the data right.
This concludes my discussion of embedded epistemic modals. In closing I note that, in
addition to cases we’ve considered and other overt modals we haven’t considered, sometimes
epistemic modals can be interpreted relative to a local context that is introduced tacitly. In
those cases I assume that a tacit operator is represented at the level of logical form whose
sole function is to introduce the local context.
10
Conclusion
In this paper I have tried to show that, pace much recent work in linguistics and philosophy,
we do not need to abandon the orthodox framework for understanding discourse dynamics
in order to account for epistemic modals: prospective contextualism provides an attractive
account of epistemic modals within that framework. In concluding, I will briefly discuss
some alternative analyses, and indicate why prospective contextualism might be preferable
to its competitors.
First, even where its competitors make similar predictions—as some do—there is some
dialectical reason to prefer prospective contextualism on the grounds of conservatism. Many
of the heterodox approaches to epistemic modals are motivated in part by arguments that
the orthodox framework can’t account for the behavior of epistemic modals. If this is wrong,
54
Sketched in Kratzer (2012); see those notes for further arguments for this approach.
32
it substantially reduces the motivation for pursuing alternative approaches. As Yalcin puts
it, ‘Because [the orthodoxy] makes no semantic or pragmatic waves, there is a certain presumption in favor of it’ (Yalcin, 2011, p. 300).
Another general positive point in favor of prospective contextualism is the generality of
the framework it provides for analyzing the language we use to negotiate about contextual
parameters; I return to this point in a moment. First, I will make some brief negative comments about three approaches that are similar to the present approach in terms of their
empirical predictions (and to which the present approach is indebted in obvious ways). In
each case I object to the style of pragmatic theory needed as a companion to the proposed
semantics for epistemic modals. These remarks merely sketch some reasons to prefer prospective contextualism over a few of its competitors; much more would need to be said to make
the case decisively.
The closest approach to mine is given in Stalnaker (2014). Stalnaker’s proposal is similar
in motivation and predictions to the present proposal, but rather than building prospectivity
into the meaning of a bare epistemic modal, Stalnaker proposes that epistemic modal claims
are coupled with a special force rule. The force rule specifies that an epistemic modal claim
is to be interpreted as a proposal to make the prospective context set satisfy the claim in
question, i.e. to make the prospective context set such that if the claim is interpreted with
that set as the domain of quantification for the epistemic modal, the claim comes out true:
The innovation comes when we specify the force rule. . . The proposal is to make
the rule prospective. . . In saying ⌜Might ϕ⌝, one is not asserting that ϕ is possible,
relative to the prior context. Rather, one is proposing to adjust the context (if
required) to bring it about that what the sentence says, relative to the posterior
context—the context as adjusted—is true. (Stalnaker, 2014, p.140)
Stalnaker’s account and prospective contextualism are empirically very similar, and much
of the work of this essay, which attempts to explore and motivate a style of account similar
to Stalnaker’s, should be friendly to Stalnaker’s proposal. There is a significant difference
between our accounts, however. On my proposal, an epistemic modal claim updates the
context set by intersection, like other assertions. By contrast, on Stalnaker’s proposal,
an epistemic modal claim is associated with the distinctive force rule just summarized,
instead of updating the context set by intersection, as other assertions do. I believe this
adds unneeded complexity to the pragmatic framework. It also entangles our pragmatic
theory with our lexical semantics in a way that looks unattractive to me, since what triggers
the distinctive prospective force rule, on Stalnaker’s account, is the specific meaning of a
given claim (because whether a claim is an epistemic modal claim or not depends just on
33
its meaning, and not on broader structural features of the claim).55 I think it would be
better to keep our pragmatics as general, and distinct from specific conventions, as possible.
Prospective contextualism gives us a similar story to Stalnaker’s account, but without a
special force rule, by building prospectivity into the semantics, and then relying on collapse
principles to derive the right force for updates with epistemic modal claims. The resulting
picture—according to which there is just one force rule for all assertions, modal or nonmodal—leaves us with a simpler and more general pragmatic theory than on Stalnaker’s
picture.
The second view I will discuss is the non-factualist expressivism developed in Yalcin (2007,
2011, 2012a). On Yalcin’s view, an epistemic modal claim does not express a proposition,
and as such is not truth-evaluable; rather, it directly makes a proposal about the context
set. I have some prima facie discomfort with the claim that epistemic modals are not truth
evaluable; this seems to clash with the fact that ordinary speakers do regularly ascribe truth
and falsity to epistemic modal claims, and this fits especially poorly with my intuitions about
‘must’-claims.56 A more substantial worry I have about Yalcin’s view is parallel to the worry
I have about Stalnaker’s view. Yalcin’s pragmatics is very much like Stalnaker’s, in that
epistemic modal claims are governed by a special force rule which dictates that they are
proposals to make the context set satisfy the claim in question, where a context set, again,
satisfies an epistemic modal claim just in case the claim is true when it is taken to quantify
over the context set. This kind of disjunctive pragmatics seems problematic to me for the
same reasons as Stalnaker’s does.57
The third view I will mention is the dynamic view advocated by Veltman (1996, 1985),
55
Thanks to Kai von Fintel for discussion. Compare the situation with imperatives and questions: in each
case, the orthodox pragmatic theory responds differently than it does to assertions, and so it must to some
degree be response to the meaning of what has been uttered. But in these cases the differences in meaning
are at a structural level, not the level of lexical semantics, since imperatives and questions do not, on
standard accounts, express propositions at all.
56
It is of course open to an expressivist to define a minimal truth predicate as a companion to her expressivism, as some expressivists about epistemic modality seem inclined to do; but, as is well known, it then
becomes difficult to specify what is distinctively expressivist about the theory in question.
57
I have glossed over some complexity here. Yalcin defines a notion ‘acceptance’ roughly as follows:
(31)
JϕKc is accepted at the context set of c, gc , iff for all worlds w in gc , JϕKc,gc ,w = 1
with truth conditions for epistemic modals that dictate that when an epistemic modal takes highest scope
in ϕ, the epistemic modal is interpreted as a quantifier over the second parameter in the index, i.e. in
this case over gc . Given this notion of acceptance, Yalcin can then formulate what looks like a unified
pragmatics: to assert ϕ is to propose to minimally change the context in order to make it accept JϕKc .
This permits an elegant statement of the pragmatics, but the underlying pragmatic mechanism is just
as disjunctive, and as sensitive to conventional markers, as on Stalnaker’s view: a non-modal claim will
update by intersection, a modal claim will update by checking to see whether the context set satisfies the
modal claim and minimally adjusting it so that it does, if needed.
34
according to which ‘might’ has an essentially dynamic, and not truth conditional, meaning.58
On this view ⌜Might ϕ⌝ is a test which, when asserted, returns the context set unchanged
just in case the context set is compatible with JϕKc , and otherwise returns the empty set
(‘crashes’). This view, again, needs to be coupled with an inelegant pragmatics—though in
somewhat different ways from Stalnaker’s and Yalcin’s views. This is easiest to see if we focus
on ‘must’. If ‘must’ is the dual of ‘might’, then the resulting theory of ‘must’ predicts that
⌜Must ϕ⌝ is a test which returns the context set just in case it entails JϕKc , and otherwise the
empty set (von Fintel and Gillies, 2007). A little empirical observation shows that ⌜Must ϕ⌝
is very often used in contexts which do not antecedently entail JϕKc . But then it follows that
almost any ‘must’ claim will precipitate a crash of the context set. This is, obviously, not
what actually happens, and so our pragmatic theory must tell a revisionary story: it must
explain why things almost never go as the semantic theory says they do. This makes the
semantic theory look like an idle wheel. It also wrongly assimilates any update with ‘must’ to
an assertion of a claim that is inconsistent with the antecedent context. Intuitively, there is a
very different feel to the update that happens when we antecedently took for granted JϕKc and
then someone asserts ⌜¬ϕ⌝, or ⌜Might ¬ϕ⌝, then when someone asserts ⌜Must ϕ⌝ in a context
that was antecedently compatible with, but didn’t entail, JϕK. Finally, it makes a distinction
where there does not seem to be one. For it looks like asserting ϕ and asserting ⌜Must ϕ⌝
update the context set in the same way (modulo issues about evidence and indirectness59 ):
by intersecting it with JϕKc . It strikes me as inelegant, therefore, to treat the two cases as
unalike, and to predict entirely different mechanisms by which they update the context set.
Another oddity of the dynamic view is that it tells a weird story about what we agree and
disagree about when we agree and disagree about epistemic modal claims. According to the
view, we are (dis)agreeing about whether to run some test on the context set, a test which
will either leave things as is or lead us to a crash. Thus, for instance, when Jack says that the
keys might be in the garage, and Ruth says ‘No’ in response, she is, on this view, proposing
that they not run the test that Jack has proposed to run. It takes quite a bit of work to
get from there to a story about why, if Ruth’s ‘No’ is accepted, Jack and Ruth subsequently
conclude that the keys aren’t in the garage—work that, it seems to me, should be done at
least partly by our semantic theory, rather than (what looks to me like) an excessively ad
hoc pragmatics.
So much for negative remarks; these are obviously inconclusive, but aim to give a sense
of why I am attracted to prospective contextualism over its competitors. Unlike these views,
58
Willer (2013) develops the dynamic view in interesting and worthwhile ways. It’s not obvious to me that
he avoids all these problems, though I cannot discuss his view here for reasons of space.
59
See von Fintel and Gillies (2010) for discussion.
35
the theory I am advocating treats assertions of modal and non-modal claims alike, from the
point of view of formal pragmatics, and thus has a simpler, less ad hoc pragmatics than
those views. One way to summarize the contribution of prospective contextualism, along
this dimension, is to present prospective contextualism as a solution to a certain functional
problem.60 According to the orthodox account of discourse dynamics, an assertion is a
proposal to update the context set with the asserted content in question. But assertions
sometimes have the effect, if accepted, of bringing it about that the context set is compatible
with (but doesn’t necessarily entail) a given proposition. Earlier researchers have concluded
that the orthodoxy can’t predict the possibility of both these kinds of update, and have thus
rejected the orthodoxy. On the contrary, I have argued that we can account for both kinds of
update without rejecting the simple and attractive story the orthodoxy tells about discourse
dynamics.
More broadly, the positive goal of this paper has been to show that the framework of
prospective contextualism provides a fruitful general strategy for analyzing the language
that we use to negotiate matters that depend, in part or in whole, on parameters of the
context. It is tempting when approaching these natural language constructions to think that
the orthodox framework is not applicable in these cases: for in these cases we seem not to
be describing how things are vis-`a-vis some contextual parameter, but rather proposing how
they ought to be. But we can predict this, and thus account for these constructions, by
taking these constructions to describe the way that parameter will be. Provided that the
identity of the parameter depends in the right way on what the interlocutors accept it to
be, constructions which describe how the parameter will come to be will have the force of
proposals about how the parameter ought to be.
References
Anscombe, G. (1963). Intention. Cornell University Press.
Austin, J. L. (1979). Performative Utterances. In Philosophical Papers. Oxford University Press.
Barker, C. (2002). Dynamics of Vagueness. Linguistics and Philosophy, 25(1):1–36.
Barker, C. (2009). Clarity and the Grammar of Skepticism. Mind and Language, 24(3):253–73.
Barker, C. (2013). Negotiating Taste. Inquiry, 56(2-3):240–257.
DeRose, K. (1991). Epistemic Possibilities. Philosophical Review, 100(4):581–605.
Dorr, C. and Hawthorne, J. (2013). Embedding Epistemic Modals. Mind, 122(488):867–913.
60
Thanks to Kai von Fintel for suggesting this perspective.
36
Egan, A., Hawthorne, J., and Weatherson, B. (2005). Epistemic Modals in Context. In Preyer, G. and
Peter, G., editors, Contextualism in Philosophy: Knowledge, Meaning and Truth, chapter 6, pages 131–
169. Oxford University Press, Oxford.
von Fintel, K. (1999). NPI Licensing, Strawson Entailment, and Context Dependency. Journal of Semantics,
16(2):97–148.
von Fintel, K. (2001). Counterfactuals in a Dynamic Context. In Ken Hale: A Life in Language. MIT Press.
von Fintel, K. (2008).
22(1):137–170.
What is Presupposition Accommodation, Again?
Philosophical Perspectives,
von Fintel, K. and Gillies, A. (2007). An Opinionated Guide to Epistemic Modality. In Gendler, T. and
Hawthorne, J., editors, Oxford Studies in Epistemology 2, pages 32–62. Oxford.
von Fintel, K. and Gillies, A. (2008). CIA Leaks. Philosophical Review, 117(1):77–98.
von Fintel, K. and Gillies, A. (2010). Must...Stay...Strong! Natural Language Semantics, 18(4):351–383.
von Fintel, K. and Gillies, A. (2011). ‘Might’ Made Right. In Egan, A. and Weatherson, B., editors, Epistemic
Modality. Oxford University Press.
Gillies, A. (2007). Counterfactual scorekeeping. Linguistics and Philosophy, 30(3):329–360.
Gillies, A. (2010). Iffiness. Semantics and Pragmatics, 3:1–42.
Hacking, I. (1967). Possibility. Philosophical Review, 76(2):143–168.
Heim, I. (1982). The Semantics of Definite and Indefinite Noun Phrases. PhD thesis, University of Massachusetts, Amherst.
Heim, I. (1983). On the Projection Problem for Presuppositions. In D. Flickinger et al., editor, Proceedings
of the Second West Coast Conference on Formal Linguistics, volume 2, pages 114–125, Stanford, CA.
Stanford University Press.
Heim, I. (1992). Presupposition Projection and the Semantics of Attitude Verbs. Journal of Semantics,
9:183–221.
Holliday, W. and Thomas Icard, I. (2013). Measure semantics and qualitative semantics for epistemic modals.
In Proceedings of Semantics and Linguistic Theory, volume 23, pages 514–534.
Karttunen, L. (1974). Presuppositions and Linguistic Context. Theoretical Linguistics, 1:181–194.
Khoo, J. and Knobe, J. (2015). Moral disagreement and exclusion. MS.
Klinedinst, N. and Rothschild, D. (2012). Connectives without truth tables. Natural Language Semantics,
20:137–175.
Knobe, J. and Yalcin, S. (2014). Epistemic modals and context: Experimental data. Semantics and Pragmatics, 7(10):1–21.
Kratzer, A. (1977). What ‘Must’ and ‘Can’ Must and Can Mean. Linguistics and Philosophy, 1(337-355).
Kratzer, A. (1981). The Notional Category of Modality. In Eikmeyer, H. and Rieser, H., editors, Words,
Worlds, and Contexts: New Approaches in Word Semantics. de Gruyter.
Kratzer, A. (1986). Conditionals. Chicago Linguistics Society, 22(2):1–15.
37
Kratzer, A. (1991). Modality. In von Stechow, A. and Wunderlich, D., editors, Semantics: An International
Handbook of Contemporary Research, pages 639–650. de Gruyter, Berlin.
Kratzer, A. (2012). The building blocks of attitude reports. Presentation to the Jowett Society.
Krifka, M. (2014). Embedding Illocutionary Acts. In Roeper, T. and Speas, M., editors, Recursion: Complexity in Cognition, volume 43 of Studies in Theoretical Psycholinguistics. Springer.
Lewis, D. (1970). General Semantics. Synthese, 22:18–67.
Lewis, D. (1975). Adverbs of Quantification. In Keenan, E., editor, Formal Semantics of Natural Language,
pages 3–15. Cambridge University Press.
MacFarlane, J. (2011). Epistemic Modals Are Assessment Sensitive. In Egan, A. and Weatherson, B., editors,
Epistemic Modality. Oxford University Press.
Mandelkern, M. (MS). Self-Determination and the Semantic Representation of Uncertainty.
Moore, G. (1962). Commonplace Book. Routledge.
Moss, S. (2013). Epistemology Formalized. Philosophical Review, 122(1):1–43.
Ninan, D. (2005). Two Puzzles About Deontic Necessity. In Gajewski, J., Hacquard, V., Nickel, B., and
Yalcin, S., editors, New Work on Modality, volume 51. MIT Working Papers in Linguistics.
Ninan, D. (2014). Relational semantics and domain semantics for epistemic modals. Notes based on lecture
at MIT in November 2012.
Ogihara, T. (2007). Tense and Aspect in truth-conditional semantics. Lingua, 117:392–418.
Price, H. (1983). Does ‘Probably’ Modify Sense? Australasian Journal of Philosophy, 61(4):396–408.
Ramsey, F. P. (1978). General Propositions and Causality. In Mellor, D., editor, Foundations: Essays in
Philosophy, Logic, Mathematics and Economics, pages 133–51. Routledge and Kegan Paul.
Schlenker, P. (2008). Be articulate: a pragmatic theory of presupposition projection. Theoretical Linguistics,
34:157–212.
Schlenker, P. (2010). Local contexts and local meanings. Philosophical Studies, 151:115–142.
Stalnaker, R. (1968). A Theory of Conditionals. In Rescher, N., editor, Studies in Logical Theory, pages
98–112. Oxford: Blackwell.
Stalnaker, R. (1970). Pragmatics. Synthese, 22:272–289.
Stalnaker, R. (1975). Indicative Conditionals. Philosophia, 5(3):269–86.
Stalnaker, R. (1978). Assertion. In Cole, P., editor, Syntax and semantics, volume 9, pages 315–322. New
York: Academic Press.
Stalnaker, R. (1984). Inquiry. MIT.
Stalnaker, R. (1998). On the Representation of Context. Journal of Logic, Language and Information,
7(1):3–19.
Stalnaker, R. (2002). Common Ground. Linguistics and Philosophy, 25:701–721.
Stalnaker, R. (2014). Context. Oxford University Press.
38
Swanson, E. (2015). The Application of Constraint Semantics to the Language of Subjective Uncertainty.
To appear in the Journal of Philosophical Logic.
Szabolcsi, A. (1982). Model Theoretic Semantics of Performatives. In Kiefer, F., editor, Hungarian General
Linguistics. John Benjamins.
Teller, P. (1972). Epistemic Possibility. Philosophia, 2(4):303–320.
Velleman, J. D. (1989). Epistemic Freedom. Pacific Philosophical Quarterly, 70:73–97.
Veltman, F. (1985). Logics for Conditionals. PhD thesis, University of Amsterdam.
Veltman, F. (1996). Defaults in Update Semantics. Journal of Philosophical Logic, 25(3):221–261.
Willer, M. (2013). Dynamics of Epistemic Modality. Philosophical Review, 122(1):45–92.
Williams, J. R. G. (2008). Conversations and conditionals. Philosophical Studies, 138:211–223.
Williamson, T. (2000). Knowledge and its Limits. Oxford University Press.
Yablo, S. (2011). A Problem About Permission and Possibility. In Epistemic Modality, pages 270–293.
Oxford University Press.
Yalcin, S. (2007). Epistemic Modals. Mind, 116(464):983–1026.
Yalcin, S. (2011). Nonfactualism About Epistemic Modality. In Egan, A. and Weatherson, B., editors,
Epistemic Modality, pages 295–332. Oxford University Press.
Yalcin, S. (2012a). Bayesian Expressivism. Proceedings of the Aristotelian Society, CXII(2):123–160.
Yalcin, S. (2012b). Context Probabilism. In Proceedings of the 18th Amsterdam Colloquium, pages 12–21.
39